Introduction
In this chapter we prove some results in commutative algebra which are less elementary than those in the first chapter on commutative algebra, see Algebra, Section 00AP. A reference is [MatCA].
Advice for the reader
More than in the chapter on commutative algebra, each of the sections in this chapter stands on its own. Starting with Section 0914 we freely use the (unbounded) derived category of modules over rings and all the machinery that comes with it.
Stably free modules
Here is what seems to be the generally accepted definition.
Definition
Let \(R\) be a ring.
Two modules \(M\), \(N\) over \(R\) are said to be stably isomorphic if there exist \(n, m \geq 0\) such that \(M \oplus R^{\oplus m} \cong N \oplus R^{\oplus n}\) as \(R\)-modules.
A module \(M\) is stably free if it is stably isomorphic to a free module.
Observe that a stably free module is projective.
Lemma
Let \(R\) be a ring. Let \(0 \to P' \to P \to P'' \to 0\) be a short exact sequence of finite projective \(R\)-modules. If \(2\) out of \(3\) of these modules are stably free, then so is the third.
Proof
Since the modules are projective, the sequence is split. Thus we can choose an isomorphism \(P = P' \oplus P''\). If \(P' \oplus R^{\oplus n}\) and \(P'' \oplus R^{\oplus m}\) are free, then we see that \(P \oplus R^{\oplus n + m}\) is free. Suppose that \(P'\) and \(P\) are stably free, say \(P \oplus R^{\oplus n}\) is free and \(P' \oplus R^{\oplus m}\) is free. Then \[P'' \oplus (P' \oplus R^{\oplus m}) \oplus R^{\oplus n} = (P'' \oplus P') \oplus R^{\oplus m} \oplus R^{\oplus n} = (P \oplus R^{\oplus n}) \oplus R^{\oplus m}\] is free. Thus \(P''\) is stably free. By symmetry we get the last of the three cases.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Assume that every element of \(1 + I\) is a unit (in other words \(I\) is contained in the Jacobson radical of \(R\)). For every finite stably free \(R/I\)-module \(E\) there exists a finite stably free \(R\)-module \(M\) such that \(M/IM \cong E\).
Proof
Choose a \(n\) and \(m\) and an isomorphism \(E \oplus (R/I)^{\oplus n} \cong (R/I)^{\oplus m}\). Choose \(R\)-linear maps \(\varphi : R^{\oplus m} \to R^{\oplus n}\) and \(\psi : R^{\oplus n} \to R^{\oplus m}\) lifting the projection \((R/I)^{\oplus m} \to (R/I)^{\oplus n}\) and injection \((R/I)^{\oplus n} \to (R/I)^{\oplus m}\). Then \(\varphi \circ \psi : R^{\oplus n} \to R^{\oplus n}\) reduces to the identity modulo \(I\). Thus the determinant of this map is invertible by our assumption on \(I\). Hence \(P = \Ker(\varphi)\) is stably free and lifts \(E\).
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Assume that every element of \(1 + I\) is a unit (in other words \(I\) is contained in the Jacobson radical of \(R\)). Let \(M\) be a finite flat \(R\)-module such that \(M/IM\) is a projective \(R/I\)-module. Then \(M\) is a finite projective \(R\)-module.
Proof
By Algebra, Lemma 00NZ we see that \(M_\mathfrak p\) is finite free for all prime ideals \(\mathfrak p \subset R\). By Algebra, Lemma 00NX it suffices to show that the function \(\rho_M : \mathfrak p \mapsto \dim_{\kappa(\mathfrak p)} M \otimes_R \kappa(\mathfrak p)\) is locally constant on \(\Spec(R)\). Because \(M/IM\) is finite projective, this is true on \(V(I) \subset \Spec(R)\). Since every closed point of \(\Spec(R)\) is in \(V(I)\) and since \(\rho_M(\mathfrak p) = \rho_M(\mathfrak q)\) whenever \(\mathfrak p \subset \mathfrak q \subset R\) are prime ideals, we conclude by an elementary argument on topological spaces which we omit.
The lift of Lemma 0BC5 is unique up to isomorphism by the following lemma.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Assume that every element of \(1 + I\) is a unit (in other words \(I\) is contained in the Jacobson radical of \(R\)). If \(P\) and \(P'\) are finite projective \(R\)-modules, then
if \(\varphi : P \to P'\) is an \(R\)-module map inducing an isomorphism \(\overline{\varphi} : P/IP \to P'/IP'\), then \(\varphi\) is an isomorphism,
if \(P/IP \cong P'/IP'\), then \(P \cong P'\).
Proof
Proof of (1). As \(P'\) is projective as an \(R\)-module we may choose a lift \(\psi : P' \to P\) of the map \(P' \to P'/IP' \xrightarrow{\overline{\varphi}^{-1}} P/IP\). By Nakayama’s lemma (Algebra, Lemma 00DV) \(\psi \circ \varphi\) and \(\varphi \circ \psi\) are surjective. Hence these maps are isomorphisms (Algebra, Lemma 05G8). Thus \(\varphi\) is an isomorphism.
Proof of (2). Choose an isomorphism \(P/IP \cong P'/IP'\). Since \(P\) is projective we can choose a lift \(\varphi : P \to P'\) of the map \(P \to P/IP \to P'/IP'\). Then \(\varphi\) is an isomorphism by (1).
A comment on the Artin-Rees property
Some of this material is taken from [conrad-dejong]. A general discussion with additional references can be found in [Eis, Section 1].
Let \(A\) be a Noetherian ring and let \(I \subset A\) be an ideal. Given a homomorphism \(f : M \to N\) of finite \(A\)-modules there exists a \(c \geq 0\) such that \[f(M) \cap I^nN \subset f(I^{n - c}M)\] for all \(n \geq c\), see Algebra, Lemma 00IO. In this situation we will say \(c\) works for \(f\) in the Artin-Rees lemma.
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal contained in the Jacobson radical of \(A\). Let \[S : L \xrightarrow{f} M \xrightarrow{g} N \quad\text{and}\quad S' : L \xrightarrow{f'} M \xrightarrow{g'} N\] be two complexes of finite \(A\)-modules as shown. Assume that
\(c\) works in the Artin-Rees lemma for \(f\) and \(g\),
the complex \(S\) is exact, and
\(f' = f \bmod I^{c + 1}M\) and \(g' = g \bmod I^{c + 1}N\).
Then \(c\) works in the Artin-Rees lemma for \(g'\) and the complex \(S'\) is exact.
Proof
We first show that \(g'(M) \cap I^nN \subset g'(I^{n - c}M)\) for \(n \geq c\). Let \(a\) be an element of \(M\) such that \(g'(a) \in I^nN\). We want to adjust \(a\) by an element of \(f'(L)\), i.e, without changing \(g'(a)\), so that \(a \in I^{n-c}M\). Assume that \(a \in I^rM\), where \(r < n - c\). Then \[g(a) = g'(a) + (g - g')(a) \in I^n N + I^{r + c + 1}N = I^{r + c + 1}N.\] By Artin-Rees for \(g\) we have \(g(a) \in g(I^{r + 1}M)\). Say \(g(a) = g(a_1)\) with \(a_1 \in I^{r + 1}M\). Since the sequence \(S\) is exact, \(a - a_1 \in f(L)\). Accordingly, we write \(a = f(b) + a_1\) for some \(b \in L\). Then \(f(b) = a - a_1 \in I^rM\). Artin-Rees for \(f\) shows that if \(r \geq c\), we may replace \(b\) by an element of \(I^{r - c}L\). Then in all cases, \(a = f'(b) + a_2\), where \(a_2 = (f - f')(b) + a_1 \in I^{r + 1}M\). (Namely, either \(c \geq r\) and \((f - f')(b) \in I^{r + 1}M\) by assumption, or \(c < r\) and \(b \in I^{r - c}\), whence again \((f - f')(b) \in I^{c + 1} I^{r - c} M = I^{r + 1}M\).) So we can adjust \(a\) by the element \(f'(b) \in f'(L)\) to increase \(r\) by \(1\).
In fact, the argument above shows that \((g')^{-1}(I^nN) \subset f'(L) + I^{n - c}M\) for all \(n \geq c\). Hence \(S'\) is exact because \[(g')^{-1}(0) = (g')^{-1}(\bigcap I^nN) \subset \bigcap f'(L) + I^{n - c}M = f'(L)\] as \(I\) is contained in the Jacobson radical of \(A\), see Algebra, Lemma 00IQ.
Given an ideal \(I \subset A\) of a ring \(A\) and an \(A\)-module \(M\) we set \[\text{Gr}_I(M) = \bigoplus I^nM/I^{n + 1}M.\] We think of this as a graded \(\text{Gr}_I(A)\)-module.
Lemma
Assumptions as in Lemma 07VE. Let \(Q = \Coker(g)\) and \(Q' = \Coker(g')\). Then \(\text{Gr}_I(Q) \cong \text{Gr}_I(Q')\) as graded \(\text{Gr}_I(A)\)-modules.
Proof
In degree \(n\) we have \(\text{Gr}_I(Q)_n = I^nN/(I^{n + 1}N + g(M) \cap I^nN)\) and similarly for \(Q'\). We claim that \[g(M) \cap I^nN \subset I^{n + 1}N + g'(M) \cap I^nN.\] By symmetry (the proof of the claim will only use that \(c\) works for \(g\) which also holds for \(g'\) by the lemma) this will imply that \[I^{n + 1}N + g(M) \cap I^nN = I^{n + 1}N + g'(M) \cap I^nN\] whence \(\text{Gr}_I(Q)_n\) and \(\text{Gr}_I(Q')_n\) agree as subquotients of \(N\), implying the lemma. Observe that the claim is clear for \(n \leq c\) as \(g = g' \bmod I^{c + 1}N\). If \(n > c\), then suppose \(b \in g(M) \cap I^nN\). Write \(b = g(a)\) for \(a \in I^{n - c}M\). Set \(b' = g'(a)\). We have \(b - b' = (g - g')(a) \in I^{n + 1}N\) as desired.
Lemma
Let \(A \to B\) be a flat map of Noetherian rings. Let \(I \subset A\) be an ideal. Let \(f : M \to N\) be a homomorphism of finite \(A\)-modules. Assume that \(c\) works for \(f\) in the Artin-Rees lemma. Then \(c\) works for \(f \otimes 1 : M \otimes_A B \to N \otimes_A B\) in the Artin-Rees lemma for the ideal \(IB\).
Proof
Note that \[(f \otimes 1)(M) \cap I^n N \otimes_A B = (f \otimes 1)\left((f \otimes 1)^{-1}(I^n N \otimes_A B)\right)\] On the other hand, \[\begin{align*} (f \otimes 1)^{-1}(I^n N \otimes_A B) & = \Ker(M \otimes_A B \to N \otimes_A B/(I^n N \otimes_A B)) \\ & = \Ker(M \otimes_A B \to (N/I^nN) \otimes_A B) \end{align*}\] As \(A \to B\) is flat taking kernels and cokernels commutes with tensoring with \(B\), whence this is equal to \(f^{-1}(I^nN) \otimes_A B\). By assumption \(f^{-1}(I^nN)\) is contained in \(\Ker(f) + I^{n - c}M\). Thus the lemma holds.
Fibre products of rings, I
Fibre products of rings have to do with pushouts of schemes. Some cases of pushouts of schemes are discussed in More on Morphisms, Section 07RS.
Lemma
Let \(R\) be a ring. Let \(A \to B\) and \(C \to B\) be \(R\)-algebra maps. Assume
\(R\) is Noetherian,
\(A\), \(B\), \(C\) are of finite type over \(R\),
\(A \to B\) is surjective, and
\(B\) is finite over \(C\).
Then \(A \times_B C\) is of finite type over \(R\).
Proof
Set \(D = A \times_B C\). There is a commutative diagram \[\xymatrix{ 0 & B \ar[l] & A \ar[l] & I \ar[l] & 0 \ar[l] \\ 0 & C \ar[l] \ar[u] & D \ar[l] \ar[u] & I \ar[l] \ar[u] & 0 \ar[l] }\] with exact rows. Choose \(y_1, \ldots, y_n \in B\) which are generators for \(B\) as a \(C\)-module. Choose \(x_i \in A\) mapping to \(y_i\). Then \(1, x_1, \ldots, x_n\) are generators for \(A\) as a \(D\)-module. The map \(D \to A \times C\) is injective, and the ring \(A \times C\) is finite as a \(D\)-module (because it is the direct sum of the finite \(D\)-modules \(A\) and \(C\)). Hence the lemma follows from the Artin-Tate lemma (Algebra, Lemma 00IS).
Lemma
Let \(R\) be a Noetherian ring. Let \(I\) be a finite set. Suppose given a cartesian diagram \[\xymatrix{ \prod B_i & \prod A_i \ar[l]^{\prod \varphi_i} \\ Q \ar[u]^{\prod \psi_i} & P \ar[u] \ar[l] }\] with \(\psi_i\) and \(\varphi_i\) surjective, and \(Q\), \(A_i\), \(B_i\) of finite type over \(R\). Then \(P\) is of finite type over \(R\).
Proof
Follows from Lemma 00IT and induction on the size of \(I\). Namely, let \(I = I' \amalg \{i_0\}\). Let \(P'\) be the ring defined by the diagram of the lemma using \(I'\). Then \(P'\) is of finite type by induction hypothesis. Finally, \(P\) sits in a fibre product diagram \[\xymatrix{ B_{i_0} & A_{i_0} \ar[l] \\ P' \ar[u] & P \ar[u] \ar[l] & }\] to which the lemma applies.
Lemma
Suppose given a cartesian diagram of rings \[\xymatrix{ R & R' \ar[l]^t \\ B \ar[u]_s & B'\ar[u] \ar[l] }\] i.e., \(B' = B \times_R R'\). If \(h \in B'\) corresponds to \(g \in B\) and \(f \in R'\) such that \(s(g) = t(f)\), then the diagram \[\xymatrix{ R_{s(g)} = R_{t(f)} & (R')_f \ar[l]^-t \\ B_g \ar[u]_s & (B')_h \ar[u] \ar[l] }\] is cartesian too.
Proof
The equality \(B' = B \times_R R'\) tells us that \[0 \to B' \to B \oplus R' \xrightarrow{s, -t} R\] is an exact sequence of \(B'\)-modules. We have \(B_g = B_h\), \(R'_f = R'_h\), and \(R_{s(g)} = R_{t(f)} = R_h\) as \(B'\)-modules. By exactness of localization (Algebra, Proposition 00CS) we find that \[0 \to B'_h \to B_g \oplus R'_f \xrightarrow{s, -t} R_{s(g)} = R_{t(f)}\] is an exact sequence. This proves the lemma.
Consider a commutative diagram of rings \[\xymatrix{ R & R' \ar[l] \\ B \ar[u] & B' \ar[u] \ar[l] }\] Consider the functor (where the fibre product of categories is as constructed in Categories, Example 003R) [0D2E]\[\begin{equation} \text{Mod}_{B'} \longrightarrow \text{Mod}_B \times_{\text{Mod}_R} \text{Mod}_{R'},\quad L' \longmapsto (L' \otimes_{B'} B, L' \otimes_{B'} R', can) \end{equation}\] where \(can\) is the canonical identification \(L' \otimes_{B'} B \otimes_B R = L' \otimes_{B'} R' \otimes_{R'} R\). In the following we will write \((N, M', \varphi)\) for an object of the right hand side, i.e., \(N\) is a \(B\)-module, \(M'\) is an \(R'\)-module and \(\varphi : N \otimes_B R \to M' \otimes_{R'} R\) is an isomorphism.
Lemma
Given a commutative diagram of rings \[\xymatrix{ R & R' \ar[l] \\ B \ar[u] & B' \ar[u] \ar[l] }\] the functor (0D2E) has a right adjoint, namely the functor \[F : (N, M', \varphi) \longmapsto N \times_\varphi M'\] (see proof for elucidation).
Proof
Given an object \((N, M', \varphi)\) of the category \(\text{Mod}_B \times_{\text{Mod}_R} \text{Mod}_{R'}\) we set \[N \times_\varphi M' = \{(n, m') \in N \times M' \mid \varphi(n \otimes 1) = m' \otimes 1\text{ in }M' \otimes_{R'} R\}\] viewed as a \(B'\)-module. The adjointness statement is that for a \(B'\)-module \(L'\) and a triple \((N, M', \varphi)\) we have \[\Hom_{B'}(L', N \times_\varphi M') = \Hom_B(L' \otimes_{B'} B, N) \times_{\Hom_R(L' \otimes_{B'} R, M' \otimes_{R'} R)} \Hom_{R'}(L' \otimes_{B'} R', M')\] By Algebra, Lemma 05DQ the right hand side is equal to \[\Hom_{B'}(L', N) \times_{\Hom_{B'}(L', M' \otimes_{R'} R)} \Hom_{B'}(L', M')\] Thus it is clear that for a pair \((g, f')\) of elements of this fibre product we get an \(B'\)-linear map \(L' \to N \times_\varphi M'\), \(l' \mapsto (g(l'), f'(l'))\). Conversely, given a \(B'\) linear map \(g' : L' \to N \times_\varphi M'\) we can set \(g\) equal to the composition \(L' \to N \times_\varphi M' \to N\) and \(f'\) equal to the composition \(L' \to N \times_\varphi M' \to M'\). These constructions are mutually inverse to each other and define the desired isomorphism.
Fibre products of rings, II
In this section we discuss fibre products in the following situation.
Situation
In the following we will consider ring maps \[\xymatrix{ B \ar[r] & A & A' \ar[l] }\] where we assume \(A' \to A\) is surjective with kernel \(I\). In this situation we set \(B' = B \times_A A'\) to obtain a cartesian square \[\xymatrix{ A & A' \ar[l] \\ B \ar[u] & B' \ar[l] \ar[u] }\]
Lemma
In Situation 08KH we have \[\Spec(B') = \Spec(B) \amalg_{\Spec(A)} \Spec(A')\] as topological spaces.
Proof
Since \(B' = B \times_A A'\) we obtain a commutative square of spectra, which induces a continuous map \[can : \Spec(B) \amalg_{\Spec(A)} \Spec(A') \longrightarrow \Spec(B')\] as the source is a pushout in the category of topological spaces (which exists by Topology, Section 0B1W).
To show the map \(can\) is surjective, let \(\mathfrak q' \subset B'\) be a prime ideal. If \(I \subset \mathfrak q'\) (here and below we take the liberty of considering \(I\) as an ideal of \(B'\) as well as an ideal of \(A'\)), then \(\mathfrak q'\) corresponds to a prime ideal of \(B\) and is in the image. If not, then pick \(h \in I\), \(h \not \in \mathfrak q'\). In this case \(B_h = A_h = 0\) and the ring map \(B'_h \to A'_h\) is an isomorphism, see Lemma 01Z8. Thus we see that \(\mathfrak q'\) corresponds to a unique prime ideal \(\mathfrak p' \subset A'\) which does not contain \(I\).
Since \(B' \to B\) is surjective, we see that \(can\) is injective on the summand \(\Spec(B)\). We have seen above that \(\Spec(A') \to \Spec(B')\) is injective on the complement of \(V(I) \subset \Spec(A')\). Since \(V(I) \subset \Spec(A')\) is exactly the image of \(\Spec(A) \to \Spec(A')\) a trivial set theoretic argument shows that \(can\) is injective.
To finish the proof we have to show that \(can\) is open. To do this, observe that an open of the pushout is of the form \(V \amalg U'\) where \(V \subset \Spec(B)\) and \(U' \subset \Spec(A')\) are opens whose inverse images in \(\Spec(A)\) agree. Let \(v \in V\). We can find a \(g \in B\) such that \(v \in D(g) \subset V\). Let \(f \in A\) be the image. Pick \(f' \in A'\) mapping to \(f\). Then \(D(f') \cap U' \cap V(I) = D(f') \cap V(I)\). Hence \(V(I) \cap D(f')\) and \(D(f') \cap (U')^c\) are disjoint closed subsets of \(D(f') = \Spec(A'_{f'})\). Write \((U')^c = V(J)\) for some ideal \(J \subset A'\). Since \(A'_{f'} \to A'_{f'}/IA'_{f'} \times A'_{f'}/JA'_{f'}\) is surjective by the disjointness just shown, we can find an \(a'' \in A'_{f'}\) mapping to \(1\) in \(A'_{f'}/IA'_{f'}\) and mapping to zero in \(A'_{f'}/JA'_{f'}\). Clearing denominators, we find an element \(a' \in J\) mapping to \(f^n\) in \(A\). Then \(D(a'f') \subset U'\). Let \(h' = (g^{n + 1}, a'f') \in B'\). Since \(B'_{h'} = B_{g^{n + 1}} \times_{A_{f^{n + 1}}} A'_{a'f'}\) by a previously cited lemma, we see that \(D(h')\) pulls back to an open neighbourhood of \(v\) in the pushout, i.e., the image of \(V \amalg U'\) contains an open neighbourhood of the image of \(v\). We omit the (easier) proof that the same thing is true for \(u' \in U'\) with \(u' \not \in V(I)\).
Lemma
In Situation 08KH if \(B \to A\) is integral, then \(B' \to A'\) is integral.
Proof
Let \(a' \in A'\) with image \(a \in A\). Let \(x^d + b_1 x^{d - 1} + \ldots + b_d\) be a monical polynomial with coefficients in \(B\) satisfied by \(a\). Choose \(b'_i \in B'\) mapping to \(b_i \in B\) (possible). Then \((a')^d + b'_1 (a')^{d - 1} + \ldots + b'_d\) is in the kernel of \(A' \to A\). Since \(\Ker(B' \to B) = \Ker(A' \to A)\) we can modify our choice of \(b'_d\) to get \((a')^d + b'_1 (a')^{d - 1} + \ldots + b'_d = 0\) as desired.
In Situation 08KH we’d like to understand \(B'\)-modules in terms of modules over \(A'\), \(A\), and \(B\). In order to do this we consider the functor (where the fibre product of categories as constructed in Categories, Example 003R) [08KI]\[\begin{equation} \text{Mod}_{B'} \longrightarrow \text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'},\quad L' \longmapsto (L' \otimes_{B'} B, L' \otimes_{B'} A', can) \end{equation}\] where \(can\) is the canonical identification \(L' \otimes_{B'} B \otimes_B A = L' \otimes_{B'} A' \otimes_{A'} A\). In the following we will write \((N, M', \varphi)\) for an object of the right hand side, i.e., \(N\) is a \(B\)-module, \(M'\) is an \(A'\)-module and \(\varphi : N \otimes_B A \to M' \otimes_{A'} A\) is an isomorphism. However, it is often more convenient think of \(\varphi\) as a \(B\)-linear map \(\varphi : N \to M'/IM'\) which induces an isomorphism \(N \otimes_B A \to M' \otimes_{A'} A = M'/IM'\).
Lemma
In Situation 08KH the functor (08KI) has a right adjoint, namely the functor \[F : (N, M', \varphi) \longmapsto N \times_{\varphi, M} M'\] where \(M = M'/IM'\). Moreover, the composition of \(F\) with (08KI) is the identity functor on \(\text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'}\). In other words, setting \(N' = N \times_{\varphi, M} M'\) we have \(N' \otimes_{B'} B = N\) and \(N' \otimes_{B'} A' = M'\).
Proof
The adjointness statement follows from the more general Lemma 0D2F. To prove the final assertion, recall that \(B' = B \times_A A'\) and \(N' = N \times_{\varphi, M} M'\) and extend these equalities to \[\vcenter{ \xymatrix{ A & A' \ar[l] & I \ar[l] \\ B \ar[u] & B' \ar[l] \ar[u] & J \ar[l] \ar[u] } } \quad\text{and}\quad \vcenter{ \xymatrix{ M & M' \ar[l] & K \ar[l] \\ N \ar[u]_\varphi & N' \ar[l] \ar[u] & L \ar[l] \ar[u] } }\] where \(I, J, K, L\) are the kernels of the horizontal maps of the original diagrams. We present the proof as a sequence of observations:
\(K = IM'\) (see statement lemma),
\(B' \to B\) is surjective with kernel \(J\) and \(J \to I\) is bijective,
\(N' \to N\) is surjective with kernel \(L\) and \(L \to K\) is bijective,
\(JN' \subset L\),
\(\Im(N \to M)\) generates \(M\) as an \(A\)-module (because \(N \otimes_B A = M\)),
\(\Im(N' \to M')\) generates \(M'\) as an \(A'\)-module (because it holds modulo \(K\) and \(L\) maps isomorphically to \(K\)),
\(JN' = L\) (because \(L \cong K = I M'\) is generated by images of elements \(x n'\) with \(x \in I\) and \(n' \in N'\) by the previous statement),
\(N' \otimes_{B'} B = N\) (because \(N = N'/L\), \(B = B'/J\), and the previous statement),
there is a map \(\gamma : N' \otimes_{B'} A' \to M'\),
\(\gamma\) is surjective (see above),
the kernel of the composition \(N' \otimes_{B'} A' \to M' \to M\) is generated by elements \(l \otimes 1\) and \(n' \otimes x\) with \(l \in K\), \(n' \in N'\), \(x \in I\) (because \(M = N \otimes_B A\) by assumption and because \(N' \to N\) and \(A' \to A\) are surjective with kernels \(L\) and \(I\)),
any element of \(N' \otimes_{B'} A'\) in the submodule generated by the elements \(l \otimes 1\) and \(n' \otimes x\) with \(l \in L\), \(n' \in N'\), \(x \in I\) can be written as \(l \otimes 1\) for some \(l \in L\) (because \(J\) maps isomorphically to \(I\) we see that \(n' \otimes x = n'x \otimes 1\) in \(N' \otimes_{B'} A'\); similarly \(x n' \otimes a' = n' \otimes xa' = n'(xa') \otimes 1\) in \(N' \otimes_{B'} A'\) when \(n' \in N'\), \(x \in J\) and \(a' \in A'\); since we have seen that \(JN' = L\) this proves the assertion),
the kernel of \(\gamma\) is zero (because by (10) and (11) any element of the kernel is of the form \(l \otimes 1\) with \(l \in L\) which is mapped to \(l \in K \subset M'\) by \(\gamma\)).
This finishes the proof.
Lemma
In the situation of Lemma 07RU for a \(B'\)-module \(L'\) the adjunction map \[L' \longrightarrow (L' \otimes_{B'} B) \times_{(L' \otimes_{B'} A)} (L' \otimes_{B'} A')\] is surjective but in general not injective.
Proof
As in the proof of Lemma 07RU let \(J \subset B'\) be the kernel of the map \(B' \to B\). Then \(L' \otimes_{B'} B = L'/JL'\). Hence to prove surjectivity it suffices to show that elements of the form \((0, z)\) of the fibre product are in the image of the map of the lemma. The kernel of the map \(L' \otimes_{B'} A' \to L' \otimes_{B'} A\) is the image of \(L' \otimes_{B'} I \to L' \otimes_{B'} A'\). Since the map \(J \to I\) induced by \(B' \to A'\) is an isomorphism the composition \[L' \otimes_{B'} J \to L' \to (L' \otimes_{B'} B) \times_{(L' \otimes_{B'} A)} (L' \otimes_{B'} A')\] induces a surjection of \(L' \otimes_{B'} J\) onto the set of elements of the form \((0, z)\). To see the map is not injective in general we present a simple example. Namely, take a field \(k\), set \(B' = k[x, y]/(xy)\), \(A' = B'/(x)\), \(B = B'/(y)\), \(A = B'/(x, y)\) and \(L' = B'/(x - y)\). In that case the class of \(x\) in \(L'\) is nonzero but is mapped to zero under the displayed arrow.
Lemma
In Situation 08KH let \((N_1, M'_1, \varphi_1) \to (N_2, M'_2, \varphi_2)\) be a morphism of \(\text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'}\) with \(N_1 \to N_2\) and \(M'_1 \to M'_2\) surjective. Then \[N_1 \times_{\varphi_1, M_1} M'_1 \to N_2 \times_{\varphi_2, M_2} M'_2\] where \(M_1 = M'_1/IM'_1\) and \(M_2 = M'_2/IM'_2\) is surjective.
Proof
Pick \((x_2, y_2) \in N_2 \times_{\varphi_2, M_2} M'_2\). Choose \(x_1 \in N_1\) mapping to \(x_2\). Since \(M'_1 \to M_1\) is surjective we can find \(y_1 \in M'_1\) mapping to \(\varphi_1(x_1)\). Then \((x_1, y_1)\) maps to \((x_2, y'_2)\) in \(N_2 \times_{\varphi_2, M_2} M'_2\). Thus it suffices to show that elements of the form \((0, y_2)\) are in the image of the map. Here we see that \(y_2 \in IM'_2\). Write \(y_2 = \sum t_i y_{2, i}\) with \(t_i \in I\). Choose \(y_{1, i} \in M'_1\) mapping to \(y_{2, i}\). Then \(y_1 = \sum t_iy_{1, i} \in IM'_1\) and the element \((0, y_1)\) does the job.
Lemma
Let \(A, A', B, B', I, M, M', N, \varphi\) be as in Lemma 07RU. If \(N\) finite over \(B\) and \(M'\) finite over \(A'\), then \(N' = N \times_{\varphi, M} M'\) is finite over \(B'\).
Proof
We will use the results of Lemma 07RU without further mention. Choose generators \(y_1, \ldots, y_r\) of \(N\) over \(B\) and generators \(x_1, \ldots, x_s\) of \(M'\) over \(A'\). Using that \(N = N' \otimes_{B'} B\) and \(B' \to B\) is surjective we can find \(u_1, \ldots, u_r \in N'\) mapping to \(y_1, \ldots, y_r\) in \(N\). Using that \(M' = N' \otimes_{B'} A'\) we can find \(v_1, \ldots, v_t \in N'\) such that \(x_i = \sum v_j \otimes a'_{ij}\) for some \(a'_{ij} \in A'\). In particular we see that the images \(\overline{v}_j \in M'\) of the \(v_j\) generate \(M'\) over \(A'\). We claim that \(u_1, \ldots, u_r, v_1, \ldots, v_t\) generate \(N'\) as a \(B'\)-module. Namely, pick \(\xi \in N'\). We first choose \(b'_1, \ldots, b'_r \in B'\) such that \(\xi\) and \(\sum b'_i u_i\) map to the same element of \(N\). This is possible because \(B' \to B\) is surjective and \(y_1, \ldots, y_r\) generate \(N\) over \(B\). The difference \(\xi - \sum b'_i u_i\) is of the form \((0, \theta)\) for some \(\theta\) in \(IM'\). Say \(\theta\) is \(\sum t_j\overline{v}_j\) with \(t_j \in I\). As \(J = \Ker(B' \to B)\) maps isomorphically to \(I\) we can choose \(s_j \in J \subset B'\) mapping to \(t_j\). Because \(N' = N \times_{\varphi, M} M'\) it follows that \(\xi = \sum b'_i u_i + \sum s_j v_j\) as desired.
Lemma
With \(A, A', B, B', I\) as in Situation 08KH.
Let \((N, M', \varphi)\) be an object of \(\text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'}\). If \(M'\) is flat over \(A'\) and \(N\) is flat over \(B\), then \(N' = N \times_{\varphi, M} M'\) is flat over \(B'\).
If \(L'\) is a flat \(B'\)-module, then \(L' = (L \otimes_{B'} B) \times_{(L \otimes_{B'} A)} (L \otimes_{B'} A')\).
The category of flat \(B'\)-modules is equivalent to the full subcategory of \(\text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'}\) consisting of triples \((N, M', \varphi)\) with \(N\) flat over \(B\) and \(M'\) flat over \(A'\).
Proof
In the proof we will use Lemma 07RU without further mention.
Proof of (1). Set \(J = \Ker(B' \to B)\). This is an ideal of \(B'\) mapping isomorphically to \(I = \Ker(A' \to A)\). Let \(\mathfrak b' \subset B'\) be an ideal. We have to show that \(\mathfrak b' \otimes_{B'} N' \to N'\) is injective, see Algebra, Lemma 00HD. We know that \[\mathfrak b'/(\mathfrak b' \cap J) \otimes_{B'} N' = \mathfrak b'/(\mathfrak b' \cap J) \otimes_B N \to N\] is injective as \(N\) is flat over \(B\). As \(\mathfrak b' \cap J \to \mathfrak b' \to \mathfrak b'/(\mathfrak b' \cap J) \to 0\) is exact, we conclude that it suffices to show that \((\mathfrak b' \cap J) \otimes_{B'} N' \to N'\) is injective. Thus we may assume that \(\mathfrak b' \subset J\). Next, since \(J \to I\) is an isomorphism we have \[J \otimes_{B'} N' = I \otimes_{A'} A' \otimes_{B'} N' = I \otimes_{A'} M'\] which maps injectively into \(M'\) as \(M'\) is a flat \(A'\)-module. Hence \(J \otimes_{B'} N' \to N'\) is injective and we conclude that \(\text{Tor}_1^{B'}(B'/J, N') = 0\), see Algebra, Remark 00M6. Thus we may apply Algebra, Lemma 051C to \(N'\) over \(B'\) and the ideal \(J\). Going back to our ideal \(\mathfrak b' \subset J\), let \(\mathfrak b' \subset \mathfrak b'' \subset J\) be the smallest ideal whose image in \(I\) is an \(A'\)-submodule of \(I\). In other words, we have \(\mathfrak b'' = A' \mathfrak b'\) if we view \(J = I\) as \(A'\)-module. Then \(\mathfrak b''/\mathfrak b'\) is killed by \(J\) and we get a short exact sequence \[0 \to \mathfrak b' \otimes_{B'} N' \to \mathfrak b'' \otimes_{B'} N' \to \mathfrak b''/\mathfrak b' \otimes_{B'} N' \to 0\] by the vanishing of \(\text{Tor}_1^{B'}(\mathfrak b''/\mathfrak b', N')\) we get from the application of the lemma. Thus we may replace \(\mathfrak b'\) by \(\mathfrak b''\). In particular we may assume \(\mathfrak b'\) is an \(A'\)-module and maps to an ideal of \(A'\). Then \[\mathfrak b' \otimes_{B'} N' = \mathfrak b' \otimes_{A'} A' \otimes_{B'} N' = \mathfrak b' \otimes_{A'} M'\] This tensor product maps injectively into \(M'\) by our assumption that \(M'\) is flat over \(A'\). We conclude that \(\mathfrak b' \otimes_{B'} N' \to N' \to M'\) is injective and hence the first map is injective as desired.
Proof of (2). This follows by tensoring the short exact sequence \(0 \to B' \to B \oplus A' \to A \to 0\) with \(L'\) over \(B'\).
Proof of (3). Immediate consequence of (1) and (2).
Lemma
Let \(A, A', B, B', I\) be as in Situation 08KH. The category of finite projective \(B'\)-modules is equivalent to the full subcategory of \(\text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'}\) consisting of triples \((N, M', \varphi)\) with \(N\) finite projective over \(B\) and \(M'\) finite projective over \(A'\).
Proof
Recall that a module is finite projective if and only if it is finitely presented and flat, see Algebra, Lemma 00NX. Using Lemmas 0D2I and 0D2H we reduce to showing that \(N' = N \times_{\varphi, M} M'\) is a \(B'\)-module of finite presentation if \(N\) finite projective over \(B\) and \(M'\) finite projective over \(A'\).
By Lemma 0D2H the module \(N'\) is finite over \(B'\). Choose a surjection \((B')^{\oplus n} \to N'\) with kernel \(K'\). By base change we obtain maps \(B^{\oplus n} \to N\), \((A')^{\oplus n} \to M'\), and \(A^{\oplus n} \to M\) with kernels \(K_B\), \(K_{A'}\), and \(K_A\). There is a canonical map \[K' \longrightarrow K_B \times_{K_A} K_{A'}\] On the other hand, since \(N' = N \times_{\varphi, M} M'\) and \(B' = B \times_A A'\) there is also a canonical map \(K_B \times_{K_A} K_{A'} \to K'\) inverse to the displayed arrow. Hence the displayed map is an isomorphism. By Algebra, Lemma 0519 the modules \(K_B\) and \(K_{A'}\) are finite. We conclude from Lemma 0D2H that \(K'\) is a finite \(B'\)-module provided that \(K_B \to K_A\) and \(K_{A'} \to K_A\) induce isomorphisms \(K_B \otimes_B A = K_A = K_{A'} \otimes_{A'} A\). This is true because the flatness assumptions implies the sequences \[0 \to K_B \to B^{\oplus n} \to N \to 0 \quad\text{and}\quad 0 \to K_{A'} \to (A')^{\oplus n} \to M' \to 0\] stay exact upon tensoring, see Algebra, Lemma 00HL.
Fibre products of rings, III
In this section we discuss fibre products in the following situation.
Situation
Let \(A, A', B, B', I\) be as in Situation 08KH. Let \(B' \to D'\) be a ring map. Set \(D = D' \otimes_{B'} B\), \(C' = D' \otimes_{B'} A'\), and \(C = D' \otimes_{B'} A\). This leads to a big commutative diagram \[\xymatrix{ C & & & C' \ar[lll] \\ & A \ar[ul] & A' \ar[l] \ar[ru] \\ & B \ar[u] \ar[ld] & B' \ar[l] \ar[u] \ar[rd] \\ D \ar[uuu] & & & D' \ar[lll] \ar[uuu] }\] of rings. Observe that we do not assume that the map \(D' \to D \times_C C'\) is an isomorphism1. In this situation we have the functor [08KL]\[\begin{equation} \text{Mod}_{D'} \longrightarrow \text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'},\quad L' \longmapsto (L' \otimes_{D'} D, L' \otimes_{D'} C', can) \end{equation}\] analogous to (08KI). Note that \(L' \otimes_{D'} D = L \otimes_{D'} (D' \otimes_{B'} B) = L \otimes_{B'} B\) and similarly \(L' \otimes_{D'} C' = L \otimes_{D'} (D' \otimes_{B'} A') = L \otimes_{B'} A'\) hence the diagram \[\xymatrix{ \text{Mod}_{D'} \ar[r] \ar[d] & \text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'} \ar[d] \\ \text{Mod}_{B'} \ar[r] & \text{Mod}_B \times_{\text{Mod}_A} \text{Mod}_{A'} }\] is commutative. In the following we will write \((N, M', \varphi)\) for an object of \(\text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'}\), i.e., \(N\) is a \(D\)-module, \(M'\) is an \(C'\)-module and \(\varphi : N \otimes_B A \to M' \otimes_{A'} A\) is an isomorphism of \(C\)-modules. However, it is often more convenient think of \(\varphi\) as a \(D\)-linear map \(\varphi : N \to M'/IM'\) which induces an isomorphism \(N \otimes_B A \to M' \otimes_{A'} A = M'/IM'\).
Lemma
In Situation 08KK the functor (08KL) has a right adjoint, namely the functor \[F : (N, M', \varphi) \longmapsto N \times_{\varphi, M} M'\] where \(M = M'/IM'\). Moreover, the composition of \(F\) with (08KL) is the identity functor on \(\text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'}\). In other words, setting \(N' = N \times_{\varphi, M} M'\) we have \(N' \otimes_{D'} D = N\) and \(N' \otimes_{D'} C' = M'\).
Proof
The adjointness statement follows from the more general Lemma 0D2F. The final assertion follows from the corresponding assertion of Lemma 07RU because \(N' \otimes_{D'} D = N' \otimes_{D'} D' \otimes_{B'} B = N' \otimes_{B'} B\) and \(N' \otimes_{D'} C' = N' \otimes_{D'} D' \otimes_{B'} A' = N' \otimes_{B'} A'\).
Lemma
In Situation 08KK the map \(JD' \to IC'\) is surjective where \(J = \Ker(B' \to B)\).
Proof
Since \(C' = D' \otimes_{B'} A'\) we have that \(IC'\) is the image of \(D' \otimes_{B'} I = C' \otimes_{A'} I \to C'\). As the ring map \(B' \to A'\) induces an isomorphism \(J \to I\) the lemma follows.
Lemma
Let \(A, A', B, B', C, C', D, D', I, M', M, N, \varphi\) be as in Lemma 08KM. If \(N\) finite over \(D\) and \(M'\) finite over \(C'\), then \(N' = N \times_{\varphi, M} M'\) is finite over \(D'\).
Proof
Recall that \(D' \to D \times_C C'\) is surjective by Lemma 08IG. Observe that \(N' = N \times_{\varphi, M} M'\) is a module over \(D \times_C C'\). We can apply Lemma 0D2H to the data \(C, C', D, D', IC', M', M, N, \varphi\) to see that \(N' = N \times_{\varphi, M} M'\) is finite over \(D \times_C C'\). Thus it is finite over \(D'\).
Lemma
With \(A, A', B, B', C, C', D, D', I\) as in Situation 08KK.
Let \((N, M', \varphi)\) be an object of \(\text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'}\). If \(M'\) is flat over \(A'\) and \(N\) is flat over \(B\), then \(N' = N \times_{\varphi, M} M'\) is flat over \(B'\).
If \(L'\) is a \(D'\)-module flat over \(B'\), then \(L' = (L \otimes_{D'} D) \times_{(L \otimes_{D'} C)} (L \otimes_{D'} C')\).
The category of \(D'\)-modules flat over \(B'\) is equivalent to the categories of objects \((N, M', \varphi)\) of \(\text{Mod}_D \times_{\text{Mod}_C} \text{Mod}_{C'}\) with \(N\) flat over \(B\) and \(M'\) flat over \(A'\).
Proof
Part (1) follows from part (1) of Lemma 0D2I.
Part (2) follows from part (2) of Lemma 0D2I using that \(L' \otimes_{D'} D = L' \otimes_{B'} B\), \(L' \otimes_{D'} C' = L' \otimes_{B'} A'\), and \(L' \otimes_{D'} C = L' \otimes_{B'} A\), see discussion in Situation 08KK.
Part (3) is an immediate consequence of (1) and (2).
The following lemma is a good deal more interesting than its counter part in the absolute case (Lemma 0D2J), although the proof is essentially the same.
Lemma
Let \(A, A', B, B', C, C', D, D', I, M', M, N, \varphi\) be as in Lemma 08KM. If
\(N\) is finitely presented over \(D\) and flat over \(B\),
\(M'\) finitely presented over \(C'\) and flat over \(A'\), and
the ring map \(B' \to D'\) factors as \(B' \to D'' \to D'\) with \(B' \to D''\) flat and \(D'' \to D'\) of finite presentation,
then \(N' = N \times_M M'\) is finitely presented over \(D'\).
Proof
Choose a surjection \(D''' = D''[x_1, \ldots, x_n] \to D'\) with finitely generated kernel \(J\). By Algebra, Lemma 0564 it suffices to show that \(N'\) is finitely presented as a \(D'''\)-module. Moreover, \(D''' \otimes_{B'} B \to D' \otimes_{B'} B = D\) and \(D''' \otimes_{B'} A' \to D' \otimes_{B'} A' = C'\) are surjections whose kernels are generated by the image of \(J\), hence \(N\) is a finitely presented \(D''' \otimes_{B'} B\)-module and \(M'\) is a finitely presented \(D''' \otimes_{B'} A'\)-module by Algebra, Lemma 0564 again. Thus we may replace \(D'\) by \(D'''\) and \(D\) by \(D''' \otimes_{B'} B\), etc. Since \(D'''\) is flat over \(B'\), it follows that we may assume that \(B' \to D'\) is flat.
Assume \(B' \to D'\) is flat. By Lemma 08IH the module \(N'\) is finite over \(D'\). Choose a surjection \((D')^{\oplus n} \to N'\) with kernel \(K'\). By base change we obtain maps \(D^{\oplus n} \to N\), \((C')^{\oplus n} \to M'\), and \(C^{\oplus n} \to M\) with kernels \(K_D\), \(K_{C'}\), and \(K_C\). There is a canonical map \[K' \longrightarrow K_D \times_{K_C} K_{C'}\] On the other hand, since \(N' = N \times_M M'\) and \(D' = D \times_C C'\) (by Lemma 0D2I; applied to the flat \(B'\)-module \(D'\)) there is also a canonical map \(K_D \times_{K_C} K_{C'} \to K'\) inverse to the displayed arrow. Hence the displayed map is an isomorphism. By Algebra, Lemma 0519 the modules \(K_D\) and \(K_{C'}\) are finite. We conclude from Lemma 08IH that \(K'\) is a finite \(D'\)-module provided that \(K_D \to K_C\) and \(K_{C'} \to K_C\) induce isomorphisms \(K_D \otimes_B A = K_C = K_{C'} \otimes_{A'} A\). This is true because the flatness assumptions implies the sequences \[0 \to K_D \to D^{\oplus n} \to N \to 0 \quad\text{and}\quad 0 \to K_{C'} \to (C')^{\oplus n} \to M' \to 0\] stay exact upon tensoring, see Algebra, Lemma 00HL.
Lemma
Let \(A, A', B, B', I\) be as in Situation 08KH. Let \((D, C', \varphi)\) be a system consisting of an \(B\)-algebra \(D\), a \(A'\)-algebra \(C'\) and an isomorphism \(D \otimes_B A \to C'/IC' = C\). Set \(D' = D \times_C C'\) (as in Lemma 07RU). Then
\(B' \to D'\) is finite type if and only if \(B \to D\) and \(A' \to C'\) are finite type,
\(B' \to D'\) is flat if and only if \(B \to D\) and \(A' \to C'\) are flat,
\(B' \to D'\) is flat and of finite presentation if and only if \(B \to D\) and \(A' \to C'\) are flat and of finite presentation,
\(B' \to D'\) is smooth if and only if \(B \to D\) and \(A' \to C'\) are smooth,
\(B' \to D'\) is étale if and only if \(B \to D\) and \(A' \to C'\) are étale.
Moreover, if \(D'\) is a flat \(B'\)-algebra, then \(D' \to (D' \otimes_{B'} B) \times_{(D' \otimes_{B'} A)} (D' \otimes_{B'} A')\) is an isomorphism. In this way the category of flat \(B'\)-algebras is equivalent to the categories of systems \((D, C', \varphi)\) as above with \(D\) flat over \(B\) and \(C'\) flat over \(A'\).
Proof
The implication “\(\Rightarrow\)” follows from Algebra, Lemmas 05G5, 00HI, 00T4, and 00U2 because we have \(D' \otimes_{B'} B = D\) and \(D' \otimes_{B'} A' = C'\) by Lemma 07RU. Thus it suffices to prove the implications in the other direction.
Ad (1). Assume \(D\) of finite type over \(B\) and \(C'\) of finite type over \(A'\). We will use the results of Lemma 07RU without further mention. Choose generators \(x_1, \ldots, x_r\) of \(D\) over \(B\) and generators \(y_1, \ldots, y_s\) of \(C'\) over \(A'\). Using that \(D = D' \otimes_{B'} B\) and \(B' \to B\) is surjective we can find \(u_1, \ldots, u_r \in D'\) mapping to \(x_1, \ldots, x_r\) in \(D\). Using that \(C' = D' \otimes_{B'} A'\) we can find \(v_1, \ldots, v_t \in D'\) such that \(y_i = \sum v_j \otimes a'_{ij}\) for some \(a'_{ij} \in A'\). In particular, the images of \(v_j\) in \(C'\) generate \(C'\) as an \(A'\)-algebra. Set \(N = r + t\) and consider the cube of rings \[\xymatrix{ A[x_1, \ldots, x_N] & & A'[x_1, \ldots, x_N] \ar[ll] \\ & A \ar[lu] & & A' \ar[ll] \ar[lu] \\ B[x_1, \ldots, x_N] \ar[uu] & & B'[x_1, \ldots, x_N] \ar[uu] \ar[ll] \\ & B \ar[uu] \ar[lu] & & B' \ar[ll] \ar[uu] \ar[lu] }\] Observe that the back square is cartesian as well. Consider the ring map \[B'[x_1, \ldots, x_N] \to D',\quad x_i \mapsto u_i \quad\text{and}\quad x_{r + j} \mapsto v_j.\] Then we see that the induced maps \(B[x_1, \ldots, x_N] \to D\) and \(A'[x_1, \ldots, x_N] \to C'\) are surjective, in particular finite. We conclude from Lemma 08IH that \(B'[x_1, \ldots, x_N] \to D'\) is finite, which implies that \(D'\) is of finite type over \(B'\) for example by Algebra, Lemma 00F4.
Ad (2). The implication “\(\Leftarrow\)” follows from Lemma 07RW. Moreover, the final statement follows from the final statement of Lemma 07RW.
Ad (3). Assume \(B \to D\) and \(A' \to C'\) are flat and of finite presentation. The flatness of \(B' \to D'\) we’ve seen in (2). We know \(B' \to D'\) is of finite type by (1). Choose a surjection \(B'[x_1, \ldots, x_N] \to D'\). By Algebra, Lemma 00R2 the ring \(D\) is of finite presentation as a \(B[x_1, \ldots, x_N]\)-module and the ring \(C'\) is of finite presentation as a \(A'[x_1, \ldots, x_N]\)-module. By Lemma 08KP we see that \(D'\) is of finite presentation as a \(B'[x_1, \ldots, x_N]\)-module, i.e., \(B' \to D'\) is of finite presentation.
Ad (4). Assume \(B \to D\) and \(A' \to C'\) smooth. By (3) we see that \(B' \to D'\) is flat and of finite presentation. By Algebra, Lemma 00TF it suffices to check that \(D' \otimes_{B'} k\) is smooth for any field \(k\) over \(B'\). If the composition \(J \to B' \to k\) is zero, then \(B' \to k\) factors as \(B' \to B \to k\) and we see that \[D' \otimes_{B'} k = D' \otimes_{B'} B \otimes_B k = D \otimes_B k\] is smooth as \(B \to D\) is smooth. If the composition \(J \to B' \to k\) is nonzero, then there exists an \(h \in J\) which does not map to zero in \(k\). Then \(B' \to k\) factors as \(B' \to B'_h \to k\). Observe that \(h\) maps to zero in \(B\), hence \(B_h = 0\). Thus by Lemma 01Z8 we have \(B'_h = A'_h\) and we get \[D' \otimes_{B'} k = D' \otimes_{B'} B'_h \otimes_{B'_h} k = C'_h \otimes_{A'_h} k\] is smooth as \(A' \to C'\) is smooth.
Ad (5). Assume \(B \to D\) and \(A' \to C'\) are étale. By (4) we see that \(B' \to D'\) is smooth. As we can read off whether or not a smooth map is étale from the dimension of fibres we see that (5) holds (argue as in the proof of (4) to identify fibres – some details omitted).
Remark
In Situation 08KK. Assume \(B' \to D'\) is of finite presentation and suppose we are given a \(D'\)-module \(L'\). We claim there is a bijective correspondence between
surjections of \(D'\)-modules \(L' \to Q'\) with \(Q'\) of finite presentation over \(D'\) and flat over \(B'\), and
pairs of surjections of modules \((L' \otimes_{D'} D \to Q_1, L' \otimes_{D'} C' \to Q_2)\) with
\(Q_1\) of finite presentation over \(D\) and flat over \(B\),
\(Q_2\) of finite presentation over \(C'\) and flat over \(A'\),
\(Q_1 \otimes_D C = Q_2 \otimes_{C'} C\) as quotients of \(L' \otimes_{D'} C\).
The correspondence between these is given by \(Q \mapsto (Q_1, Q_2)\) with \(Q_1 = Q \otimes_{D'} D\) and \(Q_2 = Q \otimes_{D'} C'\). And for the converse we use \(Q = Q_1 \times_{Q_{12}} Q_2\) where \(Q_{12}\) the common quotient \(Q_1 \otimes_D C = Q_2 \otimes_{C'} C\) of \(L' \otimes_{D'} C\). As quotient map we use \[L' \longrightarrow (L' \otimes_{D'} D) \times_{(L' \otimes_{D'} C)} (L' \otimes_{D'} C') \longrightarrow Q_1 \times_{Q_{12}} Q_2 = Q\] where the first arrow is surjective by Lemma 08IG and the second by Lemma 08KJ. The claim follows by Lemmas 07RW and 08KP.
Fitting ideals
The Fitting ideals of a finite module are the ideals determined by the construction of Lemma 07Z8.
Lemma
Let \(R\) be a ring. Let \(A\) be an \(n \times m\) matrix with coefficients in \(R\). Let \(I_r(A)\) be the ideal generated by the \(r \times r\)-minors of \(A\) with the convention that \(I_0(A) = R\) and \(I_r(A) = 0\) if \(r > \min(n, m)\). Then
\(I_0(A) \supset I_1(A) \supset I_2(A) \supset \ldots\),
if \(B\) is an \((n + n') \times m\) matrix, and \(A\) is the first \(n\) rows of \(B\), then \(I_{r + n'}(B) \subset I_r(A)\),
if \(C\) is an \(n \times n\) matrix then \(I_r(CA) \subset I_r(A)\).
If \(A\) is a block matrix \[\left( \begin{matrix} A_1 & 0 \\ 0 & A_2 \end{matrix} \right)\] then \(I_r(A) = \sum_{r_1 + r_2 = r} I_{r_1}(A_1) I_{r_2}(A_2)\).
Add more here.
Proof
Omitted. (Hint: Use that a determinant can be computed by expanding along a column or a row.)
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Choose a presentation \[\bigoplus\nolimits_{j \in J} R \longrightarrow R^{\oplus n} \longrightarrow M \longrightarrow 0.\] of \(M\). Let \(A = (a_{ij})_{i = 1, \ldots, n, j \in J}\) be the matrix of the map \(\bigoplus_{j \in J} R \to R^{\oplus n}\). The ideal \(\text{Fit}_k(M)\) generated by the \((n - k) \times (n - k)\) minors of \(A\) is independent of the choice of the presentation.
Proof
Let \(K \subset R^{\oplus n}\) be the kernel of the surjection \(R^{\oplus n} \to M\). Pick \(z_1, \ldots, z_{n - k} \in K\) and write \(z_j = (z_{1j}, \ldots, z_{nj})\). Another description of the ideal \(\text{Fit}_k(M)\) is that it is the ideal generated by the \((n - k) \times (n - k)\) minors of all the matrices \((z_{ij})\) we obtain in this way.
Suppose we change the surjection into the surjection \(R^{\oplus n + n'} \to M\) with kernel \(K'\) where we use the original map on the first \(n\) standard basis elements of \(R^{\oplus n + n'}\) and \(0\) on the last \(n'\) basis vectors. Then the corresponding ideals are the same. Namely, if \(z_1, \ldots, z_{n - k} \in K\) as above, let \(z'_j = (z_{1j}, \ldots, z_{nj}, 0, \ldots, 0) \in K'\) for \(j = 1, \ldots, n - k\) and \(z'_{n + j'} = (0, \ldots, 0, 1, 0, \ldots, 0) \in K'\). Then we see that the ideal of \((n - k) \times (n - k)\) minors of \((z_{ij})\) agrees with the ideal of \((n + n' - k) \times (n + n' - k)\) minors of \((z'_{ij})\). This gives one of the inclusions. Conversely, given \(z'_1, \ldots, z'_{n + n' - k}\) in \(K'\) we can project these to \(R^{\oplus n}\) to get \(z_1, \ldots, z_{n + n' - k}\) in \(K\). By Lemma 07Z7 we see that the ideal generated by the \((n + n' - k) \times (n + n' - k)\) minors of \((z'_{ij})\) is contained in the ideal generated by the \((n - k) \times (n - k)\) minors of \((z_{ij})\). This gives the other inclusion.
Let \(R^{\oplus m} \to M\) be another surjection with kernel \(L\). By Schanuel’s lemma (Algebra, Lemma 00O3) and the results of the previous paragraph, we may assume \(m = n\) and that there is an isomorphism \(R^{\oplus n} \to R^{\oplus m}\) commuting with the surjections to \(M\). Let \(C = (c_{li})\) be the (invertible) matrix of this map (it is a square matrix as \(n = m\)). Then given \(z'_1, \ldots, z'_{n - k} \in L\) as above we can find \(z_1, \ldots, z_{n - k} \in K\) with \(z_1' = Cz_1, \ldots, z'_{n - k} = Cz_{n - k}\). By Lemma 07Z7 we get one of the inclusions. By symmetry we get the other.
Definition
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(k \geq 0\). The \(k\)th Fitting ideal of \(M\) is the ideal \(\text{Fit}_k(M)\) constructed in Lemma 07Z8. Set \(\text{Fit}_{-1}(M) = 0\).
Since the Fitting ideals are the ideals of minors of a big matrix (numbered in reverse ordering from the ordering in Lemma 07Z7) we see that \[0 = \text{Fit}_{-1}(M) \subset \text{Fit}_0(M) \subset \text{Fit}_1(M) \subset \ldots \subset \text{Fit}_t(M) = R\] for some \(t \gg 0\). Here are some basic properties of Fitting ideals.
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module.
If \(M\) can be generated by \(n\) elements, then \(\text{Fit}_n(M) = R\).
Given a second finite \(R\)-module \(M'\) we have \(\text{Fit}_0(M \oplus M') = \text{Fit}_0(M)\text{Fit}_0(M')\) and more generally \[\text{Fit}_l(M \oplus M') = \sum\nolimits_{k + k' = l} \text{Fit}_k(M)\text{Fit}_{k'}(M')\]
If \(R \to R'\) is a ring map, then \(\text{Fit}_k(M \otimes_R R')\) is the ideal of \(R'\) generated by the image of \(\text{Fit}_k(M)\).
If \(M\) is of finite presentation, then \(\text{Fit}_k(M)\) is a finitely generated ideal.
If \(M \to M'\) is a surjection, then \(\text{Fit}_k(M) \subset \text{Fit}_k(M')\).
We have \(\text{Fit}_0(M) \subset \text{Ann}_R(M)\).
We have \(V(\text{Fit}_0(M)) = \text{Supp}(M)\).
If \(I\) is an ideal of \(R\), then \(\text{Fit}_0(R/I) = I\).
Add more here.
Proof
Part (1) follows from the fact that \(I_0(A) = R\) in Lemma 07Z7.
Part (2) follows form the corresponding statement in Lemma 07Z7.
Part (3) follows from the fact that \(\otimes_R R'\) is right exact, so the base change of a presentation of \(M\) is a presentation of \(M \otimes_R R'\).
Proof of (4). Let \(R^{\oplus m} \xrightarrow{A} R^{\oplus n} \to M \to 0\) be a presentation. Then \(\text{Fit}_k(M)\) is the ideal generated by the \((n - k) \times (n - k)\) minors of the matrix \(A\).
Part (5) is immediate from the definition.
Proof of (6). Choose a presentation of \(M\) with matrix \(A\) as in Lemma 07Z8. Let \(J' \subset J\) be a subset of cardinality \(n\). It suffices to show that \(f = \det(a_{ij})_{i = 1, \ldots, n, j \in J'}\) annihilates \(M\). This is clear because the cokernel of \[R^{\oplus n} \xrightarrow{A' = (a_{ij})_{i = 1, \ldots, n, j \in J'}} R^{\oplus n} \to M \to 0\] is killed by \(f\) as there is a matrix \(B\) with \(A' B = f1_{n \times n}\).
Proof of (7). Choose a presentation of \(M\) with matrix \(A\) as in Lemma 07Z8. By Nakayama’s lemma (Algebra, Lemma 00DV) we have \[M_\mathfrak p \not = 0 \Leftrightarrow M \otimes_R \kappa(\mathfrak p) \not = 0 \Leftrightarrow \text{rank}(\text{image }A\text{ in }\kappa(\mathfrak p)) < n\] Clearly \(\text{Fit}_0(M)\) exactly cuts out the set of primes with this property.
Example
Let \(R\) be a ring. The Fitting ideals of the finite free module \(M = R^{\oplus n}\) are \(\text{Fit}_k(M) = 0\) for \(k < n\) and \(\text{Fit}_k(M) = R\) for \(k \geq n\).
Example
Let \(R\) be a ring and let \(I, J\subset R\) be ideals. Then \(\text{Fit}_0(R/I) = I\), \(\text{Fit}_0(R/I \oplus R/J) = IJ\), \(\text{Fit}_1(R/I\oplus R/J) = I+J\).
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(k \geq 0\). Let \(\mathfrak p \subset R\) be a prime ideal. The following are equivalent
\(\text{Fit}_k(M) \not \subset \mathfrak p\),
\(\dim_{\kappa(\mathfrak p)} M \otimes_R \kappa(\mathfrak p) \leq k\),
\(M_\mathfrak p\) can be generated by \(k\) elements over \(R_\mathfrak p\), and
\(M_f\) can be generated by \(k\) elements over \(R_f\) for some \(f \in R\), \(f \not \in \mathfrak p\).
Proof
By Nakayama’s lemma (Algebra, Lemma 00DV) we see that \(M_f\) can be generated by \(k\) elements over \(R_f\) for some \(f \in R\), \(f \not \in \mathfrak p\) if \(M \otimes_R \kappa(\mathfrak p)\) can be generated by \(k\) elements. Hence (2), (3), and (4) are equivalent. Using Lemma 07ZA part (3) this reduces the problem to the case where \(R\) is a field and \(\mathfrak p = (0)\). In this case the result follows from Example 07ZB.
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(r \geq 0\). The following are equivalent
\(M\) is finite locally free of rank \(r\) (Algebra, Definition 00NW),
\(\text{Fit}_{r - 1}(M) = 0\) and \(\text{Fit}_r(M) = R\), and
\(\text{Fit}_k(M) = 0\) for \(k < r\) and \(\text{Fit}_k(M) = R\) for \(k \geq r\).
Proof
It is immediate that (2) is equivalent to (3) because the Fitting ideals form an increasing sequence of ideals. Since the formation of \(\text{Fit}_k(M)\) commutes with base change (Lemma 07ZA) we see that (1) implies (2) by Example 07ZB and glueing results (Algebra, Section 00EN). Conversely, assume (2). By Lemma 07ZC we may assume that \(M\) is generated by \(r\) elements. Thus a presentation \(\bigoplus_{j \in J} R \to R^{\oplus r} \to M \to 0\). But now the assumption that \(\text{Fit}_{r - 1}(M) = 0\) implies that all entries of the matrix of the map \(\bigoplus_{j \in J} R \to R^{\oplus r}\) are zero. Thus \(M\) is free.
Lemma
Let \(R\) be a local ring. Let \(M\) be a finite \(R\)-module. Let \(k \geq 0\). Assume that \(\text{Fit}_k(M) = (f)\) for some \(f \in R\). Let \(M'\) be the quotient of \(M\) by \(\{x \in M \mid fx = 0\}\). Then \(M'\) can be generated by \(k\) elements.
Proof
Choose generators \(x_1, \ldots, x_n \in M\) corresponding to the surjection \(R^{\oplus n} \to M\). Since \(R\) is local if a set of elements \(E \subset (f)\) generates \((f)\), then some \(e \in E\) generates \((f)\), see Algebra, Lemma 00DV. Hence we may pick \(z_1, \ldots, z_{n - k}\) in the kernel of \(R^{\oplus n} \to M\) such that some \((n - k) \times (n - k)\) minor of the \(n \times (n - k)\) matrix \(A = (z_{ij})\) generates \((f)\). After renumbering the \(x_i\) we may assume the first minor \(\det(z_{ij})_{1 \leq i, j \leq n - k}\) generates \((f)\), i.e., \(\det(z_{ij})_{1 \leq i, j \leq n - k} = uf\) for some unit \(u \in R\). Every other minor is a multiple of \(f\). By Algebra, Lemma 080R there exists a \(n - k \times n - k\) matrix \(B\) such that \[AB = f \left( \begin{matrix} u 1_{n - k \times n - k} \\ C \end{matrix} \right)\] for some matrix \(C\) with coefficients in \(R\). This implies that for every \(i \leq n - k\) the element \(y_i = ux_i + \sum_j c_{ji}x_j\) is annihilated by \(f\). Since \(M/\sum Ry_i\) is generated by the images of \(x_{n - k + 1}, \ldots, x_n\) we win.
Lemma
Let \(R\) be a ring. Let \(M\) be a finitely presented \(R\)-module. Let \(k \geq 0\). Assume that \(\text{Fit}_k(M) = (f)\) for some nonzerodivisor \(f \in R\) and \(\text{Fit}_{k - 1}(M) = 0\). Then
\(M\) has projective dimension \(\leq 1\),
\(M' = \Ker(f : M \to M)\) is the \(f\)-power torsion submodule of \(M\),
\(M'\) has projective dimension \(\leq 1\),
\(M/M'\) is finite locally free of rank \(k\), and
\(M \cong M/M' \oplus M'\).
Proof
Choose a presentation \[R^{\oplus m} \xrightarrow{A} R^{\oplus n} \to M \to 0\] for some matrix \(A\) with coefficients in \(R\).
We first prove the lemma when \(R\) is local. Set \(M' = \{x \in M \mid fx = 0\}\) as in the statement. By Lemma 080Z we can choose \(x_1, \ldots, x_k \in M\) which generate \(M/M'\). Then \(x_1, \ldots, x_k\) generate \(M_f = (M/M')_f\). Hence, if there is a relation \(\sum a_ix_i = 0\) in \(M\), then we see that \(a_1, \ldots, a_k\) map to zero in \(R_f\) since otherwise \(\text{Fit}_{k - 1}(M) R_f = \text{Fit}_{k - 1}(M_f)\) would be nonzero. Since \(f\) is a nonzerodivisor, we conclude \(a_1 = \ldots = a_k = 0\). Thus \(M \cong R^{\oplus k} \oplus M'\). After a change of basis in our presentation above, we may assume the first \(n - k\) basis vectors of \(R^{\oplus n}\) map into the summand \(M'\) of \(M\) and the last \(k\)-basis vectors of \(R^{\oplus n}\) map to basis elements of the summand \(R^{\oplus k}\) of \(M\). Having done so, the last \(k\) rows of the matrix \(A\) vanish. In this way we see that, replacing \(M\) by \(M'\), \(k\) by \(0\), \(n\) by \(n - k\), and \(A\) by the submatrix where we delete the last \(k\) rows, we reduce to the case discussed in the next paragraph.
Assume \(R\) is local, \(k = 0\), and \(M\) annihilated by \(f\). Now the \(0\)th Fitting ideal of \(M\) is \((f)\) and is generated by the \(n \times n\) minors of the matrix \(A\) of size \(n \times m\). (This in particular implies \(m \geq n\).) Since \(R\) is local, some \(n \times n\) minor of \(A\) is \(uf\) for a unit \(u \in R\). After renumbering we may assume this minor is the first one. Moreover, we know all other \(n \times n\) minors of \(A\) are divisible by \(f\). Write \(A = (A_1 A_2)\) in block form where \(A_1\) is an \(n \times n\) matrix and \(A_2\) is an \(n \times (m - n)\) matrix. By Algebra, Lemma 080R applied to the transpose of \(A\) (!) we find there exists an \(n \times n\) matrix \(B\) such that \[BA = B(A_1 A_2) = f \left( \begin{matrix} u 1_{n \times n} & C \end{matrix} \right)\] for some \(n \times (m - n)\) matrix \(C\) with coefficients in \(R\). Then we first conclude \(BA_1 = fu 1_{n \times n}\). Thus \[BA_2 = fC = u^{-1}fuC = u^{-1}BA_1C\] Since the determinant of \(B\) is a nonzerodivisor we conclude that \(A_2 = u^{-1}A_1C\). Therefore the image of \(A\) is equal to the image of \(A_1\) which is isomorphic to \(R^{\oplus n}\) because the determinant of \(A_1\) is a nonzerodivisor. Hence \(M\) has projective dimension \(\leq 1\).
We return to the case of a general ring \(R\). By the local case we see that \(M/M'\) is a finite locally free module of rank \(k\), see Algebra, Lemma 00NX. Hence the extension \(0 \to M' \to M \to M/M' \to 0\) splits. It follows that \(M'\) is a finitely presented module. Choose a short exact sequence \(0 \to K \to R^{\oplus a} \to M' \to 0\). Then \(K\) is a finite \(R\)-module, see Algebra, Lemma 0519. By the local case we see that \(K_\mathfrak p \cong R_\mathfrak p^{\oplus a}\) for all primes. Hence by Algebra, Lemma 00NX again we see that \(K\) is finite locally free of rank \(a\). It follows that \(M'\) has projective dimension \(\leq 1\) and the lemma is proved.
Lifting
In this section we collection some lemmas concerning lifting statements of the following kind: If \(A\) is a ring and \(I \subset A\) is an ideal, and \(\overline{\xi}\) is some kind of structure over \(A/I\), then we can lift \(\overline{\xi}\) to a similar kind of structure \(\xi\) over \(A\) or over some étale extension of \(A\). Here are some types of structure for which we have already proved some results:
finite stably free modules, see Lemma 0BC5,
ring maps, i.e., proving certain algebras are formally smooth, see Algebra, Lemma 00TK, Proposition 00TN, and Lemma 07K4,
syntomic ring maps, see Algebra, Lemma 00T0,
smooth ring maps, see Algebra, Lemma 04B1,
étale ring maps, see Algebra, Lemma 04D1,
factoring polynomials, see Algebra, Lemma 00UH, and
Algebra, Section 04GE discusses henselian local rings.
The interested reader will find more results of this nature in Smoothing Ring Maps, Section 07CD in particular Smoothing Ring Maps, Proposition 07M8.
Let \(A\) be a ring and let \(I \subset A\) be an ideal. Let \(\overline{\xi}\) be some kind of structure over \(A/I\). In the following lemmas we look for étale ring maps \(A \to A'\) which induce isomorphisms \(A/I \to A'/IA'\) and objects \(\xi'\) over \(A'\) lifting \(\overline{\xi}\). A general remark is that given étale ring maps \(A \to A' \to A''\) such that \(A/I \cong A'/IA'\) and \(A'/IA' \cong A''/IA''\) the composition \(A \to A''\) is also étale (Algebra, Lemma 00U2) and also satisfies \(A/I \cong A''/IA''\). We will frequently use this in the following lemmas without further mention. Here is a trivial example of the type of result we are looking for.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal, let \(\overline{u} \in A/I\) be an invertible element. There exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and an invertible element \(u' \in A'\) lifting \(\overline{u}\).
Proof
Choose any lift \(f \in A\) of \(\overline{u}\) and set \(A' = A_f\) and \(u\) the image of \(f\) in \(A'\).
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal, let \(\overline{e} \in A/I\) be an idempotent. There exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and an idempotent \(e' \in A'\) lifting \(\overline{e}\).
Proof
Choose any lift \(x \in A\) of \(\overline{e}\). Set \[A' = A[t]/(t^2 - t)\left[\frac{1}{t - 1 + x}\right].\] The ring map \(A \to A'\) is étale because \((2t - 1)\text{d}t = 0\) and \((2t - 1)(2t - 1) = 1\) which is invertible. We have \(A'/IA' = A/I[t]/(t^2 - t)[\frac{1}{t - 1 + \overline{e}}] \cong A/I\) the last map sending \(t\) to \(\overline{e}\) which works as \(\overline{e}\) is a root of \(t^2 - t\). This also shows that setting \(e'\) equal to the class of \(t\) in \(A'\) works.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Let \(\Spec(A/I) = \coprod_{j \in J} \overline{U}_j\) be a finite disjoint open covering. Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and a finite disjoint open covering \(\Spec(A') = \coprod_{j \in J} U'_j\) lifting the given covering.
Proof
This follows from Lemma 07LY and the fact that open and closed subsets of Spectra correspond to idempotents, see Algebra, Lemma 00EE.
Lemma
Let \(A \to B\) be a ring map and \(J \subset B\) an ideal. If \(A \to B\) is étale at every prime of \(V(J)\), then there exists a \(g \in B\) mapping to an invertible element of \(B/J\) such that \(A' = B_g\) is étale over \(A\).
Proof
The set of points of \(\Spec(B)\) where \(A \to B\) is not étale is a closed subset of \(\Spec(B)\), see Algebra, Definition 00U1. Write this as \(V(J')\) for some ideal \(J' \subset B\). Then \(V(J') \cap V(J) = \emptyset\) hence \(J + J' = B\) by Algebra, Lemma 00E0. Write \(1 = f + g\) with \(f \in J\) and \(g \in J'\). Then \(g\) works.
Next we have three lemmas saying we can lift factorizations of polynomials.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Let \(f \in A[x]\) be a monic polynomial. Let \(\overline{f} = \overline{g} \overline{h}\) be a factorization of \(f\) in \(A/I[x]\) such that \(\overline{g}\) and \(\overline{h}\) are monic and generate the unit ideal in \(A/I[x]\). Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and a factorization \(f = g' h'\) in \(A'[x]\) with \(g'\), \(h'\) monic lifting the given factorization over \(A/I\).
Proof
We will deduce this from results on the universal factorization proved earlier; however, we encourage the reader to find their own proof not using this trick. Say \(\deg(\overline{g}) = n\) and \(\deg(\overline{h}) = m\) so that \(\deg(f) = n + m\). Write \(f = x^{n + m} + \sum \alpha_i x^{n + m - i}\) for some \(\alpha_1, \ldots, \alpha_{n + m} \in A\). Consider the ring map \[R = \mathbf{Z}[a_1, \ldots, a_{n + m}] \longrightarrow S = \mathbf{Z}[b_1, \ldots, b_n, c_1, \ldots, c_m]\] of Algebra, Example 00UA. Let \(R \to A\) be the ring map which sends \(a_i\) to \(\alpha_i\). Set \[B = A \otimes_R S\] By construction the image \(f_B\) of \(f\) in \(B[x]\) factors, say \(f_B = g_B h_B\) with \(g_B = x^n + \sum (1 \otimes b_i) x^{n - i}\) and similarly for \(h_B\). Write \(\overline{g} = x^n + \sum \overline{\beta}_i x^{n - i}\) and \(\overline{h} = x^m + \sum \overline{\gamma}_i x^{m - i}\). The \(A\)-algebra map \[B \longrightarrow A/I, \quad 1 \otimes b_i \mapsto \overline{\beta}_i, \quad 1 \otimes c_i \mapsto \overline{\gamma}_i\] maps \(g_B\) and \(h_B\) to \(\overline{g}\) and \(\overline{h}\) in \(A/I[x]\). The displayed map is surjective; denote \(J \subset B\) its kernel. From the discussion in Algebra, Example 00UA it is clear that \(A \to B\) is etale at all points of \(V(J) \subset \Spec(B)\). Choose \(g \in B\) as in Lemma 07M0 and consider the \(A\)-algebra \(B_g\). Since \(g\) maps to a unit in \(B/J = A/I\) we obtain also a map \(B_g/I B_g \to A/I\) of \(A/I\)-algebras. Since \(A/I \to B_g/I B_g\) is étale, also \(B_g/IB_g \to A/I\) is étale (Algebra, Lemma 00U7). Hence there exists an idempotent \(e \in B_g/I B_g\) such that \(A/I = (B_g/I B_g)_e\) (Algebra, Lemma 00U8). Choose a lift \(h \in B_g\) of \(e\). Then \(A \to A' = (B_g)_h\) with factorization given by the image of the factorization \(f_B = g_B h_B\) in \(A'\) is a solution to the problem posed by the lemma.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Let \(f \in A[x]\) be a monic polynomial. Let \(\overline{f} = \overline{g} \overline{h}\) be a factorization of \(f\) in \(A/I[x]\) and assume
the leading coefficient of \(\overline{g}\) is an invertible element of \(A/I\), and
\(\overline{g}\), \(\overline{h}\) generate the unit ideal in \(A/I[x]\).
Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and a factorization \(f = g' h'\) in \(A'[x]\) lifting the given factorization over \(A/I\).
Proof
Applying Lemma 07LX we may assume that the leading coefficient of \(\overline{g}\) is the reduction of an invertible element \(u \in A\). Then we may replace \(\overline{g}\) by \(\overline{u}^{-1}\overline{g}\) and \(\overline{h}\) by \(\overline{u}\overline{h}\). Thus we may assume that \(\overline{g}\) is monic. Since \(f\) is monic we conclude that \(\overline{h}\) is monic too. In this case the result follows from Lemma 0ALH.
Example
This example shows that one cannot drop the condition on the leading coefficient in Lemma 07M1. Namely, let \(A = \mathbf{Z}\), \(I = 4\mathbf{Z}\), \(f = 1\) and \(\bar{g} = \bar{h} = 2x^2 + 2x + 1\) in \(A/I[x]\). If \(A \to A'\) is étale and \(A'/I A' = A/I\), then the \(2\)-adic completion of \(A'\) is isomorphic to \(\mathbf{Z}_2\). Now any polynomial \(g'\) in \(\mathbf{Z}_2[x]\) congruent to \(2x^2 + 2x + 1\) mod \(4\) will not have a polynomial multiplicative inverse in \(\mathbf{Z}_2[x]\), whence the conclusion of the lemma does not hold.
Lemma
Let \(R \to S\) be a ring map. Let \(I \subset R\) be an ideal of \(R\) and let \(J \subset S\) be an ideal of \(S\). If the closure of the image of \(V(J)\) in \(\Spec(R)\) is disjoint from \(V(I)\), then there exists an element \(f \in R\) which maps to \(1\) in \(R/I\) and to an element of \(J\) in \(S\).
Proof
Let \(I' \subset R\) be an ideal such that \(V(I')\) is the closure of the image of \(V(J)\). Then \(V(I) \cap V(I') = \emptyset\) by assumption and hence \(I + I' = R\) by Algebra, Lemma 00E0. Write \(1 = g + f\) with \(g \in I\) and \(f \in I'\). We have \(V(f') \supset V(J)\) where \(f'\) is the image of \(f\) in \(S\). Hence \((f')^n \in J\) for some \(n\), see Algebra, Lemma 00E0. Replacing \(f\) by \(f^n\) we win.
Lemma
Let \(I\) be an ideal of a ring \(A\). Let \(A \to B\) be an integral ring map. Let \(b \in B\) map to an idempotent in \(B/IB\). Then there exists a monic \(f \in A[x]\) with \(f(b) = 0\) and \(f \bmod I = x^d(x - 1)^d\) for some \(d \geq 1\).
Proof
Observe that \(z = b^2 - b\) is an element of \(IB\). By Algebra, Lemma 00H5 there exist a monic polynomial \(g(x) = x^d + \sum a_j x^j\) of degree \(d\) with \(a_j \in I\) such that \(g(z) = 0\) in \(B\). Hence \(f(x) = g(x^2 - x) \in A[x]\) is a monic polynomial such that \(f(x) \equiv x^d(x - 1)^d \bmod I\) and such that \(f(b) = 0\) in \(B\).
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Let \(A \to B\) be an integral ring map. Let \(\overline{e} \in B/IB\) be an idempotent. Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and an idempotent \(e' \in B \otimes_A A'\) lifting \(\overline{e}\).
Proof
Choose an element \(y \in B\) lifting \(\overline{e}\). Choose \(f \in A[x]\) as in Lemma 09XG for \(y\). By Lemma 07M1 we can find an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and such that \(f = gh\) in \(A[x]\) with \(g(x) = x^d \bmod IA'\) and \(h(x) = (x - 1)^d \bmod IA'\). After replacing \(A\) by \(A'\) we may assume that the factorization is defined over \(A\). In that case we see that \(b_1 = g(y) \in B\) is a lift of \(\overline{e}^d = \overline{e}\) and \(b_2 = h(y) \in B\) is a lift of \((\overline{e} - 1)^d = (-1)^d (1 - \overline{e})^d = (-1)^d(1 - \overline{e})\) and moreover \(b_1b_2 = 0\). Thus \((b_1, b_2)B/IB = B/IB\) and \(V(b_1, b_2) \subset \Spec(B)\) is disjoint from \(V(IB)\). Since \(\Spec(B) \to \Spec(A)\) is closed (see Algebra, Lemmas 00GU and 00HZ) we can find an \(a \in A\) which maps to an invertible element of \(A/I\) whose image in \(B\) lies in \((b_1, b_2)\), see Lemma 07M3. After replacing \(A\) by the localization \(A_a\) we get that \((b_1, b_2) = B\). Then \(\Spec(B) = D(b_1) \amalg D(b_2)\); disjoint union because \(b_1b_2 = 0\) and covers \(\Spec(B)\) because \((b_1, b_2) = B\). Let \(e \in B\) be the idempotent corresponding to the open and closed subset \(D(b_1)\), see Algebra, Lemma 00EE. Since \(b_1\) is a lift of \(\overline{e}\) and \(b_2\) is a lift of \(\pm (1 - \overline{e})\) we conclude that \(e\) is a lift of \(\overline{e}\) by the uniqueness statement in Algebra, Lemma 00EE.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Let \(\overline{P}\) be a finite projective \(A/I\)-module. Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and a finite projective \(A'\)-module \(P'\) lifting \(\overline{P}\).
Proof
We can choose an integer \(n\) and a direct sum decomposition \((A/I)^{\oplus n} = \overline{P} \oplus \overline{K}\) for some \(R/I\)-module \(\overline{K}\). Choose a lift \(\varphi : A^{\oplus n} \to A^{\oplus n}\) of the projector \(\overline{p}\) associated to the direct summand \(\overline{P}\). Let \(f \in A[x]\) be the characteristic polynomial of \(\varphi\). Set \(B = A[x]/(f)\). By Cayley-Hamilton (Algebra, Lemma 00DX) there is a map \(B \to \text{End}_A(A^{\oplus n})\) mapping \(x\) to \(\varphi\). For every prime \(\mathfrak p \supset I\) the image of \(f\) in \(\kappa(\mathfrak p)\) is \((x - 1)^rx^{n - r}\) where \(r\) is the dimension of \(\overline{P} \otimes_{A/I} \kappa(\mathfrak p)\). Hence \((x - 1)^nx^n\) maps to zero in \(B \otimes_A \kappa(\mathfrak p)\) for all \(\mathfrak p \supset I\). Thus \(x(1 - x)\) is contained in every prime ideal of \(B/IB\). Hence \(x^N(1 - x)^N\) is contained in \(IB\) for some \(N \geq 1\). It follows that \(x^N + (1 - x)^N\) is a unit in \(B/IB\) and that \[\overline{e} = \text{image of }\frac{x^N}{x^N + (1 - x)^N}\text{ in }B/IB\] is an idempotent as both assertions hold in \(\mathbf{Z}[x]/(x^N(x - 1)^N)\). The image of \(\overline{e}\) in \(\text{End}_{A/I}((A/I)^{\oplus n})\) is \[\frac{\overline{p}^N}{\overline{p}^N + (1 - \overline{p})^N} = \overline{p}\] as \(\overline{p}\) is an idempotent. After replacing \(A\) by an étale extension \(A'\) as in the lemma, we may assume there exists an idempotent \(e \in B\) which maps to \(\overline{e}\) in \(B/IB\), see Lemma 07M4. Then the image of \(e\) under the map \[B = A[x]/(f) \longrightarrow \text{End}_A(A^{\oplus n}).\] is an idempotent element \(p\) which lifts \(\overline{p}\). Setting \(P = \Im(p)\) we win.
Lemma
Let \(A\) be a ring. Let \(0 \to K \to A^{\oplus m} \to M \to 0\) be a sequence of \(A\)-modules. Consider the \(A\)-algebra \(C = \text{Sym}^*_A(M)\) with its presentation \(\alpha : A[y_1, \ldots, y_m] \to C\) coming from the surjection \(A^{\oplus m} \to M\). Then \[\NL(\alpha) = (K \otimes_A C \to \bigoplus\nolimits_{j = 1, \ldots, m} C \text{d}y_j)\] (see Algebra, Section 00S0) in particular \(\Omega_{C/A} = M \otimes_A C\).
Proof
Let \(J = \Ker(\alpha)\). The lemma asserts that \(J/J^2 \cong K \otimes_A C\). Note that \(\alpha\) is a homomorphism of graded algebras. We will prove that in degree \(d\) we have \((J/J^2)_d = K \otimes_A C_{d - 1}\). Note that \[J_d = \Ker(\text{Sym}^d_A(A^{\oplus m}) \to \text{Sym}^d_A(M)) = \Im(K \otimes_A \text{Sym}^{d - 1}_A(A^{\oplus m}) \to \text{Sym}^d_A(A^{\oplus m})),\] see Algebra, Lemma 00DO. It follows that \((J^2)_d = \sum_{a + b = d} J_a \cdot J_b\) is the image of \[K \otimes_A K \otimes_A \text{Sym}^{d - 2}_A(A^{\otimes m}) \to \text{Sym}^d_A(A^{\oplus m}).\] The cokernel of the map \(K \otimes_A \text{Sym}^{d - 2}_A(A^{\otimes m}) \to \text{Sym}^{d - 1}_A(A^{\oplus m})\) is \(\text{Sym}^{d - 1}_A(M)\) by the lemma referenced above. Hence it is clear that \((J/J^2)_d = J_d/(J^2)_d\) is equal to \[\begin{align*} \Coker( K \otimes_A K \otimes_A \text{Sym}^{d - 2}_A(A^{\otimes m}) \to K \otimes_A \text{Sym}^{d - 1}_A(A^{\otimes m})) & = K \otimes_A \text{Sym}^{d - 1}_A(M) \\ & = K \otimes_A C_{d -1} \end{align*}\] as desired.
Lemma
Let \(A\) be a ring. Let \(M\) be an \(A\)-module. Then \(C = \text{Sym}_A^*(M)\) is smooth over \(A\) if and only if \(M\) is a finite projective \(A\)-module.
Proof
Let \(\sigma : C \to A\) be the projection onto the degree \(0\) part of \(C\). Then \(J = \Ker(\sigma)\) is the part of degree \(> 0\) and we see that \(J/J^2 = M\) as an \(A\)-module. Hence if \(A \to C\) is smooth then \(M\) is a finite projective \(A\)-module by Algebra, Lemma 05D5.
Conversely, assume that \(M\) is finite projective and choose a surjection \(A^{\oplus n} \to M\) with kernel \(K\). Of course the sequence \(0 \to K \to A^{\oplus n} \to M \to 0\) is split as \(M\) is projective. In particular we see that \(K\) is a finite \(A\)-module and hence \(C\) is of finite presentation over \(A\) as \(C\) is a quotient of \(A[x_1, \ldots, x_n]\) by the ideal generated by \(K \subset \bigoplus Ax_i\). The computation of Lemma 07EV shows that \(\NL_{C/A}\) is homotopy equivalent to \((K \to M) \otimes_A C\). Hence \(\NL_{C/A}\) is quasi-isomorphic to \(C \otimes_A M\) placed in degree \(0\) which means that \(C\) is smooth over \(A\) by Algebra, Definition 00T2.
Lemma
Let \(A\) be a ring, let \(I \subset A\) be an ideal. Consider a commutative diagram \[\xymatrix{ B \ar[rd] \\ A \ar[u] \ar[r] & A/I }\] where \(B\) is a smooth \(A\)-algebra. Then there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and an \(A\)-algebra map \(B \to A'\) lifting the ring map \(B \to A/I\).
Proof
Let \(J \subset B\) be the kernel of \(B \to A/I\) so that \(B/J = A/I\). By Algebra, Lemma 06A9 the sequence \[0 \to I/I^2 \to J/J^2 \to \Omega_{B/A} \otimes_B B/J \to 0\] is split exact. Thus \(\overline{P} = J/(J^2 + IB) = \Omega_{B/A} \otimes_B B/J\) is a finite projective \(A/I\)-module. Choose an integer \(n\) and a direct sum decomposition \(A/I^{\oplus n} = \overline{P} \oplus \overline{K}\). By Lemma 07M5 we can find an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) and a finite projective \(A\)-module \(K\) which lifts \(\overline{K}\). We may and do replace \(A\) by \(A'\). Set \(B' = B \otimes_A \text{Sym}_A^*(K)\). Since \(A \to \text{Sym}_A^*(K)\) is smooth by Lemma 07M6 we see that \(B \to B'\) is smooth which in turn implies that \(A \to B'\) is smooth (see Algebra, Lemmas 00T4 and 00TC). Moreover the section \(\text{Sym}^*_A(K) \to A\) determines a section \(B' \to B\) and we let \(B' \to A/I\) be the composition \(B' \to B \to A/I\). Let \(J' \subset B'\) be the kernel of \(B' \to A/I\). We have \(JB' \subset J'\) and \(B \otimes_A K \subset J'\). These maps combine to give an isomorphism \[(A/I)^{\oplus n} \cong J/(J^2 + IB) \oplus \overline{K} \longrightarrow J'/((J')^2 + IB')\] Thus, after replacing \(B\) by \(B'\) we may assume that \(J/(J^2 + IB) = \Omega_{B/A} \otimes_B B/J\) is a free \(A/I\)-module of rank \(n\).
In this case, choose \(f_1, \ldots, f_n \in J\) which map to a basis of \(J/(J^2 + IB)\). Consider the finitely presented \(A\)-algebra \(C = B/(f_1, \ldots, f_n)\). Note that we have an exact sequence \[0 \to H_1(L_{C/A}) \to (f_1, \ldots, f_n)/(f_1, \ldots, f_n)^2 \to \Omega_{B/A} \otimes_B C \to \Omega_{C/A} \to 0\] see Algebra, Lemma 00S2 (note that \(H_1(L_{B/A}) = 0\) and that \(\Omega_{B/A}\) is finite projective, in particular flat so the Tor group vanishes). For any prime \(\mathfrak q \supset J\) of \(B\) the module \(\Omega_{B/A, \mathfrak q}\) is free of rank \(n\) because \(\Omega_{B/A}\) is finite projective and because \(\Omega_{B/A} \otimes_B B/J\) is free of rank \(n\) (see Algebra, Lemma 00NX). By our choice of \(f_1, \ldots, f_n\) the map \[\left((f_1, \ldots, f_n)/(f_1, \ldots, f_n)^2\right)_{\mathfrak q} \to \Omega_{B/A, \mathfrak q}\] is surjective modulo \(J\). Hence we see that this map of modules over the local ring \(C_{\mathfrak q}\) has to be an isomorphism (by Algebra, Lemma 00DV the map is surjective and for example by Algebra, Lemma 05G8 because \(((f_1, \ldots, f_n)/(f_1, \ldots, f_n)^2)_{\mathfrak q}\) is generated by \(n\) elements the map is injective). Thus \(H_1(L_{C/A})_{\mathfrak q} = 0\) and \(\Omega_{C/A, \mathfrak q} = 0\). By Algebra, Lemma 07BU we see that \(A \to C\) is smooth at the prime \(\overline{\mathfrak q}\) of \(C\) corresponding to \(\mathfrak q\). Since \(\Omega_{C/A, \mathfrak q} = 0\) it is actually étale at \(\overline{\mathfrak q}\). Thus \(A \to C\) is étale at all primes of \(C\) containing \(JC\). By Lemma 07M0 we can find an \(f \in C\) mapping to an invertible element of \(C/JC\) such that \(A \to C_f\) is étale. By our choice of \(f\) it is still true that \(C_f/JC_f = A/I\). The map \(C_f/IC_f \to A/I\) is surjective and étale by Algebra, Lemma 00U7. Hence \(A/I\) is isomorphic to the localization of \(C_f/IC_f\) at some element \(g \in C\), see Algebra, Lemma 00U8. Set \(A' = C_{fg}\) to conclude the proof.
Zariski pairs
In this section and the next a pair is a pair \((A, I)\) where \(A\) is a ring and \(I \subset A\) is an ideal. A morphism of pairs \((A, I) \to (B, J)\) is a ring map \(\varphi : A \to B\) with \(\varphi(I) \subset J\).
Definition
A Zariski pair is a pair \((A, I)\) such that \(I\) is contained in the Jacobson radical of \(A\).
Lemma
Let \((A, I)\) be a Zariski pair. Then the map from idempotents of \(A\) to idempotents of \(A/I\) is injective.
Proof
An idempotent of a local ring is either \(0\) or \(1\). Thus an idempotent is determined by the set of maximal ideals where it vanishes, by Algebra, Lemma 00HN.
Lemma
Let \((A, I)\) be a Zariski pair. Let \(A \to B\) be a flat, integral, finitely presented ring map such that \(A/I \to B/IB\) is an isomorphism. Then \(A \to B\) is an isomorphism.
Proof
The ring map \(A \to B\) is finite by Algebra, Lemma 02JJ. Hence \(B\) is finitely presented as an \(A\)-module by Algebra, Lemma 0564. Hence \(B\) is a finite locally free \(A\)-module by Algebra, Lemma 00NX. Since the module \(B\) has rank \(1\) along \(V(I)\) (see rank function described in Algebra, Lemma 00NX), and as \((A, I)\) is a Zariski pair, we conclude that the rank is \(1\) everywhere. It follows that \(A \to B\) is an isomorphism: it is a pleasant exercise to show that a ring map \(R \to S\) such that \(S\) is a locally free \(R\)-module of rank \(1\) is an isomorphism (hint: look at local rings).
Lemma
Let \((A, I)\) be a Zariski pair. Let \(A \to B\) be a finite ring map. Assume
\(B/IB = B_1 \times B_2\) is a product of \(A/I\)-algebras
\(A/I \to B_1/IB_1\) is surjective,
\(b \in B\) maps to \((1, 0)\) in the product.
Then there exists a monic \(f \in A[x]\) with \(f(b) = 0\) and \(f \bmod I = (x - 1)x^d\) for some \(d \geq 1\).
Proof
By Lemma 07M4 we can find an étale ring map \(A \to A'\) inducing an isomorphism \(A/I \to A'/IA'\) such that \(B' = B \otimes_A A'\) contains an idempotent \(e'\) lifting the image of \(b\) in \(B'/IB'\). Consider the corresponding \(A'\)-algebra decomposition \[B' = B'_1 \times B'_2\] which is compatible with the one given in the lemma upon reduction modulo \(I\). The map \(A' \to B'_1\) is surjective modulo \(IA'\). By Nakayama’s lemma (Algebra, Lemma 00DV) we can find \(i \in IA'\) such that after replacing \(A'\) by \(A'_{1 + i}\) the map \(A' \to B'_1\) is surjective. Observe that the image \(b'_1 \in B'_1\) of \(b\) satisfies \(b'_1 - 1 \in IB'_1\). Thus we may pick \(a' \in IA'\) mapping to \(b'_1 - 1\). On the other hand, the image \(b'_2 \in B'_2\) of \(b\) is in \(IB'_2\). By Algebra, Lemma 00H5 there exist a monic polynomial \(g(x) = x^d + \sum a'_j x^j\) of degree \(d\) with \(a'_j \in IA'\) such that \(g(b'_2) = 0\) in \(B'_2\). Thus the image \(b' = (b'_1, b'_2) \in B'\) of \(b\) is a root of the polynomial \((x - 1 - a')g(x)\). We conclude that \[(b' - 1)(b')^d \in \sum\nolimits_{j = 0, \ldots, d} IA' \cdot (b')^j\] We claim that this implies \[(b - 1)b^d \in \sum\nolimits_{j = 0, \ldots, d} I \cdot b^j\] in \(B\). For this it is enough to see that the ring map \(A \to A'\) is faithfully flat, because the condition is that the image of \((b - 1)b^d\) is zero in \(B/\sum_{j = 0, \ldots, d} Ib^j\) (use Algebra, Lemma 05CK). The map \(A \to A'\) flat because it is étale (Algebra, Lemma 00U2). On the other hand, the induced map on spectra is open (see Algebra, Proposition 00I1 and use previous lemma referenced) and the image contains \(V(I)\). Since \(I\) is contained in the Jacobson radical of \(A\) the image is all of \(\Spec(A)\) by Algebra, Lemma algebra-lemma-open-containing-vanishing-jacobson-radical. Thus \(A \to A'\) is faithfully flat by Algebra, Lemma 00HQ, and we conclude.
Lemma
Let \((A, I)\) be a Zariski pair with \(A\) Noetherian. Let \(f \in I\). Then \(A_f\) is a Jacobson ring.
Proof
We will use the criterion of Algebra, Lemma 00KX. Let \(\mathfrak p \subset A\) be a prime ideal such that \(\mathfrak p_f = \mathfrak p A_f\) is prime and not maximal. We have to show that \(A_f/\mathfrak p_f = (A/\mathfrak p)_f\) has infinitely many prime ideals. After replacing \(A\) by \(A/\mathfrak p\) we may assume \(A\) is a domain, \(\dim A_f > 0\), and our goal is to show that \(\Spec(A_f)\) is infinite. Since \(\dim A_f > 0\) we can find a nonzero prime ideal \(\mathfrak q \subset A\) not containing \(f\). Choose a maximal ideal \(\mathfrak m \subset A\) containing \(\mathfrak q\). Since \((A, I)\) is a Zariski pair, we see \(I \subset \mathfrak m\). Hence \(\mathfrak m \not = \mathfrak q\) and \(\dim(A_\mathfrak m) > 1\). Hence \(\Spec((A_\mathfrak m)_f) \subset \Spec(A_f)\) is infinite by Algebra, Lemma 02IG and we win.
Henselian pairs
Some of the results of Section 07LW may be viewed as results about henselian pairs. In this section a pair is a pair \((A, I)\) where \(A\) is a ring and \(I \subset A\) is an ideal. A morphism of pairs \((A, I) \to (B, J)\) is a ring map \(\varphi : A \to B\) with \(\varphi(I) \subset J\). As in Section 07LW given an object \(\xi\) over \(A\) we denote \(\overline{\xi}\) the “base change” of \(\xi\) to an object over \(A/I\) (provided this makes sense).
Definition
A henselian pair is a pair \((A, I)\) satisfying
\(I\) is contained in the Jacobson radical of \(A\), and
for any monic polynomial \(f \in A[T]\) and factorization \(\overline{f} = g_0h_0\) with \(g_0, h_0 \in A/I[T]\) monic generating the unit ideal in \(A/I[T]\), there exists a factorization \(f = gh\) in \(A[T]\) with \(g, h\) monic and \(g_0 = \overline{g}\) and \(h_0 = \overline{h}\).
Observe that if \(A\) is a local ring and \(I = \mathfrak m\) is the maximal ideal, then \((A, I)\) is a henselian pair if and only if \(A\) is a henselian local ring, see Algebra, Lemma 04GG. In Lemma 09XI we give a number of equivalent characterizations of henselian pairs (and we will add more as time goes on).
Lemma
Let \((A, I)\) be a pair with \(I\) locally nilpotent. Then the functor \(B \mapsto B/IB\) induces an equivalence between the category of étale algebras over \(A\) and the category of étale algebras over \(A/I\). Moreover, the pair is henselian.
Proof
Essential surjectivity holds by Algebra, Lemma 04D1. If \(B\), \(B'\) are étale over \(A\) and \(B/IB \to B'/IB'\) is a morphism of \(A/I\)-algebras, then we can lift this by Algebra, Lemma 07K4. Finally, suppose that \(f, g : B \to B'\) are two \(A\)-algebra maps with \(f \bmod I = g \bmod I\). Choose an idempotent \(e \in B \otimes_A B\) generating the kernel of the multiplication map \(B \otimes_A B \to B\), see Algebra, Lemmas 02FL and 00UV (to see that étale is unramified). Then \((f \otimes g)(e) \in IB'\). Since \(IB'\) is locally nilpotent (Algebra, Lemma 0544) this implies \((f \otimes g)(e) = 0\) by Algebra, Lemma 00J9. Thus \(f = g\).
It is clear that \(I\) is contained in the Jacobson radical of \(A\). Let \(f \in A[T]\) be a monic polynomial and let \(\overline{f} = g_0h_0\) be a factorization of \(\overline{f} = f \bmod I\) with \(g_0, h_0 \in A/I[T]\) monic generating the unit ideal in \(A/I[T]\). By Lemma 0ALH there exists an étale ring map \(A \to A'\) which induces an isomorphism \(A/I \to A'/IA'\) such that the factorization lifts to a factorization into monic polynomials over \(A'\). By the above we have \(A = A'\) and the factorization is over \(A\).
Lemma
Let \(A = \lim A_n\) where \((A_n)\) is an inverse system of rings whose transition maps are surjective and have locally nilpotent kernels. Then \((A, I_n)\) is a henselian pair, where \(I_n = \Ker(A \to A_n)\).
Proof
Fix \(n\). Let \(a \in A\) be an element which maps to \(1\) in \(A_n\). By Algebra, Lemma 0AMG we see that \(a\) maps to a unit in \(A_m\) for all \(m \geq n\). Hence \(a\) is a unit in \(A\). Thus by Algebra, Lemma 0AME the ideal \(I_n\) is contained in the Jacobson radical of \(A\). Let \(f \in A[T]\) be a monic polynomial and let \(\overline{f} = g_nh_n\) be a factorization of \(\overline{f} = f \bmod I_n\) with \(g_n, h_n \in A_n[T]\) monic generating the unit ideal in \(A_n[T]\). By Lemma 0ALI we can successively lift this factorization to \(f \bmod I_m = g_m h_m\) with \(g_m, h_m\) monic in \(A_m[T]\) for all \(m \geq n\). At each step we have to verify that our lifts \(g_m, h_m\) generate the unit ideal in \(A_n[T]\); this follows from the corresponding fact for \(g_n, h_n\) and the fact that \(\Spec(A_n[T]) = \Spec(A_m[T])\) because the kernel of \(A_m \to A_n\) is locally nilpotent. As \(A = \lim A_m\) this finishes the proof.
Lemma
Let \((A, I)\) be a pair. If \(A\) is \(I\)-adically complete, then the pair is henselian.
Proof
By Algebra, Lemma 05GI the ideal \(I\) is contained in the Jacobson radical of \(A\). Let \(f \in A[T]\) be a monic polynomial and let \(\overline{f} = g_0h_0\) be a factorization of \(\overline{f} = f \bmod I\) with \(g_0, h_0 \in A/I[T]\) monic generating the unit ideal in \(A/I[T]\). By Lemma 0ALI we can successively lift this factorization to \(f \bmod I^n = g_n h_n\) with \(g_n, h_n\) monic in \(A/I^n[T]\) for all \(n \geq 1\). As \(A = \lim A/I^n\) this finishes the proof.
Lemma
Let \((A, I)\) be a pair. Let \(A \to B\) be a finite type ring map such that \(B/IB = C_1 \times C_2\) with \(A/I \to C_1\) finite. Let \(B'\) be the integral closure of \(A\) in \(B\). Then we can write \(B'/IB' = C_1 \times C'_2\) such that the map \(B'/IB' \to B/IB\) preserves product decompositions and there exists a \(g \in B'\) mapping to \((1, 0)\) in \(C_1 \times C'_2\) with \(B'_g \to B_g\) an isomorphism.
Proof
Observe that \(A \to B\) is quasi-finite at every prime of the closed subset \(T = \Spec(C_1) \subset \Spec(B)\) (this follows by looking at fibre rings, see Algebra, Definition 00PL). Consider the diagram of topological spaces \[\xymatrix{ \Spec(B) \ar[rr]_\phi \ar[rd]_\psi & & \Spec(B') \ar[ld]^{\psi'} \\ & \Spec(A) }\] By Algebra, Theorem 00Q9 for every \(\mathfrak p \in T\) there is a \(h_\mathfrak p \in B'\), \(h_\mathfrak p \not \in \mathfrak p\) such that \(B'_h \to B_h\) is an isomorphism. The union \(U = \bigcup D(h_\mathfrak p)\) gives an open \(U \subset \Spec(B')\) such that \(\phi^{-1}(U) \to U\) is a homeomorphism and \(T \subset \phi^{-1}(U)\). Since \(T\) is open in \(\psi^{-1}(V(I))\) we conclude that \(\phi(T)\) is open in \(U \cap (\psi')^{-1}(V(I))\). Thus \(\phi(T)\) is open in \((\psi')^{-1}(V(I))\). On the other hand, since \(C_1\) is finite over \(A/I\) it is finite over \(B'\). Hence \(\phi(T)\) is a closed subset of \(\Spec(B')\) by Algebra, Lemmas 00HZ and 00GU. We conclude that \(\Spec(B'/IB') \supset \phi(T)\) is open and closed. By Algebra, Lemma 00EM we get a corresponding product decomposition \(B'/IB' = C'_1 \times C'_2\). The map \(B'/IB' \to B/IB\) maps \(C'_1\) into \(C_1\) and \(C'_2\) into \(C_2\) as one sees by looking at what happens on spectra (hint: the inverse image of \(\phi(T)\) is exactly \(T\); some details omitted). Pick a \(g \in B'\) mapping to \((1, 0)\) in \(C'_1 \times C'_2\) such that \(D(g) \subset U\); this is possible because \(\Spec(C'_1)\) and \(\Spec(C'_2)\) are disjoint and closed in \(\Spec(B')\) and \(\Spec(C'_1)\) is contained in \(U\). Then \(B'_g \to B_g\) defines a homeomorphism on spectra and an isomorphism on local rings (by our choice of \(U\) above). Hence it is an isomorphism, as follows for example from Algebra, Lemma 00HN. Finally, it follows that \(C'_1 = C_1\) and the proof is complete.
Lemma
Let \((A, I)\) be a pair. The following are equivalent
\((A, I)\) is a henselian pair,
given an étale ring map \(A \to A'\) and an \(A\)-algebra map \(\sigma : A' \to A/I\), there exists an \(A\)-algebra map \(A' \to A\) lifting \(\sigma\),
for any finite \(A\)-algebra \(B\) the map \(B \to B/IB\) induces a bijection on idempotents,
for any integral \(A\)-algebra \(B\) the map \(B \to B/IB\) induces a bijection on idempotents, and
(Gabber) \(I\) is contained in the Jacobson radical of \(A\) and every monic polynomial \(f(T) \in A[T]\) of the form \[f(T) = T^n(T - 1) + a_n T^n + \ldots + a_1 T + a_0\] with \(a_n, \ldots, a_0 \in I\) and \(n \ge 1\) has a root \(\alpha \in 1 + I\).
Moreover, in part (5) the root is unique.
Proof
Assume (2) holds. Then \(I\) is contained in the Jacobson radical of \(A\), since otherwise there would be a nonunit \(f \in A\) congruent to \(1\) modulo \(I\) and the map \(A \to A_f\) would contradict (2). Hence \(IB \subset B\) is contained in the Jacobson radical of \(B\) for \(B\) integral over \(A\) because \(\Spec(B) \to \Spec(A)\) is closed by Algebra, Lemmas 00HZ and 00GU. Thus the map from idempotents of \(B\) to idempotents of \(B/IB\) is injective by Lemma 09XF. On the other hand, since (2) holds, every idempotent of \(B/IB\) lifts to an idempotent of \(B\) by Lemma 07M4. In this way we see that (2) implies (4).
The implication (4) \(\Rightarrow\) (3) is trivial.
Assume (3). Let \(\mathfrak m\) be a maximal ideal and consider the finite map \(A \to B = A/(I \cap \mathfrak m)\). The condition that \(B \to B/IB\) induces a bijection on idempotents implies that \(I \subset \mathfrak m\) (if not, then \(B = A/I \times A/\mathfrak m\) and \(B/IB = A/I\)). Thus we see that \(I\) is contained in the Jacobson radical of \(A\). Let \(f \in A[T]\) be monic and suppose given a factorization \(\overline{f} = g_0h_0\) with \(g_0, h_0 \in A/I[T]\) monic generating the unit ideal in \(A/I[T]\). Set \(B = A[T]/(f)\). Let \(\overline{e}\) be the idempotent of \(B/IB\) corresponding to the decomposition \[B/IB = A/I[T]/(g_0) \times A/I[T]/(h_0)\] of \(A\)-algebras. Let \(e \in B\) be an idempotent lifting \(\overline{e}\) which exists as we assumed (3). This gives a product decomposition \[B = eB \times (1 - e)B\] Note that \(B\) is free of rank \(\deg(f)\) as an \(A\)-module. Hence \(eB\) and \((1 - e)B\) are finite locally free \(A\)-modules. However, since \(eB\) and \((1 - e)B\) have constant rank \(\deg(g_0)\) and \(\deg(h_0)\) over \(A/I\) we find that the same is true over \(\Spec(A)\). We conclude that \[\begin{align*} f & = \text{CharPol}_A(T : B \to B) \\ & = \text{CharPol}_A(T : eB \to eB) \text{CharPol}_A(T : (1 - e)B \to (1 - e)B) \end{align*}\] is a factorization into monic polynomials reducing to the given factorization modulo \(I\). Here \(\text{CharPol}_A\) denotes the characteristic polynomial of an endomorphism of a finite locally free module over \(A\). If the module is free the \(\text{CharPol}_A\) is defined as the characteristic polynomial of the corresponding matrix and in general one uses Algebra, Lemma 00EJ to glue. Details omitted. Thus (3) implies (1).
Assume (1). Let \(f\) be as in (5). The factorization of \(f \bmod I\) as \(T^n\) times \(T - 1\) lifts to a factorization \(f = gh\) with \(g\) and \(h\) monic by Definition 09XE. Then \(h\) has to have degree \(1\) and we see that \(f\) has a root reducing to \(1\) modulo \(I\). Finally, \(I\) is contained in the Jacobson radical by the definition of a henselian pair. Thus (1) implies (5).
Before we give the proof of the last step, let us show that the root \(\alpha\) in (5), if it exists, is unique. Namely, due to the explicit shape of \(f(T)\), we have \(f'(\alpha) \in 1 + I\) where \(f'\) is the derivative of \(f\) with respect to \(T\). An elementary argument shows that \[f(T) = f(\alpha + T - \alpha) = f(\alpha) + f'(\alpha) \cdot (T - \alpha) \bmod (T - \alpha)^2 A[T]\] This shows that any other root \(\alpha' \in 1 + I\) of \(f(T)\) satisfies \(0 = f(\alpha') - f(\alpha) = (\alpha' - \alpha)(1 + i)\) for some \(i \in I\), so that, since \(1 + i\) is a unit in \(A\), we have \(\alpha = \alpha'\).
Assume (5). We will show that (2) holds, in other words, that for every étale map \(A \to A'\), every section \(\sigma : A' \to A/I\) modulo \(I\) lifts to a section \(A' \to A\). Since \(A \to A'\) is étale, the section \(\sigma\) determines a decomposition [0EM1]\[\begin{equation} A'/IA' \cong A/I \times C \end{equation}\] of \(A/I\)-algebras. Namely, the surjective ring map \(A'/IA' \to A/I\) is étale by Algebra, Lemma 00U7 and then we get the desired idempotent by Algebra, Lemma 00U8. We will show that this decomposition lifts to a decomposition [0EM2]\[\begin{equation} A' \cong A'_1 \times A'_2 \end{equation}\] of \(A\)-algebras with \(A'_1\) integral over \(A\). Then \(A \to A'_1\) is integral and étale and \(A/I \to A'_1/IA'_1\) is an isomorphism, thus \(A \to A'_1\) is an isomorphism by Lemma 0ELZ (here we also use that an étale ring map is flat and of finite presentation, see Algebra, Lemma 00U2).
Let \(B'\) be the integral closure of \(A\) in \(A'\). By Lemma 09XH we may decompose [0EM3]\[\begin{equation} B'/IB' \cong A/I \times C' \end{equation}\] as \(A/I\)-algebras compatibly with (0EM1) and we may find \(b \in B'\) that lifts \((1, 0)\) such that \(B'_b \to A'_b\) is an isomorphism. If the decomposition (0EM3) lifts to a decomposition [0EM4]\[\begin{equation} B' \cong B'_1 \times B'_2 \end{equation}\] of \(A\)-algebras, then the induced decomposition \(A' = A'_1 \times A'_2\) will give the desired (0EM2): indeed, since \(b\) is a unit in \(B'_1\) (details omitted), we will have \(B'_1 \cong A'_1\), so that \(A'_1\) will be integral over \(A\).
Choose a finite \(A\)-subalgebra \(B'' \subset B'\) containing \(b\) (observe that any finitely generated \(A\)-subalgebra of \(B'\) is finite over \(A\)). After enlarging \(B''\) we may assume \(b\) maps to an idempotent in \(B''/IB''\) producing [0EM5]\[\begin{equation} B''/IB'' \cong C''_1 \times C''_2 \end{equation}\] Since \(B'_b \cong A'_b\) we see that \(B'_b\) is of finite type over \(A\). Say \(B'_b\) is generated by \(b_1/b^n, \ldots, b_t/b^n\) over \(A\) and enlarge \(B''\) so that \(b_1, \ldots, b_t \in B''\). Then \(B''_b \to B'_b\) is surjective as well as injective, hence an isomorphism. In particular, we see that \(C''_1 = A/I\)! Therefore \(A/I \to C''_1\) is an isomorphism, in particular surjective. By Lemma 0EM0 we can find an \(f(T) \in A[T]\) of the form \[f(T) = T^n(T - 1) + a_n T^n + \ldots + a_1 T + a_0\] with \(a_n, \ldots, a_0 \in I\) and \(n \ge 1\) such that \(f(b) = 0\). In particular, we find that \(B'\) is a \(A[T]/(f)\)-algebra. By (5) we deduce there is a root \(a \in 1 + I\) of \(f\). This produces a product decomposition \(A[T]/(f) = A[T]/(T - a) \times D\) compatible with the splitting (0EM3) of \(B'/IB'\). The induced splitting of \(B'\) is then a desired (0EM4).
Lemma
Let \(A\) be a ring. Let \(I, J \subset A\) be ideals with \(V(I) = V(J)\). Then \((A, I)\) is henselian if and only if \((A, J)\) is henselian.
Proof
For any integral ring map \(A \to B\) we see that \(V(IB) = V(JB)\). Hence idempotents of \(B/IB\) and \(B/JB\) are in bijective correspondence (Algebra, Lemma 00EE). It follows that \(B \to B/IB\) induces a bijection on sets of idempotents if and only if \(B \to B/JB\) induces a bijection on sets of idempotents. Thus we conclude by Lemma 09XI.
Lemma
Let \((A, I)\) be a henselian pair and let \(A \to B\) be an integral ring map. Then \((B, IB)\) is a henselian pair.
Proof
Immediate from the fourth characterization of henselian pairs in Lemma 09XI and the fact that the composition of integral ring maps is integral.
Lemma
Let \(I \subset J \subset A\) be ideals of a ring \(A\). The following are equivalent
\((A, I)\) and \((A/I, J/I)\) are henselian pairs, and
\((A, J)\) is an henselian pair.
Proof
Assume (1). Let \(B\) be an integral \(A\)-algebra. Consider the ring maps \[B \to B/IB \to B/JB\] By Lemma 09XI we find that both arrows induce bijections on idempotents. Hence so does the composition. Whence \((A, J)\) is a henselian pair by Lemma 09XI.
Conversely, assume (2) holds. Then \((A/I, J/I)\) is a henselian pair by Lemma 09XK. Let \(B\) be an integral \(A\)-algebra. Consider the ring maps \[B \to B/IB \to B/JB\] By Lemma 09XI we find that the composition and the second arrow induce bijections on idempotents. Hence so does the first arrow. It follows that \((A, I)\) is a henselian pair (by the lemma again).
Lemma
Let \(A\) be a ring and let \((A, I)\) and \((A, I')\) be henselian pairs. Then \((A, I + I')\) is an henselian pair.
Proof
By Lemma 09XK the pair \((A/I, (I' + I)/I)\) is henselian. Thus we get the conclusion from Lemma 0DYD.
Lemma
Let \(J\) be a set and let \(\{ (A_j, I_j)\}_{j \in J}\) be a collection of pairs. Then \((\prod_{j \in J} A_j, \prod_{j\in J} I_j)\) is Henselian if and only if so is each \((A_j, I_j)\).
Proof
For every \(j \in J\), the projection \(\prod_{j \in J} A_j \rightarrow A_j\) is an integral ring map, so Lemma 09XK proves that each \((A_j, I_j)\) is Henselian if \((\prod_{j \in J} A_j, \prod_{j\in J} I_j)\) is Henselian.
Conversely, suppose that each \((A_j, I_j)\) is a Henselian pair. Then every \(1 + x\) with \(x \in \prod_{j \in J} I_j\) is a unit in \(\prod_{j \in J} A_j\) because it is so componentwise by Algebra, Lemma 0AME and Definition 09XE. Thus, by Algebra, Lemma 0AME again, \(\prod_{j \in J} I_j\) is contained in the Jacobson radical of \(\prod_{j \in J} A_j\). Continuing to work componentwise, it likewise follows that for every monic \(f \in (\prod_{j \in J} A_j)[T]\) and every factorization \(\overline{f} = g_0h_0\) with monic \(g_0, h_0 \in (\prod_{j \in J} A_j / \prod_{j \in J} I_j)[T] = (\prod_{j \in J} A_j/I_j)[T]\) that generate the unit ideal in \((\prod_{j \in J} A_j / \prod_{j \in J} I_j)[T]\), there exists a factorization \(f = gh\) in \((\prod_{j \in J} A_j)[T]\) with \(g\), \(h\) monic and reducing to \(g_0\), \(h_0\). In conclusion, according to Definition 09XE \((\prod_{j \in J} A_j, \prod_{j\in J} I_j)\) is a Henselian pair.
Lemma
The property of being Henselian is preserved under limits of pairs. More precisely, let \(J\) be a preordered set and let \((A_j, I_j)\) be an inverse system of henselian pairs over \(J\). Then \(A = \lim A_j\) equipped with the ideal \(I = \lim I_j\) is a henselian pair \((A, I)\).
Proof
By Categories, Lemma 002N, we only need to consider products and equalizers. For products, the claim follows from Lemma 0ATD. Thus, consider an equalizer diagram \[\xymatrix{ (A, I) \ar[r] & (A', I') \ar@<1ex>[r]^{\varphi} \ar@<-1ex>[r]_{\psi} & (A'', I'') }\] in which the pairs \((A', I')\) and \((A'', I'')\) are henselian. To check that the pair \((A, I)\) is also henselian, we will use the Gabber’s criterion in Lemma 09XI. Every element of \(1 + I\) is a unit in \(A\) because, due to the uniqueness of the inverses of units, this may be checked in \((A', I')\). Thus \(I\) is contained in the Jacobson radical of \(A\), see Algebra, Lemma 0AME. Thus, let \[f(T) = T^{N - 1}(T - 1) + a_{N - 1} T^{N - 1} + \dotsb + a_1 T + a_0\] be a polynomial in \(A[T]\) with \(a_{N - 1}, \dotsc, a_0 \in I\) and \(N \ge 1\). The image of \(f(T)\) in \(A'[T]\) has a unique root \(\alpha' \in 1 + I'\) and likewise for the further image in \(A''[T]\). Thus, due to the uniqueness, \(\varphi(\alpha') = \psi(\alpha')\), to the effect that \(\alpha'\) defines a root of \(f(T)\) in \(1 + I\), as desired.
Lemma
The property of being Henselian is preserved under filtered colimits of pairs. More precisely, let \(J\) be a directed set and let \((A_j, I_j)\) be a system of henselian pairs over \(J\). Then \(A = \colim A_j\) equipped with the ideal \(I = \colim I_j\) is a henselian pair \((A, I)\).
Proof
If \(u \in 1 + I\) then for some \(j \in J\) we see that \(u\) is the image of some \(u_j \in 1 + I_j\). Then \(u_j\) is invertible in \(A_j\) by Algebra, Lemma 0AME and the assumption that \(I_j\) is contained in the Jacobson radical of \(A_j\). Hence \(u\) is invertible in \(A\). Thus \(I\) is contained in the Jacobson radical of \(A\) (by the lemma).
Let \(f \in A[T]\) be a monic polynomial and let \(\overline{f} = g_0 h_0\) be a factorization with \(g_0, h_0 \in A/I[T]\) monic generating the unit ideal in \(A/I[T]\). Write \(1 = g_0 g'_0 + h_0 h'_0\) for some \(g'_0, h'_0 \in A/I[T]\). Since \(A = \colim A_j\) and \(A/I = \colim A_j/I_j\) are filtered colimits we can find a \(j \in J\) and \(f_j \in A_j\) and a factorization \(\overline{f}_j = g_{j, 0} h_{j, 0}\) with \(g_{j, 0}, h_{j, 0} \in A_j/I_j[T]\) monic and \(1 = g_{j, 0} g'_{j, 0} + h_{j, 0} h'_{j, 0}\) for some \(g'_{j, 0}, h'_{j, 0} \in A_j/I_j[T]\) with \(f_j, g_{j, 0}, h_{j, 0}, g'_{j, 0}, h'_{j, 0}\) mapping to \(f, g_0, h_0, g'_0, h'_0\). Since \((A_j, I_j)\) is a henselian pair, we can lift \(\overline{f}_j = g_{j, 0} h_{j, 0}\) to a factorization over \(A_j\) and taking the image in \(A\) we obtain a corresponding factorization in \(A\). Hence \((A, I)\) is henselian.
Example
Lemma 0FWT is wrong if the colimit isn’t filtered. For example, if we take the coproduct of the henselian pairs \((\mathbf{Z}_p, (p))\) and \((\mathbf{Z}_p, (p))\), then we obtain \((A, pA)\) with \(A = \mathbf{Z}_p \otimes_\mathbf{Z} \mathbf{Z}_p\). This isn’t a henselian pair: \(A/pA = \mathbf{F}_p\) hence if \((A, pA)\) where henselian, then \(A\) would have to be local. However, \(\Spec(A)\) is disconnected; for example for odd primes \(p\) we have the nontrivial idempotent \[(1/2 \otimes 1) \left(1 \otimes 1 - (1 + p)^{-1}u \otimes u\right)\] where \(u \in \mathbf{Z}_p\) is a square root of \(1 + p\). Some details omitted.
Lemma
Let \(A\) be a ring. There exists a largest ideal \(I \subset A\) such that \((A, I)\) is a henselian pair.
Proof
Lemma
Let \((A, I)\) be a henselian pair. Let \(\mathfrak p \subset A\) be a prime ideal. Then \(V(\mathfrak p + I)\) is connected.
Proof
By Lemma 09XK we see that \((A/\mathfrak p, I + \mathfrak p/\mathfrak p)\) is a henselian pair. Thus it suffices to prove: If \((A, I)\) is a henselian pair and \(A\) is a domain, then \(\Spec(A/I) = V(I)\) is connected. If not, then \(A/I\) has a nontrivial idempotent by Algebra, Lemma 00EF. By Lemma 09XI this would imply \(A\) has a nontrivial idempotent. This is a contradiction.
Henselization of pairs
We continue the discussion started in Section 09XD.
Lemma
The inclusion functor \[\text{category of henselian pairs} \longrightarrow \text{category of pairs}\] has a left adjoint \((A, I) \mapsto (A^h, I^h)\).
Proof
Let \((A, I)\) be a pair. Consider the category \(\mathcal{C}\) consisting of étale ring maps \(A \to B\) such that \(A/I \to B/IB\) is an isomorphism. We will show that the category \(\mathcal{C}\) is directed and that \(A^h = \colim_{B \in \mathcal{C}} B\) with ideal \(I^h = IA^h\) gives the desired adjoint.
We first prove that \(\mathcal{C}\) is directed (Categories, Definition 002V). It is nonempty because \(\text{id} : A \to A\) is an object. If \(B\) and \(B'\) are two objects of \(\mathcal{C}\), then \(B'' = B \otimes_A B'\) is an object of \(\mathcal{C}\) (use Algebra, Lemma 00U2) and there are morphisms \(B \to B''\) and \(B' \to B''\). Suppose that \(f, g : B \to B'\) are two maps between objects of \(\mathcal{C}\). Then a coequalizer is \[(B' \otimes_{f, B, g} B') \otimes_{(B' \otimes_A B')} B'\] which is étale over \(A\) by Algebra, Lemmas 00U2 and 00U7. Thus the category \(\mathcal{C}\) is directed.
Since \(B/IB = A/I\) for all objects \(B\) of \(\mathcal{C}\) we see that \(A^h/I^h = A^h/IA^h = \colim B/IB = \colim A/I = A/I\).
Next, we show that \(A^h = \colim_{B \in \mathcal{C}} B\) with \(I^h = IA^h\) is a henselian pair. To do this we will verify condition (2) of Lemma 09XI. Namely, suppose given an étale ring map \(A^h \to A'\) and \(A^h\)-algebra map \(\sigma : A' \to A^h/I^h\). Then there exists a \(B \in \mathcal{C}\) and an étale ring map \(B \to B'\) such that \(A' = B' \otimes_B A^h\). See Algebra, Lemma 00U2. Since \(A^h/I^h = A/I\), the map \(\sigma\) induces an \(A\)-algebra map \(s : B' \to A/I\). Then \(B'/IB' = A/I \times C\) as \(A/I\)-algebra, where \(C\) is the kernel of the map \(B'/IB' \to A/I\) induced by \(s\). Let \(g \in B'\) map to \((1, 0) \in A/I \times C\). Then \(B \to B'_g\) is étale and \(A/I \to B'_g/IB'_g\) is an isomorphism, i.e., \(B'_g\) is an object of \(\mathcal{C}\). Thus we obtain a canonical map \(B'_g \to A^h\) such that \[\vcenter{ \xymatrix{ B'_g \ar[r] & A^h \\ B \ar[u] \ar[ur] } } \quad\text{and}\quad \vcenter{ \xymatrix{ B' \ar[r] \ar[rrd]_s & B'_g \ar[r] & A^h \ar[d] \\ & & A/I } }\] commute. This induces a map \(A' = B' \otimes_B A^h \to A^h\) compatible with \(\sigma\) as desired.
Let \((A, I) \to (A', I')\) be a morphism of pairs with \((A', I')\) henselian. We will show there is a unique factorization \(A \to A^h \to A'\) which will finish the proof. Namely, for each \(A \to B\) in \(\mathcal{C}\) the ring map \(A' \to B' = A' \otimes_A B\) is étale and induces an isomorphism \(A'/I' \to B'/I'B'\). Hence there is a section \(\sigma_B : B' \to A'\) by Lemma 09XI. Given a morphism \(B_1 \to B_2\) in \(\mathcal{C}\) we claim the diagram \[\xymatrix{ B'_1 \ar[rr] \ar[rd]_{\sigma_{B_1}} & & B'_2 \ar[ld]^{\sigma_{B_2}} \\ & A' }\] commutes. This follows once we prove that for every \(B\) in \(\mathcal{C}\) the section \(\sigma_B\) is the unique \(A'\)-algebra map \(B' \to A'\). We have \(B' \otimes_{A'} B' = B' \times R\) for some ring \(R\), see Algebra, Lemma 02FL. In our case \(R/I'R = 0\) as \(B'/I'B' = A'/I'\). Thus given two \(A'\)-algebra maps \(\sigma_B, \sigma_B' : B' \to A'\) then \(e = (\sigma_B \otimes \sigma_B')(0, 1) \in A'\) is an idempotent contained in \(I'\). We conclude that \(e = 0\) by Lemma 09XF. Hence \(\sigma_B = \sigma_B'\) as desired. Using the commutativity we obtain \[A^h = \colim_{B \in \mathcal{C}} B \to \colim_{B \in \mathcal{C}} A' \otimes_A B \xrightarrow{\colim \sigma_B} A'\] as desired. The uniqueness of the maps \(\sigma_B\) also guarantees that this map is unique. Hence \((A, I) \mapsto (A^h, I^h)\) is the desired adjoint.
Lemma
Let \((A, I)\) be a pair. Let \((A^h, I^h)\) be as in Lemma 0A02. Then \(A \to A^h\) is flat, \(I^h = IA^h\) and \(A/I^n \to A^h/I^nA^h\) is an isomorphism for all \(n\).
Proof
In the proof of Lemma 0A02 we have seen that \(A^h\) is a filtered colimit of étale \(A\)-algebras \(B\) such that \(A/I \to B/IB\) is an isomorphism and we have seen that \(I^h = IA^h\). As an étale ring map is flat (Algebra, Lemma 00U2) we conclude that \(A \to A^h\) is flat by Algebra, Lemma 05UT. Since each \(A \to B\) is flat we find that the maps \(A/I^n \to B/I^nB\) are isomorphisms as well (for example by Algebra, Lemma 051H). Taking the colimit we find that \(A/I^n = A^h/I^nA^h\) as desired.
Lemma
The functor of Lemma 0A02 associates to a local ring \((A, \mathfrak m)\) its henselization.
Proof
Let \((A^h, \mathfrak m^h)\) be the henselization of the pair \((A, \mathfrak m)\) constructed in Lemma 0A02. Then \(\mathfrak m^h = \mathfrak m A^h\) is a maximal ideal by Lemma 0AGU and since it is contained in the Jacobson radical, we conclude \(A^h\) is local with maximal ideal \(\mathfrak m^h\). Having said this there are two ways to finish the proof.
First proof: observe that the construction in the proof of Algebra, Lemma 04GN as a colimit is the same as the colimit used to construct \(A^h\) in Lemma 0A02. Second proof: Both the henselization \(A \to S\) and \(A \to A^h\) of Lemma 0A02 are local ring homomorphisms, both \(S\) and \(A^h\) are filtered colimits of étale \(A\)-algebras, both \(S\) and \(A^h\) are henselian local rings, and both \(S\) and \(A^h\) have residue fields equal to \(\kappa(\mathfrak m)\) (by Lemma 0AGU for the second case). Hence they are canonically isomorphic by Algebra, Lemma 08HT.
Lemma
Let \((A, I)\) be a pair with \(A\) Noetherian. Let \((A^h, I^h)\) be as in Lemma 0A02. Then the map of \(I\)-adic completions \[A^\wedge \to (A^h)^\wedge\] is an isomorphism. Moreover, \(A^h\) is Noetherian, the maps \(A \to A^h \to A^\wedge\) are flat, and \(A^h \to A^\wedge\) is faithfully flat.
Proof
The first statement is an immediate consequence of Lemma 0AGU and in fact holds without assuming \(A\) is Noetherian. In the proof of Lemma 0A02 we have seen that \(A^h\) is a filtered colimit of étale \(A\)-algebras \(B\) such that \(A/I \to B/IB\) is an isomorphism. For each such \(A \to B\) the induced map \(A^\wedge \to B^\wedge\) is an isomorphism (see proof of Lemma 0AGU). By Algebra, Lemma 00MB the ring map \(B \to A^\wedge = B^\wedge = (A^h)^\wedge\) is flat for each \(B\). Thus \(A^h \to A^\wedge = (A^h)^\wedge\) is flat by Algebra, Lemma 05UU. Since \(I^h = IA^h\) is contained in the Jacobson radical of \(A^h\) and since \(A^h \to A^\wedge\) induces an isomorphism \(A^h/I^h \to A/I\) we see that \(A^h \to A^\wedge\) is faithfully flat by Algebra, Lemma 00HP. By Algebra, Lemma 0316 the ring \(A^\wedge\) is Noetherian. Hence we conclude that \(A^h\) is Noetherian by Algebra, Lemma 033E.
Lemma
Let \((A, I) = \colim (A_i, I_i)\) be a filtered colimit of pairs. The functor of Lemma 0A02 gives \(A^h = \colim A_i^h\) and \(I^h = \colim I_i^h\).
This lemma is false for non-filtered colimits, see Example 0FWU.
Proof
By Categories, Lemma 0038 we see that \((A^h, I^h)\) is the colimit of the system \((A_i^h, I_i^h)\) in the category of henselian pairs. Thus for a henselian pair \((B, J)\) we have \[\Mor((A^h, I^h), (B, J)) = \lim \Mor((A_i^h, I_i^h), (B, J)) = \Mor(\colim (A_i^h, I_i^h), (B, J))\] Here the colimit is in the category of pairs. Since the colimit is filtered we obtain \(\colim (A_i^h, I_i^h) = (\colim A_i^h, \colim I_i^h)\) in the category of pairs; details omitted. Again using the colimit is filtered, this is a henselian pair (Lemma 0FWT). Hence by the Yoneda lemma we find \((A^h, I^h) = (\colim A_i^h, \colim I_i^h)\).
Lemma
Let \(A\) be a ring with ideals \(I\) and \(J\). If \(V(I) = V(J)\) then the functor of Lemma 0A02 produces the same ring for the pair \((A, I)\) as for the pair \((A, J)\).
Proof
Let \((A', IA')\) be the pair produced by Lemma 0A02 starting with the pair \((A, I)\), see Lemma 0AGU. Let \((A'', JA'')\) be the pair produced by Lemma 0A02 starting with the pair \((A, J)\). By Lemma 09XJ we see that \((A', JA')\) is a henselian pair and \((A'', IA'')\) is a henselian pair. By the universal property of the construction we obtain unique \(A\)-algebra maps \(A'' \to A'\) and \(A' \to A''\). The uniqueness shows that these are mutually inverse.
Lemma
Let \((A, I) \to (B, J)\) be a map of pairs such that \(V(J) = V(IB)\). Let \((A^h , I^h) \to (B^h, J^h)\) be the induced map on henselizations (Lemma 0A02). If \(A \to B\) is integral, then the induced map \(A^h \otimes_A B \to B^h\) is an isomorphism.
Proof
By Lemma 0F0L we may assume \(J = IB\). By Lemma 09XK the pair \((A^h \otimes_A B, I^h(A^h \otimes_A B))\) is henselian. By the universal property of \((B^h, IB^h)\) we obtain a map \(B^h \to A^h \otimes_A B\). We omit the proof that this map is the inverse of the map in the lemma.
Lemma
Let \(I_1, I_2, \dots, I_n \subset A\) be ideals in a ring \(A\). Suppose \(I_i + I_j = A\) for \(i \not = j\), \(1 \leq i, j \leq n\). Denote \(A^h\) the henselization of the pair \((A, I_1 \cap \ldots \cap I_n)\) and \(A_i^h\) the henselization of the pair \((A, I_i)\). Then the natural morphism \[A^h \to \prod\nolimits_{i = 1, \ldots, n} A^h_i\] is an isomorphism.
Proof
The map comes from the functoriality of the construction. If \(n = 1\) the result is immediate. If \(n > 1\), observe that \((I_1 \cap \ldots \cap I_{n - 1}) + I_n = A\) by our condition on the ideals. Thus if we prove the lemma for \(n = 2\), then the lemma follows for all \(n\) by induction.
The case \(n = 2\). Choose \(f_i \in I_i\) such that \(f_1 + f_2 = 1\). Recall that \(A^h\) is constructed as the colimit of \(A\)-algebras \(B\) such that \(A \to B\) is étale and \(A/I_1 \cap I_2 \to B/(I_1 \cap I_2)B\) is an isomorphism. See proof of Lemma 0A02. Denote \(Hens\) the category of such \(A\)-algebras \(B\). Similarly, denote \(Hens_i\), \(i = 1, 2\) the category of étale \(A\)-algebras \(B\) such that \(A/I_i \to B/I_iB\) is an isomorphism.
Consider the subcategory \(Hens_1(f_2) \subset Hens_1\) consisting of those \(B\) such that \(f_2\) is invertible. This is a cofinal subcategory since if \(B\) is in \(Hens_1\), then \(B_{f_2}\) is in \(Hens_1\) as well (hint: \(B_{f_2}/I_1B_{f_2} = (B/I_1B)_{f_2} = B/I_1B\) because \(f_2\) is a unit modulo \(I_1\)). Similarly, the subcategory \(Hens_2(f_1) \subset Hens_2\) is cofinal. Hence to compute \(A_1^h\) and \(A_2^h\) we may work with this smaller categories, see Categories, Lemma 04E7.
Consider the functor \[Hens_1(f_2) \times Hens_2(f_1) \to Hens,\quad (B_1, B_2) \mapsto B_1 \times B_2\] This functor makes sense as \[\begin{align*} (B_1 \times B_2)/(I_1 \cap I_2)(B_1 \times B_2) & = B_1/(I_1 \cap I_2)B_1 \times B_2/(I_1 \cap I_2)B_2 \\ & = B_1/I_1B_1 \times B_2/I_2B_2 \\ & = A/I_1 \times A/I_2 \\ & = A/(I_1 \cap I_2) \end{align*}\] The second equality holds because \((I_1 \cap I_2)B_1 = I_1B_1 \cap I_2B_1 = I_1B_1\) as \(f_2 \in I_2\) is invertible in \(B_1\) (we have also used Algebra, Lemma 0BBY). The last equality holds by Algebra, Lemma 00DT. To finish the proof, it suffices to prove that the displayed functor is cofinal (see lemma quoted above). The first condition of Categories, Definition 04E6 holds because an object \(B\) of \(Hens\) maps to \(B_{f_2} \times B_{f_1}\) which is in the image of the functor. The second one always holds because if we have \(B \to B_1 \times B_2\) and \(B \to B'_1 \times B'_2\) with \(B_1, B'_1 \in Hens_1(f_2)\) and \(B_2, B'_2 \in Hens_2(f_1)\), then we can set \(B''_1 = B_1 \otimes_B B'_1\) which is in \(Hens_1(f_2)\) and similarly \(B''_2 = B_2 \otimes_B B''_2\) in \(Hens_2(f_1)\) to find a map \(B \to B''_1 \times B''_2\) fitting into the commutative diagram \[\xymatrix{ & B \ar[ld] \ar[d] \ar[rd] \\ B_1 \times B_2 \ar[r] & B_1'' \times B_2'' & B_1' \times B_2' \ar[l] }\] This finishes the proof.
Lifting and henselian pairs
In this section we mostly combine results from Sections 07LW and 09XD.
Lemma
Let \((R, I)\) be a henselian pair. The map \[P \longrightarrow P/IP\] induces a bijection between the sets of isomorphism classes of finite projective \(R\)-modules and finite projective \(R/I\)-modules. In particular, any finite projective \(R/I\)-module is isomorphic to \(P/IP\) for some finite projective \(R\)-module \(P\).
Proof
We first prove the final statement. Let \(\overline{P}\) be a finite projective \(R/I\)-module. We can find a finite projective module \(P'\) over some \(R'\) étale over \(R\) with \(R/I = R'/IR'\) such that \(P'/IP'\) is isomorphic to \(\overline{P}\), see Lemma 07M5. Then, since \((R, I)\) is a henselian pair, the étale ring map \(R \to R'\) has a section \(\tau : R' \to R\) (Lemma 09XI). Setting \(P = P' \otimes_{R', \tau} R\) we conclude that \(P/IP\) is isomorphic to \(\overline{P}\). Of course, this tells us that the map in the statement of the lemma is surjective.
Injectivity. Suppose that \(P_1\) and \(P_2\) are finite projective \(R\)-modules such that \(P_1/IP_1 \cong P_2/IP_2\) as \(R/I\)-modules. Since \(P_1\) is projective, we can find an \(R\)-module map \(u : P_1 \to P_2\) lifting the given isomorphism. Then \(u\) is surjective by Nakayama’s lemma (Algebra, Lemma 00DV). We similarly find a surjection \(v : P_2 \to P_1\). By Algebra, Lemma 05G8 the map \(v \circ u\) is an isomorphism and we conclude \(u\) is an isomorphism.
Lemma
Let \((A, I)\) be a henselian pair. The functor \(B \to B/IB\) determines an equivalence between finite étale \(A\)-algebras and finite étale \(A/I\)-algebras.
Proof
Let \(B, B'\) be two \(A\)-algebras finite étale over \(A\). Then \(B' \to B'' = B \otimes_A B'\) is finite étale as well (Algebra, Lemmas 00U2 and 02JK). Now we have \(1\)-to-\(1\) correspondences between
\(A\)-algebra maps \(B \to B'\),
sections of \(B' \to B''\), and
idempotents \(e\) of \(B''\) such that \(B' \to B'' \to eB''\) is an isomorphism.
The bijection between (2) and (3) sends \(\sigma : B'' \to B'\) to \(e\) such that \((1 - e)\) is the idempotent that generates the kernel of \(\sigma\) which exists by Algebra, Lemmas 00U7 and 00U8. There is a similar correspondence between \(A/I\)-algebra maps \(B/IB \to B'/IB'\) and idempotents \(\overline{e}\) of \(B''/IB''\) such that \(B'/IB' \to B''/IB'' \to \overline{e}(B''/IB'')\) is an isomorphism. However every idempotent \(\overline{e}\) of \(B''/IB''\) lifts uniquely to an idempotent \(e\) of \(B''\) (Lemma 09XI). Moreover, if \(B'/IB' \to \overline{e}(B''/IB'')\) is an isomorphism, then \(B' \to eB''\) is an isomorphism too by Nakayama’s lemma (Algebra, Lemma 00DV). In this way we see that the functor is fully faithful.
Essential surjectivity. Let \(A/I \to C\) be a finite étale map. By Algebra, Lemma 04D1 there exists an étale map \(A \to B\) such that \(B/IB \cong C\). Let \(B'\) be the integral closure of \(A\) in \(B\). By Lemma 09XH we have \(B'/IB' = C \times C'\) for some ring \(C'\) and \(B'_g \cong B_g\) for some \(g \in B'\) mapping to \((1, 0) \in C \times C'\). Since idempotents lift (Lemma 09XI) we get \(B' = B'_1 \times B'_2\) with \(C = B'_1/IB'_1\) and \(C' = B'_2/IB'_2\). The image of \(g\) in \(B'_1\) is invertible. Then \(B_g = B'_g = B'_1 \times (B_2)_g\) and this implies that \(A \to B'_1\) is étale. We conclude that \(B'_1\) is finite étale over \(A\) (integral étale implies finite étale by Algebra, Lemma 02JJ for example) and the proof is done.
Lemma
Let \(R\) be a ring and \(S\) a smooth \(R\)-algebra. Assume that \(A\) is an \(R\)-algebra and \((A,I)\) is a henselian pair. Then any \(R\)-algebra map \(S \to A/I\) can be lifted to an \(R\)-algebra map \(S \to A\).
Proof
Let \(\tau : S \to A/I\) be an \(R\)-algebra map. Observe that \(S \otimes_R A\) is a smooth \(A\)-algebra by Algebra, Lemma 00T4. Thus by Lemma 07M7 we can lift the induced map \(S \otimes_R A \to A/I\) to an \(A\)-algebra homorphism \(S \otimes_R A \to A'\) where \(A \to A'\) is étale and induces an isomorphism \(A/I \to A'/IA'\). Since \((A, I)\) is henselian there is an \(A\)-algebra map \(A' \to A\), see Lemma 09XI. The composition \(S \to S \otimes_R A \to A' \to A\) is the desired lift.
Lemma
Let \(A = \lim A_n\) be a limit of an inverse system \((A_n)\) of rings. Suppose given \(A_n\)-modules \(M_n\) and \(A_{n + 1}\)-module maps \(M_{n + 1} \to M_n\). Assume
the transition maps \(A_{n + 1} \to A_n\) are surjective with locally nilpotent kernels,
\(M_1\) is a finite projective \(A_1\)-module,
\(M_n\) is a finite flat \(A_n\)-module, and
the maps induce isomorphisms \(M_{n + 1} \otimes_{A_{n + 1}} A_n \to M_n\).
Then \(M = \lim M_n\) is a finite projective \(A\)-module and \(M \otimes_A A_n \to M_n\) is an isomorphism for all \(n\).
Proof
By Lemma 0CT7 the pair \((A, \Ker(A \to A_1))\) is henselian. By Lemma 0D4A we can choose a finite projective \(A\)-module \(P\) and an isomorphism \(P \otimes_A A_1 \to M_1\). Since \(P\) is projective, we can successively lift the \(A\)-module map \(P \to M_1\) to \(A\)-module maps \(P \to M_2\), \(P \to M_3\), and so on. Thus we obtain a map \[P \longrightarrow M\] Since \(P\) is finite projective, we can write \(A^{\oplus m} = P \oplus Q\) for some \(m \geq 0\) and \(A\)-module \(Q\). Since \(A = \lim A_n\) we conclude that \(P = \lim P \otimes_A A_n\). Hence, in order to show that the displayed \(A\)-module map is an isomorphism, it suffices to show that the maps \(P \otimes_A A_n \to M_n\) are isomorphisms. From Lemma 0D48 we see that \(M_n\) is a finite projective module. By Lemma 0BC6 the maps \(P \otimes_A A_n \to M_n\) are isomorphisms.
Absolute integral closure
Here is our definition.
Definition
A ring \(A\) is absolutely integrally closed if every monic \(f \in A[T]\) is a product of linear factors.
Be careful: it may be possible to write \(f\) as a product of linear factors in many different ways.
Lemma
Let \(A\) be a ring. The following are equivalent
\(A\) is absolutely integrally closed, and
any monic \(f \in A[T]\) has a root in \(A\).
Proof
Omitted.
Lemma
Let \(A\) be absolutely integrally closed.
Any quotient ring \(A/I\) of \(A\) is absolutely integrally closed.
Any localization \(S^{-1}A\) is absolutely integrally closed.
Proof
Omitted.
Lemma
Let \(A\) be a ring. Let \(S \subset A\) be a multiplicative subset consisting of nonzerodivisors. If \(S^{-1}A\) is absolutely integrally closed and \(A \subset S^{-1}A\) is integrally closed in \(S^{-1}A\), then \(A\) is absolutely integrally closed.
Proof
Omitted.
Lemma
Let \(A\) be a normal domain. Then \(A\) is absolutely integrally closed if and only if its fraction field is algebraically closed.
Proof
Observe that a field is algebraically closed if and only if it is absolutely integrally closed as a ring. Hence the lemma follows from Lemmas 0DCN and 0DCP.
Lemma
For any ring \(A\) there exists an extension \(A \subset B\) such that
\(B\) is a filtered colimit of finite free \(A\)-algebras,
\(B\) is free as an \(A\)-module, and
\(B\) is absolutely integrally closed.
Proof
Let \(I\) be the set of monic polynomials over \(A\). For \(i \in I\) denote \(x_i\) a variable and \(P_i\) the corresponding monic polynomial in the variable \(x_i\). Then we set \[F(A) = A[x_i; i \in I]/(P_i; i \in I)\] As the notation suggests \(F\) is a functor from the category of rings to itself. Note that \(A \subset F(A)\), that \(F(A)\) is free as an \(A\)-module, and that \(F(A)\) is a filtered colimit of finite free \(A\)-algebras. Then we take \[B = \colim F^n(A)\] where the transition maps are the inclusions \(F^n(A) \subset F(F^n(A)) = F^{n + 1}(A)\). Any monic polynomial with coefficients in \(B\) actually has coefficients in \(F^n(A)\) for some \(n\) and hence has a solution in \(F^{n + 1}(A)\) by construction. This implies that \(B\) is absolutely integrally closed by Lemma 0DCM. We omit the proof of the other properties.
Lemma
Let \(A\) be absolutely integrally closed. Let \(\mathfrak p \subset A\) be a prime. Then the local ring \(A_\mathfrak p\) is strictly henselian.
Proof
By Lemma 0DCN we may assume \(A\) is a local ring and \(\mathfrak p\) is its maximal ideal. The residue field is algebraically closed by Lemma 0DCN. Every monic polynomial decomposes completely into linear factors hence Algebra, Definition 04GF applies directly.
Lemma
Let \(A\) be absolutely integrally closed. Let \(I \subset A\) be an ideal. Then \((A, I)\) is a henselian pair if (and only if) the following conditions hold
\(I\) is contained in the Jacobson radical of \(A\),
\(A \to A/I\) induces a bijection on idempotents.
Proof
Let \(f \in A[T]\) be a monic polynomial and let \(f \bmod I = g_0 h_0\) be a factorization over \(A/I\) with \(g_0\), \(h_0\) monic such that \(g_0\) and \(h_0\) generate the unit ideal of \(A/I[T]\). This means that \[A/I[T]/(f) = A/I[T]/(g_0) \times A/I[T]/(h_0)\] Denote \(e \in A/I[T]/(f)\) the element correspoing to the idempotent \((1, 0)\) in the ring on the right. Write \(f = (T - a_1) \ldots (T - a_d)\) with \(a_i \in A\). For each \(i \in \{1, \ldots, d\}\) we obtain an \(A\)-algebra map \(\varphi_i : A[T]/(f) \to A\), \(T \mapsto a_i\) which induces a similar \(A/I\)-algebra map \(\overline{\varphi}_i : A/I[T]/(f) \to A/I\). Denote \(e_i = \overline{\varphi}_i(e) \in A/I\). These are idempotents. By our assumption (2) we can lift \(e_i\) to an idempotent in \(A\). This means we can write \(A = \prod A_j\) as a finite product of rings such that in \(A_j/IA_j\) each \(e_i\) is either \(0\) or \(1\). Some details omitted. Observe that \(A_j\) is absolutely integrally closed as a factor ring of \(A\). It suffices to lift the factorization of \(f\) over \(A_j/IA_j\) to \(A_j\). This reduces us to the situation discussed in the next paragraph.
Assume \(e_i = 1\) for \(i = 1, \ldots, r\) and \(e_i = 0\) for \(i = r + 1, \ldots, d\). From \((g_0, h_0) = A/I[T]\) we have that there are \(k_0, l_0 \in A/I[T]\) such that \(g_0 k_0 + h_0 l_0 = 1\). We see that \(e = h_0 l_0\) and \(e_i = h_0(a_i) l_0(a_i)\). We conclude that \(h_0(a_i)\) is a unit for \(i = 1, \dots ,r\). Since \(f(a_i) = 0\) we find \(0 = h_0(a_i)g_0(a_i)\) and we conclude that \(g_0(a_i) = 0\) for \(i = 1, \ldots, r\). Thus \((T - a_1)\) divides \(g_0\) in \(A/I[T]\), say \(g_0 = (T - a_1) g_0'\). Set \(f' = (T - a_2) \ldots (T - a_d)\) and \(h'_0 = h_0\). By induction on \(d\) we can lift the factorization \(f' \bmod I = g'_0 h'_0\) to a factorization of \(f' = g' h'\) over over \(A\) which gives the factorization \(f = (T - a_1) g' h'\) lifting the factorization \(f \bmod I = g_0 h_0\) as desired.
Auto-associated rings
Some of this material is in [Autour].
Definition
A ring \(R\) is said to be auto-associated if \(R\) is local and its maximal ideal \(\mathfrak m\) is weakly associated to \(R\).
Lemma
An auto-associated ring \(R\) has the following property: (P) Every proper finitely generated ideal \(I \subset R\) has a nonzero annihilator.
Proof
By assumption there exists a nonzero element \(x \in R\) such that for every \(f \in \mathfrak m\) we have \(f^n x = 0\). Say \(I = (f_1, \ldots, f_r)\). Then \(x\) is in the kernel of \(R \to \bigoplus R_{f_i}\). Hence we see that there exists a nonzero \(y \in R\) such that \(f_i y = 0\) for all \(i\), see Algebra, Lemma 0565. As \(y \in \text{Ann}_R(I)\) we win.
Lemma
Let \(R\) be a ring having property (P) of Lemma 05GN. Let \(u : N \to M\) be a homomorphism of projective \(R\)-modules. Then \(u\) is universally injective if and only if \(u\) is injective.
Proof
Assume \(u\) is injective. Our goal is to show \(u\) is universally injective. First we choose a module \(Q\) such that \(N \oplus Q\) is free. On considering the map \(N \oplus Q \to M \oplus Q\) we see that it suffices to prove the lemma in case \(N\) is free. In this case \(N\) is a directed colimit of finite free \(R\)-modules. Thus we reduce to the case that \(N\) is a finite free \(R\)-module, say \(N = R^{\oplus n}\). We prove the lemma by induction on \(n\). The case \(n = 0\) is trivial.
Let \(u : R^{\oplus n} \to M\) be an injective module map with \(M\) projective. Choose an \(R\)-module \(Q\) such that \(M \oplus Q\) is free. After replacing \(u\) by the composition \(R^{\oplus n} \to M \to M \oplus Q\) we see that we may assume that \(M\) is free. Then we can find a direct summand \(R^{\oplus m} \subset M\) such that \(u(R^{\oplus n}) \subset R^{\oplus m}\). Hence we may assume that \(M = R^{\oplus m}\). In this case \(u\) is given by a matrix \(A = (a_{ij})\) so that \(u(x_1, \ldots, x_n) = (\sum x_i a_{i1}, \ldots, \sum x_i a_{im})\). As \(u\) is injective, in particular \(u(x, 0, \ldots, 0) = (xa_{11}, xa_{12}, \ldots, xa_{1m}) \not = 0\) if \(x \not = 0\), and as \(R\) has property (P) we see that \(a_{11}R + a_{12}R + \ldots + a_{1m}R = R\). Hence see that \(R(a_{11}, \ldots, a_{1m}) \subset R^{\oplus m}\) is a direct summand of \(R^{\oplus m}\), in particular \(R^{\oplus m}/R(a_{11}, \ldots, a_{1m})\) is a projective \(R\)-module. We get a commutative diagram \[\xymatrix{ 0 \ar[r] & R \ar[rr] \ar[d]^1 & & R^{\oplus n} \ar[r] \ar[d]^u & R^{\oplus n - 1} \ar[r] \ar[d] & 0 \\ 0 \ar[r] & R \ar[rr]^{(a_{11}, \ldots, a_{1m})} & & R^{\oplus m} \ar[r] & R^{\oplus m}/R(a_{11}, \ldots, a_{1m}) \ar[r] & 0 }\] with split exact rows. Thus the right vertical arrow is injective and we may apply the induction hypothesis to conclude that the right vertical arrow is universally injective. It follows that the middle vertical arrow is universally injective.
Lemma
Let \(R\) be a ring. The following are equivalent
\(R\) has property (P) of Lemma 05GN,
any injective map of projective \(R\)-modules is universally injective,
if \(u : N \to M\) is injective and \(N\), \(M\) are finite projective \(R\)-modules then \(\Coker(u)\) is a finite projective \(R\)-module,
if \(N \subset M\) and \(N\), \(M\) are finite projective as \(R\)-modules, then \(N\) is a direct summand of \(M\), and
any injective map \(R \to R^{\oplus n}\) is a split injection.
Proof
The implication (1) \(\Rightarrow\) (2) is Lemma 05GP. It is clear that (3) and (4) are equivalent. We have (2) \(\Rightarrow\) (3), (4) by Algebra, Lemma 058L. Part (5) is a special case of (4). Assume (5). Let \(I = (a_1, \ldots, a_n)\) be a proper finitely generated ideal of \(R\). As \(I \not = R\) we see that \(R \to R^{\oplus n}\), \(x \mapsto (xa_1, \ldots, xa_n)\) is not a split injection. Hence it has a nonzero kernel and we conclude that \(\text{Ann}_R(I) \not = 0\). Thus (1) holds.
Example
If the equivalent conditions of Lemma 05GQ hold, then it is not always the case that every injective map of free \(R\)-modules is a split injection. For example suppose that \(R = k[x_1, x_2, x_3, \ldots]/(x_i^2)\). This is an auto-associated ring. Consider the map of free \(R\)-modules \[u : \bigoplus\nolimits_{i \geq 1} Re_i \longrightarrow \bigoplus\nolimits_{i \geq 1} Rf_i, \quad e_i \longmapsto f_i - x_if_{i + 1}.\] For any integer \(n\) the restriction of \(u\) to \(\bigoplus_{i = 1, \ldots, n} Re_i\) is injective as the images \(u(e_1), \ldots, u(e_n)\) are \(R\)-linearly independent. Hence \(u\) is injective and hence universally injective by the lemma. Since \(u \otimes \text{id}_k\) is bijective we see that if \(u\) were a split injection then \(u\) would be surjective. But \(u\) is not surjective because the inverse image of \(f_1\) would be the element \[\sum\nolimits_{i \geq 0} x_1 \ldots x_ie_{i + 1} = e_1 + x_1e_2 + x_1x_2e_3 + \ldots\] which is not an element of the direct sum. A side remark is that \(\Coker(u)\) is a flat (because \(u\) is universally injective), countably generated \(R\)-module which is not projective (as \(u\) is not split), hence not Mittag-Leffler (see Algebra, Lemma 059X).
The following lemma is a special case of Algebra, Proposition 00N1 in case the local ring is Noetherian.
Lemma
Let \((R, \mathfrak m)\) be a local ring. Suppose that \(\varphi : R^m \to R^n\) is a map of finite free modules. The following are equivalent
\(\varphi\) is injective,
the rank of \(\varphi\) is \(m\) and the annihilator of \(I(\varphi)\) in \(R\) is zero.
If \(R\) is Noetherian these are also equivalent to
the rank of \(\varphi\) is \(m\) and either \(I(\varphi) = R\) or it contains a nonzerodivisor.
Here the rank of \(\varphi\) and \(I(\varphi)\) are defined as in Algebra, Definition 00MV.
Proof
If any matrix coefficient of \(\varphi\) is not in \(\mathfrak m\), then we apply Algebra, Lemma 00MT to write \(\varphi\) as the sum of \(1 : R \to R\) and a map \(\varphi' : R^{m-1} \to R^{n-1}\). It is easy to see that the lemma for \(\varphi'\) implies the lemma for \(\varphi\). Thus we may assume from the outset that all the matrix coefficients of \(\varphi\) are in \(\mathfrak m\).
Suppose \(\varphi\) is injective. We may assume \(m > 0\). Let \(\mathfrak q \in \text{WeakAss}(R)\) so that \(R_\mathfrak q\) is an auto-associated ring. Then \(\varphi\) induces an injective map \(R_\mathfrak q^m \to R_\mathfrak q^n\) which is universally injective by Lemmas 05GN and 05GP. Thus \(\varphi : \kappa(\mathfrak q)^m \to \kappa(\mathfrak q)^n\) is injective. Hence the rank of \(\varphi \bmod \mathfrak q\) is \(m\) and \(I(\varphi \otimes \kappa(\mathfrak q))\) is not the zero ideal. Since \(m\) is the maximum rank \(\varphi\) can have, we conclude that \(\varphi\) has rank \(m\) as well (ranks of matrices can only drop after base change). Hence \(I(\varphi) \cdot \kappa(\mathfrak q) = I(\varphi \otimes \kappa(\mathfrak q))\) is not zero. Thus \(I(\varphi)\) is not contained in \(\mathfrak q\). Thus none of the weakly associated primes of \(R\) are weakly associated primes of the \(R\)-module \(\text{Ann}_R I(\varphi)\). Thus \(\text{Ann}_R I(\varphi)\) has no weakly associated primes, see Algebra, Lemma 0548. It follows from Algebra, Lemma 0588 that \(\text{Ann}_R I(\varphi)\) is zero.
Conversely, assume (2). The rank being \(m\) implies \(n \geq m\). Write \(I(\varphi) = (f_1, \ldots, f_r)\) which is possible as \(I(\varphi)\) is finitely generated. By Algebra, Lemma 07DQ we can find maps \(\psi_i : R^n \to R^m\) such that \(\psi \circ \varphi = f_i \text{id}_{R^m}\). Thus \(\varphi(x) = 0\) implies \(f_i x = 0\) for \(i = 1, \ldots, r\). This implies \(x = 0\) and hence \(\varphi\) is injective.
For the equivalence of (1) and (3) in the Noetherian local case we refer to Algebra, Proposition 00N1. If the ring \(R\) is Noetherian but not local, then the reader can deduce it from the local case; details omitted. Another option is to redo the argument above using associated primes, using that there are finitely many of these, using prime avoidance, and using the characterization of nonzerodivisors as elements of a Noetherian ring not contained in any associated prime.
Lemma
Let \(R\) be a ring. Suppose that \(\varphi : R^n \to R^n\) be an injective map of finite free modules of the same rank. Then \(\Hom_R(\Coker(\varphi), R) = 0\).
Proof
Let \(\varphi^t : R^n \to R^n\) be the transpose of \(\varphi\). The lemma claims that \(\varphi^t\) is injective. With notation as in Lemma 00MX we see that the rank of \(\varphi^t\) is \(n\) and that \(I(\varphi) = I(\varphi^t)\). Thus we conclude by the equivalence of (1) and (2) of the lemma.
Flattening stratification
Let \(R \to S\) be a ring map and let \(M\) be an \(S\)-module. For any \(R\)-algebra \(R'\) we can consider the base changes \(S' = S \otimes_R R'\) and \(M' = M \otimes_R R'\). We say \(R \to R'\) flattens \(M\) if the module \(M'\) is flat over \(R'\). We would like to understand the structure of the collection of ring maps \(R \to R'\) which flatten \(M\). In particular we would like to know if there exists a universal flattening \(R \to R_{univ}\) of \(M\), i.e., a ring map \(R \to R_{univ}\) which flattens \(M\) and has the property that any ring map \(R \to R'\) which flattens \(M\) factors through \(R \to R_{univ}\). It turns out that such a universal solution usually does not exist.
We will discuss universal flattenings and flattening stratifications in a scheme theoretic setting \(\mathcal{F}/X/S\) in More on Flatness, Section 052F. If the universal flattening \(R \to R_{univ}\) exists then the morphism of schemes \(\Spec(R_{univ}) \to \Spec(R)\) is the universal flattening of the quasi-coherent module \(\widetilde{M}\) on \(\Spec(S)\).
In this and the next few sections we prove some basic algebra facts related to this. The most basic result is perhaps the following.
Lemma
Let \(R\) be a ring. Let \(M\) be an \(R\)-module. Let \(I_1\), \(I_2\) be ideals of \(R\). If \(M/I_1M\) is flat over \(R/I_1\) and \(M/I_2M\) is flat over \(R/I_2\), then \(M/(I_1 \cap I_2)M\) is flat over \(R/(I_1 \cap I_2)\).
Proof
By replacing \(R\) with \(R/(I_1 \cap I_2)\) and \(M\) by \(M/(I_1 \cap I_2)M\) we may assume that \(I_1 \cap I_2 = 0\). Let \(J \subset R\) be an ideal. To prove that \(M\) is flat over \(R\) we have to show that \(J \otimes_R M \to M\) is injective, see Algebra, Lemma 00HD. By flatness of \(M/I_1M\) over \(R/I_1\) the map \[J/(J \cap I_1) \otimes_R M = (J + I_1)/I_1 \otimes_{R/I_1} M/I_1M \longrightarrow M/I_1M\] is injective. As \(0 \to (J \cap I_1) \to J \to J/(J \cap I_1) \to 0\) is exact we obtain a diagram \[\xymatrix{ (J \cap I_1) \otimes_R M \ar[r] \ar[d] & J \otimes_R M \ar[r] \ar[d] & J/(J \cap I_1) \otimes_R M \ar[r] \ar[d] & 0 \\ M \ar@{=}[r] & M \ar[r] & M/I_1M }\] hence it suffices to show that \((J \cap I_1) \otimes_R M \to M\) is injective. Since \(I_1 \cap I_2 = 0\) the ideal \(J \cap I_1\) maps isomorphically to an ideal \(J' \subset R/I_2\) and we see that \((J \cap I_1) \otimes_R M = J' \otimes_{R/I_2} M/I_2M\). By flatness of \(M/I_2M\) over \(R/I_2\) the map \(J' \otimes_{R/I_2} M/I_2M \to M/I_2M\) is injective, which clearly implies that \((J \cap I_1) \otimes_R M \to M\) is injective.
Flattening over an Artinian ring
A universal flattening exists when the base ring is an Artinian local ring. It exists for an arbitrary module. Hence, as we will see later, a flatting stratification exists when the base scheme is the spectrum of an Artinian local ring.
Lemma
Let \(R\) be an Artinian ring. Let \(M\) be an \(R\)-module. Then there exists a smallest ideal \(I \subset R\) such that \(M/IM\) is flat over \(R/I\).
Proof
This follows directly from Lemma 0522 and the Artinian property.
This ideal has the following universal property.
Lemma
Let \(R\) be an Artinian ring. Let \(M\) be an \(R\)-module. Let \(I \subset R\) be the smallest ideal \(I \subset R\) such that \(M/IM\) is flat over \(R/I\). Then \(I\) has the following universal property: For every ring map \(\varphi : R \to R'\) we have \[R' \otimes_R M\text{ is flat over }R' \Leftrightarrow \text{we have }\varphi(I) = 0.\]
Proof
Note that \(I\) exists by Lemma 0524. The implication \(\Rightarrow\) follows from Algebra, Lemma 00HI. Let \(\varphi : R \to R'\) be such that \(M \otimes_R R'\) is flat over \(R'\). Let \(J = \Ker(\varphi)\). By Algebra, Lemma 051L and as \(R' \otimes_R M = R' \otimes_{R/J} M/JM\) is flat over \(R'\) we conclude that \(M/JM\) is flat over \(R/J\). Hence \(I \subset J\) as desired.
Flattening over a closed subset of the base
Let \(R \to S\) be a ring map. Let \(I \subset R\) be an ideal. Let \(M\) be an \(S\)-module. In the following we will consider the following condition [052W]\[\begin{equation} \forall \mathfrak q \in V(IS) \subset \Spec(S) : M_{\mathfrak q}\text{ is flat over }R. \end{equation}\] Geometrically, this means that \(M\) is flat over \(R\) along the inverse image of \(V(I)\) in \(\Spec(S)\). If \(R\) and \(S\) are Noetherian rings and \(M\) is a finite \(S\)-module, then (052W) is equivalent to the condition that \(M/I^nM\) is flat over \(R/I^n\) for all \(n \geq 1\), see Algebra, Lemma 0523.
Lemma
Let \(R \to S\) be a ring map. Let \(I \subset R\) be an ideal. Let \(M\) be an \(S\)-module. Let \(R \to R'\) be a ring map and \(IR' \subset I' \subset R'\) an ideal. If (052W) holds for \((R \to S, I, M)\), then (052W) holds for \((R' \to S \otimes_R R', I', M \otimes_R R')\).
Proof
Assume (052W) holds for \((R \to S, I \subset R, M)\). Let \(I'(S \otimes_R R') \subset \mathfrak q'\) be a prime of \(S \otimes_R R'\). Let \(\mathfrak q \subset S\) be the corresponding prime of \(S\). Then \(IS \subset \mathfrak q\). Note that \((M \otimes_R R')_{\mathfrak q'}\) is a localization of the base change \(M_{\mathfrak q} \otimes_R R'\). Hence \((M \otimes_R R')_{\mathfrak q'}\) is flat over \(R'\) as a localization of a flat module, see Algebra, Lemmas 00HI and 00HT.
Lemma
Let \(R \to S\) be a ring map. Let \(I \subset R\) be an ideal. Let \(M\) be an \(S\)-module. Let \(R \to R'\) be a ring map and \(IR' \subset I' \subset R'\) an ideal such that
the map \(V(I') \to V(I)\) induced by \(\Spec(R') \to \Spec(R)\) is surjective, and
\(R'_{\mathfrak p'}\) is flat over \(R\) for all primes \(\mathfrak p' \in V(I')\).
If (052W) holds for \((R' \to S \otimes_R R', I', M \otimes_R R')\), then (052W) holds for \((R \to S, I, M)\).
Proof
Assume (052W) holds for \((R' \to S \otimes_R R', IR', M \otimes_R R')\). Pick a prime \(IS \subset \mathfrak q \subset S\). Let \(I \subset \mathfrak p \subset R\) be the corresponding prime of \(R\). By assumption there exists a prime \(\mathfrak p' \in V(I')\) of \(R'\) lying over \(\mathfrak p\) and \(R_{\mathfrak p} \to R'_{\mathfrak p'}\) is flat. Choose a prime \(\overline{\mathfrak q}' \subset \kappa(\mathfrak q) \otimes_{\kappa(\mathfrak p)} \kappa(\mathfrak p')\) which corresponds to a prime \(\mathfrak q' \subset S \otimes_R R'\) which lies over \(\mathfrak q\) and over \(\mathfrak p'\). Note that \((S \otimes_R R')_{\mathfrak q'}\) is a localization of \(S_{\mathfrak q} \otimes_{R_{\mathfrak p}} R'_{\mathfrak p'}\). By assumption the module \((M \otimes_R R')_{\mathfrak q'}\) is flat over \(R'_{\mathfrak p'}\). Hence Algebra, Lemma 00MQ implies that \(M_{\mathfrak q}\) is flat over \(R_{\mathfrak p}\) which is what we wanted to prove.
Lemma
Let \(R \to S\) be a ring map of finite presentation. Let \(M\) be an \(S\)-module of finite presentation. Let \(R' = \colim_{\lambda \in \Lambda} R_\lambda\) be a directed colimit of \(R\)-algebras. Let \(I_\lambda \subset R_\lambda\) be ideals such that \(I_\lambda R_\mu \subset I_\mu\) for all \(\mu \geq \lambda\) and set \(I' = \colim_\lambda I_\lambda\). If (052W) holds for \((R' \to S \otimes_R R', I', M \otimes_R R')\), then there exists a \(\lambda \in \Lambda\) such that (052W) holds for \((R_\lambda \to S \otimes_R R_\lambda, I_\lambda, M \otimes_R R_\lambda)\).
Proof
We are going to write \(S_\lambda = S \otimes_R R_\lambda\), \(S' = S \otimes_R R'\), \(M_\lambda = M \otimes_R R_\lambda\), and \(M' = M \otimes_R R'\). The base change \(S'\) is of finite presentation over \(R'\) and \(M'\) is of finite presentation over \(S'\) and similarly for the versions with subscript \(\lambda\), see Algebra, Lemma 05G5. By Algebra, Theorem 00RC the set \[U' = \{\mathfrak q' \in \Spec(S') \mid M'_{\mathfrak q'}\text{ is flat over }R'\}\] is open in \(\Spec(S')\). Note that \(V(I'S')\) is a quasi-compact space which is contained in \(U'\) by assumption. Hence there exist finitely many \(g'_j \in S'\), \(j = 1, \ldots, m\) such that \(D(g'_j) \subset U'\) and such that \(V(I'S') \subset \bigcup D(g'_j)\). Note that in particular \((M')_{g'_j}\) is a flat module over \(R'\).
We are going to pick increasingly large elements \(\lambda \in \Lambda\). First we pick it large enough so that we can find \(g_{j, \lambda} \in S_{\lambda}\) mapping to \(g'_j\). The inclusion \(V(I'S') \subset \bigcup D(g'_j)\) means that \(I'S' + (g'_1, \ldots, g'_m) = S'\) which can be expressed as \(1 = \sum z_sh_s + \sum f_jg'_j\) for some \(z_s \in I'\), \(h_s, f_j \in S'\). After increasing \(\lambda\) we may assume such an equation holds in \(S_\lambda\). Hence we may assume that \(V(I_\lambda S_\lambda) \subset \bigcup D(g_{j, \lambda})\). By Algebra, Lemma 02JO we see that for some sufficiently large \(\lambda\) the modules \((M_\lambda)_{g_{j, \lambda}}\) are flat over \(R_\lambda\). In particular the module \(M_\lambda\) is flat over \(R_\lambda\) at all the primes lying over the ideal \(I_\lambda\).
Flattening over a closed subsets of source and base
In this section we slightly generalize the discussion in Section 05LK. We strongly suggest the reader first read and understand that section.
Situation
Let \(R \to S\) be a ring map. Let \(J \subset S\) be an ideal. Let \(M\) be an \(S\)-module.
In this situation, given an \(R\)-algebra \(R'\) and an ideal \(I' \subset R'\) we set \(S' = S \otimes_R R'\) and \(M' = M \otimes_R R'\). We will consider the condition [05LQ]\[\begin{equation} \forall \mathfrak q' \in V(I'S' + JS') \subset \Spec(S') : M'_{\mathfrak q'}\text{ is flat over }R'. \end{equation}\] Geometrically, this means that \(M'\) is flat over \(R'\) along the intersection of the inverse image of \(V(I')\) with the inverse image of \(V(J)\). Since \((R \to S, J, M)\) are fixed, condition (05LQ) only depends on the pair \((R', I')\) where \(R'\) is viewed as an \(R\)-algebra.
Lemma
In Situation 05LP let \(R' \to R''\) be an \(R\)-algebra map. Let \(I' \subset R'\) and \(I'R'' \subset I'' \subset R''\) be ideals. If (05LQ) holds for \((R', I')\), then (05LQ) holds for \((R'', I'')\).
Proof
Assume (05LQ) holds for \((R', I')\). Let \(I''S'' + JS'' \subset \mathfrak q''\) be a prime of \(S''\). Let \(\mathfrak q' \subset S'\) be the corresponding prime of \(S'\). Then both \(I'S' \subset \mathfrak q'\) and \(JS' \subset \mathfrak q'\) because the corresponding conditions hold for \(\mathfrak q''\). Note that \((M'')_{\mathfrak q''}\) is a localization of the base change \(M'_{\mathfrak q'} \otimes_R R''\). Hence \((M'')_{\mathfrak q''}\) is flat over \(R''\) as a localization of a flat module, see Algebra, Lemmas 00HI and 00HT.
Lemma
In Situation 05LP let \(R' \to R''\) be an \(R\)-algebra map. Let \(I' \subset R'\) and \(I'R'' \subset I'' \subset R''\) be ideals. Assume
the map \(V(I'') \to V(I')\) induced by \(\Spec(R'') \to \Spec(R')\) is surjective, and
\(R''_{\mathfrak p''}\) is flat over \(R'\) for all primes \(\mathfrak p'' \in V(I'')\).
If (05LQ) holds for \((R'', I'')\), then (05LQ) holds for \((R', I')\).
Proof
Assume (05LQ) holds for \((R'', I'')\). Pick a prime \(I'S' + JS' \subset \mathfrak q' \subset S'\). Let \(I' \subset \mathfrak p' \subset R'\) be the corresponding prime of \(R'\). By assumption there exists a prime \(\mathfrak p'' \in V(I'')\) of \(R''\) lying over \(\mathfrak p'\) and \(R'_{\mathfrak p'} \to R''_{\mathfrak p''}\) is flat. Choose a prime \(\overline{\mathfrak q}'' \subset \kappa(\mathfrak q') \otimes_{\kappa(\mathfrak p')} \kappa(\mathfrak p'')\). This corresponds to a prime \(\mathfrak q'' \subset S'' = S' \otimes_{R'} R''\) which lies over \(\mathfrak q'\) and over \(\mathfrak p''\). In particular we see that \(I''S'' \subset \mathfrak q''\) and that \(JS'' \subset \mathfrak q''\). Note that \((S' \otimes_{R'} R'')_{\mathfrak q''}\) is a localization of \(S'_{\mathfrak q'} \otimes_{R'_{\mathfrak p'}} R''_{\mathfrak p''}\). By assumption the module \((M' \otimes_{R'} R'')_{\mathfrak q''}\) is flat over \(R''_{\mathfrak p''}\). Hence Algebra, Lemma 00MQ implies that \(M'_{\mathfrak q'}\) is flat over \(R'_{\mathfrak p'}\) which is what we wanted to prove.
Lemma
In Situation 05LP assume \(R \to S\) is essentially of finite presentation and \(M\) is an \(S\)-module of finite presentation. Let \(R' = \colim_{\lambda \in \Lambda} R_\lambda\) be a directed colimit of \(R\)-algebras. Let \(I_\lambda \subset R_\lambda\) be ideals such that \(I_\lambda R_\mu \subset I_\mu\) for all \(\mu \geq \lambda\) and set \(I' = \colim_\lambda I_\lambda\). If (05LQ) holds for \((R', I')\), then there exists a \(\lambda \in \Lambda\) such that (05LQ) holds for \((R_\lambda, I_\lambda)\).
Proof
We first prove the lemma in case \(R \to S\) is of finite presentation and then we explain what needs to be changed in the general case. We are going to write \(S_\lambda = S \otimes_R R_\lambda\), \(S' = S \otimes_R R'\), \(M_\lambda = M \otimes_R R_\lambda\), and \(M' = M \otimes_R R'\). The base change \(S'\) is of finite presentation over \(R'\) and \(M'\) is of finite presentation over \(S'\) and similarly for the versions with subscript \(\lambda\), see Algebra, Lemma 05G5. By Algebra, Theorem 00RC the set \[U' = \{\mathfrak q' \in \Spec(S') \mid M'_{\mathfrak q'}\text{ is flat over }R'\}\] is open in \(\Spec(S')\). Note that \(V(I'S' + JS')\) is a quasi-compact space which is contained in \(U'\) by assumption. Hence there exist finitely many \(g'_j \in S'\), \(j = 1, \ldots, m\) such that \(D(g'_j) \subset U'\) and such that \(V(I'S' + JS') \subset \bigcup D(g'_j)\). Note that in particular \((M')_{g'_j}\) is a flat module over \(R'\).
We are going to pick increasingly large elements \(\lambda \in \Lambda\). First we pick it large enough so that we can find \(g_{j, \lambda} \in S_{\lambda}\) mapping to \(g'_j\). The inclusion \(V(I'S' + JS') \subset \bigcup D(g'_j)\) means that \(I'S' + JS' + (g'_1, \ldots, g'_m) = S'\) which can be expressed as \[1 = \sum y_tk_t + \sum z_sh_s + \sum f_jg'_j\] for some \(z_s \in I'\), \(y_t \in J\), \(k_t, h_s, f_j \in S'\). After increasing \(\lambda\) we may assume such an equation holds in \(S_\lambda\). Hence we may assume that \(V(I_\lambda S_\lambda + J S_\lambda) \subset \bigcup D(g_{j, \lambda})\). By Algebra, Lemma 02JO we see that for some sufficiently large \(\lambda\) the modules \((M_\lambda)_{g_{j, \lambda}}\) are flat over \(R_\lambda\). In particular the module \(M_\lambda\) is flat over \(R_\lambda\) at all the primes corresponding to points of \(V(I_\lambda S_\lambda + J S_\lambda)\).
In the case that \(S\) is essentially of finite presentation, we can write \(S = \Sigma^{-1}C\) where \(R \to C\) is of finite presentation and \(\Sigma \subset C\) is a multiplicative subset. We can also write \(M = \Sigma^{-1}N\) for some finitely presented \(C\)-module \(N\), see Algebra, Lemma 05N5. At this point we introduce \(C_\lambda\), \(C'\), \(N_\lambda\), \(N'\). Then in the discussion above we obtain an open \(U' \subset \Spec(C')\) over which \(N'\) is flat over \(R'\). The assumption that (05LQ) is true means that \(V(I'S' + JS')\) maps into \(U'\), because for a prime \(\mathfrak q' \subset S'\), corresponding to a prime \(\mathfrak r' \subset C'\) we have \(M'_{\mathfrak q'} = N'_{\mathfrak r'}\). Thus we can find \(g'_j \in C'\) such that \(\bigcup D(g'_j)\) contains the image of \(V(I'S' + JS')\). The rest of the proof is exactly the same as before.
Lemma
In Situation 05LP. Let \(I \subset R\) be an ideal. Assume
\(R\) is a Noetherian ring,
\(S\) is a Noetherian ring,
\(M\) is a finite \(S\)-module, and
for each \(n \geq 1\) and any prime \(\mathfrak q \in V(J + IS)\) the module \((M/I^n M)_{\mathfrak q}\) is flat over \(R/I^n\).
Then (05LQ) holds for \((R, I)\), i.e., for every prime \(\mathfrak q \in V(J + IS)\) the localization \(M_{\mathfrak q}\) is flat over \(R\).
Proof
Let \(\mathfrak q \in V(J + IS)\). Then Algebra, Lemma 0523 applied to \(R \to S_{\mathfrak q}\) and \(M_{\mathfrak q}\) implies that \(M_{\mathfrak q}\) is flat over \(R\).
Flattening over a Noetherian complete local ring
The following three lemmas give a completely algebraic proof of the existence of the “local” flattening stratification when the base is a complete local Noetherian ring \(R\) and the given module is finite over a finite type \(R\)-algebra \(S\).
Lemma
Let \(R \to S\) be a ring map. Let \(M\) be an \(S\)-module. Assume
\((R, \mathfrak m)\) is a complete local Noetherian ring,
\(S\) is a Noetherian ring, and
\(M\) is finite over \(S\).
Then there exists an ideal \(I \subset \mathfrak m\) such that
\((M/IM)_{\mathfrak q}\) is flat over \(R/I\) for all primes \(\mathfrak q\) of \(S/IS\) lying over \(\mathfrak m\), and
if \(J \subset R\) is an ideal such that \((M/JM)_{\mathfrak q}\) is flat over \(R/J\) for all primes \(\mathfrak q\) lying over \(\mathfrak m\), then \(I \subset J\).
In other words, \(I\) is the smallest ideal of \(R\) such that (052W) holds for \((\overline{R} \to \overline{S}, \overline{\mathfrak m}, \overline{M})\) where \(\overline{R} = R/I\), \(\overline{S} = S/IS\), \(\overline{\mathfrak m} = \mathfrak m/I\) and \(\overline{M} = M/IM\).
Proof
Let \(J \subset R\) be an ideal. Apply Algebra, Lemma 0523 to the module \(M/JM\) over the ring \(R/J\). Then we see that \((M/JM)_{\mathfrak q}\) is flat over \(R/J\) for all primes \(\mathfrak q\) of \(S/JS\) if and only if \(M/(J + \mathfrak m^n)M\) is flat over \(R/(J + \mathfrak m^n)\) for all \(n \geq 1\). We will use this remark below.
For every \(n \geq 1\) the local ring \(R/\mathfrak m^n\) is Artinian. Hence, by Lemma 0524 there exists a smallest ideal \(I_n \supset \mathfrak m^n\) such that \(M/I_nM\) is flat over \(R/I_n\). It is clear that \(I_{n + 1} + \mathfrak m^n\) is contains \(I_n\) and applying Lemma 0522 we see that \(I_n = I_{n + 1} + \mathfrak m^n\). Since \(R = \lim_n\ R/\mathfrak m^n\) we see that \(I = \lim_n\ I_n/\mathfrak m^n\) is an ideal in \(R\) such that \(I_n = I + \mathfrak m^n\) for all \(n \geq 1\). By the initial remarks of the proof we see that \(I\) verifies (1) and (2). Some details omitted.
Lemma
With notation \(R \to S\), \(M\), and \(I\) and assumptions as in Lemma 0526. Consider a local homomorphism of local rings \(\varphi : (R, \mathfrak m) \to (R', \mathfrak m')\) such that \(R'\) is Noetherian. Then the following are equivalent
condition (052W) holds for \((R' \to S \otimes_R R', \mathfrak m', M \otimes_R R')\), and
\(\varphi(I) = 0\).
Proof
The implication (2) \(\Rightarrow\) (1) follows from Lemma 052X. Let \(\varphi : R \to R'\) be as in the lemma satisfying (1). We have to show that \(\varphi(I) = 0\). This is equivalent to the condition that \(\varphi(I)R' = 0\). By Artin-Rees in the Noetherian local ring \(R'\) (see Algebra, Lemma 00IP) this is equivalent to the condition that \(\varphi(I)R' + (\mathfrak m')^n = (\mathfrak m')^n\) for all \(n > 0\). Hence this is equivalent to the condition that the composition \(\varphi_n : R \to R' \to R'/(\mathfrak m')^n\) annihilates \(I\) for each \(n\). Now assumption (1) for \(\varphi\) implies assumption (1) for \(\varphi_n\) by Lemma 052X. This reduces us to the case where \(R'\) is Artinian local.
Assume \(R'\) Artinian. Let \(J = \Ker(\varphi)\). We have to show that \(I \subset J\). By the construction of \(I\) in Lemma 0526 it suffices to show that \((M/JM)_{\mathfrak q}\) is flat over \(R/J\) for every prime \(\mathfrak q\) of \(S/JS\) lying over \(\mathfrak m\). As \(R'\) is Artinian, condition (1) signifies that \(M \otimes_R R'\) is flat over \(R'\). As \(R'\) is Artinian and \(R/J \to R'\) is a local injective ring map, it follows that \(R/J\) is Artinian too. Hence the flatness of \(M \otimes_R R' = M/JM \otimes_{R/J} R'\) over \(R'\) implies that \(M/JM\) is flat over \(R/J\) by Algebra, Lemma 051L. This concludes the proof.
Lemma
With notation \(R \to S\), \(M\), and \(I\) and assumptions as in Lemma 0526. In addition assume that \(R \to S\) is of finite type. Then for any local homomorphism of local rings \(\varphi : (R, \mathfrak m) \to (R', \mathfrak m')\) the following are equivalent
condition (052W) holds for \((R' \to S \otimes_R R', \mathfrak m', M \otimes_R R')\), and
\(\varphi(I) = 0\).
Proof
The implication (2) \(\Rightarrow\) (1) follows from Lemma 052X. Let \(\varphi : R \to R'\) be as in the lemma satisfying (1). As \(R\) is Noetherian we see that \(R \to S\) is of finite presentation and \(M\) is an \(S\)-module of finite presentation. Write \(R' = \colim_\lambda R_\lambda\) as a directed colimit of local \(R\)-subalgebras \(R_\lambda \subset R'\), with maximal ideals \(\mathfrak m_\lambda = R_\lambda \cap \mathfrak m'\) such that each \(R_\lambda\) is essentially of finite type over \(R\). By Lemma 05LM we see that condition (052W) holds for \((R_\lambda \to S \otimes_R R_\lambda, \mathfrak m_\lambda, M \otimes_R R_\lambda)\) for some \(\lambda\). Hence Lemma 0527 applies to the ring map \(R \to R_\lambda\) and we see that \(I\) maps to zero in \(R_\lambda\), a fortiori it maps to zero in \(R'\).
Descent of flatness along integral maps
First a few simple lemmas.
Lemma
Let \(R\) be a ring. Let \(P(T)\) be a monic polynomial with coefficients in \(R\). Let \(\alpha \in R\) be such that \(P(\alpha) = 0\). Then \(P(T) = (T - \alpha)Q(T)\) for some monic polynomial \(Q(T) \in R[T]\).
Proof
By induction on the degree of \(P\). If \(\deg(P) = 1\), then \(P(T) = T - \alpha\) and the result is true. If \(\deg(P) > 1\), then we can write \(P(T) = (T - \alpha)Q(T) + r\) for some polynomial \(Q \in R[T]\) of degree \(< \deg(P)\) and some \(r \in R\) by long division. By assumption \(0 = P(\alpha) = (\alpha - \alpha)Q(\alpha) + r = r\) and we conclude that \(r = 0\) as desired.
Lemma
Let \(R\) be a ring. Let \(P(T)\) be a monic polynomial with coefficients in \(R\). There exists a finite free ring map \(R \to R'\) such that \(P(T) = (T - \alpha)Q(T)\) for some \(\alpha \in R'\) and some monic polynomial \(Q(T) \in R'[T]\).
Proof
Write \(P(T) = T^d + a_1T^{d - 1} + \ldots + a_0\). Set \(R' = R[x]/(x^d + a_1x^{d - 1} + \ldots + a_0)\). Set \(\alpha\) equal to the congruence class of \(x\). Then it is clear that \(P(\alpha) = 0\). Thus we win by Lemma 052Z.
Lemma
Let \(R \to S\) be a finite ring map. There exists a finite free ring extension \(R \subset R'\) such that \(S \otimes_R R'\) is a quotient of a ring of the form \[R'[T_1, \ldots, T_n]/(P_1(T_1), \ldots, P_n(T_n))\] with \(P_i(T) = \prod_{j = 1, \ldots, d_i} (T - \alpha_{ij})\) for some \(\alpha_{ij} \in R'\).
Proof
Let \(x_1, \ldots, x_n \in S\) be generators of \(S\) over \(R\). For each \(i\) we can choose a monic polynomial \(P_i(T) \in R[T]\) such that \(P_i(x_i) = 0\) in \(S\), see Algebra, Lemma 00GK. Say \(\deg(P_i) = d_i\). By Lemma 0530 (applied \(\sum d_i\) times) there exists a finite free ring extension \(R \subset R'\) such that each \(P_i\) splits completely: \[P_i(T) = \prod\nolimits_{j = 1, \ldots, d_i} (T - \alpha_{ij})\] for certain \(\alpha_{ik} \in R'\). Let \(R'[T_1, \ldots, T_n] \to S \otimes_R R'\) be the \(R'\)-algebra map which maps \(T_i\) to \(x_i \otimes 1\). As this maps \(P_i(T_i)\) to zero, this induces the desired surjection.
Lemma
Let \(R\) be a ring. Let \(S = R[T_1, \ldots, T_n]/J\). Assume \(J\) contains elements of the form \(P_i(T_i)\) with \(P_i(T) = \prod_{j = 1, \ldots, d_i} (T - \alpha_{ij})\) for some \(\alpha_{ij} \in R\). For \(\underline{k} = (k_1, \ldots, k_n)\) with \(1 \leq k_i \leq d_i\) consider the ring map \[\Phi_{\underline{k}} : R[T_1, \ldots, T_n] \to R, \quad T_i \longmapsto \alpha_{ik_i}\] Set \(J_{\underline{k}} = \Phi_{\underline{k}}(J)\). Then the image of \(\Spec(S) \to \Spec(R)\) is equal to \(V(\bigcap J_{\underline{k}})\).
Proof
This lemma proves itself. Hint: \(V(\bigcap J_{\underline{k}}) = \bigcup V(J_{\underline{k}})\).
The following result is due to Ferrand, see [Ferrand].
Lemma
Let \(R \to S\) be a finite injective homomorphism of Noetherian rings. Let \(M\) be an \(R\)-module. If \(M \otimes_R S\) is a flat \(S\)-module, then \(M\) is a flat \(R\)-module.
Proof
Let \(M\) be an \(R\)-module such that \(M \otimes_R S\) is flat over \(S\). By Algebra, Lemma 00HJ in order to prove that \(M\) is flat we may replace \(R\) by any faithfully flat ring extension. By Lemma 0531 we can find a finite locally free ring extension \(R \subset R'\) such that \(S' = S \otimes_R R' = R'[T_1, \ldots, T_n]/J\) for some ideal \(J \subset R'[T_1, \ldots, T_n]\) which contains the elements of the form \(P_i(T_i)\) with \(P_i(T) = \prod_{j = 1, \ldots, d_i} (T - \alpha_{ij})\) for some \(\alpha_{ij} \in R'\). Note that \(R'\) is Noetherian and that \(R' \subset S'\) is a finite extension of rings. Hence we may replace \(R\) by \(R'\) and assume that \(S\) has a presentation as in Lemma 0532. Note that \(\Spec(S) \to \Spec(R)\) is surjective, see Algebra, Lemma 00GQ. Thus, using Lemma 0532 we conclude that \(I = \bigcap J_{\underline{k}}\) is an ideal such that \(V(I) = \Spec(R)\). This means that \(I \subset \sqrt{(0)}\), and since \(R\) is Noetherian that \(I\) is nilpotent. The maps \(\Phi_{\underline{k}}\) induce commutative diagrams \[\xymatrix{ S \ar[rr] & & R/J_{\underline{k}} \\ & R \ar[lu] \ar[ru] }\] from which we conclude that \(M/J_{\underline{k}}M\) is flat over \(R/J_{\underline{k}}\). By Lemma 0522 we see that \(M/IM\) is flat over \(R/I\). Finally, applying Algebra, Lemma 051J we conclude that \(M\) is flat over \(R\).
Lemma
Let \(R \to S\) be an injective integral ring map. Let \(M\) be a finitely presented module over \(R[x_1, \ldots, x_n]\). If \(M \otimes_R S\) is flat over \(S\), then \(M\) is flat over \(R\).
Proof
Choose a presentation \[R[x_1, \ldots, x_n]^{\oplus t} \to R[x_1, \ldots, x_n]^{\oplus r} \to M \to 0.\] Let’s say that the first map is given by the \(r \times t\)-matrix \(T = (f_{ij})\) with \(f_{ij} \in R[x_1, \ldots, x_n]\). Write \(f_{ij} = \sum f_{ij, I} x^I\) with \(f_{ij, I} \in R\) (multi-index notation). Consider diagrams \[\xymatrix{ R \ar[r] & S \\ R_\lambda \ar[u] \ar[r] & S_\lambda \ar[u] }\] where \(R_\lambda\) is a finitely generated \(\mathbf{Z}\)-subalgebra of \(R\) containing all \(f_{ij, I}\) and \(S_\lambda\) is a finite \(R_\lambda\)-subalgebra of \(S\). Let \(M_\lambda\) be the finite \(R_\lambda[x_1, \ldots, x_n]\)-module defined by a presentation as above, using the same matrix \(T\) but now viewed as a matrix over \(R_\lambda[x_1, \ldots, x_n]\). Note that \(S\) is the directed colimit of the \(S_\lambda\) (details omitted). By Algebra, Lemma 02JO we see that for some \(\lambda\) the module \(M_\lambda \otimes_{R_\lambda} S_\lambda\) is flat over \(S_\lambda\). By Lemma 0533 we conclude that \(M_\lambda\) is flat over \(R_\lambda\). Since \(M = M_\lambda \otimes_{R_\lambda} R\) we win by Algebra, Lemma 00HI.
Lemma
Let \(R \to S\) be a finite injective homomorphism of Noetherian rings. Let \(P\) be an \(R\)-module. If \(P \otimes_R S\) is a projective \(S\)-module, then \(P\) is a projective \(R\)-module.
Proof
Let \(P\) be an \(R\)-module such that \(P \otimes_R S\) is projective over \(S\). By Algebra, Theorem 05A9 in order to prove that \(P\) is projective we may replace \(R\) by any faithfully flat ring extension. By Lemma 0531 we can find a finite locally free ring extension \(R \subset R'\) such that \(S' = S \otimes_R R' = R'[T_1, \ldots, T_n]/J\) for some ideal \(J \subset R'[T_1, \ldots, T_n]\) which contains the elements of the form \(P_i(T_i)\) with \(P_i(T) = \prod_{j = 1, \ldots, d_i} (T - \alpha_{ij})\) for some \(\alpha_{ij} \in R'\). Note that \(R'\) is Noetherian and that \(R' \subset S'\) is a finite extension of rings. Hence we may replace \(R\) by \(R'\) and assume that \(S\) has a presentation as in Lemma 0532. Note that \(\Spec(S) \to \Spec(R)\) is surjective, see Algebra, Lemma 00GQ. Thus, using Lemma 0532 we conclude that \(I = \bigcap J_{\underline{k}}\) is an ideal such that \(V(I) = \Spec(R)\). This means that \(I \subset \sqrt{(0)}\), and since \(R\) is Noetherian that \(I\) is nilpotent. The maps \(\Phi_{\underline{k}}\) induce commutative diagrams \[\xymatrix{ S \ar[rr] & & R/J_{\underline{k}} \\ & R \ar[lu] \ar[ru] }\] from which we conclude that \(P/J_{\underline{k}}P\) is projective over \(R/J_{\underline{k}}\). By Algebra, Lemma 0H7M we see that \(P/IP\) is projective over \(R/I\). Since \(P\) is flat over \(R\) by Lemma 0533, we may apply Algebra, Lemma 05CG to conclude that \(P\) is projective over \(R\).
Torsion free modules
In this section we discuss torsion free modules and the relationship with flatness (especially over dimension 1 rings).
Definition
Let \(R\) be a domain. Let \(M\) be an \(R\)-module.
We say an element \(x \in M\) is torsion if there exists a nonzero \(f \in R\) such that \(fx = 0\).
We say \(M\) is torsion free if the only torsion element of \(M\) is \(0\).
We say \(M\) is a torsion module if every element of \(M\) is torsion.
Let \(R\) be a domain and let \(S = R \setminus \{0\}\) be the multiplicative set of nonzero elements of \(R\). Then an \(R\)-module \(M\) is torsion free if and only if \(M \to S^{-1}M\) is injective. In other words, if and only if the map \(M \to M \otimes_R K\) is injective where \(K = S^{-1}R\) is the fraction field of \(R\).
Lemma
Let \(R\) be a domain. Let \(M\) be an \(R\)-module. The set of torsion elements \(M_{tors}\) of \(M\) is the kernel of the map \(M \to M \otimes_R K\). Thus \(M_{tors}\) is an \(R\)-submodule of \(M\). The quotient module \(M/M_{tors}\) is torsion free.
Proof
See discussion above.
Lemma
Let \(R\) be a domain. Let \(M\) be a torsion free \(R\)-module. For any multiplicative set \(S \subset R\) the module \(S^{-1}M\) is a torsion free \(S^{-1}R\)-module.
Proof
Omitted.
Lemma
Let \(R \to R'\) be a flat homomorphism of domains. If \(M\) is a torsion free \(R\)-module, then \(M \otimes_R R'\) is a torsion free \(R'\)-module.
Proof
If \(M\) is torsion free, then \(M \subset M \otimes_R K\) is injective where \(K\) is the fraction field of \(R\). Since \(R'\) is flat over \(R\) we see that \(M \otimes_R R' \to (M \otimes_R K) \otimes_R R'\) is injective. Since \(M \otimes_R K\) is isomorphic to a direct sum of copies of \(K\), it suffices to see that \(K \otimes_R R'\) is torsion free. This is true because it is a localization of \(R'\).
Lemma
Let \(R\) be a domain. Let \(0 \to M \to M' \to M'' \to 0\) be a short exact sequence of \(R\)-modules. If \(M\) and \(M''\) are torsion free, then \(M'\) is torsion free.
Proof
Omitted.
Lemma
Let \(R\) be a domain. Let \(M\) be an \(R\)-module. Then \(M\) is torsion free if and only if \(M_\mathfrak m\) is a torsion free \(R_\mathfrak m\)-module for all maximal ideals \(\mathfrak m\) of \(R\).
Proof
Lemma
Let \(R\) be a domain. Let \(M\) be a finite \(R\)-module. Then \(M\) is torsion free if and only if \(M\) is a submodule of a finite free module.
Proof
If \(M\) is a submodule of \(R^{\oplus n}\), then \(M\) is torsion free. For the converse, assume \(M\) is torsion free. Let \(K\) be the fraction field of \(R\). Then \(M \otimes_R K\) is a finite dimensional \(K\)-vector space. Choose a basis \(e_1, \ldots, e_r\) for this vector space. Let \(x_1, \ldots, x_n\) be generators of \(M\). Write \(x_i = \sum (a_{ij}/b_{ij}) e_j\) for some \(a_{ij}, b_{ij} \in R\) with \(b_{ij} \not = 0\). Set \(b = \prod_{i, j} b_{ij}\). Since \(M\) is torsion free the map \(M \to M \otimes_R K\) is injective and the image is contained in \(R^{\oplus r} = R e_1/b \oplus \ldots \oplus Re_r/b\).
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a nonzero finite \(R\)-module. The following are equivalent
\(M\) is torsion free,
\(M\) is a submodule of a finite free module,
\((0)\) is the only associated prime of \(M\),
\((0)\) is in the support of \(M\) and \(M\) has property \((S_1)\), and
\((0)\) is in the support of \(M\) and \(M\) has no embedded associated prime.
Proof
We have seen the equivalence of (1) and (2) in Lemma 0AUU. We have seen the equivalence of (4) and (5) in Algebra, Lemma 031Q. The equivalence between (3) and (5) is immediate from the definition. A localization of a torsion free module is torsion free (Lemma 0AUR), hence it is clear that a \(M\) has no associated primes different from \((0)\). Thus (1) implies (5). Conversely, assume (5). If \(M\) has torsion, then there exists an embedding \(R/I \subset M\) for some nonzero ideal \(I\) of \(R\). Hence \(M\) has an associated prime different from \((0)\) (see Algebra, Lemmas 02M3 and 0587). This is an embedded associated prime which contradicts the assumption.
Lemma
Let \(R\) be a domain. Any flat \(R\)-module is torsion free.
Proof
If \(x \in R\) is nonzero, then \(x : R \to R\) is injective, and hence if \(M\) is flat over \(R\), then \(x : M \to M\) is injective. Thus if \(M\) is flat over \(R\), then \(M\) is torsion free.
Lemma
Let \(A\) be a valuation ring. An \(A\)-module \(M\) is flat over \(A\) if and only if \(M\) is torsion free.
Proof
The implication “flat \(\Rightarrow\) torsion free” is Lemma 0538. For the converse, assume \(M\) is torsion free. By the equational criterion of flatness (see Algebra, Lemma 00HK) we have to show that every relation in \(M\) is trivial. To do this assume that \(\sum_{i = 1, \ldots, n} a_i x_i = 0\) with \(x_i \in M\) and \(a_i \in A\). After renumbering we may assume that \(v(a_1) \leq v(a_i)\) for all \(i\). Hence we can write \(a_i = a'_i a_1\) for some \(a'_i \in A\). Note that \(a'_1 = 1\). As \(M\) is torsion free we see that \(x_1 = - \sum_{i \geq 2} a'_i x_i\). Thus, if we choose \(y_i = x_i\), \(i = 2, \ldots, n\) then \[x_1 = \sum\nolimits_{j \geq 2} -a'_j y_j, \quad x_i = y_i, (i \geq 2)\quad 0 = a_1 \cdot (-a'_j) + a_j \cdot 1 (j \geq 2)\] shows that the relation was trivial (to be explicit the elements \(a_{ij}\) are defined by setting \(a_{11} = 0\), \(a_{1j} = -a'_j\) for \(j > 1\), and \(a_{ij} = \delta_{ij}\) for \(i, j \geq 2\)).
Lemma
Let \(A\) be a Dedekind domain (for example a discrete valuation ring or more generally a PID).
An \(A\)-module is flat if and only if it is torsion free.
A finite torsion free \(A\)-module is finite locally free.
A finite torsion free \(A\)-module is finite free if \(A\) is a PID.
Proof
(For the parenthetical remark in the statement of the lemma, see Algebra, Lemma 0AUQ.) Proof of (1). By Lemma 0AUT and Algebra, Lemma 00HT it suffices to check the statement over \(A_\mathfrak m\) for \(\mathfrak m \subset A\) maximal. Since \(A_\mathfrak m\) is a discrete valuation ring (Algebra, Lemma 034X) we win by Lemma 0539.
Proof of (2). Follows from Algebra, Lemma 00NX and (1).
Proof of (3). Let \(A\) be a PID and let \(M\) be a finite torsion free module. By Lemma 0AUU we see that \(M \subset A^{\oplus n}\) for some \(n\). We argue that \(M\) is free by induction on \(n\). The case \(n = 1\) expresses exactly the fact that \(A\) is a PID. If \(n > 1\) let \(M' \subset A^{\oplus n - 1}\) be the image of the projection onto the last \(n - 1\) summands of \(A^{\oplus n}\). Then we obtain a short exact sequence \(0 \to I \to M \to M' \to 0\) where \(I\) is the intersection of \(M\) with the first summand \(A\) of \(A^{\oplus n}\). By induction we see that \(M\) is an extension of finite free \(A\)-modules, whence finite free.
Lemma
Let \(R\) be a domain. Let \(M\), \(N\) be \(R\)-modules. If \(N\) is torsion free, so is \(\Hom_R(M, N)\).
Proof
Choose a surjection \(\bigoplus_{i \in I} R \to M\). Then \(\Hom_R(M, N) \subset \prod_{i \in I} N\).
Ranks of modules
Here is our definition.
Definition
Let \(R\) be a domain with fraction field \(K\). The rank of an \(R\)-module \(M\) is \(\dim_K(M \otimes_R K)\).
This definition does not conflict with the notion of a locally free module of rank \(r\) in Algebra, Definition 00NW when both definitions apply.
Lemma
Let \(R\) be a domain. If \(M \to M'\) is a map of \(R\)-modules whose kernel and cokernel are torsion, then the rank of \(M\) equals the rank of \(M'\).
Proof
Omitted. Hint: the induced map \(M \otimes_R K \to M' \otimes_R K\) is an isomorphism if \(K\) is the fraction field of \(R\).
Lemma
Let \(R\) be a domain. Let \(0 \to M \to M' \to M'' \to 0\) be a short exact sequence of \(R\)-modules. Then the rank of \(M'\) is the sum of the ranks of \(M\) and \(M''\).
Proof
Omitted.
Lemma
Let \(R\) be a domain. Let \(M\) and \(N\) be \(R\)-modules.
The rank of \(M \otimes_R N\) is the product of the ranks of \(M\) and \(N\).
If \(M\) is a finitely presented \(R\)-module, then the rank of \(\Hom_R(M, N)\) is the product of the ranks of \(M\) and \(N\).
Proof
Part (1) follows from the same fact for vector spaces. Assume \(M\) is finitely presented as an \(R\)-module. Then \(\Hom_R(M, N) \otimes_R K\) is isomorphic to \(\Hom_K(M \otimes_R K, N \otimes_R K)\) as a \(K\)-vector space by Algebra, Lemma 0583. Since also \(M \otimes_R K\) is finite dimensional, we conclude from the corresponding fact for vector spaces.
Lemma
Let \(R \subset R'\) be an extension of domains. If \(M\) is an \(R\)-module, then the rank of \(M\) over \(R\) is equal to the rank of \(M \otimes_R R'\) over \(R'\).
Proof
This is true because the dimension of a vector space is invariant under extension of ground field.
Lemma
Let \(R\) be a domain. Let \(M\) be a finite \(R\)-module. Then
\(M\) has finite rank \(r \geq 0\),
there is a map \(M \to R^{\oplus r}\) whose kernel and cokernel are torsion modules,
there is an \(f \in R\), \(f \not = 0\) such that \(M_f\) is free of rank \(r\), and
there is an injective map \(R^{\oplus r} \to M\) whose cokernel is a torsion module.
Proof
Since \(M\) is finite, \(M \otimes_R K\) is a finite dimensional vector space over the fraction field \(K\) of \(R\). Thus the rank \(r\) is finite.
Choose generators \(x_1, \ldots, x_n\) of \(M\). Choose a basis \(e_1, \ldots, e_r\) for \(M \otimes_R K\). We may write \(x_i \otimes 1 = \sum (a_{ij}/b_{ij}) e_j\) for some \(a_{ij} , b_{ij} \in R\) with \(b_{ij} \not = 0\). Setting \(b = \prod b_{ij}\) and replacing \(a_{ij}\) by \(a_{ij}b/b_{ij}\) we may assume that \(x_i \otimes 1 = \sum (a_{ij}/b) e_j\). Note that if we have a relation \(\sum r_i x_i = 0\) in \(M\) then \(\sum r_ia_{ij} = 0\) in \(R\) for all \(j = 1, \ldots, r\).
Proof of (2). The last comment above tells us the map \(M \to \bigoplus Re_j\) sending \(x_i\) to \(\sum a_{ij} e_j\). is well defined. The fact that \(e_1, \ldots, e_r\) forms a basis of \(M \otimes_R K\) tells us that this map becomes an isomorphism upon tensoring by \(K\). Hence this is a map as in (2).
Proof of (3). The cokernel of the map \(M \to R^{\oplus r}\) constructed in (2) is finite and torsion. Hence we can find a \(g \in R\), \(g \not = 0\) which annihilates this cokernel. In other words, such that the map \(M_g \to R_g^{\oplus r}\) is surjective. Then we can choose a splitting \(M_g = N \oplus R_g^{\oplus r}\). Then \(N\) is a finite \(R_g\)-module with \(N \otimes_{R_g} K = 0\). In other words, \(N\) is torsion and finite over \(R_g\). Thus we can find a \(g' \in R\) which annihilates \(N\). Setting \(f= gg'\) we see that (3) holds.
To prove (4) choose a basis \(y_1, \ldots, y_r\) of \(M_f\) where \(f\) is as in (3). Then write \(y_j = m_j/f^{t_j}\) for some \(m_j \in M\) and \(t_j \geq 0\). The map \(R^{\oplus r} \to M\) sending the \(j\)th basis vector to \(m_j\) will be as in (4). Details omitted.
Reflexive modules
Here is our definition.
Definition
Let \(R\) be a domain. We say an \(R\)-module \(M\) is reflexive if the natural map \[j : M \longrightarrow \Hom_R(\Hom_R(M, R), R)\] which sends \(m \in M\) to the map sending \(\varphi \in \Hom_R(M, R)\) to \(\varphi(m) \in R\) is an isomorphism.
We can make this definition for more general rings, but already the definition above has drawbacks. It would be wise to restrict to Noetherian domains and finite torsion free modules and (perhaps) impose some regularity conditions on \(R\) (e.g., \(R\) is normal).
Lemma
Let \(R\) be a domain and let \(M\) be an \(R\)-module.
If \(M\) is reflexive, then \(M\) is torsion free.
If \(M\) is finite, then the kernel and cokernel of \(j : M \to \Hom_R(\Hom_R(M, R), R)\) are torsion modules.
If \(M\) is finite, then \(j\) is injective if and only if \(M\) is torsion free.
Proof
To see (1) use that the double dual of \(M\) is torsion free by Lemma 0AUX. Assume that \(M\) is finite. Choose generators \(x_1, \ldots, x_n\) of \(M\) over \(R\). Let \(K\) be the fraction field of \(R\). After renumbering we may assume that for some \(0 \leq r \leq n\) the elements \(x_1, \ldots, x_r\) map to a basis of \(M \otimes_R K\) over \(K\). For \(r < i \leq n\) write \(x_i \otimes 1 = \sum_{1 \leq j \leq r} x_j \otimes k_{ij}\) with \(k_{ij} \in K\). Choose \(c \in R\) such that \(a_{ij} = ck_{ij} \in R\) and such that \(cx_i = \sum_{1 \leq j \leq r} a_{ij} x_j\) in \(M\) (small detail omitted). Then we get maps \[R^{\oplus r} \xrightarrow{\alpha} M \xrightarrow{\beta} R^{\oplus r} \xrightarrow{\alpha} M\] with both \(\alpha \circ \beta\) and \(\beta \circ \alpha\) given by multiplication with \(c\). Here \(\alpha\) uses \(x_1, \ldots, x_r\) and \(\beta\) sends \(x_i\) to \(c\) times the \(i\)th basis vector for \(1 \leq i \leq r\) and to the vector \((a_{i1}, \ldots, a_{ir})\) for \(r < i \leq n\). Consider the commutative diagram \[\xymatrix{ R^{\oplus r} \ar[r] \ar[d]^{\text{id}} & M \ar[r] \ar[d]^j & R^{\oplus r} \ar[d]^{\text{id}} \ar[r] & M \ar[d]^j \\ R^{\oplus r} \ar[r] & \Hom_R(\Hom_R(M, R), R) \ar[r] & R^{\oplus r} \ar[r] & \Hom_R(\Hom_R(M, R), R) }\] Of course the compositions of consecutive arrows on the bottom row are equal to multiplication by \(c\). The diagram shows that the kernel and cokernel of \(j\) are annihilated by \(c\) which proves (2). For (3) note that if \(M\) is torsion free, then \(\beta\) is injective which implies that \(j\) is injective. The converse is clear because we’ve seen already that the double dual of \(M\) is torsion free.
Lemma
Let \(R\) be a discrete valuation ring and let \(M\) be a finite \(R\)-module. Then the map \(j : M \to \Hom_R(\Hom_R(M, R), R)\) is surjective.
Proof
Let \(M_{tors} \subset M\) be the torsion submodule. Then we have \(\Hom_R(M, R) = \Hom_R(M/M_{tors}, R)\) (holds over any domain). Hence we may assume that \(M\) is torsion free. Then \(M\) is free by Lemma 0AUW and the lemma is clear.
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a finite \(R\)-module. The following are equivalent:
\(M\) is reflexive,
\(M_\mathfrak p\) is a reflexive \(R_\mathfrak p\)-module for all primes \(\mathfrak p \subset R\), and
\(M_\mathfrak m\) is a reflexive \(R_\mathfrak m\)-module for all maximal ideals \(\mathfrak m\) of \(R\).
Proof
The localization of \(j : M \to \Hom_R(\Hom_R(M, R), R)\) at a prime \(\mathfrak p\) is the corresponding map for the module \(M_\mathfrak p\) over the Noetherian local domain \(R_\mathfrak p\). See Algebra, Lemma 0583. Thus the lemma holds by Algebra, Lemma 00HN.
Lemma
Let \(R\) be a Noetherian domain. Let \(0 \to M \to M' \to M''\) an exact sequence of finite \(R\)-modules. If \(M'\) is reflexive and \(M''\) is torsion free, then \(M\) is reflexive.
Proof
We will use without further mention that \(\Hom_R(N, N')\) is a finite \(R\)-module for any finite \(R\)-modules \(N\) and \(N'\), see Algebra, Lemma 08YR. We take duals to get a sequence \[\Hom_R(M, R) \leftarrow \Hom_R(M', R) \leftarrow \Hom_R(M'', R)\] Dualizing again we obtain a commutative diagram \[\xymatrix{ \Hom_R(\Hom_R(M, R), R) \ar[r]_j & \Hom_R(\Hom_R(M', R), R) \ar[r] & \Hom_R(\Hom_R(M'', R), R) \\ M \ar[u] \ar[r] & M' \ar[u] \ar[r] & M'' \ar[u] }\] We do not know the top row is exact. But, by assumption the middle vertical arrow is an isomorphism and the right vertical arrow is injective (Lemma 0AV0). We claim \(j\) is injective. Assuming the claim a diagram chase shows that the left vertical arrow is an isomorphism, i.e., \(M\) is reflexive.
Proof of the claim. Consider the exact sequence \(\Hom_R(M',R)\to \Hom_R(M,R)\to Q \to 0\) defining \(Q\). One applies Algebra, Lemma 0583 to obtain \[\Hom_K(M'\otimes_R K,K) \to \Hom_K(M\otimes_R K,K) \to Q\otimes_R K\to 0\] But \(M \otimes_R K \to M' \otimes_R K\) is an injective map of vector spaces, hence split injective, so \(Q \otimes_R K = 0\), that is, \(Q\) is torsion. Then one gets the exact sequence \[0 \to \Hom_R(Q,R) \to \Hom_R(\Hom_R(M,R),R) \to \Hom_R(\Hom_R(M',R),R)\] and \(\Hom_R(Q,R)=0\) because \(Q\) is torsion.
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a finite \(R\)-module. The following are equivalent
\(M\) is reflexive,
there exists a short exact sequence \(0 \to M \to F \to N \to 0\) with \(F\) finite free and \(N\) torsion free.
Proof
Observe that a finite free module is reflexive. By Lemma 0EB8 we see that (2) implies (1). Assume \(M\) is reflexive. Choose a presentation \(R^{\oplus m} \to R^{\oplus n} \to \Hom_R(M, R) \to 0\). Dualizing we get an exact sequence \[0 \to \Hom_R(\Hom_R(M, R), R) \to R^{\oplus n} \to N \to 0\] with \(N = \Im(R^{\oplus n} \to R^{\oplus m})\) a torsion free module. As \(M = \Hom_R(\Hom_R(M, R), R)\) we get an exact sequence as in (2).
Lemma
Let \(R \to R'\) be a flat homomorphism of Noetherian domains. If \(M\) is a finite reflexive \(R\)-module, then \(M \otimes_R R'\) is a finite reflexive \(R'\)-module.
Proof
Choose a short exact sequence \(0 \to M \to F \to N \to 0\) with \(F\) finite free and \(N\) torsion free, see Lemma 0AV2. Since \(R \to R'\) is flat we obtain a short exact sequence \(0 \to M \otimes_R R' \to F \otimes_R R' \to N \otimes_R R' \to 0\) with \(F \otimes_R R'\) finite free and \(N \otimes_R R'\) torsion free (Lemma 0AXM). Thus \(M \otimes_R R'\) is reflexive by Lemma 0AV2.
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a finite \(R\)-module. Let \(N\) be a finite reflexive \(R\)-module. Then \(\Hom_R(M, N)\) is reflexive.
Proof
Choose a presentation \(R^{\oplus m} \to R^{\oplus n} \to M \to 0\). Then we obtain \[0 \to \Hom_R(M, N) \to N^{\oplus n} \to N' \to 0\] with \(N' = \Im(N^{\oplus n} \to N^{\oplus m})\) torsion free. We conclude by Lemma 0EB8.
Definition
Let \(R\) be a Noetherian domain. Let \(M\) be a finite \(R\)-module. The module \(M^{**} = \Hom_R(\Hom_R(M, R), R)\) is called the reflexive hull of \(M\).
This makes sense because the reflexive hull is reflexive by Lemma 0AV3. The assignment \(M \mapsto M^{**}\) is a functor. If \(\varphi : M \to N\) is an \(R\)-module map into a reflexive \(R\)-module \(N\), then \(\varphi\) factors \(M \to M^{**} \to N\) through the reflexive hull of \(M\). Another way to say this is that taking the reflexive hull is the left adjoint to the inclusion functor \[\text{finite reflexive modules} \subset \text{finite modules}\] over a Noetherian domain \(R\).
Lemma
Let \(R\) be a Noetherian local ring. Let \(M\), \(N\) be finite \(R\)-modules.
If \(N\) has depth \(\geq 1\), then \(\Hom_R(M, N)\) has depth \(\geq 1\).
If \(N\) has depth \(\geq 2\), then \(\Hom_R(M, N)\) has depth \(\geq 2\).
Proof
Choose a presentation \(R^{\oplus m} \to R^{\oplus n} \to M \to 0\). Dualizing we get an exact sequence \[0 \to \Hom_R(M, N) \to N^{\oplus n} \to N' \to 0\] with \(N' = \Im(N^{\oplus n} \to N^{\oplus m})\). A submodule of a module with depth \(\geq 1\) has depth \(\geq 1\); this follows immediately from the definition. Thus part (1) is clear. For (2) note that here the assumption and the previous remark implies \(N'\) has depth \(\geq 1\). The module \(N^{\oplus n}\) has depth \(\geq 2\). From Algebra, Lemma 00LX we conclude \(\Hom_R(M, N)\) has depth \(\geq 2\).
Lemma
Let \(R\) be a Noetherian ring. Let \(M\), \(N\) be finite \(R\)-modules.
If \(N\) has property \((S_1)\), then \(\Hom_R(M, N)\) has property \((S_1)\).
If \(N\) has property \((S_2)\), then \(\Hom_R(M, N)\) has property \((S_2)\).
If \(R\) is a domain, \(N\) is torsion free and \((S_2)\), then \(\Hom_R(M, N)\) is torsion free and has property \((S_2)\).
Proof
Since localizing at primes commutes with taking \(\Hom_R\) for finite \(R\)-modules (Algebra, Lemma 08YR) parts (1) and (2) follow immediately from Lemma 0AV5. Part (3) follows from (2) and Lemma 0AUX.
Lemma
Let \(R\) be a Noetherian ring. Let \(\varphi : M \to N\) be a map of \(R\)-modules. Assume that for every prime \(\mathfrak p\) of \(R\) at least one of the following happens
\(M_\mathfrak p \to N_\mathfrak p\) is injective, or
\(\mathfrak p \not \in \text{Ass}(M)\).
Then \(\varphi\) is injective.
Proof
Let \(\mathfrak p\) be an associated prime of \(\Ker(\varphi)\). Then there exists an element \(x \in M_\mathfrak p\) which is in the kernel of \(M_\mathfrak p \to N_\mathfrak p\) and whose annihilator is \(\mathfrak pR_\mathfrak p\) (Algebra, Lemma 0310). This is impossible in both cases. Hence \(\text{Ass}(\Ker(\varphi)) = \emptyset\) and we conclude \(\Ker(\varphi) = 0\) by Algebra, Lemma 0587.
Lemma
Let \(R\) be a Noetherian ring. Let \(\varphi : M \to N\) be a map of \(R\)-modules. Assume \(M\) is finite and that for every prime \(\mathfrak p\) of \(R\) one of the following happens
\(M_\mathfrak p \to N_\mathfrak p\) is an isomorphism, or
\(\text{depth}(M_\mathfrak p) \geq 2\) and \(\mathfrak p \not \in \text{Ass}(N)\).
Then \(\varphi\) is an isomorphism.
Proof
By Lemma 0AV7 we see that \(\varphi\) is injective. Let \(N' \subset N\) be an finitely generated \(R\)-module containing the image of \(M\). Then \(\text{Ass}(N_\mathfrak p) = \emptyset\) implies \(\text{Ass}(N'_\mathfrak p) = \emptyset\). Hence the assumptions of the lemma hold for \(M \to N'\). In order to prove that \(\varphi\) is an isomorphism, it suffices to prove the same thing for every such \(N' \subset N\). Thus we may assume \(N\) is a finite \(R\)-module. In this case, \(\mathfrak p \not \in \text{Ass}(N) \Rightarrow \text{depth}(N_\mathfrak p) \geq 1\), see Algebra, Lemma 00LL. Consider the short exact sequence \[0 \to M \to N \to Q \to 0\] defining \(Q\). Looking at the conditions we see that either \(Q_\mathfrak p = 0\) in case (1) or \(\text{depth}(Q_\mathfrak p) \geq 1\) in case (2) by Algebra, Lemma 00LX. This implies that \(Q\) does not have any associated primes, hence \(Q = 0\) by Algebra, Lemma 0587.
Lemma
Let \(R\) be a Noetherian domain. Let \(\varphi : M \to N\) be a map of \(R\)-modules. Assume \(M\) is finite, \(N\) is torsion free, and that for every prime \(\mathfrak p\) of \(R\) one of the following happens
\(M_\mathfrak p \to N_\mathfrak p\) is an isomorphism, or
\(\text{depth}(M_\mathfrak p) \geq 2\).
Then \(\varphi\) is an isomorphism.
Proof
This is a special case of Lemma 0AV8.
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a finite \(R\)-module. The following are equivalent
\(M\) is reflexive,
for every prime \(\mathfrak p\) of \(R\) one of the following happens
\(M_\mathfrak p\) is a reflexive \(R_\mathfrak p\)-module, or
\(\text{depth}(M_\mathfrak p) \geq 2\).
Proof
If (1) is true, then \(M_\mathfrak p\) is a reflexive module for all primes of \(\mathfrak p\) by Lemma 0AV1. Thus (1) \(\Rightarrow\) (2). Assume (2). Set \(N = \Hom_R(\Hom_R(M, R), R)\) so that \[N_\mathfrak p = \Hom_{R_\mathfrak p}( \Hom_{R_\mathfrak p}(M_\mathfrak p, R_\mathfrak p), R_\mathfrak p)\] for every prime \(\mathfrak p\) of \(R\). See Algebra, Lemma 0583. We apply Lemma 0AV9 to the map \(j : M \to N\). This is allowed because \(M\) is finite and \(N\) is torsion free by Lemma 0AUX. In case (2)(a) the map \(M_\mathfrak p \to N_\mathfrak p\) is an isomorphism and in case (2)(b) we have \(\text{depth}(M_\mathfrak p) \geq 2\).
Lemma
Let \(R\) be a Noetherian domain. Let \(M\) be a finite reflexive \(R\)-module. Let \(\mathfrak p \subset R\) be a prime ideal.
If \(\text{depth}(R_\mathfrak p) \geq 2\), then \(\text{depth}(M_\mathfrak p) \geq 2\).
If \(R\) is \((S_2)\), then \(M\) is \((S_2)\).
Proof
Since formation of reflexive hull \(\Hom_R(\Hom_R(M, R), R)\) commutes with localization (Algebra, Lemma 0583) part (1) follows from Lemma 0AV5. Part (2) is immediate from Lemma 0AV6.
Example
The results above and below suggest reflexivity is related to the \((S_2)\) condition; here is an example to prevent too optimistic conjectures. Let \(k\) be a field. Let \(R\) be the \(k\)-subalgebra of \(k[x, y]\) generated by \(1, y, x^2, xy, x^3\). Then \(R\) is not \((S_2)\). So \(R\) as an \(R\)-module is an example of a reflexive \(R\)-module which is not \((S_2)\). Let \(M = k[x, y]\) viewed as an \(R\)-module. Then \(M\) is a reflexive \(R\)-module because \[\Hom_R(M, R) = \mathfrak m = (y, x^2, xy, x^3) \quad\text{and}\quad \Hom_R(\mathfrak m, R) = M\] and \(M\) is \((S_2)\) as an \(R\)-module (computations omitted). Thus \(R\) is a Noetherian domain possessing a reflexive \((S_2)\) module but \(R\) is not \((S_2)\) itself.
Lemma
Let \(R\) be a Noetherian normal domain with fraction field \(K\). Let \(M\) be a finite \(R\)-module. The following are equivalent
\(M\) is reflexive,
\(M\) is torsion free and has property \((S_2)\),
\(M\) is torsion free and \(M = \bigcap_{\text{height}(\mathfrak p) = 1} M_{\mathfrak p}\) where the intersection happens in \(M_K = M \otimes_R K\).
Proof
By Algebra, Lemma 031S we see that \(R\) satisfies \((R_1)\) and \((S_2)\).
Assume (1). Then \(M\) is torsion free by Lemma 0AV0 and satisfies \((S_2)\) by Lemma 0EBA. Thus (2) holds.
Assume (2). By definition \(M' = \bigcap_{\text{height}(\mathfrak p) = 1} M_{\mathfrak p}\) is the kernel of the map \[M_K \longrightarrow \bigoplus\nolimits_{\text{height}(\mathfrak p) = 1} M_K/M_\mathfrak p \subset \prod\nolimits_{\text{height}(\mathfrak p) = 1} M_K/M_\mathfrak p\] Observe that our map indeed factors through the direct sum as indicated since given \(a/b \in K\) there are at most finitely many height \(1\) primes \(\mathfrak p\) with \(b \in \mathfrak p\). Let \(\mathfrak p_0\) be a prime of height \(1\). Then \((M_K/M_\mathfrak p)_{\mathfrak p_0} = 0\) unless \(\mathfrak p = \mathfrak p_0\) in which case we get \((M_K/M_\mathfrak p)_{\mathfrak p_0} = M_K/M_{\mathfrak p_0}\). Thus by exactness of localization and the fact that localization commutes with direct sums, we see that \(M'_{\mathfrak p_0} = M_{\mathfrak p_0}\). Since \(M\) has depth \(\geq 2\) at primes of height \(> 1\), we see that \(M \to M'\) is an isomorphism by Lemma 0AV9. Hence (3) holds.
Assume (3). Let \(\mathfrak p\) be a prime of height \(1\). Then \(R_\mathfrak p\) is a discrete valuation ring by \((R_1)\). By Lemma 0AUW we see that \(M_\mathfrak p\) is finite free, in particular reflexive. Hence the map \(M \to M^{**}\) induces an isomorphism at all the primes \(\mathfrak p\) of height \(1\). Thus the condition \(M = \bigcap_{\text{height}(\mathfrak p) = 1} M_{\mathfrak p}\) implies that \(M = M^{**}\) and (1) holds.
Lemma
Let \(R\) be a Noetherian normal domain. Let \(M\) be a finite \(R\)-module. Then the reflexive hull of \(M\) is the intersection \[M^{**} = \bigcap\nolimits_{\text{height}(\mathfrak p) = 1} M_{\mathfrak p}/(M_\mathfrak p)_{tors} = \bigcap\nolimits_{\text{height}(\mathfrak p) = 1} (M/M_{tors})_\mathfrak p\] taken in \(M \otimes_R K\).
Proof
Let \(\mathfrak p\) be a prime of height \(1\). The kernel of \(M_\mathfrak p \to M \otimes_R K\) is the torsion submodule \((M_\mathfrak p)_{tors}\) of \(M_\mathfrak p\). Moreover, we have \((M/M_{tors})_\mathfrak p = M_\mathfrak p/(M_\mathfrak p)_{tors}\) and this is a finite free module over the discrete valuation ring \(R_\mathfrak p\) (Lemma 0AUW). Then \(M_\mathfrak p/(M_\mathfrak p)_{tors} \to (M_\mathfrak p)^{**} = (M^{**})_\mathfrak p\) is an isomorphism, hence the lemma is a consequence of Lemma 0AVB.
Lemma
Let \(A\) be a Noetherian normal domain with fraction field \(K\). Let \(L\) be a finite extension of \(K\). If the integral closure \(B\) of \(A\) in \(L\) is finite over \(A\), then \(B\) is reflexive as an \(A\)-module.
Proof
It suffices to show that \(B = \bigcap B_\mathfrak p\) where the intersection is over height \(1\) primes \(\mathfrak p \subset A\), see Lemma 0AVB. Let \(b \in \bigcap B_\mathfrak p\). Let \(x^d + a_1x^{d - 1} + \ldots + a_d\) be the minimal polynomial of \(b\) over \(K\). We want to show \(a_i \in A\). By Algebra, Lemma 00H7 we see that \(a_i \in A_\mathfrak p\) for all \(i\) and all height one primes \(\mathfrak p\). Hence we get what we want from Algebra, Lemma 031T (or the lemma already cited as \(A\) is a reflexive module over itself).
Content ideals
The definition may not be what you expect.
Definition
Let \(A\) be a ring. Let \(M\) be a flat \(A\)-module. Let \(x \in M\). If the set of ideals \(I\) in \(A\) such that \(x \in IM\) has a smallest element, we call it the content ideal of \(x\).
Note that since \(M\) is flat over \(A\), for a pair of ideals \(I, I'\) of \(A\) we have \(IM \cap I'M = (I \cap I')M\) as can be seen by tensoring the exact sequence \(0 \to I \cap I' \to I \oplus I' \to I + I' \to 0\) by \(M\).
Lemma
Let \(A\) be a ring. Let \(M\) be a flat \(A\)-module. Let \(x \in M\). The content ideal of \(x\), if it exists, is finitely generated.
Proof
Say \(x \in IM\). Then we can write \(x = \sum_{i = 1, \ldots, n} f_i x_i\) with \(f_i \in I\) and \(x_i \in M\). Hence \(x \in I'M\) with \(I' = (f_1, \ldots, f_n)\).
Lemma
Let \((A, \mathfrak m)\) be a local ring. Let \(u : M \to N\) be a map of flat \(A\)-modules such that \(\overline{u} : M/\mathfrak m M \to N/\mathfrak m N\) is injective. If \(x \in M\) has content ideal \(I\), then \(u(x)\) has content ideal \(I\) as well.
Proof
It is clear that \(u(x) \in IN\). If \(u(x) \in I'N\), then \(u(x) \in (I' \cap I)N\), see discussion following Definition 0ASA. Hence it suffices to show: if \(x \in I'N\) and \(I' \subset I\), \(I' \not = I\), then \(u(x) \not \in I'N\). Since \(I/I'\) is a nonzero finite \(A\)-module (Lemma 0ASB) there is a nonzero map \(\chi : I/I' \to A/\mathfrak m\) of \(A\)-modules by Nakayama’s lemma (Algebra, Lemma 00DV). Since \(I\) is the content ideal of \(x\) we see that \(x \not \in I''M\) where \(I'' = \Ker(\chi)\). Hence \(x\) is not in the kernel of the map \[IM = I \otimes_A M \xrightarrow{\chi \otimes 1} A/\mathfrak m \otimes M \cong M/\mathfrak m M\] Applying our hypothesis on \(\overline{u}\) we conclude that \(u(x)\) does not map to zero under the map \[IN = I \otimes_A N \xrightarrow{\chi \otimes 1} A/\mathfrak m \otimes N \cong N/\mathfrak m N\] and we conclude.
Lemma
Let \(A\) be a ring. Let \(M\) be a flat Mittag-Leffler module. Then every element of \(M\) has a content ideal.
Proof
This is a special case of Algebra, Lemma 059S.
Flatness and finiteness conditions
In this section we discuss some implications of the type “flat \(+\) finite type \(\Rightarrow\) finite presentation”. We will revisit this result in the chapter on flatness, see More on Flatness, Section 057N. A first result of this type was proved in Algebra, Lemma 052U.
Lemma
Let \(R\) be a ring. Let \(S = R[x_1, \ldots, x_n]\) be a polynomial ring over \(R\). Let \(M\) be an \(S\)-module. Assume
there exist finitely many primes \(\mathfrak p_1, \ldots, \mathfrak p_m\) of \(R\) such that the map \(R \to \prod R_{\mathfrak p_j}\) is injective,
\(M\) is a finite \(S\)-module,
\(M\) flat over \(R\), and
for every prime \(\mathfrak p\) of \(R\) the module \(M_{\mathfrak p}\) is of finite presentation over \(S_{\mathfrak p}\).
Then \(M\) is of finite presentation over \(S\).
Proof
Choose a presentation \[0 \to K \to S^{\oplus r} \to M \to 0\] of \(M\) as an \(S\)-module. Let \(\mathfrak q\) be a prime ideal of \(S\) lying over a prime \(\mathfrak p\) of \(R\). By assumption there exist finitely many elements \(k_1, \ldots, k_t \in K\) such that if we set \(K' = \sum Sk_j \subset K\) then \(K'_{\mathfrak p} = K_{\mathfrak p}\) and \(K'_{\mathfrak p_j} = K_{\mathfrak p_j}\) for \(j = 1, \ldots, m\). Setting \(M' = S^{\oplus r}/K'\) we deduce that in particular \(M'_{\mathfrak q} = M_{\mathfrak q}\). By openness of flatness, see Algebra, Theorem 00RC we conclude that there exists a \(g \in S\), \(g \not \in \mathfrak q\) such that \(M'_g\) is flat over \(R\). Thus \(M'_g \to M_g\) is a surjective map of flat \(R\)-modules. Consider the commutative diagram \[\xymatrix{ M'_g \ar[r] \ar[d] & M_g \ar[d] \\ \prod (M'_g)_{\mathfrak p_j} \ar[r] & \prod (M_g)_{\mathfrak p_j} }\] The bottom arrow is an isomorphism by choice of \(k_1, \ldots, k_t\). The left vertical arrow is an injective map as \(R \to \prod R_{\mathfrak p_j}\) is injective and \(M'_g\) is flat over \(R\). Hence the top horizontal arrow is injective, hence an isomorphism. This proves that \(M_g\) is of finite presentation over \(S_g\). We conclude by applying Algebra, Lemma 00EO.
Lemma
Let \(R \to S\) be a ring homomorphism. Assume
there exist finitely many primes \(\mathfrak p_1, \ldots, \mathfrak p_m\) of \(R\) such that the map \(R \to \prod R_{\mathfrak p_j}\) is injective,
\(R \to S\) is of finite type,
\(S\) flat over \(R\), and
for every prime \(\mathfrak p\) of \(R\) the ring \(S_{\mathfrak p}\) is of finite presentation over \(R_{\mathfrak p}\).
Then \(S\) is of finite presentation over \(R\).
Proof
By assumption \(S\) is a quotient of a polynomial ring over \(R\). Thus the result follows directly from Lemma 053A.
Lemma
Let \(R\) be a ring. Let \(S = R[x_1, \ldots, x_n]\) be a graded polynomial algebra over \(R\), i.e., \(\deg(x_i) > 0\) but not necessarily equal to \(1\). Let \(M\) be a graded \(S\)-module. Assume
\(R\) is a local ring,
\(M\) is a finite \(S\)-module, and
\(M\) is flat over \(R\).
Then \(M\) is finitely presented as an \(S\)-module.
Proof
Let \(M = \bigoplus M_d\) be the grading on \(M\). Pick homogeneous generators \(m_1, \ldots, m_r \in M\) of \(M\). Say \(\deg(m_i) = d_i \in \mathbf{Z}\). This gives us a presentation \[0 \to K \to \bigoplus\nolimits_{i = 1, \ldots, r} S(-d_i) \to M \to 0\] which in each degree \(d\) leads to the short exact sequence \[0 \to K_d \to \bigoplus\nolimits_{i = 1, \ldots, r} S_{d - d_i} \to M_d \to 0.\] By assumption each \(M_d\) is a finite flat \(R\)-module. By Algebra, Lemma 00NZ this implies each \(M_d\) is a finite free \(R\)-module. Hence we see each \(K_d\) is a finite \(R\)-module. Also each \(K_d\) is flat over \(R\) by Algebra, Lemma 00HM. Hence we conclude that each \(K_d\) is finite free by Algebra, Lemma 00NZ again.
Let \(\mathfrak m\) be the maximal ideal of \(R\). By the flatness of \(M\) over \(R\) the short exact sequences above remain short exact after tensoring with \(\kappa = \kappa(\mathfrak m)\). As the ring \(S \otimes_R \kappa\) is Noetherian we see that there exist homogeneous elements \(k_1, \ldots, k_t \in K\) such that the images \(\overline{k}_j\) generate \(K \otimes_R \kappa\) over \(S \otimes_R \kappa\). Say \(\deg(k_j) = e_j\). Thus for any \(d\) the map \[\bigoplus\nolimits_{j = 1, \ldots, t} S_{d - e_j} \longrightarrow K_d\] becomes surjective after tensoring with \(\kappa\). By Nakayama’s lemma (Algebra, Lemma 00DV) this implies the map is surjective over \(R\). Hence \(K\) is generated by \(k_1, \ldots, k_t\) over \(S\) and we win.
Lemma
Let \(R\) be a ring. Let \(S = \bigoplus_{n \geq 0} S_n\) be a graded \(R\)-algebra. Let \(M = \bigoplus_{d \in \mathbf{Z}} M_d\) be a graded \(S\)-module. Assume \(S\) is finitely generated as an \(R\)-algebra, assume \(S_0\) is a finite \(R\)-algebra, and assume there exist finitely many primes \(\mathfrak p_j\), \(i = 1, \ldots, m\) such that \(R \to \prod R_{\mathfrak p_j}\) is injective.
If \(S\) is flat over \(R\), then \(S\) is a finitely presented \(R\)-algebra.
If \(M\) is flat as an \(R\)-module and finite as an \(S\)-module, then \(M\) is finitely presented as an \(S\)-module.
Proof
As \(S\) is finitely generated as an \(R\)-algebra, it is finitely generated as an \(S_0\) algebra, say by homogeneous elements \(t_1, \ldots, t_n \in S\) of degrees \(d_1, \ldots, d_n > 0\). Set \(P = R[x_1, \ldots, x_n]\) with \(\deg(x_i) = d_i\). The ring map \(P \to S\), \(x_i \to t_i\) is finite as \(S_0\) is a finite \(R\)-module. To prove (1) it suffices to prove that \(S\) is a finitely presented \(P\)-module. To prove (2) it suffices to prove that \(M\) is a finitely presented \(P\)-module. Thus it suffices to prove that if \(S = P\) is a graded polynomial ring and \(M\) is a finite \(S\)-module flat over \(R\), then \(M\) is finitely presented as an \(S\)-module. By Lemma 053C we see \(M_{\mathfrak p}\) is a finitely presented \(S_{\mathfrak p}\)-module for every prime \(\mathfrak p\) of \(R\). Thus the result follows from Lemma 053A.
Remark
Let \(R\) be a ring. When does \(R\) satisfy the condition mentioned in Lemmas 053A, 053B, and 053D? This holds if
\(R\) is local,
\(R\) is Noetherian,
\(R\) is a domain,
\(R\) is a reduced ring with finitely many minimal primes, or
\(R\) has finitely many weakly associated primes, see Algebra, Lemma 05CB.
Thus these lemmas hold in all cases listed above.
The following lemma will be improved on in More on Flatness, Proposition 053G.
Lemma
Let \(A\) be a valuation ring. Let \(A \to B\) be a ring map of finite type. Let \(M\) be a finite \(B\)-module.
If \(B\) is flat over \(A\), then \(B\) is a finitely presented \(A\)-algebra.
If \(M\) is flat as an \(A\)-module, then \(M\) is finitely presented as a \(B\)-module.
Proof
We are going to use that an \(A\)-module is flat if and only if it is torsion free, see Lemma 0539. By Algebra, Lemma 052N we can find a graded \(A\)-algebra \(S\) with \(S_0 = A\) and generated by finitely many elements in degree \(1\), an element \(f \in S_1\) and a finite graded \(S\)-module \(N\) such that \(B \cong S_{(f)}\) and \(M \cong N_{(f)}\). If \(M\) is torsion free, then we can take \(N\) torsion free by replacing it by \(N/N_{tors}\), see Lemma 0537. Similarly, if \(B\) is torsion free, then we can take \(S\) torsion free by replacing it by \(S/S_{tors}\). Hence in case (1), we may apply Lemma 053D to see that \(S\) is a finitely presented \(A\)-algebra, which implies that \(B = S_{(f)}\) is a finitely presented \(A\)-algebra. To see (2) we may first replace \(S\) by a graded polynomial ring, and then we may apply Lemma 053C to conclude.
Lemma
Let \(A\) be a valuation ring. Let \(A \to B\) be a local homomorphism which is essentially of finite type. Let \(M\) be a finite \(B\)-module.
If \(B\) is flat over \(A\), then \(B\) is essentially of finite presentation over \(A\).
If \(M\) is flat as an \(A\)-module, then \(M\) is finitely presented as a \(B\)-module.
Proof
By assumption we can write \(B\) as a quotient of the localization of a polynomial algebra \(P = A[x_1, \ldots, x_n]\) at a prime ideal \(\mathfrak q\). In case (1) we consider \(M = B\) as a finite module over \(P_\mathfrak q\) and in case (2) we consider \(M\) as a finite module over \(P_\mathfrak q\). In both cases, we have to show that this is a finitely presented \(P_\mathfrak q\)-module, see Algebra, Lemma 0561 for case (2).
Choose a presentation \(0 \to K \to P_\mathfrak q^{\oplus r} \to M \to 0\) which is possible because \(M\) is finite over \(P_\mathfrak q\). Let \(L = P^{\oplus r} \cap K\). Then \(K = L_\mathfrak q\), see Algebra, Lemma 00CU. Then \(N = P^{\oplus r}/L\) is a submodule of \(M\) and hence flat by Lemma 0539. Since also \(N\) is a finite \(P\)-module, we see that \(N\) is finitely presented as a \(P\)-module by Lemma 053E. Since localization is exact (Algebra, Proposition 00CS) we see that \(N_\mathfrak q = M\) and we conclude.
Blowing up and flatness
In this section we begin our discussion of results of the form: “After a blowup the strict transform becomes flat”. More results of this type may be found in Divisors, Section 0F84 and More on Flatness, Section 080X.
Definition
Let \(R\) be a ring. Let \(I \subset R\) be an ideal and \(a \in I\). Let \(R[\frac{I}{a}]\) be the affine blowup algebra, see Algebra, Definition 052Q. Let \(M\) be an \(R\)-module. The strict transform of \(M\) along \(R \to R[\frac{I}{a}]\) is the \(R[\frac{I}{a}]\)-module \[M' = \left(M \otimes_R R[\textstyle{\frac{I}{a}}]\right)/a\text{-power torsion}\]
The following is a very weak version of flattening by blowing up, but it is already sometimes a useful result.
Lemma
Let \((R, \mathfrak m)\) be a local domain with fraction field \(K\). Let \(S\) be a finite type \(R\)-algebra. Let \(M\) be a finite \(S\)-module. For every valuation ring \(A \subset K\) dominating \(R\) there exists an ideal \(I \subset \mathfrak m\) and a nonzero element \(a \in I\) such that
\(I\) is finitely generated,
\(A\) has center on \(R[\frac{I}{a}]\),
the fibre ring of \(R \to R[\frac{I}{a}]\) at \(\mathfrak m\) is not zero, and
the strict transform \(S_{I, a}\) of \(S\) along \(R \to R[\frac{I}{a}]\) is flat and of finite presentation over \(R\), and the strict transform \(M_{I, a}\) of \(M\) along \(R \to R[\frac{I}{a}]\) is flat over \(R\) and finitely presented over \(S_{I, a}\).
Proof
Write \(S = R[x_1, \ldots, x_n]/J\) and denote \(N = S \oplus M\) viewed as a module over \(P = R[x_1, \ldots, x_n]\). If we can prove the lemma in case \(S\) is a polynomial algebra over \(R\), then we can find \(I, a\) satisfying (1), (2), (3) such that the strict transform \(N_{I, a}\) of \(N\) along \(R \to R[\frac{I}{a}]\) is flat over \(R\) and finitely presented as a module over the strict transform \(P_{I, a}\) of \(P\). Since \(P_{I, a} = R[\frac{I}{a}][x_1, \ldots, x_n]\) (small detail omitted) we find that the summand \(S_{I, a} \subset N_{I, a}\) is flat over \(R\) and finitely presented as a module over \(R[\frac{I}{a}][x_1, \ldots, x_n]\). Hence \(S_{I, a}\) is finitely presented as an \(R[\frac{I}{a}]\)-algebra. Moreover, the summand \(M_{I, a} \subset N_{I, a}\) is flat over \(R\) and finitely presented as a module over \(P_{I, a}\) hence also finitely presented as a module over \(S_{I, a}\), see Algebra, Lemma 0561. This reduces us to the case discussed in the next paragraph.
Assume \(S = R[x_1, \ldots, x_n]\). Choose a presentation \[0 \to K \to S^{\oplus r} \to M \to 0.\] Let \(M_A\) be the quotient of \(M \otimes_R A\) by its torsion submodule, see Lemma 0537. Then \(M_A\) is a finite module over \(S_A = A[x_1, \ldots, x_n]\). By Lemma 0539 we see that \(M_A\) is flat over \(A\). By Lemma 053E we see that \(M_A\) is finitely presented. Hence there exist finitely many elements \(k_1, \ldots, k_t \in S_A^{\oplus r}\) which generate the kernel of the presentation \(S_A^{\oplus r} \to M_A\) as an \(S_A\)-module. For any choice of \(a \in I \subset \mathfrak m\) satisfying (1), (2), and (3) we denote \(M_{I, a}\) the strict transform of \(M\) along \(R \to R[\frac{I}{a}]\). It is a finite module over \(S_{I, a} = R[\frac{I}{a}][x_1, \ldots, x_n]\). By Algebra, Lemma 052M we have \(A = \colim_{I, a} R[\frac{I}{a}]\). This implies that \(S_A = \colim S_{I, a}\) and \[\colim M \otimes_R R[\textstyle{\frac{I}{a}}] = M \otimes_R A\] Choose \(I, a\) and lifts \(k_1, \ldots, k_t \in S_{I, a}^{\oplus r}\). Since \(M_A\) is the quotient of \(M \otimes_R A\) by torsion, we see that the images of \(k_1, \ldots, k_t\) in \(M \otimes_R A\) are annihilated by a nonzero element \(\alpha \in A\). After replacing \(I, a\) by a different pair (recall that the colimit is filtered), we may assume \(\alpha = x/a^n\) for some \(x \in I^n\) nonzero. Then we find that \(x k_1, \ldots, x k_t\) map to zero in \(M \otimes_R A\). Hence after replacing \(I, a\) by a different pair we may assume \(x k_1, \ldots, x k_t\) map to zero in \(M \otimes_R R[\frac{I}{a}]\) for some nonzero \(x \in R\). Then finally replacing \(I, a\) by \(xI, xa\) we find that we may assume \(k_1, \ldots, k_t\) map to \(a\)-power torsion elements of \(M \otimes_R R[\frac{I}{a}]\). For any such pair \((I, a)\) we set \[M'_{I, a} = S_{I, a}^{\oplus r}/ \sum S_{I, a}k_j.\] Since \(M_A = S_A^{\oplus r}/ \sum S_Ak_j\) we see that \(M_A = \colim_{I, a} M'_{I, a}\). At this point we finally apply Algebra, Lemma 02JO (3) to conclude that \(M'_{I, a}\) is flat for some pair \((I, a)\) as above. This lemma does not apply a priori to the system of strict transforms \[M_{I, a} = (M \otimes_R R[\textstyle{\frac{I}{a}}])/a\text{-power torsion}\] as the transition maps may not satisfy the assumptions of the lemma. But now, since flatness implies torsion free (Lemma 0538) and since \(M_{I, a}\) is the quotient of \(M'_{I, a}\) (because we arranged it so the elements \(k_1, \ldots, k_t\) map to zero in \(M_{I, a}\)) by the \(a\)-power torsion submodule we also conclude that \(M'_{I, a} = M_{I, a}\) for such a pair and we win.
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(k \geq 0\) and \(I = \text{Fit}_k(M)\). For every \(a \in I\) with \(R' = R[\frac{I}{a}]\) the strict transform \[M' = (M \otimes_R R')/a\text{-power torsion}\] has \(\text{Fit}_k(M') = R'\).
Proof
First observe that \(\text{Fit}_k(M \otimes_R R') = IR' = aR'\). The first equality by Lemma 07ZA part (3) and the second equality by Algebra, Lemma 07Z3. From Lemma 080Z and exactness of localization we see that \(M'_{\mathfrak p'}\) can be generated by \(\leq k\) elements for every prime \(\mathfrak p'\) of \(R'\). Then \(\text{Fit}_k(M') = R'\) for example by Lemma 07ZC.
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(k \geq 0\) and \(I = \text{Fit}_k(M)\). Assume that \(M_\mathfrak p\) is free of rank \(k\) for every \(\mathfrak p \not \in V(I)\). Then for every \(a \in I\) with \(R' = R[\frac{I}{a}]\) the strict transform \[M' = (M \otimes_R R')/a\text{-power torsion}\] is locally free of rank \(k\).
Proof
By Lemma 0CZM we have \(\text{Fit}_k(M') = R'\). By Lemma 07ZD it suffices to show that \(\text{Fit}_{k - 1}(M') = 0\). Recall that \(R' \subset R'_a = R_a\), see Algebra, Lemma 07Z3. Hence it suffices to prove that \(\text{Fit}_{k - 1}(M')\) maps to zero in \(R'_a = R_a\). Since clearly \((M')_a = M_a\) this reduces us to showing that \(\text{Fit}_{k - 1}(M_a) = 0\) because formation of Fitting ideals commutes with base change according to Lemma 07ZA part (3). This is true by our assumption that \(M_a\) is finite locally free of rank \(k\) (see Algebra, Lemma 00NX) and the already cited Lemma 07ZD.
Lemma
Let \(R\) be a ring. Let \(M\) be a finite \(R\)-module. Let \(f \in R\) be an element such that \(M_f\) is finite locally free of rank \(r\). Then there exists a finitely generated ideal \(I \subset R\) with \(V(f) = V(I)\) such that for all \(a \in I\) with \(R' = R[\frac{I}{a}]\) the strict transform \[M' = (M \otimes_R R')/a\text{-power torsion}\] is locally free of rank \(r\).
Proof
Choose a surjection \(R^{\oplus n} \to M\). Choose a finite submodule \(K \subset \Ker(R^{\oplus n} \to M)\) such that \(R^{\oplus n}/K \to M\) becomes an isomorphism after inverting \(f\). This is possible because \(M_f\) is of finite presentation for example by Algebra, Lemma 00NX. Set \(M_1 = R^{\oplus n}/K\) and suppose we can prove the lemma for \(M_1\). Say \(I \subset R\) is the corresponding ideal. Then for \(a \in I\) the map \[M_1' = (M_1 \otimes_R R')/a\text{-power torsion} \longrightarrow M' = (M \otimes_R R')/a\text{-power torsion}\] is surjective. It is also an isomorphism after inverting \(a\) in \(R'\) as \(R'_a = R_f\), see Algebra, Lemma 080U. But \(a\) is a nonzerodivisor on \(M'_1\), whence the displayed map is an isomorphism. Thus it suffices to prove the lemma in case \(M\) is a finitely presented \(R\)-module.
Assume \(M\) is a finitely presented \(R\)-module. Then \(J = \text{Fit}_r(M) \subset S\) is a finitely generated ideal. We claim that \(I = fJ\) works.
We first check that \(V(f) = V(I)\). The inclusion \(V(f) \subset V(I)\) is clear. Conversely, if \(f \not \in \mathfrak p\), then \(\mathfrak p\) is not an element of \(V(J)\) by Lemma 07ZC. Thus \(\mathfrak p \not \in V(fJ) = V(I)\).
Let \(a \in I\) and set \(R' = R[\frac{I}{a}]\). We may write \(a = fb\) for some \(b \in J\). By Algebra, Lemmas 07Z3 and 0BBI we see that \(J R' = b R'\) and \(b\) is a nonzerodivisor in \(R'\). Let \(\mathfrak p' \subset R' = R[\frac{I}{a}]\) be a prime ideal. Then \(JR'_{\mathfrak p'}\) is generated by \(b\). It follows from Lemma 080Z that \(M'_{\mathfrak p'}\) can be generated by \(r\) elements. Since \(M'\) is finite, there exist \(m_1, \ldots, m_r \in M'\) and \(g \in R'\), \(g \not \in \mathfrak p'\) such that the corresponding map \((R')^{\oplus r} \to M'\) becomes surjective after inverting \(g\).
Finally, consider the ideal \(J' = \text{Fit}_{k - 1}(M')\). Note that \(J' R'_g\) is generated by the coefficients of relations between \(m_1, \ldots, m_r\) (compatibility of Fitting ideal with base change). Thus it suffices to show that \(J' = 0\), see Lemma 07ZD. Since \(R'_a = R_f\) (Algebra, Lemma 080U) and \(M'_a = M_f\) is free of rank \(r\) we see that \(J'_a = 0\). Since \(a\) is a nonzerodivisor in \(R'\) we conclude that \(J' = 0\) and we win.
Completion and flatness
In this section we discuss when the completion of a “big” flat module is flat.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(A\) be a set. Assume \(R\) is Noetherian and complete with respect to \(I\). There is a canonical map \[\left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge \longrightarrow \prod\nolimits_{\alpha \in A} R\] from the \(I\)-adic completion of the direct sum into the product which is universally injective.
Proof
By definition an element \(x\) of the left hand side is \(x = (x_n)\) where \(x_n = (x_{n, \alpha}) \in \bigoplus\nolimits_{\alpha \in A} R/I^n\) such that \(x_{n, \alpha} = x_{n + 1, \alpha} \bmod I^n\). As \(R = R^\wedge\) we see that for any \(\alpha\) there exists a \(y_\alpha \in R\) such that \(x_{n, \alpha} = y_\alpha \bmod I^n\). Note that for each \(n\) there are only finitely many \(\alpha\) such that the elements \(x_{n, \alpha}\) are nonzero. Conversely, given \((y_\alpha) \in \prod_\alpha R\) such that for each \(n\) there are only finitely many \(\alpha\) such that \(y_{\alpha} \bmod I^n\) is nonzero, then this defines an element of the left hand side. Hence we can think of an element of the left hand side as infinite “convergent sums” \(\sum_\alpha y_\alpha\) with \(y_\alpha \in R\) such that for each \(n\) there are only finitely many \(y_\alpha\) which are nonzero modulo \(I^n\). The displayed map maps this element to the element to \((y_\alpha)\) in the product. In particular the map is injective.
Let \(Q\) be a finite \(R\)-module. We have to show that the map \[Q \otimes_R \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge \longrightarrow Q \otimes_R \left(\prod\nolimits_{\alpha \in A} R\right)\] is injective, see Algebra, Theorem 058K. Choose a presentation \(R^{\oplus k} \to R^{\oplus m} \to Q \to 0\) and denote \(q_1, \ldots, q_m \in Q\) the corresponding generators for \(Q\). By Artin-Rees (Algebra, Lemma 00IN) there exists a constant \(c\) such that \(\Im(R^{\oplus k} \to R^{\oplus m}) \cap (I^N)^{\oplus m} \subset \Im((I^{N - c})^{\oplus k} \to R^{\oplus m})\). Let us contemplate the diagram \[\xymatrix{ \bigoplus_{l = 1}^k \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge \ar[r] \ar[d] & \bigoplus_{j = 1}^m \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge \ar[r] \ar[d] & Q \otimes_R \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge \ar[r] \ar[d] & 0 \\ \bigoplus_{l = 1}^k \left(\prod\nolimits_{\alpha \in A} R\right) \ar[r] & \bigoplus_{j = 1}^m \left(\prod\nolimits_{\alpha \in A} R\right) \ar[r] & Q \otimes_R \left(\prod\nolimits_{\alpha \in A} R\right) \ar[r] & 0 }\] with exact rows. Pick an element \(\sum_j \sum_\alpha y_{j, \alpha}\) of \(\bigoplus_{j = 1, \ldots, m} \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge\). If this element maps to zero in the module \(Q \otimes_R \left(\prod\nolimits_{\alpha \in A} R\right)\), then we see in particular that \(\sum_j q_j \otimes y_{j, \alpha} = 0\) in \(Q\) for each \(\alpha\). Thus we can find an element \((z_{1, \alpha}, \ldots, z_{k, \alpha}) \in \bigoplus_{l = 1, \ldots, k} R\) which maps to \((y_{1, \alpha}, \ldots, y_{m, \alpha}) \in \bigoplus_{j = 1, \ldots, m} R\). Moreover, if \(y_{j, \alpha} \in I^{N_\alpha}\) for \(j = 1, \ldots, m\), then we may assume that \(z_{l, \alpha} \in I^{N_\alpha - c}\) for \(l = 1, \ldots, k\). Hence the sum \(\sum_l \sum_\alpha z_{l, \alpha}\) is “convergent” and defines an element of \(\bigoplus_{l = 1, \ldots, k} \left(\bigoplus\nolimits_{\alpha \in A} R\right)^\wedge\) which maps to the element \(\sum_j \sum_\alpha y_{j, \alpha}\) we started out with. Thus the right vertical arrow is injective and we win.
The following lemma can also be deduced from Lemma 0912 below.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(A\) be a set. Assume \(R\) is Noetherian. The completion \((\bigoplus\nolimits_{\alpha \in A} R)^\wedge\) is a flat \(R\)-module.
Proof
Denote \(R^\wedge\) the completion of \(R\) with respect to \(I\). As \(R \to R^\wedge\) is flat by Algebra, Lemma 00MB it suffices to prove that \((\bigoplus\nolimits_{\alpha \in A} R)^\wedge\) is a flat \(R^\wedge\)-module (use Algebra, Lemma 00HC). Since \[(\bigoplus\nolimits_{\alpha \in A} R)^\wedge = (\bigoplus\nolimits_{\alpha \in A} R^\wedge)^\wedge\] we may replace \(R\) by \(R^\wedge\) and assume that \(R\) is complete with respect to \(I\) (see Algebra, Lemma 031C). In this case Lemma 05BC tells us the map \((\bigoplus\nolimits_{\alpha \in A} R)^\wedge \to \prod_{\alpha \in A} R\) is universally injective. Thus, by Algebra, Lemma 058P it suffices to show that \(\prod_{\alpha \in A} R\) is flat. By Algebra, Proposition 05CZ (and Algebra, Lemma 05CY) we see that \(\prod_{\alpha \in A} R\) is flat.
Lemma
Let \(A\) be a Noetherian ring. Let \(I\) be an ideal of \(A\). Let \(M\) be a finite \(A\)-module. For every \(p > 0\) there exists a \(c > 0\) such that \(\text{Tor}_p^A(M, A/I^n) \to \text{Tor}_p^A(M, A/I^{n - c})\) is zero for all \(n \geq c\).
Proof
Proof for \(p = 1\). Choose a short exact sequence \(0 \to K \to A^{\oplus t} \to M \to 0\). Then \(\text{Tor}_1^A(M, A/I^n) = K \cap (I^n)^{\oplus t}/I^nK\). By Artin-Rees (Algebra, Lemma 00IN) there is a constant \(c \geq 0\) such that \(K \cap (I^n)^{\oplus t} \subset I^{n - c}K\) for \(n \geq c\). Thus the result for \(p = 1\). For \(p > 1\) we have \(\text{Tor}_p^A(M, A/I^n) = \text{Tor}^A_{p - 1}(K, A/I^n)\). Thus the lemma follows by induction.
Lemma
Let \(A\) be a Noetherian ring. Let \(I\) be an ideal of \(A\). Let \((M_n)\) be an inverse system of \(A\)-modules such that
\(M_n\) is a flat \(A/I^n\)-module,
\(M_{n + 1} \to M_n\) is surjective.
Then \(M = \lim M_n\) is a flat \(A\)-module and \(Q \otimes_A M = \lim Q \otimes_A M_n\) for every finite \(A\)-module \(Q\).
Proof
We first show that \(Q \otimes_A M = \lim Q \otimes_A M_n\) for every finite \(A\)-module \(Q\). Choose a resolution \(F_2 \to F_1 \to F_0 \to Q \to 0\) by finite free \(A\)-modules \(F_i\). Then \[F_2 \otimes_A M_n \to F_1 \otimes_A M_n \to F_0 \otimes_A M_n\] is a chain complex whose homology in degree \(0\) is \(Q \otimes_A M_n\) and whose homology in degree \(1\) is \[\text{Tor}_1^A(Q, M_n) = \text{Tor}_1^A(Q, A/I^n) \otimes_{A/I^n} M_n\] as \(M_n\) is flat over \(A/I^n\). By Lemma 0911 we see that this system is essentially constant (with value \(0\)). It follows from Homology, Lemma 070E that \(\lim Q \otimes_A M_n = \Coker(\lim F_1 \otimes_A M_n \to \lim F_0 \otimes_A M_n)\). Since \(F_i\) is finite free this equals \(\Coker(F_1 \otimes_A M \to F_0 \otimes_A M) = Q \otimes_A M\).
Next, let \(Q \to Q'\) be an injective map of finite \(A\)-modules. We have to show that \(Q \otimes_A M \to Q' \otimes_A M\) is injective (Algebra, Lemma 00HD). By the above we see \[\Ker(Q \otimes_A M \to Q' \otimes_A M) = \Ker(\lim Q \otimes_A M_n \to \lim Q' \otimes_A M_n).\] For each \(n\) we have an exact sequence \[\text{Tor}_1^A(Q', M_n) \to \text{Tor}_1^A(Q'', M_n) \to Q \otimes_A M_n \to Q' \otimes_A M_n\] where \(Q'' = \Coker(Q \to Q')\). Above we have seen that the inverse systems of Tor’s are essentially constant with value \(0\). It follows from Homology, Lemma 070E that the inverse limit of the right most maps is injective.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(M\) be an \(R\)-module. Assume
\(I\) is finitely generated,
\(R/I\) is Noetherian,
\(M/IM\) is flat over \(R/I\),
\(\text{Tor}_1^R(M, R/I) = 0\).
Then the \(I\)-adic completion \(R^\wedge\) is a Noetherian ring and \(M^\wedge\) is flat over \(R^\wedge\).
Proof
By Algebra, Lemma 051C the modules \(M/I^nM\) are flat over \(R/I^n\) for all \(n\). By Algebra, Lemma 05GG we have (a) \(R^\wedge\) and \(M^\wedge\) are \(I\)-adically complete and (b) \(R/I^n = R^\wedge/I^nR^\wedge\) for all \(n\). By Algebra, Lemma 05GH the ring \(R^\wedge\) is Noetherian. Applying Lemma 0912 we conclude that \(M^\wedge = \lim M/I^nM\) is flat as an \(R^\wedge\)-module.
The Koszul complex
We define the Koszul complex as follows.
Definition
Let \(R\) be a ring. Let \(\varphi : E \to R\) be an \(R\)-module map. The Koszul complex \(K_\bullet(\varphi)\) associated to \(\varphi\) is the commutative differential graded algebra defined as follows:
the underlying graded algebra is the exterior algebra \(K_\bullet(\varphi) = \wedge(E)\),
the differential \(d : K_\bullet(\varphi) \to K_\bullet(\varphi)\) is the unique derivation such that \(d(e) = \varphi(e)\) for all \(e \in E = K_1(\varphi)\).
Explicitly, if \(e_1 \wedge \ldots \wedge e_n\) is one of the generators of degree \(n\) in \(K_\bullet(\varphi)\), then \[d(e_1 \wedge \ldots \wedge e_n) = \sum\nolimits_{i = 1, \ldots, n} (-1)^{i + 1} \varphi(e_i)e_1 \wedge \ldots \wedge \widehat{e_i} \wedge \ldots \wedge e_n.\] It is straightforward to see that this gives a well defined derivation on the tensor algebra, which annihilates \(e \otimes e\) and hence factors through the exterior algebra.
We often assume that \(E\) is a finite free module, say \(E = R^{\oplus n}\). In this case the map \(\varphi\) is given by a sequence of elements \(f_1, \ldots, f_n \in R\).
Definition
Let \(R\) be a ring and let \(f_1, \ldots, f_r \in R\). The Koszul complex on \(f_1, \ldots, f_r\) is the Koszul complex associated to the map \((f_1, \ldots, f_r) : R^{\oplus r} \to R\). Notation \(K_\bullet(f_\bullet)\), \(K_\bullet(f_1, \ldots, f_r)\), \(K_\bullet(R, f_1, \ldots, f_r)\), or \(K_\bullet(R, f_\bullet)\).
Of course, if \(E\) is finite locally free, then \(K_\bullet(\varphi)\) is locally on \(\Spec(R)\) isomorphic to a Koszul complex \(K_\bullet(f_1, \ldots, f_r)\). This complex has many interesting formal properties.
Lemma
Let \(\varphi : E \to R\) and \(\varphi' : E' \to R\) be \(R\)-module maps. Let \(\psi : E \to E'\) be an \(R\)-module map such that \(\varphi' \circ \psi = \varphi\). Then \(\psi\) induces a homomorphism of differential graded algebras \(K_\bullet(\varphi) \to K_\bullet(\varphi')\).
Proof
This is immediate from the definitions.
Lemma
Let \(f_1, \ldots, f_r \in R\) be a sequence. Let \((x_{ij})\) be an invertible \(r \times r\)-matrix with coefficients in \(R\). Then the complexes \(K_\bullet(f_\bullet)\) and \[K_\bullet(\sum x_{1j}f_j, \sum x_{2j}f_j, \ldots, \sum x_{rj}f_j)\] are isomorphic.
Proof
Set \(g_i = \sum x_{ij}f_j\). The matrix \((x_{ji})\) gives an isomorphism \(x : R^{\oplus r} \to R^{\oplus r}\) such that \((g_1, \ldots, g_r) = (f_1, \ldots, f_r) \circ x\). Hence this follows from the functoriality of the Koszul complex described in Lemma 0624.
Lemma
Let \(R\) be a ring. Let \(\varphi : E \to R\) be an \(R\)-module map. Let \(e \in E\) with image \(f = \varphi(e)\) in \(R\). Then \[f = de + ed\] as endomorphisms of \(K_\bullet(\varphi)\).
Proof
This is true because \(d(ea) = d(e)a - ed(a) = fa - ed(a)\).
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\) be a sequence. Multiplication by \(f_i\) on \(K_\bullet(f_\bullet)\) is homotopic to zero, and in particular the cohomology modules \(H_i(K_\bullet(f_\bullet))\) are annihilated by the ideal \((f_1, \ldots, f_r)\).
Proof
Special case of Lemma 0626.
In Derived Categories, Section 014D we defined the cone of a morphism of cochain complexes. The cone \(C(f)_\bullet\) of a morphism of chain complexes \(f : A_\bullet \to B_\bullet\) is the complex \(C(f)_\bullet\) given by \(C(f)_n = B_n \oplus A_{n - 1}\) and differential [0627]\[\begin{equation} d_{C(f), n} = \left( \begin{matrix} d_{B, n} & f_{n - 1} \\ 0 & -d_{A, n - 1} \end{matrix} \right) \end{equation}\] It comes equipped with canonical morphisms of complexes \(i : B_\bullet \to C(f)_\bullet\) and \(p : C(f)_\bullet \to A_\bullet[-1]\) induced by the obvious maps \(B_n \to C(f)_n \to A_{n - 1}\).
Lemma
Let \(R\) be a ring. Let \(\varphi : E \to R\) be an \(R\)-module map. Let \(f \in R\). Set \(E' = E \oplus R\) and define \(\varphi' : E' \to R\) by \(\varphi\) on \(E\) and multiplication by \(f\) on \(R\). The complex \(K_\bullet(\varphi')\) is isomorphic to the cone of the map of complexes \[f : K_\bullet(\varphi) \longrightarrow K_\bullet(\varphi).\]
Proof
Denote \(e_0 \in E'\) the element \(1 \in R \subset R \oplus E\). By our definition of the cone above we see that \[C(f)_n = K_n(\varphi) \oplus K_{n - 1}(\varphi) = \wedge^n(E) \oplus \wedge^{n - 1}(E) = \wedge^n(E')\] where in the last \(=\) we map \((0, e_1 \wedge \ldots \wedge e_{n - 1})\) to \(e_0 \wedge e_1 \wedge \ldots \wedge e_{n - 1}\) in \(\wedge^n(E')\). A computation shows that this isomorphism is compatible with differentials. Namely, this is clear for elements of the first summand as \(\varphi'|_E = \varphi\) and \(d_{C(f)}\) restricted to the first summand is just \(d_{K_\bullet(\varphi)}\). On the other hand, if \(e_1 \wedge \ldots \wedge e_{n - 1}\) is in the second summand, then \[d_{C(f)}(0, e_1 \wedge \ldots \wedge e_{n - 1}) = fe_1 \wedge \ldots \wedge e_{n - 1} - d_{K_\bullet(\varphi)}(e_1 \wedge \ldots \wedge e_{n - 1})\] and on the other hand \[\begin{align*} & d_{K_\bullet(\varphi')}(0, e_0 \wedge e_1 \wedge \ldots \wedge e_{n - 1}) \\ & = \sum\nolimits_{i = 0, \ldots, n - 1} (-1)^i \varphi'(e_i)e_0 \wedge \ldots \wedge \widehat{e_i} \wedge \ldots \wedge e_{n - 1} \\ & = fe_1 \wedge \ldots \wedge e_{n - 1} + \sum\nolimits_{i = 1, \ldots, n - 1} (-1)^i \varphi(e_i)e_0 \wedge \ldots \wedge \widehat{e_i} \wedge \ldots \wedge e_{n - 1} \\ & = fe_1 \wedge \ldots \wedge e_{n - 1} - e_0 \left(\sum\nolimits_{i = 1, \ldots, n - 1} (-1)^{i + 1} \varphi(e_i)e_1 \wedge \ldots \wedge \widehat{e_i} \wedge \ldots \wedge e_{n - 1}\right) \end{align*}\] which is the image of the result of the previous computation.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r\) be a sequence of elements of \(R\). The complex \(K_\bullet(f_1, \ldots, f_r)\) is isomorphic to the cone of the map of complexes \[f_r : K_\bullet(f_1, \ldots, f_{r - 1}) \longrightarrow K_\bullet(f_1, \ldots, f_{r - 1}).\]
Proof
Special case of Lemma 0628.
Lemma
Let \(R\) be a ring. Let \(A_\bullet\) be a complex of \(R\)-modules. Let \(f, g \in R\). Let \(C(f)_\bullet\) be the cone of \(f : A_\bullet \to A_\bullet\). Define similarly \(C(g)_\bullet\) and \(C(fg)_\bullet\). Then \(C(fg)_\bullet\) is homotopy equivalent to the cone of a map \[C(f)_\bullet[1] \longrightarrow C(g)_\bullet\]
Proof
We first prove this if \(A_\bullet\) is the complex consisting of \(R\) placed in degree \(0\). In this case the complex \(C(f)_\bullet\) is the complex \[\ldots \to 0 \to R \xrightarrow{f} R \to 0 \to \ldots\] with \(R\) placed in (homological) degrees \(1\) and \(0\). The map of complexes we use is \[\xymatrix{ 0 \ar[r] \ar[d] & 0 \ar[r] \ar[d] & R \ar[r]^f \ar[d]^1 & R \ar[r] \ar[d] & 0 \ar[d] \\ 0 \ar[r] & R \ar[r]^g & R \ar[r] & 0 \ar[r] & 0 }\] The cone of this is the chain complex consisting of \(R^{\oplus 2}\) placed in degrees \(1\) and \(0\) and differential (0627) \[\left( \begin{matrix} g & 1 \\ 0 & -f \end{matrix} \right) : R^{\oplus 2} \longrightarrow R^{\oplus 2}\] To see this chain complex is homotopic to \(C(fg)_\bullet\), i.e., to \(R \xrightarrow{fg} R\), consider the maps of complexes \[\xymatrix{ R \ar[d]_{(1, -g)} \ar[r]_{fg} & R \ar[d]^{(0, 1)} \\ R^{\oplus 2} \ar[r] & R^{\oplus 2} } \quad\quad \xymatrix{ R^{\oplus 2} \ar[d]_{(1, 0)} \ar[r] & R^{\oplus 2} \ar[d]^{(f, 1)} \\ R \ar[r]^{fg} & R }\] with obvious notation. The composition of these two maps in one direction is the identity on \(C(fg)_\bullet\), but in the other direction it isn’t the identity. We omit writing out the required homotopy.
To see the result holds in general, we use that we have a functor \(K_\bullet \mapsto \text{Tot}(A_\bullet \otimes_R K_\bullet)\) on the category of complexes which is compatible with homotopies and cones. Then we write \(C(f)_\bullet\) and \(C(g)_\bullet\) as the total complex of the double complexes \[(R \xrightarrow{f} R) \otimes_R A_\bullet \quad\text{and}\quad (R \xrightarrow{g} R) \otimes_R A_\bullet\] and in this way we deduce the result from the special case discussed above. Some details omitted.
Lemma
Let \(R\) be a ring. Let \(\varphi : E \to R\) be an \(R\)-module map. Let \(f, g \in R\). Set \(E' = E \oplus R\) and define \(\varphi'_f, \varphi'_g, \varphi'_{fg} : E' \to R\) by \(\varphi\) on \(E\) and multiplication by \(f, g, fg\) on \(R\). The complex \(K_\bullet(\varphi'_{fg})\) is homotopy equivalent to the cone of a map of complexes \[K_\bullet(\varphi'_f)[1] \longrightarrow K_\bullet(\varphi'_g).\]
Proof
By Lemma 0628 the complex \(K_\bullet(\varphi'_f)\) is isomorphic to the cone of multiplication by \(f\) on \(K_\bullet(\varphi)\) and similarly for the other two cases. Hence the lemma follows from Lemma 062A.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_{r - 1}\) be a sequence of elements of \(R\). Let \(f, g \in R\). The complex \(K_\bullet(f_1, \ldots, f_{r - 1}, fg)\) is homotopy equivalent to the cone of a map of complexes \[K_\bullet(f_1, \ldots, f_{r - 1}, f)[1] \longrightarrow K_\bullet(f_1, \ldots, f_{r - 1}, g)\]
Proof
Special case of Lemma 062B.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r\), \(g_1, \ldots, g_s\) be elements of \(R\). Then there is an isomorphism of Koszul complexes \[K_\bullet(R, f_1, \ldots, f_r, g_1, \ldots, g_s) = \text{Tot}(K_\bullet(R, f_1, \ldots, f_r) \otimes_R K_\bullet(R, g_1, \ldots, g_s)).\]
Proof
Omitted. Hint: If \(K_\bullet(R, f_1, \ldots, f_r)\) is generated as a differential graded algebra by \(x_1, \ldots, x_r\) with \(\text{d}(x_i) = f_i\) and \(K_\bullet(R, g_1, \ldots, g_s)\) is generated as a differential graded algebra by \(y_1, \ldots, y_s\) with \(\text{d}(y_j) = g_j\), then we can think of \(K_\bullet(R, f_1, \ldots, f_r, g_1, \ldots, g_s)\) as the differential graded algebra generated by the sequence of elements \(x_1, \ldots, x_r, y_1, \ldots, y_s\) with \(\text{d}(x_i) = f_i\) and \(\text{d}(y_j) = g_j\).
The extended alternating Čech complex
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). The extended alternating Čech complex of \(R\) is the cochain complex \[R \to \bigoplus\nolimits_{i_0} R_{f_{i_0}} \to \bigoplus\nolimits_{i_0 < i_1} R_{f_{i_0}f_{i_1}} \to \ldots \to R_{f_1\ldots f_r}\] where \(R\) is in degree \(0\), the term \(\bigoplus_{i_0} R_{f_{i_0}}\) is in degre \(1\), and so on. The maps are defined as follows
The map \(R \to \bigoplus\nolimits_{i_0} R_{f_{i_0}}\) is given by the canonical maps \(R \to R_{f_{i_0}}\).
Given \(1 \leq i_0 < \ldots < i_{p + 1} \leq r\) and \(0 \leq j \leq p + 1\) we have the canonical localization map \[R_{f_{i_0} \ldots \hat f_{i_j} \ldots f_{i_{p + 1}}} \to R_{f_{i_0} \ldots f_{i_{p + 1}}}\]
The differentials use the canonical maps of (2) with sign \((-1)^j\).
If \(M\) is any \(R\)-module, the extended alternating Čech complex of \(M\) is the similarly constructed cochain complex \[M \to \bigoplus\nolimits_{i_0} M_{f_{i_0}} \to \bigoplus\nolimits_{i_0 < i_1} M_{f_{i_0}f_{i_1}} \to \ldots \to M_{f_1\ldots f_r}\] where \(M\) is in degree \(0\) as before.
Lemma
The extended alternating Čech complexes defined above are complexes of \(R\)-modules.
Proof
Omitted.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). Let \(M\) be an \(R\)-module. The extended alternating Čech complex of \(M\) is the tensor product over \(R\) of \(M\) with the extended alternating Čech complex of \(R\).
Proof
Omitted.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). Let \(M\) be an \(R\)-module. Let \(R \to S\) be a ring map, denote \(g_1, \ldots, g_r \in S\) the images of \(f_1, \ldots, f_r\), and set \(N = M \otimes_R S\). The extended alternating Čech complex constructed using \(S\), \(g_1, \ldots, g_r\), and \(N\) is the tensor product of the extended alternating Čech complex of \(M\) with \(S\) over \(R\).
Proof
Omitted.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). Let \(M\) be an \(R\)-module. If there exists an \(i \in \{1, \ldots, r\}\) such that \(f_i\) is a unit, then the extended alternating Čech complex of \(M\) is homotopy equivalent to \(0\).
Proof
We will use the following notation: a cochain \(x\) of degree \(p + 1\) in the extended alternating Čech complex of \(M\) is \(x = (x_{i_0 \ldots i_p})\) where \(x_{i_0 \ldots i_p}\) is in \(M_{f_{i_0} \ldots f_{i_p}}\). With this notation we have \[d(x)_{i_0 \ldots i_{p + 1}} = \sum\nolimits_j (-1)^j x_{i_0 \ldots \hat i_j \ldots i_{p + 1}}\] As homotopy we use the maps \[h : \text{cochains of degree }p + 2 \to \text{cochains of degree }p + 1\] given by the rule \[h(x)_{i_0 \ldots i_p} = 0 \text{ if } i \in \{i_0, \ldots, i_p\} \text{ and } h(x)_{i_0 \ldots i_p} = (-1)^j x_{i_0 \ldots i_j i i_{j + 1} \ldots i_p} \text{ if not}\] Here \(j\) is the unique index such that \(i_j < i < i_{j + 1}\) in the second case; also, since \(f_i\) is a unit we have the equality \[M_{f_{i_0} \ldots f_{i_p}} = M_{f_{i_0} \ldots f_{i_j} f_i f_{i_{j + 1}} \ldots f_{i_p}}\] which we can use to make sense of thinking of \((-1)^j x_{i_0 \ldots i_j i i_{j + 1} \ldots i_p}\) as an element of \(M_{f_{i_0} \ldots f_{i_p}}\). We will show by a computation that \(d h + h d\) equals the negative of the identity map which finishes the proof. To do this fix \(x\) a cochain of degree \(p + 1\) and let \(1 \leq i_0 < \ldots < i_p \leq r\).
Case I: \(i \in \{i_0, \ldots, i_p\}\). Say \(i = i_t\). Then we have \(h(d(x))_{i_0 \ldots i_p} = 0\). On the other hand we have \[d(h(x))_{i_0 \ldots i_p} = \sum (-1)^j h(x)_{i_0 \ldots \hat i_j \ldots i_p} = (-1)^t h(x)_{i_0 \ldots \hat i \ldots i_p} = (-1)^t (-1)^{t - 1} x_{i_0 \ldots i_p}\] Thus \((dh + hd)(x)_{i_0 \ldots i_p} = -x_{i_0 \ldots i_p}\) as desired.
Case II: \(i \not \in \{i_0, \ldots, i_p\}\). Let \(j\) be such that \(i_j < i < i_{j + 1}\). Then we see that \[\begin{align*} h(d(x))_{i_0 \ldots i_p} & = (-1)^j d(x)_{i_0 \ldots i_j i i_{j + 1} \ldots i_p} \\ & = \sum\nolimits_{j' \leq j} (-1)^{j + j'} x_{i_0 \ldots \hat i_{j'} \ldots i_j i i_{j + 1} \ldots i_p} - x_{i_0 \ldots i_p} \\ & + \sum\nolimits_{j' > j} (-1)^{j + j' + 1} x_{i_0 \ldots i_j i i_{j + 1} \ldots \hat i_{j'} \ldots i_p} \end{align*}\] On the other hand we have \[\begin{align*} d(h(x))_{i_0 \ldots i_p} & = \sum\nolimits_{j'} (-1)^{j'} h(x)_{i_0 \ldots \hat i_{j'} \ldots i_p} \\ & = \sum\nolimits_{j' \leq j} (-1)^{j' + j - 1} x_{i_0 \ldots \hat i_{j'} \ldots i_j i i_{j + 1} \ldots i_p} \\ & + \sum\nolimits_{j' > j} (-1)^{j' + j} x_{i_0 \ldots i_j i i_{j + 1} \ldots \hat i_{j'} \ldots i_p} \end{align*}\] Adding these up we obtain \((dh + hd)(x)_{i_0 \ldots i_p} = - x_{i_0 \ldots i_p}\) as desired.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). Let \(M\) be an \(R\)-module. Let \(H^q\) be the \(q\)th cohomology module of the extended alternation Čech complex of \(M\). Then
\(H^q = 0\) if \(q \not \in [0, r]\),
for \(x \in H^i\) there exists an \(n \geq 1\) such that \(f_i^n x = 0\) for \(i = 1, \ldots, r\),
the support of \(H^q\) is contained in \(V(f_1, \ldots, f_r)\),
if there is an \(f \in (f_1, \ldots, f_r)\) which acts invertibly on \(M\), then \(H^q = 0\).
Proof
Part (1) follows from the fact that the extended alternating Čech complex is zero in degrees \(< 0\) and \(> r\). To prove (2) it suffices to show that for each \(i\) there exists an \(n \geq 1\) such that \(f_i^n x = 0\). To see this it suffices to show that \((H^q)_{f_i} = 0\). Since localization is exact, \((H^q)_{f_i}\) is the \(q\)th cohomology module of the localization of the extended alternating complex of \(M\) at \(f_i\). By Lemma 0G6I this localization is the extended alternating Čech complex of \(M_{f_i}\) over \(R_{f_i}\) with respect to the images of \(f_1, \ldots, f_r\) in \(R_{f_i}\). Thus we reduce to showing that \(H^q\) is zero if \(f_i\) is invertible, which follows from Lemma 0G6J. Part (3) follows from the observation that \((H^q)_{f_i} = 0\) for all \(i\) that we just proved. To see part (4) note that in this case \(f\) acts invertibly on \(H^q\) and \(H^q\) is supported on \(V(f)\) by (3). This forces \(H^q\) to be zero (small detail omitted).
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\). The extended alternating Čech complex \[R \to \bigoplus\nolimits_{i_0} R_{f_{i_0}} \to \bigoplus\nolimits_{i_0 < i_1} R_{f_{i_0}f_{i_1}} \to \ldots \to R_{f_1\ldots f_r}\] is a colimit of the Koszul complexes \(K(R, f_1^n, \ldots, f_r^n)\); see proof for a precise statement. For an \(R\)-module \(M\), the same statement holds with the Koszul complexes replaced by \(M \otimes_R K(R, f_1^n, \ldots, f_r^n)\) and the localization terms replaced by the corresponding localizations of \(M\).
Proof
We urge the reader to prove this for themselves. Denote \(K(R, f_1^n, \ldots, f_r^n)\) the Koszul complex of Definition 0623 viewed as a cochain complex sitting in degrees \(0, \ldots, r\). Thus we have \[K(R, f_1^n, \ldots, f_r^n) : 0 \to \wedge^r(R^{\oplus r}) \to \wedge^{r - 1}(R^{\oplus r}) \to \ldots \to R^{\oplus r} \to R \to 0\] with the term \(\wedge^r(R^{\oplus r})\) sitting in degree \(0\). Let \(e^n_1, \ldots, e^n_r\) be the standard basis of \(R^{\oplus r}\). Then the elements \(e^n_{j_1} \wedge \ldots \wedge e^n_{j_{r - p}}\) for \(1 \leq j_1 < \ldots < j_{r - p} \leq r\) form a basis for the term in degree \(p\) of the Koszul complex. Further, observe that \[d(e^n_{j_1} \wedge \ldots \wedge e^n_{j_{r - p}}) = \sum (-1)^{a + 1} f_{j_a}^n e^n_{j_1} \wedge \ldots \wedge \hat e^n_{j_a} \wedge \ldots \wedge e^n_{j_{r - p}}\] by our construction of the Koszul complex in Section 0621. The transition maps of our system \[K(R, f_1^n, \ldots, f_r^n) \to K(R, f_1^{n + 1}, \ldots, f_r^{n + 1})\] are given by the rule \[e^n_{j_1} \wedge \ldots \wedge e^n_{j_{r - p}} \longmapsto f_{i_0} \ldots f_{i_{p - 1}} e^{n + 1}_{j_1} \wedge \ldots \wedge e^{n + 1}_{j_{r - p}}\] where the indices \(1 \leq i_0 < \ldots < i_{p - 1} \leq r\) are such that \(\{1, \ldots r\} = \{i_0, \ldots, i_{p - 1}\} \amalg \{j_1, \ldots, j_{r - p}\}\). We omit the short computation that shows this is compatible with differentials. Observe that the transition maps are always \(1\) in degree \(0\) and equal to \(f_1 \ldots f_r\) in degree \(r\).
Denote \(K^p(R, f_1^n, \ldots, f_r^n)\) the term of degree \(p\) in the Koszul complex. Observe that for any \(f \in R\) we have \[R_f = \colim (R \xrightarrow{f} R \xrightarrow{f} R \to \ldots )\] Hence we see that in degree \(p\) we obtain \[\colim K^p(R, f_1^n, \ldots f_r^n) = \bigoplus\nolimits_{1 \leq i_0 < \ldots < i_{p - 1} \leq r} R_{f_{i_0} \ldots f_{i_{p - 1}}}\] Here the element \(e^n_{j_1} \wedge \ldots \wedge e^n_{j_{r - p}}\) of the Koszul complex above maps in the colimit to the element \((f_{i_0} \ldots f_{i_{p - 1}})^{-n}\) in the summand \(R_{f_{i_0} \ldots f_{i_{p - 1}}}\) where the indices are chosen such that \(\{1, \ldots r\} = \{i_0, \ldots, i_{p - 1}\} \amalg \{j_1, \ldots, j_{r - p}\}\). Thus the differential on this complex is given by \[d(1\text{ in }R_{f_{i_0} \ldots f_{i_{p - 1}}}) = \sum\nolimits_{i \not \in \{i_0, \ldots, i_{p - 1}\}} (-1)^{i - t}\text{ in } R_{f_{i_0} \ldots f_{i_t} f_i f_{i_{t + 1}} \ldots f_{i_{p - 1}}}\] Thus if we consider the map of complexes given in degree \(p\) by the map \[\bigoplus\nolimits_{1 \leq i_0 < \ldots < i_{p - 1} \leq r} R_{f_{i_0} \ldots f_{i_{p - 1}}} \longrightarrow \bigoplus\nolimits_{1 \leq i_0 < \ldots < i_{p - 1} \leq r} R_{f_{i_0} \ldots f_{i_{p - 1}}}\] determined by the rule \[1\text{ in }R_{f_{i_0} \ldots f_{i_{p - 1}}} \longmapsto (-1)^{i_0 + \ldots + i_{p - 1} + p}\text{ in }R_{f_{i_0} \ldots f_{i_{p - 1}}}\] then we get an isomorphism of complexes from \(\colim K(R, f_1^n, \ldots, f_r^n)\) to the extended alternating Čech complex defined in this section. We omit the verification that the signs work out.
The module statement follows by tensoring the directed system with \(M\), using that tensor product commutes with colimits, and applying Lemma 0G6H.
Koszul regular sequences
Please take a look at Algebra, Sections 0AUH, 061M, and 00LE before looking at this one.
Definition
Let \(R\) be a ring. Let \(r \geq 0\) and let \(f_1, \ldots, f_r \in R\) be a sequence of elements. Let \(M\) be an \(R\)-module. The sequence \(f_1, \ldots, f_r\) is called
\(M\)-Koszul-regular if \(H_i(K_\bullet(f_1, \ldots, f_r) \otimes_R M) = 0\) for all \(i \not = 0\),
\(M\)-\(H_1\)-regular if \(H_1(K_\bullet(f_1, \ldots, f_r) \otimes_R M) = 0\),
Koszul-regular if \(H_i(K_\bullet(f_1, \ldots, f_r)) = 0\) for all \(i \not = 0\), and
\(H_1\)-regular if \(H_1(K_\bullet(f_1, \ldots, f_r)) = 0\).
We will see in Lemmas 062F, 0CEM, and 062I that for elements \(f_1, \ldots, f_r\) of a ring \(R\) we have the following implications \[\begin{align*} f_1, \ldots, f_r\text{ is a regular sequence} & \Rightarrow f_1, \ldots, f_r\text{ is a Koszul-regular sequence} \\ & \Rightarrow f_1, \ldots, f_r\text{ is an }H_1\text{-regular sequence} \\ & \Rightarrow f_1, \ldots, f_r\text{ is a quasi-regular sequence.} \end{align*}\] In general none of these implications can be reversed, but if \(R\) is a Noetherian local ring and \(f_1, \ldots, f_r \in \mathfrak m_R\), then the four conditions are all equivalent (Lemma 09CC). If \(f = f_1 \in R\) is a length \(1\) sequence and \(f\) is not a unit of \(R\) then it is clear that the following are all equivalent
\(f\) is a regular sequence of length one,
\(f\) is a Koszul-regular sequence of length one, and
\(f\) is a \(H_1\)-regular sequence of length one.
It is also clear that these imply that \(f\) is a quasi-regular sequence of length one. But there do exist quasi-regular sequences of length \(1\) which are not regular sequences. Namely, let \[R = k[x, y_0, y_1, \ldots]/(xy_0, xy_1 - y_0, xy_2 - y_1, \ldots)\] and let \(f\) be the image of \(x\) in \(R\). Then \(f\) is a zerodivisor, but \(\bigoplus_{n \geq 0} (f^n)/(f^{n + 1}) \cong k[x]\) is a polynomial ring.
Lemma
Let \(R\) be a ring, \(M\) an \(R\)-module, and \(f_1, \ldots, f_r \in R\) such that for \(i = 1, \ldots, r\) multiplication by \(f_i\) is injective on \(M/(f_1, \ldots, f_{i - 1})M\). Then \(f_1, \ldots, f_r\) is \(M\)-Koszul regular. In particular, an \(M\)-regular sequence is \(M\)-Koszul-regular and any regular sequence is Koszul-regular.
Proof
Let \(R\), \(M\), \(f_1, \ldots, f_r\) be as in the first sentence of the lemma. If \(r = 1\), it is immediate that \(f_1\) is \(M\)-Koszul-regular. Assume \(r > 1\). Since \(f_1\) is a nonzerodivisor on \(M\), we obtain a short exact sequence of complexes: \[0 \to K_\bullet(f_2, \ldots, f_r) \otimes M \xrightarrow{f_1} K_\bullet(f_2, \ldots, f_r) \otimes M \to K_\bullet(\overline{f}_2, \ldots, \overline{f}_r) \otimes M/f_1M \to 0\] Here \(\overline{f}_i\) is the image of \(f_i\) in \(R/(f_1)\). By Lemma 0629 the complex \(K_\bullet(f_1, \ldots, f_r)\) is isomorphic to the cone of multiplication by \(f_1\) on \(K_\bullet(f_2, \ldots, f_r)\). Thus \(K_\bullet(R, f_1, \ldots, f_r) \otimes M\) is isomorphic to the cone on the first map. Hence \(K_\bullet(\overline{f}_2, \ldots, \overline{f}_r) \otimes M/f_1M\) is quasi-isomorphic to \(K_\bullet(f_1, \ldots, f_r) \otimes M\). As \(R/(f_1)\), \(M/f_1M\), \(\overline{f}_2, \ldots, \overline{f}_r\) satisfy the conditions of the lemma, by induction we conclude this complex is acyclic in postive degrees. This finishes the proof of the first statement. The second statement immediately follows from the first.
Lemma
Let \(S\) be a graded ring. Let \(f_1,\ldots,f_s\in S\) be homogeneous elements of the same degree \(d\), where \(s\geq1\), and assume that this sequence is Koszul-regular. Set \(J=(f_1,\ldots,f_s)\). For \(0\leq q\leq s-1\), let \[L^q=\bigoplus_{1\leq i_0<\ldots<i_q\leq s}S(-d(q+1))\] with basis elements \(e_{i_0\ldots i_q}\), and define \[\begin{align*} d(e_{i_0\ldots i_{q+1}})&= \sum_{j=0}^{q+1}(-1)^j f_{i_j} e_{i_0\ldots\widehat{i_j}\ldots i_{q+1}},\\ \epsilon(e_i)&=f_i. \end{align*}\] Then \[0\longrightarrow L^{s-1}\longrightarrow\ldots\longrightarrow L^0\xrightarrow{\epsilon}J\longrightarrow0\] is a graded free resolution with degree-zero differentials. Consequently, for every graded \(S\)-module \(M\), the module \(\text{GrExt}^q_S(J,M)\) is the \(q\)th cohomology module of \[\text{GrHom}(L^0,M)\longrightarrow \text{GrHom}(L^1,M)\longrightarrow\ldots\longrightarrow \text{GrHom}(L^{s-1},M).\]
Proof
The shifts make the displayed differentials homogeneous of degree zero. Before the final arrow to \(J\), the displayed complex is the positive homological truncation of the Koszul complex \(K_\bullet(S,f_1,\ldots,f_s)\). Koszul regularity says that this complex has no homology in positive degrees, while its zeroth homology is \(S/J\). The image of \(K_1\to K_0=S\) is \(J\), so removing \(S\to S/J\) gives the asserted resolution of \(J\). Applying internal graded Hom and using Algebra, Definition algebra-definition-graded-ext proves the final assertion.
Lemma
Let \(S\) be a graded ring and let \(f_1,\ldots,f_c\in S\) be a homogeneous regular sequence of positive degrees \(m_1,\ldots,m_c\). Put \[A=S/(f_1,\ldots,f_c),\qquad m=m_1+\ldots+m_c.\] For \(0\leq q\leq c\), let \[K_q=\bigoplus_{1\leq i_1<\ldots<i_q\leq c} S(-m_{i_1}-\ldots-m_{i_q}),\] where the summand indexed by the empty tuple is \(K_0=S\), and set \[d(e_{i_1\ldots i_q})= \sum_{j=1}^q(-1)^{j+1}f_{i_j} e_{i_1\ldots\widehat{i_j}\ldots i_q}.\] Then \[0\longrightarrow K_c\longrightarrow\ldots\longrightarrow K_1\longrightarrow K_0\longrightarrow A\longrightarrow0\] is a graded free resolution with degree-zero differentials. For every \(b\in\mathbf Z\) and \(q\geq0\) there are isomorphisms of graded \(S\)-modules \[\text{GrExt}^q_S(A,S(b))\cong \left\{ \begin{matrix} A(m+b) & \text{if }q=c,\\ 0 & \text{if }q\ne c. \end{matrix} \right.\] In particular, \(A\) has projective dimension \(c\) over \(S\).
Proof
The displayed shifts make every differential homogeneous of degree zero. By Lemma 062F, the regular sequence is Koszul-regular, so its augmented Koszul complex is the asserted resolution.
Let \(E=\bigoplus_{i=1}^cS(-m_i)e_i\), so that \(K_q=\wedge^qE\). Using the ordered generator \(e_1\wedge\ldots\wedge e_c\) of \(\wedge^cE\cong S(-m)\), the exterior product gives perfect pairings \[K_q\otimes_SK_{c-q}\longrightarrow S(-m).\] They identify \(\text{GrHom}_S(K_q,S(b))\) with \(K_{c-q}(m+b)\). If the term in cohomological degree \(q\) is multiplied by \((-1)^{q(q-1)/2}\), these identifications carry the Hom differential to the differential of the reversed Koszul complex. Hence the cohomology of \(\text{GrHom}_S(K_\bullet,S(b))\) is the homology of \(K_\bullet(m+b)\) with degrees reversed. Koszul regularity leaves only \(H_0(K_\bullet)=A\), which proves the Ext formula. Since \(A\ne0\), the nonzero top Ext module shows that the displayed free resolution has minimal possible length \(c\).
Lemma
Let \(K\) be a field, let \(r\geq1\), and let \(S=K[t_0,\ldots,t_r]\) with the standard grading. For a graded \(S\)-module \(M\), let \(C^\bullet_+(M)\) be the positive alternating Čech complex with \[C^q_+(M)= \bigoplus_{0\leq i_0<\ldots<i_q\leq r} M_{t_{i_0}\cdots t_{i_q}},\qquad 0\leq q\leq r,\] and the usual alternating localization differential. Set \(B^q(M)=H^q(C^\bullet_+(M))\) and let \({}^\vee\) denote the graded \(K\)-dual of Algebra, Definition algebra-definition-graded-K-dual. There is a functorial isomorphism of graded \(S\)-modules \[\nu_M:\text{GrHom}(M,S(-r-1)) \longrightarrow B^r(M)^\vee.\]
Proof
Put \(\Omega=S(-r-1)\). The class \[\xi=\left[\frac{1}{t_0\cdots t_r}\right] \in B^r(\Omega)_0\] is a basis of the one-dimensional vector space \(B^r(\Omega)_0\). If \(\varphi\in\text{GrHom}(M,\Omega)_n\) and \(h\in B^r(M)_{-n}\), then the map induced by \(\varphi\) sends \(h\) to a unique scalar multiple of \(\xi\). Write \[B^r(\varphi)(h)=\langle h,\varphi\rangle\xi.\] This defines \(\nu_M(\varphi)\in(B^r(M)^\vee)_n\). Composition makes the construction functorial in \(M\), and \(\langle sh,\varphi\rangle=\langle h,s\varphi\rangle\) for homogeneous \(s\in S\) shows that \(\nu_M\) is \(S\)-linear.
Suppose first that \(M=S(m)\). For every \(n\) we have \[\text{GrHom}(S(m),\Omega)_n=S_{n-m-r-1},\] with basis the monomials \(t_0^{\gamma_0}\cdots t_r^{\gamma_r}\) such that \(\gamma_i\geq0\) and \(\sum\gamma_i=n-m-r-1\). On the other hand, the top cohomology is the quotient of \(S(m)_{t_0\cdots t_r}\) by the sum of the localizations in which at least one variable is not inverted. Hence \(B^r(S(m))_{-n}\) has basis \[\frac{1}{t_0^{\beta_0}\cdots t_r^{\beta_r}}, \qquad \beta_i\geq1, \qquad \sum\beta_i=n-m.\] Multiplication followed by passage to top cohomology gives \[\left\langle \frac{1}{t_0^{\beta_0}\cdots t_r^{\beta_r}}, t_0^{\gamma_0}\cdots t_r^{\gamma_r} \right\rangle = \left\{ \begin{matrix} 1 & \text{if }\beta_i=\gamma_i+1\text{ for every }i,\\ 0 & \text{otherwise.} \end{matrix} \right.\] Indeed, in the first case the product is \(\xi\); in every other case one variable has nonnegative exponent and the class vanishes in the top quotient. Thus \(\nu_{S(m)}\) identifies two dual bases and is an isomorphism.
Let \(L=\bigoplus_\alpha S(m_\alpha)\) be an arbitrary graded free module. Localization commutes with direct sums, and direct sums are exact, so \[B^r(L)=\bigoplus_\alpha B^r(S(m_\alpha)).\] In every internal degree, both \(\text{GrHom}(L,\Omega)\) and the graded dual of \(B^r(L)\) are the corresponding products over \(\alpha\). Therefore \(\nu_L\) is an isomorphism, without any restriction on the cardinality of the homogeneous basis.
Finally, choose a graded free presentation \[L^1\longrightarrow L^0\longrightarrow M\longrightarrow0.\] The complex \(C^\bullet_+(-)\) is degreewise exact as a functor of \(M\), because localization is exact. Put \(N=\mathop{\rm Im}(L^1\to L^0)\) and \(P=\mathop{\rm Ker}(L^1\to N)\). Apply the cohomology long exact sequence first to \(0\to P\to L^1\to N\to0\) and then to \(0\to N\to L^0\to M\to0\). Since \(C^q_+=0\) for \(q>r\), the first gives a surjection \(B^r(L^1)\to B^r(N)\) and the second gives an exact sequence \[B^r(N)\longrightarrow B^r(L^0)\longrightarrow B^r(M) \longrightarrow0.\] Consequently, \[B^r(L^1)\longrightarrow B^r(L^0)\longrightarrow B^r(M) \longrightarrow0\] is exact. Applying the exact graded \(K\)-dual and using left exactness of internal graded Hom gives a commutative diagram with exact rows \[\xymatrix{ 0 \ar[r] & \text{GrHom}(M,\Omega) \ar[r] \ar[d]^{\nu_M} & \text{GrHom}(L^0,\Omega) \ar[r] \ar[d]^{\nu_{L^0}} & \text{GrHom}(L^1,\Omega) \ar[d]^{\nu_{L^1}} \\ 0 \ar[r] & B^r(M)^\vee \ar[r] & B^r(L^0)^\vee \ar[r] & B^r(L^1)^\vee. }\] The two right vertical maps are isomorphisms by the free case. Comparing kernels proves that \(\nu_M\) is an isomorphism as well.
Lemma
Let \(K\) be a field, let \(r\geq1\), and let \(S=K[t_0,\ldots,t_r]\) with the standard grading. Put \(\Omega=S(-r-1)\). For a graded \(S\)-module \(M\), use the notation \(C^\bullet_+(M)\) and \(B^q(M)\) of Lemma more-algebra-lemma-top-graded-duality-polynomial-ring, put \(T^q(M)=B^q(M)^\vee\), and let \(a_M:M\to C^0_+(M)\) be the sum of the localization maps. Its image consists of cocycles; denote the induced map by \[\alpha_M:M\longrightarrow B^0(M)\] and consider the augmented complex \[A^\bullet(M)= \left(M\xrightarrow{a_M} C^0_+(M)\longrightarrow\cdots\longrightarrow C^r_+(M)\right),\] where \(M\) is placed in degree \(-1\). Define a complex \(D^\bullet(M)\) by \[D^q(M)=A^{r-q}(M)^\vee\quad(0\leq q\leq r+1), \qquad D^q(M)=0\quad\text{otherwise},\] with differentials obtained by dualizing those of \(A^\bullet(M)\) and reversing their direction. Then the following statements hold.
A short exact sequence \(0\to M\to N\to P\to0\) gives a short exact sequence of complexes \[0\longrightarrow D^\bullet(P)\longrightarrow D^\bullet(N) \longrightarrow D^\bullet(M)\longrightarrow0\] and hence a functorial exact sequence \[\begin{matrix} \ldots\longrightarrow H^q(D^\bullet(P))\longrightarrow H^q(D^\bullet(N))\longrightarrow H^q(D^\bullet(M))\\ \phantom{\ldots}\longrightarrow H^{q+1}(D^\bullet(P)) \longrightarrow\ldots. \end{matrix}\]
There are functorial isomorphisms of graded \(S\)-modules \[H^q(D^\bullet(M))\cong\text{GrExt}^q_S(M,\Omega), \qquad q\geq0,\] compatible with the exact sequences in (1) and the internal grading.
More explicitly, there are functorial isomorphisms \[\text{GrExt}^q_S(M,\Omega)\cong T^{r-q}(M), \qquad 0\leq q<r,\] a functorial exact sequence \[0\longrightarrow\text{GrExt}^r_S(M,\Omega) \longrightarrow T^0(M)\xrightarrow{\alpha_M^\vee}M^\vee \longrightarrow\text{GrExt}^{r+1}_S(M,\Omega) \longrightarrow0,\] and \(\text{GrExt}^q_S(M,\Omega)=0\) for \(q>r+1\).
If \(M\) is a finite graded \(S\)-module, then for \(1\leq p\leq r\) there are functorial perfect pairings of finite dimensional vector spaces \[B^p(M)_0\times \text{GrExt}^{r-p}_S(M,\Omega)_0\longrightarrow K.\] Moreover, \(T^p(M)\) is a finite graded \(S\)-module for every \(p\geq1\), and \(T^0(M)_{\geq n}\) is a finite graded \(S\)-module for all sufficiently large \(n\).
Proof
Localization is exact, so a short exact sequence of graded modules gives a termwise short exact sequence of the augmented complexes \(A^\bullet\). The graded \(K\)-dual is exact by Algebra, Lemma algebra-lemma-graded-K-dual; dualizing and reversing therefore gives the short exact sequence in (1). Its cohomology long exact sequence proves the rest of (1).
Exactness of the graded dual also gives \[H^q(D^\bullet(M))=B^{r-q}(M)^\vee=T^{r-q}(M) \quad\text{for }0\leq q<r.\] At the two remaining terms, the differential \(C^0_+(M)^\vee\to M^\vee\) descends to \(\alpha_M^\vee:T^0(M)\to M^\vee\), and hence \[\begin{align*} H^r(D^\bullet(M))&=\Ker(\alpha_M^\vee),\\ H^{r+1}(D^\bullet(M))&=\Coker(\alpha_M^\vee). \end{align*}\] For a short exact sequence \(0\to M\to N\to P\to0\), the two connecting maps at this splice are the cohomology boundary maps arising from the following commutative diagram with exact rows: \[\xymatrix{ T^1(M)\ar[r] & T^0(P)\ar[r]\ar[d]_{\alpha_P^\vee} & T^0(N)\ar[r]\ar[d]_{\alpha_N^\vee} & T^0(M)\ar[r]\ar[d]_{\alpha_M^\vee} & 0\\ 0\ar[r] & P^\vee\ar[r] & N^\vee\ar[r] & M^\vee\ar[r] & 0. }\] Thus the reversed-complex construction is exactly the intrinsic kernel–cokernel construction represented by this diagram.
We next prove (2). Regard the two families \[M\longmapsto H^q(D^\bullet(M)) \quad\text{and}\quad M\longmapsto\text{GrExt}^q_S(M,\Omega)\] as covariant cohomological delta functors on the opposite of the abelian category of graded \(S\)-modules and degree-zero maps. The first is a delta functor by (1), and the second is one by Algebra, Lemma algebra-lemma-graded-ext-properties.
If \(L=S(m)\), then \(A^\bullet(L)\) is a shift of the extended alternating Čech complex for the regular sequence \(t_0,\ldots,t_r\). More on Algebra, Lemmas 062F and 0913, together with exactness of filtered colimits, show that this complex has cohomology only in its top degree. The same holds for an arbitrary direct sum of twists, because localization and cohomology commute with direct sums. Consequently \[H^q(D^\bullet(L))=0\quad\text{for }q>0\] whenever \(L\) is graded free. Internal graded Ext in positive degree also vanishes on such \(L\).
Every graded module is a quotient of a graded free module by Algebra, Lemma algebra-lemma-graded-free-resolutions. In the opposite category this is an embedding into an object on which every positive member of either delta functor vanishes. Both delta functors are therefore effaceable and hence universal by Homology, Lemma 010T. Finally, Lemma more-algebra-lemma-top-graded-duality-polynomial-ring identifies their degree-zero members by the functorial isomorphism \[\text{GrExt}^0_S(M,\Omega)=\text{GrHom}(M,\Omega) \xrightarrow{\ \nu_M\ }B^r(M)^\vee=H^0(D^\bullet(M)).\] The uniqueness of a universal delta functor, Homology, Lemma 010U, extends this map uniquely to the isomorphisms in (2).
Part (3) follows from (2), the explicit cohomology calculation preceding it, and the fact that \(D^q(M)=0\) for \(q>r+1\).
Suppose now that \(M\) is finite. For \(p\geq1\), the degree-zero part \(B^p(M)_0\) is the cohomology of the associated coherent sheaf on \(\mathbf P^r_K\), computed on the standard affine covering. It is finite dimensional by Cohomology of Schemes, Lemmas 01XD and 01YS. Taking the degree-zero part of the isomorphism \(T^p(M)\cong\text{GrExt}^{r-p}_S(M,\Omega)\) and using biduality gives the perfect pairing in (4). Algebra, Lemma algebra-lemma-graded-ext-properties shows that the module on the right is finite; this proves that \(T^p(M)\) is finite for \(1\leq p\leq r\), while \(T^p(M)=0\) for \(p>r\) by the length of \(C^\bullet_+(M)\). Finally, \(M^\vee_n=0\) for \(n\gg0\), because a finite graded module over the nonnegatively graded ring \(S\) is bounded below. The exact sequence in (3) then identifies \(T^0(M)_n\) with \(\text{GrExt}^r_S(M,\Omega)_n\) for \(n\gg0\), proving the last assertion.
Lemma
A \(M\)-Koszul-regular sequence is \(M\)-\(H_1\)-regular. A Koszul-regular sequence is \(H_1\)-regular.
Proof
This is immediate from the definition.
Lemma
Let \(f_1, \ldots, f_{r - 1} \in R\) be a sequence and \(f, g \in R\). Let \(M\) be an \(R\)-module.
If \(f_1, \ldots, f_{r - 1}, f\) and \(f_1, \ldots, f_{r - 1}, g\) are \(M\)-\(H_1\)-regular then \(f_1, \ldots, f_{r - 1}, fg\) is \(M\)-\(H_1\)-regular too.
If \(f_1, \ldots, f_{r - 1}, f\) and \(f_1, \ldots, f_{r - 1}, g\) are \(M\)-Koszul-regular then \(f_1, \ldots, f_{r - 1}, fg\) is \(M\)-Koszul-regular too.
Proof
By Lemma 062C we have exact sequences \[H_i(K_\bullet(f_1, \ldots, f_{r - 1}, f) \otimes M) \to H_i(K_\bullet(f_1, \ldots, f_{r - 1}, fg) \otimes M) \to H_i(K_\bullet(f_1, \ldots, f_{r - 1}, g) \otimes M)\] for all \(i\).
Lemma
Let \(\varphi : R \to S\) be a flat ring map. Let \(f_1, \ldots, f_r \in R\). Let \(M\) be an \(R\)-module and set \(N = M \otimes_R S\).
If \(f_1, \ldots, f_r\) in \(R\) is an \(M\)-\(H_1\)-regular sequence, then \(\varphi(f_1), \ldots, \varphi(f_r)\) is an \(N\)-\(H_1\)-regular sequence in \(S\).
If \(f_1, \ldots, f_r\) is an \(M\)-Koszul-regular sequence in \(R\), then \(\varphi(f_1), \ldots, \varphi(f_r)\) is an \(N\)-Koszul-regular sequence in \(S\).
Proof
This is true because \(K_\bullet(f_1, \ldots, f_r) \otimes_R S = K_\bullet(\varphi(f_1), \ldots, \varphi(f_r))\) and therefore \((K_\bullet(f_1, \ldots, f_r) \otimes_R M) \otimes_R S = K_\bullet(\varphi(f_1), \ldots, \varphi(f_r)) \otimes_S N\).
Lemma
An \(M\)-\(H_1\)-regular sequence is \(M\)-quasi-regular.
Proof
Let \(R\) be a ring and let \(M\) be an \(R\)-module. Let \(f_1, \ldots, f_r\) be an \(M\)-\(H_1\)-regular sequence. Denote \(J = (f_1, \ldots, f_r)\). The assumption means that we have an exact sequence \[\wedge^2(R^r) \otimes M \to R^{\oplus r} \otimes M \to JM \to 0\] where the first arrow is given by \(e_i \wedge e_j \otimes m \mapsto (f_ie_j - f_je_i) \otimes m\). Tensoring the sequence with \(R/J\) we see that \[JM/J^2M = (R/J)^{\oplus r} \otimes_R M = (M/JM)^{\oplus r}\] is a finite free module. To finish the proof we have to prove for every \(n \geq 2\) the following: if \[\xi = \sum\nolimits_{|I| = n, I = (i_1, \ldots, i_r)} m_I f_1^{i_1} \ldots f_r^{i_r} \in J^{n + 1}M\] then \(m_I \in JM\) for all \(I\). In the next paragraph, we prove \(m_I \in JM\) for \(I = (0, \ldots, 0, n)\) and in the last paragraph we deduce the general case from this special case.
Let \(I = (0, \ldots, 0, n)\). Let \(\xi\) be as above. We can write \(\xi = m_1 f_1 + \ldots + m_{r - 1}f_{r - 1} + m_I f_r^n\). As we have assumed \(\xi \in J^{n + 1}M\), we can also write \(\xi = \sum_{1 \leq i \leq j \leq r - 1} m_{ij}f_if_j + \sum_{1 \leq i \leq r - 1}m'_i f_if_r^n + m'' f_r^{n + 1}\). Then we see that \[\begin{matrix} (m_1 - m_{11}f_1 - m'_1f_r^n)f_1 + \\ (m_2 - m_{12}f_1 - m_{22}f_2 - m'_2f_r^n)f_2 + \\ \ldots + \\ (m_{r - 1} - m_{1 r - 1}f_1 - \ldots - m_{r - 1 r - 1}f_{r - 1} - m'_{r - 1}f_r^n)f_{r - 1} + \\ (m_I - m'' f_r)f_r^n = 0 \end{matrix}\] Since \(f_1, \ldots, f_{r - 1}, f_r^n\) is \(M\)-\(H_1\)-regular by Lemma 062G we see that \(m_I - m'' f_r\) is in the submodule \(f_1M + \ldots + f_{r - 1}M + f_r^nM\). Thus \(m_I \in f_1M + \ldots + f_rM\).
Let \(S = R[x_1, x_2, \ldots, x_r, 1/x_r]\). The ring map \(R \to S\) is faithfully flat, hence \(f_1, \ldots, f_r\) is an \(M\)-\(H_1\)-regular sequence in \(S\), see Lemma 062H. By Lemma 0625 we see that \[g_1 = f_1 - \frac{x_1}{x_r} f_r, \ \ldots, \ g_{r - 1} = f_{r - 1} - \frac{x_{r - 1}}{x_r} f_r, \ g_r = \frac{1}{x_r}f_r\] is an \(M\)-\(H_1\)-regular sequence in \(S\). Finally, note that our element \(\xi\) can be rewritten \[\xi = \sum\nolimits_{|I| = n, I = (i_1, \ldots, i_r)} m_I (g_1 + x_1 g_r)^{i_1} \ldots (g_{r - 1} + x_{r - 1} g_r)^{i_{r - 1}} (x_rg_r)^{i_r}\] and the coefficient of \(g_r^n\) in this expression is \[\sum m_I x_1^{i_1} \ldots x_r^{i_r}\] By the case discussed in the previous paragraph this sum is in \(J(M \otimes_R S)\). Since the monomials \(x_1^{i_1} \ldots x_r^{i_r}\) form part of an \(R\)-basis of \(S\) over \(R\) we conclude that \(m_I \in J\) for all \(I\) as desired.
For nonzero finite modules over Noetherian local rings all of the types of regular sequences introduced so far are equivalent.
Lemma
Let \((R, \mathfrak m)\) be a Noetherian local ring. Let \(M\) be a nonzero finite \(R\)-module. Let \(f_1, \ldots, f_r \in \mathfrak m\). The following are equivalent
\(f_1, \ldots, f_r\) is an \(M\)-regular sequence,
\(f_1, \ldots, f_r\) is a \(M\)-Koszul-regular sequence,
\(f_1, \ldots, f_r\) is an \(M\)-\(H_1\)-regular sequence,
\(f_1, \ldots, f_r\) is an \(M\)-quasi-regular sequence.
In particular the sequence \(f_1, \ldots, f_r\) is a regular sequence in \(R\) if and only if it is a Koszul regular sequence, if and only if it is a \(H_1\)-regular sequence, if and only if it is a quasi-regular sequence.
Proof
The implication (1) \(\Rightarrow\) (2) is Lemma 062F. The implication (2) \(\Rightarrow\) (3) is Lemma 0CEM. The implication (3) \(\Rightarrow\) (4) is Lemma 062I. The implication (4) \(\Rightarrow\) (1) is Algebra, Lemma 061S.
Lemma
Let \(A\) be a ring. Let \(I \subset A\) be an ideal. Let \(g_1, \ldots, g_m\) be a sequence in \(A\) whose image in \(A/I\) is \(H_1\)-regular. Then \(I \cap (g_1, \ldots, g_m) = I(g_1, \ldots, g_m)\).
Proof
Consider the exact sequence of complexes \[0 \to I \otimes_A K_\bullet(A, g_1, \ldots, g_m) \to K_\bullet(A, g_1, \ldots, g_m) \to K_\bullet(A/I, g_1, \ldots, g_m) \to 0\] Since the complex on the right has \(H_1 = 0\) by assumption we see that \[\Coker(I^{\oplus m} \to I) \longrightarrow \Coker(A^{\oplus m} \to A)\] is injective. This is equivalent to the assertion of the lemma.
Lemma
Let \(A\) be a ring. Let \(I \subset J \subset A\) be ideals. Assume that \(J/I \subset A/I\) is generated by an \(H_1\)-regular sequence. Then \(I \cap J^2 = IJ\).
Proof
To prove this choose \(g_1, \ldots, g_m \in J\) whose images in \(A/I\) form a \(H_1\)-regular sequence which generates \(J/I\). In particular \(J = I + (g_1, \ldots, g_m)\). Suppose that \(x \in I \cap J^2\). Because \(x \in J^2\) can write \[x = \sum a_{ij} g_ig_j + \sum a_j g_j + a\] with \(a_{ij} \in A\), \(a_j \in I\) and \(a \in I^2\). Then \(\sum a_{ij}g_ig_j \in I \cap (g_1, \ldots, g_m)\) hence by Lemma 0665 we see that \(\sum a_{ij}g_ig_j \in I(g_1, \ldots, g_m)\). Thus \(x \in IJ\) as desired.
Lemma
Let \(A\) be a ring. Let \(I\) be an ideal generated by a quasi-regular sequence \(f_1, \ldots, f_n\) in \(A\). Let \(g_1, \ldots, g_m \in A\) be elements whose images \(\overline{g}_1, \ldots, \overline{g}_m\) form an \(H_1\)-regular sequence in \(A/I\). Then \(f_1, \ldots, f_n, g_1, \ldots, g_m\) is a quasi-regular sequence in \(A\).
Proof
We claim that \(g_1, \ldots, g_m\) forms an \(H_1\)-regular sequence in \(A/I^d\) for every \(d\). By induction assume that this holds in \(A/I^{d - 1}\). We have a short exact sequence of complexes \[0 \to K_\bullet(A, g_\bullet) \otimes_A I^{d - 1}/I^d \to K_\bullet(A/I^d, g_\bullet) \to K_\bullet(A/I^{d - 1}, g_\bullet) \to 0\] Since \(f_1, \ldots, f_n\) is quasi-regular we see that the first complex is a direct sum of copies of \(K_\bullet(A/I, g_1, \ldots, g_m)\) hence acyclic in degree \(1\). By induction hypothesis the last complex is acyclic in degree \(1\). Hence also the middle complex is. In particular, the sequence \(g_1, \ldots, g_m\) forms a quasi-regular sequence in \(A/I^d\) for every \(d \geq 1\), see Lemma 062I. Now we are ready to prove that \(f_1, \ldots, f_n, g_1, \ldots, g_m\) is a quasi-regular sequence in \(A\). Namely, set \(J = (f_1, \ldots, f_n, g_1, \ldots, g_m)\) and suppose that (with multinomial notation) \[\sum\nolimits_{|N| + |M| = d} a_{N, M} f^N g^M \in J^{d + 1}\] for some \(a_{N, M} \in A\). We have to show that \(a_{N, M} \in J\) for all \(N, M\). Let \(e \in \{0, 1, \ldots, d\}\). Then \[\sum\nolimits_{|N| = d - e, \ |M| = e} a_{N, M} f^N g^M \in (g_1, \ldots, g_m)^{e + 1} + I^{d - e + 1}\] Because \(g_1, \ldots, g_m\) is a quasi-regular sequence in \(A/I^{d - e + 1}\) we deduce \[\sum\nolimits_{|N| = d - e} a_{N, M} f^N \in (g_1, \ldots, g_m) + I^{d - e + 1}\] for each \(M\) with \(|M| = e\). By Lemma 0665 applied to \(I^{d - e}/I^{d - e + 1}\) in the ring \(A/I^{d - e + 1}\) this implies \(\sum_{|N| = d - e} a_{N, M} f^N \in I^{d - e}(g_1, \ldots, g_m)\). Since \(f_1, \ldots, f_n\) is quasi-regular in \(A\) this implies that \(a_{N, M} \in J\) for each \(N, M\) with \(|N| = d - e\) and \(|M| = e\). This proves the lemma.
Lemma
Let \(A\) be a ring. Let \(I\) be an ideal generated by an \(H_1\)-regular sequence \(f_1, \ldots, f_n\) in \(A\). Let \(g_1, \ldots, g_m \in A\) be elements whose images \(\overline{g}_1, \ldots, \overline{g}_m\) form an \(H_1\)-regular sequence in \(A/I\). Then \(f_1, \ldots, f_n, g_1, \ldots, g_m\) is an \(H_1\)-regular sequence in \(A\).
Proof
We have to show that \(H_1(A, f_1, \ldots, f_n, g_1, \ldots, g_m) = 0\). To do this consider the commutative diagram \[\xymatrix{ \wedge^2(A^{\oplus n + m}) \ar[r] \ar[d] & A^{\oplus n + m} \ar[r] \ar[d] & A \ar[r] \ar[d] & 0 \\ \wedge^2(A/I^{\oplus m}) \ar[r] & A/I^{\oplus m} \ar[r] & A/I \ar[r] & 0 }\] Consider an element \((a_1, \ldots, a_{n + m}) \in A^{\oplus n + m}\) which maps to zero in \(A\). Because \(\overline{g}_1, \ldots, \overline{g}_m\) form an \(H_1\)-regular sequence in \(A/I\) we see that \((\overline{a}_{n + 1}, \ldots, \overline{a}_{n + m})\) is the image of some element \(\overline{\alpha}\) of \(\wedge^2(A/I^{\oplus m})\). We can lift \(\overline{\alpha}\) to an element \(\alpha \in \wedge^2(A^{\oplus n + m})\) and subtract the image of it in \(A^{\oplus n + m}\) from our element \((a_1, \ldots, a_{n + m})\). Thus we may assume that \(a_{n + 1}, \ldots, a_{n + m} \in I\). Since \(I = (f_1, \ldots, f_n)\) we can modify our element \((a_1, \ldots, a_{n + m})\) by linear combinations of the elements \[(0, \ldots, g_j, 0, \ldots, 0, f_i, 0, \ldots, 0)\] in the image of the top left horizontal arrow to reduce to the case that \(a_{n + 1}, \ldots, a_{n + m}\) are zero. In this case \((a_1, \ldots, a_n, 0, \ldots, 0)\) defines an element of \(H_1(A, f_1, \ldots, f_n)\) which we assumed to be zero.
Lemma
Let \(A\) be a ring. Let \(f_1, \ldots, f_n, g_1, \ldots, g_m \in A\) be an \(H_1\)-regular sequence. Then the images \(\overline{g}_1, \ldots, \overline{g}_m\) in \(A/(f_1, \ldots, f_n)\) form an \(H_1\)-regular sequence.
Proof
Set \(I = (f_1, \ldots, f_n)\). We have to show that any relation \(\sum_{j = 1, \ldots, m} \overline{a}_j \overline{g}_j\) in \(A/I\) is a linear combination of trivial relations. Because \(I = (f_1, \ldots, f_n)\) we can lift this relation to a relation \[\sum\nolimits_{j = 1, \ldots, m} a_j g_j + \sum\nolimits_{i = 1, \ldots, n} b_if_i = 0\] in \(A\). By assumption this relation in \(A\) is a linear combination of trivial relations. Taking the image in \(A/I\) we obtain what we want.
Lemma
Let \(A\) be a ring. Let \(I\) be an ideal generated by a Koszul-regular sequence \(f_1, \ldots, f_n\) in \(A\). Let \(g_1, \ldots, g_m \in A\) be elements whose images \(\overline{g}_1, \ldots, \overline{g}_m\) form a Koszul-regular sequence in \(A/I\). Then \(f_1, \ldots, f_n, g_1, \ldots, g_m\) is a Koszul-regular sequence in \(A\).
Proof
Our assumptions say that \(K_\bullet(A, f_1, \ldots, f_n)\) is a finite free resolution of \(A/I\) and \(K_\bullet(A/I, \overline{g}_1, \ldots, \overline{g}_m)\) is a finite free resolution of \(A/(f_i, g_j)\) over \(A/I\). Then \[\begin{align*} K_\bullet(A, f_1, \ldots, f_n, g_1, \ldots, g_m) & = \text{Tot}(K_\bullet(A, f_1, \ldots, f_n) \otimes_A K_\bullet(A, g_1, \ldots, g_m)) \\ & \cong A/I \otimes_A K_\bullet(A, g_1, \ldots, g_m) \\ & = K_\bullet(A/I, \overline{g}_1, \ldots, \overline{g}_m) \\ & \cong A/(f_i, g_j) \end{align*}\] The first equality by Lemma 0664. The first quasi-isomorphism \(\cong\) by (the dual of) Homology, Lemma 0133 as the \(q\)th row of the double complex \(K_\bullet(A, f_1, \ldots, f_n) \otimes_A K_\bullet(A, g_1, \ldots, g_m)\) is a resolution of \(A/I \otimes_A K_q(A, g_1, \ldots, g_m)\). The second equality is clear. The last quasi-isomorphism by assumption. Hence we win.
To conclude in the following lemma it is necessary to assume that both \(f_1, \ldots, f_n\) and \(f_1, \ldots, f_n, g_1, \ldots, g_m\) are Koszul-regular. A counter example to dropping the assumption that \(f_1, \ldots, f_n\) is Koszul-regular is Examples, Lemma 0640.
Lemma
Let \(A\) be a ring. Let \(f_1, \ldots, f_n, g_1, \ldots, g_m \in A\). If both \(f_1, \ldots, f_n\) and \(f_1, \ldots, f_n, g_1, \ldots, g_m\) are Koszul-regular sequences in \(A\), then \(\overline{g}_1, \ldots, \overline{g}_m\) in \(A/(f_1, \ldots, f_n)\) form a Koszul-regular sequence.
Proof
Set \(I = (f_1, \ldots, f_n)\). Our assumptions say that \(K_\bullet(A, f_1, \ldots, f_n)\) is a finite free resolution of \(A/I\) and \(K_\bullet(A, f_1, \ldots, f_n, g_1, \ldots, g_m)\) is a finite free resolution of \(A/(f_i, g_j)\) over \(A\). Then \[\begin{align*} A/(f_i, g_j) & \cong K_\bullet(A, f_1, \ldots, f_n, g_1, \ldots, g_m) \\ & = \text{Tot}(K_\bullet(A, f_1, \ldots, f_n) \otimes_A K_\bullet(A, g_1, \ldots, g_m)) \\ & \cong A/I \otimes_A K_\bullet(A, g_1, \ldots, g_m) \\ & = K_\bullet(A/I, \overline{g}_1, \ldots, \overline{g}_m) \end{align*}\] The first quasi-isomorphism \(\cong\) by assumption. The first equality by Lemma 0664. The second quasi-isomorphism by (the dual of) Homology, Lemma 0133 as the \(q\)th row of the double complex \(K_\bullet(A, f_1, \ldots, f_n) \otimes_A K_\bullet(A, g_1, \ldots, g_m)\) is a resolution of \(A/I \otimes_A K_q(A, g_1, \ldots, g_m)\). The second equality is clear. Hence we win.
Lemma
Let \(R\) be a ring. Let \(I\) be an ideal generated by \(f_1, \ldots, f_r \in R\).
If \(I\) can be generated by a quasi-regular sequence of length \(r\), then \(f_1, \ldots, f_r\) is a quasi-regular sequence.
If \(I\) can be generated by an \(H_1\)-regular sequence of length \(r\), then \(f_1, \ldots, f_r\) is an \(H_1\)-regular sequence.
If \(I\) can be generated by a Koszul-regular sequence of length \(r\), then \(f_1, \ldots, f_r\) is a Koszul-regular sequence.
Proof
If \(I\) can be generated by a quasi-regular sequence of length \(r\), then \(I/I^2\) is free of rank \(r\) over \(R/I\). Since \(f_1, \ldots, f_r\) generate by assumption we see that the images \(\overline{f}_i\) form a basis of \(I/I^2\) over \(R/I\). It follows that \(f_1, \ldots, f_r\) is a quasi-regular sequence as all this means, besides the freeness of \(I/I^2\), is that the maps \(\text{Sym}^n_{R/I}(I/I^2) \to I^n/I^{n + 1}\) are isomorphisms.
We continue to assume that \(I\) can be generated by a quasi-regular sequence, say \(g_1, \ldots, g_r\). Write \(g_j = \sum a_{ij}f_i\). As \(f_1, \ldots, f_r\) is quasi-regular according to the previous paragraph, we see that \(\det(a_{ij})\) is invertible mod \(I\). The matrix \(a_{ij}\) gives a map \(R^{\oplus r} \to R^{\oplus r}\) which induces a map of Koszul complexes \(\alpha : K_\bullet(R, f_1, \ldots, f_r) \to K_\bullet(R, g_1, \ldots, g_r)\), see Lemma 0624. This map becomes an isomorphism on inverting \(\det(a_{ij})\). Since the cohomology modules of both \(K_\bullet(R, f_1, \ldots, f_r)\) and \(K_\bullet(R, g_1, \ldots, g_r)\) are annihilated by \(I\), see Lemma 0663, we see that \(\alpha\) is a quasi-isomorphism.
Now assume that \(g_1, \ldots, g_r\) is a \(H_1\)-regular sequence generating \(I\). Then \(g_1, \ldots, g_r\) is a quasi-regular sequence by Lemma 062I. By the previous paragraph we conclude that \(f_1, \ldots, f_r\) is a \(H_1\)-regular sequence. Similarly for Koszul-regular sequences.
Lemma
Let \(R\) be a ring. Let \(a_1, \ldots, a_n \in R\) be elements such that \(R \to R^{\oplus n}\), \(x \mapsto (xa_1, \ldots, xa_n)\) is injective. Then the element \(\sum a_i t_i\) of the polynomial ring \(R[t_1, \ldots, t_n]\) is a nonzerodivisor.
Proof
If one of the \(a_i\) is a unit this is just the statement that any element of the form \(t_1 + a_2 t_2 + \ldots + a_n t_n\) is a nonzerodivisor in the polynomial ring over \(R\).
Case I: \(R\) is Noetherian. Let \(\mathfrak q_j\), \(j = 1, \ldots, m\) be the associated primes of \(R\). We have to show that each of the maps \[\sum a_i t_i : \text{Sym}^d(R^{\oplus n}) \longrightarrow \text{Sym}^{d + 1}(R^{\oplus n})\] is injective. As \(\text{Sym}^d(R^{\oplus n})\) is a free \(R\)-module its associated primes are \(\mathfrak q_j\), \(j = 1, \ldots, m\). For each \(j\) there exists an \(i = i(j)\) such that \(a_i \not \in \mathfrak q_j\) because there exists an \(x \in R\) with \(\mathfrak q_jx = 0\) but \(a_i x \not = 0\) for some \(i\) by assumption. Hence \(a_i\) is a unit in \(R_{\mathfrak q_j}\) and the map is injective after localizing at \(\mathfrak q_j\). Thus the map is injective, see Algebra, Lemma 0311.
Case II: \(R\) general. We can write \(R\) as the union of Noetherian rings \(R_\lambda\) with \(a_1, \ldots, a_n \in R_\lambda\). For each \(R_\lambda\) the result holds, hence the result holds for \(R\).
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_n\) be a Koszul-regular sequence in \(R\) such that \((f_1, \ldots, f_n) \not = R\). Consider the faithfully flat, smooth ring map \[R \longrightarrow S = R[\{t_{ij}\}_{i \leq j}, t_{11}^{-1}, t_{22}^{-1}, \ldots, t_{nn}^{-1}]\] For \(1 \leq i \leq n\) set \[g_i = \sum\nolimits_{i \leq j} t_{ij} f_j \in S.\] Then \(g_1, \ldots, g_n\) is a regular sequence in \(S\) and \((f_1, \ldots, f_n)S = (g_1, \ldots, g_n)\).
Proof
The equality of ideals is obvious as the matrix \[\left( \begin{matrix} t_{11} & t_{12} & t_{13} & \ldots \\ 0 & t_{22} & t_{23} & \ldots \\ 0 & 0 & t_{33} & \ldots \\ \ldots & \ldots & \ldots & \ldots \end{matrix} \right)\] is invertible in \(S\). Because \(f_1, \ldots, f_n\) is a Koszul-regular sequence we see that the kernel of \(R \to R^{\oplus n}\), \(x \mapsto (xf_1, \ldots, xf_n)\) is zero (as it computes the \(n\)the Koszul homology of \(R\) w.r.t. \(f_1, \ldots, f_n\)). Hence by Lemma 068P we see that \(g_1 = f_1 t_{11} + \ldots + f_n t_{1n}\) is a nonzerodivisor in \(S' = R[t_{11}, t_{12}, \ldots, t_{1n}, t_{11}^{-1}]\). We see that \(g_1, f_2, \ldots, f_n\) is a Koszul-sequence in \(S'\) by Lemma 062H and 066A. We conclude that \(\overline{f}_2, \ldots, \overline{f}_n\) is a Koszul-regular sequence in \(S'/(g_1)\) by Lemma 068M. Hence by induction on \(n\) we see that the images \(\overline{g}_2, \ldots, \overline{g}_n\) of \(g_2, \ldots, g_n\) in \(S'/(g_1)[\{t_{ij}\}_{2 \leq i \leq j}, t_{22}^{-1}, \ldots, t_{nn}^{-1}]\) form a regular sequence. This in turn means that \(g_1, \ldots, g_n\) forms a regular sequence in \(S\).
More on Koszul regular sequences
We continue the discussion from Section 062D.
Lemma
Let \(R\) be a ring and let \(M\) be an \(R\)-module. Let \(f_1, \ldots, f_r \in R\) be an \(M\)-Koszul-regular sequence. Then the extended alternating Čech complex \[M \to \bigoplus\nolimits_{i_0} M_{f_{i_0}} \to \bigoplus\nolimits_{i_0 < i_1} M_{f_{i_0}f_{i_1}} \to \ldots \to M_{f_1\ldots f_r}\] from Section 0G6F only has cohomology in degree \(r\).
Proof
By Lemma 062G and induction the sequence \(f_1, \ldots, f_{r - 1}, f_r^n\) is \(M\)-Koszul-regular for all \(n \geq 1\). By Lemma 0625 and tensoring with \(M\), any permutation of an \(M\)-Koszul-regular sequence is \(M\)-Koszul-regular. Hence we see that we may replace any (or all) \(f_i\) by its \(n\)th power and still have an \(M\)-Koszul-regular sequence. Thus \(K_\bullet(R, f_1^n, \ldots, f_r^n) \otimes_R M\) has nonzero cohomology only in homological degree \(0\). This implies what we want by Lemma 0913.
Lemma
Let \(a, a_2, \ldots, a_r\) be an \(H_1\)-regular sequence in a ring \(R\) (for example a Koszul regular sequence or a regular sequence, see Lemmas 062F and 0CEM). With \(I = (a, a_2, \ldots, a_r)\) the blowup algebra \(R' = R[\frac{I}{a}]\) is isomorphic to \(R'' = R[y_2, \ldots, y_r]/(a y_i - a_i)\).
Proof
By Algebra, Lemma 0G8S it suffices to show that \(R''\) is \(a\)-torsion free.
We claim \(a, ay_2 - a_2, \ldots, ay_n - a_r\) is a \(H_1\)-regular sequence in \(R[y_2, \ldots, y_r]\). Namely, the map \[(a, ay_2 - a_2, \ldots, ay_n - a_r) : R[y_2, \ldots, y_r]^{\oplus r} \longrightarrow R[y_2, \ldots, y_r]\] used to define the Koszul complex on \(a, ay_2 - a_2, \ldots, ay_n - a_r\) is isomorphic to the map \[(a, a_2, \ldots, a_r) : R[y_2, \ldots, y_r]^{\oplus r} \longrightarrow R[y_2, \ldots, y_r]\] used to the define the Koszul complex on \(a, a_2, \ldots, a_r\) via the isomorphism \[R[y_2, \ldots, y_r]^{\oplus r} \longrightarrow R[y_2, \ldots, y_r]^{\oplus r}\] sending \((b_1, \ldots, b_r)\) to \((b_1 - b_2y_2 \ldots - b_ry_r, -b_2, \ldots, - b_r)\). By Lemma 0624 these Koszul complexes are isomorphic. By Lemma 062H applied to the flat ring map \(R \to R[y_2, \ldots, y_r]\) we conclude our claim is true. By Lemma 0629 we see that the Koszul complex \(K\) on \(a, ay_2 - a_2, \ldots, ay_n - a_r\) is the cone on \(a : L \to L\) where \(L\) is the Koszul complex on \(ay_2 - a_2, \ldots, ay_n - a_r\). Since \(H_1(K) = 0\) by the claim, we conclude that \(a : H_0(L) \to H_0(L)\) is injective, in other words that \(R'' = R[y_2, \ldots, y_r]/(a y_i - a_i)\) has no nonzero \(a\)-torsion elements as desired.
Lemma
Let \(A \to B\) be a ring map. Let \(f_1, \ldots, f_r\) be a sequence in \(B\) such that \(B/(f_1, \ldots, f_r)\) is \(A\)-flat. Let \(A \to A'\) be a ring map. Then the canonical map \[H_1(K_\bullet(B, f_1, \ldots, f_r)) \otimes_A A' \longrightarrow H_1(K_\bullet(B', f'_1, \ldots, f'_r))\] is surjective. Here \(B' = B \otimes_A A'\) and \(f_i' \in B'\) is the image of \(f_i\).
Proof
The sequence \[\wedge^2(B^{\oplus r}) \to B^{\oplus r} \to B \to B/J \to 0\] is a complex of \(A\)-modules with \(B/J\) flat over \(A\) and cohomology group \(H_1 = H_1(K_\bullet(B, f_1, \ldots, f_r))\) in the spot \(B^{\oplus r}\). If we tensor this with \(A'\) we obtain a complex \[\wedge^2((B')^{\oplus r}) \to (B')^{\oplus r} \to B' \to B'/J' \to 0\] which is exact at \(B'\) and \(B'/J'\). In order to compute its cohomology group \(H'_1 = H_1(K_\bullet(B', f'_1, \ldots, f'_r))\) at \((B')^{\oplus r}\) we split the first sequence above into the exact sequences \(0 \to J \to B \to B/J \to 0\), \(0 \to K \to B^{\oplus r} \to J \to 0\), and \(\wedge^2(B^{\oplus r}) \to K \to H_1 \to 0\). Tensoring over \(A\) with \(A'\) we obtain the exact sequences \[\begin{matrix} 0 \to J \otimes_A A' \to B \otimes_A A' \to (B/J) \otimes_A A' \to 0 \\ K \otimes_A A' \to B^{\oplus r} \otimes_A A' \to J \otimes_A A' \to 0 \\ \wedge^2(B^{\oplus r}) \otimes_A A' \to K \otimes_A A' \to H_1 \otimes_A A' \to 0 \end{matrix}\] where the first one is exact as \(B/J\) is flat over \(A\), see Algebra, Lemma 00HL. We conclude that \(J' = J \otimes_A A' \subset B'\) and that \(K \otimes_A A' \to \Ker((B')^{\oplus r} \to B')\) is surjective. Thus \[\begin{align*} H_1 \otimes_A A' & = \Coker\left(\wedge^2(B^{\oplus r}) \otimes_A A' \to K \otimes_A A'\right) \\ & \to \Coker\left( \wedge^2((B')^{\oplus r}) \to \Ker((B')^{\oplus r} \to B') \right) = H'_1 \end{align*}\] is surjective too.
Lemma
Let \(A \to B\) and \(A \to A'\) be ring maps. Set \(B' = B \otimes_A A'\). Let \(f_1, \ldots, f_r \in B\). Assume \(B/(f_1, \ldots, f_r)B\) is flat over \(A\)
If \(f_1, \ldots, f_r\) is a quasi-regular sequence, then the image in \(B'\) is a quasi-regular sequence.
If \(f_1, \ldots, f_r\) is a \(H_1\)-regular sequence, then the image in \(B'\) is a \(H_1\)-regular sequence.
Proof
Assume \(f_1, \ldots, f_r\) is quasi-regular. Set \(J = (f_1, \ldots, f_r)\). By assumption \(J^n/J^{n + 1}\) is isomorphic to a direct sum of copies of \(B/J\) hence flat over \(A\). By induction and Algebra, Lemma 00HM we conclude that \(B/J^n\) is flat over \(A\). The ideal \((J')^n\) is equal to \(J^n \otimes_A A'\), see Algebra, Lemma 00HL. Hence \((J')^n/(J')^{n + 1} = J^n/J^{n + 1} \otimes_A A'\) which clearly implies that \(f_1, \ldots, f_r\) is a quasi-regular sequence in \(B'\).
Assume \(f_1, \ldots, f_r\) is \(H_1\)-regular. By Lemma 063Q the vanishing of the Koszul homology group \(H_1(K_\bullet(B, f_1, \ldots, f_r))\) implies the vanishing of \(H_1(K_\bullet(B', f'_1, \ldots, f'_r))\) and we win.
Lemma
Let \(A' \to B'\) be a ring map. Let \(I \subset A'\) be an ideal. Set \(A = A'/I\) and \(B = B'/IB'\). Let \(f'_1, \ldots, f'_r \in B'\). Assume
\(A' \to B'\) is flat and of finite presentation,
\(I\) is locally nilpotent,
the images \(f_1, \ldots, f_r \in B\) form a quasi-regular sequence,
\(B/(f_1, \ldots, f_r)\) is flat over \(A\).
Then \(B'/(f'_1, \ldots, f'_r)\) is flat over \(A'\).
Proof
Set \(C' = B'/(f'_1, \ldots, f'_r)\). We have to show \(A' \to C'\) is flat. Let \(\mathfrak r' \subset C'\) be a prime ideal lying over \(\mathfrak p' \subset A'\). We let \(\mathfrak q' \subset B'\) be the inverse image of \(\mathfrak r'\). By Algebra, Lemma 00HT it suffices to show that \(A'_{\mathfrak p'} \to C'_{\mathfrak q'}\) is flat. Algebra, Lemma 0470 tells us it suffices to show that \(f'_1, \ldots, f'_r\) map to a regular sequence in \[B'_{\mathfrak q'}/\mathfrak p'B'_{\mathfrak q'} = B_\mathfrak q/\mathfrak p B_\mathfrak q = (B \otimes_A \kappa(\mathfrak p))_\mathfrak q\] with obvious notation. What we know is that \(f_1, \ldots, f_r\) is a quasi-regular sequence in \(B\) and that \(B/(f_1, \ldots, f_r)\) is flat over \(A\). By Lemma 0CEP the images \(\overline{f}_1, \ldots, \overline{f}_r\) of \(f'_1, \ldots, f'_r\) in \(B \otimes_A \kappa(\mathfrak p)\) form a quasi-regular sequence. Since \((B \otimes_A \kappa(\mathfrak p))_\mathfrak q\) is a Noetherian local ring, we conclude by Lemma 09CC.
Lemma
Let \(A' \to B'\) be a ring map. Let \(I \subset A'\) be an ideal. Set \(A = A'/I\) and \(B = B'/IB'\). Let \(f'_1, \ldots, f'_r \in B'\). Assume
\(A' \to B'\) is flat and of finite presentation (for example smooth),
\(I\) is locally nilpotent,
the images \(f_1, \ldots, f_r \in B\) form a quasi-regular sequence,
\(B/(f_1, \ldots, f_r)\) is smooth over \(A\).
Then \(B'/(f'_1, \ldots, f'_r)\) is smooth over \(A'\).
Proof
Set \(C' = B'/(f'_1, \ldots, f'_r)\) and \(C = B/(f_1, \ldots, f_r)\). Then \(A' \to C'\) is of finite presentation. By Lemma 0CEQ we see that \(A' \to C'\) is flat. The fibre rings of \(A' \to C'\) are equal to the fibre rings of \(A \to C\) and hence smooth by assumption (4). It follows that \(A' \to C'\) is smooth by Algebra, Lemma 00TF.
Regular ideals
We will discuss the notion of a regular ideal sheaf in great generality in Divisors, Section 067M. Here we define the corresponding notion in the affine case, i.e., in the case of an ideal in a ring.
Definition
Let \(R\) be a ring and let \(I \subset R\) be an ideal.
We say \(I\) is a regular ideal if for every \(\mathfrak p \in V(I)\) there exists a \(g \in R\), \(g \not \in \mathfrak p\) and a regular sequence \(f_1, \ldots, f_r \in R_g\) such that \(I_g\) is generated by \(f_1, \ldots, f_r\).
We say \(I\) is a Koszul-regular ideal if for every \(\mathfrak p \in V(I)\) there exists a \(g \in R\), \(g \not \in \mathfrak p\) and a Koszul-regular sequence \(f_1, \ldots, f_r \in R_g\) such that \(I_g\) is generated by \(f_1, \ldots, f_r\).
We say \(I\) is a \(H_1\)-regular ideal if for every \(\mathfrak p \in V(I)\) there exists a \(g \in R\), \(g \not \in \mathfrak p\) and an \(H_1\)-regular sequence \(f_1, \ldots, f_r \in R_g\) such that \(I_g\) is generated by \(f_1, \ldots, f_r\).
We say \(I\) is a quasi-regular ideal if for every \(\mathfrak p \in V(I)\) there exists a \(g \in R\), \(g \not \in \mathfrak p\) and a quasi-regular sequence \(f_1, \ldots, f_r \in R_g\) such that \(I_g\) is generated by \(f_1, \ldots, f_r\).
It is clear that given \(I \subset R\) we have the implications \[\begin{align*} I\text{ is a regular ideal} & \Rightarrow I\text{ is a Koszul-regular ideal} \\ & \Rightarrow I\text{ is a }H_1\text{-regular ideal} \\ & \Rightarrow I\text{ is a quasi-regular ideal} \end{align*}\] see Lemmas 062F, 0CEM, and 062I. Such an ideal is always finitely generated.
Lemma
A quasi-regular ideal is finitely generated.
Proof
Let \(I \subset R\) be a quasi-regular ideal. Since \(V(I)\) is quasi-compact, there exist \(g_1, \ldots, g_m \in R\) such that \(V(I) \subset D(g_1) \cup \ldots \cup D(g_m)\) and such that \(I_{g_j}\) is generated by a quasi-regular sequence \(g_{j1}, \ldots, g_{jr_j} \in R_{g_j}\). Write \(g_{ji} = g'_{ji}/g_j^{e_{ij}}\) for some \(g'_{ij} \in I\). Write \(1 + x = \sum g_j h_j\) for some \(x \in I\) which is possible as \(V(I) \subset D(g_1) \cup \ldots \cup D(g_m)\). Note that \(\Spec(R) = D(g_1) \cup \ldots \cup D(g_m) \bigcup D(x)\) Then \(I\) is generated by the elements \(g'_{ij}\) and \(x\) as these generate on each of the pieces of the cover, see Algebra, Lemma 00EO.
Lemma
Let \(I \subset R\) be a quasi-regular ideal of a ring. Then \(I/I^2\) is a finite projective \(R/I\)-module.
Proof
This follows from Algebra, Lemma 00NX and the definitions.
We prove flat descent for Koszul-regular, \(H_1\)-regular, quasi-regular ideals.
Lemma
Let \(A \to B\) be a faithfully flat ring map. Let \(I \subset A\) be an ideal. If \(IB\) is a Koszul-regular (resp. \(H_1\)-regular, resp. quasi-regular) ideal in \(B\), then \(I\) is a Koszul-regular (resp. \(H_1\)-regular, resp. quasi-regular) ideal in \(A\).
Proof
We fix the prime \(\mathfrak p \supset I\) throughout the proof. Assume \(IB\) is quasi-regular. By Lemma 07CW \(IB\) is a finite module, hence \(I\) is a finite \(A\)-module by Algebra, Lemma 03C4. As \(A \to B\) is flat we see that \[I/I^2 \otimes_{A/I} B/IB = I/I^2 \otimes_A B = IB/(IB)^2.\] As \(IB\) is quasi-regular, the \(B/IB\)-module \(IB/(IB)^2\) is finite locally free. Hence \(I/I^2\) is finite projective, see Algebra, Proposition 058S. In particular, after replacing \(A\) by \(A_f\) for some \(f \in A\), \(f \not \in \mathfrak p\) we may assume that \(I/I^2\) is free of rank \(r\). Pick \(f_1, \ldots, f_r \in I\) which give a basis of \(I/I^2\). By Nakayama’s lemma (see Algebra, Lemma 00DV) we see that, after another replacement \(A \leadsto A_f\) as above, \(I\) is generated by \(f_1, \ldots, f_r\).
Proof of the “quasi-regular” case. Above we have seen that \(I/I^2\) is free on the \(r\)-generators \(f_1, \ldots, f_r\). To finish the proof in this case we have to show that the maps \(\text{Sym}^d(I/I^2) \to I^d/I^{d + 1}\) are isomorphisms for each \(d \geq 2\). This is clear as the faithfully flat base changes \(\text{Sym}^d(IB/(IB)^2) \to (IB)^d/(IB)^{d + 1}\) are isomorphisms locally on \(B\) by assumption. Details omitted.
Proof of the “\(H_1\)-regular” and “Koszul-regular” case. Consider the sequence of elements \(f_1, \ldots, f_r\) generating \(I\) we constructed above. By Lemma 066A we see that \(f_1, \ldots, f_r\) map to a \(H_1\)-regular or Koszul-regular sequence in \(B_g\) for any \(g \in B\) such that \(IB\) is generated by an \(H_1\)-regular or Koszul-regular sequence. Hence \(K_\bullet(A, f_1, \ldots, f_r) \otimes_A B_g\) has vanishing \(H_1\) or \(H_i\), \(i > 0\). Since the homology of \(K_\bullet(B, f_1, \ldots, f_r) = K_\bullet(A, f_1, \ldots, f_r) \otimes_A B\) is annihilated by \(IB\) (see Lemma 0663) and since \(V(IB) \subset \bigcup_{g\text{ as above}} D(g)\) we conclude that \(K_\bullet(A, f_1, \ldots, f_r) \otimes_A B\) has vanishing homology in degree \(1\) or all positive degrees. Using that \(A \to B\) is faithfully flat we conclude that the same is true for \(K_\bullet(A, f_1, \ldots, f_r)\).
Lemma
Let \(A\) be a ring. Let \(I \subset J \subset A\) be ideals. Assume that \(J/I \subset A/I\) is a \(H_1\)-regular ideal. Then \(I \cap J^2 = IJ\).
Proof
Follows immediately from Lemma 0666 by localizing.
Local complete intersection maps
We can use the material above to define a local complete intersection map between rings using presentations by (finite) polynomial algebras.
Lemma
Let \(A \to B\) be a finite type ring map. If for some presentation \(\alpha : A[x_1, \ldots, x_n] \to B\) the kernel \(I\) is a Koszul-regular ideal then for any presentation \(\beta : A[y_1, \ldots, y_m] \to B\) the kernel \(J\) is a Koszul-regular ideal.
Proof
Choose \(f_j \in A[x_1, \ldots, x_n]\) with \(\alpha(f_j) = \beta(y_j)\) and \(g_i \in A[y_1, \ldots, y_m]\) with \(\beta(g_i) = \alpha(x_i)\). Then we get a commutative diagram \[\xymatrix{ A[x_1, \ldots, x_n, y_1, \ldots, y_m] \ar[d]^{x_i \mapsto g_i} \ar[rr]_-{y_j \mapsto f_j} & & A[x_1, \ldots, x_n] \ar[d] \\ A[y_1, \ldots, y_m] \ar[rr] & & B }\] Note that the kernel \(K\) of \(A[x_i, y_j] \to B\) is equal to \(K = (I, y_j - f_j) = (J, x_i - g_i)\). In particular, as \(I\) is finitely generated by Lemma 07CW we see that \(J = K/(x_i - g_i)\) is finitely generated too.
Pick a prime \(\mathfrak q \subset B\). Since \(I/I^2 \oplus B^{\oplus m} = J/J^2 \oplus B^{\oplus n}\) (Algebra, Lemma 00S5) we see that \[\dim J/J^2 \otimes_B \kappa(\mathfrak q) + n = \dim I/I^2 \otimes_B \kappa(\mathfrak q) + m.\] Pick \(p_1, \ldots, p_t \in I\) which map to a basis of \(I/I^2 \otimes \kappa(\mathfrak q) = I \otimes_{A[x_i]} \kappa(\mathfrak q)\). Pick \(q_1, \ldots, q_s \in J\) which map to a basis of \(J/J^2 \otimes \kappa(\mathfrak q) = J \otimes_{A[y_j]} \kappa(\mathfrak q)\). So \(s + n = t + m\). By Nakayama’s lemma there exist \(h \in A[x_i]\) and \(h' \in A[y_j]\) both mapping to a nonzero element of \(\kappa(\mathfrak q)\) such that \(I_h = (p_1, \ldots, p_t)\) in \(A[x_i, 1/h]\) and \(J_{h'} = (q_1, \ldots, q_s)\) in \(A[y_j, 1/h']\). As \(I\) is Koszul-regular we may also assume that \(I_h\) is generated by a Koszul regular sequence. This sequence must necessarily have length \(t = \dim I/I^2 \otimes_B \kappa(\mathfrak q)\), hence we see that \(p_1, \ldots, p_t\) is a Koszul-regular sequence by Lemma 066A. As also \(y_1 - f_1, \ldots, y_m - f_m\) is a regular sequence we conclude \[y_1 - f_1, \ldots, y_m - f_m, p_1, \ldots, p_t\] is a Koszul-regular sequence in \(A[x_i, y_j, 1/h]\) (see Lemma 0669). This sequence generates the ideal \(K_h\). Hence the ideal \(K_{hh'}\) is generated by a Koszul-regular sequence of length \(m + t = n + s\). But it is also generated by the sequence \[x_1 - g_1, \ldots, x_n - g_n, q_1, \ldots, q_s\] of the same length which is thus a Koszul-regular sequence by Lemma 066A. Finally, by Lemma 068M we conclude that the images of \(q_1, \ldots, q_s\) in \[A[x_i, y_j, 1/hh']/(x_1 - g_1, \ldots, x_n - g_n) \cong A[y_j, 1/h'']\] form a Koszul-regular sequence generating \(J_{h''}\). Since \(h''\) is the image of \(hh'\) it doesn’t map to zero in \(\kappa(\mathfrak q)\) and we win.
This lemma allows us to make the following definition.
Definition
A ring map \(A \to B\) is called a local complete intersection if it is of finite type and for some (equivalently any) presentation \(B = A[x_1, \ldots, x_n]/I\) the ideal \(I\) is Koszul-regular.
This notion is local.
Lemma
Let \(R \to S\) be a ring map. Let \(g_1, \ldots, g_m \in S\) generate the unit ideal. If each \(R \to S_{g_j}\) is a local complete intersection so is \(R \to S\).
Proof
Let \(S = R[x_1, \ldots, x_n]/I\) be a presentation. Pick \(h_j \in R[x_1, \ldots, x_n]\) mapping to \(g_j\) in \(S\). Then \(R[x_1, \ldots, x_n, x_{n + 1}]/(I, x_{n + 1}h_j - 1)\) is a presentation of \(S_{g_j}\). Hence \(I_j = (I, x_{n + 1}h_j - 1)\) is a Koszul-regular ideal in \(R[x_1, \ldots, x_n, x_{n + 1}]\). Pick a prime \(I \subset \mathfrak q \subset R[x_1, \ldots, x_n]\). Then \(h_j \not \in \mathfrak q\) for some \(j\) and \(\mathfrak q_j = (\mathfrak q, x_{n + 1}h_j - 1)\) is a prime ideal of \(V(I_j)\) lying over \(\mathfrak q\). Pick \(f_1, \ldots, f_r \in I\) which map to a basis of \(I/I^2 \otimes \kappa(\mathfrak q)\). Then \(x_{n + 1}h_j - 1, f_1, \ldots, f_r\) is a sequence of elements of \(I_j\) which map to a basis of \(I_j \otimes \kappa(\mathfrak q_j)\), see Algebra, Lemma 08JZ. By Nakayama’s lemma there exists an \(h \in R[x_1, \ldots, x_n, x_{n + 1}]\) such that \((I_j)_h\) is generated by \(x_{n + 1}h_j - 1, f_1, \ldots, f_r\). We may also assume that \((I_j)_h\) is generated by a Koszul regular sequence of some length \(e\). Looking at the dimension of \(I_j \otimes \kappa(\mathfrak q_j)\) we see that \(e = r + 1\). Hence by Lemma 066A we see that \(x_{n + 1}h_j - 1, f_1, \ldots, f_r\) is a Koszul-regular sequence generating \((I_j)_h\) for some \(h \in R[x_1, \ldots, x_n, x_{n + 1}]\), \(h \not \in \mathfrak q_j\). By Lemma 068M we see that \(I_{h'}\) is generated by a Koszul-regular sequence for some \(h' \in R[x_1, \ldots, x_n]\), \(h' \not \in \mathfrak q\) as desired.
Lemma
Let \(R\) be a ring. If \(R[x_1, \ldots, x_n]/(f_1, \ldots, f_c)\) is a relative global complete intersection, then \(f_1, \ldots, f_c\) is a Koszul regular sequence.
Proof
Recall that the homology groups \(H_i(K_\bullet(f_\bullet))\) are annihilated by the ideal \((f_1, \ldots, f_c)\). Hence it suffices to show that \(H_i(K_\bullet(f_\bullet))_\mathfrak q\) is zero for all primes \(\mathfrak q \subset R[x_1, \ldots, x_n]\) containing \((f_1, \ldots, f_c)\). This follows from Algebra, Lemma 00SV and the fact that a regular sequence is Koszul regular (Lemma 062F).
Lemma
Let \(R \to S\) be a ring map. The following are equivalent
\(R \to S\) is syntomic (Algebra, Definition 00SL), and
\(R \to S\) is flat and a local complete intersection.
Proof
Assume (1). Then \(R \to S\) is flat by definition. By Algebra, Lemma 00SY and Lemma 07D1 we see that it suffices to show a relative global complete intersection is a local complete intersection homomorphism which is Lemma 07D2.
Assume (2). A local complete intersection is of finite presentation because a Koszul-regular ideal is finitely generated. Let \(R \to k\) be a map to a field. It suffices to show that \(S' = S \otimes_R k\) is a local complete intersection over \(k\), see Algebra, Definition 00S9. Choose a prime \(\mathfrak q' \subset S'\). Write \(S = R[x_1, \ldots, x_n]/I\). Then \(S' = k[x_1, \ldots, x_n]/I'\) where \(I' \subset k[x_1, \ldots, x_n]\) is the image of \(I\). Let \(\mathfrak p' \subset k[x_1, \ldots, x_n]\), \(\mathfrak q \subset S\), and \(\mathfrak p \subset R[x_1, \ldots, x_n]\) be the corresponding primes. By Definition 07CV exists an \(g \in R[x_1, \ldots, x_n]\), \(g \not \in \mathfrak p\) and \(f_1, \ldots, f_r \in R[x_1, \ldots, x_n]_g\) which form a Koszul-regular sequence generating \(I_g\). Since \(S\) and hence \(S_g\) is flat over \(R\) we see that the images \(f'_1, \ldots, f'_r\) in \(k[x_1, \ldots, x_n]_g\) form a \(H_1\)-regular sequence generating \(I'_g\), see Lemma 0CEP. Thus \(f'_1, \ldots, f'_r\) map to a regular sequence in \(k[x_1, \ldots, x_n]_{\mathfrak p'}\) generating \(I'_{\mathfrak p'}\) by Lemma 09CC. Applying Algebra, Lemma 00SC we conclude \(S'_{gg'}\) for some \(g' \in S\), \(g' \not \in \mathfrak q'\) is a global complete intersection over \(k\) as desired.
For a local complete intersection \(R \to S\) we have \(H_n(L_{S/R}) = 0\) for \(n \geq 2\). Since we haven’t (yet) defined the full cotangent complex we can’t state and prove this, but we can deduce one of the consequences.
Lemma
Let \(A \to B \to C\) be ring maps. Assume \(B \to C\) is a local complete intersection homomorphism. Choose a presentation \(\alpha : A[x_s, s \in S] \to B\) with kernel \(I\). Choose a presentation \(\beta : B[y_1, \ldots, y_m] \to C\) with kernel \(J\). Let \(\gamma : A[x_s, y_t] \to C\) be the induced presentation of \(C\) with kernel \(K\). Then we get a canonical commutative diagram \[\xymatrix{ 0 \ar[r] & \Omega_{A[x_s]/A} \otimes C \ar[r] & \Omega_{A[x_s, y_t]/A} \otimes C \ar[r] & \Omega_{B[y_t]/B} \otimes C \ar[r] & 0 \\ 0 \ar[r] & I/I^2 \otimes C \ar[r] \ar[u] & K/K^2 \ar[r] \ar[u] & J/J^2 \ar[r] \ar[u] & 0 }\] with exact rows. In particular, the six term exact sequence of Algebra, Lemma 00S2 can be completed with a zero on the left, i.e., the sequence \[0 \to H_1(\NL_{B/A} \otimes_B C) \to H_1(L_{C/A}) \to H_1(L_{C/B}) \to \Omega_{B/A} \otimes_B C \to \Omega_{C/A} \to \Omega_{C/B} \to 0\] is exact.
Proof
The only thing to prove is the injectivity of the map \(I/I^2 \otimes C \to K/K^2\). By assumption the ideal \(J\) is Koszul-regular. Hence we have \(IA[x_s, y_j] \cap K^2 = IK\) by Lemma 07CX. This means that the kernel of \(K/K^2 \to J/J^2\) is isomorphic to \(IA[x_s, y_j]/IK\). Since \(I/I^2 \otimes_A C = IA[x_s, y_j]/IK\) by right exactness of tensor product, this provides us with the desired injectivity of \(I/I^2 \otimes_A C \to K/K^2\).
Lemma
Let \(A \to B \to C\) be ring maps. If \(B \to C\) is a filtered colimit of local complete intersection homomorphisms then the conclusion of Lemma 07D4 remains valid.
Proof
Lemma
Let \(A \to B\) be a local homomorphism of local rings. Let \(A^h \to B^h\), resp. \(A^{sh} \to B^{sh}\) be the induced map on henselizations, resp. strict henselizations (Algebra, Lemma 04GS, resp. Lemma 04GU). Then \(\NL_{B/A} \otimes_B B^h \to \NL_{B^h/A^h}\) and \(\NL_{B/A} \otimes_B B^{sh} \to \NL_{B^{sh}/A^{sh}}\) induce isomorphisms on cohomology groups.
Proof
Since \(A^h\) is a filtered colimit of étale algebras over \(A\) we see that \(\NL_{A^h/A}\) is an acyclic complex by Algebra, Lemma 07BQ and Algebra, Definition 00U1. The same is true for \(B^h/B\). Using the Jacobi-Zariski sequence (Algebra, Lemma 00S2) for \(A \to A^h \to B^h\) we find that \(\NL_{B^h/A} \to \NL_{B^h/A^h}\) induces isomorphisms on cohomology groups. Moreover, an étale ring map is a local complete intersection as it is even a global complete intersection, see Algebra, Lemma 00U9. By Lemma 07D5 we get a six term exact Jacobi-Zariski sequence associated to \(A \to B \to B^h\) which proves that \(\NL_{B/A} \otimes_B B^h \to \NL_{B^h/A}\) induces isomorphisms on cohomology groups. This finishes the proof in the case of the map on henselizations. The case of strict henselization is proved in exactly the same manner.
Cartier’s equality and geometric regularity
A reference for this section and the next is [MatCA, Section 39]. In order to comfortably read this section the reader should be familiar with the naive cotangent complex and its properties, see Algebra, Section 00S0.
Lemma
Let \(K/k\) be a finitely generated field extension. Then \(\Omega_{K/k}\) and \(H_1(L_{K/k})\) are finite dimensional and \(\text{trdeg}_k(K) = \dim_K \Omega_{K/k} - \dim_K H_1(L_{K/k})\).
Proof
We can find a global complete intersection \(A = k[x_1, \ldots, x_n]/(f_1, \ldots, f_c)\) over \(k\) such that \(K\) is isomorphic to the fraction field of \(A\), see Algebra, Lemma 07BV and its proof. In this case we see that \(\NL_{K/k}\) is homotopy equivalent to the complex \[\bigoplus\nolimits_{j = 1, \ldots, c} K \longrightarrow \bigoplus\nolimits_{i = 1, \ldots, n} K\text{d}x_i\] by Algebra, Lemmas 00S1 and 00S7. The transcendence degree of \(K\) over \(k\) is the dimension of \(A\) (by Algebra, Lemma 00P0) which is \(n - c\) and we win.
Lemma
Let \(M/L/K\) be field extensions. Then the Jacobi-Zariski sequence \[0 \to H_1(L_{L/K}) \otimes_L M \to H_1(L_{M/K}) \to H_1(L_{M/L}) \to \Omega_{L/K} \otimes_L M \to \Omega_{M/K} \to \Omega_{M/L} \to 0\] is exact.
Proof
Lemma
Given a commutative diagram of fields \[\xymatrix{ K \ar[r] & K' \\ k \ar[u] \ar[r] & k' \ar[u] }\] with \(k'/k\) and \(K'/K\) finitely generated field extensions the kernel and cokernel of the maps \[\alpha : \Omega_{K/k} \otimes_K K' \to \Omega_{K'/k'} \quad\text{and}\quad \beta : H_1(L_{K/k}) \otimes_K K' \to H_1(L_{K'/k'})\] are finite dimensional and \[\dim \Ker(\alpha) - \dim \Coker(\alpha) -\dim \Ker(\beta) + \dim \Coker(\beta) = \text{trdeg}_k(k') - \text{trdeg}_K(K')\]
Proof
The Jacobi-Zariski sequences for \(k \subset k' \subset K'\) and \(k \subset K \subset K'\) are \[0 \to H_1(L_{k'/k}) \otimes K' \to H_1(L_{K'/k}) \to H_1(L_{K'/k'}) \to \Omega_{k'/k} \otimes K' \to \Omega_{K'/k} \to \Omega_{K'/k'} \to 0\] and \[0 \to H_1(L_{K/k}) \otimes K' \to H_1(L_{K'/k}) \to H_1(L_{K'/K}) \to \Omega_{K/k} \otimes K' \to \Omega_{K'/k} \to \Omega_{K'/K} \to 0\] By Lemma 07E1 the vector spaces \(\Omega_{k'/k}\), \(\Omega_{K'/K}\), \(H_1(L_{K'/K})\), and \(H_1(L_{k'/k})\) are finite dimensional and the alternating sum of their dimensions is \(\text{trdeg}_k(k') - \text{trdeg}_K(K')\). The lemma follows.
Geometric regularity
Let \(k\) be a field. Let \((A, \mathfrak m, K)\) be a Noetherian local \(k\)-algebra. The Jacobi-Zariski sequence (Algebra, Lemma 00S2) is a canonical exact sequence \[H_1(L_{K/k}) \to \mathfrak m/\mathfrak m^2 \to \Omega_{A/k} \otimes_A K \to \Omega_{K/k} \to 0\] because \(H_1(L_{K/A}) = \mathfrak m/\mathfrak m^2\) by Algebra, Lemma 07BP. We will show that exactness on the left of this sequence characterizes whether or not a regular local ring \(A\) is geometrically regular over \(k\). We will link this to the notion of formal smoothness in Section 07EK.
Proposition
Let \(k\) be a field of characteristic \(p > 0\). Let \((A, \mathfrak m, K)\) be a Noetherian local \(k\)-algebra. The following are equivalent
\(A\) is geometrically regular over \(k\),
for all \(k \subset k' \subset k^{1/p}\) finite over \(k\) the ring \(A \otimes_k k'\) is regular,
\(A\) is regular and the canonical map \(H_1(L_{K/k}) \to \mathfrak m/\mathfrak m^2\) is injective, and
\(A\) is regular and the map \(\Omega_{k/\mathbf{F}_p} \otimes_k K \to \Omega_{A/\mathbf{F}_p} \otimes_A K\) is injective.
Proof
Proof of (3) \(\Rightarrow\) (1). Assume (3). Let \(k'/k\) be a finite purely inseparable extension. Set \(A' = A \otimes_k k'\). This is a local ring with maximal ideal \(\mathfrak m'\). Set \(K' = A'/\mathfrak m'\). We get a commutative diagram \[\xymatrix{ 0 \ar[r] & H_1(L_{K/k}) \otimes K' \ar[r] \ar[d]_\beta & \mathfrak m/\mathfrak m^2 \otimes K' \ar[r] \ar[d] & \Omega_{A/k} \otimes_A K' \ar[r] \ar[d]_{\cong} & \Omega_{K/k} \otimes K' \ar[r] \ar[d]_\alpha & 0 \\ & H_1(L_{K'/k'}) \ar[r] & \mathfrak m'/(\mathfrak m')^2 \ar[r] & \Omega_{A'/k'} \otimes_{A'} K' \ar[r] & \Omega_{K'/k'} \ar[r] & 0 }\] with exact rows. The third vertical arrow is an isomorphism by base change for modules of differentials (Algebra, Lemma 00RV). Thus \(\alpha\) is surjective. By Lemma 07E3 we have \[\dim \Ker(\alpha) - \dim \Ker(\beta) + \dim \Coker(\beta) = 0\] (and these dimensions are all finite). A diagram chase shows that \(\dim \mathfrak m'/(\mathfrak m')^2 \leq \dim \mathfrak m/\mathfrak m^2\). However, since \(A \to A'\) is finite flat we see that \(\dim(A) = \dim(A')\), see Algebra, Lemma 00OM. Hence \(A'\) is regular by definition.
Equivalence of (3) and (4). Consider the Jacobi-Zariski sequences for rows of the commutative diagram \[\xymatrix{ \mathbf{F}_p \ar[r] & A \ar[r] & K \\ \mathbf{F}_p \ar[r] \ar[u] & k \ar[r] \ar[u] & K \ar[u] }\] to get a commutative diagram \[\xymatrix{ 0 \ar[r] & \mathfrak m/\mathfrak m^2 \ar[r] & \Omega_{A/\mathbf{F}_p} \otimes_A K \ar[r] & \Omega_{K/\mathbf{F}_p} \ar[r] & 0 & \\ 0 \ar[r] & H_1(L_{K/k}) \ar[r] \ar[u] & \Omega_{k/\mathbf{F}_p} \otimes_k K \ar[r] \ar[u] & \Omega_{K/\mathbf{F}_p} \ar[r] \ar[u] & \Omega_{K/k} \ar[r] \ar[u] & 0 }\] with exact rows. We have used that \(H_1(L_{K/A}) = \mathfrak m/\mathfrak m^2\) and that \(H_1(L_{K/\mathbf{F}_p}) = 0\) as \(K/\mathbf{F}_p\) is separable, see Algebra, Proposition 0322. Thus it is clear that the kernels of \(H_1(L_{K/k}) \to \mathfrak m/\mathfrak m^2\) and \(\Omega_{k/\mathbf{F}_p} \otimes_k K \to \Omega_{A/\mathbf{F}_p} \otimes_A K\) have the same dimension.
Proof of (2) \(\Rightarrow\) (4) following Faltings, see [Faltings-einfacher]. Let \(a_1, \ldots, a_n \in k\) be elements such that \(\text{d}a_1, \ldots, \text{d}a_n\) are linearly independent in \(\Omega_{k/\mathbf{F}_p}\). Consider the field extension \(k' = k(a_1^{1/p}, \ldots, a_n^{1/p})\). By Algebra, Lemma 07DZ we see that \(k' = k[x_1, \ldots, x_n]/(x_1^p - a_1, \ldots, x_n^p - a_n)\). In particular we see that the naive cotangent complex of \(k'/k\) is homotopic to the complex \(\bigoplus_{j = 1, \ldots, n} k' \rightarrow \bigoplus_{i = 1, \ldots, n} k'\) with the zero differential as \(\text{d}(x_j^p - a_j) = 0\) in \(\Omega_{k[x_1, \ldots, x_n]/k}\). Set \(A' = A \otimes_k k'\) and \(K' = A'/\mathfrak m'\) as above. By Algebra, Lemma 00S4 we see that \(\NL_{A'/A}\) is homotopy equivalent to the complex \(\bigoplus_{j = 1, \ldots, n} A' \rightarrow \bigoplus_{i = 1, \ldots, n} A'\) with the zero differential, i.e., \(H_1(L_{A'/A})\) and \(\Omega_{A'/A}\) are free of rank \(n\). The Jacobi-Zariski sequence for \(\mathbf{F}_p \to A \to A'\) is \[H_1(L_{A'/A}) \to \Omega_{A/\mathbf{F}_p} \otimes_A A' \to \Omega_{A'/\mathbf{F}_p} \to \Omega_{A'/A} \to 0\] Using the presentation \(A[x_1, \ldots, x_n] \to A'\) with kernel \((x_j^p - a_j)\) we see, unwinding the maps in Algebra, Lemma 00S2, that the \(j\)th basis vector of \(H_1(L_{A'/A})\) maps to \(\text{d}a_j \otimes 1\) in \(\Omega_{A/\mathbf{F}_p} \otimes A'\). As \(\Omega_{A'/A}\) is free (hence flat) we get on tensoring with \(K'\) an exact sequence \[K'^{\oplus n} \to \Omega_{A/\mathbf{F}_p} \otimes_A K' \xrightarrow{\beta} \Omega_{A'/\mathbf{F}_p} \otimes_{A'} K' \to K'^{\oplus n} \to 0\] We conclude that the elements \(\text{d}a_j \otimes 1\) generate \(\Ker(\beta)\) and we have to show that are linearly independent, i.e., we have to show \(\dim(\Ker(\beta)) = n\). Consider the following big diagram \[\xymatrix{ 0 \ar[r] & \mathfrak m'/(\mathfrak m')^2 \ar[r] & \Omega_{A'/\mathbf{F}_p} \otimes K' \ar[r] & \Omega_{K'/\mathbf{F}_p} \ar[r] & 0 \\ 0 \ar[r] & \mathfrak m/\mathfrak m^2 \otimes K' \ar[r] \ar[u]^\alpha & \Omega_{A/\mathbf{F}_p} \otimes K' \ar[r] \ar[u]^\beta & \Omega_{K/\mathbf{F}_p} \otimes K' \ar[r] \ar[u]^\gamma & 0 }\] By Lemma 07E1 and the Jacobi-Zariski sequence for \(\mathbf{F}_p \to K \to K'\) we see that the kernel and cokernel of \(\gamma\) have the same finite dimension. By assumption \(A'\) is regular (and of the same dimension as \(A\), see above) hence the kernel and cokernel of \(\alpha\) have the same dimension. It follows that the kernel and cokernel of \(\beta\) have the same dimension which is what we wanted to show.
The implication (1) \(\Rightarrow\) (2) is trivial. This finishes the proof of the proposition.
Lemma
Let \(k\) be a field of characteristic \(p > 0\). Let \((A, \mathfrak m, K)\) be a Noetherian local \(k\)-algebra. Assume \(A\) is geometrically regular over \(k\). Let \(K/F/k\) be a finitely generated subextension. Let \(\varphi : k[y_1, \ldots, y_m] \to A\) be a \(k\)-algebra map such that \(y_i\) maps to an element of \(F\) in \(K\) and such that \(\text{d}y_1, \ldots, \text{d}y_m\) map to a basis of \(\Omega_{F/k}\). Set \(\mathfrak p = \varphi^{-1}(\mathfrak m)\). Then \[k[y_1, \ldots, y_m]_\mathfrak p \to A\] is flat and \(A/\mathfrak pA\) is regular.
Proof
Set \(A_0 = k[y_1, \ldots, y_m]_\mathfrak p\) with maximal ideal \(\mathfrak m_0\) and residue field \(K_0\). Note that \(\Omega_{A_0/k}\) is free of rank \(m\) and \(\Omega_{A_0/k} \otimes K_0 \to \Omega_{K_0/k}\) is an isomorphism. It is clear that \(A_0\) is geometrically regular over \(k\). Hence \(H_1(L_{K_0/k}) \to \mathfrak m_0/\mathfrak m_0^2\) is an isomorphism, see Proposition 07E5. Now consider \[\xymatrix{ H_1(L_{K_0/k}) \otimes K \ar[d] \ar[r] & \mathfrak m_0/\mathfrak m_0^2 \otimes K \ar[d] \\ H_1(L_{K/k}) \ar[r] & \mathfrak m/\mathfrak m^2 }\] Since the left vertical arrow is injective by Lemma 07E2 and the lower horizontal by Proposition 07E5 we conclude that the right vertical one is too. Hence a regular system of parameters in \(A_0\) maps to part of a regular system of parameters in \(A\). We win by Algebra, Lemmas 07DY and 00NQ.
Topological rings and modules
Let’s quickly discuss some properties of topological abelian groups. An abelian group \(M\) is a topological abelian group if \(M\) is endowed with a topology such that addition \(M \times M \to M\), \((x, y) \mapsto x + y\) and inverse \(M \to M\), \(x \mapsto -x\) are continuous. A homomorphism of topological abelian groups is just a homomorphism of abelian groups which is continuous. The category of commutative topological groups is additive and has kernels and cokernels, but is not abelian (as the axiom \(\Im = \Coim\) doesn’t hold). If \(N \subset M\) is a subgroup, then we think of \(N\) and \(M/N\) as topological groups also, namely using the induced topology on \(N\) and the quotient topology on \(M/N\) (i.e., such that \(M \to M/N\) is submersive). Note that if \(N \subset M\) is an open subgroup, then the topology on \(M/N\) is discrete.
We say the topology on \(M\) is linear if there exists a fundamental system of neighbourhoods of \(0\) consisting of subgroups. If so then these subgroups are also open. An example is the following. Let \(I\) be a directed set and let \(G_i\) be an inverse system of (discrete) abelian groups over \(I\). Then \[G = \lim_{i \in I} G_i\] with the inverse limit topology is linearly topologized with a fundamental system of neighbourhoods of \(0\) given by \(\Ker(G \to G_i)\). Conversely, let \(M\) be a linearly topologized abelian group. Choose any fundamental system of open subgroups \(U_i \subset M\), \(i \in I\) (i.e., the \(U_i\) form a fundamental system of open neighbourhoods and each \(U_i\) is a subgroup of \(M\)). Setting \(i \geq i' \Leftrightarrow U_i \subset U_{i'}\) we see that \(I\) is a directed set. We obtain a homomorphism of linearly topologized abelian groups \[c : M \longrightarrow \lim_{i \in I} M/U_i.\] It is clear that \(M\) is separated (as a topological space) if and only if \(c\) is injective. We say that \(M\) is complete if \(c\) is an isomorphism2. We leave it to the reader to check that this condition is independent of the choice of fundamental system of open subgroups \(\{U_i\}_{i \in I}\) chosen above. In fact the topological abelian group \(M^\wedge = \lim_{i \in I} M/U_i\) is independent of this choice and is sometimes called the completion of \(M\). Any \(G = \lim G_i\) as above is complete, in particular, the completion \(M^\wedge\) is always complete.
Definition
Let \(R\) be a ring and let \(M\) be an \(R\)-module.
We say \(R\) is a topological ring if \(R\) is endowed with a topology such that both addition and multiplication are continuous as maps \(R \times R \to R\) where \(R \times R\) has the product topology. In this case we say \(M\) is a topological module if \(M\) is endowed with a topology such that addition \(M \times M \to M\) and scalar multiplication \(R \times M \to M\) are continuous.
A homomorphism of topological modules is just a continuous \(R\)-module map. A homomorphism of topological rings is a ring homomorphism which is continuous for the given topologies.
We say \(M\) is linearly topologized if \(0\) has a fundamental system of neighbourhoods consisting of submodules. We say \(R\) is linearly topologized if \(0\) has a fundamental system of neighbourhoods consisting of ideals.
If \(R\) is linearly topologized, we say that \(I \subset R\) is an ideal of definition if \(I\) is open and if every neighbourhood of \(0\) contains \(I^n\) for some \(n\).
If \(R\) is linearly topologized, we say that \(R\) is pre-admissible if \(R\) has an ideal of definition.
If \(R\) is linearly topologized, we say that \(R\) is admissible if it is pre-admissible and complete3.
If \(R\) is linearly topologized, we say that \(R\) is pre-adic if there exists an ideal of definition \(I\) such that \(\{I^n\}_{n \geq 0}\) forms a fundamental system of neighbourhoods of \(0\).
If \(R\) is linearly topologized, we say that \(R\) is adic if \(R\) is pre-adic and complete.
Note that a (pre)adic topological ring is the same thing as a (pre)admissible topological ring which has an ideal of definition \(I\) such that \(I^n\) is open for all \(n \geq 1\).
Let \(R\) be a ring and let \(M\) be an \(R\)-module. Let \(I \subset R\) be an ideal. Then we can consider the linear topology on \(R\) which has \(\{I^n\}_{n \geq 0}\) as a fundamental system of neighbourhoods of \(0\). This topology is called the \(I\)-adic topology; \(R\) is a pre-adic topological ring in the \(I\)-adic topology4. Moreover, the linear topology on \(M\) which has \(\{I^nM\}_{n \geq 0}\) as a fundamental system of open neighbourhoods of \(0\) turns \(M\) into a topological \(R\)-module. This is called the \(I\)-adic topology on \(M\). We see that \(M\) is \(I\)-adically complete (as defined in Algebra, Definition 0317) if and only if \(M\) is complete in the \(I\)-adic topology5. In particular, we see that \(R\) is \(I\)-adically complete if and only if \(R\) is an adic topological ring in the \(I\)-adic topology.
As a special case, note that the discrete topology is the \(0\)-adic topology and that any ring in the discrete topology is adic.
Lemma
Let \(\varphi : R \to S\) be a ring map. Let \(I \subset R\) and \(J \subset S\) be ideals and endow \(R\) with the \(I\)-adic topology and \(S\) with the \(J\)-adic topology. Then \(\varphi\) is a homomorphism of topological rings if and only if \(\varphi(I^n) \subset J\) for some \(n \geq 1\).
Proof
Omitted.
Lemma
Let \(M\) be a topological abelian group. Assume \(M\) is linearly topologized, complete, and has a countable fundamental system of neighbourhoods of \(0\). If \(U_n \subset M\), \(n \geq 1\) are open dense subsets, then \(\bigcap_{n \geq 1} U_n\) is dense.
Proof
Let \(U_n\) be as in the statement of the lemma. After replacing \(U_n\) by \(U_1 \cap \ldots \cap U_n\), we may assume that \(U_1 \supset U_2 \supset \ldots\). Let \(M_n\), \(n \in \mathbf{N}\) be a fundamental system of neighbourhoods of \(0\). We may assume that \(M_{n + 1} \subset M_n\). Pick \(x \in M\). We will show that for every \(k \geq 1\) there exists a \(y \in \bigcap_{n \geq 1} U_n\) with \(x - y \in M_k\).
To construct \(y\) we argue as follows. First, we pick a \(y_1 \in U_1\) with \(y_1 \in x + M_k\). This is possible because \(U_1\) is dense and \(x + M_k\) is open. Then we pick a \(k_1 > k\) such that \(y_1 + M_{k_1} \subset U_1\). This is possible because \(U_1\) is open. Next, we pick a \(y_2 \in U_2\) with \(y_2 \in y_1 + M_{k_1}\). This is possible because \(U_2\) is dense and \(y_2 + M_{k_1}\) is open. Then we pick a \(k_2 > k_1\) such that \(y_2 + M_{k_2} \subset U_2\). This is possible because \(U_2\) is open.
Continuing in this fashion we get a converging sequence \(y_i\) of elements of \(M\) with limit \(y\). By construction \(x - y \in M_k\). Since \[y - y_i = (y_{i + 1} - y_i) + (y_{i + 2} - y_{i + 1}) + \ldots\] is in \(M_{k_i}\) we see that \(y \in y_i + M_{k_i} \subset U_i\) for all \(i\) as desired.
Lemma
With same assumptions as Lemma 0CQU if \(M = \bigcup_{n \geq 1} N_n\) for some closed subgroups \(N_n\), then \(N_n\) is open for some \(n\).
Proof
If not, then \(U_n = M \setminus N_n\) is dense for all \(n\) and we get a contradiction with Lemma 0CQU.
Lemma
Let \(u : N \to M\) be a continuous map of topologized abelian groups. Assume that \(M\) separated and that \(N\) is complete, linearly topologized, and has a countable fundamental system of neighbourhoods of \(0\). Then exactly one of the following holds
\(u\) is open, or
for some open subgroup \(N' \subset N\) the image \(u(N')\) is nowhere dense in \(M\).
Proof
Let \(N_n\), \(n \in \mathbf{N}\) be a fundamental system of neighbourhoods of \(0\). We may assume that \(N_n\) is a subgroup and that \(N_{n + 1} \subset N_n\). If (2) does not hold, then the closure \(M_n\) of \(u(N_n)\) is an open subgroup for \(n = 1, 2, 3, \ldots\). Since \(u\) is continuous, we see that \(M_n\), \(n \in \mathbf{N}\) must be a fundamental system of open neighbourhoods of \(0\) in \(M\). Also, since \(M_n\) is the closure of \(u(N_n)\) we see that \[u(N_n) + M_{n + 1} = M_n\] for all \(n \geq 1\). Pick \(x_1 \in M_1\). Then we can inductively choose \(y_i \in N_i\) and \(x_{i + 1} \in M_{i + 1}\) such that \[u(y_i) + x_{i + 1} = x_i\] The element \(y = y_1 + y_2 + y_3 + \ldots\) of \(N\) exists because \(N\) is complete. Whereupon we see that \(x_1 = u(y)\) because \(M\) is separated. Thus \(M_1 = u(N_1)\). In exactly the same way the reader shows that \(M_i = u(N_i)\) for all \(i \geq 2\) and we see that \(u\) is open.
Formally smooth maps of topological rings
There is a version of formal smoothness which applies to homomorphisms of topological rings.
Definition
Let \(R \to S\) be a homomorphism of topological rings with \(R\) and \(S\) linearly topologized. We say \(S\) is formally smooth over \(R\) if for every commutative solid diagram \[\xymatrix{ S \ar[r] \ar@{-->}[rd] & A/J \\ R \ar[r] \ar[u] & A \ar[u] }\] of homomorphisms of topological rings where \(A\) is a discrete ring and \(J \subset A\) is an ideal of square zero, a dotted arrow exists which makes the diagram commute.
We will mostly use this notion when given ideals \(\mathfrak m \subset R\) and \(\mathfrak n \subset S\) and we endow \(R\) with the \(\mathfrak m\)-adic topology and \(S\) with the \(\mathfrak n\)-adic topology. Continuity of \(\varphi : R \to S\) holds if and only if \(\varphi(\mathfrak m^m) \subset \mathfrak n\) for some \(m \geq 1\), see Lemma 07E9. It turns out that in this case only the topology on \(S\) is relevant.
Lemma
Let \(\varphi : R \to S\) be a ring map.
If \(R \to S\) is formally smooth in the sense of Algebra, Definition 00TI, then \(R \to S\) is formally smooth for any linear topology on \(R\) and any pre-adic topology on \(S\) such that \(R \to S\) is continuous.
Let \(\mathfrak n \subset S\) and \(\mathfrak m \subset R\) ideals such that \(\varphi\) is continuous for the \(\mathfrak m\)-adic topology on \(R\) and the \(\mathfrak n\)-adic topology on \(S\). Then the following are equivalent
\(\varphi\) is formally smooth for the \(\mathfrak m\)-adic topology on \(R\) and the \(\mathfrak n\)-adic topology on \(S\), and
\(\varphi\) is formally smooth for the discrete topology on \(R\) and the \(\mathfrak n\)-adic topology on \(S\).
Proof
Assume \(R \to S\) is formally smooth in the sense of Algebra, Definition 00TI. If \(S\) has a pre-adic topology, then there exists an ideal \(\mathfrak n \subset S\) such that \(S\) has the \(\mathfrak n\)-adic topology. Suppose given a solid commutative diagram as in Definition 07EB. Continuity of \(S \to A/J\) means that \(\mathfrak n^k\) maps to zero in \(A/J\) for some \(k \geq 1\), see Lemma 07E9. We obtain a ring map \(\psi : S \to A\) from the assumed formal smoothness of \(S\) over \(R\). Then \(\psi(\mathfrak n^k) \subset J\) hence \(\psi(\mathfrak n^{2k}) = 0\) as \(J^2 = 0\). Hence \(\psi\) is continuous by Lemma 07E9. This proves (1).
The proof of (2)(b) \(\Rightarrow\) (2)(a) is the same as the proof of (1). Assume (2)(a). Suppose given a solid commutative diagram as in Definition 07EB where we use the discrete topology on \(R\). Since \(\varphi\) is continuous we see that \(\varphi(\mathfrak m^n) \subset \mathfrak n\) for some \(n \geq 1\). As \(S \to A/J\) is continuous we see that \(\mathfrak n^k\) maps to zero in \(A/J\) for some \(k \geq 1\). Hence \(\mathfrak m^{nk}\) maps into \(J\) under the map \(R \to A\). Thus \(\mathfrak m^{2nk}\) maps to zero in \(A\) and we see that \(R \to A\) is continuous in the \(\mathfrak m\)-adic topology. Thus (2)(a) gives a dotted arrow as desired.
Definition
Let \(R \to S\) be a ring map. Let \(\mathfrak n \subset S\) be an ideal. If the equivalent conditions (2)(a) and (2)(b) of Lemma 07EC hold, then we say \(R \to S\) is formally smooth for the \(\mathfrak n\)-adic topology.
This property is inherited by the completions.
Lemma
Let \((R, \mathfrak m)\) and \((S, \mathfrak n)\) be rings endowed with finitely generated ideals. Endow \(R\) and \(S\) with the \(\mathfrak m\)-adic and \(\mathfrak n\)-adic topologies. Let \(R \to S\) be a homomorphism of topological rings. The following are equivalent
\(R \to S\) is formally smooth for the \(\mathfrak n\)-adic topology,
\(R \to S^\wedge\) is formally smooth for the \(\mathfrak n^\wedge\)-adic topology,
\(R^\wedge \to S^\wedge\) is formally smooth for the \(\mathfrak n^\wedge\)-adic topology.
Here \(R^\wedge\) and \(S^\wedge\) are the \(\mathfrak m\)-adic and \(\mathfrak n\)-adic completions of \(R\) and \(S\).
Proof
The assumption that \(\mathfrak m\) is finitely generated implies that \(R^\wedge\) is \(\mathfrak mR^\wedge\)-adically complete, that \(\mathfrak mR^\wedge = \mathfrak m^\wedge\) and that \(R^\wedge/\mathfrak m^nR^\wedge = R/\mathfrak m^n\), see Algebra, Lemma 05GG and its proof. Similarly for \((S, \mathfrak n)\). Thus it is clear that diagrams as in Definition 07EB for the cases (1), (2), and (3) are in 1-to-1 correspondence.
The advantage of working with adic rings is that one gets a stronger lifting property.
Lemma
Let \(R \to S\) be a ring map. Let \(\mathfrak n\) be an ideal of \(S\). Assume that \(R \to S\) is formally smooth in the \(\mathfrak n\)-adic topology. Consider a solid commutative diagram \[\xymatrix{ S \ar[r]_\psi \ar@{-->}[rd] & A/J \\ R \ar[r] \ar[u] & A \ar[u] }\] of homomorphisms of topological rings where \(A\) is adic and \(A/J\) is the quotient (as topological ring) of \(A\) by a closed ideal \(J \subset A\) such that \(J^t\) is contained in an ideal of definition of \(A\) for some \(t \geq 1\). Then there exists a dotted arrow in the category of topological rings which makes the diagram commute.
Proof
Let \(I \subset A\) be an ideal of definition so that \(I \supset J^t\) for some \(t\). Then \(A = \lim A/I^n\) and \(A/J = \lim A/J + I^n\) because \(J\) is assumed closed. Consider the following diagram of discrete \(R\) algebras \(A_{n, m} = A/J^n + I^m\): \[\xymatrix{ A/J^3 + I^3 \ar[r] \ar[d] & A/J^2 + I^3 \ar[r] \ar[d] & A/J + I^3 \ar[d] \\ A/J^3 + I^2 \ar[r] \ar[d] & A/J^2 + I^2 \ar[r] \ar[d] & A/J + I^2 \ar[d] \\ A/J^3 + I \ar[r] & A/J^2 + I \ar[r] & A/J + I }\] Note that each of the commutative squares defines a surjection \[A_{n + 1, m + 1} \longrightarrow A_{n + 1, m} \times_{A_{n, m}} A_{n, m + 1}\] of \(R\)-algebras whose kernel has square zero. We will inductively construct \(R\)-algebra maps \(\varphi_{n, m} : S \to A_{n, m}\). Namely, we have the maps \(\varphi_{1, m} = \psi \bmod J + I^m\). Note that each of these maps is continuous as \(\psi\) is. We can inductively choose the maps \(\varphi_{n, 1}\) by starting with our choice of \(\varphi_{1, 1}\) and lifting up, using the formal smoothness of \(S\) over \(R\), along the bottom row of the diagram above. We construct the remaining maps \(\varphi_{n, m}\) by induction on \(n + m\). Namely, we choose \(\varphi_{n + 1, m + 1}\) by lifting the pair \((\varphi_{n + 1, m}, \varphi_{n, m + 1})\) along the displayed surjection above (again using the formal smoothness of \(S\) over \(R\)). In this way all of the maps \(\varphi_{n, m}\) are compatible with the transition maps of the system. As \(J^t \subset I\) we see that for example \(\varphi_n = \varphi_{nt, n} \bmod I^n\) induces a map \(S \to A/I^n\). Taking the limit \(\varphi = \lim \varphi_n\) we obtain a map \(S \to A = \lim A/I^n\). The composition into \(A/J\) agrees with \(\psi\) as we have seen that \(A/J = \lim A/J + I^n\). Finally we show that \(\varphi\) is continuous. Namely, we know that \(\psi(\mathfrak n^r) \subset J + I/J\) for some \(r \geq 1\) by our assumption that \(\psi\) is a morphism of topological rings, see Lemma 07E9. Hence \(\varphi(\mathfrak n^r) \subset J + I\) hence \(\varphi(\mathfrak n^{rt}) \subset I\) as desired.
Lemma
Let \(R \to S\) be a ring map. Let \(\mathfrak n \subset \mathfrak n' \subset S\) be ideals. If \(R \to S\) is formally smooth for the \(\mathfrak n\)-adic topology, then \(R \to S\) is formally smooth for the \(\mathfrak n'\)-adic topology.
Proof
Omitted.
Lemma
A composition of formally smooth continuous homomorphisms of linearly topologized rings is formally smooth.
Proof
Omitted. (Hint: This is completely formal, and follows from considering a suitable diagram.)
Lemma
Let \(R\), \(S\) be rings. Let \(\mathfrak n \subset S\) be an ideal. Let \(R \to S\) be formally smooth for the \(\mathfrak n\)-adic topology. Let \(R \to R'\) be any ring map. Then \(R' \to S' = S \otimes_R R'\) is formally smooth in the \(\mathfrak n' = \mathfrak nS'\)-adic topology.
Proof
Let a solid diagram \[\xymatrix{ S \ar[r] \ar@{-->}[rrd] & S' \ar[r] \ar@{-->}[rd] & A/J \\ R \ar[u] \ar[r] & R' \ar[r] \ar[u] & A \ar[u] }\] as in Definition 07EB be given. Then the composition \(S \to S' \to A/J\) is continuous. By assumption the longer dotted arrow exists. By the universal property of tensor product we obtain the shorter dotted arrow.
We have seen descent for formal smoothness along faithfully flat ring maps in Algebra, Lemma 06CM. Something similar holds in the current setting of topological rings. However, here we just prove the following very simple and easy to prove version which is already quite useful.
Lemma
Let \(R\), \(S\) be rings. Let \(\mathfrak n \subset S\) be an ideal. Let \(R \to R'\) be a ring map. Set \(S' = S \otimes_R R'\) and \(\mathfrak n' = \mathfrak nS\). If
the map \(R \to R'\) embeds \(R\) as a direct summand of \(R'\) as an \(R\)-module, and
\(R' \to S'\) is formally smooth for the \(\mathfrak n'\)-adic topology,
then \(R \to S\) is formally smooth in the \(\mathfrak n\)-adic topology.
Proof
Let a solid diagram \[\xymatrix{ S \ar[r] & A/J \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Definition 07EB be given. Set \(A' = A \otimes_R R'\) and \(J' = \Im(J \otimes_R R' \to A')\). The base change of the diagram above is the diagram \[\xymatrix{ S' \ar[r] \ar@{-->}[rd]^{\psi'} & A'/J' \\ R' \ar[u] \ar[r] & A' \ar[u] }\] with continuous arrows. By condition (2) we obtain the dotted arrow \(\psi' : S' \to A'\). Using condition (1) choose a direct summand decomposition \(R' = R \oplus C\) as \(R\)-modules. (Warning: \(C\) isn’t an ideal in \(R'\).) Then \(A' = A \oplus A \otimes_R C\). Set \[J'' = \Im(J \otimes_R C \to A \otimes_R C) \subset J' \subset A'.\] Then \(J' = J \oplus J''\) as \(A\)-modules. The image of the composition \(\psi : S \to A'\) of \(\psi'\) with \(S \to S'\) is contained in \(A + J' = A \oplus J''\). However, in the ring \(A + J' = A \oplus J''\) the \(A\)-submodule \(J''\) is an ideal! (Use that \(J^2 = 0\).) Hence the composition \(S \to A + J' \to (A + J')/J'' = A\) is the arrow we were looking for.
Formally smooth maps of local rings
In the case of a local homomorphism of local rings one can limit the diagrams for which the lifting property has to be checked. Please compare with Algebra, Lemma 02HT.
Lemma
Let \((R, \mathfrak m) \to (S, \mathfrak n)\) be a local homomorphism of local rings. The following are equivalent
\(R \to S\) is formally smooth in the \(\mathfrak n\)-adic topology,
for every solid commutative diagram \[\xymatrix{ S \ar[r] \ar@{-->}[rd] & A/J \\ R \ar[r] \ar[u] & A \ar[u] }\] of local homomorphisms of local rings where \(J \subset A\) is an ideal of square zero, \(\mathfrak m_A^n = 0\) for some \(n > 0\), and \(S \to A/J\) induces an isomorphism on residue fields, a dotted arrow exists which makes the diagram commute.
If \(S\) is Noetherian these conditions are also equivalent to
same as in (2) but only for diagrams where in addition \(A \to A/J\) is a small extension (Algebra, Definition 02HS).
Proof
The implication (1) \(\Rightarrow\) (2) follows from the definitions. Consider a diagram \[\xymatrix{ S \ar[r] \ar@{-->}[rd] & A/J \\ R \ar[r] \ar[u] & A \ar[u] }\] as in Definition 07EB for the \(\mathfrak m\)-adic topology on \(R\) and the \(\mathfrak n\)-adic topology on \(S\). Pick \(m > 0\) with \(\mathfrak n^m(A/J) = 0\) (possible by continuity of maps in diagram). Consider the subring \(A'\) of \(A\) which is the inverse image of the image of \(S\) in \(A/J\). Set \(J' = J\) viewed as an ideal in \(A'\). Then \(J'\) is an ideal of square zero in \(A'\) and \(A'/J'\) is a quotient of \(S/\mathfrak n^m\). Hence \(A'\) is local and \(\mathfrak m_{A'}^{2m} = 0\). Thus we get a diagram \[\xymatrix{ S \ar[r] \ar@{-->}[rd] & A'/J' \\ R \ar[r] \ar[u] & A' \ar[u] }\] as in (2). If we can construct the dotted arrow in this diagram, then we obtain the dotted arrow in the original one by composing with \(A' \to A\). In this way we see that (2) implies (1).
Assume \(S\) Noetherian. The implication (1) \(\Rightarrow\) (3) is immediate. Assume (3) and suppose a diagram as in (2) is given. Then \(\mathfrak m_A^n J = 0\) for some \(n > 0\). Considering the maps \[A \to A/\mathfrak m_A^{n - 1}J \to \ldots \to A/\mathfrak mJ \to A/J\] we see that it suffices to produce the lifting if \(\mathfrak m_A J = 0\). Assume \(\mathfrak m_A J = 0\) and let \(A' \subset A\) be the ring constructed above. Then \(A'/J'\) is Artinian as a quotient of the Artinian local ring \(S/\mathfrak n^m\). Thus it suffices to show that given property (3) we can find the dotted arrow in diagrams as in (2) with \(A/J\) Artinian and \(\mathfrak m_A J = 0\). Let \(\kappa\) be the common residue field of \(A\), \(A/J\), and \(S\). By (3), if \(J_0 \subset J\) is an ideal with \(\dim_\kappa(J/J_0) = 1\), then we can produce a dotted arrow \(S \to A/J_0\). Taking the product we obtain \[S \longrightarrow \prod\nolimits_{J_0 \text{ as above}} A/J_0\] Clearly the image of this arrow is contained in the sub \(R\)-algebra \(A'\) of elements which map into the small diagonal \(A/J \subset \prod_{J_0} A/J\). Let \(J' \subset A'\) be the elements mapping to zero in \(A/J\). Then \(J'\) is an ideal of square zero and as \(\kappa\)-vector space equal to \[J' = \prod\nolimits_{J_0 \text{ as above}} J/J_0\] Thus the map \(J \to J'\) is injective. By the theory of vector spaces we can choose a splitting \(J' = J \oplus M\). It follows that \[A' = A \oplus M\] as an \(R\)-algebra. Hence the map \(S \to A'\) can be composed with the projection \(A' \to A\) to give the desired dotted arrow thereby finishing the proof of the lemma.
The following lemma will be improved on in Section 07EK.
Lemma
Let \(k\) be a field and let \((A, \mathfrak m, K)\) be a Noetherian local \(k\)-algebra. If \(k \to A\) is formally smooth for the \(\mathfrak m\)-adic topology, then \(A\) is a regular local ring.
Proof
Let \(k_0 \subset k\) be the prime field. Then \(k_0\) is perfect, hence \(k / k_0\) is separable, hence formally smooth by Algebra, Lemma 0320. By Lemmas 07EC and 07EF we see that \(k_0 \to A\) is formally smooth for the \(\mathfrak m\)-adic topology on \(A\). Hence we may assume \(k = \mathbf{Q}\) or \(k = \mathbf{F}_p\).
By Algebra, Lemmas 00MC and 00OF it suffices to prove the completion \(A^\wedge\) is regular. By Lemma 07ED we may replace \(A\) by \(A^\wedge\). Thus we may assume that \(A\) is a Noetherian complete local ring. By the Cohen structure theorem (Algebra, Theorem 032A) there exist a map \(K \to A\). As \(k\) is the prime field we see that \(K \to A\) is a \(k\)-algebra map.
Let \(x_1, \ldots, x_n \in \mathfrak m\) be elements whose images form a basis of \(\mathfrak m/\mathfrak m^2\). Set \(T = K[[X_1, \ldots, X_n]]\). Note that \[A/\mathfrak m^2 \cong K[x_1, \ldots, x_n]/(x_ix_j)\] and \[T/\mathfrak m_T^2 \cong K[X_1, \ldots, X_n]/(X_iX_j).\] Let \(A/\mathfrak m^2 \to T/m_T^2\) be the local \(K\)-algebra isomorphism given by mapping the class of \(x_i\) to the class of \(X_i\). Denote \(f_1 : A \to T/\mathfrak m_T^2\) the composition of this isomorphism with the quotient map \(A \to A/\mathfrak m^2\). The assumption that \(k \to A\) is formally smooth in the \(\mathfrak m\)-adic topology means we can lift \(f_1\) to a map \(f_2 : A \to T/\mathfrak{m}_T^3\), then to a map \(f_3 : A \to T/\mathfrak{m}_T^4\), and so on, for all \(n \geq 1\). Warning: the maps \(f_n\) are continuous \(k\)-algebra maps and may not be \(K\)-algebra maps. We get an induced map \(f : A \to T = \lim T/\mathfrak m_T^n\) of local \(k\)-algebras. By our choice of \(f_1\), the map \(f\) induces an isomorphism \(\mathfrak m/\mathfrak m^2 \to \mathfrak m_T/\mathfrak m_T^2\) hence each \(f_n\) is surjective and we conclude \(f\) is surjective as \(A\) is complete. This implies \(\dim(A) \geq \dim(T) = n\). Hence \(A\) is regular by definition. (It also follows that \(f\) is an isomorphism.)
Lemma
Let \(k\) be a field. Let \((A, \mathfrak m, \kappa)\) be a complete local \(k\)-algebra. If \(\kappa/k\) is separable, then there exists a \(k\)-algebra map \(\kappa \to A\) such that \(\kappa \to A \to \kappa\) is \(\text{id}_\kappa\).
Proof
By Algebra, Proposition 0322 the extension \(\kappa/k\) is formally smooth. By Lemma 07EC \(k \to \kappa\) is formally smooth in the sense of Definition 07EB. Then we get \(\kappa \to A\) from Lemma 07NJ.
Lemma
Let \(k\) be a field. Let \((A, \mathfrak m, \kappa)\) be a complete local \(k\)-algebra. If \(\kappa/k\) is separable and \(A\) regular, then there exists an isomorphism of \(A \cong \kappa[[t_1, \ldots, t_d]]\) as \(k\)-algebras.
Proof
Choose \(\kappa \to A\) as in Lemma 0C34 and apply Algebra, Lemma 0C0S.
The following result will be improved on in Section 07EK
Lemma
Let \(k\) be a field. Let \((A, \mathfrak m, K)\) be a regular local \(k\)-algebra such that \(K/k\) is separable. Then \(k \to A\) is formally smooth in the \(\mathfrak m\)-adic topology.
Proof
It suffices to prove that the completion of \(A\) is formally smooth over \(k\), see Lemma 07ED. Hence we may assume that \(A\) is a complete local regular \(k\)-algebra with residue field \(K\) separable over \(k\). By Lemma 0C35 we see that \(A = K[[x_1, \ldots, x_n]]\).
The power series ring \(K[[x_1, \ldots, x_n]]\) is formally smooth over \(k\). Namely, \(K\) is formally smooth over \(k\) and \(K[x_1, \ldots, x_n]\) is formally smooth over \(K\) as a polynomial algebra. Hence \(K[x_1, \ldots, x_n]\) is formally smooth over \(k\) by Algebra, Lemma 031H. It follows that \(k \to K[x_1, \ldots, x_n]\) is formally smooth for the \((x_1, \ldots, x_n)\)-adic topology by Lemma 07EC. Finally, it follows that \(k \to K[[x_1, \ldots, x_n]]\) is formally smooth for the \((x_1, \ldots, x_n)\)-adic topology by Lemma 07ED.
Lemma
Let \(A \to B\) be a finite type ring map with \(A\) Noetherian. Let \(\mathfrak q \subset B\) be a prime ideal lying over \(\mathfrak p \subset A\). The following are equivalent
\(A \to B\) is smooth at \(\mathfrak q\), and
\(A_\mathfrak p \to B_\mathfrak q\) is formally smooth in the \(\mathfrak q\)-adic topology.
Proof
The implication (2) \(\Rightarrow\) (1) follows from Algebra, Lemma 02HT. Conversely, if \(A \to B\) is smooth at \(\mathfrak q\), then \(A \to B_g\) is smooth for some \(g \in B\), \(g \not \in \mathfrak q\). Then \(A \to B_g\) is formally smooth by Algebra, Proposition 00TN. Hence \(A_\mathfrak p \to B_\mathfrak q\) is formally smooth as localization preserves formal smoothness (for example by the criterion of Algebra, Proposition 031J and the fact that the cotangent complex behaves well with respect to localization, see Algebra, Lemmas 07BS and 00S7). Finally, Lemma 07EC implies that \(A_\mathfrak p \to B_\mathfrak q\) is formally smooth in the \(\mathfrak q\)-adic topology.
Some results on power series rings
Questions on formally smooth maps between Noetherian local rings can often be reduced to questions on maps between power series rings. In this section we prove some helper lemmas to facilitate this kind of argument.
Lemma
Let \(K\) be a field of characteristic \(0\) and \(A = K[[x_1, \ldots, x_n]]\). Let \(L\) be a field of characteristic \(p > 0\) and \(B = L[[x_1, \ldots, x_n]]\). Let \(\Lambda\) be a Cohen ring. Let \(C = \Lambda[[x_1, \ldots, x_n]]\).
\(\mathbf{Q} \to A\) is formally smooth in the \(\mathfrak m_A\)-adic topology.
\(\mathbf{F}_p \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology.
\(\mathbf{Z} \to C\) is formally smooth in the \(\mathfrak m_C\)-adic topology.
Proof
By the universal property of power series rings it suffices to prove:
\(\mathbf{Q} \to K\) is formally smooth.
\(\mathbf{F}_p \to L\) is formally smooth.
\(\mathbf{Z} \to \Lambda\) is formally smooth in the \(\mathfrak m_\Lambda\)-adic topology.
The first two are Algebra, Proposition 0322. The third follows from Algebra, Lemma 0329 since for any test diagram as in Definition 07EB some power of \(p\) will be zero in \(A/J\) and hence some power of \(p\) will be zero in \(A\).
Lemma
Let \(K\) be a field and \(A = K[[x_1, \ldots, x_n]]\). Let \(\Lambda\) be a Cohen ring and let \(B = \Lambda[[x_1, \ldots, x_n]]\).
If \(y_1, \ldots, y_n \in A\) is a regular system of parameters then \(K[[y_1, \ldots, y_n]] \to A\) is an isomorphism.
If \(z_1, \ldots, z_r \in A\) form part of a regular system of parameters for \(A\), then \(r \leq n\) and \(A/(z_1, \ldots, z_r) \cong K[[y_1, \ldots, y_{n - r}]]\).
If \(p, y_1, \ldots, y_n \in B\) is a regular system of parameters then \(\Lambda[[y_1, \ldots, y_n]] \to B\) is an isomorphism.
If \(p, z_1, \ldots, z_r \in B\) form part of a regular system of parameters for \(B\), then \(r \leq n\) and \(B/(z_1, \ldots, z_r) \cong \Lambda[[y_1, \ldots, y_{n - r}]]\).
Proof
Proof of (1). Set \(A' = K[[y_1, \ldots, y_n]]\). It is clear that the map \(A' \to A\) induces an isomorphism \(A'/\mathfrak m_{A'}^n \to A/\mathfrak m_A^n\) for all \(n \geq 1\). Since \(A\) and \(A'\) are both complete we deduce that \(A' \to A\) is an isomorphism. Proof of (2). Extend \(z_1, \ldots, z_r\) to a regular system of parameters \(z_1, \ldots, z_r, y_1, \ldots, y_{n - r}\) of \(A\). Consider the map \(A' = K[[z_1, \ldots, z_r, y_1, \ldots, y_{n - r}]] \to A\). This is an isomorphism by (1). Hence (2) follows as it is clear that \(A'/(z_1, \ldots, z_r) \cong K[[y_1, \ldots, y_{n - r}]]\). The proofs of (3) and (4) are exactly the same as the proofs of (1) and (2).
Lemma
Let \(A \to B\) be a local homomorphism of Noetherian complete local rings. Then there exists a commutative diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] with the following properties:
the horizontal arrows are surjective,
if the characteristic of \(A/\mathfrak m_A\) is zero, then \(S\) and \(R\) are power series rings over fields,
if the characteristic of \(A/\mathfrak m_A\) is \(p > 0\), then \(S\) and \(R\) are power series rings over Cohen rings, and
\(R \to S\) maps a regular system of parameters of \(R\) to part of a regular system of parameters of \(S\).
In particular \(R \to S\) is flat (see Algebra, Lemma 07DY) with regular fibre \(S/\mathfrak m_R S\) (see Algebra, Lemma 00NQ).
Proof
Use the Cohen structure theorem (Algebra, Theorem 032A) to choose a surjection \(S \to B\) as in the statement of the lemma where we choose \(S\) to be a power series over a Cohen ring if the residue characteristic is \(p > 0\) and a power series over a field else. Let \(J \subset S\) be the kernel of \(S \to B\). Next, choose a surjection \(R = \Lambda[[x_1, \ldots, x_n]] \to A\) where we choose \(\Lambda\) to be a Cohen ring if the residue characteristic of \(A\) is \(p > 0\) and \(\Lambda\) equal to the residue field of \(A\) otherwise. We lift the composition \(\Lambda[[x_1, \ldots, x_n]] \to A \to B\) to a map \(\varphi : R \to S\). This is possible because \(\Lambda[[x_1, \ldots, x_n]]\) is formally smooth over \(\mathbf{Z}\) in the \(\mathfrak m\)-adic topology (see Lemma 07NL) by an application of Lemma 07NJ. Finally, we replace \(\varphi\) by the map \(\varphi' : R = \Lambda[[x_1, \ldots, x_n]] \to S' = S[[y_1, \ldots, y_n]]\) with \(\varphi'|_\Lambda = \varphi|_\Lambda\) and \(\varphi'(x_i) = \varphi(x_i) + y_i\). We also replace \(S \to B\) by the map \(S' \to B\) which maps \(y_i\) to zero. After this replacement it is clear that a regular system of parameters of \(R\) maps to part of a regular sequence in \(S'\) and we win.
There should be an elementary proof of the following lemma.
Lemma
Let \(S \to R\) and \(S' \to R\) be surjective maps of complete Noetherian local rings. Then \(S \times_R S'\) is a complete Noetherian local ring.
Proof
Let \(k\) be the residue field of \(R\). If the characteristic of \(k\) is \(p > 0\), then we denote \(\Lambda\) a Cohen ring (Algebra, Definition 0327) with residue field \(k\) (Algebra, Lemma 0328). If the characteristic of \(k\) is \(0\) we set \(\Lambda = k\). Choose a surjection \(\Lambda[[x_1, \ldots, x_n]] \to R\) (as in the Cohen structure theorem, see Algebra, Theorem 032A) and lift this to maps \(\Lambda[[x_1, \ldots, x_n]] \to S\) and \(\varphi : \Lambda[[x_1, \ldots, x_n]] \to S\) and \(\varphi' : \Lambda[[x_1, \ldots, x_n]] \to S'\) using Lemmas 07NL and 07NJ. Next, choose \(f_1, \ldots, f_m \in S\) generating the kernel of \(S \to R\) and \(f'_1, \ldots, f'_{m'} \in S'\) generating the kernel of \(S' \to R\). Then the map \[\Lambda[[x_1, \ldots, x_n, y_1, \ldots, y_m, z_1, \ldots, z_{m'}]] \longrightarrow S \times_R S,\] which sends \(x_i\) to \((\varphi(x_i), \varphi'(x_i))\) and \(y_j\) to \((f_j, 0)\) and \(z_{j'}\) to \((0, f'_j)\) is surjective. Thus \(S \times_R S'\) is a quotient of a complete local ring, whence complete.
Geometric regularity and formal smoothness
In this section we combine the results of the previous sections to prove the following characterization of geometrically regular local rings over fields. We then recycle some of our arguments to prove a characterization of formally smooth maps in the \(\mathfrak m\)-adic topology between Noetherian local rings.
Theorem
Let \(k\) be a field. Let \((A, \mathfrak m, K)\) be a Noetherian local \(k\)-algebra. If the characteristic of \(k\) is zero then the following are equivalent
\(A\) is a regular local ring, and
\(k \to A\) is formally smooth in the \(\mathfrak m\)-adic topology.
If the characteristic of \(k\) is \(p > 0\) then the following are equivalent
\(A\) is geometrically regular over \(k\),
\(k \to A\) is formally smooth in the \(\mathfrak m\)-adic topology.
for all \(k \subset k' \subset k^{1/p}\) finite over \(k\) the ring \(A \otimes_k k'\) is regular,
\(A\) is regular and the canonical map \(H_1(L_{K/k}) \to \mathfrak m/\mathfrak m^2\) is injective, and
\(A\) is regular and the map \(\Omega_{k/\mathbf{F}_p} \otimes_k K \to \Omega_{A/\mathbf{F}_p} \otimes_A K\) is injective.
Proof
If the characteristic of \(k\) is zero, then the equivalence of (1) and (2) follows from Lemmas 07EI and 07EJ.
If the characteristic of \(k\) is \(p > 0\), then it follows from Proposition 07E5 that (1), (3), (4), and (5) are equivalent. Assume (2) holds. By Lemma 07EG we see that \(k' \to A' = A \otimes_k k'\) is formally smooth for the \(\mathfrak m' = \mathfrak mA'\)-adic topology. Hence if \(k \subset k'\) is finite purely inseparable, then \(A'\) is a regular local ring by Lemma 07EI. Thus we see that (1) holds.
Finally, we will prove that (5) implies (2). Choose a solid diagram \[\xymatrix{ A \ar[r]_{\bar\psi} \ar@{-->}[rd] & B/J \\ k \ar[u]^i \ar[r]^\varphi & B \ar[u]_\pi }\] as in Definition 07EB. As \(J^2 = 0\) we see that \(J\) has a canonical \(B/J\) module structure and via \(\bar\psi\) an \(A\)-module structure. As \(\bar\psi\) is continuous for the \(\mathfrak m\)-adic topology we see that \(\mathfrak m^nJ = 0\) for some \(n\). Hence we can filter \(J\) by \(B/J\)-submodules \(0 \subset J_1 \subset J_2 \subset \ldots \subset J_n = J\) such that each quotient \(J_{t + 1}/J_t\) is annihilated by \(\mathfrak m\). Considering the sequence of ring maps \(B \to B/J_1 \to B/J_2 \to \ldots \to B/J\) we see that it suffices to prove the existence of the dotted arrow when \(J\) is annihilated by \(\mathfrak m\), i.e., when \(J\) is a \(K\)-vector space.
Assume given a diagram as above such that \(J\) is annihilated by \(\mathfrak m\). By Lemma 07EJ we see that \(\mathbf{F}_p \to A\) is formally smooth in the \(\mathfrak m\)-adic topology. Hence we can find a ring map \(\psi : A \to B\) such that \(\pi \circ \psi = \bar \psi\). Then \(\psi \circ i, \varphi : k \to B\) are two maps whose compositions with \(\pi\) are equal. Hence \(D = \psi \circ i - \varphi : k \to J\) is a derivation. By Algebra, Lemma 00RO we can write \(D = \xi \circ \text{d}\) for some \(k\)-linear map \(\xi : \Omega_{k/\mathbf{F}_p} \to J\). Using the \(K\)-vector space structure on \(J\) we extend \(\xi\) to a \(K\)-linear map \(\xi' : \Omega_{k/\mathbf{F}_p} \otimes_k K \to J\). Using (5) we can find a \(K\)-linear map \(\xi'' : \Omega_{A/\mathbf{F}_p} \otimes_A K\) whose restriction to \(\Omega_{k/\mathbf{F}_p} \otimes_k K\) is \(\xi'\). Write \[D' : A \xrightarrow{\text{d}} \Omega_{A/\mathbf{F}_p} \to \Omega_{A/\mathbf{F}_p} \otimes_A K \xrightarrow{\xi''} J.\] Finally, set \(\psi' = \psi - D' : A \to B\). The reader verifies that \(\psi'\) is a ring map such that \(\pi \circ \psi' = \bar \psi\) and such that \(\psi' \circ i = \varphi\) as desired.
Example
Let \(k\) be a field of characteristic \(p > 0\). Suppose that \(a \in k\) is an element which is not a \(p\)th power. A standard example of a geometrically regular local \(k\)-algebra whose residue field is purely inseparable over \(k\) is the ring \[A = k[x, y]_{(x, y^p - a)}/(y^p - a - x)\] Namely, \(A\) is a localization of a smooth algebra over \(k\) hence \(k \to A\) is formally smooth, hence \(k \to A\) is formally smooth for the \(\mathfrak m\)-adic topology. A closely related example is the following. Let \(k = \mathbf{F}_p(s)\) and \(K = \mathbf{F}_p(t)^{perf}\). We claim the ring map \[k \longrightarrow A = K[[x]],\quad s \longmapsto t + x\] is formally smooth for the \((x)\)-adic topology on \(A\). Namely, \(\Omega_{k/\mathbf{F}_p}\) is \(1\)-dimensional with basis \(\text{d}s\). It maps to the element \(\text{d}x + \text{d}t = \text{d}x\) in \(\Omega_{A/\mathbf{F}_p}\). We leave it to the reader to show that \(\Omega_{A/\mathbf{F}_p}\) is free on \(\text{d}x\) as an \(A\)-module. Hence we see that condition (5) of Theorem 07EL holds and we conclude that \(k \to A\) is formally smooth in the \((x)\)-adic topology.
Lemma
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Assume \(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology. Then \(A \to B\) is flat.
Proof
We may assume that \(A\) and \(B\) a Noetherian complete local rings by Lemma 07ED and Algebra, Lemma 0316 (this also uses Algebra, Lemma 0584 and 00MC to see that flatness of the map on completions implies flatness of \(A \to B\)). Choose a commutative diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Lemma 07NN with \(R \to S\) flat. Let \(I \subset R\) be the kernel of \(R \to A\). Because \(B\) is formally smooth over \(A\) we see that the \(A\)-algebra map \[S/IS \longrightarrow B\] has a section, see Lemma 07NJ. Hence \(B\) is a direct summand of the flat \(A\)-module \(S/IS\) (by base change of flatness, see Algebra, Lemma 00HI), whence flat.
Lemma
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Assume \(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology. Let \(K\) be the residue field of \(B\). Then the Jacobi-Zariski sequence for \(A \to B \to K\) gives an exact sequence \[0 \to H_1(\NL_{K/A}) \to \mathfrak m_B/\mathfrak m_B^2 \to \Omega_{B/A} \otimes_B K \to \Omega_{K/A} \to 0\]
Proof
Observe that \(\mathfrak m_B/\mathfrak m_B^2 = H_1(\NL_{K/B})\) by Algebra, Lemma 07BP. By Algebra, Lemma 00S2 it remains to show injectivity of \(H_1(\NL_{K/A}) \to \mathfrak m_B/\mathfrak m_B^2\). With \(k\) the residue field of \(A\), the Jacobi-Zariski sequence for \(A \to k \to K\) gives \(\Omega_{K/A} = \Omega_{K/k}\) and an exact sequence \[\mathfrak m_A/\mathfrak m_A^2 \otimes_k K \to H_1(\NL_{K/A}) \to H_1(\NL_{K/k}) \to 0\] Set \(\overline{B} = B \otimes_A k\). Since \(\overline{B}\) is regular the ideal \(\mathfrak m_{\overline{B}}\) is generated by a regular sequence. Applying Lemmas 0666 and 09CC to \(\mathfrak m_A B \subset \mathfrak m_B\) we find \(\mathfrak m_A B / (\mathfrak m_AB \cap \mathfrak m_B^2) = \mathfrak m_A B / \mathfrak m_A \mathfrak m_B\) which is equal to \(\mathfrak m_A/\mathfrak m_A^2 \otimes_k K\) as \(A \to B\) is flat by Lemma 07NP. Thus we obtain a short exact sequence \[0 \to \mathfrak m_A/\mathfrak m_A^2 \otimes_k K \to \mathfrak m_B/\mathfrak m_B^2 \to \mathfrak m_{\overline{B}}/\mathfrak m_{\overline{B}}^2 \to 0\] Functoriality of the Jacobi-Zariski sequences shows that we obtain a commutative diagram \[\xymatrix{ & \mathfrak m_A/\mathfrak m_A^2 \otimes_k K \ar[d] \ar[r] & H_1(\NL_{K/A}) \ar[d] \ar[r] & H_1(\NL_{K/k}) \ar[d] \ar[r] & 0 \\ 0 \ar[r] & \mathfrak m_A/\mathfrak m_A^2 \otimes_k K \ar[r] & \mathfrak m_B/\mathfrak m_B^2 \ar[r] & \mathfrak m_{\overline{B}}/\mathfrak m_{\overline{B}}^2 \ar[r] & 0 }\] The left vertical arrow is injective by Theorem 07EL as \(k \to \overline{B}\) is formally smooth in the \(\mathfrak m_{\overline{B}}\)-adic topology by Lemma 07EG. This finishes the proof by the snake lemma.
Proposition
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Let \(k\) be the residue field of \(A\) and \(\overline{B} = B \otimes_A k\) the special fibre. The following are equivalent
\(A \to B\) is flat and \(\overline{B}\) is geometrically regular over \(k\),
\(A \to B\) is flat and \(k \to \overline{B}\) is formally smooth in the \(\mathfrak m_{\overline{B}}\)-adic topology, and
\(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology.
Proof
The equivalence of (1) and (2) follows from Theorem 07EL.
Assume (3). By Lemma 07NP we see that \(A \to B\) is flat. By Lemma 07EG we see that \(k \to \overline{B}\) is formally smooth in the \(\mathfrak m_{\overline{B}}\)-adic topology. Thus (2) holds.
Assume (2). Lemma 07ED tells us formal smoothness is preserved under completion. The same is true for flatness by Algebra, Lemma 00MC. Hence we may replace \(A\) and \(B\) by their respective completions and assume that \(A\) and \(B\) are Noetherian complete local rings. In this case choose a diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Lemma 07NN. We will use all of the properties of this diagram without further mention. Fix a regular system of parameters \(t_1, \ldots, t_d\) of \(R\) with \(t_1 = p\) in case the characteristic of \(k\) is \(p > 0\). Set \(\overline{S} = S \otimes_R k\). Consider the short exact sequence \[0 \to J \to S \to B \to 0\] As \(\overline{B}\) and \(\overline{S}\) are regular, the kernel of \(\overline{S} \to \overline{B}\) is generated by elements \(\overline{x}_1, \ldots, \overline{x}_r\) which form part of a regular system of parameters of \(\overline{S}\), see Algebra, Lemma 00NR. Lift these elements to \(x_1, \ldots, x_r \in J\). Then \(t_1, \ldots, t_d, x_1, \ldots, x_r\) is part of a regular system of parameters for \(S\). Hence \(S/(x_1, \ldots, x_r)\) is a power series ring over a field (if the characteristic of \(k\) is zero) or a power series ring over a Cohen ring (if the characteristic of \(k\) is \(p > 0\)), see Lemma 07NM. Moreover, it is still the case that \(R \to S/(x_1, \ldots, x_r)\) maps \(t_1, \ldots, t_d\) to a part of a regular system of parameters of \(S/(x_1, \ldots, x_r)\). In other words, we may replace \(S\) by \(S/(x_1, \ldots, x_r)\) and assume we have a diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Lemma 07NN with moreover \(\overline{S} = \overline{B}\). In this case the map \[S \otimes_R A \longrightarrow B\] is an isomorphism as it is surjective, an isomorphism on special fibres, and source and target are flat over \(A\) (for example use Algebra, Lemma 00ME or use that tensoring the short exact sequence \(0 \to I \to S \otimes_R A \to B \to 0\) over \(A\) with \(k\) we find \(I \otimes_A k = 0\) hence \(I = 0\) by Nakayama). Thus by Lemma 07EG it suffices to show that \(R \to S\) is formally smooth in the \(\mathfrak m_S\)-adic topology. Of course, since \(\overline{S} = \overline{B}\), we have that \(\overline{S}\) is formally smooth over \(k = R/\mathfrak m_R\).
Choose elements \(y_1, \ldots, y_m \in S\) such that \(t_1, \ldots, t_d, y_1, \ldots, y_m\) is a regular system of parameters for \(S\). If the characteristic of \(k\) is zero, choose a coefficient field \(K \subset S\) and if the characteristic of \(k\) is \(p > 0\) choose a Cohen ring \(\Lambda \subset S\) with residue field \(K\). At this point the map \(K[[t_1, \ldots, t_d, y_1, \ldots, y_m]] \to S\) (characteristic zero case) or \(\Lambda[[t_2, \ldots, t_d, y_1, \ldots, y_m]] \to S\) (characteristic \(p > 0\) case) is an isomorphism, see Lemma 07NM. From now on we think of \(S\) as the above power series ring.
The rest of the proof is analogous to the argument in the proof of Theorem 07EL. Choose a solid diagram \[\xymatrix{ S \ar[r]_{\bar\psi} \ar@{-->}[rd] & N/J \\ R \ar[u]^i \ar[r]^\varphi & N \ar[u]_\pi }\] as in Definition 07EB. As \(J^2 = 0\) we see that \(J\) has a canonical \(N/J\) module structure and via \(\bar\psi\) a \(S\)-module structure. As \(\bar\psi\) is continuous for the \(\mathfrak m_S\)-adic topology we see that \(\mathfrak m_S^nJ = 0\) for some \(n\). Hence we can filter \(J\) by \(N/J\)-submodules \(0 \subset J_1 \subset J_2 \subset \ldots \subset J_n = J\) such that each quotient \(J_{t + 1}/J_t\) is annihilated by \(\mathfrak m_S\). Considering the sequence of ring maps \(N \to N/J_1 \to N/J_2 \to \ldots \to N/J\) we see that it suffices to prove the existence of the dotted arrow when \(J\) is annihilated by \(\mathfrak m_S\), i.e., when \(J\) is a \(K\)-vector space.
Assume given a diagram as above such that \(J\) is annihilated by \(\mathfrak m_S\). As \(\mathbf{Q} \to S\) (characteristic zero case) or \(\mathbf{Z} \to S\) (characteristic \(p > 0\) case) is formally smooth in the \(\mathfrak m_S\)-adic topology (see Lemma 07NL), we can find a ring map \(\psi : S \to N\) such that \(\pi \circ \psi = \bar \psi\). Since \(S\) is a power series ring in \(t_1, \ldots, t_d\) (characteristic zero) or \(t_2, \ldots, t_d\) (characteristic \(p > 0\)) over a subring, it follows from the universal property of power series rings that we can change our choice of \(\psi\) so that \(\psi(t_i)\) equals \(\varphi(t_i)\) (automatic for \(t_1 = p\) in the characteristic \(p\) case). Then \(\psi \circ i\) and \(\varphi : R \to N\) are two maps whose compositions with \(\pi\) are equal and which agree on \(t_1, \ldots, t_d\). Hence \(D = \psi \circ i - \varphi : R \to J\) is a derivation which annihilates \(t_1, \ldots, t_d\). By Algebra, Lemma 00RO we can write \(D = \xi \circ \text{d}\) for some \(R\)-linear map \(\xi : \Omega_{R/\mathbf{Z}} \to J\) which annihilates \(\text{d}t_1, \ldots, \text{d}t_d\) (by construction) and \(\mathfrak m_R \Omega_{R/\mathbf{Z}}\) (as \(J\) is annihilated by \(\mathfrak m_R\)). Hence \(\xi\) factors as a composition \[\Omega_{R/\mathbf{Z}} \to \Omega_{k/\mathbf{Z}} \xrightarrow{\xi'} J\] where \(\xi'\) is \(k\)-linear. Using the \(K\)-vector space structure on \(J\) we extend \(\xi'\) to a \(K\)-linear map \[\xi'' : \Omega_{k/\mathbf{Z}} \otimes_k K \longrightarrow J.\] Using that \(\overline{S}/k\) is formally smooth we see that \[\Omega_{k/\mathbf{Z}} \otimes_k K \to \Omega_{\overline{S}/\mathbf{Z}} \otimes_S K\] is injective by Theorem 07EL (this is true also in the characteristic zero case as it is even true that \(\Omega_{k/\mathbf{Z}} \to \Omega_{K/\mathbf{Z}}\) is injective in characteristic zero, see Algebra, Proposition 0322). Hence we can find a \(K\)-linear map \(\xi''' : \Omega_{\overline{S}/\mathbf{Z}} \otimes_S K \to J\) whose restriction to \(\Omega_{k/\mathbf{Z}} \otimes_k K\) is \(\xi''\). Write \[D' : S \xrightarrow{\text{d}} \Omega_{S/\mathbf{Z}} \to \Omega_{\overline{S}/\mathbf{Z}} \to \Omega_{\overline{S}/\mathbf{Z}} \otimes_S K \xrightarrow{\xi'''} J.\] Finally, set \(\psi' = \psi - D' : S \to N\). The reader verifies that \(\psi'\) is a ring map such that \(\pi \circ \psi' = \bar \psi\) and such that \(\psi' \circ i = \varphi\) as desired.
As an application of the result above we prove that deformations of formally smooth algebras are unobstructed.
Lemma
Let \(A\) be a Noetherian complete local ring with residue field \(k\). Let \(B\) be a Noetherian complete local \(k\)-algebra. Assume \(k \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology. Then there exists a Noetherian complete local ring \(C\) and a local homomorphism \(A \to C\) which is formally smooth in the \(\mathfrak m_C\)-adic topology such that \(C \otimes_A k \cong B\).
Proof
Choose a diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Lemma 07NN. Let \(t_1, \ldots, t_d\) be a regular system of parameters for \(R\) with \(t_1 = p\) in case the characteristic of \(k\) is \(p > 0\). As \(B\) and \(\overline{S} = S \otimes_R k\) are regular we see that \(\Ker(\overline{S} \to B)\) is generated by elements \(\overline{x}_1, \ldots, \overline{x}_r\) which form part of a regular system of parameters of \(\overline{S}\), see Algebra, Lemma 00NR. Lift these elements to \(x_1, \ldots, x_r \in S\). Then \(t_1, \ldots, t_d, x_1, \ldots, x_r\) is part of a regular system of parameters for \(S\). Hence \(S/(x_1, \ldots, x_r)\) is a power series ring over a field (if the characteristic of \(k\) is zero) or a power series ring over a Cohen ring (if the characteristic of \(k\) is \(p > 0\)), see Lemma 07NM. Moreover, it is still the case that \(R \to S/(x_1, \ldots, x_r)\) maps \(t_1, \ldots, t_d\) to a part of a regular system of parameters of \(S/(x_1, \ldots, x_r)\). In other words, we may replace \(S\) by \(S/(x_1, \ldots, x_r)\) and assume we have a diagram \[\xymatrix{ S \ar[r] & B \\ R \ar[u] \ar[r] & A \ar[u] }\] as in Lemma 07NN with moreover \(\overline{S} = B\). In this case \(R \to S\) is formally smooth in the \(\mathfrak m_S\)-adic topology by Proposition 07NQ. Hence the base change \(C = S \otimes_R A\) is formally smooth over \(A\) in the \(\mathfrak m_C\)-adic topology by Lemma 07EG.
Remark
The assertion of Lemma 07NR is quite strong. Namely, suppose that we have a diagram \[\xymatrix{ & B \\ A \ar[r] & A' \ar[u] }\] of local homomorphisms of Noetherian complete local rings where \(A \to A'\) induces an isomorphism of residue fields \(k = A/\mathfrak m_A = A'/\mathfrak m_{A'}\) and with \(B \otimes_{A'} k\) formally smooth over \(k\). Then we can extend this to a commutative diagram \[\xymatrix{ C \ar[r] & B \\ A \ar[r] \ar[u] & A' \ar[u] }\] of local homomorphisms of Noetherian complete local rings where \(A \to C\) is formally smooth in the \(\mathfrak m_C\)-adic topology and where \(C \otimes_A k \cong B \otimes_{A'} k\). Namely, pick \(A \to C\) as in Lemma 07NR lifting \(B \otimes_{A'} k\) over \(k\). By formal smoothness we can find the arrow \(C \to B\), see Lemma 07NJ. Denote \(C \otimes_A^\wedge A'\) the completion of \(C \otimes_A A'\) with respect to the ideal \(C \otimes_A \mathfrak m_{A'}\). Note that \(C \otimes_A^\wedge A'\) is a Noetherian complete local ring (see Algebra, Lemma 05GH) which is flat over \(A'\) (see Algebra, Lemma 0523). We have moreover
\(C \otimes_A^\wedge A' \to B\) is surjective,
if \(A \to A'\) is surjective, then \(C \to B\) is surjective,
if \(A \to A'\) is finite, then \(C \to B\) is finite, and
if \(A' \to B\) is flat, then \(C \otimes_A^\wedge A' \cong B\).
Namely, by Nakayama’s lemma for nilpotent ideals (see Algebra, Lemma 00DV) we see that \(C \otimes_A k \cong B \otimes_{A'} k\) implies that \(C \otimes_A A'/\mathfrak m_{A'}^n \to B/\mathfrak m_{A'}^nB\) is surjective for all \(n\). This proves (1). Parts (2) and (3) follow from part (1). Part (4) follows from Algebra, Lemma 00ME.
Regular ring maps
Let \(k\) be a field. Recall that a Noetherian \(k\)-algebra \(A\) is said to be geometrically regular over \(k\) if and only if \(A \otimes_k k'\) is regular for all finite purely inseparable extensions \(k'\) of \(k\), see Algebra, Definition 0382. Moreover, if this is the case then \(A \otimes_k k'\) is regular for every finitely generated field extension \(k'/k\), see Algebra, Lemma 0381. We use this notion in the following definition.
Definition
A ring map \(R \to \Lambda\) is regular if it is flat and for every prime \(\mathfrak p \subset R\) the fibre ring \[\Lambda \otimes_R \kappa(\mathfrak p) = \Lambda_\mathfrak p/\mathfrak p\Lambda_\mathfrak p\] is Noetherian and geometrically regular over \(\kappa(\mathfrak p)\).
If \(R \to \Lambda\) is a ring map with \(\Lambda\) Noetherian, then the fibre rings are always Noetherian.
Lemma
Let \(R \to \Lambda\) be a ring map with \(\Lambda\) Noetherian. The following are equivalent
\(R \to \Lambda\) is regular,
\(R_\mathfrak p \to \Lambda_\mathfrak q\) is regular for all \(\mathfrak q \subset \Lambda\) lying over \(\mathfrak p \subset R\), and
\(R_\mathfrak m \to \Lambda_{\mathfrak m'}\) is regular for all maximal ideals \(\mathfrak m' \subset \Lambda\) lying over \(\mathfrak m\) in \(R\).
Proof
This is true because a Noetherian ring is regular if and only if all the local rings are regular local rings, see Algebra, Definition 00OD and a ring map is flat if and only if all the induced maps of local rings are flat, see Algebra, Lemma 00HT.
Lemma
Let \(R \to \Lambda\) be a regular ring map. For any finite type ring map \(R \to R'\) the base change \(R' \to \Lambda \otimes_R R'\) is regular too.
Proof
Flatness is preserved under any base change, see Algebra, Lemma 00HI. Consider a prime \(\mathfrak p' \subset R'\) lying over \(\mathfrak p \subset R\). The residue field extension \(\kappa(\mathfrak p')/\kappa(\mathfrak p)\) is finitely generated as \(R'\) is of finite type over \(R\). Hence the fibre ring \[(\Lambda \otimes_R R') \otimes_{R'} \kappa(\mathfrak p') = \Lambda \otimes_R \kappa(\mathfrak p) \otimes_{\kappa(\mathfrak p)} \kappa(\mathfrak p')\] is Noetherian by Algebra, Lemma 045I and the assumption on the fibre rings of \(R \to \Lambda\). Geometric regularity of the fibres is preserved by Algebra, Lemma 0381.
Lemma
Let \(A \to B\) and \(B \to C\) be regular ring maps. If the fibre rings of \(A \to C\) are Noetherian, then \(A \to C\) is regular.
Proof
Let \(\mathfrak p \subset A\) be a prime. Let \(\kappa(\mathfrak p) \subset k\) be a finite purely inseparable extension. We have to show that \(C \otimes_A k\) is regular. By Lemma 07C1 we may assume that \(A = k\) and we reduce to proving that \(C\) is regular. The assumption is that \(B\) is regular and that \(B \to C\) is flat with regular fibres. Then \(C\) is regular by Algebra, Lemma 031E. Some details omitted.
Lemma
Let \(R\) be a ring. Let \((A_i, \varphi_{ii'})\) be a directed system of smooth \(R\)-algebras. Set \(\Lambda = \colim A_i\). If the fibre rings \(\Lambda \otimes_R \kappa(\mathfrak p)\) are Noetherian for all \(\mathfrak p \subset R\), then \(R \to \Lambda\) is regular.
Proof
Note that \(\Lambda\) is flat over \(R\) by Algebra, Lemmas 05UT and 00TA. Let \(\kappa(\mathfrak p) \subset k\) be a finite purely inseparable extension. Note that \[\Lambda \otimes_R \kappa(\mathfrak p) \otimes_{\kappa(\mathfrak p)} k = \Lambda \otimes_R k = \colim A_i \otimes_R k\] is a colimit of smooth \(k\)-algebras, see Algebra, Lemma 00T4. Since each local ring of a smooth \(k\)-algebra is regular by Algebra, Lemma 00TT we conclude that all local rings of \(\Lambda \otimes_R k\) are regular by Algebra, Lemma 07DX. This proves the lemma.
Let’s see when a field extension defines a regular ring map.
Lemma
Let \(K/k\) be a field extension. Then \(k \to K\) is a regular ring map if and only if \(K\) is a separable field extension of \(k\).
Proof
If \(k \to K\) is regular, then \(K\) is geometrically reduced over \(k\), hence \(K\) is separable over \(k\) by Algebra, Proposition 0322. Conversely, if \(K/k\) is separable, then \(K\) is a colimit of smooth \(k\)-algebras, see Algebra, Lemma 07BV hence is regular by Lemma 07EP.
Lemma
Let \(A \to B \to C\) be ring maps. If \(A \to C\) is regular and \(B \to C\) is flat and surjective on spectra, then \(A \to B\) is regular.
Proof
By Algebra, Lemma 039V we see that \(A \to B\) is flat. Let \(\mathfrak p \subset A\) be a prime. The ring map \(B \otimes_A \kappa(\mathfrak p) \to C \otimes_A \kappa(\mathfrak p)\) is flat and surjective on spectra. Hence \(B \otimes_A \kappa(\mathfrak p)\) is geometrically regular by Algebra, Lemma 07NH.
Ascending properties along regular ring maps
This section is the analogue of Algebra, Section 0336 but where the ring map \(R \to S\) is regular.
Lemma
Let \(\varphi : R \to S\) be a ring map. Assume
\(\varphi\) is regular,
\(S\) is Noetherian, and
\(R\) is Noetherian and reduced.
Then \(S\) is reduced.
Proof
For Noetherian rings being reduced is the same as having properties \((S_1)\) and \((R_0)\), see Algebra, Lemma 031R. Hence we may apply Algebra, Lemmas 0339 and 033A.
Lemma
Let \(\varphi : R \to S\) be a ring map. Assume
\(\varphi\) is regular,
\(S\) is Noetherian, and
\(R\) is Noetherian and normal.
Then \(S\) is normal.
Proof
For Noetherian rings being normal is the same as having properties \((S_2)\) and \((R_1)\), see Algebra, Lemma 031S. Hence we may apply Algebra, Lemmas 0339 and 033A.
Lemma
Let \(\varphi : R \to S\) be a ring map. Assume
\(\varphi\) is regular,
\(S\) is Noetherian, and
\(R\) is Noetherian and regular.
Then \(S\) is regular.
Proof
For Noetherian rings being regular is the same as having properties \((R_k)\) for all \(k\). Hence we may apply Algebra, Lemma 033A.
Lemma
Let \(\varphi : R \to S\) be a ring map. Assume
\(\varphi\) is regular,
\(S\) is Noetherian, and
\(R\) is Noetherian and Cohen-Macaulay.
Then \(S\) is Cohen-Macaulay.
Proof
For Noetherian rings being Cohen-Macaulay is the same as having properties \((S_k)\) for all \(k\). Hence we may apply Algebra, Lemma 0339.
Permanence of properties under completion
Given a Noetherian local ring \((A, \mathfrak m)\) we denote \(A^\wedge\) the completion of \(A\) with respect to \(\mathfrak m\). We will use without further mention that \(A^\wedge\) is a Noetherian complete local ring with maximal ideal \(\mathfrak m^\wedge = \mathfrak m A^\wedge\) and that \(A \to A^\wedge\) is faithfully flat. See Algebra, Lemmas 0316, 031C, and 00MC.
Lemma
Let \(A\) be a Noetherian local ring. Then \(\dim(A) = \dim(A^\wedge)\).
Proof
By Algebra, Lemma 031C the map \(A \to A^\wedge\) induces isomorphisms \(A/\mathfrak m^n = A^\wedge/(\mathfrak m^\wedge)^n\) for \(n \geq 1\). By Algebra, Lemma 02M0 this implies that \[\text{length}_A(A/\mathfrak m^n) = \text{length}_{A^\wedge}(A^\wedge/(\mathfrak m^\wedge)^n)\] for all \(n \geq 1\). Thus \(d(A) = d(A^\wedge)\) and we conclude by Algebra, Proposition 00KQ. An alternative proof is to use Algebra, Lemma 00ON.
Lemma
Let \(A\) be a Noetherian local ring. Then \(\text{depth}(A) = \text{depth}(A^\wedge)\).
Proof
See Algebra, Lemma 0337.
Lemma
Let \(A\) be a Noetherian local ring. Then \(A\) is Cohen-Macaulay if and only if \(A^\wedge\) is so.
Proof
A local ring \(A\) is Cohen-Macaulay if and only if \(\dim(A) = \text{depth}(A)\). As both of these invariants are preserved under completion (Lemmas 07NV and 07NW) the claim follows.
Lemma
Let \(A\) be a Noetherian local ring. Then \(A\) is regular if and only if \(A^\wedge\) is so.
Proof
If \(A^\wedge\) is regular, then \(A\) is regular by Algebra, Lemma 00OF. Assume \(A\) is regular. Let \(\mathfrak m\) be the maximal ideal of \(A\). Then \(\dim_{\kappa(\mathfrak m)} \mathfrak m/\mathfrak m^2 = \dim(A) = \dim(A^\wedge)\) (Lemma 07NV). On the other hand, \(\mathfrak mA^\wedge\) is the maximal ideal of \(A^\wedge\) and hence \(\mathfrak m_{A^\wedge}\) is generated by at most \(\dim(A^\wedge)\) elements. Thus \(A^\wedge\) is regular. (You can also use Algebra, Lemma 031E.)
Lemma
Let \(A\) be a Noetherian local ring. Then \(A\) is a discrete valuation ring if and only if \(A^\wedge\) is so.
Proof
This follows from Lemmas 07NV and 07NY and Algebra, Lemma 00PD.
Lemma
Let \(A\) be a Noetherian local ring.
If \(A^\wedge\) is reduced, then so is \(A\).
In general \(A\) reduced does not imply \(A^\wedge\) is reduced.
If \(A\) is Nagata, then \(A\) is reduced if and only if \(A^\wedge\) is reduced.
Proof
As \(A \to A^\wedge\) is faithfully flat we have (1) by Algebra, Lemma 033F. For (2) see Algebra, Example 00PB (there are also examples in characteristic zero, see Algebra, Remark 00PC). For (3) see Algebra, Lemmas 0331 and 032Y.
Lemma
Let \(A\) be a Noetherian local ring. If \(A^\wedge\) is normal, then so is \(A\).
Proof
As \(A \to A^\wedge\) is faithfully flat this follows from Algebra, Lemma 033G.
Lemma
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Then the induced map of completions \(A^\wedge \to B^\wedge\) is flat if and only if \(A \to B\) is flat.
Proof
Consider the commutative diagram \[\xymatrix{ A^\wedge \ar[r] & B^\wedge \\ A \ar[r] \ar[u] & B \ar[u] }\] The vertical arrows are faithfully flat. Assume that \(A^\wedge \to B^\wedge\) is flat. Then \(A \to B^\wedge\) is flat. Hence \(B\) is flat over \(A\) by Algebra, Lemma 0584.
Assume that \(A \to B\) is flat. Then \(A \to B^\wedge\) is flat. Hence \(B^\wedge/\mathfrak m_A^n B^\wedge\) is flat over \(A/\mathfrak m_A^n\) for all \(n \geq 1\). Note that \(\mathfrak m_A^n A^\wedge\) is the \(n\)th power of the maximal ideal \(\mathfrak m_A^\wedge\) of \(A^\wedge\) and \(A/\mathfrak m_A^n = A^\wedge/(\mathfrak m_A^\wedge)^n\). Thus we see that \(B^\wedge\) is flat over \(A^\wedge\) by applying Algebra, Lemma 0523 (with \(R = A^\wedge\), \(I = \mathfrak m_A^\wedge\), \(S = B^\wedge\), \(M = S\)).
Lemma
Let \(A \to B\) be a flat local homomorphism of Noetherian local rings such that \(\mathfrak m_A B = \mathfrak m_B\) and \(\kappa(\mathfrak m_A) = \kappa(\mathfrak m_B)\). Then \(A \to B\) induces an isomorphism \(A^\wedge \to B^\wedge\) of completions.
Proof
By Algebra, Lemma 0394 we see that \(B^\wedge\) is the \(\mathfrak m_A\)-adic completion of \(B\) and that \(A^\wedge \to B^\wedge\) is finite. Since \(A \to B\) is flat we have \(\text{Tor}_1^A(B, \kappa(\mathfrak m_A)) = 0\). Hence we see that \(B^\wedge\) is flat over \(A^\wedge\) by Lemma 0AGW. Thus \(B^\wedge\) is a free \(A^\wedge\)-module by Algebra, Lemma 00NZ. Since \(A^\wedge \to B^\wedge\) induces an isomorphism \(\kappa(\mathfrak m_A) = A^\wedge/\mathfrak m_A A^\wedge \to B^\wedge/\mathfrak m_A B^\wedge = B^\wedge/\mathfrak m_B B^\wedge = \kappa(\mathfrak m_B)\) by our assumptions (and Algebra, Lemma 05GG), we see that \(B^\wedge\) is free of rank \(1\). Thus \(A^\wedge \to B^\wedge\) is an isomorphism.
Permanence of properties under étale maps
In this section we consider an étale ring map \(\varphi : A \to B\) and we study which properties of \(A\) are inherited by \(B\) and which properties of the local ring of \(B\) at \(\mathfrak q\) are inherited by the local ring of \(A\) at \(\mathfrak p = \varphi^{-1}(\mathfrak q)\). Basically, this section reviews and collects earlier results and does not add any new material.
We will use without further mention that an étale ring map is flat (Algebra, Lemma 00U2) and that a flat local homomorphism of local rings is faithfully flat (Algebra, Lemma 00HR).
Lemma
If \(A \to B\) is an étale ring map and \(\mathfrak q\) is a prime of \(B\) lying over \(\mathfrak p \subset A\), then \(A_{\mathfrak p}\) is Noetherian if and only if \(B_{\mathfrak q}\) is Noetherian.
Proof
Since \(A_\mathfrak p \to B_\mathfrak q\) is faithfully flat we see that \(B_\mathfrak q\) Noetherian implies that \(A_\mathfrak p\) is Noetherian, see Algebra, Lemma 033E. Conversely, if \(A_\mathfrak p\) is Noetherian, then \(B_\mathfrak q\) is Noetherian as it is a localization of a finite type \(A_\mathfrak p\)-algebra.
Lemma
If \(A \to B\) is an étale ring map and \(\mathfrak q\) is a prime of \(B\) lying over \(\mathfrak p \subset A\), then \(\dim(A_{\mathfrak p}) = \dim(B_{\mathfrak q})\).
Proof
Namely, because \(A_{\mathfrak p} \to B_{\mathfrak q}\) is flat we have going down, and hence the inequality \(\dim(A_{\mathfrak p}) \leq \dim(B_{\mathfrak q})\), see Algebra, Lemma 00OH. On the other hand, suppose that \(\mathfrak q_0 \subset \mathfrak q_1 \subset \ldots \subset \mathfrak q_n\) is a chain of primes in \(B_{\mathfrak q}\). Then the corresponding sequence of primes \(\mathfrak p_0 \subset \mathfrak p_1 \subset \ldots \subset \mathfrak p_n\) (with \(\mathfrak p_i = \mathfrak q_i \cap A_{\mathfrak p}\)) is chain also (i.e., no equalities in the sequence) as an étale ring map is quasi-finite (see Algebra, Lemma 00U5) and a quasi-finite ring map induces a map of spectra with discrete fibres (by definition). This means that \(\dim(A_{\mathfrak p}) \geq \dim(B_{\mathfrak q})\) as desired.
Lemma
If \(A \to B\) is an étale ring map and \(\mathfrak q\) is a prime of \(B\) lying over \(\mathfrak p \subset A\), then \(A_{\mathfrak p}\) is regular if and only if \(B_{\mathfrak q}\) is regular.
Proof
By Lemma 0AGZ we may assume both \(A_\mathfrak p\) and \(B_\mathfrak q\) are Noetherian in order to prove the equivalence. Let \(x_1, \ldots, x_t \in \mathfrak pA_\mathfrak p\) be a minimal set of generators. As \(A_\mathfrak p \to B_\mathfrak q\) is faithfully flat we see that the images \(y_1, \ldots, y_t\) in \(B_\mathfrak q\) form a minimal system of generators for \(\mathfrak pB_\mathfrak q = \mathfrak q B_\mathfrak q\) (Algebra, Lemma 00U4). Regularity of \(A_\mathfrak p\) by definition means \(t = \dim(A_\mathfrak p)\) and similarly for \(B_\mathfrak q\). Hence the lemma follows from the equality \(\dim(A_\mathfrak p) = \dim(B_\mathfrak q)\) of Lemma 07QP.
Lemma
If \(A \to B\) is an étale ring map and \(A\) is a Dedekind domain, then \(B\) is a finite product of Dedekind domains. In particular, the localizations \(B_\mathfrak q\) for \(\mathfrak q \subset B\) maximal are discrete valuation rings.
Proof
The statement on the local rings follows from Lemmas 07QP and 0AH0 and Algebra, Lemma 00PD. It follows that \(B\) is a Noetherian normal ring of dimension \(1\). By Algebra, Lemma 030C we conclude that \(B\) is a finite product of normal domains of dimension \(1\). These are Dedekind domains by Algebra, Lemma 034X.
Permanence of properties under henselization
Given a local ring \(R\) we denote \(R^h\), resp. \(R^{sh}\) the henselization, resp. strict henselization of \(R\), see Algebra, Definition 04GQ. Many of the properties of \(R\) are reflected in \(R^h\) and \(R^{sh}\) as we will show in this section.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a local ring. Then we have the following
\(R \to R^h \to R^{sh}\) are faithfully flat ring maps,
\(\mathfrak m R^h = \mathfrak m^h\) and \(\mathfrak m R^{sh} = \mathfrak m^h R^{sh} = \mathfrak m^{sh}\),
\(R/\mathfrak m^n = R^h/\mathfrak m^nR^h\) for all \(n\),
there exist elements \(x_i \in R^{sh}\) such that \(R^{sh}/\mathfrak m^nR^{sh}\) is a free \(R/\mathfrak m^n\)-module on \(x_i \bmod \mathfrak m^nR^{sh}\).
Proof
By construction \(R^h\) is a colimit of étale \(R\)-algebras, see Algebra, Lemma 04GN. Since étale ring maps are flat (Algebra, Lemma 00U2) we see that \(R^h\) is flat over \(R\) by Algebra, Lemma 05UT. As a flat local ring homomorphism is faithfully flat (Algebra, Lemma 00HR) we see that \(R \to R^h\) is faithfully flat. The ring map \(R^h \to R^{sh}\) is a colimit of finite étale ring maps, see proof of Algebra, Lemma 04GP. Hence the same arguments as above show that \(R^h \to R^{sh}\) is faithfully flat.
Part (2) follows from Algebra, Lemmas 04GN and 04GP. Part (3) follows from Algebra, Lemma 051F because \(R/\mathfrak m \to R^h/\mathfrak mR^h\) is an isomorphism and \(R/\mathfrak m^n \to R^h/\mathfrak m^nR^h\) is flat as a base change of the flat ring map \(R \to R^h\) (Algebra, Lemma 00HI). Let \(\kappa^{sep}\) be the residue field of \(R^{sh}\) (it is a separable algebraic closure of \(\kappa\)). Choose \(x_i \in R^{sh}\) mapping to a basis of \(\kappa^{sep}\) as a \(\kappa\)-vector space. Then (4) follows from Algebra, Lemma 051F in exactly the same way as above.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a local ring. Then
\(R \to R^h\), \(R^h \to R^{sh}\), and \(R \to R^{sh}\) are formally étale,
\(R \to R^h\), \(R^h \to R^{sh}\), resp. \(R \to R^{sh}\) are formally smooth in the \(\mathfrak m^h\), \(\mathfrak m^{sh}\), resp. \(\mathfrak m^{sh}\)-topology.
Proof
Part (1) follows from the fact that \(R^h\) and \(R^{sh}\) are directed colimits of étale algebras (by construction), that étale algebras are formally étale (Algebra, Lemma 00UR), and that colimits of formally étale algebras are formally étale (Algebra, Lemma 031N). Part (2) follows from the fact that a formally étale ring map is formally smooth and Lemma 07EC.
Lemma
Let \(R\) be a local ring. The following are equivalent
\(R\) is Noetherian,
\(R^h\) is Noetherian, and
\(R^{sh}\) is Noetherian.
In this case we have
\((R^h)^\wedge\) and \((R^{sh})^\wedge\) are Noetherian complete local rings,
\(R^\wedge \to (R^h)^\wedge\) is an isomorphism,
\(R^h \to (R^h)^\wedge\) and \(R^{sh} \to (R^{sh})^\wedge\) are flat,
\(R^\wedge \to (R^{sh})^\wedge\) is formally smooth in the \(\mathfrak m_{(R^{sh})^\wedge}\)-adic topology,
\((R^\wedge)^{sh} = R^\wedge \otimes_{R^h} R^{sh}\), and
\(((R^\wedge)^{sh})^\wedge = (R^{sh})^\wedge\).
Proof
Since \(R \to R^h \to R^{sh}\) are faithfully flat (Lemma 07QM), we see that \(R^h\) or \(R^{sh}\) being Noetherian implies that \(R\) is Noetherian, see Algebra, Lemma 033E. In the rest of the proof we assume \(R\) is Noetherian.
As \(\mathfrak m \subset R\) is finitely generated it follows that \(\mathfrak m^h = \mathfrak m R^h\) and \(\mathfrak m^{sh} = \mathfrak mR^{sh}\) are finitely generated, see Lemma 07QM. Hence \((R^h)^\wedge\) and \((R^{sh})^\wedge\) are Noetherian by Algebra, Lemma 032B. This proves (a).
Note that (b) is immediate from Lemma 07QM. In particular we see that \((R^h)^\wedge\) is flat over \(R\), see Algebra, Lemma 00MC.
Next, we show that \(R^h \to (R^h)^\wedge\) is flat. Write \(R^h = \colim_i R_i\) as a directed colimit of localizations of étale \(R\)-algebras. By Algebra, Lemma 05UU if \((R^h)^\wedge\) is flat over each \(R_i\), then \(R^h \to (R^h)^\wedge\) is flat. Note that \(R^h = R_i^h\) (by construction). Hence \(R_i^\wedge = (R^h)^\wedge\) by part (b) is flat over \(R_i\) as desired. To finish the proof of (c) we show that \(R^{sh} \to (R^{sh})^\wedge\) is flat. To do this, by a limit argument as above, it suffices to show that \((R^{sh})^\wedge\) is flat over \(R\). Note that it follows from Lemma 07QM that \((R^{sh})^\wedge\) is the completion of a free \(R\)-module. By Lemma 06LE we see this is flat over \(R\) as desired. This finishes the proof of (c).
At this point we know (c) is true and that \((R^h)^\wedge\) and \((R^{sh})^\wedge\) are Noetherian. It follows from Algebra, Lemma 033E that \(R^h\) and \(R^{sh}\) are Noetherian.
Part (d) follows from Lemma 07QN and Lemma 07ED.
Part (e) follows from Algebra, Lemma 0C2Z and the fact that \(R^\wedge\) is henselian by Algebra, Lemma 04GM.
Proof of (f). Using (e) there is a map \(R^{sh} \to (R^\wedge)^{sh}\) which induces a map \((R^{sh})^\wedge \to ((R^\wedge)^{sh})^\wedge\) upon completion. Using (e) there is a map \(R^\wedge \to (R^{sh})^\wedge\). Since \((R^{sh})^\wedge\) is strictly henselian (see above) this map induces a map \((R^\wedge)^{sh} \to (R^{sh})^\wedge\) by Algebra, Lemma 04GU. Completing we obtain a map \(((R^\wedge)^{sh})^\wedge \to (R^{sh})^\wedge\). We omit the verification that these two maps are mutually inverse.
Lemma
Let \(R\) be a local ring. The following are equivalent: \(R\) is reduced, the henselization \(R^h\) of \(R\) is reduced, and the strict henselization \(R^{sh}\) of \(R\) is reduced.
Proof
The ring maps \(R \to R^h \to R^{sh}\) are faithfully flat. Hence one direction of the implications follows from Algebra, Lemma 033F. Conversely, assume \(R\) is reduced. Since \(R^h\) and \(R^{sh}\) are filtered colimits of étale, hence smooth \(R\)-algebras, the result follows from Algebra, Lemma 033B.
Lemma
Let \(R\) be a local ring. Let \(nil(R)\) denote the ideal of nilpotent elements of \(R\). Then \(nil(R)R^h = nil(R^h)\) and \(nil(R)R^{sh} = nil(R^{sh})\).
Proof
Note that \(nil(R)\) is the biggest ideal consisting of nilpotent elements such that the quotient \(R/nil(R)\) is reduced. Note that \(nil(R)R^h\) consists of nilpotent elements by Algebra, Lemma 0544. Also, note that \(R^h/nil(R) R^h\) is the henselization of \(R/nil(R)\) by Algebra, Lemma 05WQ. Hence \(R^h/nil(R)R^h\) is reduced by Lemma 06DH. We conclude that \(nil(R) R^h = nil(R^h)\) as desired. Similarly for the strict henselization but using Algebra, Lemma 05WS.
Lemma
Let \(R\) be a local ring. The following are equivalent: \(R\) is a normal domain, the henselization \(R^h\) of \(R\) is a normal domain, and the strict henselization \(R^{sh}\) of \(R\) is a normal domain.
Proof
A preliminary remark is that a local ring is normal if and only if it is a normal domain (see Algebra, Definition 00GV). The ring maps \(R \to R^h \to R^{sh}\) are faithfully flat. Hence one direction of the implications follows from Algebra, Lemma 033G. Conversely, assume \(R\) is normal. Since \(R^h\) and \(R^{sh}\) are filtered colimits of étale hence smooth \(R\)-algebras, the result follows from Algebra, Lemmas 033C and 037D.
Lemma
Given any local ring \(R\) we have \(\dim(R) = \dim(R^h) = \dim(R^{sh})\).
Proof
Since \(R \to R^{sh}\) is faithfully flat (Lemma 07QM) we see that \(\dim(R^{sh}) \geq \dim(R)\) by going down, see Algebra, Lemma 00OH. For the converse, we write \(R^{sh} = \colim R_i\) as a directed colimit of local rings \(R_i\) each of which is a localization of an étale \(R\)-algebra. Now if \(\mathfrak q_0 \subset \mathfrak q_1 \subset \ldots \subset \mathfrak q_n\) is a chain of prime ideals in \(R^{sh}\), then for some sufficiently large \(i\) the sequence \[R_i \cap \mathfrak q_0 \subset R_i \cap \mathfrak q_1 \subset \ldots \subset R_i \cap \mathfrak q_n\] is a chain of primes in \(R_i\). Thus we see that \(\dim(R^{sh}) \leq \sup_i \dim(R_i)\). But by the result of Lemma 07QP we have \(\dim(R_i) = \dim(R)\) for each \(i\) and we win.
Lemma
Given a Noetherian local ring \(R\) we have \(\text{depth}(R) = \text{depth}(R^h) = \text{depth}(R^{sh})\).
Proof
By Lemma 06LJ we know that \(R^h\) and \(R^{sh}\) are Noetherian. Hence the lemma follows from Algebra, Lemma 0337.
Lemma
Let \(R\) be a Noetherian local ring. The following are equivalent: \(R\) is Cohen-Macaulay, the henselization \(R^h\) of \(R\) is Cohen-Macaulay, and the strict henselization \(R^{sh}\) of \(R\) is Cohen-Macaulay.
Proof
By Lemma 06LJ we know that \(R^h\) and \(R^{sh}\) are Noetherian, hence the lemma makes sense. Since we have \(\text{depth}(R) = \text{depth}(R^h) = \text{depth}(R^{sh})\) and \(\dim(R) = \dim(R^h) = \dim(R^{sh})\) by Lemmas 06LL and 06LK we conclude.
Lemma
Let \(R\) be a Noetherian local ring. The following are equivalent: \(R\) is a regular local ring, the henselization \(R^h\) of \(R\) is a regular local ring, and the strict henselization \(R^{sh}\) of \(R\) is a regular local ring.
Proof
By Lemma 06LJ we know that \(R^h\) and \(R^{sh}\) are Noetherian, hence the lemma makes sense. Let \(\mathfrak m\) be the maximal ideal of \(R\). Let \(x_1, \ldots, x_t \in \mathfrak m\) be a minimal system of generators of \(\mathfrak m\), i.e., such that the images in \(\mathfrak m/\mathfrak m^2\) form a basis over \(\kappa = R/\mathfrak m\). Because \(R \to R^h\) and \(R \to R^{sh}\) are faithfully flat, it follows that the images \(x_1^h, \ldots, x_t^h\) in \(R^h\), resp. \(x_1^{sh}, \ldots, x_t^{sh}\) in \(R^{sh}\) are a minimal system of generators for \(\mathfrak m^h = \mathfrak mR^h\), resp. \(\mathfrak m^{sh} = \mathfrak mR^{sh}\). Regularity of \(R\) by definition means \(t = \dim(R)\) and similarly for \(R^h\) and \(R^{sh}\). Hence the lemma follows from the equality of dimensions \(\dim(R) = \dim(R^h) = \dim(R^{sh})\) of Lemma 06LK
Lemma
Let \(R\) be a Noetherian local ring. Then \(R\) is a discrete valuation ring if and only if \(R^h\) is a discrete valuation ring if and only if \(R^{sh}\) is a discrete valuation ring.
Proof
This follows from Lemmas 06LK and 06LN and Algebra, Lemma 00PD.
Lemma
Let \(A\) be a ring. Let \(B\) be a filtered colimit of étale \(A\)-algebras. Let \(\mathfrak p\) be a prime of \(A\). If \(B\) is Noetherian, then there are finitely many primes \(\mathfrak q_1, \ldots, \mathfrak q_r\) lying over \(\mathfrak p\), we have \(B \otimes_A \kappa(\mathfrak p) = \prod \kappa(\mathfrak q_i)\), and each of the field extensions \(\kappa(\mathfrak q_i)/\kappa(\mathfrak p)\) is separable algebraic.
Proof
Write \(B\) as a filtered colimit \(B = \colim B_i\) with \(A \to B_i\) étale. Then on the one hand \(B \otimes_A \kappa(\mathfrak p) = \colim B_i \otimes_A \kappa(\mathfrak p)\) is a filtered colimit of étale \(\kappa(\mathfrak p)\)-algebras, and on the other hand it is Noetherian. An étale \(\kappa(\mathfrak p)\)-algebra is a finite product of finite separable field extensions (Algebra, Lemma 00U3). Hence there are no nontrivial specializations between the primes (which are all maximal and minimal primes) of the algebras \(B_i \otimes_A \kappa(\mathfrak p)\) and hence there are no nontrivial specializations between the primes of \(B \otimes_A \kappa(\mathfrak p)\). Thus \(B \otimes_A \kappa(\mathfrak p)\) is reduced and has finitely many primes which all minimal. Thus it is a finite product of fields (use Algebra, Lemma 02LX or Algebra, Proposition 00KJ). Each of these fields is a colimit of finite separable extensions and hence the final statement of the lemma follows.
Lemma
Let \(R\) be a Noetherian local ring. Let \(\mathfrak p \subset R\) be a prime. Then \[R^h \otimes_R \kappa(\mathfrak p) = \prod\nolimits_{i = 1, \ldots, t} \kappa(\mathfrak q_i) \quad\text{resp.}\quad R^{sh} \otimes_R \kappa(\mathfrak p) = \prod\nolimits_{i = 1, \ldots, s} \kappa(\mathfrak r_i)\] where \(\mathfrak q_1, \ldots, \mathfrak q_t\), resp. \(\mathfrak r_1, \ldots, \mathfrak r_s\) are the prime of \(R^h\), resp. \(R^{sh}\) lying over \(\mathfrak p\). Moreover, the field extensions \(\kappa(\mathfrak q_i)/\kappa(\mathfrak p)\) resp. \(\kappa(\mathfrak r_i)/\kappa(\mathfrak p)\) are separable algebraic.
Proof
This can be deduced from the more general Lemma 0AH1 using that the henselization and strict henselization are Noetherian (as we’ve seen above). But we also give a direct proof as follows.
We will use without further mention the results of Lemmas 07QM and 06LJ. Note that \(R^h/\mathfrak pR^h\), resp. \(R^{sh}/\mathfrak pR^{sh}\) is the henselization, resp. strict henselization of \(R/\mathfrak p\), see Algebra, Lemma 05WQ resp. Algebra, Lemma 05WS. Hence we may replace \(R\) by \(R/\mathfrak p\) and assume that \(R\) is a Noetherian local domain and that \(\mathfrak p = (0)\). Since \(R^h\), resp. \(R^{sh}\) is Noetherian, it has finitely many minimal primes \(\mathfrak q_1, \ldots, \mathfrak q_t\), resp. \(\mathfrak r_1, \ldots, \mathfrak r_s\). Since \(R \to R^h\), resp. \(R \to R^{sh}\) is flat these are exactly the primes lying over \(\mathfrak p = (0)\) (by going down). Finally, as \(R\) is a domain, we see that \(R^h\), resp. \(R^{sh}\) is reduced, see Lemma 06DH. Thus we see that \(R^h \otimes_R \kappa(\mathfrak p)\) resp. \(R^{sh} \otimes_R \kappa(\mathfrak p)\) is a reduced Noetherian ring with finitely many primes, all of which are minimal (and hence maximal). Thus these rings are Artinian and are products of their localizations at maximal ideals, each necessarily a field (see Algebra, Proposition 00KJ and Algebra, Lemma 00EU).
The final statement follows from the fact that \(R \to R^h\), resp. \(R \to R^{sh}\) is a colimit of étale ring maps and hence the induced residue field extensions are colimits of finite separable extensions, see Algebra, Lemma 00U4.
Field extensions, revisited
In this section we study some peculiarities of field extensions in characteristic \(p > 0\).
Definition
Let \(p\) be a prime number. Let \(k \to K\) be an extension of fields of characteristic \(p\). Denote \(kK^p\) the compositum of \(k\) and \(K^p\) in \(K\).
A subset \(\{x_i\} \subset K\) is called p-independent over \(k\) if the elements \(x^E = \prod x_i^{e_i}\) where \(0 \leq e_i < p\) are linearly independent over \(kK^p\).
A subset \(\{x_i\}\) of \(K\) is called a p-basis of \(K\) over \(k\) if the elements \(x^E\) form a basis of \(K\) over \(kK^p\).
This is related to the notion of a \(p\)-basis of a \(\mathbf{F}_p\)-algebra which we will discuss later.
Lemma
Let \(K/k\) be a field extension. Assume \(k\) has characteristic \(p > 0\). Let \(\{x_i\}\) be a subset of \(K\). The following are equivalent
the elements \(\{x_i\}\) are \(p\)-independent over \(k\), and
the elements \(\text{d}x_i\) are \(K\)-linearly independent in \(\Omega_{K/k}\).
Any \(p\)-independent collection can be extended to a \(p\)-basis of \(K\) over \(k\). In particular, the field \(K\) has a \(p\)-basis over \(k\). Moreover, the following are equivalent:
\(\{x_i\}\) is a \(p\)-basis of \(K\) over \(k\), and
\(\text{d}x_i\) is a basis of the \(K\)-vector space \(\Omega_{K/k}\).
Proof
Assume (2) and suppose that \(\sum a_E x^E = 0\) is a linear relation with \(a_E \in k K^p\). Let \(\theta_i : K \to K\) be a \(k\)-derivation such that \(\theta_i(x_j) = \delta_{ij}\) (Kronecker delta). Note that any \(k\)-derivation of \(K\) annihilates \(kK^p\). Applying \(\theta_i\) to the given relation we obtain new relations \[\sum\nolimits_{E, e_i > 0} e_i a_E x_1^{e_1}\ldots x_i^{e_i - 1} \ldots x_n^{e_n} = 0\] Hence if we pick \(\sum a_E x^E\) as the relation with minimal total degree \(|E| = \sum e_i\) for some \(a_E \not = 0\), then we get a contradiction. Hence (1) holds.
If \(\{x_i\}\) is a \(p\)-basis for \(K\) over \(k\), then \(K \cong kK^p[X_i]/(X_i^p - x_i^p)\). Hence we see that \(\text{d}x_i\) forms a basis for \(\Omega_{K/k}\) over \(K\). Thus (a) implies (b).
Let \(\{x_i\}\) be a \(p\)-independent subset of \(K\) over \(k\). An application of Zorn’s lemma shows that we can enlarge this to a maximal \(p\)-independent subset of \(K\) over \(k\). We claim that any maximal \(p\)-independent subset \(\{x_i\}\) of \(K\) is a \(p\)-basis of \(K\) over \(k\). The claim will imply that (1) implies (2) and establish the existence of \(p\)-bases. To prove the claim let \(L\) be the subfield of \(K\) generated by \(kK^p\) and the \(x_i\). We have to show that \(L = K\). If \(x \in K\) but \(x \not \in L\), then \(x^p \in L\) and \(L(x) \cong L[z]/(z^p - x^p)\). Hence \(\{x_i\} \cup \{x\}\) is \(p\)-independent over \(k\), a contradiction.
Finally, we have to show that (b) implies (a). By the equivalence of (1) and (2) we see that \(\{x_i\}\) is a maximal \(p\)-independent subset of \(K\) over \(k\). Hence by the claim above it is a \(p\)-basis.
Lemma
Let \(K/k\) be a field extension. Let \(\{K_\alpha\}_{\alpha \in A}\) be a collection of subfields of \(K\) with the following properties
\(k \subset K_\alpha\) for all \(\alpha \in A\),
\(k = \bigcap_{\alpha \in A} K_\alpha\),
for \(\alpha, \alpha' \in A\) there exists an \(\alpha'' \in A\) such that \(K_{\alpha''} \subset K_\alpha \cap K_{\alpha'}\).
Then for \(n \geq 1\) and \(V \subset K^{\oplus n}\) a \(K\)-vector space we have \(V \cap k^{\oplus n} \not = 0\) if and only if \(V \cap K_\alpha^{\oplus n} \not = 0\) for all \(\alpha \in A\).
Proof
By induction on \(n\). The case \(n = 1\) follows from the assumptions. Assume the result proven for subspaces of \(K^{\oplus n - 1}\). Assume that \(V \subset K^{\oplus n}\) has nonzero intersection with \(K_\alpha^{\oplus n}\) for all \(\alpha \in A\). If \(V \cap 0 \oplus k^{\oplus n - 1}\) is nonzero then we win. Hence we may assume this is not the case. By induction hypothesis we can find an \(\alpha\) such that \(V \cap 0 \oplus K_\alpha^{\oplus n - 1}\) is zero. Let \(v = (x_1, \ldots, x_n) \in V \cap K_\alpha^{\oplus n}\) be a nonzero element. By our choice of \(\alpha\) we see that \(x_1\) is not zero. Replace \(v\) by \(x_1^{-1}v\) so that \(v = (1, x_2, \ldots, x_n)\). Note that if \(v' = (x_1', \ldots, x'_n) \in V \cap K_\alpha^{\oplus n}\), then \(v' - x_1'v = 0\) by our choice of \(\alpha\). Hence we see that \(V \cap K_\alpha^{\oplus n} = K_\alpha v\). If we choose some \(\alpha'\) such that \(K_{\alpha'} \subset K_\alpha\), then we see that necessarily \(v \in V \cap K_{\alpha'}^{\oplus n}\) (by the same arguments applied to \(\alpha'\)). Hence \[x_2, \ldots, x_n \in \bigcap\nolimits_{\alpha' \in A, K_{\alpha'} \subset K_\alpha} K_{\alpha'}\] which equals \(k\) by (2) and (3).
Lemma
Let \(K\) be a field of characteristic \(p\). Let \(\{K_\alpha\}_{\alpha \in A}\) be a collection of subfields of \(K\) with the following properties
\(K^p \subset K_\alpha\) for all \(\alpha \in A\),
\(K^p = \bigcap_{\alpha \in A} K_\alpha\),
for \(\alpha, \alpha' \in A\) there exists an \(\alpha'' \in A\) such that \(K_{\alpha''} \subset K_\alpha \cap K_{\alpha'}\).
Then
the intersection of the kernels of the maps \(\Omega_{K/\mathbf{F}_p} \to \Omega_{K/K_\alpha}\) is zero,
for any finite extension \(L/K\) we have \(L^p = \bigcap_{\alpha \in A} L^pK_\alpha\).
Proof
Proof of (1). Choose a \(p\)-basis \(\{x_i\}\) for \(K\) over \(\mathbf{F}_p\). Suppose that \(\eta = \sum_{i \in I'} y_i \text{d}x_i\) maps to zero in \(\Omega_{K/K_\alpha}\) for every \(\alpha \in A\). Here the index set \(I'\) is finite. By Lemma 07P2 this means that for every \(\alpha\) there exists a relation \[\sum\nolimits_E a_{E, \alpha} x^E = 0,\quad a_{E, \alpha} \in K_\alpha\] where \(E\) runs over multi-indices \(E = (e_i)_{i \in I'}\) with \(0 \leq e_i < p\). On the other hand, Lemma 07P2 guarantees there is no such relation \(\sum a_E x^E = 0\) with \(a_E \in K^p\). This is a contradiction by Lemma 07P3.
Proof of (2). Suppose that we have a tower \(L/M/K\) of finite extensions of fields. Set \(M_\alpha = M^p K_\alpha\) and \(L_\alpha = L^p K_\alpha = L^p M_\alpha\). Then we can first prove that \(M^p = \bigcap_{\alpha \in A} M_\alpha\), and after that prove that \(L^p = \bigcap_{\alpha \in A} L_\alpha\). Hence it suffices to prove (2) for primitive field extensions having no nontrivial subfields. First, assume that \(L = K(\theta)\) is separable over \(K\). Then \(L\) is generated by \(\theta^p\) over \(K\), hence we may assume that \(\theta \in L^p\). In this case we see that \[L^p = K^p \oplus K^p\theta \oplus \ldots K^p\theta^{d - 1} \quad\text{and}\quad L^pK_\alpha = K_\alpha \oplus K_\alpha \theta \oplus \ldots K_\alpha\theta^{d - 1}\] where \(d = [L : K]\). Thus the conclusion is clear in this case. The other case is where \(L = K(\theta)\) with \(\theta^p = t \in K\), \(t \not \in K^p\). In this case we have \[L^p = K^p \oplus K^pt \oplus \ldots K^pt^{p - 1} \quad\text{and}\quad L^pK_\alpha = K_\alpha \oplus K_\alpha t \oplus \ldots K_\alpha t^{p - 1}\] Again the result is clear.
Lemma
Let \(k\) be a field of characteristic \(p > 0\). Let \(\{x_i\}_{i \in I}\) be a \(p\)-basis for \(k\). Let \(n, m \geq 0\). Let \(K\) be the fraction field of \(A = k[[x_1, \ldots, x_n]][y_1, \ldots, y_m]\). Let \(J\) be a finite subset of \(I\). Consider the subfield \(k/k_J/k^p\) generated by \(k^p\) and \(x_i\) with \(i \in I \setminus J\). The fraction fields \(K_J\) of \[A_J = k_J[[x_1^p, \ldots, x_n^p]][y_1^p, \ldots, y_m^p]\] form a family of subfields of \(K\) as in Lemma 07P4. Moreover, each of the ring extensions \(A_J \subset A\) is finite.
Proof
Since \(k/k_J\) is finite, the ring extension \(k_J[[x_1^p, \ldots, x_d^p]] \subset k[[x_1, \ldots, x_d]]\) is finite by Algebra, Lemma 0394. This implies that \(A_J \to A\) is finite.
Let us check properties (1), (2), (3) of Lemma 07P4. Proof of (1). For \(a \in A\) we see that \(a^p \in A_J\). Hence \(K^p \subset K_J\). Proof of (2). Suppose that \(f/g^p \in K\), \(f, g \in A\), \(g \not = 0\) is contained in \(K_J\) for every choice of \(J\). Fix \(J\) for the moment. Since \(f/g^p \in K_J\) we can write \(f/g^p = a/b^p\) with \(a \in A_J\) and \(b \in A\) nonzero. Hence \(b^p f \in A_J\). For any \(A_J\)-derivation \(D : A \to A\) we see that \(0 = D(b^pf) = b^p D(f)\) hence \(D(f) = 0\) as \(A\) is a domain. Taking \(D = \partial_{x_i}\) and \(D = \partial_{y_j}\) we conclude that \(f \in k[[x_1^p, \ldots, x_n^p]][y_1^p, \ldots, y_m^p]\). Applying a \(k_J\)-derivation \(\theta : k \to k\) we similarly conclude that all coefficients of \(f\) are in \(k_J\), i.e., \(f \in A_J\). Since it is clear that \(A^p = \bigcap\nolimits_J A_J\) where \(J\) ranges over all subfields as in the lemma we conclude \(f \in A^p\) as desired. Proof of (3). This is clear because \(K_{J \cup J'} \subset K_J \cap K_{J'}\).
The singular locus
Let \(R\) be a Noetherian ring. The regular locus \(\text{Reg}(X)\) of \(X = \Spec(R)\) is the set of primes \(\mathfrak p\) such that \(R_\mathfrak p\) is a regular local ring. The singular locus \(\text{Sing}(X)\) of \(X = \Spec(R)\) is the complement \(X \setminus \text{Reg}(X)\), i.e., the set of primes \(\mathfrak p\) such that \(R_\mathfrak p\) is not a regular local ring. By the discussion preceding Algebra, Definition 00OD we see that \(\text{Reg}(X)\) is stable under generalization. In this section we study conditions that guarantee that \(\text{Reg}(X)\) is open.
Definition
Let \(R\) be a Noetherian ring. Let \(X = \Spec(R)\).
We say \(R\) is J-0 if \(\text{Reg}(X)\) contains a nonempty open.
We say \(R\) is J-1 if \(\text{Reg}(X)\) is open.
We say \(R\) is J-2 if any finite type \(R\)-algebra is J-1.
The ring \(\mathbf{Q}[x]/(x^2)\) does not satisfy J-0, but it does satisfy J-1. On the other hand, J-1 implies J-0 for Noetherian domains and more generally nonzero reduced Noetherian rings as such a ring is regular at the minimal primes. Here is a characterization of the J-1 property.
Lemma
Let \(R\) be a Noetherian ring. Let \(X = \Spec(R)\). The ring \(R\) is J-1 if and only if \(V(\mathfrak p) \cap \text{Reg}(X)\) contains a nonempty open subset of \(V(\mathfrak p)\) for all \(\mathfrak p \in \text{Reg}(X)\).
Proof
This follows from Topology, Lemma 0541 and the fact that \(\text{Reg}(X)\) is stable under generalization by Algebra, Lemma 0AFS.
Lemma
Let \(R\) be a Noetherian ring. Let \(X = \Spec(R)\). Assume that for all primes \(\mathfrak p \subset R\) the ring \(R/\mathfrak p\) is J-0. Then \(R\) is J-1.
Proof
We will show that the criterion of Lemma 07P8 applies. Let \(\mathfrak p \in \text{Reg}(X)\) be a prime of height \(r\). Pick \(f_1, \ldots, f_r \in \mathfrak p\) which map to generators of \(\mathfrak pR_\mathfrak p\). Since \(\mathfrak p \in \text{Reg}(X)\) we see that \(f_1, \ldots, f_r\) maps to a regular sequence in \(R_\mathfrak p\), see Algebra, Lemma 00NQ. Thus by Algebra, Lemma 061L we see that after replacing \(R\) by \(R_g\) for some \(g \in R\), \(g \not \in \mathfrak p\) the sequence \(f_1, \ldots, f_r\) is a regular sequence in \(R\). After another replacement we may also assume \(f_1, \ldots, f_r\) generate \(\mathfrak p\). Next, let \(\mathfrak p \subset \mathfrak q\) be a prime ideal such that \((R/\mathfrak p)_\mathfrak q\) is a regular local ring. By the assumption of the lemma there exists a non-empty open subset of \(V(\mathfrak p)\) consisting of such primes, hence it suffices to prove \(R_\mathfrak q\) is regular. Note that \(f_1, \ldots, f_r\) is a regular sequence in \(R_\mathfrak q\) such that \(R_\mathfrak q/(f_1, \ldots, f_r)R_\mathfrak q\) is regular. Hence \(R_\mathfrak q\) is regular by Algebra, Lemma 00NU.
Lemma
Let \(R \to S\) be a ring map. Assume that
\(R\) is a Noetherian domain,
\(R \to S\) is injective and of finite type, and
\(S\) is a domain and J-0.
Then \(R\) is J-0.
Proof
After replacing \(S\) by \(S_g\) for some nonzero \(g \in S\) we may assume that \(S\) is a regular ring. By generic flatness we may assume that also \(R \to S\) is faithfully flat, see Algebra, Lemma 051R. Then \(R\) is regular by Algebra, Lemma 07NG.
Lemma
Let \(R \to S\) be a ring map. Assume that
\(R\) is a Noetherian domain and J-0,
\(R \to S\) is injective and of finite type, and
\(S\) is a domain, and
the induced extension of fraction fields is separable.
Then \(S\) is J-0.
Proof
We may replace \(R\) by a principal localization and assume \(R\) is a regular ring. By Algebra, Lemma 07ND the ring map \(R \to S\) is smooth at \((0)\). Hence after replacing \(S\) by a principal localization we may assume that \(S\) is smooth over \(R\). Then \(S\) is regular too, see Algebra, Lemma 07NF.
Lemma
Let \(R\) be a Noetherian ring. The following are equivalent
\(R\) is J-2,
every finite type \(R\)-algebra which is a domain is J-0,
every finite \(R\)-algebra is J-1,
for every prime \(\mathfrak p\) and every finite purely inseparable extension \(L/\kappa(\mathfrak p)\) there exists a finite \(R\)-algebra \(R'\) which is a domain, which is J-0, and whose field of fractions is \(L\).
Proof
It is clear that we have the implications (1) \(\Rightarrow\) (2) and (2) \(\Rightarrow\) (4). Recall that a domain which is J-1 is J-0. Hence we also have the implications (1) \(\Rightarrow\) (3) and (3) \(\Rightarrow\) (4).
Let \(R \to S\) be a finite type ring map and let’s try to show \(S\) is J-1. By Lemma 07P9 it suffices to prove that \(S/\mathfrak q\) is J-0 for every prime \(\mathfrak q\) of \(S\). In this way we see (2) \(\Rightarrow\) (1).
Assume (4). We will show that (2) holds which will finish the proof. Let \(R \to S\) be a finite type ring map with \(S\) a domain. Let \(\mathfrak p = \Ker(R \to S)\). Let \(K\) be the fraction field of \(S\). There exists a diagram of fields \[\xymatrix{ K \ar[r] & K' \\ \kappa(\mathfrak p) \ar[u] \ar[r] & L \ar[u] }\] where the horizontal arrows are finite purely inseparable field extensions and where \(K'/L\) is separable, see Algebra, Lemma 04KM. Choose \(R' \subset L\) as in (4) and let \(S'\) be the image of the map \(S \otimes_R R' \to K'\). Then \(S'\) is a domain whose fraction field is \(K'\), hence \(S'\) is J-0 by Lemma 07PB and our choice of \(R'\). Then we apply Lemma 07PA to see that \(S\) is J-0 as desired.
Regularity and derivations
Let \(R \to S\) be a ring map. Let \(D : R \to R\) be a derivation. We say that \(D\) extends to \(S\) if there exists a derivation \(D' : S \to S\) such that \[\xymatrix{ S \ar[r]_{D'} & S \\ R \ar[u] \ar[r]^D & R \ar[u] }\] is commutative.
Lemma
Let \(R\) be a ring. Let \(D : R \to R\) be a derivation.
For any ideal \(I \subset R\) the derivation \(D\) extends canonically to a derivation \(D^\wedge : R^\wedge \to R^\wedge\) on the \(I\)-adic completion.
For any multiplicative subset \(S \subset R\) the derivation \(D\) extends uniquely to the localization \(S^{-1}R\) of \(R\).
If \(R \subset R'\) is a finite type extension of rings such that \(R_g \cong R'_g\) for some \(g \in R\) which is a nonzerodivisor in \(R'\), then \(g^ND\) extends to \(R'\) for some \(N \geq 0\).
Proof
Proof of (1). For \(n \geq 2\) we have \(D(I^n) \subset I^{n - 1}\) by the Leibniz rule. Hence \(D\) induces maps \(D_n : R/I^n \to R/I^{n - 1}\). Taking the limit we obtain \(D^\wedge\). We omit the verification that \(D^\wedge\) is a derivation.
Proof of (2). To extend \(D\) to \(S^{-1}R\) just set \(D(r/s) = D(r)/s - rD(s)/s^2\) and check the axioms.
Proof of the final statement. Let \(x_1, \ldots, x_n \in R'\) be generators of \(R'\) over \(R\). Choose an \(N\) such that \(g^Nx_i \in R\). Consider \(g^{N + 1}D\). By (2) this extends to \(R_g\). Moreover, by the Leibniz rule and our construction of the extension above we have \[g^{N + 1}D(x_i) = g^{N + 1}D(g^{-N} g^Nx_i) = -Ng^Nx_iD(g) + gD(g^Nx_i)\] and both terms are in \(R\). This implies that \[g^{N + 1}D(x_1^{e_1} \ldots x_n^{e_n}) = \sum e_i x_1^{e_1} \ldots x_i^{e_i - 1} \ldots x_n^{e_n} g^{N + 1}D(x_i)\] is an element of \(R'\). Hence every element of \(R'\) (which can be written as a sum of monomials in the \(x_i\) with coefficients in \(R\)) is mapped to an element of \(R'\) by \(g^{N + 1}D\) and we win.
Lemma
Let \(R\) be a regular ring. Let \(f \in R\). Assume there exists a derivation \(D : R \to R\) such that \(D(f)\) is a unit of \(R/(f)\). Then \(R/(f)\) is regular.
Proof
It suffices to prove this when \(R\) is a local ring with maximal ideal \(\mathfrak m\) and residue field \(\kappa\). In this case it suffices to prove that \(f \not \in \mathfrak m^2\), see Algebra, Lemma 00NQ. However, if \(f \in \mathfrak m^2\) then \(D(f) \in \mathfrak m\) by the Leibniz rule, a contradiction.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a regular local ring. Let \(m \geq 1\). Let \(f_1, \ldots, f_m \in \mathfrak m\). Assume there exist derivations \(D_1, \ldots, D_m : R \to R\) such that \(\det_{1 \leq i, j \leq m}(D_i(f_j))\) is a unit of \(R\). Then \(R/(f_1, \ldots, f_m)\) is regular and \(f_1, \ldots, f_m\) is a regular sequence.
Proof
It suffices to prove that \(f_1, \ldots, f_m\) are \(\kappa\)-linearly independent in \(\mathfrak m/\mathfrak m^2\), see Algebra, Lemma 00NQ. However, if there is a nontrivial linear relation then we get \(\sum a_i f_i \in \mathfrak m^2\) for some \(a_i \in R\) but not all \(a_i \in \mathfrak m\). Observe that \(D_i(\mathfrak m^2) \subset \mathfrak m\) and \(D_i(a_j f_j) \equiv a_j D_i(f_j) \bmod \mathfrak m\) by the Leibniz rule for derivations. Hence this would imply \[\sum a_j D_i(f_j) \in \mathfrak m\] which would contradict the assumption on the determinant.
Lemma
Let \(R\) be a regular ring. Let \(f \in R\). Assume there exists a derivation \(D : R \to R\) such that \(D(f)\) is a unit of \(R\). Then \(R[z]/(z^n - f)\) is regular for any integer \(n \geq 1\). More generally, \(R[z]/(p(z) - f)\) is regular for any \(p \in \mathbf{Z}[z]\).
Proof
By Algebra, Lemma 07NF we see that \(R[z]\) is a regular ring. Apply Lemma 07PF to the extension of \(D\) to \(R[z]\) which maps \(z\) to zero. This works because \(D\) annihilates any polynomial with integer coefficients and sends \(f\) to a unit.
Lemma
Let \(p\) be a prime number. Let \(B\) be a domain with \(p = 0\) in \(B\). Let \(f \in B\) be an element which is not a \(p\)th power in the fraction field of \(B\). If \(B\) is of finite type over a Noetherian complete local ring, then there exists a derivation \(D : B \to B\) such that \(D(f)\) is not zero.
Proof
Let \(R\) be a Noetherian complete local ring such that there exists a finite type ring map \(R \to B\). Of course we may replace \(R\) by its image in \(B\), hence we may assume \(R\) is a domain of characteristic \(p > 0\) (as well as Noetherian complete local). By Algebra, Lemma 032D we can write \(R\) as a finite extension of \(k[[x_1, \ldots, x_n]]\) for some field \(k\) and integer \(n\). Hence we may replace \(R\) by \(k[[x_1, \ldots, x_n]]\). Next, we use Algebra, Lemma 07NA to factor \(R \to B\) as \[R \subset R[y_1, \ldots, y_d] \subset B' \subset B\] with \(B'\) finite over \(R[y_1, \ldots, y_d]\) and \(B'_g \cong B_g\) for some nonzero \(g \in R\). Note that \(f' = g^{pN} f \in B'\) for some large integer \(N\). It is clear that \(f'\) is not a \(p\)th power in the fraction field of \(B'\). If we can find a derivation \(D' : B' \to B'\) with \(D'(f') \not = 0\), then Lemma 07PE guarantees that \(D = g^MD'\) extends to \(B\) for some \(M > 0\). Then \(D(f) = g^MD'(f) = g^MD'(g^{-pN}f') = g^{M - pN}D'(f')\) is nonzero. Thus it suffices to prove the lemma in case \(B\) is a finite extension of \(A = k[[x_1, \ldots, x_n]][y_1, \ldots, y_m]\).
Assume \(B\) is a finite extension of \(A = k[[x_1, \ldots, x_n]][y_1, \ldots, y_m]\). Denote \(L\) the fraction field of \(B\). Note that \(\text{d}f\) is not zero in \(\Omega_{L/\mathbf{F}_p}\), see Algebra, Lemma 031W. We apply Lemma 07P5 to find a subfield \(k' \subset k\) of finite index such that with \(A' = k'[[x_1^p, \ldots, x_n^p]][y_1^p, \ldots, y_m^p]\) the element \(\text{d}f\) does not map to zero in \(\Omega_{L/K'}\) where \(K'\) is the fraction field of \(A'\). Thus we can choose a \(K'\)-derivation \(D' : L \to L\) with \(D'(f) \not = 0\). Since \(A' \subset A\) and \(A \subset B\) are finite by construction we see that \(A' \subset B\) is finite. Choose \(b_1, \ldots, b_t \in B\) which generate \(B\) as an \(A'\)-module. Then \(D'(b_i) = f_i/g_i\) for some \(f_i, g_i \in B\) with \(g_i \not = 0\). Setting \(D = g_1 \ldots g_t D'\) we win.
Lemma
Let \(A\) be a Noetherian complete local domain. Then \(A\) is J-0.
Proof
By Algebra, Lemma 032D we can find a regular subring \(A_0 \subset A\) with \(A\) finite over \(A_0\). The induced extension \(K/K_0\) of fraction fields is finite. If \(K/K_0\) is separable, then we are done by Lemma 07PB. If not, then \(A_0\) and \(A\) have characteristic \(p > 0\). For any subextension \(K/M/K_0\) there exists a finite subextension \(A_0 \subset B \subset A\) whose fraction field is \(M\). Hence, arguing by induction on \([K : K_0]\) we may assume there exists \(A_0 \subset B \subset A\) such that \(B\) is J-0 and \(K/M\) has no nontrivial subextensions. In this case, if \(K/M\) is separable, then we see that \(A\) is J-0 by Lemma 07PB. If not, then \(K = M[z]/(z^p - b_1/b_2)\) for some \(b_1, b_2 \in B\) with \(b_2 \not = 0\) and \(b_1/b_2\) not a \(p\)th power in \(M\). Choose \(a \in A\) nonzero such that \(az \in A\). After replacing \(z\) by \(b_2 a^p z\) we obtain \(K = M[z]/(z^p - b)\) with \(z \in A\) and \(b \in B\) not a \(p\)th power in \(M\). By Lemma 07PH we can find a derivation \(D : B \to B\) with \(D(b) \not = 0\). Applying Lemma 07PG we see that \(A_\mathfrak p\) is regular for any prime \(\mathfrak p\) of \(A\) lying over a regular prime of \(B\) and not containing \(D(b)\). As \(B\) is J-0 we conclude \(A\) is too.
Proposition
The following types of rings are J-2:
fields,
Noetherian complete local rings,
\(\mathbf{Z}\),
Noetherian local rings of dimension \(1\),
Nagata rings of dimension \(1\),
Dedekind domains with fraction field of characteristic zero,
finite type ring extensions of any of the above.
Proof
For cases (1), (3), (5), and (6) this is proved by checking condition (4) of Lemma 07PC. We will only do this in case \(R\) is a Nagata ring of dimension \(1\). Let \(\mathfrak p \subset R\) be a prime ideal and let \(L/\kappa(\mathfrak p)\) be a finite purely inseparable extension. If \(\mathfrak p \subset R\) is a maximal ideal, then \(R \to L\) is finite and \(L\) is a regular ring and we’ve checked the condition. If \(\mathfrak p \subset R\) is a minimal prime, then the Nagata condition insures that the integral closure \(R' \subset L\) of \(R\) in \(L\) is finite over \(R\). Then \(R'\) is a normal domain of dimension \(1\) (Algebra, Lemma 00OJ) hence regular (Algebra, Lemma 031S) and we’ve checked the condition in this case as well.
For case (2), we will use condition (3) of Lemma 07PC. Let \(R\) be a Noetherian complete local ring. Note that if \(R \to R'\) is finite, then \(R'\) is a product of Noetherian complete local rings, see Algebra, Lemma 0325. Thus by Lemma 07P9 it suffices to prove that a Noetherian complete local ring which is a domain is J-0, which is Lemma 07PI.
For case (4), we also use condition (3) of Lemma 07PC. Namely, if \(R\) is a local Noetherian ring of dimension \(1\) and \(R \to R'\) is finite, then \(\Spec(R')\) is finite. Since the regular locus is stable under generalization, we see that \(R'\) is J-1.
Formal smoothness and regularity
The title of this section refers to André’s Theorem 0H7U.
Lemma
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Let \(D : A \to A\) be a derivation. Assume that \(B\) is complete and \(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology. Then there exists an extension \(D' : B \to B\) of \(D\).
Proof
Denote \(B[\epsilon] = B[x]/(x^2)\) the ring of dual numbers over \(B\). Consider the ring map \(\psi : A \to B[\epsilon]\), \(a \mapsto a + \epsilon D(a)\). Consider the commutative diagram \[\xymatrix{ B \ar[r]_1 & B \\ A \ar[u] \ar[r]^\psi & B[\epsilon] \ar[u] }\] By Lemma 07NJ and the assumption of formal smoothness of \(B/A\) we find a map \(\varphi : B \to B[\epsilon]\) fitting into the diagram. Write \(\varphi(b) = b + \epsilon D'(b)\). Then \(D' : B \to B\) is the desired extension.
Proposition
Let \(A \to B\) be a local homomorphism of Noetherian complete local rings. Let \(k\) be the residue field of \(A\) and \(\overline{B} = B \otimes_A k\) the special fibre. The following are equivalent
\(A \to B\) is regular,
\(A \to B\) is flat and \(\overline{B}\) is geometrically regular over \(k\),
\(A \to B\) is flat and \(k \to \overline{B}\) is formally smooth in the \(\mathfrak m_{\overline{B}}\)-adic topology, and
\(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology.
Proof
We have seen the equivalence of (2), (3), and (4) in Proposition 07NQ. It is clear that (1) implies (2). Thus we assume the equivalent conditions (2), (3), and (4) hold and we prove (1).
Let \(\mathfrak p\) be a prime of \(A\). We will show that \(B \otimes_A \kappa(\mathfrak p)\) is geometrically regular over \(\kappa(\mathfrak p)\). By Lemma 07EG we may replace \(A\) by \(A/\mathfrak p\) and \(B\) by \(B/\mathfrak pB\). Thus we may assume that \(A\) is a domain and that \(\mathfrak p = (0)\).
Choose \(A_0 \subset A\) as in Algebra, Lemma 032D. We will use all the properties stated in that lemma without further mention. As \(A_0 \to A\) induces an isomorphism on residue fields, and as \(B/\mathfrak m_A B\) is geometrically regular over \(A/\mathfrak m_A\) we can find a diagram \[\xymatrix{ C \ar[r] & B \\ A_0 \ar[r] \ar[u] & A \ar[u] }\] with \(A_0 \to C\) formally smooth in the \(\mathfrak m_C\)-adic topology such that \(B = C \otimes_{A_0} A\), see Remark 07NS. (Completion in the tensor product is not needed as \(A_0 \to A\) is finite, see Algebra, Lemma 00MA.) Hence it suffices to show that \(C \otimes_{A_0} K_0\) is a geometrically regular algebra over the fraction field \(K_0\) of \(A_0\).
The upshot of the preceding paragraph is that we may assume that \(A = k[[x_1, \ldots, x_n]]\) where \(k\) is a field or \(A = \Lambda[[x_1, \ldots, x_n]]\) where \(\Lambda\) is a Cohen ring. In this case \(B\) is a regular ring, see Algebra, Lemma 031E. Hence \(B \otimes_A K\) is a regular ring too (where \(K\) is the fraction field of \(A\)) and we win if the characteristic of \(K\) is zero.
Thus we are left with the case where \(A = k[[x_1, \ldots, x_n]]\) and \(k\) is a field of characteristic \(p > 0\). Let \(L/K\) be a finite purely inseparable field extension. We will show by induction on \([L : K]\) that \(B \otimes_A L\) is regular. The base case is \(L = K\) which we’ve seen above. Let \(K \subset M \subset L\) be a subfield such that \(L\) is a degree \(p\) extension of \(M\) obtained by adjoining a \(p\)th root of an element \(f \in M\). Let \(A'\) be a finite \(A\)-subalgebra of \(M\) with fraction field \(M\). Clearing denominators, we may and do assume \(f \in A'\). Set \(A'' = A'[z]/(z^p -f)\) and note that \(A' \subset A''\) is finite and that the fraction field of \(A''\) is \(L\). By induction we know that \(B \otimes_A M\) ring is regular. We have \[B \otimes_A L = B \otimes_A M[z]/(z^p - f)\] By Lemma 07PH we know there exists a derivation \(D : A' \to A'\) such that \(D(f) \not = 0\). As \(A' \to B \otimes_A A'\) is formally smooth in the \(\mathfrak m\)-adic topology by Lemma 07EH we can use Lemma 07PL to extend \(D\) to a derivation \(D' : B \otimes_A A' \to B \otimes_A A'\). Note that \(D'(f) = D(f)\) is a unit in \(B \otimes_A M\) as \(D(f)\) is not zero in \(A' \subset M\). Hence \(B \otimes_A L\) is regular by Lemma 07PG and we win.
Theorem
Let \(A \to B\) be a local homomorphism of Noetherian local rings. Let \(k\) be the residue field of \(A\) and \(\overline{B} = B \otimes_A k\) the special fibre. Assume \(A \to A^\wedge\) is regular6. The following are equivalent
\(A \to B\) is regular,
\(A \to B\) is flat and \(\overline{B}\) is geometrically regular over \(k\),
\(A \to B\) is flat and \(k \to \overline{B}\) is formally smooth in the \(\mathfrak m_{\overline{B}}\)-adic topology, and
\(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology.
Proof
We have seen the equivalence of (2), (3), and (4) in Proposition 07NQ. It is clear that (1) implies (2). Thus we assume (4) holds and we prove (1).
By Lemma 07ED we see that \(A^\wedge \to B^\wedge\) is formally smooth. By Proposition 07PM we see that \(A^\wedge \to B^\wedge\) is regular. By assumption \(A \to A^\wedge\) is regular, hence \(A \to B^\wedge\) is regular by Lemma 07QI. Since \(B \to B^\wedge\) is faithfully flat, we conclude that \(A \to B\) is regular by Lemma 07NT.
G-rings
Let \(A\) be a Noetherian local ring. In Section 07NU we have seen that some but not all properties of \(A\) are reflected in the completion \(A^\wedge\) of \(A\). To study this further we introduce some terminology. For a prime \(\mathfrak q\) of \(A\) the fibre ring \[A^\wedge \otimes_A \kappa(\mathfrak q) = (A^\wedge)_\mathfrak q/\mathfrak q(A^\wedge)_\mathfrak q = (A/\mathfrak q)^\wedge \otimes_{A/\mathfrak q} \kappa(\mathfrak q)\] is called a formal fibre of \(A\). We think of the formal fibre as an algebra over \(\kappa(\mathfrak q)\). Thus \(A \to A^\wedge\) is a regular ring homomorphism if and only if all the formal fibres are geometrically regular algebras.
Definition
A ring \(R\) is called a G-ring if \(R\) is Noetherian and for every prime \(\mathfrak p\) of \(R\) the ring map \(R_\mathfrak p \to (R_\mathfrak p)^\wedge\) is regular.
By the discussion above we see that \(R\) is a G-ring if and only if every local ring \(R_\mathfrak p\) has geometrically regular formal fibres. Note that if \(\mathbf{Q} \subset R\), then it suffices to check the formal fibres are regular. Another way to express the G-ring condition is described in the following lemma.
Lemma
Let \(R\) be a Noetherian ring. Then \(R\) is a G-ring if and only if for every pair of primes \(\mathfrak q \subset \mathfrak p \subset R\) the algebra \[(R/\mathfrak q)_\mathfrak p^\wedge \otimes_{R/\mathfrak q} \kappa(\mathfrak q)\] is geometrically regular over \(\kappa(\mathfrak q)\).
Proof
This follows from the fact that \[R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) = (R/\mathfrak q)_\mathfrak p^\wedge \otimes_{R/\mathfrak q} \kappa(\mathfrak q)\] as algebras over \(\kappa(\mathfrak q)\).
Lemma
Let \(R \to R'\) be a finite type map of Noetherian rings and let \[\xymatrix{ \mathfrak q' \ar[r] & \mathfrak p' \ar[r] & R' \\ \mathfrak q \ar[r] \ar@{-}[u] & \mathfrak p \ar[r] \ar@{-}[u] & R \ar[u] }\] be primes. Assume \(R \to R'\) is quasi-finite at \(\mathfrak p'\).
If the formal fibre \(R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q)\) is geometrically regular over \(\kappa(\mathfrak q)\), then the formal fibre \((R'_{\mathfrak p'})^\wedge \otimes_{R'} \kappa(\mathfrak q')\) is geometrically regular over \(\kappa(\mathfrak q')\).
If the formal fibres of \(R_\mathfrak p\) are geometrically regular, then the formal fibres of \(R'_{\mathfrak p'}\) are geometrically regular.
If \(R \to R'\) is quasi-finite and \(R\) is a G-ring, then \(R'\) is a G-ring.
Proof
It is clear that (1) \(\Rightarrow\) (2) \(\Rightarrow\) (3). Assume \(R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q)\) is geometrically regular over \(\kappa(\mathfrak q)\). By Algebra, Lemma 07NC we see that \[R_\mathfrak p^\wedge \otimes_R R' = (R'_{\mathfrak p'})^\wedge \times B\] for some \(R_\mathfrak p^\wedge\)-algebra \(B\). Hence \(R'_{\mathfrak p'} \to (R'_{\mathfrak p'})^\wedge\) is a factor of a base change of the map \(R_\mathfrak p \to R_\mathfrak p^\wedge\). It follows that \((R'_{\mathfrak p'})^\wedge \otimes_{R'} \kappa(\mathfrak q')\) is a factor of \[R_\mathfrak p^\wedge \otimes_R R' \otimes_{R'} \kappa(\mathfrak q') = R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) \otimes_{\kappa(\mathfrak q)} \kappa(\mathfrak q').\] Thus the result follows as extension of base field preserves geometric regularity, see Algebra, Lemma 0381.
Lemma
Let \(R\) be a Noetherian ring. Then \(R\) is a G-ring if and only if for every finite free ring map \(R \to S\) the formal fibres of \(S\) are regular rings.
Proof
Assume that for any finite free ring map \(R \to S\) the ring \(S\) has regular formal fibres. Let \(\mathfrak q \subset \mathfrak p \subset R\) be primes and let \(\kappa(\mathfrak q) \subset L\) be a finite purely inseparable extension. To show that \(R\) is a G-ring it suffices to show that \[R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) \otimes_{\kappa(\mathfrak q)} L\] is a regular ring. Choose a finite free extension \(R \to R'\) such that \(\mathfrak q' = \mathfrak qR'\) is a prime and such that \(\kappa(\mathfrak q')\) is isomorphic to \(L\) over \(\kappa(\mathfrak q)\), see Algebra, Lemma 07NE. By Algebra, Lemma 07N9 we have \[R_\mathfrak p^\wedge \otimes_R R' = \prod (R'_{\mathfrak p_i'})^\wedge\] where \(\mathfrak p_i'\) are the primes of \(R'\) lying over \(\mathfrak p\). Thus we have \[R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) \otimes_{\kappa(\mathfrak q)} L = R_\mathfrak p^\wedge \otimes_R R' \otimes_{R'} \kappa(\mathfrak q') = \prod (R'_{\mathfrak p_i'})^\wedge \otimes_{R'_{\mathfrak p'_i}} \kappa(\mathfrak q')\] Our assumption is that the rings on the right are regular, hence the ring on the left is regular too. Thus \(R\) is a G-ring. The converse follows from Lemma 07PP.
Lemma
Let \(k\) be a field of characteristic \(p\). Let \(A = k[[x_1, \ldots, x_n]][y_1, \ldots, y_m]\) and denote \(K\) the fraction field of \(A\). Let \(\mathfrak p \subset A\) be a prime. Then \(A_\mathfrak p^\wedge \otimes_A K\) is geometrically regular over \(K\).
Proof
Let \(L/K\) be a finite purely inseparable field extension. We will show by induction on \([L : K]\) that \(A_\mathfrak p^\wedge \otimes L\) is regular. The base case is \(L = K\): as \(A\) is regular, \(A_\mathfrak p^\wedge\) is regular (Lemma 07NY), hence the localization \(A_\mathfrak p^\wedge \otimes K\) is regular. Let \(K \subset M \subset L\) be a subfield such that \(L\) is a degree \(p\) extension of \(M\) obtained by adjoining a \(p\)th root of an element \(f \in M\). Let \(B\) be a finite \(A\)-subalgebra of \(M\) with fraction field \(M\). Clearing denominators, we may and do assume \(f \in B\). Set \(C = B[z]/(z^p -f)\) and note that \(B \subset C\) is finite and that the fraction field of \(C\) is \(L\). Since \(A \subset B \subset C\) are finite and \(L/M/K\) are purely inseparable we see that for every element of \(B\) or \(C\) some power of it lies in \(A\). Hence there is a unique prime \(\mathfrak r \subset B\), resp. \(\mathfrak q \subset C\) lying over \(\mathfrak p\). Note that \[A_\mathfrak p^\wedge \otimes_A M = B_\mathfrak r^\wedge \otimes_B M\] see Algebra, Lemma 07N9. By induction we know that this ring is regular. In the same manner we have \[A_\mathfrak p^\wedge \otimes_A L = C_\mathfrak q^\wedge \otimes_C L = B_\mathfrak r^\wedge \otimes_B M[z]/(z^p - f)\] the last equality because the completion of \(C = B[z]/(z^p - f)\) equals \(B_\mathfrak r^\wedge[z]/(z^p -f)\). By Lemma 07PH we know there exists a derivation \(D : B \to B\) such that \(D(f) \not = 0\). In other words, \(g = D(f)\) is a unit in \(M\)! By Lemma 07PE \(D\) extends to a derivation of \(B_\mathfrak r\), \(B_\mathfrak r^\wedge\) and \(B_\mathfrak r^\wedge \otimes_B M\) (successively extending through a localization, a completion, and a localization). Since it is an extension we end up with a derivation of \(B_\mathfrak r^\wedge \otimes_B M\) which maps \(f\) to \(g\) and \(g\) is a unit of the ring \(B_\mathfrak r^\wedge \otimes_B M\). Hence \(A_\mathfrak p^\wedge \otimes_A L\) is regular by Lemma 07PG and we win.
Proposition
A Noetherian complete local ring is a G-ring.
Proof
Let \(A\) be a Noetherian complete local ring. By Lemma 07PN it suffices to check that \(B = A/\mathfrak q\) has geometrically regular formal fibres over the minimal prime \((0)\) of \(B\). Thus we may assume that \(A\) is a domain and it suffices to check the condition for the formal fibres over the minimal prime \((0)\) of \(A\). Let \(K\) be the fraction field of \(A\).
We can choose a subring \(A_0 \subset A\) which is a regular complete local ring such that \(A\) is finite over \(A_0\), see Algebra, Lemma 032D. Moreover, we may assume that \(A_0\) is a power series ring over a field or a Cohen ring. By Lemma 07PP we see that it suffices to prove the result for \(A_0\).
Assume that \(A\) is a power series ring over a field or a Cohen ring. Since \(A\) is regular the localizations \(A_\mathfrak p\) are regular (see Algebra, Definition 00OD and the discussion preceding it). Hence the completions \(A_\mathfrak p^\wedge\) are regular, see Lemma 07NY. Hence the fibre \(A_{\mathfrak p}^\wedge \otimes_A K\) is, as a localization of \(A_\mathfrak p^\wedge\), also regular. Thus we are done if the characteristic of \(K\) is \(0\). The positive characteristic case is the case \(A = k[[x_1, \ldots, x_d]]\) which is a special case of Lemma 07PR.
Lemma
Let \(R\) be a Noetherian ring. Then \(R\) is a G-ring if and only if \(R_\mathfrak m\) has geometrically regular formal fibres for every maximal ideal \(\mathfrak m\) of \(R\).
Proof
Assume \(R_\mathfrak m \to R_\mathfrak m^\wedge\) is regular for every maximal ideal \(\mathfrak m\) of \(R\). Let \(\mathfrak p\) be a prime of \(R\) and choose a maximal ideal \(\mathfrak p \subset \mathfrak m\). Since \(R_\mathfrak m \to R_\mathfrak m^\wedge\) is faithfully flat we can choose a prime \(\mathfrak p'\) in \(R_\mathfrak m^\wedge\) lying over \(\mathfrak pR_\mathfrak m\). Consider the commutative diagram \[\xymatrix{ R_\mathfrak m^\wedge \ar[r] & (R_\mathfrak m^\wedge)_{\mathfrak p'} \ar[r] & (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge \\ R_\mathfrak m \ar[u] \ar[r] & R_\mathfrak p \ar[u] \ar[r] & R_\mathfrak p^\wedge \ar[u] }\] By assumption the ring map \(R_\mathfrak m \to R_\mathfrak m^\wedge\) is regular. By Proposition 07PS \((R_\mathfrak m^\wedge)_{\mathfrak p'} \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) is regular. The localization \(R_\mathfrak m^\wedge \to (R_\mathfrak m^\wedge)_{\mathfrak p'}\) is regular. Hence \(R_\mathfrak m \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) is regular by Lemma 07QI. Since it factors through the localization \(R_\mathfrak p\), also the ring map \(R_\mathfrak p \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) is regular. Thus we may apply Lemma 07NT to see that \(R_\mathfrak p \to R_\mathfrak p^\wedge\) is regular.
Lemma
Let \(R\) be a Noetherian local ring which is a G-ring. Then the henselization \(R^h\) and the strict henselization \(R^{sh}\) are G-rings.
Proof
We will use the criterion of Lemma 07PT. Let \(\mathfrak q \subset R^h\) be a prime and set \(\mathfrak p = R \cap \mathfrak q\). Set \(\mathfrak q_1 = \mathfrak q\) and let \(\mathfrak q_2, \ldots, \mathfrak q_t\) be the other primes of \(R^h\) lying over \(\mathfrak p\), so that \(R^h \otimes_R \kappa(\mathfrak p) = \prod\nolimits_{i = 1, \ldots, t} \kappa(\mathfrak q_i)\), see Lemma 07QQ. Using that \((R^h)^\wedge = R^\wedge\) (Lemma 06LJ) we see \[\prod\nolimits_{i = 1, \ldots, t} (R^h)^\wedge \otimes_{R^h} \kappa(\mathfrak q_i) = (R^h)^\wedge \otimes_{R^h} (R^h \otimes_R \kappa(\mathfrak p)) = R^\wedge \otimes_R \kappa(\mathfrak p)\] Hence \((R^h)^\wedge \otimes_{R^h} \kappa(\mathfrak q_i)\) is geometrically regular over \(\kappa(\mathfrak p)\) by assumption. Since \(\kappa(\mathfrak q_i)\) is separable algebraic over \(\kappa(\mathfrak p)\) it follows from Algebra, Lemma 07QH that \((R^h)^\wedge \otimes_{R^h} \kappa(\mathfrak q_i)\) is geometrically regular over \(\kappa(\mathfrak q_i)\).
Let \(\mathfrak r \subset R^{sh}\) be a prime and set \(\mathfrak p = R \cap \mathfrak r\). Set \(\mathfrak r_1 = \mathfrak r\) and let \(\mathfrak r_2, \ldots, \mathfrak r_s\) be the other primes of \(R^{sh}\) lying over \(\mathfrak p\), so that \(R^{sh} \otimes_R \kappa(\mathfrak p) = \prod\nolimits_{i = 1, \ldots, s} \kappa(\mathfrak r_i)\), see Lemma 07QQ. Then we see that \[\prod\nolimits_{i = 1, \ldots, s} (R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r_i) = (R^{sh})^\wedge \otimes_{R^{sh}} (R^{sh} \otimes_R \kappa(\mathfrak p)) = (R^{sh})^\wedge \otimes_R \kappa(\mathfrak p)\] Note that \(R^\wedge \to (R^{sh})^\wedge\) is formally smooth in the \(\mathfrak m_{(R^{sh})^\wedge}\)-adic topology, see Lemma 06LJ. Hence \(R^\wedge \to (R^{sh})^\wedge\) is regular by Proposition 07PM. We conclude that \((R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r_i)\) is regular over \(\kappa(\mathfrak p)\) by Lemma 07QI as \(R^\wedge \otimes_R \kappa(\mathfrak p)\) is regular over \(\kappa(\mathfrak p)\) by assumption. Since \(\kappa(\mathfrak r_i)\) is separable algebraic over \(\kappa(\mathfrak p)\) it follows from Algebra, Lemma 07QH that \((R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r_i)\) is geometrically regular over \(\kappa(\mathfrak r_i)\).
Lemma
Let \(p\) be a prime number. Let \(A\) be a Noetherian complete local domain with fraction field \(K\) of characteristic \(p\). Let \(\mathfrak q \subset A[x]\) be a maximal ideal lying over the maximal ideal of \(A\) and let \((0) \not = \mathfrak r \subset \mathfrak q\) be a prime lying over \((0) \subset A\). Then \(A[x]_\mathfrak q^\wedge \otimes_{A[x]} \kappa(\mathfrak r)\) is geometrically regular over \(\kappa(\mathfrak r)\).
Proof
Note that \(K \subset \kappa(\mathfrak r)\) is finite. Hence, given a finite purely inseparable extension \(L/\kappa(\mathfrak r)\) there exists a finite extension of Noetherian complete local domains \(A \subset B\) such that \(\kappa(\mathfrak r) \otimes_A B\) surjects onto \(L\). Namely, you take \(B \subset L\) a finite \(A\)-subalgebra whose field of fractions is \(L\). Denote \(\mathfrak r' \subset B[x]\) the kernel of the map \(B[x] = A[x] \otimes_A B \to \kappa(\mathfrak r) \otimes_A B \to L\) so that \(\kappa(\mathfrak r') = L\). Then \[A[x]_\mathfrak q^\wedge \otimes_{A[x]} L = A[x]_\mathfrak q^\wedge \otimes_{A[x]} B[x] \otimes_{B[x]} \kappa(\mathfrak r') = \prod B[x]_{\mathfrak q_i}^\wedge \otimes_{B[x]} \kappa(\mathfrak r')\] where \(\mathfrak q_1, \ldots, \mathfrak q_t\) are the primes of \(B[x]\) lying over \(\mathfrak q\), see Algebra, Lemma 07N9. Thus we see that it suffices to prove the rings \(B[x]_{\mathfrak q_i}^\wedge \otimes_{B[x]} \kappa(\mathfrak r')\) are regular. This reduces us to showing that \(A[x]_\mathfrak q^\wedge \otimes_{A[x]} \kappa(\mathfrak r)\) is regular in the special case that \(K = \kappa(\mathfrak r)\).
Assume \(K = \kappa(\mathfrak r)\). In this case we see that \(\mathfrak r K[x]\) is generated by \(x - f\) for some \(f \in K\) and \[A[x]_\mathfrak q^\wedge \otimes_{A[x]} \kappa(\mathfrak r) = (A[x]_\mathfrak q^\wedge \otimes_A K)/(x - f)\] The derivation \(D = \text{d}/\text{d}x\) of \(A[x]\) extends to \(K[x]\) and maps \(x - f\) to a unit of \(K[x]\). Moreover \(D\) extends to \(A[x]_\mathfrak q^\wedge \otimes_A K\) by Lemma 07PE. As \(A \to A[x]_\mathfrak q^\wedge\) is formally smooth (see Lemmas 07EC and 07ED) the ring \(A[x]_\mathfrak q^\wedge \otimes_A K\) is regular by Proposition 07PM (the arguments of the proof of that proposition simplify significantly in this particular case). We conclude by Lemma 07PF.
Proposition
Let \(R\) be a G-ring. If \(R \to S\) is essentially of finite type then \(S\) is a G-ring.
Proof
Since being a G-ring is a property of the local rings it is clear that a localization of a G-ring is a G-ring. Conversely, if every localization at a prime is a G-ring, then the ring is a G-ring. Thus it suffices to show that \(S_\mathfrak q\) is a G-ring for every finite type \(R\)-algebra \(S\) and every prime \(\mathfrak q\) of \(S\). Writing \(S\) as a quotient of \(R[x_1, \ldots, x_n]\) we see from Lemma 07PP that it suffices to prove that \(R[x_1, \ldots, x_n]\) is a G-ring. By induction on \(n\) it suffices to prove that \(R[x]\) is a G-ring. Let \(\mathfrak q \subset R[x]\) be a maximal ideal. By Lemma 07PT it suffices to show that \[R[x]_\mathfrak q \longrightarrow R[x]_\mathfrak q^\wedge\] is regular. If \(\mathfrak q\) lies over \(\mathfrak p \subset R\), then we may replace \(R\) by \(R_\mathfrak p\). Hence we may assume that \(R\) is a Noetherian local G-ring with maximal ideal \(\mathfrak m\) and that \(\mathfrak q \subset R[x]\) lies over \(\mathfrak m\). Note that there is a unique prime \(\mathfrak q' \subset R^\wedge[x]\) lying over \(\mathfrak q\). Consider the diagram \[\xymatrix{ R[x]_\mathfrak q^\wedge \ar[r] & (R^\wedge[x]_{\mathfrak q'})^\wedge \\ R[x]_\mathfrak q \ar[r] \ar[u] & R^\wedge[x]_{\mathfrak q'} \ar[u] }\] Since \(R\) is a G-ring the lower horizontal arrow is regular (as a localization of a base change of the regular ring map \(R \to R^\wedge\)). Suppose we can prove the right vertical arrow is regular. Then it follows that the composition \(R[x]_\mathfrak q \to (R^\wedge[x]_{\mathfrak q'})^\wedge\) is regular, and hence the left vertical arrow is regular by Lemma 07NT. Hence we see that we may assume \(R\) is a Noetherian complete local ring and \(\mathfrak q\) a prime lying over the maximal ideal of \(R\).
Let \(R\) be a Noetherian complete local ring and let \(\mathfrak q \subset R[x]\) be a maximal ideal lying over the maximal ideal of \(R\). Let \(\mathfrak r \subset \mathfrak q\) be a prime ideal. We want to show that \(R[x]_\mathfrak q^\wedge \otimes_{R[x]} \kappa(\mathfrak r)\) is a geometrically regular algebra over \(\kappa(\mathfrak r)\). Set \(\mathfrak p = R \cap \mathfrak r\). Then we can replace \(R\) by \(R/\mathfrak p\) and \(\mathfrak q\) and \(\mathfrak r\) by their images in \(R/\mathfrak p[x]\), see Lemma 07PN. Hence we may assume that \(R\) is a domain and that \(\mathfrak r \cap R = (0)\).
By Algebra, Lemma 032D we can find \(R_0 \subset R\) which is regular and such that \(R\) is finite over \(R_0\). Applying Lemma 07PP we see that it suffices to prove \(R[x]_\mathfrak q^\wedge \otimes_{R[x]} \kappa(\mathfrak r)\) is geometrically regular over \(\kappa(\mathfrak r)\) when, in addition to the above, \(R\) is a regular complete local ring.
Now \(R\) is a regular complete local ring, we have \(\mathfrak r \subset \mathfrak q \subset R[x]\), we have \((0) = R \cap \mathfrak r\) and \(\mathfrak q\) is a maximal ideal lying over the maximal ideal of \(R\). Since \(R\) is regular the ring \(R[x]\) is regular (Algebra, Lemma 07NF). Hence the localization \(R[x]_\mathfrak q\) is regular. Hence the completions \(R[x]_\mathfrak q^\wedge\) are regular, see Lemma 07NY. Hence the fibre \(R[x]_{\mathfrak q}^\wedge \otimes_{R[x]} \kappa(\mathfrak r)\) is, as a localization of \(R[x]_\mathfrak q^\wedge\), also regular. Thus we are done if the characteristic of the fraction field of \(R\) is \(0\).
If the characteristic of \(R\) is positive, then \(R = k[[x_1, \ldots, x_n]]\). In this case we split the argument in two subcases:
Remark
Let \(R\) be a G-ring and let \(I \subset R\) be an ideal. In general it is not the case that the \(I\)-adic completion \(R^\wedge\) is a G-ring. An example was given by Nishimura in [Nishimura]. A generalization and, in some sense, clarification of this example can be found in the last section of [Dumitrescu].
Proposition
The following types of rings are G-rings:
fields,
Noetherian complete local rings,
\(\mathbf{Z}\),
Dedekind domains with fraction field of characteristic zero,
finite type ring extensions of any of the above.
Proof
For fields, \(\mathbf{Z}\) and Dedekind domains of characteristic zero this follows immediately from the definition and the fact that the completion of a discrete valuation ring is a discrete valuation ring. A Noetherian complete local ring is a G-ring by Proposition 07PS. The statement on finite type overrings is Proposition 07PV.
Lemma
Let \((A, \mathfrak m)\) be a henselian local ring. Then \(A\) is a filtered colimit of a system of henselian local G-rings with local transition maps.
Proof
Write \(A = \colim A_i\) as a filtered colimit of finite type \(\mathbf{Z}\)-algebras. Let \(\mathfrak p_i\) be the prime ideal of \(A_i\) lying under \(\mathfrak m\). We may replace \(A_i\) by the localization of \(A_i\) at \(\mathfrak p_i\). Then \(A_i\) is a Noetherian local G-ring (Proposition 07PX). By Lemma 0A04 we see that \(A = \colim A_i^h\). By Lemma 07QR the rings \(A_i^h\) are G-rings.
Lemma
Let \(A\) be a G-ring. Let \(I \subset A\) be an ideal and let \(A^\wedge\) be the completion of \(A\) with respect to \(I\). Then \(A \to A^\wedge\) is regular.
Proof
The ring map \(A \to A^\wedge\) is flat by Algebra, Lemma 00MB. The ring \(A^\wedge\) is Noetherian by Algebra, Lemma 0316. Thus it suffices to check the third condition of Lemma 07C0. Let \(\mathfrak m' \subset A^\wedge\) be a maximal ideal lying over \(\mathfrak m \subset A\). By Algebra, Lemma 05GI we have \(IA^\wedge \subset \mathfrak m'\). Since \(A^\wedge/IA^\wedge = A/I\) we see that \(I \subset \mathfrak m\), \(\mathfrak m/I = \mathfrak m'/IA^\wedge\), and \(A/\mathfrak m = A^\wedge/\mathfrak m'\). Since \(A^\wedge/\mathfrak m'\) is a field, we conclude that \(\mathfrak m\) is a maximal ideal as well. Then \(A_\mathfrak m \to A^\wedge_{\mathfrak m'}\) is a flat local ring homomorphism of Noetherian local rings which identifies residue fields and such that \(\mathfrak m A^\wedge_{\mathfrak m'} = \mathfrak m'A^\wedge_{\mathfrak m'}\). Thus it induces an isomorphism on complete local rings, see Lemma 0AGX. Let \((A_\mathfrak m)^\wedge\) be the completion of \(A_\mathfrak m\) with respect to its maximal ideal. The ring map \[(A^\wedge)_{\mathfrak m'} \to ((A^\wedge)_{\mathfrak m'})^\wedge = (A_\mathfrak m)^\wedge\] is faithfully flat (Algebra, Lemma 00MC). Thus we can apply Lemma 07NT to the ring maps \[A_\mathfrak m \to (A^\wedge)_{\mathfrak m'} \to (A_\mathfrak m)^\wedge\] to conclude because \(A_\mathfrak m \to (A_\mathfrak m)^\wedge\) is regular as \(A\) is a G-ring.
Lemma
Let \(A\) be a G-ring. Let \(I \subset A\) be an ideal. Let \((A^h, I^h)\) be the henselization of the pair \((A, I)\), see Lemma 0A02. Then \(A^h\) is a G-ring.
Proof
Let \(\mathfrak m^h \subset A^h\) be a maximal ideal. We have to show that the map from \(A^h_{\mathfrak m^h}\) to its completion has geometrically regular fibres, see Lemma 07PT. Let \(\mathfrak m\) be the inverse image of \(\mathfrak m^h\) in \(A\). Note that \(I^h \subset \mathfrak m^h\) and hence \(I \subset \mathfrak m\) as \((A^h, I^h)\) is a henselian pair. Recall that \(A^h\) is Noetherian, \(I^h = IA^h\), and that \(A \to A^h\) induces an isomorphism on \(I\)-adic completions, see Lemma 0AGV. Then the local homomorphism of Noetherian local rings \[A_\mathfrak m \to A^h_{\mathfrak m^h}\] induces an isomorphism on completions at maximal ideals by Lemma 0AGX (details omitted). Let \(\mathfrak q^h\) be a prime of \(A^h_{\mathfrak m^h}\) lying over \(\mathfrak q \subset A_\mathfrak m\). Set \(\mathfrak q_1 = \mathfrak q^h\) and let \(\mathfrak q_2, \ldots, \mathfrak q_t\) be the other primes of \(A^h\) lying over \(\mathfrak q\), so that \(A^h \otimes_A \kappa(\mathfrak q) = \prod\nolimits_{i = 1, \ldots, t} \kappa(\mathfrak q_i)\), see Lemma 0AH1. Using that \((A^h)_{\mathfrak m^h}^\wedge = (A_\mathfrak m)^\wedge\) as discussed above we see \[\prod\nolimits_{i = 1, \ldots, t} (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q_i) = (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} (A^h_{\mathfrak m^h} \otimes_{A_{\mathfrak m}} \kappa(\mathfrak q)) = (A_{\mathfrak m})^\wedge \otimes_{A_{\mathfrak m}} \kappa(\mathfrak q)\] Hence, as one of the components, the ring \[(A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q^h)\] is geometrically regular over \(\kappa(\mathfrak q)\) by assumption on \(A\). Since \(\kappa(\mathfrak q^h)\) is separable algebraic over \(\kappa(\mathfrak q)\) it follows from Algebra, Lemma 07QH that \[(A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q^h)\] is geometrically regular over \(\kappa(\mathfrak q^h)\) as desired.
Properties of formal fibres
In this section we redo some of the arguments of Section 07GG for to be able to talk intelligently about properties of the formal fibres of Noetherian rings.
Let \(P\) be a property of ring maps \(k \to R\) where \(k\) is a field and \(R\) is Noetherian. We say \(P\) holds for the fibres of a ring homomorphism \(A \to B\) with \(B\) Noetherian if \(P\) holds for \(\kappa(\mathfrak q) \to B \otimes_A \kappa(\mathfrak q)\) for all primes \(\mathfrak q\) of \(A\). In the following we will use the following assertions
\(P(k \to R) \Rightarrow P(k' \to R \otimes_k k')\) for finitely generated field extensions \(k'/k\),
\(P(k \to R_\mathfrak p),\ \forall \mathfrak p \in \Spec(R) \Leftrightarrow P(k \to R)\),
given flat maps \(A \to B \to C\) of Noetherian rings, if the fibres of \(A \to B\) have \(P\) and \(B \to C\) is regular, then the fibres of \(A \to C\) have \(P\),
given flat maps \(A \to B \to C\) of Noetherian rings if the fibres of \(A \to C\) have \(P\) and \(B \to C\) is faithfully flat, then the fibres of \(A \to B\) have \(P\),
given \(k \to k' \to R\) with \(R\) Noetherian if \(k'/k\) is separable algebraic and \(P(k \to R)\), then \(P(k' \to R)\), and
add more here.
Given a Noetherian local ring \(A\) we say “the formal fibres of \(A\) have \(P\)” if \(P\) holds for the fibres of \(A \to A^\wedge\). We say that \(R\) is a \(P\)-ring if \(R\) is Noetherian and for all primes \(\mathfrak p\) of \(R\) the formal fibres of \(R_\mathfrak p\) have \(P\).
Lemma
Let \(R\) be a Noetherian ring. Let \(P\) be a property as above. Then \(R\) is a \(P\)-ring if and only if for every pair of primes \(\mathfrak q \subset \mathfrak p \subset R\) the \(\kappa(\mathfrak q)\)-algebra \[(R/\mathfrak q)_\mathfrak p^\wedge \otimes_{R/\mathfrak q} \kappa(\mathfrak q)\] has property \(P\).
Proof
This follows from the fact that \[R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) = (R/\mathfrak q)_\mathfrak p^\wedge \otimes_{R/\mathfrak q} \kappa(\mathfrak q)\] as algebras over \(\kappa(\mathfrak q)\).
Lemma
Let \(R \to \Lambda\) be a homomorphism of Noetherian rings. Assume \(P\) has property (B). The following are equivalent
the fibres of \(R \to \Lambda\) have \(P\),
the fibres of \(R_\mathfrak p \to \Lambda_\mathfrak q\) have \(P\) for all \(\mathfrak q \subset \Lambda\) lying over \(\mathfrak p \subset R\), and
the fibres of \(R_\mathfrak m \to \Lambda_{\mathfrak m'}\) have \(P\) for all maximal ideals \(\mathfrak m' \subset \Lambda\) lying over \(\mathfrak m\) in \(R\).
Proof
Let \(\mathfrak p \subset R\) be a prime. Then the fibre over \(\mathfrak p\) is the ring \(\Lambda \otimes_R \kappa(\mathfrak p)\) whose spectrum maps bijectively onto the subset of \(\Spec(\Lambda)\) consisting of primes \(\mathfrak q\) lying over \(\mathfrak p\), see Algebra, Remark 00E6. For such a prime \(\mathfrak q\) choose a maximal ideal \(\mathfrak q \subset \mathfrak m'\) and set \(\mathfrak m = R \cap \mathfrak m'\). Then \(\mathfrak p \subset \mathfrak m\) and we have \[(\Lambda \otimes_R \kappa(\mathfrak p))_\mathfrak q \cong (\Lambda_{\mathfrak m'} \otimes_{R_\mathfrak m} \kappa(\mathfrak p))_\mathfrak q\] as \(\kappa(\mathfrak p)\)-algebras. Thus (1), (2), and (3) are equivalent because by (B) we can check property \(P\) on local rings.
Lemma
Let \(R \to R'\) be a finite type map of Noetherian rings and let \[\xymatrix{ \mathfrak q' \ar[r] & \mathfrak p' \ar[r] & R' \\ \mathfrak q \ar[r] \ar@{-}[u] & \mathfrak p \ar[r] \ar@{-}[u] & R \ar[u] }\] be primes. Assume \(R \to R'\) is quasi-finite at \(\mathfrak p'\). Assume \(P\) satisfies (A) and (B).
If \(\kappa(\mathfrak q) \to R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q)\) has \(P\), then \(\kappa(\mathfrak q') \to (R'_{\mathfrak p'})^\wedge \otimes_{R'} \kappa(\mathfrak q')\) has \(P\).
If the formal fibres of \(R_\mathfrak p\) have \(P\), then the formal fibres of \(R'_{\mathfrak p'}\) have \(P\).
If \(R \to R'\) is quasi-finite and \(R\) is a \(P\)-ring, then \(R'\) is a \(P\)-ring.
Proof
It is clear that (1) \(\Rightarrow\) (2) \(\Rightarrow\) (3). Assume \(P\) holds for \(\kappa(\mathfrak q) \to R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q)\). By Algebra, Lemma 07NC we see that \[R_\mathfrak p^\wedge \otimes_R R' = (R'_{\mathfrak p'})^\wedge \times B\] for some \(R_\mathfrak p^\wedge\)-algebra \(B\). Hence \(R'_{\mathfrak p'} \to (R'_{\mathfrak p'})^\wedge\) is a factor of a base change of the map \(R_\mathfrak p \to R_\mathfrak p^\wedge\). It follows that \((R'_{\mathfrak p'})^\wedge \otimes_{R'} \kappa(\mathfrak q')\) is a factor of \[R_\mathfrak p^\wedge \otimes_R R' \otimes_{R'} \kappa(\mathfrak q') = R_\mathfrak p^\wedge \otimes_R \kappa(\mathfrak q) \otimes_{\kappa(\mathfrak q)} \kappa(\mathfrak q').\] Thus the result follows from the assumptions on \(P\).
Lemma
Let \(R\) be a Noetherian ring. Assume \(P\) satisfies (C) and (D). Then \(R\) is a \(P\)-ring if and only if the formal fibres of \(R_\mathfrak m\) have \(P\) for every maximal ideal \(\mathfrak m\) of \(R\).
Proof
Assume the formal fibres of \(R_\mathfrak m\) have \(P\) for all maximal ideals \(\mathfrak m\) of \(R\). Let \(\mathfrak p\) be a prime of \(R\) and choose a maximal ideal \(\mathfrak p \subset \mathfrak m\). Since \(R_\mathfrak m \to R_\mathfrak m^\wedge\) is faithfully flat we can choose a prime \(\mathfrak p'\) of \(R_\mathfrak m^\wedge\) lying over \(\mathfrak pR_\mathfrak m\). Consider the commutative diagram \[\xymatrix{ R_\mathfrak m^\wedge \ar[r] & (R_\mathfrak m^\wedge)_{\mathfrak p'} \ar[r] & (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge \\ R_\mathfrak m \ar[u] \ar[r] & R_\mathfrak p \ar[u] \ar[r] & R_\mathfrak p^\wedge \ar[u] }\] By assumption the fibres of the ring map \(R_\mathfrak m \to R_\mathfrak m^\wedge\) have \(P\). By Proposition 07PS \((R_\mathfrak m^\wedge)_{\mathfrak p'} \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) is regular. The localization \(R_\mathfrak m^\wedge \to (R_\mathfrak m^\wedge)_{\mathfrak p'}\) is regular. Hence \(R_\mathfrak m^\wedge \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) is regular by Lemma 07QI. Hence the fibres of \(R_\mathfrak m \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) have \(P\) by (C). Since \(R_\mathfrak m \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) factors through the localization \(R_\mathfrak p\), also the fibres of \(R_\mathfrak p \to (R_\mathfrak m^\wedge)_{\mathfrak p'}^\wedge\) have \(P\). Thus we may apply (D) to see that the fibres of \(R_\mathfrak p \to R_\mathfrak p^\wedge\) have \(P\).
Proposition
Let \(R\) be a \(P\)-ring where \(P\) satisfies (A), (B), (C), and (D). If \(R \to S\) is essentially of finite type then \(S\) is a \(P\)-ring.
Proof
Since being a \(P\)-ring is a property of the local rings it is clear that a localization of a \(P\)-ring is a \(P\)-ring. Conversely, if every localization at a prime is a \(P\)-ring, then the ring is a \(P\)-ring. Thus it suffices to show that \(S_\mathfrak q\) is a \(P\)-ring for every finite type \(R\)-algebra \(S\) and every prime \(\mathfrak q\) of \(S\). Writing \(S\) as a quotient of \(R[x_1, \ldots, x_n]\) we see from Lemma 0BIT that it suffices to prove that \(R[x_1, \ldots, x_n]\) is a \(P\)-ring. By induction on \(n\) it suffices to prove that \(R[x]\) is a \(P\)-ring. Let \(\mathfrak q \subset R[x]\) be a maximal ideal. By Lemma 0BIU it suffices to show that the fibres of \[R[x]_\mathfrak q \longrightarrow R[x]_\mathfrak q^\wedge\] have \(P\). If \(\mathfrak q\) lies over \(\mathfrak p \subset R\), then we may replace \(R\) by \(R_\mathfrak p\). Hence we may assume that \(R\) is a Noetherian local \(P\)-ring with maximal ideal \(\mathfrak m\) and that \(\mathfrak q \subset R[x]\) lies over \(\mathfrak m\). Note that there is a unique prime \(\mathfrak q' \subset R^\wedge[x]\) lying over \(\mathfrak q\). Consider the diagram \[\xymatrix{ R[x]_\mathfrak q^\wedge \ar[r] & (R^\wedge[x]_{\mathfrak q'})^\wedge \\ R[x]_\mathfrak q \ar[r] \ar[u] & R^\wedge[x]_{\mathfrak q'} \ar[u] }\] Since \(R\) is a \(P\)-ring the fibres of \(R[x] \to R^\wedge[x]\) have \(P\) because they are base changes of the fibres of \(R \to R^\wedge\) by a finitely generated field extension so (A) applies. Hence the fibres of the lower horizontal arrow have \(P\) for example by Lemma 0BK8. The right vertical arrow is regular because \(R^\wedge\) is a G-ring (Propositions 07PS and 07PV). It follows that the fibres of the composition \(R[x]_\mathfrak q \to (R^\wedge[x]_{\mathfrak q'})^\wedge\) have \(P\) by (C). Hence the fibres of the left vertical arrow have \(P\) by (D) and the proof is complete.
Lemma
Let \(A\) be a \(P\)-ring where \(P\) satisfies (B) and (D). Let \(I \subset A\) be an ideal and let \(A^\wedge\) be the completion of \(A\) with respect to \(I\). Then the fibres of \(A \to A^\wedge\) have \(P\).
Proof
The ring map \(A \to A^\wedge\) is flat by Algebra, Lemma 00MB. The ring \(A^\wedge\) is Noetherian by Algebra, Lemma 0316. Thus it suffices to check the third condition of Lemma 0BK8. Let \(\mathfrak m' \subset A^\wedge\) be a maximal ideal lying over \(\mathfrak m \subset A\). By Algebra, Lemma 05GI we have \(IA^\wedge \subset \mathfrak m'\). Since \(A^\wedge/IA^\wedge = A/I\) we see that \(I \subset \mathfrak m\), \(\mathfrak m/I = \mathfrak m'/IA^\wedge\), and \(A/\mathfrak m = A^\wedge/\mathfrak m'\). Since \(A^\wedge/\mathfrak m'\) is a field, we conclude that \(\mathfrak m\) is a maximal ideal as well. Then \(A_\mathfrak m \to A^\wedge_{\mathfrak m'}\) is a flat local ring homomorphism of Noetherian local rings which identifies residue fields and such that \(\mathfrak m A^\wedge_{\mathfrak m'} = \mathfrak m'A^\wedge_{\mathfrak m'}\). Thus it induces an isomorphism on complete local rings, see Lemma 0AGX. Let \((A_\mathfrak m)^\wedge\) be the completion of \(A_\mathfrak m\) with respect to its maximal ideal. The ring map \[(A^\wedge)_{\mathfrak m'} \to ((A^\wedge)_{\mathfrak m'})^\wedge = (A_\mathfrak m)^\wedge\] is faithfully flat (Algebra, Lemma 00MC). Thus we can apply (D) to the ring maps \[A_\mathfrak m \to (A^\wedge)_{\mathfrak m'} \to (A_\mathfrak m)^\wedge\] to conclude because the fibres of \(A_\mathfrak m \to (A_\mathfrak m)^\wedge\) have \(P\) as \(A\) is a \(P\)-ring.
Lemma
Let \(A\) be a \(P\)-ring where \(P\) satisfies (B), (C), (D), and (E). Let \(I \subset A\) be an ideal. Let \((A^h, I^h)\) be the henselization of the pair \((A, I)\), see Lemma 0A02. Then \(A^h\) is a \(P\)-ring.
Proof
Let \(\mathfrak m^h \subset A^h\) be a maximal ideal. We have to show that the fibres of \(A^h_{\mathfrak m^h} \to (A^h_{\mathfrak m^h})^\wedge\) have \(P\), see Lemma 0BIU. Let \(\mathfrak m\) be the inverse image of \(\mathfrak m^h\) in \(A\). Note that \(I^h \subset \mathfrak m^h\) and hence \(I \subset \mathfrak m\) as \((A^h, I^h)\) is a henselian pair. Recall that \(A^h\) is Noetherian, \(I^h = IA^h\), and that \(A \to A^h\) induces an isomorphism on \(I\)-adic completions, see Lemma 0AGV. Then the local homomorphism of Noetherian local rings \[A_\mathfrak m \to A^h_{\mathfrak m^h}\] induces an isomorphism on completions at maximal ideals by Lemma 0AGX (details omitted). Let \(\mathfrak q^h\) be a prime of \(A^h_{\mathfrak m^h}\) lying over \(\mathfrak q \subset A_\mathfrak m\). Set \(\mathfrak q_1 = \mathfrak q^h\) and let \(\mathfrak q_2, \ldots, \mathfrak q_t\) be the other primes of \(A^h\) lying over \(\mathfrak q\), so that \(A^h \otimes_A \kappa(\mathfrak q) = \prod\nolimits_{i = 1, \ldots, t} \kappa(\mathfrak q_i)\), see Lemma 0AH1. Using that \((A^h)_{\mathfrak m^h}^\wedge = (A_\mathfrak m)^\wedge\) as discussed above we see \[\prod\nolimits_{i = 1, \ldots, t} (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q_i) = (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} (A^h_{\mathfrak m^h} \otimes_{A_{\mathfrak m}} \kappa(\mathfrak q)) = (A_{\mathfrak m})^\wedge \otimes_{A_{\mathfrak m}} \kappa(\mathfrak q)\] Hence, looking at local rings and using (B), we see that \[\kappa(\mathfrak q) \longrightarrow (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q^h)\] has \(P\) as \(\kappa(\mathfrak q) \to (A_\mathfrak m)^\wedge \otimes_{A_\mathfrak m} \kappa(\mathfrak q)\) does by assumption on \(A\). Since \(\kappa(\mathfrak q^h)/\kappa(\mathfrak q)\) is separable algebraic, by (E) we find that \(\kappa(\mathfrak q^h) \to (A^h_{\mathfrak m^h})^\wedge \otimes_{A^h_{\mathfrak m^h}} \kappa(\mathfrak q^h)\) has \(P\) as desired.
Lemma
Let \(R\) be a Noetherian local ring which is a \(P\)-ring where \(P\) satisfies (B), (C), (D), and (E). Then the henselization \(R^h\) and the strict henselization \(R^{sh}\) are \(P\)-rings.
Proof
We have seen this for the henselization in Lemma 0BKA. To prove it for the strict henselization, it suffices to show that the formal fibres of \(R^{sh}\) have \(P\), see Lemma 0BIU. Let \(\mathfrak r \subset R^{sh}\) be a prime and set \(\mathfrak p = R \cap \mathfrak r\). Set \(\mathfrak r_1 = \mathfrak r\) and let \(\mathfrak r_2, \ldots, \mathfrak r_s\) be the other primes of \(R^{sh}\) lying over \(\mathfrak p\), so that \(R^{sh} \otimes_R \kappa(\mathfrak p) = \prod\nolimits_{i = 1, \ldots, s} \kappa(\mathfrak r_i)\), see Lemma 07QQ. Then we see that \[\prod\nolimits_{i = 1, \ldots, s} (R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r_i) = (R^{sh})^\wedge \otimes_{R^{sh}} (R^{sh} \otimes_R \kappa(\mathfrak p)) = (R^{sh})^\wedge \otimes_R \kappa(\mathfrak p)\] Note that \(R^\wedge \to (R^{sh})^\wedge\) is formally smooth in the \(\mathfrak m_{(R^{sh})^\wedge}\)-adic topology, see Lemma 06LJ. Hence \(R^\wedge \to (R^{sh})^\wedge\) is regular by Proposition 07PM. We conclude that property \(P\) holds for \(\kappa(\mathfrak p) \to (R^{sh})^\wedge \otimes_R \kappa(\mathfrak p)\) by (C) and our assumption on \(R\). Using property (B), using the decomposition above, and looking at local rings we conclude that property \(P\) holds for \(\kappa(\mathfrak p) \to (R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r)\). Since \(\kappa(\mathfrak r)/\kappa(\mathfrak p)\) is separable algebraic, it follows from (E) that \(P\) holds for \(\kappa(\mathfrak r) \to (R^{sh})^\wedge \otimes_{R^{sh}} \kappa(\mathfrak r)\).
Lemma
Properties (A), (B), (C), (D), and (E) hold for \(P(k \to R) =\)“\(R\) is geometrically reduced over \(k\)”.
Proof
Part (A) follows from the definition of geometrically reduced algebras (Algebra, Definition 030S). Part (B) follows too: a ring is reduced if and only if all local rings are reduced. Part (C). This follows from Lemma 07QK. Part (D). This follows from Algebra, Lemma 033F. Part (E). This follows from Algebra, Lemma 0C2Y.
Lemma
Properties (A), (B), (C), (D), and (E) hold for \(P(k \to R) =\)“\(R\) is geometrically normal over \(k\)”.
Proof
Part (A) follows from the definition of geometrically normal algebras (Algebra, Definition 0380). Part (B) follows too: a ring is normal if and only if all of its local rings are normal. Part (C). This follows from Lemma 0BFK. Part (D). This follows from Algebra, Lemma 033G. Part (E). This follows from Algebra, Lemma 0C31.
Lemma
Fix \(n \geq 1\). Properties (A), (B), (C), (D), and (E) hold for \(P(k \to R) =\)“\(R\) has \((S_n)\)”.
Proof
Let \(k \to R\) be a ring map where \(k\) is a field and \(R\) a Noetherian ring. Let \(k'/k\) be a finitely generated field extension. Then the fibres of the ring map \(R \to R \otimes_k k'\) are Cohen-Macaulay by Algebra, Lemma 045M. Hence we may apply Algebra, Lemma 0339 to the ring map \(R \to R \otimes_k k'\) to see that if \(R\) has \((S_n)\) so does \(R \otimes_k k'\). This proves (A). Part (B) follows too: a Noetherian rings has \((S_n)\) if and only if all of its local rings have \((S_n)\). Part (C). This follows from Algebra, Lemma 0339 as the fibres of a regular homomorphism are regular and in particular Cohen-Macaulay. Part (D). This follows from Algebra, Lemma 0352. Part (E). This is immediate as the condition does not refer to the ground field.
Lemma
Properties (A), (B), (C), (D), and (E) hold for \(P(k \to R) =\)“\(R\) is Cohen-Macaulay”.
Proof
Follows immediately from Lemma 0BIY and the fact that a Noetherian ring is Cohen-Macaulay if and only if it satisfies conditions \((S_n)\) for all \(n\).
Lemma
Fix \(n \geq 0\). Properties (A), (B), (C), (D), and (E) hold for \(P(k \to R) =\)“\(R \otimes_k k'\) has \((R_n)\) for all finite extensions \(k'/k\)”.
Proof
Let \(k \to R\) be a ring map where \(k\) is a field and \(R\) a Noetherian ring. Assume \(P(k \to R)\) is true. Let \(K/k\) be a finitely generated field extension. By Algebra, Lemma 030R we can find a diagram \[\xymatrix{ K \ar[r] & K' \\ k \ar[u] \ar[r] & k' \ar[u] }\] where \(k'/k\), \(K'/K\) are finite purely inseparable field extensions such that \(K'/k'\) is separable. By Algebra, Lemma 037X there exists a smooth \(k'\)-algebra \(B\) such that \(K'\) is the fraction field of \(B\). Now we can argue as follows: Step 1: \(R \otimes_k k'\) satisfies \((R_n)\) because we assumed \(P\) for \(k \to R\). Step 2: \(R \otimes_k k' \to R \otimes_k k' \otimes_{k'} B\) is a smooth ring map (Algebra, Lemma 00T4) and we conclude \(R \otimes_k k' \otimes_{k'} B\) satisfies \((R_n)\) by Algebra, Lemma 033A (and using Algebra, Lemma 00TT to see that the hypotheses are satisfied). Step 3. \(R \otimes_k k' \otimes_{k'} K' = R \otimes_k K'\) satisfies \((R_n)\) as it is a localization of a ring having \((R_n)\). Step 4. Finally \(R \otimes_k K\) satisfies \((R_n)\) by descent of \((R_n)\) along the faithfully flat ring map \(K \otimes_k R \to K' \otimes_k R\) (Algebra, Lemma 0353). This proves (A). Part (B) follows too: a Noetherian ring has \((R_n)\) if and only if all of its local rings have \((R_n)\). Part (C). This follows from Algebra, Lemma 033A as the fibres of a regular homomorphism are regular (small detail omitted). Part (D). This follows from Algebra, Lemma 0353 (small detail omitted).
Part (E). Let \(l/k\) be a separable algebraic extension of fields and let \(l \to R\) be a ring map with \(R\) Noetherian. Assume that \(k \to R\) has \(P\). We have to show that \(l \to R\) has \(P\). Let \(l'/l\) be a finite extension. First observe that there exists a finite subextension \(l/m/k\) and a finite extension \(m'/m\) such that \(l' = l \otimes_m m'\). Then \(R \otimes_l l' = R \otimes_m m'\). Hence it suffices to prove that \(m \to R\) has property \(P\), i.e., we may assume that \(l/k\) is finite. If \(l/k\) is finite, then \(l'/k\) is finite and we see that \[l' \otimes_l R = (l' \otimes_k R) \otimes_{l \otimes_k l} l\] is a localization (by Algebra, Lemma 0C2X) of the Noetherian ring \(l' \otimes_k R\) which has property \((R_n)\) by assumption \(P\) for \(k \to R\). This proves that \(l' \otimes_l R\) has property \((R_n)\) as desired.
Excellent rings
In this section we discuss Grothendieck’s notion of excellent rings. For the definitions of G-rings, J-2 rings, and universally catenary rings we refer to Definition 07GH, Definition 07P7, and Algebra, Definition 00NL.
Definition
Let \(R\) be a ring.
We say \(R\) is quasi-excellent if \(R\) is Noetherian, a G-ring, and J-2.
We say \(R\) is excellent if \(R\) is quasi-excellent and universally catenary.
Thus a Noetherian ring is quasi-excellent if it has geometrically regular formal fibres and if any finite type algebra over it has closed singular set. For such a ring to be excellent we require in addition that there exists (locally) a good dimension function. We will see later (Section 0AW1) that to be universally catenary can be formulated as a condition on the maps \(R_\mathfrak m \to R_\mathfrak m^\wedge\) for maximal ideals \(\mathfrak m\) of \(R\).
Lemma
Any localization of a finite type ring over a (quasi-)excellent ring is (quasi-)excellent.
Proof
For finite type algebras this follows from the definitions for the properties J-2 and universally catenary. For G-rings, see Proposition 07PV. We omit the proof that localization preserves (quasi-)excellency.
Proposition
The following types of rings are excellent:
fields,
Noetherian complete local rings,
\(\mathbf{Z}\),
Dedekind domains with fraction field of characteristic zero,
finite type ring extensions of any of the above.
Proof
See Propositions 07PX and 07PJ to see that these rings are G-rings and have J-2. Any Cohen-Macaulay ring is universally catenary, see Algebra, Lemma 00NM. In particular fields, Dedekind rings, and more generally regular rings are universally catenary. Via the Cohen structure theorem we see that complete local rings are universally catenary, see Algebra, Remark 032C.
The material developed above has some consequences for Nagata rings.
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring. The following are equivalent
\(A\) is Nagata, and
the formal fibres of \(A\) are geometrically reduced.
Proof
Assume (2). By Algebra, Lemma 0BI2 we have to show that if \(A \to B\) is finite, \(B\) is a domain, and \(\mathfrak m' \subset B\) is a maximal ideal, then \(B_{\mathfrak m'}\) is analytically unramified. Combining Lemmas 0BIW and 0BIU and Proposition 0BIV we see that the formal fibres of \(B_{\mathfrak m'}\) are geometrically reduced. In particular \(B_{\mathfrak m'}^\wedge \otimes_B L\) is reduced where \(L\) is the fraction field of \(B\). It follows that \(B_{\mathfrak m'}^\wedge\) is reduced, i.e., \(B_{\mathfrak m'}\) is analytically unramified.
Assume (1). Let \(\mathfrak q \subset A\) be a prime ideal and let \(K/\kappa(\mathfrak q)\) be a finite extension. We have to show that \(A^\wedge \otimes_A K\) is reduced. Let \(A/\mathfrak q \subset B \subset K\) be a local subring finite over \(A\) whose fraction field is \(K\). To construct \(B\) choose \(x_1, \ldots, x_n \in K\) which generate \(K\) over \(\kappa(\mathfrak q)\) and which satisfy monic polynomials \(P_i(T) = T^{d_i} + a_{i, 1} T^{d_i - 1} + \ldots + a_{i, d_i} = 0\) with \(a_{i, j} \in \mathfrak m\). Then let \(B\) be the \(A\)-subalgebra of \(K\) generated by \(x_1, \ldots, x_n\). (For more details see the proof of Algebra, Lemma 0BI2.) Then \[A^\wedge \otimes_A K = (A^\wedge \otimes_A B)_\mathfrak q = B^\wedge_\mathfrak q\] Since \(B^\wedge\) is reduced by Algebra, Lemma 0BI2 the proof is complete.
Lemma
A quasi-excellent ring is Nagata.
Proof
Let \(R\) be quasi-excellent. Using that a finite type algebra over \(R\) is quasi-excellent (Lemma 07QU) we see that it suffices to show that any quasi-excellent domain is N-1, see Algebra, Lemma 0351. Applying Algebra, Lemma 0333 (and using that a quasi-excellent ring is J-2) we reduce to showing that a quasi-excellent local domain \(R\) is N-1. As \(R \to R^\wedge\) is regular we see that \(R^\wedge\) is reduced by Lemma 07QK. In other words, \(R\) is analytically unramified. Hence \(R\) is N-1 by Algebra, Lemma 032Y.
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring. If \(A\) is normal and the formal fibres of \(A\) are normal (for example if \(A\) is excellent or quasi-excellent), then \(A^\wedge\) is normal.
Proof
Follows immediately from Algebra, Lemma 0C22.
Abelian categories of modules
Let \(R\) be a ring. The category \(\text{Mod}_R\) of \(R\)-modules is an abelian category. Here are some examples of subcategories of \(\text{Mod}_R\) which are abelian (we use the terminology introduced in Homology, Definition 02MO as well as Homology, Lemmas 02MP and 0754):
The category of coherent \(R\)-modules is a weak Serre subcategory of \(\text{Mod}_R\). This follows from Algebra, Lemma 05CW.
Let \(S \subset R\) be a multiplicative subset. The full subcategory consisting of \(R\)-modules \(M\) such that multiplication by \(s \in S\) is an isomorphism on \(M\) is a Serre subcategory of \(\text{Mod}_R\). This follows from Algebra, Lemma 07JY.
Let \(I \subset R\) be a finitely generated ideal. The full subcategory of \(I\)-power torsion modules is a Serre subcategory of \(\text{Mod}_R\). See Lemma 0A6K.
In some texts a torsion module is defined as a module \(M\) such that for all \(x \in M\) there exists a nonzerodivisor \(f \in R\) such that \(fx = 0\). The full subcategory of torsion modules is a Serre subcategory of \(\text{Mod}_R\).
If \(R\) is not Noetherian, then the category \(\text{Mod}^{fg}_R\) of finitely generated \(R\)-modules is not abelian. Namely, if \(I \subset R\) is a non-finitely generated ideal, then the map \(R \to R/I\) does not have a kernel in \(\text{Mod}^{fg}_R\).
If \(R\) is Noetherian, then coherent \(R\)-modules agree with finitely generated (i.e., finite) \(R\)-modules, see Algebra, Lemmas 05CY, 05CX, and 00FP. Hence \(\text{Mod}^{fg}_R\) is abelian by (1) above, but in fact,in this case the category \(\text{Mod}_R^{fg}\) is a (strong) Serre subcategory of \(\text{Mod}_R\).
Injective abelian groups
In this section we show the category of abelian groups has enough injectives. Recall that an abelian group \(M\) is divisible if and only if for every \(x \in M\) and every \(n \in \mathbf{N}\) there exists a \(y \in M\) such that \(n y = x\).
Lemma
An abelian group \(J\) is an injective object in the category of abelian groups if and only if \(J\) is divisible.
Proof
Suppose that \(J\) is not divisible. Then there exists an \(x \in J\) and \(n \in \mathbf{N}\) such that there is no \(y \in J\) with \(n y = x\). Then the morphism \(\mathbf{Z} \to J\), \(m \mapsto mx\) does not extend to \(\frac{1}{n}\mathbf{Z} \supset \mathbf{Z}\). Hence \(J\) is not injective.
Let \(A \subset B\) be abelian groups. Assume that \(J\) is a divisible abelian group. Let \(\varphi : A \to J\) be a morphism. Consider the set of homomorphisms \(\varphi' : A' \to J\) with \(A \subset A' \subset B\) and \(\varphi'|_A = \varphi\). Define \((A', \varphi') \geq (A'', \varphi'')\) if and only if \(A' \supset A''\) and \(\varphi'|_{A''} = \varphi''\). If \((A_i, \varphi_i)_{i \in I}\) is a totally ordered collection of such pairs, then we obtain a map \(\bigcup_{i \in I} A_i \to J\) defined by \(a \in A_i\) maps to \(\varphi_i(a)\). Thus Zorn’s lemma applies. To conclude we have to show that if the pair \((A', \varphi')\) is maximal then \(A' = B\). In other words, it suffices to show, given any subgroup \(A \subset B\), \(A \not = B\) and any \(\varphi : A \to J\), then we can find \(\varphi' : A' \to J\) with \(A \subset A' \subset B\) such that (a) the inclusion \(A \subset A'\) is strict, and (b) the morphism \(\varphi'\) extends \(\varphi\).
To prove this, pick \(x \in B\), \(x \not \in A\). If there exists no \(n\in \mathbf{N}\) such that \(nx \in A\), then \(A \oplus \mathbf{Z} \cong A + \mathbf{Z}x\). Hence we can extend \(\varphi\) to \(A' = A + \mathbf{Z}x\) by using \(\varphi\) on \(A\) and mapping \(x\) to zero for example. If there does exist an \(n \in \mathbf{N}\) such that \(nx \in A\), then let \(n\) be the minimal such integer. Let \(z \in J\) be an element such that \(nz = \varphi(nx)\). Define a morphism \(\tilde\varphi : A \oplus \mathbf{Z} \to J\) by \((a, m) \mapsto \varphi(a) + mz\). By our choice of \(z\) the kernel of \(\tilde \varphi\) contains the kernel of the map \(A \oplus \mathbf{Z} \to B\), \((a, m) \mapsto a + mx\). Hence \(\tilde \varphi\) factors through the image \(A' = A + \mathbf{Z}x\), and this extends the morphism \(\varphi\).
We can use this lemma to show that every abelian group can be embedded in an injective abelian group. But this is a special case of the result of the following section.
Injective modules
Some lemmas on injective modules.
Definition
Let \(R\) be a ring. An \(R\)-module \(J\) is injective if and only if the functor \(\Hom_R(-, J) : \text{Mod}_R \to \text{Mod}_R\) is an exact functor.
The functor \(\Hom_R(- , M)\) is left exact for any \(R\)-module \(M\), see Algebra, Lemma 0582. Hence the condition for \(J\) to be injective really signifies that given an injection of \(R\)-modules \(M \to M'\) the map \(\Hom_R(M', J) \to \Hom_R(M, J)\) is surjective.
Before we reformulate this in terms of \({Ext}\)-modules we discuss the relationship between \(\Ext^1_R(M, N)\) and extensions as in Homology, Section 010I.
Lemma
Let \(R\) be a ring. Let \(\mathcal{A}\) be the abelian category of \(R\)-modules. There is a canonical isomorphism \(\Ext_\mathcal{A}(M, N) = \Ext^1_R(M, N)\) compatible with the long exact sequences of Algebra, Lemmas 00LU and 065P and the \(6\)-term exact sequences of Homology, Lemma 05E2.
Proof
Omitted.
Lemma
Let \(R\) be a ring. Let \(J\) be an \(R\)-module. The following are equivalent
\(J\) is injective,
\(\Ext^1_R(M, J) = 0\) for every \(R\)-module \(M\).
Proof
Let \(0 \to M'' \to M' \to M \to 0\) be a short exact sequence of \(R\)-modules. Consider the long exact sequence \[\begin{matrix} 0 \to \Hom_R(M, J) \to \Hom_R(M', J) \to \Hom_R(M'', J) \\ \phantom{0\ } \to \Ext^1_R(M, J) \to \Ext^1_R(M', J) \to \Ext^1_R(M'', J) \to \ldots \end{matrix}\] of Algebra, Lemma 065P. Thus we see that (2) implies (1). Conversely, if \(J\) is injective then the \(\Ext\)-group is zero by Homology, Lemma 0136 and Lemma 0AUL.
Lemma
Let \(R\) be a ring. Let \(J\) be an \(R\)-module. The following are equivalent
\(J\) is injective,
\(\Ext^1_R(R/I, J) = 0\) for every ideal \(I \subset R\), and
for an ideal \(I \subset R\) and module map \(I \to J\) there exists an extension \(R \to J\).
Proof
If \(I \subset R\) is an ideal, then the short exact sequence \(0 \to I \to R \to R/I \to 0\) gives an exact sequence \[\Hom_R(R, J) \to \Hom_R(I, J) \to \Ext^1_R(R/I, J) \to 0\] by Algebra, Lemma 065P and the fact that \(\Ext^1_R(R, J) = 0\) as \(R\) is projective (Algebra, Lemma 05CF). Thus (2) and (3) are equivalent. In this proof we will show that (1) \(\Leftrightarrow\) (3) which is known as Baer’s criterion.
Assume (1). Given a module map \(I \to J\) as in (3) we find the extension \(R \to J\) because the map \(\Hom_R(R, J) \to \Hom_R(I, J)\) is surjective by definition.
Assume (3). Let \(M \subset N\) be an inclusion of \(R\)-modules. Let \(\varphi : M \to J\) be a homomorphism. We will show that \(\varphi\) extends to \(N\) which finishes the proof of the lemma. Consider the set of homomorphisms \(\varphi' : M' \to J\) with \(M \subset M' \subset N\) and \(\varphi'|_M = \varphi\). Define \((M', \varphi') \geq (M'', \varphi'')\) if and only if \(M' \supset M''\) and \(\varphi'|_{M''} = \varphi''\). If \((M_i, \varphi_i)_{i \in I}\) is a totally ordered collection of such pairs, then we obtain a map \(\bigcup_{i \in I} M_i \to J\) defined by \(a \in M_i\) maps to \(\varphi_i(a)\). Thus Zorn’s lemma applies. To conclude we have to show that if the pair \((M', \varphi')\) is maximal then \(M' = N\). In other words, it suffices to show, given any subgroup \(M \subset N\), \(M \not = N\) and any \(\varphi : M \to J\), then we can find \(\varphi' : M' \to J\) with \(M \subset M' \subset N\) such that (a) the inclusion \(M \subset M'\) is strict, and (b) the morphism \(\varphi'\) extends \(\varphi\).
To prove this, pick \(x \in N\), \(x \not \in M\). Let \(I = \{f \in R \mid fx \in M\}\). This is an ideal of \(R\). Define a homomorphism \(\psi : I \to J\) by \(f \mapsto \varphi(fx)\). Extend to a map \(\tilde\psi : R \to J\) which is possible by assumption (3). By our choice of \(I\) the kernel of \(M \oplus R \to J\), \((y, f) \mapsto \varphi(y) + \tilde\psi(f)\) contains the kernel of the map \(M \oplus R \to N\), \((y, f) \mapsto y + fx\). Hence this homomorphism factors through the image \(M' = M + Rx\) and this extends the given homomorphism as desired.
In the rest of this section we prove that there are enough injective modules over a ring \(R\). We start with the fact that \(\mathbf{Q}/\mathbf{Z}\) is an injective abelian group. This follows from Lemma 01D7.
Definition
Let \(R\) be a ring.
For any \(R\)-module \(M\) over \(R\) we denote \(M^\vee = \Hom(M, \mathbf{Q}/\mathbf{Z})\) with its natural \(R\)-module structure. We think of \(M \mapsto M^\vee\) as a contravariant functor from the category of \(R\)-modules to itself.
For any \(R\)-module \(M\) we denote \[F(M) = \bigoplus\nolimits_{m \in M} R[m]\] the free module with basis given by the elements \([m]\) with \(m \in M\). We let \(F(M)\to M\), \(\sum f_i [m_i] \mapsto \sum f_i m_i\) be the natural surjection of \(R\)-modules. We think of \(M \mapsto (F(M) \to M)\) as a functor from the category of \(R\)-modules to the category of arrows in \(R\)-modules.
Lemma
Let \(R\) be a ring. The functor \(M \mapsto M^\vee\) is exact.
Proof
This because \(\mathbf{Q}/\mathbf{Z}\) is an injective abelian group by Lemma 01D7.
There is a canonical map \(ev : M \to (M^\vee)^\vee\) given by evaluation: given \(x \in M\) we let \(ev(x) \in (M^\vee)^\vee = \Hom(M^\vee, \mathbf{Q}/\mathbf{Z})\) be the map \(\varphi \mapsto \varphi(x)\).
Lemma
For any \(R\)-module \(M\) the evaluation map \(ev : M \to (M^\vee)^\vee\) is injective.
Proof
You can check this using that \(\mathbf{Q}/\mathbf{Z}\) is an injective abelian group. Namely, if \(x \in M\) is not zero, then let \(M' \subset M\) be the cyclic group it generates. There exists a nonzero map \(M' \to \mathbf{Q}/\mathbf{Z}\) which necessarily does not annihilate \(x\). This extends to a map \(\varphi : M \to \mathbf{Q}/\mathbf{Z}\) and then \(ev(x)(\varphi) = \varphi(x) \not = 0\).
The canonical surjection \(F(M) \to M\) of \(R\)-modules turns into a canonical injection, see above, of \(R\)-modules \[(M^\vee)^\vee \longrightarrow (F(M^\vee))^\vee.\] Set \(J(M) = (F(M^\vee))^\vee\). The composition of \(ev\) with this the displayed map gives \(M \to J(M)\) functorially in \(M\).
Lemma
Let \(R\) be a ring. For every \(R\)-module \(M\) the \(R\)-module \(J(M)\) is injective.
Proof
Note that \(J(M) \cong \prod_{\varphi \in M^\vee} R^\vee\) as an \(R\)-module. As the product of injective modules is injective, it suffices to show that \(R^\vee\) is injective. For this we use that \[\Hom_R(N, R^\vee) = \Hom_R(N, \Hom_{\mathbf{Z}}(R, \mathbf{Q}/\mathbf{Z})) = N^\vee\] and the fact that \((-)^\vee\) is an exact functor by Lemma 01DA.
Lemma
Let \(R\) be a ring. The construction above defines a covariant functor \(M \mapsto (M \to J(M))\) from the category of \(R\)-modules to the category of arrows of \(R\)-modules such that for every module \(M\) the output \(M \to J(M)\) is an injective map of \(M\) into an injective \(R\)-module \(J(M)\).
Proof
Follows from the above.
In particular, for any map of \(R\)-modules \(M \to N\) there is an associated morphism \(J(M) \to J(N)\) making the following diagram commute: \[\xymatrix{ M \ar[d] \ar[r] & N \ar[d] \\ J(M) \ar[r] & J(N) }\] This is the kind of construction we would like to have in general. In Homology, Section 0134 we introduced terminology to express this. Namely, we say this means that the category of \(R\)-modules has functorial injective embeddings.
Derived categories of modules
In this section we put some generalities concerning the derived category of modules over a ring.
Let \(A\) be a ring. The category of \(A\)-modules is denoted \(\text{Mod}_A\). We will use the symbol \(K(A)\) to denote the homotopy category of complexes of \(A\)-modules, i.e., we set \(K(A) = K(\text{Mod}_A)\) as a category, see Derived Categories, Section 05RN. The bounded versions are \(K^+(A)\), \(K^-(A)\), and \(K^b(A)\). We view \(K(A)\) as a triangulated category as in Derived Categories, Section 014P. The derived category of \(A\), denoted \(D(A)\), is the category obtained from \(K(A)\) by inverting quasi-isomorphisms, i.e., we set \(D(A) = D(\text{Mod}_A)\), see Derived Categories, Section 05RR7. The bounded versions are \(D^+(A)\), \(D^-(A)\), and \(D^b(A)\).
Let \(A\) be a ring. The category of \(A\)-modules has products and products are exact. The category of \(A\)-modules has enough injectives by Lemma 01DD. Hence every complex of \(A\)-modules is quasi-isomorphic to a K-injective complex (Derived Categories, Lemma 090Y). It follows that \(D(A)\) has countable products (Derived Categories, Lemma 07KC) and in fact arbitrary products (Injectives, Lemma 07D9). This implies that every inverse system of objects of \(D(A)\) has a derived limit (well defined up to isomorphism), see Derived Categories, Section 08TB.
Lemma
Let \(R \to S\) be a flat ring map. If \(I^\bullet\) is a K-injective complex of \(S\)-modules, then \(I^\bullet\) is K-injective as a complex of \(R\)-modules.
Proof
This is true because \(\Hom_{K(R)}(M^\bullet, I^\bullet) = \Hom_{K(S)}(M^\bullet \otimes_R S, I^\bullet)\) by Algebra, Lemma 05DQ and the fact that tensoring with \(S\) is exact.
Lemma
Let \(R \to S\) be an epimorphism of rings. Let \(I^\bullet\) be a complex of \(S\)-modules. If \(I^\bullet\) is K-injective as a complex of \(R\)-modules, then \(I^\bullet\) is a K-injective complex of \(S\)-modules.
Proof
This is true because \(\Hom_{K(R)}(N^\bullet, I^\bullet) = \Hom_{K(S)}(N^\bullet, I^\bullet)\) for any complex of \(S\)-modules \(N^\bullet\), see Algebra, Lemma 08YS.
Lemma
Let \(A \to B\) be a ring map. If \(I^\bullet\) is a K-injective complex of \(A\)-modules, then \(\Hom_A(B, I^\bullet)\) is a K-injective complex of \(B\)-modules.
Proof
This is true because \(\Hom_{K(B)}(N^\bullet, \Hom_A(B, I^\bullet)) = \Hom_{K(A)}(N^\bullet, I^\bullet)\) by Algebra, Lemma 08YP.
Computing Tor
Let \(R\) be a ring. We denote \(D(R)\) the derived category of the abelian category \(\text{Mod}_R\) of \(R\)-modules. Note that \(\text{Mod}_R\) has enough projectives as every free \(R\)-module is projective. Thus we can define the left derived functors of any additive functor from \(\text{Mod}_R\) to any abelian category.
This applies in particular to the functor \(- \otimes_R M : \text{Mod}_R \to \text{Mod}_R\) whose left derived functors are the Tor functors \(\text{Tor}_i^R(-, M)\), see Algebra, Section 00LY. There is also a total left derived functor [064G]\[\begin{equation} -\otimes_R^{\mathbf{L}} M : D^{-}(R) \longrightarrow D^{-}(R) \end{equation}\] which is denoted \(-\otimes_R^{\mathbf{L}} M\). Its satellites are the Tor modules, i.e., we have \[H^{-p}(N \otimes_R^{\mathbf{L}} M) = \text{Tor}_p^R(N, M).\]
A special situation occurs when we consider the tensor product with an \(R\)-algebra \(A\). In this case we think of \(- \otimes_R A\) as a functor from \(\text{Mod}_R\) to \(\text{Mod}_A\). Hence the total left derived functor [064H]\[\begin{equation} -\otimes_R^{\mathbf{L}} A : D^{-}(R) \longrightarrow D^{-}(A) \end{equation}\] which is denoted \(-\otimes_R^{\mathbf{L}} A\). Its satellites are the tor groups, i.e., we have \[H^{-p}(N \otimes_R^{\mathbf{L}} A) = \text{Tor}_p^R(N, A).\] In particular these Tor groups naturally have the structure of \(A\)-modules.
We will generalize the material in this section to unbounded complexes in the next few sections.
Tensor products of complexes
Let \(R\) be a ring. The category \(\text{Comp}(R)\) of complexes of \(R\)-modules has a symmetric monoidal structure. Namely, suppose that we have two complexes of \(R\)-modules \(L^\bullet\) and \(M^\bullet\). Using Homology, Example 0A5J and Homology, Definition 012Z we obtain a third complex of \(R\)-modules, namely \[\text{Tot}(L^\bullet \otimes_R M^\bullet)\] Clearly this construction is functorial in both \(L^\bullet\) and \(M^\bullet\). The associativity constraint will be the canonical isomorphism of complexes \[\text{Tot}(\text{Tot}(K^\bullet \otimes_R L^\bullet) \otimes_R M^\bullet) \longrightarrow \text{Tot}(K^\bullet \otimes_R \text{Tot}(L^\bullet \otimes_R M^\bullet))\] constructed in Homology, Remark 08BI from the triple complex \(K^\bullet \otimes_R L^\bullet \otimes_R M^\bullet\). The commutativity constraint is the canonical isomorphism \[\text{Tot}(L^\bullet \otimes_R M^\bullet) \to \text{Tot}(M^\bullet \otimes_R L^\bullet)\] which uses the sign \((-1)^{pq}\) on the summand \(L^p \otimes_R M^q\). To see that it is a map of complexes we compute for \(x \in L^p\) and \(y \in M^q\) that \[\text{d}(x \otimes y) = \text{d}_L(x) \otimes y + (-1)^px \otimes \text{d}_M(y)\] Our rule says the right hand side is mapped to \[(-1)^{(p + 1)q}y \otimes \text{d}_L(x) + (-1)^{p + p(q + 1)} \text{d}_M(y) \otimes x\] On the other hand, we see that \[\text{d}((-1)^{pq}y \otimes x) = (-1)^{pq} \text{d}_M(y) \otimes x + (-1)^{pq + q} y \otimes \text{d}_L(x)\] These two expressions agree by inspection as desired.
Lemma
Let \(R\) be a ring. The category \(\text{Comp}(R)\) of complexes of \(R\)-modules endowed with the functor \((L^\bullet, M^\bullet) \mapsto \text{Tot}(L^\bullet \otimes_R M^\bullet)\) and associativity and commutativity constraints as above is a symmetric monoidal category.
Proof
Omitted. Hints: as unit \(\mathbf{1}\) we take the complex having \(R\) in degree \(0\) and zero in other degrees with obvious isomorphisms \(\text{Tot}(\mathbf{1} \otimes_R M^\bullet) = M^\bullet\) and \(\text{Tot}(K^\bullet \otimes_R \mathbf{1}) = K^\bullet\). to prove the lemma you have to check the commutativity of various diagrams, see Categories, Definitions 0FFK and 0FFW. The verifications are straightforward in each case.
Lemma
Let \(R\) be a ring. Let \(P^\bullet\) be a complex of \(R\)-modules. Let \(\alpha, \beta : L^\bullet \to M^\bullet\) be homotopic maps of complexes. Then \(\alpha\) and \(\beta\) induce homotopic maps \[\text{Tot}(\alpha \otimes \text{id}_P), \text{Tot}(\beta \otimes \text{id}_P) : \text{Tot}(L^\bullet \otimes_R P^\bullet) \longrightarrow \text{Tot}(M^\bullet \otimes_R P^\bullet).\] In particular the construction \(L^\bullet \mapsto \text{Tot}(L^\bullet \otimes_R P^\bullet)\) defines an endo-functor of the homotopy category of complexes.
Proof
Say \(\alpha = \beta + dh + hd\) for some homotopy \(h\) defined by \(h^n : L^n \to M^{n - 1}\). Set \[H^n = \bigoplus\nolimits_{a + b = n} h^a \otimes \text{id}_{P^b} : \bigoplus\nolimits_{a + b = n} L^a \otimes_R P^b \longrightarrow \bigoplus\nolimits_{a + b = n} M^{a - 1} \otimes_R P^b\] Then a straightforward computation shows that \[\text{Tot}(\alpha \otimes \text{id}_P) = \text{Tot}(\beta \otimes \text{id}_P) + dH + Hd\] as maps \(\text{Tot}(L^\bullet \otimes_R P^\bullet) \to \text{Tot}(M^\bullet \otimes_R P^\bullet)\).
Lemma
Let \(R\) be a ring. The homotopy category \(K(R)\) of complexes of \(R\)-modules endowed with the functor \((L^\bullet, M^\bullet) \mapsto \text{Tot}(L^\bullet \otimes_R M^\bullet)\) and associativity and commutativity constraints as above is a symmetric monoidal category.
Proof
Lemma
Let \(R\) be a ring. Let \(P^\bullet\) be a complex of \(R\)-modules. The functors \[K(R) \longrightarrow K(R), \quad L^\bullet \longmapsto \text{Tot}(P^\bullet \otimes_R L^\bullet)\] and \[K(R) \longrightarrow K(R), \quad L^\bullet \longmapsto \text{Tot}(L^\bullet \otimes_R P^\bullet)\] are exact functors of triangulated categories.
Proof
This follows from Derived Categories, Remark 0G6E.
Derived tensor product
We can construct the derived tensor product in greater generality. In fact, it turns out that the boundedness assumptions are not necessary, provided we choose K-flat resolutions.
Definition
Let \(R\) be a ring. A complex \(K^\bullet\) is called K-flat if for every acyclic complex \(M^\bullet\) the total complex \(\text{Tot}(M^\bullet \otimes_R K^\bullet)\) is acyclic.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a K-flat complex. Then the functor \[K(R) \longrightarrow K(R), \quad L^\bullet \longmapsto \text{Tot}(L^\bullet \otimes_R K^\bullet)\] transforms quasi-isomorphisms into quasi-isomorphisms.
Proof
Follows from Lemma 064J and the fact that quasi-isomorphisms in \(K(R)\) are characterized by having acyclic cones.
Lemma
Let \(R \to R'\) be a ring map. If \(K^\bullet\) is a K-flat complex of \(R\)-modules, then \(K^\bullet \otimes_R R'\) is a K-flat complex of \(R'\)-modules.
Proof
Follows from the definitions and the fact that \((K^\bullet \otimes_R R') \otimes_{R'} L^\bullet = K^\bullet \otimes_R L^\bullet\) for any complex \(L^\bullet\) of \(R'\)-modules.
Lemma
Let \(R\) be a ring. If \(K^\bullet\), \(L^\bullet\) are K-flat complexes of \(R\)-modules, then \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\) is a K-flat complex of \(R\)-modules.
Proof
Follows from the isomorphism \[\text{Tot}(M^\bullet \otimes_R \text{Tot}(K^\bullet \otimes_R L^\bullet)) = \text{Tot}(\text{Tot}(M^\bullet \otimes_R K^\bullet) \otimes_R L^\bullet)\] and the definition.
Lemma
Let \(R\) be a ring. Let \((K_1^\bullet, K_2^\bullet, K_3^\bullet)\) be a distinguished triangle in \(K(R)\). If two out of three of \(K_i^\bullet\) are K-flat, so is the third.
Proof
Follows from Lemma 064J and the fact that in a distinguished triangle in \(K(R)\) if two out of three are acyclic, so is the third.
Lemma
Let \(R\) be a ring. Let \(0 \to K_1^\bullet \to K_2^\bullet \to K_3^\bullet \to 0\) be a short exact sequence of complexes. If \(K_3^n\) is flat for all \(n \in \mathbf{Z}\) and two out of three of \(K_i^\bullet\) are K-flat, so is the third.
Proof
Let \(L^\bullet\) be a complex of \(R\)-modules. Then \[0 \to \text{Tot}(L^\bullet \otimes_R K_1^\bullet) \to \text{Tot}(L^\bullet \otimes_R K_2^\bullet) \to \text{Tot}(L^\bullet \otimes_R K_3^\bullet) \to 0\] is a short exact sequence of complexes. Namely, for each \(n, m\) the sequence of modules \(0 \to L^n \otimes_R K_1^m \to L^n \otimes_R K_2^m \to L^n \otimes_R K_3^m \to 0\) is exact by Algebra, Lemma 00HL and the sequence of complexes is a direct sum of these. Thus the lemma follows from this and the fact that in a short exact sequence of complexes if two out of three are acyclic, so is the third.
Lemma
Let \(R\) be a ring. Let \(P^\bullet\) be a bounded above complex of flat \(R\)-modules. Then \(P^\bullet\) is K-flat.
Proof
Let \(L^\bullet\) be an acyclic complex of \(R\)-modules. Let \(\xi \in H^n(\text{Tot}(L^\bullet \otimes_R P^\bullet))\). We have to show that \(\xi = 0\). Since \(\text{Tot}^n(L^\bullet \otimes_R P^\bullet)\) is a direct sum with terms \(L^a \otimes_R P^b\) we see that \(\xi\) comes from an element in \(H^n(\text{Tot}(\tau_{\leq m}L^\bullet \otimes_R P^\bullet))\) for some \(m \in \mathbf{Z}\). Since \(\tau_{\leq m}L^\bullet\) is also acyclic we may replace \(L^\bullet\) by \(\tau_{\leq m}L^\bullet\). Hence we may assume that \(L^\bullet\) is bounded above. In this case the spectral sequence of Homology, Lemma 0132 has \[{}'E_1^{p, q} = H^p(L^\bullet \otimes_R P^q)\] which is zero as \(P^q\) is flat and \(L^\bullet\) acyclic. Hence \(H^*(\text{Tot}(L^\bullet \otimes_R P^\bullet)) = 0\).
In the following lemma by a colimit of a system of complexes we mean the termwise colimit.
Lemma
Let \(R\) be a ring. Let \(K_1^\bullet \to K_2^\bullet \to \ldots\) be a system of K-flat complexes. Then \(\colim_i K_i^\bullet\) is K-flat. More generally any filtered colimit of K-flat complexes is K-flat.
Proof
Because we are taking termwise colimits we have \[\colim_i \text{Tot}(M^\bullet \otimes_R K_i^\bullet) = \text{Tot}(M^\bullet \otimes_R \colim_i K_i^\bullet)\] by Algebra, Lemma 00DD. Hence the lemma follows from the fact that filtered colimits are exact, see Algebra, Lemma 00DB.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. If \(K^\bullet \otimes_R M\) is acyclic for all finitely presented \(R\)-modules \(M\), then \(K^\bullet\) is K-flat.
Proof
We will use repeatedly that tensor product commute with colimits (Algebra, Lemma 00DD). Thus we see that \(K^\bullet \otimes_R M\) is acyclic for any \(R\)-module \(M\), because any \(R\)-module is a filtered colimit of finitely presented \(R\)-modules \(M\), see Algebra, Lemma 00HA. Let \(M^\bullet\) be an acyclic complex of \(R\)-modules. We have to show that \(\text{Tot}(M^\bullet \otimes_R K^\bullet)\) is acyclic. Since \(M^\bullet = \colim \tau_{\leq n} M^\bullet\) (termwise colimit) we have \[\text{Tot}(M^\bullet \otimes_R K^\bullet) = \colim \text{Tot}(\tau_{\leq n} M^\bullet \otimes_R K^\bullet)\] with truncations as in Homology, Section 0118. As filtered colimits are exact (Algebra, Lemma 00DB) we may replace \(M^\bullet\) by \(\tau_{\leq n}M^\bullet\) and assume that \(M^\bullet\) is bounded above. In the bounded above case, we can write \(M^\bullet = \colim \sigma_{\geq -n} M^\bullet\) where the complexes \(\sigma_{\geq -n} M^\bullet\) are bounded but possibly no longer acyclic. Arguing as above we reduce to the case where \(M^\bullet\) is a bounded complex. Finally, for a bounded complex \(M^a \to \ldots \to M^b\) we can argue by induction on the length \(b - a\) of the complex. The case \(b - a = 1\) we have seen above. For \(b - a > 1\) we consider the split short exact sequence of complexes \[0 \to \sigma_{\geq a + 1}M^\bullet \to M^\bullet \to M^a[-a] \to 0\] and we apply Lemma 064J to do the induction step. Some details omitted.
Lemma
Let \(R\) be a ring. For any complex \(M^\bullet\) there exists a K-flat complex \(K^\bullet\) whose terms are flat \(R\)-modules and a quasi-isomorphism \(K^\bullet \to M^\bullet\) which is termwise surjective.
Proof
Let \(\mathcal{P} \subset \Ob(\text{Mod}_R)\) be the class of flat \(R\)-modules. By Derived Categories, Lemma 06XX there exists a system \(K_1^\bullet \to K_2^\bullet \to \ldots\) and a diagram \[\xymatrix{ K_1^\bullet \ar[d] \ar[r] & K_2^\bullet \ar[d] \ar[r] & \ldots \\ \tau_{\leq 1}M^\bullet \ar[r] & \tau_{\leq 2}M^\bullet \ar[r] & \ldots }\] with the properties (1), (2), (3) listed in that lemma. These properties imply each complex \(K_i^\bullet\) is a bounded above complex of flat modules. Hence \(K_i^\bullet\) is K-flat by Lemma 064K. The induced map \(\colim_i K_i^\bullet \to M^\bullet\) is a quasi-isomorphism and termwise surjective by construction. The complex \(\colim_i K_i^\bullet\) is K-flat by Lemma 06Y3. The terms \(\colim K_i^n\) are flat because filtered colimits of flat modules are flat, see Algebra, Lemma 05UT.
Remark
In fact, we can do better than Lemma 06Y4. Namely, we can find a quasi-isomorphism \(P^\bullet \to M^\bullet\) where \(P^\bullet\) is a complex of \(R\)-modules endowed with a filtration \[0 = F_{-1}P^\bullet \subset F_0P^\bullet \subset F_1P^\bullet \subset \ldots \subset P^\bullet\] by subcomplexes such that
\(P^\bullet = \bigcup F_pP^\bullet\),
the inclusions \(F_iP^\bullet \to F_{i + 1}P^\bullet\) are termwise split injections,
the quotients \(F_{i + 1}P^\bullet/F_iP^\bullet\) are isomorphic to direct sums of shifts \(R[k]\) (as complexes, so differentials are zero).
This will be shown in Differential Graded Algebra, Lemma 09KP (or you can argue as in the proof of Lemma 0H7X). Moreover, given such a complex we obtain a distinguished triangle \[\bigoplus F_iP^\bullet \to \bigoplus F_iP^\bullet \to M^\bullet \to \bigoplus F_iP^\bullet[1]\] in \(D(R)\). Using this we can sometimes reduce statements about general complexes to statements about \(R[k]\) (this of course only works if the statement is preserved under taking direct sums). More precisely, let \(T\) be a property of objects of \(D(R)\). Suppose that
if \(K_i \in D(R)\), \(i \in I\) is a family of objects with \(T(K_i)\) for all \(i \in I\), then \(T(\bigoplus K_i)\),
if \(K \to L \to M \to K[1]\) is a distinguished triangle and \(T\) holds for two, then \(T\) holds for the third object,
\(T(R[k])\) holds for all \(k\).
Then \(T\) holds for all objects of \(D(R)\).
Lemma
Let \(R\) be a ring. Let \(\alpha : P^\bullet \to Q^\bullet\) be a quasi-isomorphism of K-flat complexes of \(R\)-modules. For every complex \(L^\bullet\) of \(R\)-modules the induced map \[\text{Tot}(\text{id}_L \otimes \alpha) : \text{Tot}(L^\bullet \otimes_R P^\bullet) \longrightarrow \text{Tot}(L^\bullet \otimes_R Q^\bullet)\] is a quasi-isomorphism.
Proof
Choose a quasi-isomorphism \(K^\bullet \to L^\bullet\) with \(K^\bullet\) a K-flat complex, see Lemma 06Y4. Consider the commutative diagram \[\xymatrix{ \text{Tot}(K^\bullet \otimes_R P^\bullet) \ar[r] \ar[d] & \text{Tot}(K^\bullet \otimes_R Q^\bullet) \ar[d] \\ \text{Tot}(L^\bullet \otimes_R P^\bullet) \ar[r] & \text{Tot}(L^\bullet \otimes_R Q^\bullet) }\] The result follows as by Lemma 06Y0 the vertical arrows and the top horizontal arrow are quasi-isomorphisms.
Let \(R\) be a ring. Let \(M^\bullet\) be an object of \(D(R)\). Choose a K-flat resolution \(K^\bullet \to M^\bullet\), see Lemma 06Y4. By Lemmas 064I and 064J we obtain an exact functor of triangulated categories \[K(R) \longrightarrow K(R), \quad L^\bullet \longmapsto \text{Tot}(L^\bullet \otimes_R K^\bullet)\] By Lemma 06Y0 this functor induces a functor \(D(R) \to D(R)\) simply because \(D(R)\) is the localization of \(K(R)\) at quasi-isomorphism. As the category of \(K\)-flat resolutions of \(M^\bullet\) is cofiltered and as we have Lemma 064L the resulting functor (up to isomorphism) does not depend on the choice of \(K^\bullet\).
Definition
Let \(R\) be a ring. Let \(M^\bullet\) be an object of \(D(R)\). The derived tensor product \[- \otimes_R^{\mathbf{L}} M^\bullet : D(R) \longrightarrow D(R)\] is the exact functor of triangulated categories described above.
This functor extends the functor (064G). It is clear from our explicit constructions that there is an isomorphism (involving a choice of signs, see below) \[M^\bullet \otimes_R^{\mathbf{L}} L^\bullet \cong L^\bullet \otimes_R^{\mathbf{L}} M^\bullet\] whenever both \(L^\bullet\) and \(M^\bullet\) are in \(D(R)\). Hence when we write \(M^\bullet \otimes_R^{\mathbf{L}} L^\bullet\) we will usually be agnostic about which variable we are using to define the derived tensor product with.
Lemma
Let \(R\) be a ring. Let \(K^\bullet, L^\bullet\) be complexes of \(R\)-modules. There is a canonical isomorphism \[K^\bullet \otimes_R^\mathbf{L} L^\bullet \longrightarrow L^\bullet \otimes_R^\mathbf{L} K^\bullet\] functorial in both complexes which uses a sign of \((-1)^{pq}\) for the map \(K^p \otimes_R L^q \to L^q \otimes_R K^p\) (see proof for explanation).
Proof
We may and do replace the complexes by K-flat complexes \(K^\bullet\) and \(L^\bullet\) and then we use the commutativity constraint discussed in Section 0GWN.
Lemma
Let \(R\) be a ring. Let \(K^\bullet, L^\bullet, M^\bullet\) be complexes of \(R\)-modules. There is a canonical isomorphism \[(K^\bullet \otimes_R^\mathbf{L} L^\bullet) \otimes_R^\mathbf{L} M^\bullet = K^\bullet \otimes_R^\mathbf{L} (L^\bullet \otimes_R^\mathbf{L} M^\bullet)\] functorial in all three complexes.
Proof
Replace the complexes by K-flat complexes and use the associativity constraint in Section 0GWN.
Lemma
Let \(R\) be a ring. Let \(a : K^\bullet \to L^\bullet\) be a map of complexes of \(R\)-modules. If \(K^\bullet\) is K-flat, then there exist a complex \(N^\bullet\) and maps of complexes \(b : K^\bullet \to N^\bullet\) and \(c : N^\bullet \to L^\bullet\) such that
\(N^\bullet\) is K-flat,
\(c\) is a quasi-isomorphism,
\(a\) is homotopic to \(c \circ b\).
If the terms of \(K^\bullet\) are flat, then we may choose \(N^\bullet\), \(b\), and \(c\) such that the same is true for \(N^\bullet\).
Proof
We will use that the homotopy category \(K(R)\) is a triangulated category, see Derived Categories, Proposition 014S. Choose a distinguished triangle \(K^\bullet \to L^\bullet \to C^\bullet \to K^\bullet[1]\). Choose a quasi-isomorphism \(M^\bullet \to C^\bullet\) with \(M^\bullet\) K-flat with flat terms, see Lemma 06Y4. By the axioms of triangulated categories, we may fit the composition \(M^\bullet \to C^\bullet \to K^\bullet[1]\) into a distinguished triangle \(K^\bullet \to N^\bullet \to M^\bullet \to K^\bullet[1]\). By Lemma 06Y2 we see that \(N^\bullet\) is K-flat. Again using the axioms of triangulated categories, we can choose a map \(N^\bullet \to L^\bullet\) fitting into the following morphism of distinguished triangles \[\xymatrix{ K^\bullet \ar[r] \ar[d] & N^\bullet \ar[r] \ar[d] & M^\bullet \ar[r] \ar[d] & K^\bullet[1] \ar[d] \\ K^\bullet \ar[r] & L^\bullet \ar[r] & C^\bullet \ar[r] & K^\bullet[1] }\] Since two out of three of the arrows are quasi-isomorphisms, so is the third arrow \(N^\bullet \to L^\bullet\) by the long exact sequences of cohomology associated to these distinguished triangles (or you can look at the image of this diagram in \(D(R)\) and use Derived Categories, Lemma 014A if you like). This finishes the proof of (1), (2), and (3). To prove the final assertion, we may choose \(N^\bullet\) such that \(N^n \cong M^n \oplus K^n\), see Derived Categories, Lemma 0G6C. Hence we get the desired flatness if the terms of \(K^\bullet\) are flat.
Derived change of rings
Let \(R \to A\) be a ring map. Let \(N^\bullet\) be a complex of \(A\)-modules. We can also use K-flat resolutions to define a functor \[- \otimes_R^{\mathbf{L}} N^\bullet : D(R) \to D(A)\] as the left derived functor of the functor \(K(R) \to K(A)\), \(M^\bullet \mapsto \text{Tot}(M^\bullet \otimes_R N^\bullet)\). In particular, taking \(N^\bullet = A[0]\) we obtain a derived base change functor \[- \otimes_R^{\mathbf{L}} A : D(R) \to D(A)\] extending the functor (064H). Namely, for every complex of \(R\)-modules \(M^\bullet\) we can choose a K-flat resolution \(K^\bullet \to M^\bullet\) and set \[M^\bullet \otimes_R^{\mathbf{L}} N^\bullet = \text{Tot}(K^\bullet \otimes_R N^\bullet).\] You can use Lemmas 06Y4 and 064L to see that this is well defined. However, to cross all the t’s and dot all the i’s it is perhaps more convenient to use some general theory.
Lemma
The construction above is independent of choices and defines an exact functor of triangulated categories \(- \otimes_R^\mathbf{L} N^\bullet : D(R) \to D(A)\). There is a functorial isomorphism \[E^\bullet \otimes_R^\mathbf{L} N^\bullet = (E^\bullet \otimes_R^\mathbf{L} A) \otimes_A^\mathbf{L} N^\bullet\] for \(E^\bullet\) in \(D(R)\).
Proof
To prove the existence of the derived functor \(- \otimes_R^\mathbf{L} N^\bullet\) we use the general theory developed in Derived Categories, Section 05S7. Set \(\mathcal{D} = K(R)\) and \(\mathcal{D}' = D(A)\). Let us write \(F : \mathcal{D} \to \mathcal{D}'\) the exact functor of triangulated categories defined by the rule \(F(M^\bullet) = \text{Tot}(M^\bullet \otimes_R N^\bullet)\). To prove the stated properties of \(F\) use Lemmas 064I and 064J. We let \(S\) be the set of quasi-isomorphisms in \(\mathcal{D} = K(R)\). This gives a situation as in Derived Categories, Situation 05S8 so that Derived Categories, Definition 05S9 applies. We claim that \(LF\) is everywhere defined. This follows from Derived Categories, Lemma 06XN with \(\mathcal{P} \subset \Ob(\mathcal{D})\) the collection of K-flat complexes: (1) follows from Lemma 06Y4 and (2) follows from Lemma 064L. Thus we obtain a derived functor \[LF : D(R) = S^{-1}\mathcal{D} \longrightarrow \mathcal{D}' = D(A)\] see Derived Categories, Equation (05SW). Finally, Derived Categories, Lemma 06XN guarantees that \(LF(K^\bullet) = F(K^\bullet) = \text{Tot}(K^\bullet \otimes_R N^\bullet)\) when \(K^\bullet\) is K-flat, i.e., \(LF\) is indeed computed in the way described above. Moreover, by Lemma 06Y1 the complex \(K^\bullet \otimes_R A\) is a K-flat complex of \(A\)-modules. Hence \[(K^\bullet \otimes_R^\mathbf{L} A) \otimes_A^\mathbf{L} N^\bullet = \text{Tot}((K^\bullet \otimes_R A) \otimes_A N^\bullet) = \text{Tot}(K^\bullet \otimes_A N^\bullet) = K^\bullet \otimes_A^\mathbf{L} N^\bullet\] which proves the final statement of the lemma.
Lemma
Let \(R \to A\) be a ring map. Let \(f : L^\bullet \to N^\bullet\) be a map of complexes of \(A\)-modules. Then \(f\) induces a transformation of functors \[1 \otimes f : - \otimes_A^\mathbf{L} L^\bullet \longrightarrow - \otimes_A^\mathbf{L} N^\bullet\] If \(f\) is a quasi-isomorphism, then \(1 \otimes f\) is an isomorphism of functors.
Proof
Since the functors are computing by evaluating on K-flat complexes \(K^\bullet\) we can simply use the functoriality \[\text{Tot}(K^\bullet \otimes_R L^\bullet) \to \text{Tot}(K^\bullet \otimes_R N^\bullet)\] to define the transformation. The last statement follows from Lemma 06Y0.
Lemma
Let \(R \to A\) be a ring map. The functor \(D(R) \to D(A)\), \(E \mapsto E \otimes_R^\mathbf{L} A\) of Lemma 06Y6 is left adjoint to the restriction functor \(D(A) \to D(R)\).
Proof
This follows from Derived Categories, Lemma 0FND and the fact that \(- \otimes_R A\) and restriction are adjoint by Algebra, Lemma 05DQ.
Remark
Let \(R \to A\) be a ring map, and let \(N\) and \(N'\) be \(A\)-modules. Denote \(N_R\) and \(N'_R\) the restriction of \(N\) and \(N'\) to \(R\)-modules, see Algebra, Section 05G3. In this situation, the objects \(N_R \otimes_R^\mathbf{L} N'\) and \(N \otimes_R^\mathbf{L} N'_R\) of \(D(A)\) are in general not isomorphic! In other words, one has to pay careful attention as to which of the two sides is being used to provide the \(A\)-module structure.
For a specific example, set \(R = k[x, y]\), \(A = R/(xy)\), \(N = R/(x)\) and \(N' = A = R/(xy)\). The resolution \(0 \to R \xrightarrow{xy} R \to N'_R \to 0\) shows that \(N \otimes_R^\mathbf{L} N'_R = N[1] \oplus N\) in \(D(A)\). The resolution \(0 \to R \xrightarrow{x} R \to N_R \to 0\) shows that \(N_R \otimes_R^\mathbf{L} N'\) is represented by the complex \(A \xrightarrow{x} A\). To see these two complexes are not isomorphic, one can show that the second complex is not isomorphic in \(D(A)\) to the direct sum of its cohomology groups, or one can show that the first complex is not a perfect object of \(D(A)\) whereas the second one is. Some details omitted.
Lemma
Let \(A \to B \to C\) be ring maps. Let \(N^\bullet\) be a complex of \(B\)-modules and \(K^\bullet\) a complex of \(C\)-modules. The compositions of the functors \[D(A) \xrightarrow{- \otimes_A^\mathbf{L} N^\bullet} D(B) \xrightarrow{- \otimes_B^\mathbf{L} K^\bullet} D(C)\] is the functor \(- \otimes_A^\mathbf{L} (N^\bullet \otimes_B^\mathbf{L} K^\bullet) : D(A) \to D(C)\). If \(M\), \(N\), \(K\) are modules over \(A\), \(B\), \(C\), then we have \[(M \otimes_A^\mathbf{L} N) \otimes_B^\mathbf{L} K = M \otimes_A^\mathbf{L} (N \otimes_B^\mathbf{L} K) = (M \otimes_A^\mathbf{L} C) \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} K)\] in \(D(C)\). We also have a canonical isomorphism \[(M \otimes_A^\mathbf{L} N) \otimes_B^\mathbf{L} K \longrightarrow (M \otimes_A^\mathbf{L} K) \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} C)\] using signs. Similar results holds for complexes.
Proof
Choose a K-flat complex \(P^\bullet\) of \(B\)-modules and a quasi-isomorphism \(P^\bullet \to N^\bullet\) (Lemma 06Y4). Let \(M^\bullet\) be a K-flat complex of \(A\)-modules representing an arbitrary object of \(D(A)\). Then we see that \[(M^\bullet \otimes_A^\mathbf{L} P^\bullet) \otimes_B^\mathbf{L} K^\bullet \longrightarrow (M^\bullet \otimes_A^\mathbf{L} N^\bullet) \otimes_B^\mathbf{L} K^\bullet\] is an isomorphism by Lemma 0BYK applied to the material inside the brackets. By Lemmas 06Y1 and 0795 the complex \[\text{Tot}(M^\bullet \otimes_A P^\bullet) = \text{Tot}((M^\bullet \otimes_R A) \otimes_A P^\bullet\] is K-flat as a complex of \(B\)-modules and it represents the derived tensor product in \(D(B)\) by construction. Hence we see that \((M^\bullet \otimes_A^\mathbf{L} P^\bullet) \otimes_B^\mathbf{L} K^\bullet\) is represented by the complex \[\text{Tot}(\text{Tot}(M^\bullet \otimes_A P^\bullet)\otimes_B K^\bullet) = \text{Tot}(M^\bullet \otimes_A \text{Tot}(P^\bullet \otimes_B K^\bullet))\] of \(C\)-modules. Equality by Homology, Remark 08BI. Going back the way we came we see that this is equal to \[M^\bullet \otimes_A^\mathbf{L} (P^\bullet \otimes_B^\mathbf{L} K^\bullet) \longleftarrow M^\bullet \otimes_A^\mathbf{L} (N^\bullet \otimes_B^\mathbf{L} K^\bullet)\] The arrow is an isomorphism by definition of the functor \(-\otimes_B^\mathbf{L} K^\bullet\). All of these constructions are functorial in the complex \(M^\bullet\) and hence we obtain our isomorphism of functors.
By the above we have the first equality in \[(M \otimes_A^\mathbf{L} N) \otimes_B^\mathbf{L} K = M \otimes_A^\mathbf{L} (N \otimes_B^\mathbf{L} K) = (M \otimes_A^\mathbf{L} C) \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} K)\] The second equality follows from the final statement of Lemma 06Y6. The same thing allows us to write \(N \otimes_B^\mathbf{L} K = (N \otimes_B^\mathbf{L} C) \otimes_C^\mathbf{L} K\) and substituting we get \[\begin{align*} (M \otimes_A^\mathbf{L} N) \otimes_B^\mathbf{L} K & = (M \otimes_A^\mathbf{L} C) \otimes_C^\mathbf{L} ((N \otimes_B^\mathbf{L} C) \otimes_C^\mathbf{L} K) \\ & = (M \otimes_A^\mathbf{L} C) \otimes_C^\mathbf{L} (K \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} C)) \\ & = ((M \otimes_A^\mathbf{L} C) \otimes_C^\mathbf{L} K) \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} C)) \\ & = (M \otimes_C^\mathbf{L} K) \otimes_C^\mathbf{L} (N \otimes_B^\mathbf{L} C) \end{align*}\] by Lemmas 0BYI and 0BYJ as well as the previously mentioned lemma.
Tor independence
Consider a commutative diagram \[\xymatrix{ A \ar[r] & A' \\ R \ar[r] \ar[u] & R' \ar[u] }\] of rings. Given an object \(K\) of \(D(A)\) we can consider its derived base change \(K \otimes_A^\mathbf{L} A'\) to an object of \(D(A')\). Or we can take the restriction of \(K\) to an object of \(D(R)\) and consider the derived base change of this to an object of \(D(R')\), denoted \(K \otimes_R^\mathbf{L} R'\). We claim there is a functorial comparison map [065Z]\[\begin{equation} K \otimes_R^{\mathbf{L}} R' \longrightarrow K \otimes_A^{\mathbf{L}} A' \end{equation}\] in \(D(R')\). To construct this comparison map choose a K-flat complex \(K^\bullet\) of \(A\)-modules representing \(K\). Next, choose a quasi-isomorphism \(E^\bullet \to K^\bullet\) where \(E^\bullet\) is a K-flat complex of \(R\)-modules. The map above is the map \[K \otimes_R^{\mathbf{L}} R' = E^\bullet \otimes_R R' \longrightarrow K^\bullet \otimes_A A' = K \otimes_A^{\mathbf{L}} A'\] In general there is no chance that this map is an isomorphism.
However, we often encounter the situation where the diagram above is a “base change” diagram of rings, i.e., \(A' = A \otimes_R R'\). In this situation, for any \(A\)-module \(M\) we have \(M \otimes_A A' = M \otimes_R R'\). Thus \(- \otimes_R R'\) is equal to \(- \otimes_A A'\) as a functor \(\text{Mod}_A \to \text{Mod}_{A'}\). In general this equality does not extend to derived tensor products. In other words, the comparison map is not an isomorphism. A simple example is to take \(R = k[x]\), \(A = R' = A' = k[x]/(x) = k\) and \(K^\bullet = A[0]\). Clearly, a necessary condition is that \(\text{Tor}_p^R(A, R') = 0\) for all \(p > 0\).
Definition
Let \(R\) be a ring. Let \(A\), \(B\) be \(R\)-algebras. We say \(A\) and \(B\) are Tor independent over \(R\) if \(\text{Tor}_p^R(A, B) = 0\) for all \(p > 0\).
Lemma
The comparison map (065Z) is an isomorphism if \(A' = A \otimes_R R'\) and \(A\) and \(R'\) are Tor independent over \(R\).
Proof
To prove this we choose a free resolution \(F^\bullet \to R'\) of \(R'\) as an \(R\)-module. Because \(A\) and \(R'\) are Tor independent over \(R\) we see that \(F^\bullet \otimes_R A\) is a free \(A\)-module resolution of \(A'\) over \(A\). By our general construction of the derived tensor product above we see that \[K^\bullet \otimes_A A' \cong \text{Tot}(K^\bullet \otimes_A (F^\bullet \otimes_R A)) = \text{Tot}(K^\bullet \otimes_R F^\bullet) \cong \text{Tot}(E^\bullet \otimes_R F^\bullet) \cong E^\bullet \otimes_R R'\] as desired.
Lemma
Consider a commutative diagram of rings \[\xymatrix{ A' & R' \ar[r] \ar[l] & B' \\ A \ar[u] & R \ar[l] \ar[u] \ar[r] & B \ar[u] }\] Assume that \(R'\) is flat over \(R\) and \(A'\) is flat over \(A \otimes_R R'\) and \(B'\) is flat over \(R' \otimes_R B\). Then \[\text{Tor}_i^R(A, B) \otimes_{(A \otimes_R B)} (A' \otimes_{R'} B') = \text{Tor}_i^{R'}(A', B')\]
Proof
By Algebra, Section 00M7 there are canonical maps \[\text{Tor}_i^R(A, B) \longrightarrow \text{Tor}_i^{R'}(A \otimes_R R', B \otimes_R R') \longrightarrow \text{Tor}_i^{R'}(A', B')\] These induce a map from left to right in the formula of the lemma.
Take a free resolution \(F_\bullet \to A\) of \(A\) as an \(R\)-module. Then we see that \(F_\bullet \otimes_R R'\) is a resolution of \(A \otimes_R R'\). Hence \(\text{Tor}_i^{R'}(A \otimes_R R', B \otimes_R R')\) is computed by \(F_\bullet \otimes_R B \otimes_R R'\). By our assumption that \(R'\) is flat over \(R\), this computes \(\text{Tor}_i^R(A, B) \otimes_R R'\). Thus \(\text{Tor}_i^{R'}(A \otimes_R R', B \otimes_R R') = \text{Tor}_i^R(A, B) \otimes_R R'\) (uses only flatness of \(R'\) over \(R\)).
By Lazard’s theorem (Algebra, Theorem 058G) we can write \(A'\), resp. \(B'\) as a filtered colimit of finite free \(A \otimes_R R'\), resp. \(B \otimes_R R'\)-modules. Say \(A' = \colim M_i\) and \(B' = \colim N_j\). The result above gives \[\text{Tor}_i^{R'}(M_i, N_j) = \text{Tor}_i^R(A, B) \otimes_{A \otimes_R B} (M_i \otimes_{R'} N_j)\] as one can see by writing everything out in terms of bases. Taking the colimit we get the result of the lemma.
Lemma
Let \(R \to A\) and \(R \to B\) be ring maps. Let \(R \to R'\) be a ring map and set \(A' = A \otimes_R R'\) and \(B' = B \otimes_R R'\). If \(A\) and \(B\) are tor independent over \(R\) and \(R \to R'\) is flat, then \(A'\) and \(B'\) are tor independent over \(R'\).
Proof
Lemma
Assumptions as in Lemma 08HW. For \(M \in D(A)\) there are canonical isomorphisms \[H^i((M \otimes_A^\mathbf{L} A') \otimes_{R'}^\mathbf{L} B') = H^i(M \otimes_R^\mathbf{L} B) \otimes_{(A \otimes_R B)} (A' \otimes_{R'} B')\] of \(A' \otimes_{R'} B'\)-modules.
Proof
Let us elucidate the two sides of the equation. On the left hand side we have the composition of the functors \(D(A) \to D(A') \to D(R') \to D(B')\) with the functor \(H^i : D(B') \to \text{Mod}_{B'}\). Since there is a map from \(A'\) to the endomorphisms of the object \((M \otimes_A^\mathbf{L} A') \otimes_{R'}^\mathbf{L} B'\) in \(D(B')\), we see that the left hand side is indeed an \(A' \otimes_{R'} B'\)-module. By the same arguments we see that \(H^i(M \otimes_R^\mathbf{L} B)\) has an \(A \otimes_R B\)-module structure.
We first prove the result in case \(B' = R' \otimes_R B\). In this case we choose a resolution \(F^\bullet \to B\) by free \(R\)-modules. We also choose a K-flat complex \(M^\bullet\) of \(A\)-modules representing \(M\). Then the left hand side is represented by \[\begin{align*} H^i(\text{Tot}((M^\bullet \otimes_A A') \otimes_{R'} (R' \otimes_R F^\bullet))) & = H^i(\text{Tot}(M^\bullet \otimes_A A' \otimes_R F^\bullet)) \\ & = H^i(\text{Tot}(M^\bullet \otimes_R F^\bullet) \otimes_A A') \\ & = H^i(M \otimes_R^\mathbf{L} B) \otimes_A A' \end{align*}\] The final equality because \(A \to A'\) is flat. The final module is the desired module because \(A' \otimes_{R'} B' = A' \otimes_R B\) since we’ve assumed \(B' = R' \otimes_R B\) in this paragraph.
General case. Suppose that \(B' \to B''\) is a flat ring map. Then it is easy to see that \[H^i((M \otimes_A^\mathbf{L} A') \otimes_{R'}^\mathbf{L} B'') = H^i((M \otimes_A^\mathbf{L} A') \otimes_{R'}^\mathbf{L} B') \otimes_{B'} B''\] and \[H^i(M \otimes_R^\mathbf{L} B) \otimes_{(A \otimes_R B)} (A' \otimes_{R'} B'') = \left( H^i(M \otimes_R^\mathbf{L} B) \otimes_{(A \otimes_R B)} (A' \otimes_{R'} B') \right) \otimes_{B'} B''\] Thus the result for \(B'\) implies the result for \(B''\). Since we’ve proven the result for \(R' \otimes_R B\) in the previous paragraph, this implies the result in general.
Lemma
Let \(R\) be a ring. Let \(A\), \(B\) be \(R\)-algebras. The following are equivalent
\(A\) and \(B\) are Tor independent over \(R\),
for every pair of primes \(\mathfrak p \subset A\) and \(\mathfrak q \subset B\) lying over the same prime \(\mathfrak r \subset R\) the rings \(A_\mathfrak p\) and \(B_\mathfrak q\) are Tor independent over \(R_\mathfrak r\), and
For every prime \(\mathfrak s\) of \(A \otimes_R B\) the module \[\text{Tor}_i^R(A, B)_\mathfrak s = \text{Tor}_i^{R_\mathfrak r}(A_\mathfrak p, B_\mathfrak q)_\mathfrak s\] (where \(\mathfrak p = A \cap \mathfrak s\), \(\mathfrak q = B \cap \mathfrak s\) and \(\mathfrak r = R \cap \mathfrak s\)) is zero.
Proof
Let \(\mathfrak s\) be a prime of \(A \otimes_R B\) as in (3). The equality \[\text{Tor}_i^R(A, B)_\mathfrak s = \text{Tor}_i^{R_\mathfrak r}(A_\mathfrak p, B_\mathfrak q)_\mathfrak s\] where \(\mathfrak p = A \cap \mathfrak s\), \(\mathfrak q = B \cap \mathfrak s\) and \(\mathfrak r = R \cap \mathfrak s\) follows from Lemma 08HW. Hence (2) implies (3). Since we can test the vanishing of modules by localizing at primes (Algebra, Lemma 00HN) we conclude that (3) implies (1). For (1) \(\Rightarrow\) (2) we use that \[\text{Tor}_i^{R_\mathfrak r}(A_\mathfrak p, B_\mathfrak q) = \text{Tor}_i^R(A, B) \otimes_{(A \otimes_R B)} (A_\mathfrak p \otimes_{R_{\mathfrak r}} B_\mathfrak q)\] again by Lemma 08HW.
Spectral sequences for Tor
In this section we collect various spectral sequences that come up when considering the Tor functors.
Example
Let \(R\) be a ring. Let \(K_\bullet\) be a chain complex of \(R\)-modules with \(K_n = 0\) for \(n \ll 0\). Let \(M\) be an \(R\)-module. Choose a resolution \(P_\bullet \to M\) of \(M\) by free \(R\)-modules. We obtain a double chain complex \(K_\bullet \otimes_R P_\bullet\). Applying the material in Homology, Section 012X (especially Homology, Lemma 0132) translated into the language of chain complexes we find two spectral sequences converging to \(H_*(K_\bullet \otimes_R^\mathbf{L} M)\). Namely, on the one hand a spectral sequence with \(E_2\)-page \[(E_2)_{i, j} = \text{Tor}^R_j(H_i(K_\bullet), M) \Rightarrow H_{i + j}(K_\bullet \otimes^{\mathbf{L}}_R M)\] and differential \(d_2\) given by maps \(\text{Tor}^R_j(H_i(K_\bullet), M) \to \text{Tor}^R_{j - 2}(H_{i + 1}(K_\bullet), M)\). Another spectral sequence with \(E_1\)-page \[(E_1)_{i, j} = \text{Tor}^R_j(K_i, M) \Rightarrow H_{i + j}(K_\bullet \otimes^{\mathbf{L}}_R M)\] with differential \(d_1\) given by maps \(\text{Tor}^R_j(K_i, M) \to \text{Tor}^R_j(K_{i - 1}, M)\) induced by \(K_i \to K_{i - 1}\).
Example
Let \(R \to S\) be a ring map. Let \(M\) be an \(R\)-module and let \(N\) be an \(S\)-module. Then there is a spectral sequence \[\text{Tor}^S_n(\text{Tor}^R_m(M, S), N) \Rightarrow \text{Tor}^R_{n + m}(M, N).\] To construct it choose a \(R\)-free resolution \(P^\bullet\) of \(M\). Then we have \[M \otimes_R^{\mathbf{L}} N = P^\bullet \otimes_R N = (P^\bullet \otimes_R S) \otimes_S N\] and then apply the first spectral sequence of Example 061Z.
Example
Consider a commutative diagram \[\xymatrix{ B \ar[r] & B' = B \otimes_A A' \\ A \ar[r] \ar[u] & A' \ar[u] }\] and \(B\)-modules \(M, N\). Set \(M' = M \otimes_A A' = M \otimes_B B'\) and \(N' = N \otimes_A A' = N \otimes_B B'\). Assume that \(A \to B\) is flat and that \(M\) and \(N\) are \(A\)-flat. Then there is a spectral sequence \[\text{Tor}^A_i(\text{Tor}_j^B(M, N), A') \Rightarrow \text{Tor}^{B'}_{i + j}(M', N')\] The reason is as follows. Choose free resolution \(F_\bullet \to M\) as a \(B\)-module. As \(B\) and \(M\) are \(A\)-flat we see that \(F_\bullet \otimes_A A'\) is a free \(B'\)-resolution of \(M'\). Hence we see that the groups \(\text{Tor}^{B'}_n(M', N')\) are computed by the complex \[(F_\bullet \otimes_A A') \otimes_{B'} N' = (F_\bullet \otimes_B N) \otimes_A A' = (F_\bullet \otimes_B N) \otimes^{\mathbf{L}}_A A'\] the last equality because \(F_\bullet \otimes_B N\) is a complex of flat \(A\)-modules as \(N\) is flat over \(A\). Hence we obtain the spectral sequence by applying the spectral sequence of Example 061Z.
Example
Let \(K^\bullet, L^\bullet\) be objects of \(D^{-}(R)\). Then there is a spectral sequence with \[E_2^{p, q} = H^p(K^\bullet \otimes_R^{\mathbf{L}} H^q(L^\bullet)) \Rightarrow H^{p + q}(K^\bullet \otimes_R^{\mathbf{L}} L^\bullet)\] and another spectral sequence with \[E_2^{p, q} = H^p(H^q(K^\bullet) \otimes_R^{\mathbf{L}} L^\bullet) \Rightarrow H^{p + q}(K^\bullet \otimes_R^{\mathbf{L}} L^\bullet)\] Both spectral sequences have \(d_2^{p, q} : E_2^{p, q} \to E_2^{p + 2, q - 1}\). After replacing \(K^\bullet\) and \(L^\bullet\) by bounded above complexes of projectives, these spectral sequences are simply the two spectral sequences for computing the cohomology of \(\text{Tot}(K^\bullet \otimes L^\bullet)\) discussed in Homology, Section 012X.
Products and Tor
The simplest example of the product maps comes from the following situation. Suppose that \(K^\bullet, L^\bullet \in D(R)\). Then there are maps [068H]\[\begin{equation} H^i(K^\bullet) \otimes_R H^j(L^\bullet) \longrightarrow H^{i + j}(K^\bullet \otimes_R^{\mathbf{L}} L^\bullet) \end{equation}\] Namely, to define these maps we may assume that one of \(K^\bullet, L^\bullet\) is a K-flat complex of \(R\)-modules (for example a bounded above complex of free or projective \(R\)-modules). In that case \(K^\bullet \otimes_R^{\mathbf{L}} L^\bullet\) is represented by the complex \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\), see Section 06XY (or Section 064F). Next, suppose that \(\xi \in H^i(K^\bullet)\) and \(\zeta \in H^j(L^\bullet)\). Choose \(k \in \Ker(K^i \to K^{i + 1})\) and \(l \in \Ker(L^j \to L^{j + 1})\) representing \(\xi\) and \(\zeta\). Then we set \[\xi \cup \zeta = \text{class of }k \otimes l\text{ in } H^{i + j}(\text{Tot}(K^\bullet \otimes_R L^\bullet)).\] This make sense because the formula (see Homology, Definition 012Z) for the differential \(\text{d}\) on the total complex shows that \(k \otimes l\) is a cocycle. Moreover, if \(k' = d_K(k'')\) for some \(k'' \in K^{i - 1}\), then \(k' \otimes l = \text{d}(k'' \otimes l)\) because \(l\) is a cocycle. Similarly, altering the choice of \(l\) representing \(\zeta\) does not change the class of \(k \otimes l\). It is equally clear that \(\cup\) is bilinear, and hence to a general element of \(H^i(K^\bullet) \otimes_R H^j(L^\bullet)\) we assign \[\sum \xi_i \otimes \zeta_i \longmapsto \sum \xi_i \cup \zeta_i\] in \(H^{i + j}(\text{Tot}(K^\bullet \otimes_R L^\bullet))\).
Let \(R \to A\) be a ring map. Let \(K^\bullet, L^\bullet \in D(R)\). Then we have a canonical identification [068I]\[\begin{equation} (K^\bullet \otimes_R^{\mathbf{L}} A) \otimes_A^{\mathbf{L}} (L^\bullet \otimes_R^{\mathbf{L}} A) = (K^\bullet \otimes_R^{\mathbf{L}} L^\bullet) \otimes_R^{\mathbf{L}} A \end{equation}\] in \(D(A)\). It is constructed as follows. First, choose K-flat resolutions \(P^\bullet \to K^\bullet\) and \(Q^\bullet \to L^\bullet\) over \(R\). Then the left hand side is represented by the complex \(\text{Tot}((P^\bullet \otimes_R A) \otimes_A (Q^\bullet \otimes_R A))\) and the right hand side by the complex \(\text{Tot}(P^\bullet \otimes_R Q^\bullet) \otimes_R A\). These complexes are canonically isomorphic. Thus the construction above induces products \[\text{Tor}^R_n(K^\bullet, A) \otimes_A \text{Tor}^R_m(L^\bullet, A) \longrightarrow \text{Tor}_{n + m}^R(K^\bullet \otimes_R^\mathbf{L} L^\bullet, A)\] which are occasionally useful.
Let \(M\), \(N\) be \(R\)-modules. Using the general construction above, the canonical map \(M \otimes_R^\mathbf{L} N \to M \otimes_R N\) and functoriality of \(\text{Tor}\) we obtain canonical maps [068J]\[\begin{equation} \text{Tor}^R_n(M, A) \otimes_A \text{Tor}^R_m(N, A) \longrightarrow \text{Tor}_{n + m}^R(M \otimes_R N, A) \end{equation}\] Here is a direct construction using projective resolutions. First, choose projective resolutions \[P_\bullet \to M, \quad Q_\bullet \to N, \quad T_\bullet \to M \otimes_R N\] over \(R\). We have \(H_0(\text{Tot}(P_\bullet \otimes_R Q_\bullet)) = M \otimes_R N\) by right exactness of \(\otimes_R\). Hence Derived Categories, Lemmas 0649 and 064A guarantee the existence and uniqueness of a map of complexes \(\mu : \text{Tot}(P_\bullet \otimes_R Q_\bullet) \to T_\bullet\) such that \(H_0(\mu) = \text{id}_{M \otimes_R N}\). This induces a canonical map \[\begin{align*} (M \otimes_R^{\mathbf{L}} A) \otimes_A^{\mathbf{L}} (N \otimes_R^{\mathbf{L}} A) & = \text{Tot}((P_\bullet \otimes_R A) \otimes_A (Q_\bullet \otimes_R A)) \\ & = \text{Tot}(P_\bullet \otimes_R Q_\bullet) \otimes_R A \\ & \to T_\bullet \otimes_R A \\ & = (M \otimes_R N) \otimes_R^{\mathbf{L}} A \end{align*}\] in \(D(A)\). Hence the products (068J) above are constructed using (068H) over \(A\) to construct \[\text{Tor}^R_n(M, A) \otimes_A \text{Tor}^R_m(N, A) \to H^{-n-m}((M \otimes_R^{\mathbf{L}} A) \otimes_A^{\mathbf{L}} (N \otimes_R^{\mathbf{L}} A))\] and then composing by the displayed map above to end up in \(\text{Tor}_{n + m}^R(M \otimes_R N, A)\).
An interesting special case of the above occurs when \(M = N = B\) where \(B\) is an \(R\)-algebra. In this case we obtain maps \[\text{Tor}_n^R(B, A) \otimes_A \text{Tor}_m^R(B, A) \longrightarrow \text{Tor}_{n + m}^R(B \otimes_R B, A) \longrightarrow \text{Tor}_{n + m}^R(B, A)\] the second arrow being induced by the multiplication map \(B \otimes_R B \to B\) via functoriality for \(\text{Tor}\). In other words we obtain an \(A\)-algebra structure on \(\text{Tor}^R_{\star}(B, A)\). This algebra structure has many intriguing properties (associativity, graded commutative, \(B\)-algebra structure, divided powers in some case, etc) which we will discuss elsewhere.
Lemma
Let \(R\) be a ring. Let \(A, B, C\) be \(R\)-algebras and let \(B \to C\) be an \(R\)-algebra map. Then the induced map \[\text{Tor}^R_{\star}(B, A) \longrightarrow \text{Tor}^R_{\star}(C, A)\] is an \(A\)-algebra homomorphism.
Proof
Omitted. Hint: You can prove this by working through the definitions, writing all the complexes explicitly.
Künneth spectral sequence
Let \(R\) be a ring. Let \(K^\bullet\) and \(L^\bullet\) be filtered complexes of \(R\)-modules (see Homology, Definition 012L; note that our filtrations are decreasing). Then the complex \[T^\bullet = \text{Tot}(K^\bullet \otimes_R L^\bullet)\] also has a decreasing filtration defined by the formula \[F^nT^\bullet = \Im\left( \bigoplus\nolimits_{i + j = n} \text{Tot}(F^iK^\bullet \otimes_R F^jL^\bullet) \to \text{Tot}(K^\bullet \otimes_R L^\bullet) \right)\] Under some assumptions on our filtered complexes, we will determine the spectral sequence that arises from this by the construction in Homology, Section 012K.
Assume that each \(K^n\), \(F^iK^n\), \(\text{gr}^iK^n\), \(L^m\), \(F^jL^m\), and \(\text{gr}^jL^m\) is a flat \(R\)-module. In this case the modules \(F^iK^n \otimes_R L^m\), \(K^n \otimes_R F^jL^m\), and \(F^iK^n \otimes_R F^jL^m\) are submodules of \(K^n \otimes_R L^m\). Similarly, the module \(\text{gr}^iK^n \otimes_R \text{gr}^jL^m\) is a submodule of \(K^n/F^{i + 1}K^n \otimes_R L^m/F^{j + 1}L^m\). Consider the map \[F^nT^n \longrightarrow \bigoplus\nolimits_{i + j = n} \text{Tot}(K^\bullet / F^{i + 1}K^\bullet \otimes_R L^\bullet / F^{j + 1}L^\bullet)\] We leave it to the reader to show that we do indeed end up in the direct sum and not the direct product, due to our definition of \(F^nT^\bullet\). For \(a + b = n\), the restriction of the displayed arrow to the subcomplex \(\text{Tot}(F^aK^\bullet \otimes_R F^bL^\bullet)\) of \(F^nT^\bullet\) maps into the summand with \(i = a\) and \(j = b\). Moreover, by our flatness assumptions, the image is isomorphic to \(\text{Tot}(\text{gr}^iK^\bullet \otimes_R \text{gr}^jL^\bullet)\) and the kernel of \(\text{Tot}(F^iK^\bullet \otimes_R F^jL^\bullet) \to \text{Tot}(\text{gr}^iK^\bullet \otimes_R \text{gr}^jL^\bullet)\) is \(\text{Tot}(F^{i + 1}K^\bullet \otimes_R F^jL^\bullet) + \text{Tot}(F^iK^\bullet \otimes_R F^{j + 1}L^\bullet)\). It follows that we have a short exact sequence of complexes \[0 \to F^{n + 1}T^\bullet \to F^nT^\bullet \to \bigoplus\nolimits_{i + j = n} \text{Tot}(\text{gr}^iK^\bullet \otimes_R \text{gr}^jL^\bullet) \to 0\] under our assumptions. In other words, this tells us that \(\text{gr}^nT^\bullet\) is a direct sum of the complexes \(\text{Tot}(\text{gr}^iK^\bullet \otimes_R \text{gr}^jL^\bullet)\) for \(i + j = n\).
Assume in addition that the complexes of \(R\)-modules \(K^\bullet\), \(F^iK^\bullet\), \(\text{gr}^iK^\bullet\), \(L^\bullet\), \(F^jL^\bullet\), and \(\text{gr}^jL^\bullet\) are K-flat. In this case we conclude that the spectral sequence of Homology, Section 012K associated to the filtered complex \(T^\bullet\) has terms \[E_1^{p, q} = \bigoplus\nolimits_{i + j = p} H^{p + q}(\text{gr}^iK^\bullet \otimes_R^\mathbf{L} \text{gr}^jL^\bullet)\] with differentials induced from the short exact sequences \[0 \to \text{gr}^{n + 1}T^\bullet \to F^nT^\bullet/F^{n + 2}T^\bullet \to \text{gr}^nT^\bullet \to 0\] as explained in Homology, Lemma 012N. In particular, the reader can show this means that the summand \(H^{p + q}(\text{gr}^iK^\bullet \otimes_R^\mathbf{L} \text{gr}^jL^\bullet)\) of \(E_1^{p, q}\) maps into the sum \[H^{p + q + 1}(\text{gr}^{i + 1}K^\bullet \otimes_R^\mathbf{L} \text{gr}^jL^\bullet) \oplus H^{p + q + 1}(\text{gr}^iK^\bullet \otimes_R^\mathbf{L} \text{gr}^{j + 1}L^\bullet)\] inside \(E_1^{p + 1, q}\).
Lemma
Under the assumptions above, if in addition
the filtration on \(K^\bullet\) is finite and the filtration on \(L^\bullet\) is finite, or
more generally the following are true
\(F^iK^\bullet\) is acyclic for \(i \gg 0\),
\(F^iK^\bullet \to K^\bullet\) is a quasi-isomorphism for \(i \ll 0\),
\(F^jL^\bullet\) is acyclic for \(j \gg 0\), and
\(F^jL^\bullet \to L^\bullet\) is a quasi-isomorphism for \(j \ll 0\).
Then the spectral sequence is bounded, the associated filtration on each \(H^n(T^\bullet) = H^n(K^\bullet \otimes_R^\mathbf{L} L^\bullet)\) is finite and we have convergence \[E_1^{p, q} = \bigoplus\nolimits_{i + j = p} H^{p + q}(\text{gr}^iK^\bullet \otimes_R^\mathbf{L} \text{gr}^jL^\bullet) \Rightarrow H^n(K^\bullet \otimes_R^\mathbf{L} L^\bullet)\]
Proof
In case (1) the filtration on \(T^\bullet\) is finite and the lemma follows immediately from Homology, Lemma 012W. In case (2) choose \(a < b\) such that \(F^iK^\bullet\) and \(F^jL^\bullet\) are acyclic for \(i, j > b\) and \(F^iK^\bullet \to K^\bullet\) and \(F^jL^\bullet \to L^\bullet\) are quasi-isomorphisms for \(i, j < a\). We claim that in this case the complex \(F^nT^\bullet\) is acyclic for \(n > 2b\) and that \(F^nT^\bullet \to T^\bullet\) is a quasi-isomorphism for \(n < 2a - 1\). Since the claim shows that Homology, Lemma 0BK5 applies we see that our lemma is true.
Proof of the claim. Given \(n\) and integers \(i_1 \leq i_2\) consider \[S^{n, i_1, i_2} = \Im\left( \bigoplus\nolimits_{i_1 \leq i \leq i_2} \text{Tot}(F^iK^\bullet \otimes_R F^{n - i}L^\bullet) \to \text{Tot}(K^\bullet \otimes_R L^\bullet) \right)\] viewed as a subcomplex of \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\). Observe that \(F^nT^\bullet = \colim S^{n, i_1, i_2}\) is a filtered colimit. Thus to show the claim it suffices to show that \(S^{n, i_1, i_2}\) is acyclic for \(n > 2b\) and that \(S^{n, i_1, i_2} \to \text{Tot}(K^\bullet \otimes L^\bullet)\) is a quasi-isomorphism for \(n < 2a - 1\) and a cofinal set of choices of pairs \(i_1, i_2\).
If \(i_1 = i_2 = i\) and \(n > 2b\), then \[S^{n, i, i} = \text{Tot}(F^iK^\bullet \otimes_R F^{n - i}L^\bullet)\] is acyclic: either \(i > b\) and this complex is acyclic by our assumption that \(F^{n - i}L^\bullet\) is K-flat, or \(i \leq b\) and then \(n - i \geq n - b > b\) and the same holds. Using our flatness assumptions the reader shows that there is a short exact sequence \[0 \to S^{n + 1, i_2 + 1, i_2 + 1} \to S^{n, i_1, i_2} \oplus S^{n, i_2 + 1, i_2 + 1} \to S^{n, i_1, i_2 + 1} \to 0\] of complexes. Thus we see by induction on \(i_2 - i_1\) that all of the complexes \(S^{n, i_1, i_2}\) for \(n > 2b\) are acyclic.
Assume \(n < 2a - 1\). Observe that \[S^{n, a - 1, a - 1} = \text{Tot}(F^{a - 1}K^\bullet \otimes_R F^{n - a + 1}L^\bullet) \to \text{Tot}(K^\bullet \otimes_R L^\bullet)\] is a quasi-isomorphism because both \(a - 1 < a\) and \(n - a + 1 < a\) and all our complexes are K-flat hence both sides compute the same object of \(D(R)\). Observe that for \(i_2 \geq a - 1\) and \(n < 2a - 1\) the map \[S^{n + 1, i_2 + 1, i_2 + 1} \to S^{n, i_2 + 1, i_2 + 1}\] is a quasi-isomorphism because left and right hand side are quasi-isomorphic to \(\text{Tot}(F^{i_2 + 1}K^\bullet \otimes L^\bullet)\). Thus using the short exact sequence of complexes in the previous paragraph we find that \(S^{n, a - 1, a - 1} \to S^{n, a - 1, t}\) is a quasi-isomorphism for all \(t \geq a - 1\). On the other hand, there are similarly short exact sequences \[0 \to S^{n + 1, i_1, i_1} \to S^{n, i_1 - 1, i_1 - 1} \oplus S^{n, i_1, i_2} \to S^{n, i_1 - 1, i_2} \to 0\] of complexes. Similarly to the above, we note that for \(i_1 \leq a - 1\) the map \[S^{n + 1, i_1, i_1} \to S^{n, i_1 - 1, i_1 - 1}\] is a quasi-isomorphism because left and right hand side are quasi-isomorphic to \(\text{Tot}(K^\bullet \otimes F^{n - i_1 + 1}L^\bullet)\). Thus we conclude that \(S^{n, s, t} \to S^{n, a - 1, t}\) is a quasi-isomorphism for all \(s \leq a - 1\). Combined we find that \(S^{n, s, t}\) for \(s \leq a - 1\) and \(t \geq a - 1\) maps quasi-isomorphically to \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\). This finishes the proof of the lemma.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a filtered complex. There exists a map \(f : P^\bullet \to K^\bullet\) of filtered complexes such that
each \(P^n\), \(F^iP^n\), \(\text{gr}^iP^n\) is a free \(R\)-module,
the complexes of \(R\)-modules \(P^\bullet\), \(F^iP^\bullet\), and \(\text{gr}^iP^\bullet\) are K-flat,
\(f\) induces quasi-isomorphisms \(P^\bullet \to K^\bullet\), \(F^iP^\bullet \to F^iK^\bullet\), and \(\text{gr}^iP^\bullet \to \text{gr}^iK^\bullet\).
Proof
Let us say a filtered complex \(L^\bullet\) is basic if each \(L^n\), \(F^iL^n\), \(\text{gr}^iL^n\) is a free \(R\)-module and if all differentials are zero.
There exists a basic filtered complex \(P_0^\bullet\) and a map \(f_0 : P_0^\bullet \to K^\bullet\) of filtered complexes such that \(f_0\) and \(F^if_0\), \(i \in \mathbf{Z}\) are surjective on cohomology. To see this set \[P_0^n = \bigoplus\nolimits_{z \in \Ker(d_K^n)} R \xi_z \oplus \bigoplus\nolimits_{j \in \mathbf{Z}} \bigoplus\nolimits_{z \in \Ker(d_{F^jK}^n)} R \xi_z\] with zero differentials and the map \(f_0\) defined by \(f_0(\xi_z) = z\). As for the filtration, we set \[F^iP_0^n = \bigoplus\nolimits_{j \geq i \in \mathbf{Z}} \bigoplus\nolimits_{z \in \Ker(d_{F^jK}^n)} R \xi_z\] We leave it to the reader to check that this gives \(f_0 : P_0^\bullet \to K^\bullet\) as claimed.
By induction on \(m \geq 0\) we are going to construct embeddings \[P_0^\bullet \subset \ldots P_m^\bullet \subset P_{m + 1}^\bullet\] of filtered complexes and maps \(f_m : P_m^\bullet \to K^\bullet\) with the following properties
the filtered complex \(P_{m + 1}^\bullet / P_m^\bullet\) is basic,
the kernel of \(H^n(f_m)\) and \(H^n(F^if_m)\) maps to zero in \(H^n(P_{m + 1}^\bullet)\) and \(H^n(F^iP_{m + 1}^\bullet)\),
the map \(f_{m + 1} : P_{m + 1}^\bullet \to K^\bullet\) extends the map \(f_m\).
To to this, set \[\Omega_{n, m} = \Ker\left(\Ker(d^n_{P_m}) \to H^n(K^\bullet)\right), \text{ resp. } \Omega_{n, i, m} = \Ker\left(\Ker(d^n_{F^iP_m}) \to H^n(F^iK^\bullet)\right),\] Note that \(\Omega_{n, m}\) surjects onto the kernel of \(H^n(f_m)\) and that \(\Omega_{n, i, m}\) surjects onto the kernel of \(H^n(F^if_m)\). For each \(z \in \Omega_{n, m}\), resp. \(z \in \Omega_{n, i, m}\) we choose a \(y_z \in K^{n - 1}\), resp. \(y_z \in F^iK^{n - 1}\) such that \(d_K(y_z) = f_m(z)\). Then we set \[P^n_{m + 1} = P_m^n \oplus \bigoplus\nolimits_{z \in \Omega_{n + 1, m}} R \eta_z \oplus \bigoplus\nolimits_{j \in \mathbf{Z}} \bigoplus\nolimits_{z \in \Omega_{n + 1, j, m}} R \eta_z\] We set \(d_{P_{m + 1}}(\eta_z) = z\) and we set \(f_{m + 1}(\eta_z) = y_z\). Finally, we set \[F^iP_{m + 1}^n = F^iP_m^n \oplus \bigoplus\nolimits_{j \geq i} \bigoplus\nolimits_{z \in \Omega_{n + 1, j, m}} R \eta_z\] We leave it to the reader to check that this gives \(P_m^\bullet \subset P_{m + 1}^\bullet\) and \(f_{m + 1} : P_{m + 1}^\bullet \to K^\bullet\) as claimed.
At this point we simply take \[P^\bullet = \bigcup P_m^\bullet\] as a filtered complex with map \(f : P^\bullet \to K^\bullet\) given by \(\bigcup f_m\).
Part (1) of the statement of the lemma holds because \(P^n\) as a filtered module is isomorphic to the direct sum of \(P_0^n\) and \(P_{m + 1}^n/P_m^n\) for \(m \geq 0\). Small detail omitted.
Part (3) of the statement. Observe that \(H^n(P^\bullet) = \colim H^n(P_m^\bullet)\). The map \(f_0\) is surjective on cohomology; whence the same holds for each \(f_m\). For each \(m\) by construction the embedding \(P_m^\bullet \subset P_{m + 1}^\bullet\) kills the kernel of \(H^n(f_m)\). Combining these facts the reader easily concludes that \(H^n(f)\) is an isomorphism. Similarly for \(H^n(F^if)\). Then also \(H^n(\text{gr}^if)\) must be an isomorphism because of the short exact sequence \(0 \to F^{i + 1} \to F^i \to \text{gr}^i \to 0\) (of functors on the category of filtered complexes, say). Small detail omitted.
Part (2) of the statement. To see that \(P^\bullet\) is K-flat, by Lemma 06Y3, it suffices to show that \(P_m^\bullet\) is K-flat. By Lemma 0BYH and induction it suffices to note that a complex with zero differentials and free terms is K-flat. The same argument works to show that \(F^iP^\bullet\) is K-flat for all \(i \in \mathbf{Z}\). Finally, we see that \(\text{gr}^iP^\bullet\) is K-flat by another application of Lemma 0BYH.
Proposition
Let \(R\) be a ring. Let \(K^\bullet\) and \(L^\bullet\) be filtered complexes of \(R\)-modules. Then there exists a filtered complex \(T^\bullet\) representing \(K^\bullet \otimes_R^\mathbf{L} L^\bullet\) in \(D(R)\) such that the associated spectral sequence has \(E_1\)-page \[E_1^{p, q} = \bigoplus\nolimits_{i + j = p} H^{p + q}(\text{gr}^iK^\bullet \otimes_R^\mathbf{L} \text{gr}^jL^\bullet)\] If
\(F^iK^\bullet\) is acyclic for \(i \gg 0\),
\(F^iK^\bullet \to K^\bullet\) is a quasi-isomorphism for \(i \ll 0\),
\(F^jL^\bullet\) is acyclic for \(j \gg 0\), and
\(F^jL^\bullet \to L^\bullet\) is a quasi-isomorphism for \(j \ll 0\).
then the spectral sequence is bounded, the associated filtration on each \(H^n(K^\bullet \otimes_R^\mathbf{L} L^\bullet)\) is finite and the spectral sequence convergences.
Proof
Choose \(P^\bullet \to K^\bullet\) and \(Q^\bullet \to L^\bullet\) as in Lemma 0H7X. Then we use the spectral sequence for the filtered complex \[T^\bullet = \text{Tot}(P^\bullet \otimes_R Q^\bullet)\] described in the text of this section and in Lemma 0H7W.
Lemma
Let \(R\) be a ring. Let \(K\) and \(L\) be objects of \(D^b(R)\). There exists a bigraded bounded spectral sequence \(\{E_r\}_{r \geq 2}\) with \[E_2^{p, q} = \bigoplus\nolimits_{i + j = q} \text{Tor}^R_{-p}(H^i(K), H^j(L))\] and \(d_r\) of bidegree \((r, -r + 1)\) converging to \(H^{p + q}(K \otimes_R^\mathbf{L} L)\).
Proof
Let \(K^\bullet\) be a complex of \(R\)-modules representing \(K\) and let \(L^\bullet\) be a complex of \(R\)-modules representing \(L\). Set \(F^iK^\bullet = \tau_{\leq -i}K^\bullet\) and similarly for \(L\), see Homology, Section 0118. Apply Proposition 0H7Y noting that \(\text{gr}^iK^\bullet = H^{-i}(K)[i]\) to get a bounded spectral sequence \(\{(E')_r\}_{r \geq 1}\) with \[(E')^{p, q}_1 = \bigoplus\nolimits_{i + j = p} \text{Tor}^R_{-2p - q}(H^{-i}(K), H^{-j}(L))\] converging to the cohomology of \(K \otimes_R^\mathbf{L} L\). By construction this spectral sequence has differentials \(d_r^{p, q} : (E')_r^{p, q} \to (E')_r^{p + r, q - r + 1}\) for \(r \geq 1\). To get the spectral sequence of the lemma we set for \(r \geq 2\) \[E_r^{p, q} = (E')_{r - 1}^{-q, p + 2q}\] We leave it to the reader to show that this works.
Example
If \(R = \mathbf{Z}\) or more generally if \(R\) is a Dedekind domain, then \(\text{Tor}_i^R(M, N) = 0\) for \(i \not \in \{0, 1\}\) for all \(R\)-modules \(M\) and \(N\). Hence the spectral sequence of Lemma 0H7Z degenerates at the \(E_2\) page and we get short exact sequences \[0 \to \bigoplus_{i + j = n} H^i(K) \otimes_R H^j(L) \to H^n(K \otimes_R^\mathbf{L} L) \to \bigoplus_{i + j = n + 1} \text{Tor}^R_1(H^i(K), H^j(L)) \to 0\] for all \(n \in \mathbf{Z}\).
Pseudo-coherent modules, I
Suppose that \(R\) is a ring. Recall that an \(R\)-module \(M\) is of finite type if there exists a surjection \(R^{\oplus a} \to M\) and of finite presentation if there exists a presentation \(R^{\oplus a_1} \to R^{\oplus a_0} \to M \to 0\). Similarly, we can consider those \(R\)-modules for which there exists a length \(n\) resolution [064P]\[\begin{equation} R^{\oplus a_n} \to R^{\oplus a_{n - 1}} \to \ldots \to R^{\oplus a_0} \to M \to 0 \end{equation}\] by finite free \(R\)-modules. A module is called pseudo-coherent if we can find such a resolution for every \(n\). Here is the formal definition.
Definition
Let \(R\) be a ring. Denote \(D(R)\) its derived category. Let \(m \in \mathbf{Z}\).
An object \(K^\bullet\) of \(D(R)\) is \(m\)-pseudo-coherent if there exists a bounded complex \(E^\bullet\) of finite free \(R\)-modules and a morphism \(\alpha : E^\bullet \to K^\bullet\) such that \(H^i(\alpha)\) is an isomorphism for \(i > m\) and \(H^m(\alpha)\) is surjective.
An object \(K^\bullet\) of \(D(R)\) is pseudo-coherent if it is quasi-isomorphic to a bounded above complex of finite free \(R\)-modules.
An \(R\)-module \(M\) is called \(m\)-pseudo-coherent if \(M[0]\) is an \(m\)-pseudo-coherent object of \(D(R)\).
An \(R\)-module \(M\) is called pseudo-coherent8 if \(M[0]\) is a pseudo-coherent object of \(D(R)\).
As usual we apply this terminology also to complexes of \(R\)-modules. Since any morphism \(E^\bullet \to K^\bullet\) in \(D(R)\) is represented by an actual map of complexes, see Derived Categories, Lemma 064B, there is no ambiguity. It turns out that \(K^\bullet\) is pseudo-coherent if and only if \(K^\bullet\) is \(m\)-pseudo-coherent for all \(m \in \mathbf{Z}\), see Lemma 064U. Also, if the ring is Noetherian the condition can be understood as a finite generation condition on the cohomology, see Lemma 066E. Let us first relate this to the informal discussion above.
Lemma
Let \(R\) be a ring and \(m \in \mathbf{Z}\). Let \((K^\bullet, L^\bullet, M^\bullet, f, g, h)\) be a distinguished triangle in \(D(R)\).
If \(K^\bullet\) is \((m + 1)\)-pseudo-coherent and \(L^\bullet\) is \(m\)-pseudo-coherent then \(M^\bullet\) is \(m\)-pseudo-coherent.
If \(K^\bullet, M^\bullet\) are \(m\)-pseudo-coherent, then \(L^\bullet\) is \(m\)-pseudo-coherent.
If \(L^\bullet\) is \((m + 1)\)-pseudo-coherent and \(M^\bullet\) is \(m\)-pseudo-coherent, then \(K^\bullet\) is \((m + 1)\)-pseudo-coherent.
Proof
Proof of (1). Choose \(\alpha : P^\bullet \to K^\bullet\) with \(P^\bullet\) a bounded complex of finite free modules such that \(H^i(\alpha)\) is an isomorphism for \(i > m + 1\) and surjective for \(i = m + 1\). We may replace \(P^\bullet\) by \(\sigma_{\geq m + 1}P^\bullet\) and hence we may assume that \(P^i = 0\) for \(i < m + 1\). Choose \(\beta : E^\bullet \to L^\bullet\) with \(E^\bullet\) a bounded complex of finite free modules such that \(H^i(\beta)\) is an isomorphism for \(i > m\) and surjective for \(i = m\). By Derived Categories, Lemma 064E we can find a map \(\gamma : P^\bullet \to E^\bullet\) such that the diagram \[\xymatrix{ K^\bullet \ar[r] & L^\bullet \\ P^\bullet \ar[u] \ar[r]^\gamma & E^\bullet \ar[u]_\beta }\] is commutative in \(D(R)\). The cone \(C(\gamma)^\bullet\) is a bounded complex of finite free \(R\)-modules, and the commutativity of the diagram implies that there exists a morphism of distinguished triangles \[(P^\bullet, E^\bullet, C(\gamma)^\bullet) \longrightarrow (K^\bullet, L^\bullet, M^\bullet).\] It follows from the induced map on long exact cohomology sequences and Homology, Lemmas 05QA and 05QB that \(C(\gamma)^\bullet \to M^\bullet\) induces an isomorphism on cohomology in degrees \(> m\) and a surjection in degree \(m\). Hence \(M^\bullet\) is \(m\)-pseudo-coherent.
Assertions (2) and (3) follow from (1) by rotating the distinguished triangle.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. Let \(m \in \mathbf{Z}\).
If \(K^\bullet\) is \(m\)-pseudo-coherent and \(H^i(K^\bullet) = 0\) for \(i > m\), then \(H^m(K^\bullet)\) is a finite type \(R\)-module.
If \(K^\bullet\) is \(m\)-pseudo-coherent and \(H^i(K^\bullet) = 0\) for \(i > m + 1\), then \(H^{m + 1}(K^\bullet)\) is a finitely presented \(R\)-module.
Proof
Proof of (1). Choose a bounded complex \(E^\bullet\) of finite projective \(R\)-modules and a map \(\alpha : E^\bullet \to K^\bullet\) which induces an isomorphism on cohomology in degrees \(> m\) and a surjection in degree \(m\). It is clear that it suffices to prove the result for \(E^\bullet\). Let \(n\) be the largest integer such that \(E^n \not = 0\). If \(n = m\), then the result is clear. If \(n > m\), then \(E^{n - 1} \to E^n\) is surjective as \(H^n(E^\bullet) = 0\). As \(E^n\) is finite projective we see that \(E^{n - 1} = E' \oplus E^n\). Hence it suffices to prove the result for the complex \((E')^\bullet\) which is the same as \(E^\bullet\) except has \(E'\) in degree \(n - 1\) and \(0\) in degree \(n\). We win by induction on \(n\).
Proof of (2). Choose a bounded complex \(E^\bullet\) of finite projective \(R\)-modules and a map \(\alpha : E^\bullet \to K^\bullet\) which induces an isomorphism on cohomology in degrees \(> m\) and a surjection in degree \(m\). As in the proof of (1) we can reduce to the case that \(E^i = 0\) for \(i > m + 1\). Then we see that \(H^{m + 1}(K^\bullet) \cong H^{m + 1}(E^\bullet) = \Coker(E^m \to E^{m + 1})\) which is of finite presentation.
Lemma
Let \(R\) be a ring. Let \(M\) be an \(R\)-module. Then
\(M\) is \(0\)-pseudo-coherent if and only if \(M\) is a finite \(R\)-module,
\(M\) is \((-1)\)-pseudo-coherent if and only if \(M\) is a finitely presented \(R\)-module,
\(M\) is \((-d)\)-pseudo-coherent if and only if there exists a resolution \[R^{\oplus a_d} \to R^{\oplus a_{d - 1}} \to \ldots \to R^{\oplus a_0} \to M \to 0\] of length \(d\), and
\(M\) is pseudo-coherent if and only if there exists an infinite resolution \[\ldots \to R^{\oplus a_1} \to R^{\oplus a_0} \to M \to 0\] by finite free \(R\)-modules.
Proof
If \(M\) is of finite type (resp. of finite presentation), then \(M\) is \(0\)-pseudo-coherent (resp. \((-1)\)-pseudo-coherent) as follows from the discussion preceding Definition 064Q. Conversely, if \(M\) is \(0\)-pseudo-coherent, then \(M = H^0(M[0])\) is of finite type by Lemma 064S. If \(M\) is \((-1)\)-pseudo-coherent, then it is \(0\)-pseudo-coherent hence of finite type. Choose a surjection \(R^{\oplus a} \to M\) and denote \(K = \Ker(R^{\oplus a} \to M)\). By Lemma 064R we see that \(K\) is \(0\)-pseudo-coherent, hence of finite type, whence \(M\) is of finite presentation.
To prove the third and fourth statement use induction and an argument similar to the above (details omitted).
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. The following are equivalent
\(K^\bullet\) is pseudo-coherent,
\(K^\bullet\) is \(m\)-pseudo-coherent for every \(m \in \mathbf{Z}\), and
\(K^\bullet\) is quasi-isomorphic to a bounded above complex of finite projective \(R\)-modules.
If (1), (2), and (3) hold and \(H^i(K^\bullet) = 0\) for \(i > b\), then we can find a quasi-isomorphism \(F^\bullet \to K^\bullet\) with \(F^i\) finite free \(R\)-modules and \(F^i = 0\) for \(i > b\).
Proof
We see that (1) \(\Rightarrow\) (3) as a finite free module is a finite projective \(R\)-module. Conversely, suppose \(P^\bullet\) is a bounded above complex of finite projective \(R\)-modules. Say \(P^i = 0\) for \(i > n_0\). We choose a direct sum decompositions \(F^{n_0} = P^{n_0} \oplus C^{n_0}\) with \(F^{n_0}\) a finite free \(R\)-module, and inductively \[F^{n - 1} = P^{n - 1} \oplus C^n \oplus C^{n - 1}\] for \(n \leq n_0\) with \(F^{n - 1}\) a finite free \(R\)-module. As a complex \(F^\bullet\) has maps \(F^{n - 1} \to F^n\) which agree with \(P^{n - 1} \to P^n\), induce the identity \(C^n \to C^n\), and are zero on \(C^{n - 1}\). The map \(F^\bullet \to P^\bullet\) is a quasi-isomorphism (even a homotopy equivalence) and hence (3) implies (1).
Assume (1). Let \(E^\bullet\) be a bounded above complex of finite free \(R\)-modules and let \(E^\bullet \to K^\bullet\) be a quasi-isomorphism. Then the induced maps \(\sigma_{\geq m}E^\bullet \to K^\bullet\) from the stupid truncation of \(E^\bullet\) to \(K^\bullet\) show that \(K^\bullet\) is \(m\)-pseudo-coherent. Hence (1) implies (2).
Assume (2). Since \(K^\bullet\) is \(0\)-pseudo-coherent we see in particular that \(K^\bullet\) is bounded above. Let \(b\) be an integer such that \(H^i(K^\bullet) = 0\) for \(i > b\). By descending induction on \(n \in \mathbf{Z}\) we are going to construct finite free \(R\)-modules \(F^i\) for \(i \geq n\), differentials \(d^i : F^i \to F^{i + 1}\) for \(i \geq n\), maps \(\alpha : F^i \to K^i\) compatible with differentials, such that (1) \(H^i(\alpha)\) is an isomorphism for \(i > n\) and surjective for \(i = n\), and (2) \(F^i = 0\) for \(i > b\). Picture \[\xymatrix{ & F^n \ar[r] \ar[d]^\alpha & F^{n + 1} \ar[d]^\alpha \ar[r] & \ldots \\ K^{n - 1} \ar[r] & K^n \ar[r] & K^{n + 1} \ar[r] & \ldots }\] The base case is \(n = b + 1\) where we can take \(F^i = 0\) for all \(i\). Induction step. Let \(C^\bullet\) be the cone on \(\alpha\) (Derived Categories, Definition 014E). The long exact sequence of cohomology \[0 \to H^{n - 1}(K^\bullet) \to H^{n - 1}(C^\bullet) \to H^n(F^\bullet) \to H^n(K^\bullet) \to H^n(C^\bullet) \to \ldots\] shows that \(H^i(C^\bullet) = 0\) for \(i \geq n\). By Lemma 064R we see that \(C^\bullet\) is \((n - 1)\)-pseudo-coherent. By Lemma 064S we see that \(H^{n - 1}(C^\bullet)\) is a finite \(R\)-module. In particular, we see that the kernel of \(H^n(F^\bullet) \to H^n(K^\bullet)\) is a finite \(R\)-module. Choose a finite free \(R\)-module \(F^{n - 1}\) and a map \(F^{n - 1} \to \Ker(F^n \to F^{n + 1})\) such that \(F^{n - 1}\) surjects onto the kernel of \(H^n(F^\bullet) \to H^n(K^\bullet)\). We extend our map of complexes to \[\xymatrix{ & F^{n - 1} \ar[d]^{\alpha^{n - 1}} \ar[r] & F^n \ar[r] \ar[d]^\alpha & F^{n + 1} \ar[d]^\alpha \ar[r] & \ldots \\ \ldots \ar[r] & K^{n - 1} \ar[r] & K^n \ar[r] & K^{n + 1} \ar[r] & \ldots }\] Note that \(\alpha^{n - 1}\) exists because the image of \(F^{n - 1} \to F^n \to K^n\) is in the image of \(K^{n - 1} \to K^n\) by construction. Denote again \(C^\bullet\) the cone of this extended map of complexes. At this point we see that we get an exact sequence \[H^{n - 1}(F^\bullet) \to H^{n - 1}(K^\bullet) \to H^{n - 1}(C^\bullet) \to H^n(F^\bullet) \cong H^n(K^\bullet) \to H^n(C^\bullet) \to \ldots\] In other words, we see that the cokernel of \(H^{n - 1}(F^\bullet) \to H^{n - 1}(K^\bullet)\) is a finite \(R\)-module, say generated by the classes of \(\xi_1, \ldots, \xi_r \in \Ker(K^{n - 1} \to K^n)\). Then we replace \(F^{n - 1}\) by \(F^{n - 1} \oplus R^{\oplus r}\) where the basis elements in the free summand map to zero in \(F^n\) and to \(\xi_i\) in \(K^{n - 1}\). This finishes the proof of the induction step.
Lemma
Let \(R\) be a ring. Let \((K^\bullet, L^\bullet, M^\bullet, f, g, h)\) be a distinguished triangle in \(D(R)\). If two out of three of \(K^\bullet, L^\bullet, M^\bullet\) are pseudo-coherent then the third is also pseudo-coherent.
Proof
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. Let \(m \in \mathbf{Z}\).
If \(H^i(K^\bullet) = 0\) for all \(i \geq m\), then \(K^\bullet\) is \(m\)-pseudo-coherent.
If \(H^i(K^\bullet) = 0\) for \(i > m\) and \(H^m(K^\bullet)\) is a finite \(R\)-module, then \(K^\bullet\) is \(m\)-pseudo-coherent.
If \(H^i(K^\bullet) = 0\) for \(i > m + 1\), the module \(H^{m + 1}(K^\bullet)\) is of finite presentation, and \(H^m(K^\bullet)\) is of finite type, then \(K^\bullet\) is \(m\)-pseudo-coherent.
Proof
It suffices to prove (3). Set \(M = H^{m + 1}(K^\bullet)\). Note that \(\tau_{\geq m + 1}K^\bullet\) is quasi-isomorphic to \(M[- m - 1]\). By Lemma 064T we see that \(M[- m - 1]\) is \(m\)-pseudo-coherent. Since we have the distinguished triangle \[(\tau_{\leq m}K^\bullet, K^\bullet, \tau_{\geq m + 1}K^\bullet)\] (Derived Categories, Remark 08J5) by Lemma 064R it suffices to prove that \(\tau_{\leq m}K^\bullet\) is pseudo-coherent. By assumption \(H^m(\tau_{\leq m}K^\bullet)\) is a finite type \(R\)-module. Hence we can find a finite free \(R\)-module \(E\) and a map \(E \to \Ker(d_K^m)\) such that the composition \(E \to \Ker(d_K^m) \to H^m(\tau_{\leq m}K^\bullet)\) is surjective. Then \(E[-m] \to \tau_{\leq m}K^\bullet\) witnesses the fact that \(\tau_{\leq m}K^\bullet\) is \(m\)-pseudo-coherent.
Lemma
Let \(R\) be a ring. Let \(m \in \mathbf{Z}\). If \(K^\bullet \oplus L^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) so are \(K^\bullet\) and \(L^\bullet\).
Proof
In this proof we drop the superscript \({}^\bullet\). Assume that \(K \oplus L\) is \(m\)-pseudo-coherent. It is clear that \(K, L \in D^{-}(R)\). Note that there is a distinguished triangle \[(K \oplus L, K \oplus L, L \oplus L[1]) = (K, K, 0) \oplus (L, L, L \oplus L[1])\] see Derived Categories, Lemma 05QS. By Lemma 064R we see that \(L \oplus L[1]\) is \(m\)-pseudo-coherent. Hence also \(L[1] \oplus L[2]\) is \(m\)-pseudo-coherent. By induction \(L[n] \oplus L[n + 1]\) is \(m\)-pseudo-coherent. By Lemma 064W we see that \(L[n]\) is \(m\)-pseudo-coherent for large \(n\). Hence working backwards, using the distinguished triangles \[(L[n], L[n] \oplus L[n - 1], L[n - 1])\] we conclude that \(L[n], L[n - 1], \ldots, L\) are \(m\)-pseudo-coherent as desired. The pseudo-coherent case follows from this and Lemma 064U.
Lemma
Let \(R\) be a ring. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a bounded above complex of \(R\)-modules such that \(K^i\) is \((m - i)\)-pseudo-coherent for all \(i\). Then \(K^\bullet\) is \(m\)-pseudo-coherent. In particular, if \(K^\bullet\) is a bounded above complex of pseudo-coherent \(R\)-modules, then \(K^\bullet\) is pseudo-coherent.
Proof
We may replace \(K^\bullet\) by \(\sigma_{\geq m - 1}K^\bullet\) (for example) and hence assume that \(K^\bullet\) is bounded. Then the complex \(K^\bullet\) is \(m\)-pseudo-coherent as each \(K^i[-i]\) is \(m\)-pseudo-coherent by induction on the length of the complex: use Lemma 064R and the stupid truncations. For the final statement, it suffices to prove that \(K^\bullet\) is \(m\)-pseudo-coherent for all \(m \in \mathbf{Z}\), see Lemma 064U. This follows from the first part.
Lemma
Let \(R\) be a ring. Let \(m \in \mathbf{Z}\). Let \(K^\bullet \in D^{-}(R)\) such that \(H^i(K^\bullet)\) is \((m - i)\)-pseudo-coherent (resp. pseudo-coherent) for all \(i\). Then \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent).
Proof
Assume \(K^\bullet\) is an object of \(D^{-}(R)\) such that each \(H^i(K^\bullet)\) is \((m - i)\)-pseudo-coherent. Let \(n\) be the largest integer such that \(H^n(K^\bullet)\) is nonzero. We will prove the lemma by induction on \(n\). If \(n < m\), then \(K^\bullet\) is \(m\)-pseudo-coherent by Lemma 064W. If \(n \geq m\), then we have the distinguished triangle \[(\tau_{\leq n - 1}K^\bullet, K^\bullet, H^n(K^\bullet)[-n])\] (Derived Categories, Remark 08J5) Since \(H^n(K^\bullet)[-n]\) is \(m\)-pseudo-coherent by assumption, we can use Lemma 064R to see that it suffices to prove that \(\tau_{\leq n - 1}K^\bullet\) is \(m\)-pseudo-coherent. By induction on \(n\) we win. (The pseudo-coherent case follows from this and Lemma 064U.)
Lemma
Let \(A \to B\) be a ring map. Assume that \(B\) is pseudo-coherent as an \(A\)-module. Let \(K^\bullet\) be a complex of \(B\)-modules. The following are equivalent
\(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(B\)-modules, and
\(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(A\)-modules.
The same equivalence holds for pseudo-coherence.
Proof
Assume (1). Choose a bounded complex of finite free \(B\)-modules \(E^\bullet\) and a map \(\alpha : E^\bullet \to K^\bullet\) which is an isomorphism on cohomology in degrees \(> m\) and a surjection in degree \(m\). Consider the distinguished triangle \((E^\bullet, K^\bullet, C(\alpha)^\bullet)\). By Lemma 064W \(C(\alpha)^\bullet\) is \(m\)-pseudo-coherent as a complex of \(A\)-modules. Hence it suffices to prove that \(E^\bullet\) is pseudo-coherent as a complex of \(A\)-modules, which follows from Lemma 064Y. The pseudo-coherent case of (1) \(\Rightarrow\) (2) follows from this and Lemma 064U.
Assume (2). Let \(n\) be the largest integer such that \(H^n(K^\bullet) \not = 0\). We will prove that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(B\)-modules by induction on \(n - m\). The case \(n < m\) follows from Lemma 064W. Choose a bounded complex of finite free \(A\)-modules \(E^\bullet\) and a map \(\alpha : E^\bullet \to K^\bullet\) which is an isomorphism on cohomology in degrees \(> m\) and a surjection in degree \(m\). Consider the induced map of complexes \[\alpha \otimes 1 : E^\bullet \otimes_A B \to K^\bullet.\] Note that \(C(\alpha \otimes 1)^\bullet\) is acyclic in degrees \(\geq n\) as \(H^n(E) \to H^n(E^\bullet \otimes_A B) \to H^n(K^\bullet)\) is surjective by construction and since \(H^i(E^\bullet \otimes_A B) = 0\) for \(i > n\) by the spectral sequence of Example 0662. On the other hand, \(C(\alpha \otimes 1)^\bullet\) is \(m\)-pseudo-coherent as a complex of \(A\)-modules because both \(K^\bullet\) and \(E^\bullet \otimes_A B\) (see Lemma 064Y) are so, see Lemma 064R. Hence by induction we see that \(C(\alpha \otimes 1)^\bullet\) is \(m\)-pseudo-coherent as a complex of \(B\)-modules. Finally another application of Lemma 064R shows that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(B\)-modules (as clearly \(E^\bullet \otimes_A B\) is pseudo-coherent as a complex of \(B\)-modules). The pseudo-coherent case of (2) \(\Rightarrow\) (1) follows from this and Lemma 064U.
Lemma
Let \(A \to B\) be a ring map. Let \(K^\bullet\) be an \(m\)-pseudo-coherent (resp. pseudo-coherent) complex of \(A\)-modules. Then \(K^\bullet \otimes_A^{\mathbf{L}} B\) is an \(m\)-pseudo-coherent (resp. pseudo-coherent) complex of \(B\)-modules.
Proof
First we note that the statement of the lemma makes sense as \(K^\bullet\) is bounded above and hence \(K^\bullet \otimes_A^{\mathbf{L}} B\) is defined by Equation (064H). Having said this, choose a bounded complex \(E^\bullet\) of finite free \(A\)-modules and \(\alpha : E^\bullet \to K^\bullet\) with \(H^i(\alpha)\) an isomorphism for \(i > m\) and surjective for \(i = m\). Then the cone \(C(\alpha)^\bullet\) is acyclic in degrees \(\geq m\). Since \(-\otimes_A^{\mathbf{L}} B\) is an exact functor we get a distinguished triangle \[(E^\bullet \otimes_A^{\mathbf{L}} B, K^\bullet \otimes_A^{\mathbf{L}} B, C(\alpha)^\bullet \otimes_A^{\mathbf{L}} B)\] of complexes of \(B\)-modules. By the dual to Derived Categories, Lemma 05TC we see that \(H^i(C(\alpha)^\bullet \otimes_A^{\mathbf{L}} B) = 0\) for \(i \geq m\). Since \(E^\bullet\) is a complex of projective \(A\)-modules we see that \(E^\bullet \otimes_A^{\mathbf{L}} B = E^\bullet \otimes_A B\) and hence \[E^\bullet \otimes_A B \longrightarrow K^\bullet \otimes_A^{\mathbf{L}} B\] is a morphism of complexes of \(B\)-modules that witnesses the fact that \(K^\bullet \otimes_A^{\mathbf{L}} B\) is \(m\)-pseudo-coherent. The case of pseudo-coherent complexes follows from the case of \(m\)-pseudo-coherent complexes via Lemma 064U.
Lemma
Let \(A \to B\) be a flat ring map. Let \(M\) be an \(m\)-pseudo-coherent (resp. pseudo-coherent) \(A\)-module. Then \(M \otimes_A B\) is an \(m\)-pseudo-coherent (resp. pseudo-coherent) \(B\)-module.
Proof
Immediate consequence of Lemma 0650 and the fact that \(M \otimes_A^{\mathbf{L}} B = M \otimes_A B\) because \(B\) is flat over \(A\).
The following lemma also follows from the stronger Lemma 068R.
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\) be elements which generate the unit ideal. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(R\)-modules. If for each \(i\) the complex \(K^\bullet \otimes_R R_{f_i}\) is \(m\)-pseudo-coherent (resp. pseudo-coherent), then \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent).
Proof
We will use without further mention that \(- \otimes_R R_{f_i}\) is an exact functor and that therefore \[H^i(K^\bullet)_{f_i} = H^i(K^\bullet) \otimes_R R_{f_i} = H^i(K^\bullet \otimes_R R_{f_i}).\] Assume \(K^\bullet \otimes_R R_{f_i}\) is \(m\)-pseudo-coherent for \(i = 1, \ldots, r\). Let \(n \in \mathbf{Z}\) be the largest integer such that \(H^n(K^\bullet \otimes_R R_{f_i})\) is nonzero for some \(i\). This implies in particular that \(H^i(K^\bullet) = 0\) for \(i > n\) (and that \(H^n(K^\bullet) \not = 0\)) see Algebra, Lemma 00EO. We will prove the lemma by induction on \(n - m\). If \(n < m\), then the lemma is true by Lemma 064W. If \(n \geq m\), then \(H^n(K^\bullet)_{f_i}\) is a finite \(R_{f_i}\)-module for each \(i\), see Lemma 064S. Hence \(H^n(K^\bullet)\) is a finite \(R\)-module, see Algebra, Lemma 00EO. Choose a finite free \(R\)-module \(E\) and a surjection \(E \to H^n(K^\bullet)\). As \(E\) is projective we can lift this to a map of complexes \(\alpha : E[-n] \to K^\bullet\). Then the cone \(C(\alpha)^\bullet\) has vanishing cohomology in degrees \(\geq n\). On the other hand, the complexes \(C(\alpha)^\bullet \otimes_R R_{f_i}\) are \(m\)-pseudo-coherent for each \(i\), see Lemma 064R. Hence by induction we see that \(C(\alpha)^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules. Applying Lemma 064R once more we conclude.
Lemma
Let \(R\) be a ring. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(R\)-modules. Let \(R \to R'\) be a faithfully flat ring map. If the complex \(K^\bullet \otimes_R R'\) is \(m\)-pseudo-coherent (resp. pseudo-coherent), then \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent).
Proof
We will use without further mention that \(- \otimes_R R'\) is an exact functor and that therefore \[H^i(K^\bullet) \otimes_R R' = H^i(K^\bullet \otimes_R R').\] Assume \(K^\bullet \otimes_R R'\) is \(m\)-pseudo-coherent. Let \(n \in \mathbf{Z}\) be the largest integer such that \(H^n(K^\bullet)\) is nonzero; then \(n\) is also the largest integer such that \(H^n(K^\bullet \otimes_R R')\) is nonzero. We will prove the lemma by induction on \(n - m\). If \(n < m\), then the lemma is true by Lemma 064W. If \(n \geq m\), then \(H^n(K^\bullet) \otimes_R R'\) is a finite \(R'\)-module, see Lemma 064S. Hence \(H^n(K^\bullet)\) is a finite \(R\)-module, see Algebra, Lemma 03C4. Choose a finite free \(R\)-module \(E\) and a surjection \(E \to H^n(K^\bullet)\). As \(E\) is projective we can lift this to a map of complexes \(\alpha : E[-n] \to K^\bullet\). Then the cone \(C(\alpha)^\bullet\) has vanishing cohomology in degrees \(\geq n\). On the other hand, the complex \(C(\alpha)^\bullet \otimes_R R'\) is \(m\)-pseudo-coherent, see Lemma 064R. Hence by induction we see that \(C(\alpha)^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules. Applying Lemma 064R once more we conclude.
Lemma
Let \(R\) be a ring. Let \(K, L\) be objects of \(D(R)\).
If \(K\) is \(n\)-pseudo-coherent and \(H^i(K) = 0\) for \(i > a\) and \(L\) is \(m\)-pseudo-coherent and \(H^j(L) = 0\) for \(j > b\), then \(K \otimes_R^\mathbf{L} L\) is \(t\)-pseudo-coherent with \(t = \max(m + a, n + b)\).
If \(K\) and \(L\) are pseudo-coherent, then \(K \otimes_R^\mathbf{L} L\) is pseudo-coherent.
Proof
Proof of (1). We may assume there exist bounded complexes \(K^\bullet\) and \(L^\bullet\) of finite free \(R\)-modules and maps \(\alpha : K^\bullet \to K\) and \(\beta : L^\bullet \to L\) with \(H^i(\alpha)\) and isomorphism for \(i > n\) and surjective for \(i = n\) and with \(H^i(\beta)\) and isomorphism for \(i > m\) and surjective for \(i = m\). Then the map \[\alpha \otimes^\mathbf{L} \beta : \text{Tot}(K^\bullet \otimes_R L^\bullet) \to K \otimes_R^\mathbf{L} L\] induces isomorphisms on cohomology in degree \(i\) for \(i > t\) and a surjection for \(i = t\). This follows from the spectral sequence of tors (details omitted). Part (2) follows from part (1) and Lemma 064U.
Lemma
Let \(R\) be a Noetherian ring. Then
A complex of \(R\)-modules \(K^\bullet\) is \(m\)-pseudo-coherent if and only if \(K^\bullet \in D^{-}(R)\) and \(H^i(K^\bullet)\) is a finite \(R\)-module for \(i \geq m\).
A complex of \(R\)-modules \(K^\bullet\) is pseudo-coherent if and only if \(K^\bullet \in D^{-}(R)\) and \(H^i(K^\bullet)\) is a finite \(R\)-module for all \(i\).
An \(R\)-module is pseudo-coherent if and only if it is finite.
Proof
In Algebra, Lemma 00LP we have seen that any finite \(R\)-module is pseudo-coherent. On the other hand, a pseudo-coherent module is finite, see Lemma 064T. Hence (3) holds. Suppose that \(K^\bullet\) is an \(m\)-pseudo-coherent complex. Then there exists a bounded complex of finite free \(R\)-modules \(E^\bullet\) such that \(H^i(K^\bullet)\) is isomorphic to \(H^i(E^\bullet)\) for \(i > m\) and such that \(H^m(K^\bullet)\) is a quotient of \(H^m(E^\bullet)\). Thus it is clear that each \(H^i(K^\bullet)\), \(i \geq m\) is a finite module. The converse implication in (1) follows from Lemma 066B and part (3). Part (2) follows from (1) and Lemma 064U.
Lemma
Let \(R\) be a coherent ring (Algebra, Definition 05CV). Let \(K \in D^-(R)\). The following are equivalent
\(K\) is \(m\)-pseudo-coherent,
\(H^m(K)\) is a finite \(R\)-module and \(H^i(K)\) is coherent for \(i > m\), and
\(H^m(K)\) is a finite \(R\)-module and \(H^i(K)\) is finitely presented for \(i > m\).
Thus \(K\) is pseudo-coherent if and only if \(H^i(K)\) is a coherent module for all \(i\).
Proof
Recall that an \(R\)-module \(M\) is coherent if and only if it is of finite presentation (Algebra, Lemma 05CX). This explains the equivalence of (2) and (3). If so and if we choose an exact sequence \(0 \to N \to R^{\oplus m} \to M \to 0\), then \(N\) is coherent by Algebra, Lemma 05CW. Thus in this case, repeating this procedure with \(N\) we find a resolution \[\ldots \to R^{\oplus n} \to R^{\oplus m} \to M \to 0\] by finite free \(R\)-modules. In other words, \(M\) is pseudo-coherent. The equivalence of (1) and (2) follows from this and Lemmas 066B and 064T. The final assertion follows from the equivalence of (1) and (2) combined with Lemma 064U.
Pseudo-coherent modules, II
We continue the discussion started in Section 064N.
Lemma
Let \(R\) be a ring. Let \(M = \colim M_i\) be a filtered colimit of \(R\)-modules. Let \(K \in D(R)\) be \(m\)-pseudo-coherent. Then \(\colim \Ext^n_R(K, M_i) = \Ext^n_R(K, M)\) for \(n < -m\) and \(\colim \Ext^{-m}_R(K, M_i) \to \Ext^{-m}_R(K, M)\) is injective.
Proof
By definition we can find a distinguished triangle \[E \to K \to L \to E[1]\] in \(D(R)\) such that \(E\) is represented by a bounded complex of finite free \(R\)-modules and such that \(H^i(L) = 0\) for \(i \geq m\). Then \(\Ext^n_R(L, N) = 0\) for any \(R\)-module \(N\) and \(n \leq -m\), see Derived Categories, Lemma 06XS. By the long exact sequence of \(\Ext\) associated to the distinguished triangle we see that \(\Ext^n_R(K, N) \to \Ext^n_R(E, N)\) is an isomorphism for \(n < -m\) and injective for \(n = -m\). Thus it suffices to prove that \(M \mapsto \Ext_R^n(E, M)\) commutes with filtered colimits when \(E\) can be represented by a bounded complex of finite free \(R\)-modules \(E^\bullet\). The modules \(\Ext^n_R(E, M)\) are computed by the complex \(\Hom_R(E^\bullet, M)\), see Derived Categories, Lemma 064B. The functor \(M \mapsto \Hom_R(E^p, M)\) commutes with filtered colimits as \(E^p\) is finite free. Thus \(\Hom_R(E^\bullet, M) = \colim \Hom_R(E^\bullet, M_i)\) as complexes. Since filtered colimits are exact (Algebra, Lemma 00DB) we conclude.
Lemma
Let \(R\) be a ring. Let \(K \in D^-(R)\). Let \(m \in \mathbf{Z}\). Then \(K\) is \(m\)-pseudo-coherent if and only if for any filtered colimit \(M = \colim M_i\) of \(R\)-modules we have \(\colim \Ext^n_R(K, M_i) = \Ext^n_R(K, M)\) for \(n < -m\) and \(\colim \Ext^{-m}_R(K, M_i) \to \Ext^{-m}_R(K, M)\) is injective.
Proof
One implication was shown in Lemma 0G8W. Assume for any filtered colimit \(M = \colim M_i\) of \(R\)-modules we have \(\colim \Ext^n_R(K, M_i) = \Ext^n_R(K, M)\) for \(n < -m\) and \(\colim \Ext^{-m}_R(K, M_i) \to \Ext^{-m}_R(K, M)\) is injective. We will show \(K\) is \(m\)-pseudo-coherent.
Let \(t\) be the maximal integer such that \(H^t(K)\) is nonzero. We will use induction on \(t\). If \(t < m\), then \(K\) is \(m\)-pseudo-coherent by Lemma 064W. If \(t \geq m\), then since \(\Hom_R(H^t(K), M) = \Ext^{-t}_R(K, M)\) we conclude that \(\colim \Hom_R(H^t(K), M_i) \to \Hom_R(H^t(K), M)\) is injective for any filtered colimit \(M = \colim M_i\). This implies that \(H^t(K)\) is a finite \(R\)-module by Algebra, Lemma 0G8N. Choose a finite free \(R\)-module \(F\) and a surjection \(F \to H^t(K)\). We can lift this to a morphism \(F[-t] \to K\) in \(D(R)\) and choose a distinguished triangle \[F[-t] \to K \to L \to F[-t + 1]\] in \(D(R)\). Then \(H^i(L) = 0\) for \(i \geq t\). Moreover, the long exact sequence of \(\Ext\) associated to this distinguished triangle shows that \(L\) inherits the assumption we made on \(K\) by a small argument we omit. By induction on \(t\) we conclude that \(L\) is \(m\)-pseudo-coherent. Hence \(K\) is \(m\)-pseudo-coherent by Lemma 064R.
Lemma
Let \(R\) be a ring. Let \(L\), \(M\), \(N\) be \(R\)-modules.
If \(M\) is finitely presented and \(L\) is flat, then the canonical map \(\Hom_R(M, N) \otimes_R L \to \Hom_R(M, N \otimes_R L)\) is an isomorphism.
If \(M\) is \((-m)\)-pseudo-coherent and \(L\) is flat, then the canonical map \(\Ext^i_R(M, N) \otimes_R L \to \Ext^i_R(M, N \otimes_R L)\) is an isomorphism for \(i < m\).
Proof
Choose a resolution \(F_\bullet \to M\) whose terms are free \(R\)-modules, see Algebra, Lemma 00LP. The complex \(\Hom_R(F_\bullet, N)\) computes \(\Ext^i_R(M, N)\) and the complex \(\Hom_R(F_\bullet, N \otimes_R L)\) computes \(\Ext^i_R(M, N \otimes_R L)\). There always is a map of cochain complexes \[\Hom_R(F_\bullet, N) \otimes_R L \longrightarrow \Hom_R(F_\bullet, N \otimes_R L)\] which induces canonical maps \(\Ext^i_R(M, N) \otimes_R L \to \Ext^i_R(M, N \otimes_R L)\) for all \(i \geq 0\) (canonical for example in the sense that these maps do not depend on the choice of the resolution \(F_\bullet\)). If \(L\) is flat, then the complex \(\Hom_R(F_\bullet, N) \otimes_R L\) computes \(\Ext^i_R(M, N) \otimes_R L\) since taking cohomology commutes with tensoring by \(L\).
Having said all of the above, if \(M\) is \((-m)\)-pseudo-coherent, then we may choose \(F_\bullet\) such that \(F_i\) is finite free for \(i = 0, \ldots, m\). Then the map of cochain complexes displayed above is an isomorphism in degrees \(\leq m\) and hence an isomorphism on cohomology groups in degrees \(< m\). This proves (2). If \(M\) is finitely presented, then \(M\) is \((-1)\)-pseudo-coherent by Lemma 064T and we get the result because \(\Hom = \Ext^0\).
Lemma
Let \(R \to R'\) be a flat ring map. Let \(M\), \(N\) be \(R\)-modules.
If \(M\) is a finitely presented \(R\)-module, then \(\Hom_R(M, N) \otimes_R R' = \Hom_{R'}(M \otimes_R R', N \otimes_R R')\).
If \(M\) is \((-m)\)-pseudo-coherent, then \(\Ext^i_R(M, N) \otimes_R R' = \Ext^i_{R'}(M \otimes_R R', N \otimes_R R')\) for \(i < m\).
In particular if \(R\) is Noetherian and \(M\) is a finite module this holds for all \(i\).
Proof
By Algebra, Lemma 087N we have \(\Ext^i_{R'}(M \otimes_R R', N \otimes_R R') = \Ext^i_R(M, N \otimes_R R')\). Combined with Lemma 087Q we conclude (1) and (2) holds. The final statement follows from this and Lemma 066E.
Lemma
Let \(R\) be a ring. Let \(K \in D^-(R)\). The following are equivalent:
\(K\) is pseudo-coherent,
for every family \((Q_{\alpha})_{\alpha \in A}\) of \(R\)-modules, the canonical map \[\alpha : K \otimes_R^\mathbf{L} \left( \prod\nolimits_\alpha Q_{\alpha} \right) \longrightarrow \prod\nolimits_\alpha (K \otimes_R^\mathbf{L} Q_{\alpha})\] is an isomorphism in \(D(R)\),
for every \(R\)-module \(Q\) and every set \(A\), the canonical map \[\beta : K \otimes_R^\mathbf{L} Q^A \longrightarrow (K \otimes_R^\mathbf{L} Q)^A\] is an isomorphism in \(D(R)\), and
for every set \(A\), the canonical map \[\gamma : K \otimes_R^\mathbf{L} R^A \longrightarrow K^A\] is an isomorphism in \(D(R)\).
Given \(m \in \mathbf{Z}\) the following are equivalent
\(K\) is \(m\)-pseudo-coherent,
for every family \((Q_{\alpha})_{\alpha \in A}\) of \(R\)-modules, with \(\alpha\) as above \(H^i(\alpha)\) is an isomorphism for \(i > m\) and surjective for \(i = m\),
for every \(R\)-module \(Q\) and every set \(A\), with \(\beta\) as above \(H^i(\beta)\) is an isomorphism for \(i > m\) and surjective for \(i = m\),
for every set \(A\), with \(\gamma\) as above \(H^i(\gamma)\) is an isomorphism for \(i > m\) and surjective for \(i = m\).
Proof
If \(K\) is pseudo-coherent, then \(K\) can be represented by a bounded above complex of finite free \(R\)-modules. Then the derived tensor products are computed by tensoring with this complex. Also, products in \(D(R)\) are given by taking products of any choices of representative complexes. Hence (1) implies (2), (3), (4) by the corresponding fact for modules, see Algebra, Proposition 059K.
In the same way (using the tensor product is right exact) the reader shows that (a) implies (b), (c), and (d).
Assume (4) holds. To show that \(K\) is pseudo-coherent it suffices to show that \(K\) is \(m\)-pseudo-coherent for all \(m\) (Lemma 064U). Hence to finish then proof it suffices to prove that (d) implies (a).
Assume (d). Let \(i\) be the largest integer such that \(H^i(K)\) is nonzero. If \(i < m\), then we are done. If not, then from (d) and the description of products in \(D(R)\) given above we find that \(H^i(K) \otimes_R R^A \to H^i(K)^A\) is surjective. Hence \(H^i(K)\) is a finitely generated \(R\)-module by Algebra, Proposition 059J. Thus we may choose a complex \(L\) consisting of a single finite free module sitting in degree \(i\) and a map of complexes \(L \to K\) such that \(H^i(L) \to H^i(K)\) is surjective. In particular \(L\) satisfies (1), (2), (3), and (4). Choose a distinguished triangle \[L \to K \to M \to L[1]\] Then we see that \(H^j(M) = 0\) for \(j \geq i\). On the other hand, \(M\) still has property (d) by a small argument which we omit. By induction on \(i\) we find that \(M\) is \(m\)-pseudo-coherent. Hence \(K\) is \(m\)-pseudo-coherent by Lemma 064R.
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\) be pseudo-coherent. Let \(i \in \mathbf{Z}\). There exists a finitely presented \(R\)-module \(M\) and a map \(K \to M[-i]\) in \(D(R)\) which induces an injection \(H^i(K) \to M\).
Proof
By Definition 064Q we may represent \(K\) by a complex \(P^\bullet\) of finite free \(R\)-modules. Set \(M = \Coker(P^{i - 1} \to P^i)\).
Lemma
Let \(A\) be a Noetherian ring. Let \(K \in D(A)\) be pseudo-coherent, i.e., \(K \in D^-(A)\) with finite cohomology modules. Let \(\mathfrak m\) be a maximal ideal of \(A\). If \(H^i(K)/\mathfrak m H^i(K) \not = 0\), then there exists a finite \(A\)-module \(E\) annihilated by a power of \(\mathfrak m\) and a map \(K \to E[-i]\) which is nonzero on \(H^i(K)\).
Proof
(The equivalent formulation of pseudo-coherence in the statement of the lemma is Lemma 066E.) Choose \(K \to M[-i]\) as in Lemma 0G8Y. By Artin-Rees (Algebra, Lemma 00IN) we can find an \(n\) such that \(H^i(K) \cap \mathfrak m^n M \subset \mathfrak m H^i(K)\). Take \(E = M/\mathfrak m^n M\).
Tor dimension
Instead of resolving by projective modules we can look at resolutions by flat modules. This leads to the following concept.
Definition
Let \(R\) be a ring. Denote \(D(R)\) its derived category. Let \(a, b \in \mathbf{Z}\).
An object \(K^\bullet\) of \(D(R)\) has tor-amplitude in \([a, b]\) if \(H^i(K^\bullet \otimes_R^\mathbf{L} M) = 0\) for all \(R\)-modules \(M\) and all \(i \not \in [a, b]\).
An object \(K^\bullet\) of \(D(R)\) has finite tor dimension if it has tor-amplitude in \([a, b]\) for some \(a, b\).
An \(R\)-module \(M\) has tor dimension \(\leq d\) if \(M[0]\) as an object of \(D(R)\) has tor-amplitude in \([-d, 0]\).
An \(R\)-module \(M\) has finite tor dimension if \(M[0]\) as an object of \(D(R)\) has finite tor dimension.
We observe that if \(K^\bullet\) has finite tor dimension, then \(K^\bullet \in D^b(R)\).
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a bounded above complex of flat \(R\)-modules with tor-amplitude in \([a, b]\). Then \(\Coker(d_K^{a - 1})\) is a flat \(R\)-module.
Proof
As \(K^\bullet\) is a bounded above complex of flat modules we see that \(K^\bullet \otimes_R M = K^\bullet \otimes_R^{\mathbf{L}} M\). Hence for every \(R\)-module \(M\) the sequence \[K^{a - 2} \otimes_R M \to K^{a - 1} \otimes_R M \to K^a \otimes_R M\] is exact in the middle. Since \(K^{a - 2} \to K^{a - 1} \to K^a \to \Coker(d_K^{a - 1}) \to 0\) is a flat resolution this implies that \(\text{Tor}_1^R(\Coker(d_K^{a - 1}), M) = 0\) for all \(R\)-modules \(M\). This means that \(\Coker(d_K^{a - 1})\) is flat, see Algebra, Lemma 00M5.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be an object of \(D(R)\). Let \(a, b \in \mathbf{Z}\). The following are equivalent
\(K^\bullet\) has tor-amplitude in \([a, b]\).
\(K^\bullet\) is quasi-isomorphic to a complex \(E^\bullet\) of flat \(R\)-modules with \(E^i = 0\) for \(i \not \in [a, b]\).
Proof
If (2) holds, then we may compute \(K^\bullet \otimes_R^\mathbf{L} M = E^\bullet \otimes_R M\) and it is clear that (1) holds. Assume that (1) holds. We may replace \(K^\bullet\) by a projective resolution with \(K^i = 0\) for \(i > b\). See Derived Categories, Lemma 0646. Set \(E^\bullet = \tau_{\geq a}K^\bullet\). Everything is clear except that \(E^a\) is flat which follows immediately from Lemma 0653 and the definitions.
Lemma
Let \(R\) be a ring. Let \(a \in \mathbf{Z}\) and let \(K\) be an object of \(D(R)\). The following are equivalent
\(K\) has tor-amplitude in \([a, \infty]\), and
\(K\) is quasi-isomorphic to a K-flat complex \(E^\bullet\) whose terms are flat \(R\)-modules with \(E^i = 0\) for \(i \not \in [a, \infty]\).
Proof
The implication (2) \(\Rightarrow\) (1) is immediate. Assume (1) holds. First we choose a K-flat complex \(K^\bullet\) with flat terms representing \(K\), see Lemma 06Y4. For any \(R\)-module \(M\) the cohomology of \[K^{n - 1} \otimes_R M \to K^n \otimes_R M \to K^{n + 1} \otimes_R M\] computes \(H^n(K \otimes_R^\mathbf{L} M)\). This is always zero for \(n < a\). Hence if we apply Lemma 0653 to the complex \(\ldots \to K^{a - 1} \to K^a \to K^{a + 1}\) we conclude that \(N = \Coker(K^{a - 1} \to K^a)\) is a flat \(R\)-module. We set \[E^\bullet = \tau_{\geq a}K^\bullet = (\ldots \to 0 \to N \to K^{a + 1} \to \ldots )\] The kernel \(L^\bullet\) of \(K^\bullet \to E^\bullet\) is the complex \[L^\bullet = (\ldots \to K^{a - 1} \to I \to 0 \to \ldots)\] where \(I \subset K^a\) is the image of \(K^{a - 1} \to K^a\). Since we have the short exact sequence \(0 \to I \to K^a \to N \to 0\) we see that \(I\) is a flat \(R\)-module. Thus \(L^\bullet\) is a bounded above complex of flat modules, hence K-flat by Lemma 064K. It follows that \(E^\bullet\) is K-flat by Lemma 0BYH.
Lemma
Let \(R\) be a ring. Let \((K^\bullet, L^\bullet, M^\bullet, f, g, h)\) be a distinguished triangle in \(D(R)\). Let \(a, b \in \mathbf{Z}\).
If \(K^\bullet\) has tor-amplitude in \([a + 1, b + 1]\) and \(L^\bullet\) has tor-amplitude in \([a, b]\) then \(M^\bullet\) has tor-amplitude in \([a, b]\).
If \(K^\bullet, M^\bullet\) have tor-amplitude in \([a, b]\), then \(L^\bullet\) has tor-amplitude in \([a, b]\).
If \(L^\bullet\) has tor-amplitude in \([a + 1, b + 1]\) and \(M^\bullet\) has tor-amplitude in \([a, b]\), then \(K^\bullet\) has tor-amplitude in \([a + 1, b + 1]\).
Proof
Omitted. Hint: This just follows from the long exact cohomology sequence associated to a distinguished triangle and the fact that \(- \otimes_R^{\mathbf{L}} M\) preserves distinguished triangles. The easiest one to prove is (2) and the others follow from it by translation.
Lemma
Let \(R\) be a ring. Let \(M\) be an \(R\)-module. Let \(d \geq 0\). The following are equivalent
\(M\) has tor dimension \(\leq d\), and
there exists a resolution \[0 \to F_d \to \ldots \to F_1 \to F_0 \to M \to 0\] with \(F_i\) a flat \(R\)-module.
In particular an \(R\)-module has tor dimension \(0\) if and only if it is a flat \(R\)-module.
Proof
Assume (2). Then the complex \(E^\bullet\) with \(E^{-i} = F_i\) is quasi-isomorphic to \(M\). Hence the Tor dimension of \(M\) is at most \(d\) by Lemma 0654. Conversely, assume (1). Let \(P^\bullet \to M\) be a projective resolution of \(M\). By Lemma 0653 we see that \(\tau_{\geq -d}P^\bullet\) is a flat resolution of \(M\) of length \(d\), i.e., (2) holds.
Lemma
Let \(R\) be a ring. Let \(a, b \in \mathbf{Z}\). If \(K^\bullet \oplus L^\bullet\) has tor amplitude in \([a, b]\) so do \(K^\bullet\) and \(L^\bullet\).
Proof
Clear from the fact that the Tor functors are additive.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a bounded complex of \(R\)-modules such that \(K^i\) has tor amplitude in \([a - i, b - i]\) for all \(i\). Then \(K^\bullet\) has tor amplitude in \([a, b]\). In particular if \(K^\bullet\) is a finite complex of \(R\)-modules of finite tor dimension, then \(K^\bullet\) has finite tor dimension.
Proof
Follows by induction on the length of the finite complex: use Lemma 0655 and the stupid truncations.
Lemma
Let \(R\) be a ring. Let \(a, b \in \mathbf{Z}\). Let \(K^\bullet \in D^b(R)\) such that \(H^i(K^\bullet)\) has tor amplitude in \([a - i, b - i]\) for all \(i\). Then \(K^\bullet\) has tor amplitude in \([a, b]\). In particular if \(K^\bullet \in D^b(R)\) and all its cohomology groups have finite tor dimension then \(K^\bullet\) has finite tor dimension.
Proof
Follows by induction on the length of the finite complex: use Lemma 0655 and the canonical truncations.
Lemma
Let \(A \to B\) be a ring map. Let \(K^\bullet\) and \(L^\bullet\) be complexes of \(B\)-modules. Let \(a, b, c, d \in \mathbf{Z}\). If
\(K^\bullet\) as a complex of \(B\)-modules has tor amplitude in \([a, b]\),
\(L^\bullet\) as a complex of \(A\)-modules has tor amplitude in \([c, d]\),
then \(K^\bullet \otimes^\mathbf{L}_B L^\bullet\) as a complex of \(A\)-modules has tor amplitude in \([a + c, b + d]\).
Proof
We may assume that \(K^\bullet\) is a complex of flat \(B\)-modules with \(K^i = 0\) for \(i \not \in [a, b]\), see Lemma 0654. Let \(M\) be an \(A\)-module. Choose a free resolution \(F^\bullet \to M\). Then \[(K^\bullet \otimes_B^\mathbf{L} L^\bullet) \otimes_A^{\mathbf{L}} M = \text{Tot}(\text{Tot}(K^\bullet \otimes_B L^\bullet) \otimes_A F^\bullet) = \text{Tot}(K^\bullet \otimes_B \text{Tot}(L^\bullet \otimes_A F^\bullet))\] see Homology, Remark 08BI for the second equality. By assumption (2) the complex \(\text{Tot}(L^\bullet \otimes_A F^\bullet)\) has nonzero cohomology only in degrees \([c, d]\). Hence the spectral sequence of Homology, Lemma 0130 for the double complex \(K^\bullet \otimes_B \text{Tot}(L^\bullet \otimes_A F^\bullet)\) proves that \((K^\bullet \otimes_B^\mathbf{L} L^\bullet) \otimes_A^{\mathbf{L}} M\) has nonzero cohomology only in degrees \([a + c, b + d]\).
Lemma
Let \(A \to B\) be a ring map. Assume that \(B\) is flat as an \(A\)-module. Let \(K^\bullet\) be a complex of \(B\)-modules. Let \(a, b \in \mathbf{Z}\). If \(K^\bullet\) as a complex of \(B\)-modules has tor amplitude in \([a, b]\), then \(K^\bullet\) as a complex of \(A\)-modules has tor amplitude in \([a, b]\).
Proof
This is a special case of Lemma 0B66, but can also be seen directly as follows. We have \(K^\bullet \otimes_A^{\mathbf{L}} M = K^\bullet \otimes_B^{\mathbf{L}} (M \otimes_A B)\) since any projective resolution of \(K^\bullet\) as a complex of \(B\)-modules is a flat resolution of \(K^\bullet\) as a complex of \(A\)-modules and can be used to compute \(K^\bullet \otimes_A^{\mathbf{L}} M\).
Lemma
Let \(A \to B\) be a ring map. Assume that \(B\) has tor dimension \(\leq d\) as an \(A\)-module. Let \(K^\bullet\) be a complex of \(B\)-modules. Let \(a, b \in \mathbf{Z}\). If \(K^\bullet\) as a complex of \(B\)-modules has tor amplitude in \([a, b]\), then \(K^\bullet\) as a complex of \(A\)-modules has tor amplitude in \([a - d, b]\).
Proof
This is a special case of Lemma 0B66, but can also be seen directly as follows. Let \(M\) be an \(A\)-module. Choose a free resolution \(F^\bullet \to M\). Then \[K^\bullet \otimes_A^{\mathbf{L}} M = \text{Tot}(K^\bullet \otimes_A F^\bullet) = \text{Tot}(K^\bullet \otimes_B (F^\bullet \otimes_A B)) = K^\bullet \otimes_B^{\mathbf{L}} (M \otimes_A^{\mathbf{L}} B).\] By our assumption on \(B\) as an \(A\)-module we see that \(M \otimes_A^{\mathbf{L}} B\) has cohomology only in degrees \(-d, -d + 1, \ldots, 0\). Because \(K^\bullet\) has tor amplitude in \([a, b]\) we see from the spectral sequence in Example 0662 that \(K^\bullet \otimes_B^{\mathbf{L}} (M \otimes_A^{\mathbf{L}} B)\) has cohomology only in degrees \([-d + a, b]\) as desired.
Lemma
Let \(A \to B\) be a ring map. Let \(a, b \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(A\)-modules with tor amplitude in \([a, b]\). Then \(K^\bullet \otimes_A^{\mathbf{L}} B\) as a complex of \(B\)-modules has tor amplitude in \([a, b]\).
Proof
By Lemma 0654 we can find a quasi-isomorphism \(E^\bullet \to K^\bullet\) where \(E^\bullet\) is a complex of flat \(A\)-modules with \(E^i = 0\) for \(i \not \in [a, b]\). Then \(E^\bullet \otimes_A B\) computes \(K^\bullet \otimes_A ^{\mathbf{L}} B\) by construction and each \(E^i \otimes_A B\) is a flat \(B\)-module by Algebra, Lemma 00HI. Hence we conclude by Lemma 0654.
Lemma
Let \(A \to B\) be a flat ring map. Let \(d \geq 0\). Let \(M\) be an \(A\)-module of tor dimension \(\leq d\). Then \(M \otimes_A B\) is a \(B\)-module of tor dimension \(\leq d\).
Proof
Immediate consequence of Lemma 066L and the fact that \(M \otimes_A^{\mathbf{L}} B = M \otimes_A B\) because \(B\) is flat over \(A\).
Lemma
Let \(A \to B\) be a ring map. Let \(K^\bullet\) be a complex of \(B\)-modules. Let \(a, b \in \mathbf{Z}\). The following are equivalent
\(K^\bullet\) has tor amplitude in \([a, b]\) as a complex of \(A\)-modules,
\(K^\bullet_\mathfrak q\) has tor amplitude in \([a, b]\) as a complex of \(A_\mathfrak p\)-modules for every prime \(\mathfrak q \subset B\) with \(\mathfrak p = A \cap \mathfrak q\),
\(K^\bullet_\mathfrak m\) has tor amplitude in \([a, b]\) as a complex of \(A_\mathfrak p\)-modules for every maximal ideal \(\mathfrak m \subset B\) with \(\mathfrak p = A \cap \mathfrak m\).
Proof
Assume (3) and let \(M\) be an \(A\)-module. Then \(H^i = H^i(K^\bullet \otimes_A^\mathbf{L} M)\) is a \(B\)-module and \((H^i)_\mathfrak m = H^i(K^\bullet_\mathfrak m \otimes_{A_\mathfrak p}^\mathbf{L} M_\mathfrak p)\). Hence \(H^i = 0\) for \(i \not \in [a, b]\) by Algebra, Lemma 00HN. Thus (3) \(\Rightarrow\) (1). We omit the proofs of (1) \(\Rightarrow\) (2) and (2) \(\Rightarrow\) (3).
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\) be elements which generate the unit ideal. Let \(a, b \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(R\)-modules. If for each \(i\) the complex \(K^\bullet \otimes_R R_{f_i}\) has tor amplitude in \([a, b]\), then \(K^\bullet\) has tor amplitude in \([a, b]\).
Proof
This follows immediately from Lemma 0B67 but can also be seen directly as follows. Note that \(- \otimes_R R_{f_i}\) is an exact functor and that therefore \[H^i(K^\bullet)_{f_i} = H^i(K^\bullet) \otimes_R R_{f_i} = H^i(K^\bullet \otimes_R R_{f_i}).\] and similarly for every \(R\)-module \(M\) we have \[H^i(K^\bullet \otimes_R^{\mathbf{L}} M)_{f_i} = H^i(K^\bullet \otimes_R^{\mathbf{L}} M) \otimes_R R_{f_i} = H^i(K^\bullet \otimes_R R_{f_i} \otimes_{R_{f_i}}^{\mathbf{L}} M_{f_i}).\] Hence the result follows from the fact that an \(R\)-module \(N\) is zero if and only if \(N_{f_i}\) is zero for each \(i\), see Algebra, Lemma 00EO.
Lemma
Let \(R\) be a ring. Let \(a, b \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(R\)-modules. Let \(R \to R'\) be a faithfully flat ring map. If the complex \(K^\bullet \otimes_R R'\) has tor amplitude in \([a, b]\), then \(K^\bullet\) has tor amplitude in \([a, b]\).
Proof
Let \(M\) be an \(R\)-module. Since \(R \to R'\) is flat we see that \[(M \otimes_R^{\mathbf{L}} K^\bullet) \otimes_R R' = (M \otimes_R R') \otimes_{R'}^{\mathbf{L}} (K^\bullet \otimes_R R')\] and taking cohomology commutes with tensoring with \(R'\). Hence \(\text{Tor}_i^R(M, K^\bullet) \otimes_R R' = \text{Tor}_i^{R'}(M \otimes_R R', K^\bullet \otimes_R R')\). Since \(R \to R'\) is faithfully flat, the vanishing of \(\text{Tor}_i^{R'}(M \otimes_R R', K^\bullet \otimes_R R')\) for \(i \not \in [a, b]\) implies the same thing for \(\text{Tor}_i^R(M, K^\bullet)\).
Lemma
Given ring maps \(R \to A \to B\) with \(A \to B\) faithfully flat and \(K \in D(A)\) the tor amplitude of \(K\) over \(R\) is the same as the tor amplitude of \(K \otimes_A^\mathbf{L} B\) over \(R\).
Proof
This is true because for an \(R\)-module \(M\) we have \(H^i(K \otimes_R^\mathbf{L} M) \otimes_A B = H^i((K \otimes_A^\mathbf{L} B) \otimes_R^\mathbf{L} M)\) for all \(i\). Namely, represent \(K\) by a complex \(K^\bullet\) of \(A\)-modules and choose a free resolution \(F^\bullet \to M\). Then we have the equality \[\text{Tot}(K^\bullet \otimes_A B \otimes_R F^\bullet) = \text{Tot}(K^\bullet \otimes_R F^\bullet) \otimes_A B\] The cohomology groups of the left hand side are \(H^i((K \otimes_A^\mathbf{L} B) \otimes_R^\mathbf{L} M)\) and on the right hand side we obtain \(H^i(K \otimes_R^\mathbf{L} M) \otimes_A B\).
Lemma
Let \(R\) be a ring of finite global dimension \(d\). Then
every module has tor dimension \(\leq d\),
a complex of \(R\)-modules \(K^\bullet\) with \(H^i(K^\bullet) \not = 0\) only if \(i \in [a, b]\) has tor amplitude in \([a - d, b]\), and
a complex of \(R\)-modules \(K^\bullet\) has finite tor dimension if and only if \(K^\bullet \in D^b(R)\).
Proof
The assumption on \(R\) means that every module has a finite projective resolution of length at most \(d\), in particular every module has tor dimension \(\leq d\). The second statement follows from Lemma 066I and the definitions. The third statement is a rephrasing of the second.
Lemma
Let \(R' \to R\) be a surjective ring map whose kernel is a nilpotent ideal. Let \(K' \in D(R')\) and set \(K = K' \otimes_{R'}^\mathbf{L} R\). Let \(a, b \in \mathbf{Z}\). Then \(K\) has tor amplitude in \([a, b]\) if and only if \(K'\) does.
Proof
One direction follows from Lemma 066L. For the other, assume \(K\) has tor amplitude in \([a, b]\) and let \(M'\) be an \(R'\)-module. We have to show that \(K' \otimes_{R'}^\mathbf{L} M'\) has nonzero cohomology only for degrees contained in the interval \([a, b]\).
Let \(I = \Ker(R' \to R)\). Then \(I^n = 0\) for some \(n\). If \(IM' = 0\), then we can view \(M'\) as an \(R\)-module and argue as follows \[K' \otimes_{R'}^\mathbf{L} M' = K' \otimes_{R'}^\mathbf{L} (R \otimes_R^\mathbf{L} M') = (K' \otimes_{R'}^\mathbf{L} R) \otimes_R^\mathbf{L} M' = K \otimes_R^\mathbf{L} M'\] which has nonvanishing cohomology only in the interval \([a, b]\) by assumption on \(K\). If \(I^{t + 1}M' = 0\), then we consider the short exact sequence \[0 \to IM' \to M' \to M'/IM' \to 0\] By induction on \(t\) we have that both \(K' \otimes_{R'}^\mathbf{L} IM'\) and \(K' \otimes_{R'}^\mathbf{L} M'/IM'\) have nonzero cohomology only for degrees in the interval \([a, b]\). Then the distinguished triangle \[K' \otimes_{R'}^\mathbf{L} IM' \to K' \otimes_{R'}^\mathbf{L} M' \to K' \otimes_{R'}^\mathbf{L} M'/IM' \to (K' \otimes_{R'}^\mathbf{L} IM')[1]\] proves the same is true for \(K' \otimes_{R'}^\mathbf{L} M'\) as desired.
Spectral sequences for Ext
In this section we collect various spectral sequences that come up when considering the Ext functors. For any pair of objects \(L\), \(K\) of the derived category \(D(R)\) of a ring \(R\) we denote \[\Ext^n_R(L, K) = \Hom_{D(R)}(L, K[n])\] according to our general conventions in Derived Categories, Section 06XP.
For \(M\) an \(R\)-module and \(K \in D^+(R)\) there is a spectral sequence [0AVH]\[\begin{equation} E_2^{i, j} = \Ext_R^i(M, H^j(K)) \Rightarrow \Ext_R^{i + j}(M, K) \end{equation}\] and if \(K\) is represented by the bounded below complex \(K^\bullet\) of \(R\)-modules there is a spectral sequence [0AVI]\[\begin{equation} E_1^{i, j} = \Ext_R^j(M, K^i) \Rightarrow \Ext_R^{i + j}(M, K) \end{equation}\] These spectral sequences come from applying Derived Categories, Lemma 015J to the functor \(\Hom_R(M, -)\).
Projective dimension
We defined the projective dimension of a module in Algebra, Definition 00O4.
Definition
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\). We say \(K\) has finite projective dimension if \(K\) can be represented by a bounded complex of projective modules. We say \(K\) has projective-amplitude in \([a, b]\) if \(K\) is quasi-isomorphic to a complex \[\ldots \to 0 \to P^a \to P^{a + 1} \to \ldots \to P^{b - 1} \to P^b \to 0 \to \ldots\] where \(P^i\) is a projective \(R\)-module for all \(i \in \mathbf{Z}\).
Clearly, \(K\) has finite projective dimension if and only if \(K\) has projective-amplitude in \([a, b]\) for some \(a, b \in \mathbf{Z}\). Furthermore, if \(K\) has finite projective dimension, then \(K\) is bounded. Here is a lemma to detect such objects of \(D(R)\).
Lemma
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\). Let \(a, b \in \mathbf{Z}\). The following are equivalent
\(K\) has projective-amplitude in \([a, b]\),
\(\Ext^i_R(K, N) = 0\) for all \(R\)-modules \(N\) and all \(i \not \in [-b, -a]\),
\(H^n(K) = 0\) for \(n > b\) and \(\Ext^i_R(K, N) = 0\) for all \(R\)-modules \(N\) and all \(i > -a\), and
\(H^n(K) = 0\) for \(n \not \in [a - 1, b]\) and \(\Ext^{-a + 1}_R(K, N) = 0\) for all \(R\)-modules \(N\).
Proof
Assume (1). We may assume \(K\) is the complex \[\ldots \to 0 \to P^a \to P^{a + 1} \to \ldots \to P^{b - 1} \to P^b \to 0 \to \ldots\] where \(P^i\) is a projective \(R\)-module for all \(i \in \mathbf{Z}\). In this case we can compute the ext groups by the complex \[\ldots \to 0 \to \Hom_R(P^b, N) \to \ldots \to \Hom_R(P^a, N) \to 0 \to \ldots\] and we obtain (2).
Assume (2) holds. Choose an injection \(H^n(K) \to I\) where \(I\) is an injective \(R\)-module. Since \(\Hom_R(-, I)\) is an exact functor, we see that \(\Ext^{-n}(K, I) = \Hom_R(H^n(K), I)\). We conclude in particular that \(H^n(K)\) is zero for \(n > b\). Thus (2) implies (3).
By the same argument as in (2) implies (3) gives that (3) implies (4).
Assume (4). The same argument as in (2) implies (3) shows that \(H^{a - 1}(K) = 0\), i.e., we have \(H^i(K) = 0\) unless \(i \in [a, b]\). In particular, \(K\) is bounded above and we can choose a complex \(P^\bullet\) representing \(K\) with \(P^i\) projective (for example free) for all \(i \in \mathbf{Z}\) and \(P^i = 0\) for \(i > b\). See Derived Categories, Lemma 05T7. Let \(Q = \Coker(P^{a - 1} \to P^a)\). Then \(K\) is quasi-isomorphic to the complex \[\ldots \to 0 \to Q \to P^{a + 1} \to \ldots \to P^b \to 0 \to \ldots\] as \(H^i(K) = 0\) for \(i < a\). Denote \(K' = (P^{a + 1} \to \ldots \to P^b)\) the corresponding object of \(D(R)\). We obtain a distinguished triangle \[K' \to K \to Q[-a] \to K'[1]\] in \(D(R)\). Thus for every \(R\)-module \(N\) an exact sequence \[\Ext^{-a}(K', N) \to \Ext^1(Q, N) \to \Ext^{1 - a}(K, N)\] By assumption the term on the right vanishes. By the implication (1) \(\Rightarrow\) (2) the term on the left vanishes. Thus \(Q\) is a projective \(R\)-module by Algebra, Lemma 05CF. Hence (1) holds and the proof is complete.
Example
Let \(k\) be a field and let \(R\) be the ring of dual numbers over \(k\), i.e., \(R = k[x]/(x^2)\). Denote \(\epsilon \in R\) the class of \(x\). Let \(M = R/(\epsilon)\). Then \(M\) is quasi-isomorphic to the complex \[R \xrightarrow{\epsilon} R \xrightarrow{\epsilon} R \to \ldots\] but \(M\) does not have finite projective dimension as defined in Algebra, Definition 00O4. This explains why we consider bounded (in both directions) complexes of projective modules in our definition of finite projective dimension of objects of \(D(R)\).
Injective dimension
This section is the dual of the section on projective dimension.
Definition
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\). We say \(K\) has finite injective dimension if \(K\) can be represented by a finite complex of injective \(R\)-modules. We say \(K\) has injective-amplitude in \([a, b]\) if \(K\) is isomorphic to a complex \[\ldots \to 0 \to I^a \to I^{a + 1} \to \ldots \to I^{b - 1} \to I^b \to 0 \to \ldots\] with \(I^i\) an injective \(R\)-module for all \(i \in \mathbf{Z}\).
Clearly, \(K\) has bounded injective dimension if and only if \(K\) has injective-amplitude in \([a, b]\) for some \(a, b \in \mathbf{Z}\). Furthermore, if \(K\) has bounded injective dimension, then \(K\) is bounded. Here is the obligatory lemma.
Lemma
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\). Let \(a, b \in \mathbf{Z}\). The following are equivalent
\(K\) has injective-amplitude in \([a, b]\),
\(\Ext^i_R(N, K) = 0\) for all \(R\)-modules \(N\) and all \(i \not \in [a, b]\),
\(\Ext^i(R/I, K) = 0\) for all ideals \(I \subset R\) and all \(i \not \in [a, b]\).
Proof
Assume (1). We may assume \(K\) is the complex \[\ldots \to 0 \to I^a \to I^{a + 1} \to \ldots \to I^{b - 1} \to I^b \to 0 \to \ldots\] where \(I^i\) is an injective \(R\)-module for all \(i \in \mathbf{Z}\). In this case we can compute the ext groups by the complex \[\ldots \to 0 \to \Hom_R(N, I^a) \to \ldots \to \Hom_R(N, I^b) \to 0 \to \ldots\] and we obtain (2). It is clear that (2) implies (3).
Assume (3) holds. Choose a nonzero map \(R \to H^n(K)\). Since \(\Hom_R(R, -)\) is an exact functor, we see that \(\Ext^n_R(R, K) = \Hom_R(R, H^n(K)) = H^n(K)\). We conclude that \(H^n(K)\) is zero for \(n \not \in [a, b]\). In particular, \(K\) is bounded below and we can choose a quasi-isomorphism \[K \to I^\bullet\] with \(I^i\) injective for all \(i \in \mathbf{Z}\) and \(I^i = 0\) for \(i < a\). See Derived Categories, Lemma 05T6. Let \(J = \Ker(I^b \to I^{b + 1})\). Then \(K\) is quasi-isomorphic to the complex \[\ldots \to 0 \to I^a \to \ldots \to I^{b - 1} \to J \to 0 \to \ldots\] Denote \(K' = (I^a \to \ldots \to I^{b - 1})\) the corresponding object of \(D(R)\). We obtain a distinguished triangle \[J[-b] \to K \to K' \to J[1 - b]\] in \(D(R)\). Thus for every ideal \(I \subset R\) an exact sequence \[\Ext^b(R/I, K') \to \Ext^1(R/I, J) \to \Ext^{1 + b}(R/I, K)\] By assumption the term on the right vanishes. By the implication (1) \(\Rightarrow\) (2) the term on the left vanishes. Thus \(J\) is an injective \(R\)-module by Lemma 0AVF.
Example
Let \(R\) be a Dedekind domain. Then every nonzero ideal \(I\) is a finite projective module, see Lemma 0AUW. Thus \(R/I\) has projective dimension \(1\). Hence every \(R\)-module \(M\) has injective dimension \(\leq 1\) by Lemma 0A5T. Thus \(\Ext^i_R(M, N) = 0\) for \(i \geq 2\) and any pair of \(R\)-modules \(M, N\). It follows that any object \(K\) in \(D^b(R)\) is isomorphic to the direct sum of its cohomologies: \(K \cong \bigoplus H^i(K)[-i]\), see Derived Categories, Lemma 0EWX.
Example
Let \(k\) be a field and let \(R\) be the ring of dual numbers over \(k\), i.e., \(R = k[x]/(x^2)\). Denote \(\epsilon \in R\) the class of \(x\). Let \(M = R/(\epsilon)\). Then \(M\) is quasi-isomorphic to the complex \[\ldots \to R \xrightarrow{\epsilon} R \xrightarrow{\epsilon} R\] and \(R\) is an injective \(R\)-module. However one usually does not consider \(M\) to have finite injective dimension in this situation. This explains why we consider bounded (in both directions) complexes of injective modules in our definition of bounded injective dimension of objects of \(D(R)\).
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\).
If \(K\) is in \(D^b(R)\) and \(H^i(K)\) has finite injective dimension for all \(i\), then \(K\) has finite injective dimension.
If \(K^\bullet\) represents \(K\), is a bounded complex of \(R\)-modules, and \(K^i\) has finite injective dimension for all \(i\), then \(K\) has finite injective dimension.
Proof
Omitted. Hint: Apply the spectral sequences of Derived Categories, Lemma 015J to the functor \(F = \Hom_R(N, -)\) to get a computation of \(\Ext^i_R(N, K)\) and use the criterion of Lemma 0A5T.
Lemma
Let \(R\) be a Noetherian ring. Let \(I \subset R\) be an ideal contained in the Jacobson radical of \(R\). Let \(K \in D^+(R)\) have finite cohomology modules. Then the following are equivalent
\(K\) has finite injective dimension, and
there exists a \(b\) such that \(\Ext^i_R(R/J, K) = 0\) for \(i > b\) and any ideal \(J \supset I\).
Proof
The implication (1) \(\Rightarrow\) (2) is immediate. Assume (2). Say \(H^i(K) = 0\) for \(i < a\). Then \(\Ext^i(M, K) = 0\) for \(i < a\) and all \(R\)-modules \(M\). Thus it suffices to show that \(\Ext^i(M, K) = 0\) for \(i > b\) any finite \(R\)-module \(M\), see Lemma 0A5T. By Algebra, Lemma 00L0 the module \(M\) has a finite filtration whose successive quotients are of the form \(R/\mathfrak p\) where \(\mathfrak p\) is a prime ideal. If \(0 \to M_1 \to M \to M_2 \to 0\) is a short exact sequence and \(\Ext^i(M_j, K) = 0\) for \(i > b\) and \(j = 1, 2\), then \(\Ext^i(M, K) = 0\) for \(i > b\). Thus we may assume \(M = R/\mathfrak p\). If \(I \subset \mathfrak p\), then the vanishing follows from the assumption. If not, then choose \(f \in I\), \(f \not \in \mathfrak p\). Consider the short exact sequence \[0 \to R/\mathfrak p \xrightarrow{f} R/\mathfrak p \to R/(\mathfrak p, f) \to 0\] The \(R\)-module \(R/(\mathfrak p, f)\) has a filtration whose successive quotients are \(R/\mathfrak q\) with \((\mathfrak p, f) \subset \mathfrak q\). Thus by Noetherian induction and the argument above we may assume the vanishing holds for \(R/(\mathfrak p, f)\). On the other hand, the modules \(E^i = \Ext^i(R/\mathfrak p, K)\) are finite by our assumption on \(K\) (bounded below with finite cohomology modules), the spectral sequence (0AVH), and Algebra, Lemma 08YR. Thus \(E^i\) for \(i > b\) is a finite \(R\)-module such that \(E^i/fE^i = 0\). We conclude by Nakayama’s lemma (Algebra, Lemma 00DV) that \(E^i\) is zero.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a local Noetherian ring. Let \(K \in D^+(R)\) have finite cohomology modules. Then the following are equivalent
\(K\) has finite injective dimension, and
\(\Ext^i_R(\kappa, K) = 0\) for \(i \gg 0\).
Proof
This is a special case of Lemma 0DW2.
Modules which are close to being projective
There seem to be many different of definitions in the literature of “almost projective modules”. In this section we discuss just one of the many possibilities.
Lemma
Let \(R\) be a ring. Let \(M\), \(N\) be \(R\)-modules.
Given an \(R\)-module map \(\varphi : M \to N\) the following are equivalent: (a) \(\varphi\) factors through a projective \(R\)-module, and (b) \(\varphi\) factors through a free \(R\)-module.
The set of \(\varphi : M \to N\) satisfying the equivalent conditions of (1) is an \(R\)-submodule of \(\Hom_R(M, N)\).
Given maps \(\psi : M' \to M\) and \(\xi : N \to N'\), if \(\varphi : M \to N\) satisfies the equivalent conditions of (1), then \(\xi \circ \varphi \circ \psi : M' \to N'\) does too.
Proof
The equivalence of (1)(a) and (1)(b) follows from Algebra, Lemma 05CF. If \(\varphi : M \to N\) and \(\varphi' : M \to N\) factor through the modules \(P\) and \(P'\) then \(\varphi + \varphi'\) factors through \(P \oplus P'\) and \(\lambda \varphi\) factors through \(P\) for all \(\lambda \in R\). This proves (2). If \(\varphi : M \to N\) factors through the module \(P\) and \(\psi\) and \(\xi\) are as in (3), then \(\xi \circ \varphi \circ \psi\) factors through \(P\). This proves (3).
Lemma
Let \(R\) be a ring. Let \(\varphi : M \to N\) be an \(R\)-module map. If \(\varphi\) factors through a projective module and \(M\) is a finite \(R\)-module, then \(\varphi\) factors through a finite projective module.
Proof
By Lemma 0G90 we can factor \(\varphi = \tau \circ \sigma\) where the target of \(\sigma\) is \(\bigoplus_{i \in I} R\) for some set \(I\). Choose generators \(x_1, \ldots, x_n\) for \(M\). Write \(\sigma(x_j) = (a_{ji})_{i \in I}\). For each \(j\) only a finite number of \(a_{ji}\) are nonzero. Hence \(\sigma\) has image contained in a finite free \(R\)-module and we conclude.
Let \(R\) be a ring. Observe that an \(R\)-module is projective if and only if its identity factors through a projective module.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(M\) be an \(R\)-module. The following conditions are equivalent
for every \(a \in I\) the map \(a : M \to M\) factors through a projective \(R\)-module,
for every \(a \in I\) the map \(a : M \to M\) factors through a free \(R\)-module, and
\(\Ext^1_R(M, N)\) is annihilated by \(I\) for every \(R\)-module \(N\).
Proof
The equivalence of (1) and (2) follows from Lemma 0G90. If (1) holds, then (3) holds because \(\Ext^1_R(P, N) = 0\) for any \(N\) and any projective module \(P\). Conversely, assume (3) holds. Choose a short exact sequence \(0 \to N \to P \to M \to 0\) with \(P\) projective (or even free). By assumption the corresponding element of \(\Ext^1_R(M, N)\) is annihilated by \(I\). Hence for every \(a \in I\) the map \(a : M \to M\) can be factored through the surjection \(P \to M\) and we conclude (1) holds.
In order to comfortably talk about modules satisfying the equivalent conditions of Lemma 0G92 we give the property a name.
Definition
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(M\) be an \(R\)-module. We say \(M\) is \(I\)-projective9 if the equivalent conditions of Lemma 0G92 hold.
Modules annihilated by \(I\) are \(I\)-projective.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(M\) be an \(R\)-module. If \(M\) is annihilated by \(I\), then \(M\) is \(I\)-projective.
Proof
Immediate from the definition and the fact that the zero module is projective.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \[0 \to K \to P \to M \to 0\] be a short exact sequence of \(R\)-modules. If \(M\) is \(I\)-projective and \(P\) is projective, then \(K\) is \(I\)-projective.
Proof
The element \(\text{id}_K \in \Hom_R(K, K)\) maps to the class of the given extension in \(\Ext^1_R(M, K)\). Since by assumption this class is annihilated by any \(a \in I\) we see that \(a : K \to K\) factors through \(K \to P\) and we conclude.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. If \(M\) is a finite, \(I\)-projective \(R\)-module, then \(M^\vee = \Hom_R(M, R)\) is \(I\)-projective.
Proof
Assume \(M\) is finite and \(I\)-projective. Choose a short exact sequence \(0 \to K \to R^{\oplus r} \to M \to 0\). This produces an injection \(M^\vee \to R^{\oplus r} = (R^{\oplus r})^\vee\). Since the extension class in \(\Ext^1_R(M, K)\) corresponding to the short exact sequence is annihilated by \(I\), we see that for any \(a \in I\) we can find a map \(M \to R^{\oplus r}\) such that the composition with the given map \(R^{\oplus r} \to M\) is equal to \(a : M \to M\). Taking duals we find that \(a : M^\vee \to M^\vee\) factors through the map \(M^\vee \to R^{\oplus r}\) given above and we conclude.
Hom complexes
Let \(R\) be a ring. Let \(L^\bullet\) and \(M^\bullet\) be two complexes of \(R\)-modules. We construct a complex \(\Hom^\bullet(L^\bullet, M^\bullet)\). Namely, for each \(n\) we set \[\Hom^n(L^\bullet, M^\bullet) = \prod\nolimits_{n = p + q} \Hom_R(L^{-q}, M^p)\] It is a good idea to think of \(\Hom^n\) as the \(R\)-module of all \(R\)-linear maps from \(L^\bullet\) to \(M^\bullet\) (viewed as graded modules) which are homogeneous of degree \(n\). In this terminology, we define the differential by the rule \[\text{d}(f) = \text{d}_M \circ f - (-1)^n f \circ \text{d}_L\] for \(f \in \Hom^n(L^\bullet, M^\bullet)\). We omit the verification that \(\text{d}^2 = 0\). See Section 0FNG for sign rules. This construction is a special case of Differential Graded Algebra, Example 09L9. It follows immediately from the construction that we have [0A5X]\[\begin{equation} H^n(\Hom^\bullet(L^\bullet, M^\bullet)) = \Hom_{K(R)}(L^\bullet, M^\bullet[n]) \end{equation}\] for all \(n \in \mathbf{Z}\).
Lemma
Let \(R\) be a ring. Given complexes \(K^\bullet, L^\bullet, M^\bullet\) of \(R\)-modules there is a canonical isomorphism \[\Hom^\bullet(K^\bullet, \Hom^\bullet(L^\bullet, M^\bullet)) = \Hom^\bullet(\text{Tot}(K^\bullet \otimes_R L^\bullet), M^\bullet)\] of complexes of \(R\)-modules.
Proof
Let \(\alpha\) be an element of degree \(n\) on the left hand side. Thus \[\alpha = (\alpha^{p, q}) \in \prod\nolimits_{p + q = n} \Hom_R(K^{-q}, \Hom^p(L^\bullet, M^\bullet))\] Each \(\alpha^{p, q}\) is an element \[\alpha^{p, q} = (\alpha^{r, s, q}) \in \prod\nolimits_{r + s + q = n} \Hom_R(K^{-q}, \Hom_R(L^{-s}, M^r))\] If we make the identifications [0A5Z]\[\begin{equation} \Hom_R(K^{-q}, \Hom_R(L^{-s}, M^r)) = \Hom_R(K^{-q} \otimes_R L^{-s}, M^r) \end{equation}\] then by our sign rules we get \[\begin{align*} \text{d}(\alpha^{r, s, q}) & = \text{d}_{\Hom^\bullet(L^\bullet, M^\bullet)} \circ \alpha^{r, s, q} - (-1)^n \alpha^{r, s, q} \circ \text{d}_K \\ & = \text{d}_M \circ \alpha^{r, s, q} - (-1)^{r + s} \alpha^{r, s, q} \circ \text{d}_L - (-1)^{r + s + q} \alpha^{r, s, q} \circ \text{d}_K \end{align*}\] On the other hand, if \(\beta\) is an element of degree \(n\) of the right hand side, then \[\beta = (\beta^{r, s, q}) \in \prod\nolimits_{r + s + q = n} \Hom_R(K^{-q} \otimes_R L^{-s}, M^r)\] and by our sign rule (Homology, Definition 012Z) we get \[\begin{align*} \text{d}(\beta^{r, s, q}) & = \text{d}_M \circ \beta^{r, s, q} - (-1)^n \beta^{r, s, q} \circ \text{d}_{\text{Tot}(K^\bullet \otimes L^\bullet)} \\ & = \text{d}_M \circ \beta^{r, s, q} - (-1)^{r + s + q} \left( \beta^{r, s, q} \circ \text{d}_K + (-1)^{-q} \beta^{r, s, q} \circ \text{d}_L \right) \end{align*}\] Thus we see that the map induced by the identifications (0A5Z) indeed is a morphism of complexes.
Remark
Let \(R\) be a ring. The category \(\text{Comp}(R)\) of complexes of \(R\)-modules is a symmetric monoidal category with tensor product given by \(\text{Tot}(- \otimes_R -)\), see Lemma 0FNI. Given \(L^\bullet\) and \(M^\bullet\) in \(\text{Comp}(R)\) an element \(f \in \Hom^0(L^\bullet, M^\bullet)\) defines a map of complexes \(f : L^\bullet \to M^\bullet\) if and only if \(\text{d}(f) = 0\). Hence Lemma 0A5Y also tells us that \[\Mor_{\text{Comp}(R)}(K^\bullet, \Hom^\bullet(L^\bullet, M^\bullet)) = \Mor_{\text{Comp}(R)}(\text{Tot}(K^\bullet \otimes_R L^\bullet), M^\bullet)\] functorially in \(K^\bullet, L^\bullet, M^\bullet\) in \(\text{Comp}(R)\). This means that \(\Hom^\bullet( - , -)\) is an internal hom for the symmetric monoidal category \(\text{Comp}(R)\) as discussed in Categories, Remark 0GWJ.
Lemma
Let \(R\) be a ring. Given complexes \(K^\bullet, L^\bullet, M^\bullet\) of \(R\)-modules there is a canonical morphism \[\text{Tot}\left( \Hom^\bullet(L^\bullet, M^\bullet) \otimes_R \Hom^\bullet(K^\bullet, L^\bullet) \right) \longrightarrow \Hom^\bullet(K^\bullet, M^\bullet)\] of complexes of \(R\)-modules.
Proof
Via the discussion in Remark 0GWQ the existence of such a canonical map follows from Categories, Remark 0GWJ. We also give a direct construction.
An element \(\alpha\) of degree \(n\) of the left hand side is \[\alpha = (\alpha^{p, q}) \in \bigoplus\nolimits_{p + q = n} \Hom^p(L^\bullet, M^\bullet) \otimes_R \Hom^q(K^\bullet, L^\bullet)\] The element \(\alpha^{p, q}\) is a finite sum \(\alpha^{p, q} = \sum \beta^p_i \otimes \gamma^q_i\) with \[\beta^p_i = (\beta^{r, s}_i) \in \prod\nolimits_{r + s = p} \Hom_R(L^{-s}, M^r)\] and \[\gamma^q_i = (\gamma^{u, v}_i) \in \prod\nolimits_{u + v = q} \Hom_R(K^{-v}, L^u)\] The map is given by sending \(\alpha\) to \(\delta = (\delta^{r, v})\) with \[\delta^{r, v} = \sum\nolimits_{i, s} \beta^{r, s}_i \circ \gamma^{-s, v}_i \in \Hom_R(K^{-v}, M^r)\] For given \(r + v = n\) this sum is finite as there are only finitely many nonzero \(\alpha^{p, q}\), hence only finitely many nonzero \(\beta^p_i\) and \(\gamma^q_i\). By our sign rules we have \[\begin{align*} \text{d}(\alpha^{p, q}) & = \text{d}_{\Hom^\bullet(L^\bullet, M^\bullet)}(\alpha^{p, q}) + (-1)^p \text{d}_{\Hom^\bullet(K^\bullet, L^\bullet)}(\alpha^{p, q}) \\ & = \sum \Big( \text{d}_M \circ \beta^p_i \circ \gamma^q_i - (-1)^p \beta^p_i \circ \text{d}_L \circ \gamma^q_i \Big) \\ & \quad + (-1)^p \sum \Big( \beta^p_i \circ \text{d}_L \circ \gamma^q_i - (-1)^q \beta^p_i \circ \gamma^q_i \circ \text{d}_K \Big) \\ & = \sum \Big( \text{d}_M \circ \beta^p_i \circ \gamma^q_i -(-1)^n \beta^p_i \circ \gamma^q_i \circ \text{d}_K \Big) \end{align*}\] It follows that the rules \(\alpha \mapsto \delta\) is compatible with differentials and the lemma is proved.
Lemma
Let \(R\) be a ring. Given complexes \(K^\bullet, L^\bullet, M^\bullet\) of \(R\)-modules there is a canonical morphism \[\text{Tot}(K^\bullet \otimes_R \Hom^\bullet(M^\bullet, L^\bullet)) \longrightarrow \Hom^\bullet(M^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet))\] of complexes of \(R\)-modules functorial in all three complexes.
Proof
Via the discussion in Remark 0GWQ the existence of such a canonical map follows from Categories, Remark 0GWJ. We also give a direct construction.
Let \(\alpha\) be an element of degree \(n\) of the right hand side. Thus \[\alpha = (\alpha^{p, q}) \in \prod\nolimits_{p + q = n} \Hom_R(M^{-q}, \text{Tot}^p(K^\bullet \otimes_R L^\bullet))\] Each \(\alpha^{p, q}\) is an element \[\alpha^{p, q} = (\alpha^{r, s, q}) \in \Hom_R(M^{-q}, \bigoplus\nolimits_{r + s + q = n} K^r \otimes_R L^s)\] where we think of \(\alpha^{r, s, q}\) as a family of maps such that for every \(x \in M^{-q}\) only a finite number of \(\alpha^{r, s, q}(x)\) are nonzero. By our sign rules we get \[\begin{align*} \text{d}(\alpha^{r, s, q}) & = \text{d}_{\text{Tot}(K^\bullet \otimes_R L^\bullet)} \circ \alpha^{r, s, q} - (-1)^n \alpha^{r, s, q} \circ \text{d}_M \\ & = \text{d}_K \circ \alpha^{r, s, q} + (-1)^r \text{d}_L \circ \alpha^{r, s, q} - (-1)^n \alpha^{r, s, q} \circ \text{d}_M \end{align*}\] On the other hand, if \(\beta\) is an element of degree \(n\) of the left hand side, then \[\beta = (\beta^{p, q}) \in \bigoplus\nolimits_{p + q = n} K^p \otimes_R \Hom^q(M^\bullet, L^\bullet)\] and we can write \(\beta^{p, q} = \sum \gamma_i^p \otimes \delta_i^q\) with \(\gamma_i^p \in K^p\) and \[\delta_i^q = (\delta_i^{r, s}) \in \prod\nolimits_{r + s = q} \Hom_R(M^{-s}, L^r)\] By our sign rules we have \[\begin{align*} \text{d}(\beta^{p, q}) & = \text{d}_K(\beta^{p, q}) + (-1)^p \text{d}_{\Hom^\bullet(M^\bullet, L^\bullet)}(\beta^{p, q}) \\ & = \sum \text{d}_K(\gamma_i^p) \otimes \delta_i^q + (-1)^p \sum \gamma_i^p \otimes (\text{d}_L \circ \delta_i^q - (-1)^q \delta_i^q \circ \text{d}_M) \end{align*}\] We send the element \(\beta\) to \(\alpha\) with \[\alpha^{r, s, q} = c^{r, s, q}(\sum \gamma_i^r \otimes \delta_i^{s, q})\] where \(c^{r, s, q} : K^r \otimes_R \Hom_R(M^{-q}, L^s) \to \Hom_R(M^{-q}, K^r \otimes_R L^s)\) is the canonical map. For a given \(\beta\) and \(r\) there are only finitely many nonzero \(\gamma_i^r\) hence only finitely many nonzero \(\alpha^{r, s, q}\) are nonzero (for a given \(r\)). Thus this family of maps satisfies the conditions above and the map is well defined. Comparing signs we see that this is compatible with differentials.
Lemma
Let \(R\) be a ring. Given complexes \(K^\bullet, L^\bullet\) of \(R\)-modules there is a canonical morphism \[K^\bullet \longrightarrow \Hom^\bullet(L^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet))\] of complexes of \(R\)-modules functorial in both complexes.
Proof
Via the discussion in Remark 0GWQ the existence of such a canonical map follows from Categories, Remark 0GWJ. We also give a direct construction.
Let \(\alpha\) be an element of degree \(n\) of the right hand side. Thus \[\alpha = (\alpha^{p, q}) \in \prod\nolimits_{p + q = n} \Hom_R(L^{-q}, \text{Tot}^p(K^\bullet \otimes_R L^\bullet))\] Each \(\alpha^{p, q}\) is an element \[\alpha^{p, q} = (\alpha^{r, s, q}) \in \Hom_R(L^{-q}, \bigoplus\nolimits_{r + s + q = n} K^r \otimes_R L^s)\] where we think of \(\alpha^{r, s, q}\) as a family of maps such that for every \(x \in L^{-q}\) only a finite number of \(\alpha^{r, s, q}(x)\) are nonzero. By our sign rules we get \[\begin{align*} \text{d}(\alpha^{r, s, q}) & = \text{d}_{\text{Tot}(K^\bullet \otimes_R L^\bullet)} \circ \alpha^{r, s, q} - (-1)^n \alpha^{r, s, q} \circ \text{d}_L \\ & = \text{d}_K \circ \alpha^{r, s, q} + (-1)^r \text{d}_L \circ \alpha^{r, s, q} - (-1)^n \alpha^{r, s, q} \circ \text{d}_L \end{align*}\] Now an element \(\beta \in K^n\) we send to \(\alpha\) with \(\alpha^{n, -q, q} = \beta \otimes \text{id}_{L^{-q}}\) and \(\alpha^{r, s, q} = 0\) if \(r \not = n\). This is indeed an element as above, as for fixed \(q\) there is only one nonzero \(\alpha^{r, s, q}\). The description of the differential shows this is compatible with differentials.
Lemma
Let \(R\) be a ring. Given complexes \(K^\bullet, L^\bullet, M^\bullet\) of \(R\)-modules there is a canonical morphism \[\text{Tot}(\Hom^\bullet(L^\bullet, M^\bullet) \otimes_R K^\bullet) \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, L^\bullet), M^\bullet)\] of complexes of \(R\)-modules functorial in all three complexes.
Proof
Via the discussion in Remark 0GWQ the existence of such a canonical map follows from Categories, Remark 0GWJ. We also give a direct construction.
Consider an element \(\beta\) of degree \(n\) of the right hand side. Then \[\beta = (\beta^{p, s}) \in \prod\nolimits_{p + s = n} \Hom_R(\Hom^{-s}(K^\bullet, L^\bullet), M^p)\] Our sign rules tell us that \[\begin{align*} \text{d}(\beta^{p, s}) & = \text{d}_M \circ \beta^{p, s} - (-1)^n \beta^{p, s} \circ \text{d}_{\Hom^\bullet(K^\bullet, L^\bullet)} \end{align*}\] We can describe the last term as follows \[(\beta^{p, s} \circ \text{d}_{\Hom^\bullet(K^\bullet, L^\bullet)})(f) = \beta^{p, s}(\text{d}_L \circ f - (-1)^{s + 1} f \circ \text{d}_K)\] if \(f \in \Hom^{-s - 1}(K^\bullet, L^\bullet)\). We conclude that in some unspecified sense \(\text{d}(\beta^{p, s})\) is a sum of three terms with signs as follows [0FNE]\[\begin{equation} \text{d}(\beta^{p, s}) = \text{d}_M(\beta^{p, s}) -(-1)^n\text{d}_L(\beta^{p, s}) + (-1)^{p + 1}\text{d}_K(\beta^{p, s}) \end{equation}\]
Next, we consider an element \(\alpha\) of degree \(n\) of the left hand side. We can write it like so \[\alpha = (\alpha^{t, r}) \in \bigoplus\nolimits_{t + r = n} \Hom^t(L^\bullet, M^\bullet) \otimes K^r\] Each \(\alpha^{t, r}\) maps to an element \[\alpha^{t, r} \mapsto (\alpha^{p, q, r}) \in \prod\nolimits_{p + q = t} \Hom_R(L^{-q}, M^p) \otimes_R K^r\] Our sign rules tell us that \[\begin{align*} \text{d}(\alpha^{p, q, r}) & = \text{d}_{\Hom^\bullet(L^\bullet, M^\bullet)}(\alpha^{p, q, r}) + (-1)^{p + q} \text{d}_K(\alpha^{p, q, r}) \end{align*}\] where if we further write \(\alpha^{p, q, r} = \sum g_i^{p, q} \otimes k_i^r\) then we have \[\text{d}_{\Hom^\bullet(L^\bullet, M^\bullet)}(\alpha^{p, q, r}) = \sum (\text{d}_M \circ g_i^{p, q}) \otimes k_i^r - (-1)^{p + q} \sum (g_i^{p, q} \circ \text{d}_L) \otimes k_i^r\] We conclude that in some unspecified sense \(\text{d}(\alpha^{p, q, r})\) is a sum of three terms with signs as follows [0FNF]\[\begin{equation} \text{d}(\alpha^{p, q, r}) = \text{d}_M(\alpha^{p, q, r}) -(-1)^{p + q}\text{d}_L(\alpha^{p, q, r}) + (-1)^{p + q}\text{d}_K(\alpha^{p, q, r}) \end{equation}\]
To define our map we will use the canonical maps \[c_{p, q, r} : \Hom_R(L^{-q}, M^p) \otimes_R K^r \longrightarrow \Hom_R(\Hom_R(K^r, L^{-q}), M^p)\] which sends \(\varphi \otimes k\) to the map \(\psi \mapsto \varphi(\psi(k))\). This is functorial in all three variables. With \(s = q + r\) there is an inclusion \[\Hom_R(\Hom_R(K^r, L^{-q}), M^p) \subset \Hom_R(\Hom^{-s}(K^\bullet, L^\bullet), M^p)\] coming from the projection \(\Hom^{-s}(K^\bullet, L^\bullet) \to \Hom_R(K^r, L^{-q})\). Since \(\alpha^{p, q, r}\) is nonzero only for a finite number of \(r\) we see that for a given \(s\) there is only a finite number of \(q, r\) with \(q + r = s\). Thus we can send \(\alpha\) to the element \(\beta\) with \[\beta^{p, s} = \sum\nolimits_{q + r = s} \epsilon_{p, q, r} c_{p, q, r}(\alpha^{p, q, r})\] where where the sum uses the inclusions given above and where \(\epsilon_{p, q, r} \in \{\pm 1\}\). Comparing signs in the equations (0FNE) and (0FNF) we see that
\(\epsilon_{p, q, r} = \epsilon_{p + 1, q, r}\)
\(-(-1)^n\epsilon_{p, q, r} = -(-1)^{p + q}\epsilon_{p, q - 1, r}\) or equivalently \(\epsilon_{p, q, r} = (-1)^r\epsilon_{p, q - 1, r}\)
\((-1)^{p + 1}\epsilon_{p, q, r} = (-1)^{p + q}\epsilon_{p, q, r + 1}\) or equivalently \((-1)^{q + 1}\epsilon_{p, q, r} = \epsilon_{p, q, r + 1}\).
A good solution is to take \[\epsilon_{p, q, r} = (-1)^{r + qr}\] The choice of this sign is explained in the remark following the proof.
Remark
Let us explain why the sign used in the direct construction in the proof of Lemma 0A60 agrees with the sign we get from the construction using the discussion in Remark 0GWQ and Categories, Remark 0GWJ. Denote \(- \otimes - = \text{Tot}(- \otimes_R -)\) and \(hom(-, -) = \Hom^\bullet(-, -)\). The construction using monoidal category language tells us to use the arrow \[hom(L^\bullet, M^\bullet) \otimes K^\bullet \longrightarrow hom(hom(K^\bullet, L^\bullet), M^\bullet)\] in \(\text{Comp}(R)\) corresponding to the arrow \[hom(L^\bullet, M^\bullet) \otimes K^\bullet \otimes hom(K^\bullet, L^\bullet) \longrightarrow M^\bullet\] gotten by swapping the order of the last two tensor products and then using the evaluation maps \(hom(K^\bullet, L^\bullet) \otimes K^\bullet \to L^\bullet\) and \(hom(L^\bullet, M^\bullet) \otimes L^\bullet \to M^\bullet\). Only in swapping does a sign intervene. Namely, in the isomorphism \[K^\bullet \otimes hom(K^\bullet, L^\bullet) \to hom(K^\bullet, L^\bullet) \otimes K^\bullet\] there is a sign \((-1)^{r(q + r')}\) on \(K^r \otimes_R \Hom_R(K^{-r'}, L^q)\), see Section 0FNG item (0GWR). The reader can convince themselves that, because of the correspondence we are using to describe maps into an internal hom, this sign only matters if \(r = r'\) and in this case we obtain \((-1)^{r(q + r)} = (-1)^{r + qr}\) as in the direct proof.
Sign rules
In this section we review the sign rules used so far and we discuss some of their ramifications. It also seems appropriate to discuss these issues in the setting of the category of complexes of modules over a ring, as most interesting phenomena already occur in this case. We sincerely hope the reader will not need to use the more esoteric aspects of this section.
For the rest of this section, we fix a ring \(R\) and we denote \(M^\bullet\) a complex of \(R\)-modules with differentials \(d^n_M : M^n \to M^{n + 1}\).
The \(k\)th shifted complex \(M^\bullet[k]\) has terms \((M^\bullet[k])^n = M^{n + k}\) and differentials \(d_{M[k]}^n = (-1)^kd^{n + k}_M\), see Homology, Definition 011G.
Given a map \(f : M^\bullet \to N^\bullet\) of complexes, we define \(f[k] : M^\bullet[k] \to N^\bullet[k]\) without the intervention of signs, see Homology, Definition 011G.
We identify \(H^n(M^\bullet[k])\) with \(H^{n + k}(M^\bullet)\) without the intervention of signs, see Homology, Definition 011H.
The boundary map of a short exact sequence of complexes is defined as in the snake lemma without the intervention of signs, see Homology, Lemma 0117.
The distinguished triangle associated to a termwise split short exact sequence \(0 \to K^\bullet \to L^\bullet \to M^\bullet \to 0\) of complexes is given by \[K^\bullet \to L^\bullet \to M^\bullet \to K^\bullet[1]\] where \(M^n \to K^{n + 1}\) is the map \(\pi^{n + 1} \circ d^n_L \circ s^n\) if \(s\) and \(\pi\) are compatible termwise splittings. In other words, without the intervention of signs. See Derived Categories, Definitions 014Q and 014I.
The total complex \(\text{Tot}(M^\bullet \otimes_R N^\bullet)\) has differential \(d\) satisfying the Leibniz rule \(d(x \otimes y) = d(x) \otimes y + (-1)^{\deg(x)}x \otimes d(y)\). See Homology, Example 0A5J and Homology, Definition 012Z.
There is a canonical isomorphism \[\text{Tot}(M^\bullet \otimes_R N^\bullet)[a + b] \to \text{Tot}(M^\bullet[a] \otimes_R N^\bullet[b])\] which uses the sign \((-1)^{pb}\) on the summand \(M^p \otimes_R N^q\), see Homology, Remark 0FLG. It is often more convenient to consider the corresponding shifted map \(\text{Tot}(M^\bullet \otimes_R N^\bullet) \to \text{Tot}(M^\bullet[a] \otimes_R N^\bullet[b])[-a - b]\).
There is a canonical isomorphism of complexes \[\text{Tot}( \text{Tot}(K^\bullet \otimes_R L^\bullet) \otimes_R M^\bullet) \to \text{Tot}(K^\bullet \otimes_R \text{Tot}(L^\bullet \otimes_R M^\bullet))\] defined without the intervention of signs. See Section 0GWN.
There is a canonical isomorphism \[\text{Tot}(L^\bullet \otimes_R M^\bullet) \to \text{Tot}(M^\bullet \otimes_R L^\bullet)\] which uses the sign \((-1)^{pq}\) on the summand \(L^p \otimes_R M^q\). See Section 0GWN.
Before we get into a discussion of the sign conventions regarding Hom-complexes, we construct the dual of a complex with respect to the conventions above.
Lemma
Let \(R\) be a ring. Let \(M\) be an \(R\)-module. Let \(N, \eta, \epsilon\) be a left dual of \(M\) in the monoidal category of \(R\)-modules, see Categories, Definition 0FFP. Then
\(M\) and \(N\) are finite projective \(R\)-modules,
the map \(e : \Hom_R(M, R) \to N\), \(\lambda \mapsto (\lambda \otimes 1)(\eta)\) is an isomorphism,
we have \(\epsilon(n, m) = e^{-1}(n)(m)\) for \(n \in N\) and \(m \in M\).
Proof
The assumptions mean that \[M \xrightarrow{\eta \otimes 1} M \otimes_R N \otimes_R M \xrightarrow{1 \otimes \epsilon} M \quad\text{and}\quad N \xrightarrow{1 \otimes \eta} N \otimes_R M \otimes_R N \xrightarrow{\epsilon \otimes 1} N\] are the identity map. We can choose a finite free module \(F\), an \(R\)-module map \(F \to M\), and a lift \(\tilde \eta : R \to F \otimes_R N\) of \(\eta\). We obtain a commutative diagram \[\xymatrix{ M \ar[rr]_-{\eta \otimes 1} \ar[rrd]_-{\tilde \eta \otimes 1} & & M \otimes_R N \otimes_R M \ar[r]_-{1 \otimes \epsilon} & M \\ & & F \otimes_R N \otimes_R M \ar[u] \ar[r]^-{1 \otimes \epsilon} & F \ar[u] }\] This shows that the identity on \(M\) factors through a finite free module and hence \(M\) is finite projective. By symmetry we see that \(N\) is finite projective. This proves part (1). Part (2) follows from Categories, Lemma 0FFQ and its proof. Part (3) follows from the first equality of the proof.
Lemma
Let \(R\) be a ring. Let \(M^\bullet\) be a complex of \(R\)-modules. Let \(N^\bullet, \eta, \epsilon\) be a left dual of \(M^\bullet\) in the monoidal category of complexes of \(R\)-modules. Then
\(M^\bullet\) and \(N^\bullet\) are bounded,
\(M^n\) and \(N^n\) are finite projective \(R\)-modules,
writing \(\epsilon = \sum \epsilon_n\) with \(\epsilon_n : N^{-n} \otimes_R M^n \to R\) and \(\eta = \sum \eta_n\) with \(\eta_n : R \to M^n \otimes_R N^{-n}\) then \((N^{-n}, \eta_n, \epsilon_n)\) is the left dual of \(M^n\) as in Lemma 0FNJ,
the differential \(d_N^n : N^n \to N^{n + 1}\) is equal to \(-(-1)^n\) times the map \[N^n = \Hom_R(M^{-n}, R) \xrightarrow{d_M^{-n - 1}} \Hom_R(M^{-n - 1}, R) = N^{n + 1}\] where the equality signs are the identifications from Lemma 0FNJ part (2).
Conversely, given a bounded complex \(M^\bullet\) of finite projective \(R\)-modules, setting \(N^n = \Hom_R(M^{-n}, R)\) with differentials as above, setting \(\epsilon = \sum \epsilon_n\) with \(\epsilon_n : N^{-n} \otimes_R M^n \to R\) given by evaluation, and setting \(\eta = \sum \eta_n\) with \(\eta_n : R \to M^n \otimes_R N^{-n}\) mapping \(1\) to \(\text{id}_{M^n}\) we obtain a left dual of \(M^\bullet\) in the monoidal category of complexes of \(R\)-modules.
Proof
Since \((1 \otimes \epsilon) \circ (\eta \otimes 1) = \text{id}_{M^\bullet}\) and \((\epsilon \otimes 1) \circ (1 \otimes \eta) = \text{id}_{N^\bullet}\) by Categories, Definition 0FFP we see immediately that we have \((1 \otimes \epsilon_n) \circ (\eta_n \otimes 1) = \text{id}_{M^n}\) and \((\epsilon_n \otimes 1) \circ (1 \otimes \eta_n) = \text{id}_{N^{-n}}\) which proves (3). By Lemma 0FNJ we have (2). Since the sum \(\eta = \sum \eta_n\) is finite, we get (1). Since \(\eta = \sum \eta_n\) is a map of complexes \(R \to \text{Tot}(M^\bullet \otimes_R N^\bullet)\) we see that \[(d_M^{-n - 1} \otimes 1) \circ \eta_{-n - 1} + (-1)^n (1 \otimes d_N^{-n}) \circ \eta_{-n} = 0\] by our choice of signs for the differential on \(\text{Tot}(M^\bullet \otimes_R N^\bullet)\). Unwinding definitions, this proves (4). To see the final statement of the lemma one reads the above backwards.
We will use the description of the left dual of a complex in Lemma 0FNK as a motivation for our sign rule on the \(\Hom\)-complex. Namelly, we choose the signs such that (0FNL) holds. We continue with the discussion of various sign rules as above
Given complexes \(K^\bullet\), \(M^\bullet\) we let \(\Hom^\bullet(M^\bullet, K^\bullet)\) be the complex with terms \[\Hom^n(M^\bullet, K^\bullet) = \prod\nolimits_{n = p + q} \Hom_R(M^{-q}, K^p)\] and differential given by the rule \[d(f) = d_K \circ f - (-1)^n f \circ d_M\]
The choice above is such that if \(M^\bullet\) has a left dual \(N^\bullet\) as in Lemma 0FNK, then we have a canonical isomorphism \[\text{Tot}(K^\bullet \otimes_R N^\bullet) \longrightarrow \Hom^\bullet(M^\bullet, K^\bullet)\] defined without the intervention of signs sending the summand \(K^p \otimes_R N^q\) to the summand \(\Hom_R(M^{-q}, K^p)\) via \(N^q = \Hom_R(M^{-q}, R)\) and the canonical map \(K^p \otimes_R \Hom_R(M^{-q}, R) \to \Hom_R(M^{-q}, K^p)\).
There is a composition \[\text{Tot}( \Hom^\bullet(L^\bullet, K^\bullet) \otimes_R \Hom^\bullet(M^\bullet, L^\bullet)) \longrightarrow \Hom^\bullet(M^\bullet, K^\bullet)\] defined without the intervention of signs, see Lemma 0A8I.
There is a canonical isomorphism \[\Hom^\bullet(K^\bullet, \Hom^\bullet(L^\bullet, M^\bullet)) = \Hom^\bullet(\text{Tot}(K^\bullet \otimes_R L^\bullet), M^\bullet)\] defined without the intervention of signs, see Lemma 0A5Y.
There is a canonical map \[\text{Tot}(K^\bullet \otimes_R \Hom^\bullet(M^\bullet, L^\bullet)) \longrightarrow \Hom^\bullet(M^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet))\] defined without the intervention of signs, see Lemma 0BYM.
There is a canonical map \[K^\bullet \longrightarrow \Hom^\bullet(L^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet))\] defined without the intervention of signs, see Lemma 0A62.
By Lemma 0A60 is a canonical map \[\text{Tot}(\Hom^\bullet(L^\bullet, M^\bullet) \otimes_R K^\bullet) \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, L^\bullet), M^\bullet)\] which uses a sign \((-1)^{r + qr}\) on the module \(\Hom_R(L^{-q}, M^p) \otimes_R K^r\) whose reason is explained in Remark 0A61.
Taking \(L^\bullet = M^\bullet\) and using \(R \to \Hom^\bullet(M^\bullet, M^\bullet)\) the map from the previous item becomes the evaluation map \[ev : K^\bullet \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, M^\bullet), M^\bullet)\] It sends \(x \in K^n\) to the map which sends \(f \in \Hom^m(K^\bullet, M^\bullet)\) to \((-1)^{nm}f(x)\).
There is a canonical identification \[\Hom^\bullet(M^\bullet, K^\bullet)[a - b] \to \Hom^\bullet(M^\bullet[b], K^\bullet[a])\] which uses signs. It is defined as the map whose corresponding shifted map \[\Hom^\bullet(M^\bullet, K^\bullet) \to \Hom^\bullet(M^\bullet[b], K^\bullet[a])[b - a]\] uses the sign \((-1)^{nb}\) on the module \(\Hom_R(M^{-q}, K^p)\) with \(p + q = n\). Namely, if \(f \in \Hom^n(M^\bullet, K^\bullet)\) then \[d(f) = d_K \circ f - (-1)^n f \circ d_M\] on the source, whereas on the target \(f\) lies in \(\left(\Hom^\bullet(M^\bullet[b], K^\bullet[a])[b - a]\right)^n = \Hom^{n + b -a}(M^\bullet[b], K^\bullet[a])\) and hence we get \[\begin{align*} d(f) & = (-1)^{b - a} \left(d_{K[a]} \circ f - (-1)^{n + b - a} f \circ d_{M[b]}\right) \\ & = (-1)^{b - a} \left((-1)^a d_K \circ f - (-1)^{n + b - a} f \circ (-1)^b d_M \right) \\ & = (-1)^b d_K \circ f - (-1)^{n + b} f \circ d_M \end{align*}\] and one sees that the chosen sign of \((-1)^{nb}\) in degree \(n\) produces a map of complexes for these differentials.
Derived hom
Let \(R\) be a ring. The derived hom we will define in this section is a functor \[D(R)^{opp} \times D(R) \longrightarrow D(R),\quad (K, L) \longmapsto R\Hom_R(K, L)\] This is an internal hom in the derived category of \(R\)-modules in the sense that it is characterized by the formula [0A63]\[\begin{equation} \Hom_{D(R)}(K, R\Hom_R(L, M)) = \Hom_{D(R)}(K \otimes_R^\mathbf{L} L, M) \end{equation}\] for objects \(K, L, M\) of \(D(R)\). Note that this formula characterizes the objects up to unique isomorphism by the Yoneda lemma. A construction can be given as follows. Choose a K-injective complex \(I^\bullet\) of \(R\)-modules representing \(M\), choose a complex \(L^\bullet\) representing \(L\), and set \[R\Hom_R(L, M) = \Hom^\bullet(L^\bullet, I^\bullet)\] with notation as in Section 0A8H. A generalization of this construction is discussed in Differential Graded Algebra, Section 09LF. From (0A5X) and Derived Categories, Lemma 070I that we have [0A64]\[\begin{equation} H^n(R\Hom_R(L, M)) = \Hom_{D(R)}(L, M[n]) \end{equation}\] for all \(n \in \mathbf{Z}\). In particular, the object \(R\Hom_R(L, M)\) of \(D(R)\) is well defined, i.e., independent of the choice of the K-injective complex \(I^\bullet\).
Lemma
Let \(R\) be a ring. Let \(K, L, M\) be objects of \(D(R)\). There is a canonical isomorphism \[R\Hom_R(K, R\Hom_R(L, M)) = R\Hom_R(K \otimes_R^\mathbf{L} L, M)\] in \(D(R)\) functorial in \(K, L, M\) which recovers (0A63) by taking \(H^0\).
Proof
Choose a K-injective complex \(I^\bullet\) representing \(M\) and a K-flat complex of \(R\)-modules \(L^\bullet\) representing \(L\). For any complex of \(R\)-modules \(K^\bullet\) we have \[\Hom^\bullet(K^\bullet, \Hom^\bullet(L^\bullet, I^\bullet)) = \Hom^\bullet(\text{Tot}(K^\bullet \otimes_R L^\bullet), I^\bullet)\] by Lemma 0A5Y. The lemma follows by the definition of \(R\Hom\) and because \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\) represents the derived tensor product.
Lemma
Let \(R\) be a ring. Let \(P^\bullet\) be a bounded above complex of projective \(R\)-modules. Let \(L^\bullet\) be a complex of \(R\)-modules. Then \(R\Hom_R(P^\bullet, L^\bullet)\) is represented by the complex \(\Hom^\bullet(P^\bullet, L^\bullet)\).
Proof
By (0A5X) and Derived Categories, Lemma 064B the cohomology groups of the complex are “correct”. Hence if we choose a quasi-isomorphism \(L^\bullet \to I^\bullet\) with \(I^\bullet\) a K-injective complex of \(R\)-modules then the induced map \[\Hom^\bullet(P^\bullet, L^\bullet) \longrightarrow \Hom^\bullet(P^\bullet, I^\bullet)\] is a quasi-isomorphism. As the right hand side is our definition of \(R\Hom_R(P^\bullet, L^\bullet)\) we win.
Lemma
Let \(R\) be a ring. Let \(K, L, M\) be objects of \(D(R)\). There is a canonical morphism \[R\Hom_R(L, M) \otimes_R^\mathbf{L} K \longrightarrow R\Hom_R(R\Hom_R(K, L), M)\] in \(D(R)\) functorial in \(K, L, M\).
Proof
Choose a K-injective complex \(I^\bullet\) representing \(M\), a K-injective complex \(J^\bullet\) representing \(L\), and a K-flat complex \(K^\bullet\) representing \(K\). The map is defined using the map \[\text{Tot}(\Hom^\bullet(J^\bullet, I^\bullet) \otimes_R K^\bullet) \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, J^\bullet), I^\bullet)\] of Lemma 0A60. We omit the proof that this is functorial in all three objects of \(D(R)\).
Lemma
Let \(R\) be a ring. Given \(K, L, M\) in \(D(R)\) there is a canonical morphism \[R\Hom_R(L, M) \otimes_R^\mathbf{L} R\Hom_R(K, L) \longrightarrow R\Hom_R(K, M)\] in \(D(R)\) functorial in \(K, L, M\).
Proof
Choose a K-injective complex \(I^\bullet\) representing \(M\), a K-injective complex \(J^\bullet\) representing \(L\), and any complex of \(R\)-modules \(K^\bullet\) representing \(K\). By Lemma 0A8I there is a map of complexes \[\text{Tot}\left( \Hom^\bullet(J^\bullet, I^\bullet) \otimes_R \Hom^\bullet(K^\bullet, J^\bullet) \right) \longrightarrow \Hom^\bullet(K^\bullet, I^\bullet)\] The complexes of \(R\)-modules \(\Hom^\bullet(J^\bullet, I^\bullet)\), \(\Hom^\bullet(K^\bullet, J^\bullet)\), and \(\Hom^\bullet(K^\bullet, I^\bullet)\) represent \(R\Hom_R(L, M)\), \(R\Hom_R(K, L)\), and \(R\Hom_R(K, M)\). If we choose a K-flat complex \(H^\bullet\) and a quasi-isomorphism \(H^\bullet \to \Hom^\bullet(K^\bullet, J^\bullet)\), then there is a map \[\text{Tot}\left( \Hom^\bullet(J^\bullet, I^\bullet) \otimes_R H^\bullet \right) \longrightarrow \text{Tot}\left( \Hom^\bullet(J^\bullet, I^\bullet) \otimes_R \Hom^\bullet(K^\bullet, J^\bullet) \right)\] whose source represents \(R\Hom_R(L, M) \otimes_R^\mathbf{L} R\Hom_R(K, L)\). Composing the two displayed arrows gives the desired map. We omit the proof that the construction is functorial.
Lemma
Let \(R\) be a ring. Given complexes \(K, L, M\) in \(D(R)\) there is a canonical morphism \[K \otimes_R^\mathbf{L} R\Hom_R(M, L) \longrightarrow R\Hom_R(M, K \otimes_R^\mathbf{L} L)\] in \(D(R)\) functorial in \(K\), \(L\), \(M\).
Proof
Choose a K-flat complex \(K^\bullet\) representing \(K\), and a K-injective complex \(I^\bullet\) representing \(L\), and choose any complex \(M^\bullet\) representing \(M\). Choose a quasi-isomorphism \(\text{Tot}(K^\bullet \otimes_R I^\bullet) \to J^\bullet\) where \(J^\bullet\) is K-injective. Then we use the map \[\text{Tot}\left( K^\bullet \otimes_R \Hom^\bullet(M^\bullet, I^\bullet) \right) \to \Hom^\bullet(M^\bullet, \text{Tot}(K^\bullet \otimes_R I^\bullet)) \to \Hom^\bullet(M^\bullet, J^\bullet)\] where the first map is the map from Lemma 0BYM.
Lemma
Let \(R\) be a ring. Given complexes \(K, L\) in \(D(R)\) there is a canonical morphism \[K \longrightarrow R\Hom_R(L, K \otimes_R^\mathbf{L} L)\] in \(D(R)\) functorial in both \(K\) and \(L\).
Proof
This is a special case of Lemma 0BYN but we will also prove it directly. Choose a K-flat complex \(K^\bullet\) representing \(K\) and any complex \(L^\bullet\) representing \(L\). Choose a quasi-isomorphism \(\text{Tot}(K^\bullet \otimes_R L^\bullet) \to J^\bullet\) where \(J^\bullet\) is K-injective. Then we use the map \[K^\bullet \to \Hom^\bullet(L^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet)) \to \Hom^\bullet(L^\bullet, J^\bullet)\] where the first map is the map from Lemma 0A62.
Perfect complexes
A perfect complex is a pseudo-coherent complex of finite tor dimension. We will not use this as the definition, but define perfect complexes over a ring directly as follows.
Definition
Let \(R\) be a ring. Denote \(D(R)\) the derived category of the abelian category of \(R\)-modules.
An object \(K\) of \(D(R)\) is perfect if it is quasi-isomorphic to a bounded complex of finite projective \(R\)-modules.
An \(R\)-module \(M\) is perfect if \(M[0]\) is a perfect object in \(D(R)\).
For example, over a Noetherian ring a finite module is perfect if and only if it has finite projective dimension, see Lemma 066Q and Algebra, Definition 00O4.
Lemma
Let \(K^\bullet\) be an object of \(D(R)\). The following are equivalent
\(K^\bullet\) is perfect, and
\(K^\bullet\) is pseudo-coherent and has finite tor dimension.
If (1) and (2) hold and \(K^\bullet\) has tor-amplitude in \([a, b]\), then \(K^\bullet\) is quasi-isomorphic to a complex \(E^\bullet\) of finite projective \(R\)-modules with \(E^i = 0\) for \(i \not \in [a, b]\).
Proof
It is clear that (1) implies (2), see Lemmas 064U and 0654. Assume (2) holds and that \(K^\bullet\) has tor-amplitude in \([a, b]\). In particular, \(H^i(K^\bullet) = 0\) for \(i > b\). Choose a complex \(F^\bullet\) of finite free \(R\)-modules with \(F^i = 0\) for \(i > b\) and a quasi-isomorphism \(F^\bullet \to K^\bullet\) (Lemma 064U). Set \(E^\bullet = \tau_{\geq a}F^\bullet\). Note that \(E^i\) is finite free except \(E^a\) which is a finitely presented \(R\)-module. By Lemma 0653 \(E^a\) is flat. Hence by Algebra, Lemma 00NX we see that \(E^a\) is finite projective.
Lemma
Let \(M\) be a module over a ring \(R\). The following are equivalent
\(M\) is a perfect module, and
there exists a resolution \[0 \to F_d \to \ldots \to F_1 \to F_0 \to M \to 0\] with each \(F_i\) a finite projective \(R\)-module.
Proof
Assume (2). Then the complex \(E^\bullet\) with \(E^{-i} = F_i\) is quasi-isomorphic to \(M[0]\). Hence \(M\) is perfect. Conversely, assume (1). By Lemmas 0658 and 064T we can find resolution \(E^\bullet \to M\) with \(E^{-i}\) a finite free \(R\)-module. By Lemma 0653 we see that \(F_d = \Coker(E^{d - 1} \to E^d)\) is flat for some \(d\) sufficiently large. By Algebra, Lemma 00NX we see that \(F_d\) is finite projective. Hence \[0 \to F_d \to E^{-d+1} \to \ldots \to E^0 \to M \to 0\] is the desired resolution.
Lemma
Let \(R\) be a ring. Let \((K^\bullet, L^\bullet, M^\bullet, f, g, h)\) be a distinguished triangle in \(D(R)\). If two out of three of \(K^\bullet, L^\bullet, M^\bullet\) are perfect then the third is also perfect.
Proof
Lemma
Let \(R\) be a ring. If \(K^\bullet \oplus L^\bullet\) is perfect, then so are \(K^\bullet\) and \(L^\bullet\).
Proof
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a bounded complex of perfect \(R\)-modules. Then \(K^\bullet\) is a perfect complex.
Proof
Follows by induction on the length of the finite complex: use Lemma 066R and the stupid truncations.
Lemma
Let \(R\) be a ring. If \(K^\bullet \in D^b(R)\) and all its cohomology modules are perfect, then \(K^\bullet\) is perfect.
Proof
Follows by induction on the length of the finite complex: use Lemma 066R and the canonical truncations.
Lemma
Let \(A \to B\) be a ring map. Assume that \(B\) is perfect as an \(A\)-module. Let \(K^\bullet\) be a perfect complex of \(B\)-modules. Then \(K^\bullet\) is perfect as a complex of \(A\)-modules.
Proof
Using Lemma 0658 this translates into the corresponding results for pseudo-coherent modules and modules of finite tor dimension. See Lemma 066K and Lemma 064Z for those results.
Lemma
Let \(A \to B\) be a ring map. Let \(K^\bullet\) be a perfect complex of \(A\)-modules. Then \(K^\bullet \otimes_A^{\mathbf{L}} B\) is a perfect complex of \(B\)-modules.
Proof
Using Lemma 0658 this translates into the corresponding results for pseudo-coherent modules and modules of finite tor dimension. See Lemma 066L and Lemma 0650 for those results.
Lemma
Let \(A \to B\) be a flat ring map. Let \(M\) be a perfect \(A\)-module. Then \(M \otimes_A B\) is a perfect \(B\)-module.
Proof
By Lemma 066Q the assumption implies that \(M\) has a finite resolution \(F_\bullet\) by finite projective \(R\)-modules. As \(A \to B\) is flat the complex \(F_\bullet \otimes_A B\) is a finite length resolution of \(M \otimes_A B\) by finite projective modules over \(B\). Hence \(M \otimes_A B\) is perfect.
Lemma
Let \(R\) be a ring. If \(K\) and \(L\) are perfect objects of \(D(R)\), then \(K \otimes_R^\mathbf{L} L\) is a perfect object too.
Proof
We can prove this using the definition as follows. We may represent \(K\), resp. \(L\) by a bounded complex \(K^\bullet\), resp. \(L^\bullet\) of finite projective \(R\)-modules. Then \(K \otimes_R^\mathbf{L} L\) is represented by the bounded complex \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\). The terms of this complex are direct sums of the modules \(K^a \otimes_R L^b\). Since \(K^a\) and \(L^b\) are direct summands of finite free \(R\)-modules, so is \(M^a \otimes_R L^b\). Hence we conclude the terms of the complex \(\text{Tot}(K^\bullet \otimes_R L^\bullet)\) are finite projective.
Another proof can be given using the characterization of perfect complexes in Lemma 0658 and the corresponding lemmas for pseudo-coherent complexes (Lemma 0DJE) and for tor amplitude (Lemma 0B66 used with \(A = B = R\)).
Lemma
Let \(R\) be a ring. Let \(f_1, \ldots, f_r \in R\) be elements which generate the unit ideal. Let \(K^\bullet\) be a complex of \(R\)-modules. If for each \(i\) the complex \(K^\bullet \otimes_R R_{f_i}\) is perfect, then \(K^\bullet\) is perfect.
Proof
Using Lemma 0658 this translates into the corresponding results for pseudo-coherent modules and modules of finite tor dimension. See Lemma 066N and Lemma 066D for those results.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. Let \(R \to R'\) be a faithfully flat ring map. If the complex \(K^\bullet \otimes_R R'\) is perfect, then \(K^\bullet\) is perfect.
Proof
Using Lemma 0658 this translates into the corresponding results for pseudo-coherent modules and modules of finite tor dimension. See Lemma 068S and Lemma 068R for those results.
Lemma
Let \(R\) be a regular ring. Then
an \(R\)-module is perfect if and only if it is a finite \(R\)-module, and
a complex of \(R\)-modules \(K^\bullet\) is perfect if and only if \(K^\bullet \in D^b(R)\) and each \(H^i(K^\bullet)\) is a finite \(R\)-module.
Proof
Any perfect \(R\)-module is finite by definition. Conversely, let \(M\) be a finite \(R\)-module. Choose a resolution \[\ldots \to F_2 \xrightarrow{d_2} F_1 \xrightarrow{d_1} F_0 \to M \to 0\] with \(F_i\) finite free \(R\)-modules (Algebra, Lemma 00LP). Set \(M_i = \Ker(d_i)\). Denote \(U_i \subset \Spec(R)\) the set of primes \(\mathfrak p\) such that \(M_{i, \mathfrak p}\) is free; \(U_i\) is open by Algebra, Lemma 0GWM. We have a exact sequence \(0 \to M_{i + 1} \to F_{i + 1} \to M_i \to 0\). If \(\mathfrak p \in U_i\), then \(0 \to M_{i + 1, \mathfrak p} \to F_{i + 1, \mathfrak p} \to M_{i, \mathfrak p} \to 0\) splits. Thus \(M_{i + 1, \mathfrak p}\) is finite projective, hence free (Algebra, Lemma 00NX). This shows that \(U_i \subset U_{i + 1}\). We claim that \(\Spec(R) = \bigcup U_i\). Namely, for every prime ideal \(\mathfrak p\) the regular local ring \(R_\mathfrak p\) has finite global dimension by Algebra, Proposition 00O7. It follows that \(M_{i, \mathfrak p}\) is finite projective (hence free) for \(i \gg 0\) for example by Algebra, Lemma 00O5. Since the spectrum of \(R\) is Noetherian (Algebra, Lemma 00FQ) we conclude that \(U_n = \Spec(R)\) for some \(n\). Then \(M_n\) is a projective \(R\)-module by Algebra, Lemma 00NX. Thus \[0 \to M_n \to F_n \to \ldots \to F_1 \to M \to 0\] is a bounded resolution by finite projective modules and hence \(M\) is perfect. This proves part (1).
Let \(K^\bullet\) be a complex of \(R\)-modules. If \(K^\bullet\) is perfect, then it is in \(D^b(R)\) and it is quasi-isomorphic to a finite complex of finite projective \(R\)-modules so certainly each \(H^i(K^\bullet)\) is a finite \(R\)-module (as \(R\) is Noetherian). Conversely, suppose that \(K^\bullet\) is in \(D^b(R)\) and each \(H^i(K^\bullet)\) is a finite \(R\)-module. Then by (1) each \(H^i(K^\bullet)\) is a perfect \(R\)-module, whence \(K^\bullet\) is perfect by Lemma 066U
Lemma
Let \(A\) be a ring. Let \(K \in D(A)\) be perfect. Then \(K^\vee = R\Hom_A(K, A)\) is a perfect complex and \(K \cong (K^\vee)^\vee\). There are functorial isomorphisms \[L \otimes_A^\mathbf{L} K^\vee = R\Hom_A(K, L) \quad\text{and}\quad H^0(L \otimes_A^\mathbf{L} K^\vee) = \Ext_A^0(K, L)\] for \(L \in D(A)\).
Proof
We can represent \(K\) by a bounded complex \(K^\bullet\) of finite projective \(A\)-modules. By Lemma 0A66 the object \(K^\vee\) is represented by the complex \(E^\bullet = \Hom^\bullet(K^\bullet, A)\). Note that \(E^n = \Hom_A(K^{-n}, A)\) is finite projective as well. Thus \(E^\bullet\) is a bounded complex of finite projective modules and we conclude \(K^\vee\) is perfect. There is a canonical map \[K^\bullet = \text{Tot}(\Hom^\bullet(A, A) \otimes_A K^\bullet) \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, A), A)\] which up to sign uses the evaluation map in each degree, see Lemma 0A60. (For sign rules see Section 0FNG.) Thus this map defines a canonical isomorphism \((K^\vee)^\vee \cong K\) as the double dual of a finite projective module is itself.
The second equality follows from the first by Lemma 0A65 and Derived Categories, Lemma 064B as well as the definition of Ext groups, see Derived Categories, Section 06XP. Let \(L^\bullet\) be a complex of \(A\)-modules representing \(L\). By Section 0FNG item (0FNL) there is a canonical isomorphism \[\text{Tot}(L^\bullet \otimes_A E^\bullet) \longrightarrow \Hom^\bullet(K^\bullet, L^\bullet)\] of complexes of \(A\)-modules (this uses Lemma 0FNK to see that \(E^\bullet\) is the left dual of \(K^\bullet\) in the category of complexes). This proves the first displayed equality and the proof is complete.
Remark
Let \(A\) be a ring. Let \(K \in D(A)\) be perfect. If \(K\) has tor-amplitude in \([a, b]\) then its dual \(K^\vee\) has tor-amplitude in \([-b, -a]\). This follows from the last statement in Lemma 0658 and the description of \(K^\vee\) in the proof of Lemma 07VI.
Lemma
Let \(A\) be a ring. Let \((K_n)_{n \in \mathbf{N}}\) be a system of perfect objects of \(D(A)\). Let \(K = \text{hocolim} K_n\) be the derived colimit (Derived Categories, Definition 090Z). Then for any object \(E\) of \(D(A)\) we have \[R\Hom_A(K, E) = R\lim E \otimes^\mathbf{L}_A K_n^\vee\] where \((K_n^\vee)\) is the inverse system of dual perfect complexes.
Proof
By Lemma 07VI we have \(R\lim E \otimes^\mathbf{L}_A K_n^\vee = R\lim R\Hom_A(K_n, E)\) which fits into the distinguished triangle \[R\lim R\Hom_A(K_n, E) \to \prod R\Hom_A(K_n, E) \to \prod R\Hom_A(K_n, E)\] Because \(K\) similarly fits into the distinguished triangle \(\bigoplus K_n \to \bigoplus K_n \to K\) it suffices to show that \(\prod R\Hom_A(K_n, E) = R\Hom_A(\bigoplus K_n, E)\). This is a formal consequence of (0A63) and the fact that derived tensor product commutes with direct sums.
Lemma
Let \(R = \colim_{i \in I} R_i\) be a filtered colimit of rings.
Given a perfect \(K\) in \(D(R)\) there exists an \(i \in I\) and a perfect \(K_i\) in \(D(R_i)\) such that \(K \cong K_i \otimes_{R_i}^\mathbf{L} R\) in \(D(R)\).
Given \(0 \in I\) and \(K_0, L_0 \in D(R_0)\) with \(K_0\) perfect, we have \[\Hom_{D(R)}(K_0 \otimes_{R_0}^\mathbf{L} R, L_0 \otimes_{R_0}^\mathbf{L} R) = \colim_{i \geq 0} \Hom_{D(R_i)}(K_0 \otimes_{R_0}^\mathbf{L} R_i, L_0 \otimes_{R_0}^\mathbf{L} R_i)\]
In other words, the triangulated category of perfect complexes over \(R\) is the colimit of the triangulated categories of perfect complexes over \(R_i\).
Proof
We will use the results of Algebra, Lemmas 05LI and 05N7 without further mention. These lemmas in particular say that the category of finitely presented \(R\)-modules is the colimit of the categories of finitely presented \(R_i\)-modules. Since finite projective modules can be characterized as summands of finite free modules (Algebra, Lemma 00NX) we see that the same is true for the category of finite projective modules. This proves (1) by our definition of perfect objects of \(D(R)\).
To prove (2) we may represent \(K_0\) by a bounded complex \(K_0^\bullet\) of finite projective \(R_0\)-modules. We may represent \(L_0\) by a K-flat complex \(L_0^\bullet\) (Lemma 06Y4). Then we have \[\Hom_{D(R)}(K_0 \otimes_{R_0}^\mathbf{L} R, L_0 \otimes_{R_0}^\mathbf{L} R) = \Hom_{K(R)}(K_0^\bullet \otimes_{R_0} R, L_0^\bullet \otimes_{R_0} R)\] by Derived Categories, Lemma 064B. Similarly for the \(\Hom\) with \(R\) replaced by \(R_i\). Since in the right hand side only a finite number of terms are involved, since \[\Hom_R(K_0^p \otimes_{R_0} R, L_0^q \otimes_{R_0} R) = \colim_{i \geq 0} \Hom_{R_i}(K_0^p \otimes_{R_0} R_i, L_0^q \otimes_{R_0} R_i)\] by the lemmas cited at the beginning of the proof, and since filtered colimits are exact (Algebra, Lemma 00DB) we conclude that (2) holds as well.
Lifting complexes
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. The lifting problem we will consider is the following. Suppose given an object \(K\) of \(D(R)\) and a complex \(E^\bullet\) of \(R/I\)-modules such that \(E^\bullet\) represents \(K \otimes_R^\mathbf{L} R/I\) in \(D(R)\). Question: Does there exist a complex of \(R\)-modules \(P^\bullet\) lifting \(E^\bullet\) representing \(K\) in \(D(R)\)? In general the answer to this question is no, but in good cases something can be done. We first discuss lifting acyclic complexes.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(\mathcal{P}\) be a class of \(R\)-modules. Assume
each \(P \in \mathcal{P}\) is a projective \(R\)-module,
if \(P_1 \in \mathcal{P}\) and \(P_1 \oplus P_2 \in \mathcal{P}\), then \(P_2 \in \mathcal{P}\), and
if \(f : P_1 \to P_2\), \(P_1, P_2 \in \mathcal{P}\) is surjective modulo \(I\), then \(f\) is surjective.
Then given any bounded above acyclic complex \(E^\bullet\) whose terms are of the form \(P/IP\) for \(P \in \mathcal{P}\) there exists a bounded above acyclic complex \(P^\bullet\) whose terms are in \(\mathcal{P}\) lifting \(E^\bullet\).
Proof
Say \(E^i = 0\) for \(i > b\). Assume given \(n\) and a morphism of complexes \[\xymatrix{ & & P^n \ar[r] \ar[d] & P^{n + 1} \ar[r] \ar[d] & \ldots \ar[r] & P^b \ar[r] \ar[d] & 0 \ar[r] \ar[d] & \ldots \\ \ldots \ar[r] & E^{n - 1} \ar[r] & E^n \ar[r] & E^{n + 1} \ar[r] & \ldots \ar[r] & E^b \ar[r] & 0 \ar[r] & \ldots }\] with \(P^i \in \mathcal{P}\), with \(P^n \to P^{n + 1} \to \ldots \to P^b\) acyclic in degrees \(\geq n + 1\), and with vertical maps inducing isomorphisms \(P^i/IP^i \to E^i\). In this situation one can inductively choose isomorphisms \(P^i = Z^i \oplus Z^{i + 1}\) such that the maps \(P^i \to P^{i + 1}\) are given by \(Z^i \oplus Z^{i + 1} \to Z^{i + 1} \to Z^{i + 1} \oplus Z^{i + 2}\). By property (2) and arguing inductively we see that \(Z^i \in \mathcal{P}\). Choose \(P^{n - 1} \in \mathcal{P}\) and an isomorphism \(P^{n - 1}/IP^{n - 1} \to E^{n - 1}\). Since \(P^{n - 1}\) is projective and since \(Z^n/IZ^n = \Im(E^{n - 1} \to E^n)\), we can lift the map \(P^{n - 1} \to E^{n - 1} \to E^n\) to a map \(P^{n - 1} \to Z^n\). By property (3) the map \(P^{n - 1} \to Z^n\) is surjective. Thus we obtain an extension of the diagram by adding \(P^{n - 1}\) and the maps just constructed to the left of \(P^n\). Since a diagram of the desired form exists for \(n > b\) we conclude by induction on \(n\).
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(\mathcal{P}\) be a class of \(R\)-modules. Let \(K \in D(R)\) and let \(E^\bullet\) be a complex of \(R/I\)-modules representing \(K \otimes_R^\mathbf{L} R/I\). Assume
each \(P \in \mathcal{P}\) is a projective \(R\)-module,
\(P_1 \in \mathcal{P}\) and \(P_1 \oplus P_2 \in \mathcal{P}\) if and only if \(P_1, P_2 \in \mathcal{P}\),
if \(f : P_1 \to P_2\), \(P_1, P_2 \in \mathcal{P}\) is surjective modulo \(I\), then \(f\) is surjective,
\(E^\bullet\) is bounded above and \(E^i\) is of the form \(P/IP\) for \(P \in \mathcal{P}\), and
\(K\) can be represented by a bounded above complex whose terms are in \(\mathcal{P}\).
Then there exists a bounded above complex \(P^\bullet\) whose terms are in \(\mathcal{P}\) with \(P^\bullet/IP^\bullet\) isomorphic to \(E^\bullet\) and representing \(K\) in \(D(R)\).
Proof
By assumption (5) we can represent \(K\) by a bounded above complex \(K^\bullet\) whose terms are in \(\mathcal{P}\). Then \(K \otimes_R^\mathbf{L} R/I\) is represented by \(K^\bullet/IK^\bullet\). Since \(E^\bullet\) is a bounded above complex of projective \(R/I\)-modules by (4), we can choose a quasi-isomorphism \(\delta : E^\bullet \to K^\bullet/IK^\bullet\) (Derived Categories, Lemma 064B). Let \(C^\bullet\) be cone on \(\delta\) (Derived Categories, Definition 014E). The module \(C^i\) is the direct sum \(K^i/IK^i \oplus E^{i + 1}\) hence is of the form \(P/IP\) for some \(P \in \mathcal{P}\) as (2) says in particular that \(\mathcal{P}\) is preserved under taking sums. Since \(C^\bullet\) is acyclic, we can apply Lemma 0BC9 and find a acyclic lift \(A^\bullet\) of \(C^\bullet\). The complex \(A^\bullet\) is bounded above and has terms in \(\mathcal{P}\). In \[\xymatrix{ K^\bullet \ar@{..>}[r] \ar[d] & A^\bullet \ar[d] \\ K^\bullet/IK^\bullet \ar[r] & C^\bullet \ar[r] & E^\bullet[1] }\] we can find the dotted arrow making the diagram commute by Derived Categories, Lemma 0649. We will show below that it follows from (1), (2), (3) that \(K^i \to A^i\) is the inclusion of a direct summand for every \(i\). By property (2) we see that \(P^i = \Coker(K^i \to A^i)\) is in \(\mathcal{P}\). Thus we can take \(P^\bullet = \Coker(K^\bullet \to A^\bullet)[-1]\) to conclude.
To finish the proof we have to show the following: Let \(f : P_1 \to P_2\), \(P_1, P_2 \in \mathcal{P}\) and \(P_1/IP_1 \to P_2/IP_2\) is split injective with cokernel of the form \(P_3/IP_3\) for some \(P_3 \in \mathcal{P}\), then \(f\) is split injective. Write \(E_i = P_i/IP_i\). Then \(E_2 = E_1 \oplus E_3\). Since \(P_2\) is projective we can choose a map \(g : P_2 \to P_3\) lifting the map \(E_2 \to E_3\). By condition (3) the map \(g\) is surjective, hence split as \(P_3\) is projective. Set \(P_1' = \Ker(g)\) and choose a splitting \(P_2 = P'_1 \oplus P_3\). Then \(P'_1 \in \mathcal{P}\) by (2). We do not know that \(g \circ f = 0\), but we can consider the map \[P_1 \xrightarrow{f} P_2 \xrightarrow{projection} P'_1\] The composition modulo \(I\) is an isomorphism. Since \(P'_1\) is projective we can split \(P_1 = T \oplus P'_1\). If \(T = 0\), then we are done, because then \(P_2 \to P'_1\) is a splitting of \(f\). We see that \(T \in \mathcal{P}\) by (2). Calculating modulo \(I\) we see that \(T/IT = 0\). Since \(0 \in \mathcal{P}\) (as the summand of any \(P\) in \(\mathcal{P}\)) we see the map \(0 \to T\) is surjective and we conclude that \(T = 0\) as desired.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(E^\bullet\) be a complex of \(R/I\)-modules. Let \(K\) be an object of \(D(R)\). Assume that
\(E^\bullet\) is a bounded above complex of projective \(R/I\)-modules,
\(K \otimes_R^\mathbf{L} R/I\) is represented by \(E^\bullet\) in \(D(R/I)\), and
\(I\) is a nilpotent ideal.
Then there exists a bounded above complex \(P^\bullet\) of projective \(R\)-modules representing \(K\) in \(D(R)\) such that \(P^\bullet \otimes_R R/I\) is isomorphic to \(E^\bullet\).
Proof
We apply Lemma 0BCA using the class \(\mathcal{P}\) of all projective \(R\)-modules. Properties (1) and (2) of the lemma are immediate. Property (3) follows from Nakayama’s lemma (Algebra, Lemma 00DV). Property (4) follows from the fact that we can lift projective \(R/I\)-modules to projective \(R\)-modules, see Algebra, Lemma 07LV. To see that (5) holds it suffices to show that \(K\) is in \(D^{-}(R)\). Since we are given that \(K \otimes_R^\mathbf{L} R/I\) is in \(D^{-}(R/I)\) because \(E^\bullet\) is bounded above, this follows from Lemma 0H75.
Lemma
Let \(R' \to R\) be a surjective ring map whose kernel is a nilpotent ideal. Let \(K' \in D(R')\) and set \(K = K' \otimes_{R'}^\mathbf{L} R\). Then \(K\) is pseudo-coherent if and only if \(K'\) is pseudo-coherent.
Proof
One direction follows from Lemma 0650. For the other direction, assume \(K\) is pseudo-coherent. Then by Lemma 064U we can represent \(K\) by a bounded above complex \(E^\bullet\) of finite free \(R\)-modules. By Lemma 09AR we can represent \(K'\) by a bounded above complex \(P^\bullet\) of projective \(R'\)-modules such that \(P^n \otimes_{R'} R = E^n\). By Nakayama’s lemma we see that \(P^n\) is finite free and we conclude that \(K'\) is pseudo-coherent as well.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(E^\bullet\) be a complex of \(R/I\)-modules. Let \(K\) be an object of \(D(R)\). Assume that
\(E^\bullet\) is a bounded above complex of finite stably free \(R/I\)-modules,
\(K \otimes_R^\mathbf{L} R/I\) is represented by \(E^\bullet\) in \(D(R/I)\),
\(K^\bullet\) is pseudo-coherent, and
every element of \(1 + I\) is invertible.
Then there exists a bounded above complex \(P^\bullet\) of finite stably free \(R\)-modules representing \(K\) in \(D(R)\) such that \(P^\bullet \otimes_R R/I\) is isomorphic to \(E^\bullet\). Moreover, if \(E^i\) is free, then \(P^i\) is free.
Proof
We apply Lemma 0BCA using the class \(\mathcal{P}\) of all finite stably free \(R\)-modules. Property (1) of the lemma is immediate. Property (2) follows from Lemma 0BC4. Property (3) follows from Nakayama’s lemma (Algebra, Lemma 00DV). Property (4) follows from the fact that we can lift finite stably free \(R/I\)-modules to finite stably free \(R\)-modules, see Lemma 0BC5. Part (5) holds because a pseudo-coherent complex can be represented by a bounded above complex of finite free \(R\)-modules. The final assertion of the lemma follows from Lemma 0BC6.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a local ring. Let \(K \in D(R)\) be pseudo-coherent. Set \(d_i = \dim_\kappa H^i(K \otimes_R^\mathbf{L} \kappa)\). Then \(d_i < \infty\) and for some \(b \in \mathbf{Z}\) we have \(d_i = 0\) for \(i > b\). Then there exists a complex \[\ldots \to R^{\oplus d_{b - 2}} \to R^{\oplus d_{b - 1}} \to R^{\oplus d_b} \to 0 \to \ldots\] representing \(K\) in \(D(R)\). Moreover, this complex is unique up to isomorphism(!).
Proof
Observe that \(K \otimes_R^\mathbf{L} \kappa\) is pseudo-coherent as an object of \(D(\kappa)\), see Lemma 0650. Hence the cohomology spaces are finite dimensional and vanish above some cutoff. Every object of \(D(\kappa)\) is isomorphic in \(D(\kappa)\) to a complex \(E^\bullet\) with zero differentials. In particular \(E^i \cong \kappa^{\oplus d_i}\) is finite free. Applying Lemma 0BCB we obtain the existence.
If we have two complexes \(F^\bullet\) and \(G^\bullet\) with \(F^i\) and \(G^i\) free of rank \(d_i\) representing \(K\). Then we may choose a map of complexes \(\beta : F^\bullet \to G^\bullet\) representing the isomorphism \(F^\bullet \cong K \cong G^\bullet\), see Derived Categories, Lemma 064B. The induced map of complexes \(\beta \otimes 1 : F^\bullet \otimes_R^\mathbf{L} \kappa \to G^\bullet \otimes_R^\mathbf{L} \kappa\) must be an isomorphism of complexes as the differentials in \(F^\bullet \otimes_R^\mathbf{L} \kappa\) and \(G^\bullet \otimes_R^\mathbf{L} \kappa\) are zero. Thus \(\beta^i : F^i \to G^i\) is a map of finite free \(R\)-modules whose reduction modulo \(\mathfrak m\) is an isomorphism. Hence \(\beta^i\) is an isomorphism and we win.
Lemma
Let \(R\) be a ring. Let \(\mathfrak p \subset R\) be a prime. Let \(K \in D(R)\) be perfect. Set \(d_i = \dim_{\kappa(\mathfrak p)} H^i(K \otimes_R^\mathbf{L} \kappa(\mathfrak p))\). Then \(d_i < \infty\) and only a finite number are nonzero. Then there exists an \(f \in R\), \(f \not \in \mathfrak p\) and a complex \[\ldots \to 0 \to R_f^{\oplus d_a} \to R_f^{\oplus d_{a + 1}} \to \ldots \to R_f^{\oplus d_{b - 1}} \to R_f^{\oplus d_b} \to 0 \to \ldots\] representing \(K \otimes_R^\mathbf{L} R_f\) in \(D(R_f)\).
Proof
Observe that \(K \otimes_R^\mathbf{L} \kappa(\mathfrak p)\) is perfect as an object of \(D(\kappa(\mathfrak p))\), see Lemma 066W. Hence only a finite number of \(d_i\) are nonzero and they are all finite. Applying Lemma 0BCC we get a complex representing \(K\) having the desired shape over the local ring \(R_\mathfrak p\). We have \(R_\mathfrak p = \colim R_f\) for \(f \in R\), \(f \not \in \mathfrak p\) (Algebra, Lemma 00CR). We conclude by Lemma 0BC7. Some details omitted.
Lemma
Let \(R\) be a ring. Let \(\mathfrak p \subset R\) be a prime. Let \(M^\bullet\) and \(N^\bullet\) be bounded complexes of finite projective \(R\)-modules representing the same object of \(D(R)\). Then there exists an \(f \in R\), \(f \not \in \mathfrak p\) such that there is an isomorphism (!) of complexes \[M^\bullet_f \oplus P^\bullet \cong N^\bullet_f \oplus Q^\bullet\] where \(P^\bullet\) and \(Q^\bullet\) are finite direct sums of trivial complexes, i.e., complexes of the form the form \(\ldots \to 0 \to R_f \xrightarrow{1} R_f \to 0 \to \ldots\) (placed in arbitrary degrees).
Proof
If we have an isomorphism of the type described over the localization \(R_\mathfrak p\), then using that \(R_\mathfrak p = \colim R_f\) (Algebra, Lemma 00CR) we can descend the isomorphism to an isomorphism over \(R_f\) for some \(f\). Thus we may assume \(R\) is local and \(\mathfrak p\) is the maximal ideal. In this case the result follows from the uniqueness of a “minimal” complex representing a perfect object, see Lemma 0BCC, and the fact that any complex is a direct sum of a trivial complex and a minimal one (Algebra, Lemma 00MT).
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(E^\bullet\) be a complex of \(R/I\)-modules. Let \(K\) be an object of \(D(R)\). Assume that
\(E^\bullet\) is a bounded above complex of finite projective \(R/I\)-modules,
\(K \otimes_R^\mathbf{L} R/I\) is represented by \(E^\bullet\) in \(D(R/I)\),
\(K\) is pseudo-coherent, and
\((R, I)\) is a henselian pair.
Then there exists a bounded above complex \(P^\bullet\) of finite projective \(R\)-modules representing \(K\) in \(D(R)\) such that \(P^\bullet \otimes_R R/I\) is isomorphic to \(E^\bullet\). Moreover, if \(E^i\) is free, then \(P^i\) is free.
Proof
We apply Lemma 0BCA using the class \(\mathcal{P}\) of all finite projective \(R\)-modules. Properties (1) and (2) of the lemma are immediate. Property (3) follows from Nakayama’s lemma (Algebra, Lemma 00DV). Property (4) follows from the fact that we can lift finite projective \(R/I\)-modules to finite projective \(R\)-modules, see Lemma 0D4A. Property (5) holds because a pseudo-coherent complex can be represented by a bounded above complex of finite free \(R\)-modules. Thus Lemma 0BCA applies and we find \(P^\bullet\) as desired. The final assertion of the lemma follows from Lemma 0BC6.
Splitting complexes
In this section we discuss conditions which imply an object of the derived category of a ring is a direct sum of its truncations. Our method is to use the following lemma (under suitable hypotheses) to split the canonical distinguished triangles \[\tau_{\leq i}K \to K \to \tau_{\geq i + 1}K \to (\tau_{\leq i}K)[1]\] in \(D(R)\), see Derived Categories, Remark 08J5.
Lemma
Let \(R\) be a ring. Let \(K\) and \(L\) be objects of \(D(R)\). Assume \(L\) has projective-amplitude in \([a, b]\), for example if \(L\) is perfect of tor-amplitude in \([a, b]\).
If \(H^i(K) = 0\) for \(i \geq a\), then \(\Hom_{D(R)}(L, K) = 0\).
If \(H^i(K) = 0\) for \(i \geq a + 1\), then given any distinguished triangle \(K \to M \to L \to K[1]\) there is an isomorphism \(M \cong K \oplus L\) in \(D(R)\) compatible with the maps in the distinguished triangle.
If \(H^i(K) = 0\) for \(i \geq a\), then the isomorphism in (2) exists and is unique.
Proof
The assumption that \(L\) has projective-amplitude in \([a, b]\) means we can represent \(L\) by a complex \(L^\bullet\) of projective \(R\)-modules with \(L^i = 0\) for \(i \not \in [a, b]\), see Definition 0A5N. If \(L\) is perfect of tor-amplitude in \([a, b]\), then we can represent \(L\) by a complex \(L^\bullet\) of finite projective \(R\)-modules with \(L^i = 0\) for \(i \not \in [a, b]\), see Lemma 0658. If \(H^i(K) = 0\) for \(i \geq a\), then \(K\) is quasi-isomorphic to \(\tau_{\leq a - 1}K\). Hence we can represent \(K\) by a complex \(K^\bullet\) of \(R\)-modules with \(K^i = 0\) for \(i \geq a\). Then we obtain \[\Hom_{D(R)}(L, K) = \Hom_{K(R)}(L^\bullet, K^\bullet) = 0\] by Derived Categories, Lemma 064B. This proves (1). Under the hypotheses of (2) we see that \(\Hom_{D(R)}(L, K[1]) = 0\) by (1), hence the distinguished triangle is split by Derived Categories, Lemma 05QT. The uniqueness of (3) follows from (1).
Lemma
Let \(R\) be a ring. Let \(\mathfrak p \subset R\) be a prime ideal. Let \(K^\bullet\) be a pseudo-coherent complex of \(R\)-modules. Assume that for some \(i \in \mathbf{Z}\) the map \[H^i(K^\bullet) \otimes_R \kappa(\mathfrak p) \longrightarrow H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p))\] is surjective. Then there exists an \(f \in R\), \(f \not \in \mathfrak p\) such that \(\tau_{\geq i + 1}(K^\bullet \otimes_R R_f)\) is a perfect object of \(D(R_f)\) with tor amplitude in \([i + 1, \infty]\) and a canonical isomorphism \[K^\bullet \otimes_R R_f \cong \tau_{\leq i}(K^\bullet \otimes_R R_f) \oplus \tau_{\geq i + 1}(K^\bullet \otimes_R R_f)\] in \(D(R_f)\).
Proof
In this proof all tensor products are over \(R\) and we write \(\kappa = \kappa(\mathfrak p)\). We may assume that \(K^\bullet\) is a bounded above complex of finite free \(R\)-modules. Let us inspect what is happening in degree \(i\): \[\ldots \to K^{i - 1} \xrightarrow{d^{i - 1}} K^i \xrightarrow{d^i} K^{i + 1} \to \ldots\] Let \(0 \subset V \subset W \subset K^i \otimes \kappa\) be defined by the formulas \[V = \Im\left( K^{i - 1} \otimes \kappa \to K^i \otimes \kappa \right) \quad\text{and}\quad W = \Ker\left( K^i \otimes \kappa \to K^{i + 1} \otimes \kappa \right)\] Set \(\dim(V) = r\), \(\dim(W/V) = s\), and \(\dim(K^i \otimes \kappa/W) = t\). We can pick \(x_1, \ldots, x_r \in K^{i - 1}\) which map by \(d^{i - 1}\) to a basis of \(V\). By our assumption we can pick \(y_1, \ldots, y_s \in \Ker(d^i)\) mapping to a basis of \(W/V\). Finally, choose \(z_1, \ldots, z_t \in K^i\) mapping to a basis of \(K^i \otimes \kappa/W\). Then we see that the elements \(d^i(z_1), \ldots, d^i(z_t) \in K^{i + 1}\) are linearly independent in \(K^{i + 1} \otimes_R \kappa\). By Algebra, Lemma 00O0 we may after replacing \(R\) by \(R_f\) for some \(f \in R\), \(f \not \in \mathfrak p\) assume that
\(d^i(x_a), y_b, z_c\) is an \(R\)-basis of \(K^i\),
\(d^i(z_1), \ldots, d^i(z_t)\) are \(R\)-linearly independent in \(K^{i + 1}\), and
the quotient \(E^{i + 1} = K^{i + 1}/\sum Rd^i(z_c)\) is finite projective.
Since \(d^i\) annihilates \(d^{i - 1}(x_a)\) and \(y_b\), we deduce from condition (2) that \(E^{i + 1} = \Coker(d^i : K^i \to K^{i + 1})\). Thus we see that \[\tau_{\geq i + 1}K^\bullet = (\ldots \to 0 \to E^{i + 1} \to K^{i + 2} \to \ldots)\] is a bounded complex of finite projective modules sitting in degrees \([i + 1, b]\) for some \(b\). Thus \(\tau_{\geq i + 1}K^\bullet\) is perfect of amplitude \([i + 1, b]\). Since \(\tau_{\leq i}K^\bullet\) has no cohomology in degrees \(> i\), we may apply Lemma 0BCG to the distinguished triangle \[\tau_{\leq i}K^\bullet \to K^\bullet \to \tau_{\geq i + 1}K^\bullet \to (\tau_{\leq i}K^\bullet)[1]\] (Derived Categories, Remark 08J5) to conclude.
Lemma
Let \(R\) be a ring. Let \(\mathfrak p \subset R\) be a prime ideal. Let \(K^\bullet\) be a pseudo-coherent complex of \(R\)-modules. Assume that for some \(i \in \mathbf{Z}\) the maps \[H^i(K^\bullet) \otimes_R \kappa(\mathfrak p) \longrightarrow H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p)) \quad\text{and}\quad H^{i - 1}(K^\bullet) \otimes_R \kappa(\mathfrak p) \longrightarrow H^{i - 1}(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p))\] are surjective. Then there exists an \(f \in R\), \(f \not \in \mathfrak p\) such that
\(\tau_{\geq i + 1}(K^\bullet \otimes_R R_f)\) is a perfect object of \(D(R_f)\) with tor amplitude in \([i + 1, \infty]\),
\(H^i(K^\bullet)_f\) is a finite free \(R_f\)-module, and
there is a canonical direct sum decomposition \[K^\bullet \otimes_R R_f \cong \tau_{\leq i - 1}(K^\bullet \otimes_R R_f) \oplus H^i(K^\bullet)_f[-i] \oplus \tau_{\geq i + 1}(K^\bullet \otimes_R R_f)\] in \(D(R_f)\).
Proof
We get (1) from Lemma 0A1U as well as a splitting \(K^\bullet \otimes_R R_f = \tau_{\leq i}K^\bullet \otimes_R R_f \oplus \tau_{\geq i + 1}K^\bullet \otimes_R R_f\) in \(D(R_f)\). Applying Lemma 0A1U once more to \(\tau_{\leq i}K^\bullet \otimes_R R_f\) we obtain (after suitably choosing \(f\)) a splitting \(\tau_{\leq i}K^\bullet \otimes_R R_f = \tau_{\leq i - 1}K^\bullet \otimes_R R_f \oplus H^i(K^\bullet)_f\) in \(D(R_f)\) as well as the conclusion that \(H^i(K)_f\) is a flat perfect module, i.e., finite projective.
Lemma
Let \(R\) be a ring. Let \(\mathfrak p \subset R\) be a prime ideal. Let \(i \in \mathbf{Z}\). Let \(K^\bullet\) be a pseudo-coherent complex of \(R\)-modules such that \(H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p)) = 0\). Then there exists an \(f \in R\), \(f \not \in \mathfrak p\) and a canonical direct sum decomposition \[K^\bullet \otimes_R R_f = \tau_{\geq i + 1}(K^\bullet \otimes_R R_f) \oplus \tau_{\leq i - 1}(K^\bullet \otimes_R R_f)\] in \(D(R_f)\) with \(\tau_{\geq i + 1}(K^\bullet \otimes_R R_f)\) a perfect complex with tor-amplitude in \([i + 1, \infty]\).
Proof
This is an often used special case of Lemma 0A1U. A direct proof is as follows. We may assume that \(K^\bullet\) is a bounded above complex of finite free \(R\)-modules. Let us inspect what is happening in degree \(i\): \[\ldots \to K^{i - 2} \to R^{\oplus l} \to R^{\oplus m} \to R^{\oplus n} \to K^{i + 2} \to \ldots\] Let \(A\) be the \(m \times l\) matrix corresponding to \(K^{i - 1} \to K^i\) and let \(B\) be the \(n \times m\) matrix corresponding to \(K^i \to K^{i + 1}\). The assumption is that \(A \bmod \mathfrak p\) has rank \(r\) and that \(B \bmod \mathfrak p\) has rank \(m - r\). In other words, there is some \(r \times r\) minor \(a\) of \(A\) which is not in \(\mathfrak p\) and there is some \((m - r) \times (m - r)\)-minor \(b\) of \(B\) which is not in \(\mathfrak p\). Set \(f = ab\). Then after inverting \(f\) we can find direct sum decompositions \(K^{i - 1} = R^{\oplus l - r} \oplus R^{\oplus r}\), \(K^i = R^{\oplus r} \oplus R^{\oplus m - r}\), \(K^{i + 1} = R^{\oplus m - r} \oplus R^{\oplus n - m + r}\) such that the module map \(K^{i - 1} \to K^i\) kills of \(R^{\oplus l - r}\) and induces an isomorphism of \(R^{\oplus r}\) onto the corresponding summand of \(K^i\) and such that the module map \(K^i \to K^{i + 1}\) kills of \(R^{\oplus r}\) and induces an isomorphism of \(R^{\oplus m - r}\) onto the corresponding summand of \(K^{i + 1}\). Thus \(K^\bullet\) becomes quasi-isomorphic to \[\ldots \to K^{i - 2} \to R^{\oplus l - r} \to 0 \to R^{\oplus n - m + r} \to K^{i + 2} \to \ldots\] and everything is clear.
Lemma
Let \(R\) be a ring. Let \(K \in D^-(R)\). Let \(a \in \mathbf{Z}\). Assume that for any injective \(R\)-module map \(M \to M'\) the map \(\Ext^{-a}_R(K, M) \to \Ext^{-a}_R(K, M')\) is injective. Then there is a unique direct sum decomposition \(K \cong \tau_{\leq a}K \oplus \tau_{\geq a + 1}K\) and \(\tau_{\geq a + 1}K\) has projective-amplitude in \([a + 1, b]\) for some \(b\).
Proof
Consider the distinguished triangle \[\tau_{\leq a}K \to K \to \tau_{\geq a + 1}K \to (\tau_{\leq a}K)[1]\] in \(D(R)\), see Derived Categories, Remark 08J5. Observe that \(\Ext^{-a}_R(\tau_{\leq a}K, M) = \Hom_R(H^a(K), M)\) and \(\Ext^{-a - 1}_R(\tau_{\leq a}K, M) = 0\), see Derived Categories, Lemma 06XS. Thus the long exact sequence of \(\Ext\) gives an exact sequence \[0 \to \Ext^{-a}_R(\tau_{\geq a + 1}K, M) \to \Ext^{-a}_R(K, M) \to \Hom_R(H^a(K), M)\] functorial in the \(R\)-module \(M\). Now if \(I\) is an injective \(R\)-module, then \(\Ext^{-a}_R(\tau_{\geq a + 1}K, I) = 0\) for example by Derived Categories, Lemma 06XR. Since every module injects into an injective module, we conclude that \(\Ext^{-a}_R(\tau_{\geq a + 1}K, M) = 0\) for every \(R\)-module \(M\). By Lemma 0A5P we conclude that \(\tau_{\geq a + 1}K\) has projective-amplitude in \([a + 1, b]\) for some \(b\) (this is where we use that \(K\) is bounded above). We obtain the splitting by Lemma 0BCG.
Lemma
Let \(R\) be a ring. Let \(K \in D^-(R)\). Let \(a \in \mathbf{Z}\). Assume \(\Ext^{-a}_R(K, M) = 0\) for any \(R\)-module \(M\). Then there is a unique direct sum decomposition \(K \cong \tau_{\leq a - 1}K \oplus \tau_{\geq a + 1}K\) and \(\tau_{\geq a + 1}K\) has projective-amplitude in \([a + 1, b]\) for some \(b\).
Proof
By Lemma 0G97 we have a direct sum decomposition \(K \cong \tau_{\leq a}K \oplus \tau_{\geq a + 1}K\) and \(\tau_{\geq a + 1}K\) has projective-amplitude in \([a + 1, b]\) for some \(b\). Clearly, we must have \(H^a(K) = 0\) and we conclude that \(\tau_{\leq a}K = \tau_{\leq a - 1}K\) in \(D(R)\).
Recognizing perfect complexes
Some lemmas that allow us to prove certain complexes are perfect.
Lemma
Let \(R\) be a ring and let \(\mathfrak p \subset R\) be a prime. Let \(K\) be pseudo-coherent and bounded below. Set \(d_i = \dim_{\kappa(\mathfrak p)} H^i(K \otimes_R^\mathbf{L} \kappa(\mathfrak p))\). If there exists an \(a \in \mathbf{Z}\) such that \(d_i = 0\) for \(i < a\), then there exists an \(f \in R\), \(f \not \in \mathfrak p\) and a complex \[\ldots \to 0 \to R_f^{\oplus d_a} \to R_f^{\oplus d_{a + 1}} \to \ldots \to R_f^{\oplus d_{b - 1}} \to R_f^{\oplus d_b} \to 0 \to \ldots\] representing \(K \otimes_R^\mathbf{L} R_f\) in \(D(R_f)\). In particular \(K \otimes_R^\mathbf{L} R_f\) is perfect.
Proof
After decreasing \(a\) we may assume that also \(H^i(K^\bullet) = 0\) for \(i < a\). By Lemma 068U after replacing \(R\) by \(R_f\) for some \(f \in R\), \(f \not \in \mathfrak p\) we can write \(K^\bullet = \tau_{\leq a - 1}K^\bullet \oplus \tau_{\geq a}K^\bullet\) in \(D(R)\) with \(\tau_{\geq a}K^\bullet\) perfect. Since \(H^i(K^\bullet) = 0\) for \(i < a\) we see that \(\tau_{\leq a - 1}K^\bullet = 0\) in \(D(R)\). Hence \(K^\bullet\) is perfect. Then we can conclude using Lemma 0BCD.
Lemma
Let \(R\) be a ring. Let \(a, b \in \mathbf{Z}\). Let \(K^\bullet\) be a pseudo-coherent complex of \(R\)-modules. The following are equivalent
\(K^\bullet\) is perfect with tor amplitude in \([a, b]\),
for every prime \(\mathfrak p\) we have \(H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p)) = 0\) for all \(i \not \in [a, b]\), and
for every maximal ideal \(\mathfrak m\) we have \(H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak m)) = 0\) for all \(i \not \in [a, b]\).
Proof
We omit the proof of the implications (1) \(\Rightarrow\) (2) \(\Rightarrow\) (3). Assume (3). Let \(i \in \mathbf{Z}\) with \(i \not \in [a, b]\). By Lemma 068U we see that the assumption implies that \(H^i(K^\bullet)_{\mathfrak m} = 0\) for all maximal ideals of \(R\). Hence \(H^i(K^\bullet) = 0\), see Algebra, Lemma 00HN. Moreover, Lemma 068U now also implies that for every maximal ideal \(\mathfrak m\) there exists an element \(f \in R\), \(f \not \in \mathfrak m\) such that \(K^\bullet \otimes_R R_f\) is perfect with tor amplitude in \([a, b]\). Hence we conclude by appealing to Lemmas 066Y and 066N.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a pseudo-coherent complex of \(R\)-modules. Consider the following conditions
\(K^\bullet\) is perfect,
for every prime ideal \(\mathfrak p\) the complex \(K^\bullet \otimes_R R_{\mathfrak p}\) is perfect,
for every maximal ideal \(\mathfrak m\) the complex \(K^\bullet \otimes_R R_{\mathfrak m}\) is perfect,
for every prime \(\mathfrak p\) we have \(H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak p)) = 0\) for all \(i \ll 0\),
for every maximal ideal \(\mathfrak m\) we have \(H^i(K^\bullet \otimes_R^{\mathbf{L}} \kappa(\mathfrak m)) = 0\) for all \(i \ll 0\).
We always have the implications \[(1) \Rightarrow (2) \Leftrightarrow (3) \Leftrightarrow (4) \Leftrightarrow (5)\] If \(K^\bullet\) is bounded below, then all conditions are equivalent.
Proof
By Lemma 066W we see that (1) implies (2). It is immediate that (2) \(\Rightarrow\) (3). Since every prime \(\mathfrak p\) is contained in a maximal ideal \(\mathfrak m\), we can apply Lemma 066W to the map \(R_\mathfrak m \to R_\mathfrak p\) to see that (3) implies (2). Applying Lemma 066W to the residue maps \(R_\mathfrak p \to \kappa(\mathfrak p)\) and \(R_\mathfrak m \to \kappa(\mathfrak m)\) we see that (2) implies (4) and (3) implies (5).
Assume \(R\) is local with maximal ideal \(\mathfrak m\) and residue field \(\kappa\). We will show that if \(H^i(K^\bullet \otimes^\mathbf{L} \kappa) = 0\) for \(i < a\) for some \(a\), then \(K\) is perfect. This will show that (4) implies (2) and (5) implies (3) whence the first part of the lemma. First we apply Lemma 068U with \(i = a - 1\) to see that \(K^\bullet = \tau_{\leq a - 1}K^\bullet \oplus \tau_{\geq a}K^\bullet\) in \(D(R)\) with \(\tau_{\geq a}K^\bullet\) perfect of tor-amplitude contained in \([a, \infty]\). To finish we need to show that \(\tau_{\leq a - 1}K\) is zero, i.e., that its cohomology groups are zero. If not let \(i\) be the largest index such that \(M = H^i(\tau_{\leq a - 1}K)\) is not zero. Then \(M\) is a finite \(R\)-module because \(\tau_{\leq a - 1}K^\bullet\) is pseudo-coherent (Lemmas 064S and 064X). Thus by Nakayama’s lemma (Algebra, Lemma 00DV) we find that \(M \otimes_R \kappa\) is nonzero. This implies that \[H^i((\tau_{\leq a - 1}K^\bullet) \otimes_R^\mathbf{L} \kappa) = H^i(K^\bullet \otimes_R^\mathbf{L} \kappa)\] is nonzero which is a contradiction.
Assume the equivalent conditions (2) – (5) hold and that \(K^\bullet\) is bounded below. Say \(H^i(K^\bullet) = 0\) for \(i < a\). Pick a maximal ideal \(\mathfrak m\) of \(R\). It suffices to show there exists an \(f \in R\), \(f \not \in \mathfrak m\) such that \(K^\bullet \otimes_R^\mathbf{L} R_f\) is perfect (Lemma 066Y and Algebra, Lemma 00E8). This follows from Lemma 0BYP.
Lemma
Let \(R\) be a ring. Let \(K\) be a pseudo-coherent object of \(D(R)\). Let \(a, b \in \mathbf{Z}\). The following are equivalent
\(K\) has projective-amplitude in \([a, b]\),
\(K\) is perfect of tor-amplitude in \([a, b]\),
\(\Ext^i_R(K, N) = 0\) for all finitely presented \(R\)-modules \(N\) and all \(i \not \in [-b, -a]\),
\(H^n(K) = 0\) for \(n > b\) and \(\Ext^i_R(K, N) = 0\) for all finitely presented \(R\)-modules \(N\) and all \(i > -a\), and
\(H^n(K) = 0\) for \(n \not \in [a - 1, b]\) and \(\Ext^{-a + 1}_R(K, N) = 0\) for all finitely presented \(R\)-modules \(N\).
Proof
From the final statement of Lemma 0658 we see that (2) implies (1). If (1) holds, then \(K\) can be represented by a complex of projective modules \(P^i\) with \(P^i = 0\) for \(i \not \in [a, b]\). Since projective modules are flat (as summands of free modules), we see that \(K\) has tor-amplitude in \([a, b]\), see Lemma 0654. Thus by Lemma 0658 we see that (2) holds.
In conditions (3), (4), (5) the assumed vanishing of ext groups \(\Ext^i_R(K, M)\) for \(M\) of finite presentation is equivalent to the vanishing for all \(R\)-modules \(M\) by Lemma 0G8W and Algebra, Lemma 00HA. Thus the equivalence of (1), (3), (4), and (5) follows from Lemma 0A5P.
The following lemma useful in order to find perfect complexes over a polynomial ring \(B = A[x_1, \ldots, x_d]\).
Lemma
Let \(A \to B\) be a ring map. Let \(a, b \in \mathbf{Z}\). Let \(d \geq 0\). Let \(K^\bullet\) be a complex of \(B\)-modules. Assume
the ring map \(A \to B\) is flat,
for every prime \(\mathfrak p \subset A\) the ring \(B \otimes_A \kappa(\mathfrak p)\) has finite global dimension \(\leq d\),
\(K^\bullet\) is pseudo-coherent as a complex of \(B\)-modules, and
\(K^\bullet\) has tor amplitude in \([a, b]\) as a complex of \(A\)-modules.
Then \(K^\bullet\) is perfect as a complex of \(B\)-modules with tor amplitude in \([a - d, b]\).
Proof
We may assume that \(K^\bullet\) is a bounded above complex of finite free \(B\)-modules. In particular, \(K^\bullet\) is flat as a complex of \(A\)-modules and \(K^\bullet \otimes_A M = K^\bullet \otimes_A^{\mathbf{L}} M\) for any \(A\)-module \(M\). For every prime \(\mathfrak p\) of \(A\) the complex \[K^\bullet \otimes_A \kappa(\mathfrak p)\] is a bounded above complex of finite free modules over \(B \otimes_A \kappa(\mathfrak p)\) with vanishing \(H^i\) except for \(i \in [a, b]\). As \(B \otimes_A \kappa(\mathfrak p)\) has global dimension \(d\) we see from Lemma 066P that \(K^\bullet \otimes_A \kappa(\mathfrak p)\) has tor amplitude in \([a - d, b]\). Let \(\mathfrak q\) be a prime of \(B\) lying over \(\mathfrak p\). Since \(K^\bullet \otimes_A \kappa(\mathfrak p)\) is a bounded above complex of free \(B \otimes_A \kappa(\mathfrak p)\)-modules we see that \[\begin{align*} K^\bullet \otimes_B^{\mathbf{L}} \kappa(\mathfrak q) & = K^\bullet \otimes_B \kappa(\mathfrak q) \\ & = (K^\bullet \otimes_A \kappa(\mathfrak p)) \otimes_{B \otimes_A \kappa(\mathfrak p)} \kappa(\mathfrak q) \\ & = (K^\bullet \otimes_A \kappa(\mathfrak p)) \otimes^{\mathbf{L}}_{B \otimes_A \kappa(\mathfrak p)} \kappa(\mathfrak q) \end{align*}\] Hence the arguments above imply that \(H^i(K^\bullet \otimes_B^{\mathbf{L}} \kappa(\mathfrak q)) = 0\) for \(i \not \in [a - d, b]\). We conclude by Lemma 068V.
The following lemma is a local version of Lemma 068X. It can be used to find perfect complexes over regular local rings.
Lemma
Let \(A \to B\) be a local ring homomorphism. Let \(a, b \in \mathbf{Z}\). Let \(d \geq 0\). Let \(K^\bullet\) be a complex of \(B\)-modules. Assume
the ring map \(A \to B\) is flat,
the ring \(B/\mathfrak m_AB\) is regular of dimension \(d\),
\(K^\bullet\) is pseudo-coherent as a complex of \(B\)-modules, and
\(K^\bullet\) has tor amplitude in \([a, b]\) as a complex of \(A\)-modules, in fact it suffices if \(H^i(K^\bullet \otimes_A^\mathbf{L} \kappa(\mathfrak m_A))\) is nonzero only for \(i \in [a, b]\).
Then \(K^\bullet\) is perfect as a complex of \(B\)-modules with tor amplitude in \([a - d, b]\).
Proof
By (3) we may assume that \(K^\bullet\) is a bounded above complex of finite free \(B\)-modules. We compute \[\begin{align*} K^\bullet \otimes_B^{\mathbf{L}} \kappa(\mathfrak m_B) & = K^\bullet \otimes_B \kappa(\mathfrak m_B) \\ & = (K^\bullet \otimes_A \kappa(\mathfrak m_A)) \otimes_{B/\mathfrak m_A B} \kappa(\mathfrak m_B) \\ & = (K^\bullet \otimes_A \kappa(\mathfrak m_A)) \otimes^{\mathbf{L}}_{B/\mathfrak m_A B} \kappa(\mathfrak m_B) \end{align*}\] The first equality because \(K^\bullet\) is a bounded above complex of flat \(B\)-modules. The second equality follows from basic properties of the tensor product. The third equality holds because \(K^\bullet \otimes_A \kappa(\mathfrak m_A) = K^\bullet/ \mathfrak m_A K^\bullet\) is a bounded above complex of flat \(B/\mathfrak m_A B\)-modules. Since \(K^\bullet\) is a bounded above complex of flat \(A\)-modules by (1), the cohomology modules \(H^i\) of the complex \(K^\bullet \otimes_A \kappa(\mathfrak m_A)\) are nonzero only for \(i \in [a, b]\) by assumption (4). Thus the spectral sequence of Example 061Z and the fact that \(B/\mathfrak m_AB\) has finite global dimension \(d\) (by (2) and Algebra, Proposition 00O7) shows that \(H^j(K^\bullet \otimes_B^{\mathbf{L}} \kappa(\mathfrak m_B))\) is zero for \(j \not \in [a - d, b]\). This finishes the proof by Lemma 068V.
Characterizing perfect complexes
In this section we prove that the perfect complexes are exactly the compact objects of the derived category of a ring. First we show the following.
Lemma
Let \(R\) be a ring. The full subcategory \(D_{perf}(R) \subset D(R)\) of perfect objects is the smallest strictly full, saturated, triangulated subcategory containing \(R = R[0]\). In other words \(D_{perf}(R) = \langle R \rangle\). In particular, \(R\) is a classical generator for \(D_{perf}(R)\).
Proof
To see what the statement means, please look at Derived Categories, Definitions 05RB and 09SJ. It was shown in Lemmas 066R and 066S that \(D_{perf}(R) \subset D(R)\) is a strictly full, saturated, triangulated subcategory of \(D(R)\). Of course \(R \in D_{perf}(R)\).
Recall that \(\langle R \rangle = \bigcup \langle R \rangle_n\). To finish the proof we will show that if \(M \in D_{perf}(R)\) is represented by \[\ldots \to 0 \to M^a \to M^{a + 1} \to \ldots \to M^b \to 0 \to \ldots\] with \(M^i\) finite projective, then \(M \in \langle R \rangle_{b - a + 1}\). The proof is by induction on \(b - a\). By definition \(\langle R \rangle_1\) contains any finite projective \(R\)-module placed in any degree; this deals with the base case \(b - a = 0\) of the induction. In general, we consider the distinguished triangle \[M^b[-b] \to M^\bullet \to \sigma_{\leq b - 1}M^\bullet \to M^b[-b + 1]\] By induction the truncated complex \(\sigma_{\leq b - 1}M^\bullet\) is in \(\langle R \rangle_{b - a}\) and \(M^b[-b]\) is in \(\langle R \rangle_1\). Hence \(M^\bullet \in \langle R \rangle_{b - a + 1}\) by definition.
Let \(R\) be a ring. Recall that \(D(R)\) has direct sums which are given simply by taking direct sums of complexes, see Derived Categories, Lemma 0A5L. We will use this in the lemmas of this section without further mention.
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\) be an object such that for every countable set of objects \(E_n \in D(R)\) the canonical map \[\bigoplus \Hom_{D(R)}(K, E_n) \longrightarrow \Hom_{D(R)}(K, \bigoplus E_n)\] is a bijection. Then, given any system \(L_n^\bullet\) of complexes over \(\mathbf{N}\) we have that \[\colim \Hom_{D(R)}(K, L^\bullet_n) \longrightarrow \Hom_{D(R)}(K, L^\bullet)\] is a bijection, where \(L^\bullet\) is the termwise colimit, i.e., \(L^m = \colim L_n^m\) for all \(m \in \mathbf{Z}\).
Proof
Consider the short exact sequence of complexes \[0 \to \bigoplus L_n^\bullet \to \bigoplus L_n^\bullet \to L^\bullet \to 0\] where the first map is given by \(1 - t_n\) in degree \(n\) where \(t_n : L_n^\bullet \to L_{n + 1}^\bullet\) is the transition map. By Derived Categories, Lemma 0152 this is a distinguished triangle in \(D(R)\). Apply the homological functor \(\Hom_{D(R)}(K, -)\), see Derived Categories, Lemma 0149. Thus a long exact cohomology sequence \[\xymatrix{ & \ldots \ar[r] & \Hom_{D(R)}(K, \colim L^\bullet_n[-1]) \ar[lld] \\ \Hom_{D(R)}(K, \bigoplus L^\bullet_n) \ar[r] & \Hom_{D(R)}(K, \bigoplus L^\bullet_n) \ar[r] & \Hom_{D(R)}(K, \colim L^\bullet_n) \ar[lld] \\ \Hom_{D(R)}(K, \bigoplus L^\bullet_n[1]) \ar[r] & \ldots }\] Since we have assumed that \(\Hom_{D(R)}(K, \bigoplus L^\bullet_n)\) is equal to \(\bigoplus \Hom_{D(R)}(K, L^\bullet_n)\) we see that the first map on every row of the diagram is injective (by the explicit description of this map as the sum of the maps induced by \(1 - t_n\)). Hence we conclude that \(\Hom_{D(R)}(K, \colim L^\bullet_n)\) is the cokernel of the first map of the middle row in the diagram above which is what we had to show.
The following proposition, characterizing perfect complexes as the compact objects (Derived Categories, Definition 07LS) of the derived category, shows up in various places. See for example [Rickard, proof of Proposition 6.3] (this treats the bounded case), [TT, Theorem 2.4.3] (the statement doesn’t match exactly), and [Bokstedt-Neeman, Proposition 6.4] (watch out for horrendous notational conventions).
Proposition
Let \(R\) be a ring. For an object \(K\) of \(D(R)\) the following are equivalent
\(K\) is perfect, and
\(K\) is a compact object of \(D(R)\).
Proof
Assume \(K\) is perfect, i.e., \(K\) is quasi-isomorphic to a bounded complex \(P^\bullet\) of finite projective modules, see Definition 0657. If \(E_i\) is represented by the complex \(E_i^\bullet\), then \(\bigoplus E_i\) is represented by the complex whose degree \(n\) term is \(\bigoplus E_i^n\). On the other hand, as \(P^n\) is projective for all \(n\) we have \(\Hom_{D(R)}(P^\bullet, K^\bullet) = \Hom_{K(R)}(P^\bullet, K^\bullet)\) for every complex of \(R\)-modules \(K^\bullet\), see Derived Categories, Lemma 064B. Thus \(\Hom_{D(R)}(P^\bullet, E^\bullet)\) is the cohomology of the complex \[\prod \Hom_R(P^n, E^{n - 1}) \to \prod \Hom_R(P^n, E^n) \to \prod \Hom_R(P^n, E^{n + 1}).\] Since \(P^\bullet\) is bounded we see that we may replace the \(\prod\) signs by \(\bigoplus\) signs in the complex above. Since each \(P^n\) is a finite \(R\)-module we see that \(\Hom_R(P^n, \bigoplus_i E_i^m) = \bigoplus_i \Hom_R(P^n, E_i^m)\) for all \(n, m\). Combining these remarks we see that the map of Derived Categories, Definition 07LS is a bijection.
Conversely, assume \(K\) is compact. Represent \(K\) by a complex \(K^\bullet\) and consider the map \[K^\bullet \longrightarrow \bigoplus\nolimits_{n \geq 0} \tau_{\geq n} K^\bullet\] where we have used the canonical truncations, see Homology, Section 0118. This makes sense as in each degree the direct sum on the right is finite. By assumption this map factors through a finite direct sum. We conclude that \(K \to \tau_{\geq n} K\) is zero for at least one \(n\), i.e., \(K\) is in \(D^{-}(R)\).
Since \(K \in D^{-}(R)\) and since every \(R\)-module is a quotient of a free module, we may represent \(K\) by a bounded above complex \(K^\bullet\) of free \(R\)-modules, see Derived Categories, Lemma 05T7. Note that we have \[K^\bullet = \bigcup\nolimits_{n \leq 0} \sigma_{\geq n}K^\bullet\] where we have used the stupid truncations, see Homology, Section 0118. Hence by Lemma 07LR we see that \(1 : K^\bullet \to K^\bullet\) factors through \(\sigma_{\geq n}K^\bullet \to K^\bullet\) in \(D(R)\). Thus we see that \(1 : K^\bullet \to K^\bullet\) factors as \[K^\bullet \xrightarrow{\varphi} L^\bullet \xrightarrow{\psi} K^\bullet\] in \(D(R)\) for some complex \(L^\bullet\) which is bounded and whose terms are free \(R\)-modules. Say \(L^i = 0\) for \(i \not \in [a, b]\). Fix \(a, b\) from now on. Let \(c\) be the largest integer \(\leq b + 1\) such that we can find a factorization of \(1_{K^\bullet}\) as above with \(L^i\) finite free for \(i < c\). We will show by induction that \(c = b + 1\). Namely, write \(L^c = \bigoplus_{\lambda \in \Lambda} R\). Since \(L^{c - 1}\) is finite free we can find a finite subset \(\Lambda' \subset \Lambda\) such that \(L^{c - 1} \to L^c\) factors through \(\bigoplus_{\lambda \in \Lambda'} R \subset L^c\). Consider the map of complexes \[\pi : L^\bullet \longrightarrow (\bigoplus\nolimits_{\lambda \in \Lambda \setminus \Lambda'} R)[-c]\] given by the projection onto the factors corresponding to \(\Lambda \setminus \Lambda'\) in degree \(c\). By our assumption on \(K\) we see that, after possibly replacing \(\Lambda'\) by a larger finite subset, we may assume that \(\pi \circ \varphi = 0\) in \(D(R)\). Let \((L')^\bullet \subset L^\bullet\) be the kernel of \(\pi\). Since \(\pi\) is surjective we get a short exact sequence of complexes, which gives a distinguished triangle in \(D(R)\) (see Derived Categories, Lemma 0152). Since \(\Hom_{D(R)}(K, -)\) is homological (see Derived Categories, Lemma 0149) and \(\pi \circ \varphi = 0\), we can find a morphism \(\varphi' : K^\bullet \to (L')^\bullet\) in \(D(R)\) whose composition with \((L')^\bullet \to L^\bullet\) gives \(\varphi\). Setting \(\psi'\) equal to the composition of \(\psi\) with \((L')^\bullet \to L^\bullet\) we obtain a new factorization. Since \((L')^\bullet\) agrees with \(L^\bullet\) except in degree \(c\) and since \((L')^c = \bigoplus_{\lambda \in \Lambda'} R\) the induction step is proved.
The conclusion of the discussion of the preceding paragraph is that \(1_K : K \to K\) factors as \[K \xrightarrow{\varphi} L \xrightarrow{\psi} K\] in \(D(R)\) where \(L\) can be represented by a finite complex of free \(R\)-modules. In particular we see that \(L\) is perfect. Note that \(e = \varphi \circ \psi \in \text{End}_{D(R)}(L)\) is an idempotent. By Derived Categories, Lemma 05QW we see that \(L = \Ker(e) \oplus \Ker(1 - e)\). The map \(\varphi : K \to L\) induces an isomorphism with \(\Ker(1 - e)\) in \(D(R)\). Hence we finally conclude that \(K\) is perfect by Lemma 066S.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(K\) be an object of \(D(R)\). Assume that
\(K \otimes_R^\mathbf{L} R/I\) is perfect in \(D(R/I)\), and
\(I\) is a nilpotent ideal.
Then \(K\) is perfect in \(D(R)\).
Proof
Choose a finite complex \(\overline{P}^\bullet\) of finite projective \(R/I\)-modules representing \(K \otimes_R^\mathbf{L} R/I\), see Definition 0657. By Lemma 09AR there exists a complex \(P^\bullet\) of projective \(R\)-modules representing \(K\) such that \(\overline{P}^\bullet = P^\bullet/IP^\bullet\). It follows from Nakayama’s lemma (Algebra, Lemma 00DV) that \(P^\bullet\) is a finite complex of finite projective \(R\)-modules.
Lemma
Let \(R\) be a ring. Let \(I, J \subset R\) be ideals. Let \(K\) be an object of \(D(R)\). Assume that
\(K \otimes_R^\mathbf{L} R/I\) is perfect in \(D(R/I)\), and
\(K \otimes_R^\mathbf{L} R/J\) is perfect in \(D(R/J)\).
Then \(K \otimes_R^\mathbf{L} R/IJ\) is perfect in \(D(R/IJ)\).
Proof
It is clear that we may assume replace \(R\) by \(R/IJ\) and \(K\) by \(K \otimes_R^\mathbf{L} R/IJ\). Then \(R \to R/(I \cap J)\) is a surjection whose kernel has square zero. Hence by Lemma 07LU it suffices to prove that \(K \otimes_R^\mathbf{L} R/(I \cap J)\) is perfect. Thus we may assume that \(I \cap J = 0\).
We prove the lemma in case \(I \cap J = 0\). First, we may represent \(K\) by a K-flat complex \(K^\bullet\) with all \(K^n\) flat, see Lemma 06Y4. Then we see that we have a short exact sequence of complexes \[0 \to K^\bullet \to K^\bullet/IK^\bullet \oplus K^\bullet/JK^\bullet \to K^\bullet/(I + J)K^\bullet \to 0\] Note that \(K^\bullet/IK^\bullet\) represents \(K \otimes^\mathbf{L}_R R/I\) by construction of the derived tensor product. Similarly for \(K^\bullet/JK^\bullet\) and \(K^\bullet/(I + J)K^\bullet\). Note that \(K^\bullet/(I + J)K^\bullet\) is a perfect complex of \(R/(I + J)\)-modules, see Lemma 066W. Hence the complexes \(K^\bullet/IK^\bullet\), and \(K^\bullet/JK^\bullet\) and \(K^\bullet/(I + J)K^\bullet\) have finitely many nonzero cohomology groups (since a perfect complex has finite Tor-amplitude, see Lemma 0658). We conclude that \(K \in D^b(R)\) by the long exact cohomology sequence associated to short exact sequence of complexes displayed above. In particular we assume \(K^\bullet\) is a bounded above complex of free \(R\)-modules (see Derived Categories, Lemma 05T7).
We will now show that \(K\) is perfect using the criterion of Proposition 07LT. Thus we let \(E_j \in D(R)\) be a family of objects parametrized by a set \(J\). We choose complexes \(E_j^\bullet\) with flat terms representing \(E_j\), see for example Lemma 06Y4. It is clear that \[0 \to E_j^\bullet \to E_j^\bullet/IE_j^\bullet \oplus E_j^\bullet/JE_j^\bullet \to E_j^\bullet/(I + J)E_j^\bullet \to 0\] is a short exact sequence of complexes. Taking direct sums we obtain a similar short exact sequence \[0 \to \bigoplus E_j^\bullet \to \bigoplus E_j^\bullet/IE_j^\bullet \oplus E_j^\bullet/JE_j^\bullet \to \bigoplus E_j^\bullet/(I + J)E_j^\bullet \to 0\] (Note that \(- \otimes_R R/I\) commutes with direct sums.) This short exact sequence determines a distinguished triangle in \(D(R)\), see Derived Categories, Lemma 0152. Apply the homological functor \(\Hom_{D(R)}(K, -)\) (see Derived Categories, Lemma 0149) to get a commutative diagram \[\xymatrix{ \bigoplus \Hom_{D(R)}(K^\bullet, E_j^\bullet/(I + J))[-1] \ar[r] \ar[d] & \Hom_{D(R)}(K^\bullet, \bigoplus E_j^\bullet/(I + J))[-1] \ar[d] \\ \bigoplus \Hom_{D(R)}(K^\bullet, E_j^\bullet/I \oplus E_j^\bullet/J)[-1] \ar[r] \ar[d] & \Hom_{D(R)}(K^\bullet, \bigoplus E_j^\bullet/I \oplus E_j^\bullet/J)[-1] \ar[d] \\ \bigoplus \Hom_{D(R)}(K^\bullet, E_j^\bullet) \ar[r] \ar[d] & \Hom_{D(R)}(K^\bullet, \bigoplus E_j^\bullet) \ar[d] \\ \bigoplus \Hom_{D(R)}(K^\bullet, E_j^\bullet/I \oplus E_j^\bullet/J) \ar[r] \ar[d] & \Hom_{D(R)}(K^\bullet, \bigoplus E_j^\bullet/I \oplus E_j^\bullet/J) \ar[d] \\ \bigoplus \Hom_{D(R)}(K^\bullet, E_j^\bullet/(I + J)) \ar[r] & \Hom_{D(R)}(K^\bullet, \bigoplus E_j^\bullet/(I + J)) }\] with exact columns. It is clear that, for any complex \(E^\bullet\) of \(R\)-modules we have \[\begin{align*} \Hom_{D(R)}(K^\bullet, E^\bullet/I) & = \Hom_{K(R)}(K^\bullet, E^\bullet/I) \\ & = \Hom_{K(R/I)}(K^\bullet/IK^\bullet, E^\bullet/I) \\ & = \Hom_{D(R/I)}(K^\bullet/IK^\bullet, E^\bullet/I) \end{align*}\] and similarly for when dividing by \(J\) or \(I + J\), see Derived Categories, Lemma 064B. Thus all the horizontal arrows, except for possibly the middle one, are isomorphisms as the complexes \(K^\bullet/IK^\bullet\), \(K^\bullet/JK^\bullet\), \(K^\bullet/(I + J)K^\bullet\) are perfect complexes of \(R/I\), \(R/J\), \(R/(I + J)\)-modules, see Proposition 07LT. It follows from the \(5\)-lemma (Homology, Lemma 05QB) that the middle map is an isomorphism and the lemma follows by Proposition 07LT.
Strong generators and regular rings
Let \(R\) be a ring. Denote \(D(R)_c\) the saturated full triangulated subcategory of \(D(R)\). We already know that \[\langle R \rangle = D_{perf}(R) = D(R)_c\] See Lemma 0ATI and Proposition 07LT. It turns out that if \(R\) is regular, then \(R\) is a strong generator (Derived Categories, Definition 09SJ).
Lemma
Let \(R\) be a ring. Let \(n \geq 1\). Let \(K \in \langle R \rangle_n\) with notation as in Derived Categories, Section 09SI. Consider maps \[K \xrightarrow{f_1} K_1 \xrightarrow{f_2} K_2 \xrightarrow{f_3} \ldots \xrightarrow{f_n} K_n\] in \(D(R)\). If \(H^i(f_j) = 0\) for all \(i, j\), then \(f_n \circ \ldots \circ f_1 = 0\).
Proof
If \(n = 1\), then \(K\) is a direct summand in \(D(R)\) of a bounded complex \(P^\bullet\) whose terms are finite free \(R\)-modules and whose differentials are zero. Thus it suffices to show any morphism \(f : P^\bullet \to K_1\) in \(D(R)\) with \(H^i(f) = 0\) for all \(i\) is zero. Since \(P^\bullet\) is a finite direct sum \(P^\bullet = \bigoplus R[m_j]\) it suffices to show any morphism \(g : R[m] \to K_1\) with \(H^{-m}(g) = 0\) in \(D(R)\) is zero. This follows from the fact that \(\Hom_{D(R)}(R[-m], K) = H^m(K)\).
For \(n > 1\) we proceed by induction on \(n\). Namely, we know that \(K\) is a summand in \(D(R)\) of an object \(P\) which sits in a distinguished triangle \[P' \xrightarrow{i} P \xrightarrow{p} P'' \to P'[1]\] with \(P' \in \langle R \rangle_1\) and \(P'' \in \langle R \rangle_{n - 1}\). As above we may replace \(K\) by \(P\) and assume that we have \[P \xrightarrow{f_1} K_1 \xrightarrow{f_2} K_2 \xrightarrow{f_3} \ldots \xrightarrow{f_n} K_n\] in \(D(R)\) with \(f_j\) zero on cohomology. By the case \(n = 1\) the composition \(f_1 \circ i\) is zero. Hence by Derived Categories, Lemma 0149 we can find a morphism \(h : P'' \to K_1\) such that \(f_1 = h \circ p\). Observe that \(f_2 \circ h\) is zero on cohomology. Hence by induction we find that \(f_n \circ \ldots \circ f_2 \circ h = 0\) which implies \(f_n \circ \ldots \circ f_1 = f_n \circ \ldots \circ f_2 \circ h \circ p = 0\) as desired.
Lemma
Let \(R\) be a Noetherian ring. If \(R\) is a strong generator for \(D_{perf}(R)\), then \(R\) is regular of finite dimension.
Proof
Assume \(D_{perf}(R) = \langle R \rangle_n\) for some \(n \geq 1\). For any finite \(R\)-module \(M\) we can choose a complex \[P = ( P^{-n - 1} \xrightarrow{d^{-n - 1}} P^{-n} \xrightarrow{d^{-n}} P^{-n + 1} \xrightarrow{d^{-n + 1}} \ldots \xrightarrow{d^{-1}} P^0)\] of finite free \(R\)-modules with \(H^i(P) = 0\) for \(i = -n, \ldots, - 1\) and \(M \cong \Coker(d^{-1})\). Note that \(P\) is in \(D_{perf}(R)\). For any \(R\)-module \(N\) we can compute \(\Ext^n_R(M, N)\) the finite free resolution \(P\) of \(M\), see Algebra, Section 00LO and compare with Derived Categories, Section 06XP. In particular, the sequence above defines an element \[\xi \in \Ext^n_R(\Coker(d^{-1}), \Coker(d^{-n - 1})) = \Ext^n_R(M, \Coker(d^{-n - 1}))\] and for any element \(\overline{\xi}\) in \(\Ext^n_R(M, N)\) there is a \(R\)-module map \(\varphi : \Coker(d^{-n - 1}) \to N\) such that \(\varphi\) maps \(\xi\) to \(\overline{\xi}\). For \(j = 1, \ldots, n - 1\) consider the complexes \[K_j = (\Coker(d^{-n - 1}) \to P^{-n + 1} \to \ldots \to P^{-j})\] with \(\Coker(d^{-n - 1})\) in degree \(-n\) and \(P^t\) in degree \(t\). We also set \(K_n = \Coker(d^{-n - 1})[n]\). Then we have maps \[P \to K_1 \to K_2 \to \ldots \to K_n\] which induce vanishing maps on cohomology. By Lemma 0FXH since \(P \in D_{perf}(R) = \langle R \rangle_n\) we find that the composition of this maps is zero in \(D(R)\). Since \(\Hom_{D(R)}(P, K_n) = \Hom_{K(R)}(P, K_n)\) by Derived Categories, Lemma 064B we conclude \(\xi = 0\). Hence \(\Ext^n_R(M, N) = 0\) for all \(R\)-modules \(N\), see discussion above. It follows that \(M\) has projective dimension \(\leq n - 1\) by Algebra, Lemma 065R. Since this holds for all finite \(R\)-modules \(M\) we conclude that \(R\) has finite global dimension, see Algebra, Lemma 065T. We finally conclude by Algebra, Lemma 00OE.
Lemma
Let \(R\) be a Noetherian regular ring of dimension \(d < \infty\). Let \(K, L \in D^-(R)\). Assume there exists an \(k\) such that \(H^i(K) = 0\) for \(i \leq k\) and \(H^i(L) = 0\) for \(i \geq k - d + 1\). Then \(\Hom_{D(R)}(K, L) = 0\).
Proof
Let \(K^\bullet\) be a bounded above complex representing \(K\), say \(K^i = 0\) for \(i \geq n + 1\). After replacing \(K^\bullet\) by \(\tau_{\geq k + 1}K^\bullet\) we may assume \(K^i = 0\) for \(i \leq k\). Then we may use the distinguished triangle \[K^n[-n] \to K^\bullet \to \sigma_{\leq n - 1}K^\bullet\] to see it suffices to prove the lemma for \(K^n[-n]\) and \(\sigma_{\leq n - 1}K^\bullet\). By induction on \(n\), we conclude that it suffices to prove the lemma in case \(K\) is represented by the complex \(M[-m]\) for some \(R\)-module \(M\) and some \(m \geq k + 1\). Since \(R\) has global dimension \(d\) by Algebra, Lemma 00OE we see that \(M\) has a projective resolution \(0 \to P_d \to \ldots \to P_0 \to M \to 0\). Then the complex \(P^\bullet\) having \(P_i\) in degree \(m - i\) is a bounded complex of projectives representing \(M[-m]\). On the other hand, we can choose a complex \(L^\bullet\) representing \(L\) with \(L^i = 0\) for \(i \geq k - d + 1\). Hence any map of complexes \(P^\bullet \to L^\bullet\) is zero. This implies the lemma by Derived Categories, Lemma 064B.
Lemma
Let \(R\) be a Noetherian regular ring of dimension \(1 \leq d < \infty\). Let \(K \in D(R)\) be perfect and let \(k \in \mathbf{Z}\) such that \(H^i(K) = 0\) for \(i = k - d + 2, \ldots, k\) (empty condition if \(d = 1\)). Then \(K = \tau_{\leq k - d + 1}K \oplus \tau_{\geq k + 1}K\).
Proof
The vanishing of cohomology shows that we have a distinguished triangle \[\tau_{\leq k - d + 1}K \to K \to \tau_{\geq k + 1}K \to (\tau_{\leq k - d + 1}K)[1]\] By Derived Categories, Lemma 05QT it suffices to show that the third arrow is zero. Thus it suffices to show that \(\Hom_{D(R)}(\tau_{\geq k + 1}K, (\tau_{\leq k - d + 1}K)[1]) = 0\) which follows from Lemma 0FXJ.
Lemma
Let \(R\) be a Noetherian regular ring of finite dimension. Then \(R\) is a strong generator for the full subcategory \(D_{perf}(R) \subset D(R)\) of perfect objects.
Proof
We will use that an object \(K\) of \(D(R)\) is perfect if and only if \(K\) is bounded and has finite cohomology modules, see Lemma 066Z. Strong generators of triangulated categories are defined in Derived Categories, Definition 09SJ. Let \(d = \dim(R)\).
Let \(K \in D_{perf}(R)\). We will show \(K \in \langle R \rangle_{d + 1}\). By Algebra, Lemma 00OE every finite \(R\)-module has projective dimension \(\leq d\). We will show by induction on \(0 \leq i \leq d\) that if \(H^n(K)\) has projective dimension \(\leq i\) for all \(n \in \mathbf{Z}\), then \(K\) is in \(\langle R \rangle_{i + 1}\).
Base case \(i = 0\). In this case \(H^n(K)\) is a finite \(R\)-module of projective dimension \(0\). In other words, each cohomology is a projective \(R\)-module. Thus \(\Ext^i_R(H^n(K), H^m(K)) = 0\) for all \(i > 0\) and \(m, n \in \mathbf{Z}\). By Derived Categories, Lemma 0GM4 we find that \(K\) is isomorphic to the direct sum of the shifts of its cohomology modules. Since each cohomology module is a finite projective \(R\)-module, it is a direct summand of a direct sum of copies of \(R\). Hence by definition we see that \(K\) is contained in \(\langle R \rangle_1\).
Induction step. Assume the claim holds for \(i < d\) and let \(K \in D_{perf}(R)\) have the property that \(H^n(K)\) has projective dimension \(\leq i + 1\) for all \(n \in \mathbf{Z}\). Choose \(a \leq b\) such that \(H^n(K)\) is zero for \(n \not \in [a, b]\). For each \(n \in [a, b]\) choose a surjection \(F^n \to H^n(K)\) where \(F^n\) is a finite free \(R\)-module. Since \(F^n\) is projective, we can lift \(F^n \to H^n(K)\) to a map \(F^n[-n] \to K\) in \(D(R)\) (small detail omitted). Thus we obtain a morphism \(\bigoplus_{a \leq n \leq b} F^n[-n] \to K\) which is surjective on cohomology modules. Choose a distinguished triangle \[K' \to \bigoplus\nolimits_{a \leq n \leq b} F^n[-n] \to K \to K'[1]\] in \(D(R)\). Of course, the object \(K'\) is bounded and has finite cohomology modules. The long exact sequence of cohomology breaks into short exact sequences \[0 \to H^n(K') \to F^n \to H^n(K) \to 0\] by the choices we made. By Algebra, Lemma 065S we see that the projective dimension of \(H^n(K')\) is \(\leq \max(0, i)\). Thus \(K' \in \langle R \rangle_{i + 1}\). By definition this means that \(K\) is in \(\langle R \rangle_{i + 1 + 1}\) as desired.
Proposition
Let \(R\) be a Noetherian ring. The following are equivalent
\(R\) is regular of finite dimension,
\(D_{perf}(R)\) has a strong generator, and
\(R\) is a strong generator for \(D_{perf}(R)\).
Proof
This is a formal consequence of Lemmas 0ATI, 0FXI, and 0FXL as well as Derived Categories, Lemma 0FXA.
Relatively finitely presented modules
Let \(R\) be a ring. Let \(A \to B\) be a finite map of finite type \(R\)-algebras. Let \(M\) be a finite \(B\)-module. In this case it is not true that \[M\text{ of finite presentation over }B \Leftrightarrow M\text{ of finite presentation over }A\] A counter example is \(R = k[x_1, x_2, x_3, \ldots]\), \(A = R\), \(B = R/(x_i)\), and \(M = B\). To “fix” this we introduce a relative notion of finite presentation.
Lemma
Let \(R \to A\) be a ring map of finite type. Let \(M\) be an \(A\)-module. The following are equivalent
for some presentation \(\alpha : R[x_1, \ldots, x_n] \to A\) the module \(M\) is a finitely presented \(R[x_1, \ldots, x_n]\)-module,
for all presentations \(\alpha : R[x_1, \ldots, x_n] \to A\) the module \(M\) is a finitely presented \(R[x_1, \ldots, x_n]\)-module, and
for any surjection \(A' \to A\) where \(A'\) is a finitely presented \(R\)-algebra, the module \(M\) is finitely presented as \(A'\)-module.
In this case \(M\) is a finitely presented \(A\)-module.
Proof
If \(\alpha : R[x_1, \ldots, x_n] \to A\) and \(\beta : R[y_1, \ldots, y_m] \to A\) are presentations. Choose \(f_j \in R[x_1, \ldots, x_n]\) with \(\alpha(f_j) = \beta(y_j)\) and \(g_i \in R[y_1, \ldots, y_m]\) with \(\beta(g_i) = \alpha(x_i)\). Then we get a commutative diagram \[\xymatrix{ R[x_1, \ldots, x_n, y_1, \ldots, y_m] \ar[d]^{x_i \mapsto g_i} \ar[rr]_-{y_j \mapsto f_j} & & R[x_1, \ldots, x_n] \ar[d] \\ R[y_1, \ldots, y_m] \ar[rr] & & A }\] Hence the equivalence of (1) and (2) follows by applying Algebra, Lemmas 0561 and 0564. The equivalence of (2) and (3) follows by choosing a presentation \(A' = R[x_1, \ldots, x_n]/(f_1, \ldots, f_m)\) and using Algebra, Lemma 0564 to show that \(M\) is finitely presented as \(A'\)-module if and only if \(M\) is finitely presented as a \(R[x_1, \ldots, x_n]\)-module.
Definition
Let \(R \to A\) be a finite type ring map. Let \(M\) be an \(A\)-module. We say \(M\) is an \(A\)-module finitely presented relative to \(R\) if the equivalent conditions of Lemma 05GY hold.
Note that if \(R \to A\) is of finite presentation, then \(M\) is an \(A\)-module finitely presented relative to \(R\) if and only if \(M\) is a finitely presented \(A\)-module. It is equally clear that \(A\) as an \(A\)-module is finitely presented relative to \(R\) if and only if \(A\) is of finite presentation over \(R\). If \(R\) is Noetherian the notion is uninteresting. Now we can formulate the result we were looking for.
Lemma
Let \(R\) be a ring. Let \(A \to B\) be a finite map of finite type \(R\)-algebras. Let \(M\) be a \(B\)-module. Then \(M\) is an \(A\)-module finitely presented relative to \(R\) if and only if \(M\) is a \(B\)-module finitely presented relative to \(R\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Choose \(y_1, \ldots, y_m \in B\) which generate \(B\) over \(A\). As \(A \to B\) is finite each \(y_i\) satisfies a monic equation with coefficients in \(A\). Hence we can find monic polynomials \(P_j(T) \in R[x_1, \ldots, x_n][T]\) such that \(P_j(y_j) = 0\) in \(B\). Then we get a commutative diagram \[\xymatrix{ R[x_1, \ldots, x_n] \ar[d] \ar[r] & R[x_1, \ldots, x_n, y_1, \ldots, y_m]/(P_j(y_j)) \ar[d] \\ A \ar[r] & B }\] Since the top arrow is a finite and finitely presented ring map we conclude by Algebra, Lemma 0564 and the definition.
With this result in hand we see that the relative notion makes sense and behaves well with regards to finite maps of rings of finite type over \(R\). It is also stable under localization, stable under base change, and "glues" well.
Lemma
Let \(R\) be a ring, \(f \in R\) an element, \(R_f \to A\) is a finite type ring map, \(g \in A\), and \(M\) an \(A\)-module. If \(M\) of finite presentation relative to \(R_f\), then \(M_g\) is an \(A_g\)-module of finite presentation relative to \(R\).
Proof
Choose a presentation \(R_f[x_1, \ldots, x_n] \to A\). We write \(R_f = R[x_0]/(fx_0 - 1)\). Consider the presentation \(R[x_0, x_1, \ldots, x_n, x_{n + 1}] \to A_g\) which extends the given map, maps \(x_0\) to the image of \(1/f\), and maps \(x_{n + 1}\) to \(1/g\). Choose \(g' \in R[x_0, x_1, \ldots, x_n]\) which maps to \(g\) (this is possible). Suppose that \[R_f[x_1, \ldots, x_n]^{\oplus s} \to R_f[x_1, \ldots, x_n]^{\oplus t} \to M \to 0\] is a presentation of \(M\) given by a matrix \((h_{ij})\). Pick \(h'_{ij} \in R[x_0, x_1, \ldots, x_n]\) which map to \(h_{ij}\). Then \[R[x_0, x_1, \ldots, x_n, x_{n + 1}]^{\oplus s + 2t} \to R[x_0, x_1, \ldots, x_n, x_{n + 1}]^{\oplus t} \to M_g \to 0\] is a presentation of \(M_g\). Here the \(t \times (s + 2t)\) matrix defining the map has a first \(t \times s\) block consisting of the matrix \(h'_{ij}\), a second \(t \times t\) block which is \((x_0f - 1)I_t\), and a third block which is \((x_{n + 1}g' - 1)I_t\).
Lemma
Let \(R \to A\) be a finite type ring map. Let \(M\) be an \(A\)-module finitely presented relative to \(R\). For any ring map \(R \to R'\) the \(A \otimes_R R'\)-module \[M \otimes_A (A \otimes_R R') = M \otimes_R R'\] is finitely presented relative to \(R'\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Choose a presentation \[R[x_1, \ldots, x_n]^{\oplus s} \to R[x_1, \ldots, x_n]^{\oplus t} \to M \to 0\] Then \[R'[x_1, \ldots, x_n]^{\oplus s} \to R'[x_1, \ldots, x_n]^{\oplus t} \to M \otimes_R R' \to 0\] is a presentation of the base change and we win.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(M\) be an \(A\)-module finitely presented relative to \(R\). Let \(A \to A'\) be a ring map of finite presentation. The \(A'\)-module \(M \otimes_A A'\) is finitely presented relative to \(R\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Choose a presentation \(A' = A[y_1, \ldots, y_m]/(g_1, \ldots, g_l)\). Pick \(g'_i \in R[x_1, \ldots, x_n, y_1, \ldots, y_m]\) mapping to \(g_i\). Say \[R[x_1, \ldots, x_n]^{\oplus s} \to R[x_1, \ldots, x_n]^{\oplus t} \to M \to 0\] is a presentation of \(M\) given by a matrix \((h_{ij})\). Then \[R[x_1, \ldots, x_n, y_1, \ldots, y_m]^{\oplus s + tl} \to R[x_1, \ldots, x_n, y_1, \ldots, y_m]^{\oplus t} \to M \otimes_A A' \to 0\] is a presentation of \(M \otimes_A A'\). Here the \(t \times (s + lt)\) matrix defining the map has a first \(t \times s\) block consisting of the matrix \(h_{ij}\), followed by \(l\) blocks of size \(t \times t\) which are \(g'_iI_t\).
Lemma
Let \(R \to A \to B\) be finite type ring maps. Let \(M\) be a \(B\)-module. If \(M\) is finitely presented relative to \(A\) and \(A\) is of finite presentation over \(R\), then \(M\) is finitely presented relative to \(R\).
Proof
Choose a surjection \(A[x_1, \ldots, x_n] \to B\). Choose a presentation \[A[x_1, \ldots, x_n]^{\oplus s} \to A[x_1, \ldots, x_n]^{\oplus t} \to M \to 0\] given by a matrix \((h_{ij})\). Choose a presentation \[A = R[y_1, \ldots, y_m]/(g_1, \ldots, g_u).\] Choose \(h'_{ij} \in R[y_1, \ldots, y_m, x_1, \ldots, x_n]\) mapping to \(h_{ij}\). Then we obtain the presentation \[R[y_1, \ldots, y_m, x_1, \ldots, x_n]^{\oplus s + tu} \to R[y_1, \ldots, y_m, x_1, \ldots, x_n]^{\oplus t} \to M \to 0\] where the \(t \times (s + tu)\)-matrix is given by a first \(t \times s\) block consisting of \(h'_{ij}\) followed by \(u\) blocks of size \(t \times t\) given by \(g_iI_t\), \(i = 1, \ldots, u\).
Lemma
Let \(R \to A\) be a finite type ring map. Let \(M\) be an \(A\)-module. Let \(f_1, \ldots, f_r \in A\) generate the unit ideal. The following are equivalent
each \(M_{f_i}\) is finitely presented relative to \(R\), and
\(M\) is finitely presented relative to \(R\).
Proof
The implication (2) \(\Rightarrow\) (1) is in Lemma 065A. Assume (1). Write \(1 = \sum f_ig_i\) in \(A\). Choose a surjection \(R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r] \to A\). such that \(y_i\) maps to \(f_i\) and \(z_i\) maps to \(g_i\). Then we see that there exists a surjection \[P = R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r]/(\sum y_iz_i - 1) \longrightarrow A.\] By Lemma 05GY we see that \(M_{f_i}\) is a finitely presented \(A_{f_i}\)-module, hence by Algebra, Lemma 00EO we see that \(M\) is a finitely presented \(A\)-module. Hence \(M\) is a finite \(P\)-module (with \(P\) as above). Choose a surjection \(P^{\oplus t} \to M\). We have to show that the kernel \(K\) of this map is a finite \(P\)-module. Since \(P_{y_i}\) surjects onto \(A_{f_i}\) we see by Lemma 05GY and Algebra, Lemma 0519 that the localization \(K_{y_i}\) is a finitely generated \(P_{y_i}\)-module. Choose elements \(k_{i, j} \in K\), \(i = 1, \ldots, r\), \(j = 1, \ldots, s_i\) such that the images of \(k_{i, j}\) in \(K_{y_i}\) generate. Set \(K' \subset K\) equal to the \(P\)-module generated by the elements \(k_{i, j}\). Then \(K/K'\) is a module whose localization at \(y_i\) is zero for all \(i\). Since \((y_1, \ldots, y_r) = P\) we see that \(K/K' = 0\) as desired.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(0 \to M' \to M \to M'' \to 0\) be a short exact sequence of \(A\)-modules.
If \(M', M''\) are finitely presented relative to \(R\), then so is \(M\).
If \(M'\) is a finite type \(A\)-module and \(M\) is finitely presented relative to \(R\), then \(M''\) is finitely presented relative to \(R\).
Proof
Follows immediately from Algebra, Lemma 0519.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(M, M'\) be \(A\)-modules. If \(M \oplus M'\) is finitely presented relative to \(R\), then so are \(M\) and \(M'\).
Proof
Omitted.
Relatively pseudo-coherent modules
This section is the analogue of Section 0659 for pseudo-coherence.
Lemma
Let \(R\) be a ring. Let \(K^\bullet\) be a complex of \(R\)-modules. Consider the \(R\)-algebra map \(R[x] \to R\) which maps \(x\) to zero. Then \[K^\bullet \otimes_{R[x]}^{\mathbf{L}} R \cong K^\bullet \oplus K^\bullet[1]\] in \(D(R)\).
Proof
Choose a K-flat resolution \(P^\bullet \to K^\bullet\) over \(R\) such that \(P^n\) is a flat \(R\)-module for all \(n\), see Lemma 06Y4. Then \(P^\bullet \otimes_R R[x]\) is a K-flat complex of \(R[x]\)-modules whose terms are flat \(R[x]\)-modules, see Lemma 06Y1 and Algebra, Lemma 00HI. In particular \(x : P^n \otimes_R R[x] \to P^n \otimes_R R[x]\) is injective with cokernel isomorphic to \(P^n\). Thus \[P^\bullet \otimes_R R[x] \xrightarrow{x} P^\bullet \otimes_R R[x]\] is a double complex of \(R[x]\)-modules whose associated total complex is quasi-isomorphic to \(P^\bullet\) and hence \(K^\bullet\). Moreover, this associated total complex is a K-flat complex of \(R[x]\)-modules for example by Lemma 0795 or by Lemma 06Y2. Hence \[\begin{align*} K^\bullet \otimes_{R[x]}^{\mathbf{L}} R & \cong \text{Tot}(P^\bullet \otimes_R R[x] \xrightarrow{x} P^\bullet \otimes_R R[x]) \otimes_{R[x]} R = \text{Tot}(P^\bullet \xrightarrow{0} P^\bullet) \\ & = P^\bullet \oplus P^\bullet[1] \cong K^\bullet \oplus K^\bullet[1] \end{align*}\] as desired.
Lemma
Let \(R\) be a ring and \(K^\bullet\) a complex of \(R\)-modules. Let \(m \in \mathbf{Z}\). Consider the \(R\)-algebra map \(R[x] \to R\) which maps \(x\) to zero. Then \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules if and only if \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R[x]\)-modules.
Proof
This is a special case of Lemma 064Z. We also prove it in another way as follows.
Note that \(0 \to R[x] \to R[x] \to R \to 0\) is exact. Hence \(R\) is pseudo-coherent as an \(R[x]\)-module. Thus one implication of the lemma follows from Lemma 064Z. To prove the other implication, assume that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R[x]\)-modules. By Lemma 0650 we see that \(K^\bullet \otimes^{\mathbf{L}}_{R[x]} R\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules. By Lemma 065F we see that \(K^\bullet \oplus K^\bullet[1]\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules. Finally, we conclude that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R\)-modules from Lemma 064X.
Lemma
Let \(R \to A\) be a ring map of finite type. Let \(K^\bullet\) be a complex of \(A\)-modules. Let \(m \in \mathbf{Z}\). The following are equivalent
for some presentation \(\alpha : R[x_1, \ldots, x_n] \to A\) the complex \(K^\bullet\) is an \(m\)-pseudo-coherent complex of \(R[x_1, \ldots, x_n]\)-modules,
for all presentations \(\alpha : R[x_1, \ldots, x_n] \to A\) the complex \(K^\bullet\) is an \(m\)-pseudo-coherent complex of \(R[x_1, \ldots, x_n]\)-modules.
In particular the same equivalence holds for pseudo-coherence.
Proof
If \(\alpha : R[x_1, \ldots, x_n] \to A\) and \(\beta : R[y_1, \ldots, y_m] \to A\) are presentations. Choose \(f_j \in R[x_1, \ldots, x_n]\) with \(\alpha(f_j) = \beta(y_j)\) and \(g_i \in R[y_1, \ldots, y_m]\) with \(\beta(g_i) = \alpha(x_i)\). Then we get a commutative diagram \[\xymatrix{ R[x_1, \ldots, x_n, y_1, \ldots, y_m] \ar[d]^{x_i \mapsto g_i} \ar[rr]_-{y_j \mapsto f_j} & & R[x_1, \ldots, x_n] \ar[d] \\ R[y_1, \ldots, y_m] \ar[rr] & & A }\] After a change of coordinates the ring homomorphism \(R[x_1, \ldots, x_n, y_1, \ldots, y_m] \to R[x_1, \ldots, x_n]\) is isomorphic to the ring homomorphism which maps each \(y_i\) to zero. Similarly for the left vertical map in the diagram. Hence, by induction on the number of variables this lemma follows from Lemma 065G. The pseudo-coherent case follows from this and Lemma 064U.
Definition
Let \(R \to A\) be a finite type ring map. Let \(K^\bullet\) be a complex of \(A\)-modules. Let \(M\) be an \(A\)-module. Let \(m \in \mathbf{Z}\).
We say \(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\) if the equivalent conditions of Lemma 065H hold.
We say \(K^\bullet\) is pseudo-coherent relative to \(R\) if \(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\) for all \(m \in \mathbf{Z}\).
We say \(M\) is \(m\)-pseudo-coherent relative to \(R\) if \(M[0]\) is \(m\)-pseudo-coherent relative to \(R\).
We say \(M\) is pseudo-coherent relative to \(R\) if \(M[0]\) is pseudo-coherent relative to \(R\).
Part (2) means that \(K^\bullet\) is pseudo-coherent as a complex of \(R[x_1, \ldots, x_n]\)-modules for any surjection \(R[y_1, \ldots, y_m] \to A\), see Lemma 064U. This definition has the following pleasing property.
Lemma
Let \(R\) be a ring. Let \(A \to B\) be a finite map of finite type \(R\)-algebras. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(B\)-modules. Then \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\) if and only if \(K^\bullet\) seen as a complex of \(A\)-modules is \(m\)-pseudo-coherent (pseudo-coherent) relative to \(R\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Choose \(y_1, \ldots, y_m \in B\) which generate \(B\) over \(A\). As \(A \to B\) is finite each \(y_i\) satisfies a monic equation with coefficients in \(A\). Hence we can find monic polynomials \(P_j(T) \in R[x_1, \ldots, x_n][T]\) such that \(P_j(y_j) = 0\) in \(B\). Then we get a commutative diagram \[\xymatrix{ & R[x_1, \ldots, x_n, y_1, \ldots, y_m] \ar[d] \\ R[x_1, \ldots, x_n] \ar[d] \ar[r] & R[x_1, \ldots, x_n, y_1, \ldots, y_m]/(P_j(y_j)) \ar[d] \\ A \ar[r] & B }\] The top horizontal arrow and the top right vertical arrow satisfy the assumptions of Lemma 064Z. Hence \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) as a complex of \(R[x_1, \ldots, x_n]\)-modules if and only if \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) as a complex of \(R[x_1, \ldots, x_n, y_1, \ldots, y_m]\)-modules.
Lemma
Let \(R\) be a ring. Let \(R \to A\) be a finite type ring map. Let \(m \in \mathbf{Z}\). Let \((K^\bullet, L^\bullet, M^\bullet, f, g, h)\) be a distinguished triangle in \(D(A)\).
If \(K^\bullet\) is \((m + 1)\)-pseudo-coherent relative to \(R\) and \(L^\bullet\) is \(m\)-pseudo-coherent relative to \(R\) then \(M^\bullet\) is \(m\)-pseudo-coherent relative to \(R\).
If \(K^\bullet, M^\bullet\) are \(m\)-pseudo-coherent relative to \(R\), then \(L^\bullet\) is \(m\)-pseudo-coherent relative to \(R\).
If \(L^\bullet\) is \((m + 1)\)-pseudo-coherent relative to \(R\) and \(M^\bullet\) is \(m\)-pseudo-coherent relative to \(R\), then \(K^\bullet\) is \((m + 1)\)-pseudo-coherent relative to \(R\).
Moreover, if two out of three of \(K^\bullet, L^\bullet, M^\bullet\) are pseudo-coherent relative to \(R\), the so is the third.
Proof
Follows immediately from Lemma 064R and the definitions.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(M\) be an \(A\)-module. Then
\(M\) is \(0\)-pseudo-coherent relative to \(R\) if and only if \(M\) is a finite type \(A\)-module,
\(M\) is \((-1)\)-pseudo-coherent relative to \(R\) if and only if \(M\) is a finitely presented relative to \(R\),
\(M\) is \((-d)\)-pseudo-coherent relative to \(R\) if and only if for every surjection \(R[x_1, \ldots, x_n] \to A\) there exists a resolution \[R[x_1, \ldots, x_n]^{\oplus a_d} \to R[x_1, \ldots, x_n]^{\oplus a_{d - 1}} \to \ldots \to R[x_1, \ldots, x_n]^{\oplus a_0} \to M \to 0\] of length \(d\), and
\(M\) is pseudo-coherent relative to \(R\) if and only if for every presentation \(R[x_1, \ldots, x_n] \to A\) there exists an infinite resolution \[\ldots \to R[x_1, \ldots, x_n]^{\oplus a_1} \to R[x_1, \ldots, x_n]^{\oplus a_0} \to M \to 0\] by finite free \(R[x_1, \ldots, x_n]\)-modules.
Proof
Follows immediately from Lemma 064T and the definitions.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(m \in \mathbf{Z}\). Let \(K^\bullet, L^\bullet \in D(A)\). If \(K^\bullet \oplus L^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\) so are \(K^\bullet\) and \(L^\bullet\).
Proof
Immediate from Lemma 064X and the definitions.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a bounded above complex of \(A\)-modules such that \(K^i\) is \((m - i)\)-pseudo-coherent relative to \(R\) for all \(i\). Then \(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\). In particular, if \(K^\bullet\) is a bounded above complex of \(A\)-modules pseudo-coherent relative to \(R\), then \(K^\bullet\) is pseudo-coherent relative to \(R\).
Proof
Immediate from Lemma 064Y and the definitions.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(m \in \mathbf{Z}\). Let \(K^\bullet \in D^{-}(A)\) such that \(H^i(K^\bullet)\) is \((m - i)\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\) for all \(i\). Then \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\).
Proof
Immediate from Lemma 066B and the definitions.
Lemma
Let \(R\) be a ring, \(f \in R\) an element, \(R_f \to A\) is a finite type ring map, \(g \in A\), and \(K^\bullet\) a complex of \(A\)-modules. If \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R_f\), then \(K^\bullet \otimes_A A_g\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\).
Proof
First we show that \(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\). Namely, suppose \(R_f[x_1, \ldots, x_n] \to A\) is surjective. Write \(R_f = R[x_0]/(fx_0 - 1)\). Then \(R[x_0, x_1, \ldots, x_n] \to A\) is surjective, and \(R_f[x_1, \ldots, x_n]\) is pseudo-coherent as an \(R[x_0, \ldots, x_n]\)-module. Hence by Lemma 064Z we see that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R[x_0, x_1, \ldots, x_n]\)-modules.
Choose an element \(g' \in R[x_0, x_1, \ldots, x_n]\) which maps to \(g \in A\). By Lemma 0650 we see that \[\begin{align*} K^\bullet \otimes_{R[x_0, x_1, \ldots, x_n]}^{\mathbf{L}} R[x_0, x_1, \ldots, x_n, \frac{1}{g'}] & = K^\bullet \otimes_{R[x_0, x_1, \ldots, x_n]} R[x_0, x_1, \ldots, x_n, \frac{1}{g'}] \\ & = K^\bullet \otimes_A A_g \end{align*}\] is \(m\)-pseudo-coherent as a complex of \(R[x_0, x_1, \ldots, x_n, \frac{1}{g'}]\)-modules. write \[R[x_0, x_1, \ldots, x_n, \frac{1}{g'}] = R[x_0, \ldots, x_n, x_{n + 1}]/(x_{n + 1}g' - 1).\] As \(R[x_0, x_1, \ldots, x_n, \frac{1}{g'}]\) is pseudo-coherent as a \(R[x_0, \ldots, x_n, x_{n + 1}]\)-module we conclude (see Lemma 064Z) that \(K^\bullet \otimes_A A_g\) is \(m\)-pseudo-coherent as a complex of \(R[x_0, \ldots, x_n, x_{n + 1}]\)-modules as desired.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(A\)-modules which is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\). Let \(R \to R'\) be a ring map such that \(A\) and \(R'\) are Tor independent over \(R\). Set \(A' = A \otimes_R R'\). Then \(K^\bullet \otimes_A^{\mathbf{L}} A'\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R'\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Note that \[K^\bullet \otimes_A^{\mathbf{L}} A' = K^\bullet \otimes_R^{\mathbf{L}} R' = K^\bullet \otimes_{R[x_1, \ldots, x_n]}^{\mathbf{L}} R'[x_1, \ldots, x_n]\] by Lemma 0661 applied twice. Hence we win by Lemma 0650.
Lemma
Let \(R \to A \to B\) be finite type ring maps. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(A\)-modules. Assume \(B\) as a \(B\)-module is pseudo-coherent relative to \(A\). If \(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\), then \(K^\bullet \otimes_A^{\mathbf{L}} B\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\).
Proof
Choose a surjection \(A[y_1, \ldots, y_m] \to B\). Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Combined we get a surjection \(R[x_1, \ldots, x_n, y_1, \ldots y_m] \to B\). Choose a resolution \(E^\bullet \to B\) of \(B\) by a complex of finite free \(A[y_1, \ldots, y_m]\)-modules (which is possible by our assumption on the ring map \(A \to B\)). We may assume that \(K^\bullet\) is a bounded above complex of flat \(A\)-modules. Then \[\begin{align*} K^\bullet \otimes_A^{\mathbf{L}} B & = \text{Tot}(K^\bullet \otimes_A B[0]) \\ & = \text{Tot}(K^\bullet \otimes_A A[y_1, \ldots, y_m] \otimes_{A[y_1, \ldots, y_m]} B[0]) \\ & \cong \text{Tot}\left( (K^\bullet \otimes_A A[y_1, \ldots, y_m]) \otimes_{A[y_1, \ldots, y_m]} E^\bullet \right) \\ & = \text{Tot}(K^\bullet \otimes_A E^\bullet) \end{align*}\] in \(D(A[y_1, \ldots, y_m])\). The quasi-isomorphism \(\cong\) comes from an application of Lemma 064K. Thus we have to show that \(\text{Tot}(K^\bullet \otimes_A E^\bullet)\) is \(m\)-pseudo-coherent as a complex of \(R[x_1, \ldots, x_n, y_1, \ldots y_m]\)-modules. Note that \(\text{Tot}(K^\bullet \otimes_A E^\bullet)\) has a filtration by subcomplexes with successive quotients the complexes \(K^\bullet \otimes_A E^i[-i]\). Note that for \(i \ll 0\) the complexes \(K^\bullet \otimes_A E^i[-i]\) have zero cohomology in degrees \(\leq m\) and hence are \(m\)-pseudo-coherent (over any ring). Hence, applying Lemma 0674 and induction, it suffices to show that \(K^\bullet \otimes_A E^i[-i]\) is pseudo-coherent relative to \(R\) for all \(i\). Note that \(E^i = 0\) for \(i > 0\). Since also \(E^i\) is finite free this reduces to proving that \(K^\bullet \otimes_A A[y_1, \ldots, y_m]\) is \(m\)-pseudo-coherent relative to \(R\) which follows from Lemma 067A for instance.
Lemma
Let \(R \to A \to B\) be finite type ring maps. Let \(m \in \mathbf{Z}\). Let \(M\) be an \(A\)-module. Assume \(B\) is flat over \(A\) and \(B\) as a \(B\)-module is pseudo-coherent relative to \(A\). If \(M\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\), then \(M \otimes_A B\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\).
Proof
Immediate from Lemma 067B.
Lemma
Let \(R\) be a ring. Let \(A \to B\) be a map of finite type \(R\)-algebras. Let \(m \in \mathbf{Z}\). Let \(K^\bullet\) be a complex of \(B\)-modules. Assume \(A\) is pseudo-coherent relative to \(R\). Then the following are equivalent
\(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(A\), and
\(K^\bullet\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\).
Proof
Choose a surjection \(R[x_1, \ldots, x_n] \to A\). Choose a surjection \(A[y_1, \ldots, y_m] \to B\). Then we get a surjection \[R[x_1, \ldots, x_n, y_1, \ldots, y_m] \to A[y_1, \ldots, y_m]\] which is a flat base change of \(R[x_1, \ldots, x_n] \to A\). By assumption \(A\) is a pseudo-coherent module over \(R[x_1, \ldots, x_n]\) hence by Lemma 066C we see that \(A[y_1, \ldots, y_m]\) is pseudo-coherent over \(R[x_1, \ldots, x_n, y_1, \ldots, y_m]\). Thus the lemma follows from Lemma 064Z and the definitions.
Lemma
Let \(R \to A\) be a finite type ring map. Let \(K^\bullet\) be a complex of \(A\)-modules. Let \(m \in \mathbf{Z}\). Let \(f_1, \ldots, f_r \in A\) generate the unit ideal. The following are equivalent
each \(K^\bullet \otimes_A A_{f_i}\) is \(m\)-pseudo-coherent relative to \(R\), and
\(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\).
The same equivalence holds for pseudo-coherence relative to \(R\).
Proof
The implication (2) \(\Rightarrow\) (1) is in Lemma 0679. Assume (1). Write \(1 = \sum f_ig_i\) in \(A\). Choose a surjection \(R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r] \to A\). such that \(y_i\) maps to \(f_i\) and \(z_i\) maps to \(g_i\). Then we see that there exists a surjection \[P = R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r]/(\sum y_iz_i - 1) \longrightarrow A.\] Note that \(P\) is pseudo-coherent as an \(R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r]\)-module and that \(P[1/y_i]\) is pseudo-coherent as an \(R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r, 1/y_i]\)-module. Hence by Lemma 064Z we see that \(K^\bullet \otimes_A A_{f_i}\) is an \(m\)-pseudo-coherent complex of \(P[1/y_i]\)-modules for each \(i\). Thus by Lemma 066D we see that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(P\)-modules, and Lemma 064Z shows that \(K^\bullet\) is \(m\)-pseudo-coherent as a complex of \(R[x_1, \ldots, x_n, y_1, \ldots, y_r, z_1, \ldots, z_r]\)-modules.
Lemma
Let \(R\) be a Noetherian ring. Let \(R \to A\) be a finite type ring map. Then
A complex of \(A\)-modules \(K^\bullet\) is \(m\)-pseudo-coherent relative to \(R\) if and only if \(K^\bullet \in D^{-}(A)\) and \(H^i(K^\bullet)\) is a finite \(A\)-module for \(i \geq m\).
A complex of \(A\)-modules \(K^\bullet\) is pseudo-coherent relative to \(R\) if and only if \(K^\bullet \in D^{-}(A)\) and \(H^i(K^\bullet)\) is a finite \(A\)-module for all \(i\).
An \(A\)-module is pseudo-coherent relative to \(R\) if and only if it is finite.
Proof
Immediate consequence of Lemma 066E and the definitions.
Pseudo-coherent and perfect ring maps
We can define these types of ring maps as follows.
Definition
Let \(A \to B\) be a ring map.
We say \(A \to B\) is a pseudo-coherent ring map if it is of finite type and \(B\), as a \(B\)-module, is pseudo-coherent relative to \(A\).
We say \(A \to B\) is a perfect ring map if it is a pseudo-coherent ring map such that \(B\) as an \(A\)-module has finite tor dimension.
This terminology may be nonstandard. Using Lemma 0675 we see that \(A \to B\) is pseudo-coherent if and only if \(B = A[x_1, \ldots, x_n]/I\) and \(B\) as an \(A[x_1, \ldots, x_n]\)-module has a resolution by finite free \(A[x_1, \ldots, x_n]\)-modules. The motivation for the definition of a perfect ring map is Lemma 0658. The following lemmas gives a more useful and intuitive characterization of a perfect ring map.
Lemma
A ring map \(A \to B\) is perfect if and only if \(B = A[x_1, \ldots, x_n]/I\) and \(B\) as an \(A[x_1, \ldots, x_n]\)-module has a finite resolution by finite projective \(A[x_1, \ldots, x_n]\)-modules.
Proof
If \(A \to B\) is perfect, then \(B = A[x_1, \ldots, x_n]/I\) and \(B\) is pseudo-coherent as an \(A[x_1, \ldots, x_n]\)-module and has finite tor dimension as an \(A\)-module. Hence Lemma 068X implies that \(B\) is perfect as a \(A[x_1, \ldots, x_n]\)-module, i.e., it has a finite resolution by finite projective \(A[x_1, \ldots, x_n]\)-modules (Lemma 066Q). Conversely, if \(B = A[x_1, \ldots, x_n]/I\) and \(B\) as an \(A[x_1, \ldots, x_n]\)-module has a finite resolution by finite projective \(A[x_1, \ldots, x_n]\)-modules then \(B\) is pseudo-coherent as an \(A[x_1, \ldots, x_n]\)-module, hence \(A \to B\) is pseudo-coherent. Moreover, the given resolution over \(A[x_1, \ldots, x_n]\) is a finite resolution by flat \(A\)-modules and hence \(B\) has finite tor dimension as an \(A\)-module.
Lots of the results of the preceding sections can be reformulated in terms of this terminology. We also refer to More on Morphisms, Sections 067X and 0685 for the corresponding discussion concerning morphisms of schemes.
Lemma
A finite type ring map of Noetherian rings is pseudo-coherent.
Proof
See Lemma 067F.
Lemma
A ring map which is flat and of finite presentation is perfect.
Proof
Let \(A \to B\) be a ring map which is flat and of finite presentation. It is clear that \(B\) has finite tor dimension. By Algebra, Lemma 02JO there exists a finite type \(\mathbf{Z}\)-algebra \(A_0 \subset A\) and a flat finite type ring map \(A_0 \to B_0\) such that \(B = B_0 \otimes_{A_0} A\). By Lemma 067F we see that \(A_0 \to B_0\) is pseudo-coherent. As \(A_0 \to B_0\) is flat we see that \(B_0\) and \(A\) are tor independent over \(A_0\), hence we may use Lemma 067A to conclude that \(A \to B\) is pseudo-coherent.
Lemma
Let \(A \to B\) be a finite type ring map with \(A\) a regular ring of finite dimension. Then \(A \to B\) is perfect.
Proof
By Algebra, Lemma 00OE the assumption on \(A\) means that \(A\) has finite global dimension. Hence every module has finite tor dimension, see Lemma 066P, in particular \(B\) does. By Lemma 067I the map is pseudo-coherent.
Lemma
A local complete intersection homomorphism is perfect.
Proof
Let \(A \to B\) be a local complete intersection homomorphism. By Definition 07D0 this means that \(B = A[x_1, \ldots, x_n]/I\) where \(I\) is a Koszul ideal in \(A[x_1, \ldots, x_n]\). By Lemmas 068Y and 066Q it suffices to show that \(I\) is a perfect module over \(A[x_1, \ldots, x_n]\). By Lemma 066Y this is a local question. Hence we may assume that \(I\) is generated by a Koszul-regular sequence (by Definition 07CV). Of course this means that \(I\) has a finite free resolution and we win.
Lemma
Let \(R \to A\) be a pseudo-coherent ring map. Let \(K \in D(A)\). The following are equivalent
\(K\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) relative to \(R\), and
\(K\) is \(m\)-pseudo-coherent (resp. pseudo-coherent) in \(D(A)\).
Proof
Reformulation of a special case of Lemma 067D.
Lemma
Let \(R \to B \to A\) be ring maps with \(\varphi : B \to A\) surjective and \(R \to B\) and \(R \to A\) flat and of finite presentation. For \(K \in D(A)\) denote \(\varphi_*K \in D(B)\) the restriction. The following are equivalent
\(K\) is pseudo-coherent,
\(K\) is pseudo-coherent relative to \(R\),
\(K\) is pseudo-coherent relative to \(A\),
\(\varphi_*K\) is pseudo-coherent,
\(\varphi_*K\) is pseudo-coherent relative to \(R\).
Similar holds for \(m\)-pseudo-coherence.
Proof
Observe that \(R \to A\) and \(R \to B\) are perfect ring maps (Lemma 067J) hence a fortiori pseudo-coherent ring maps. Thus (1) \(\Leftrightarrow\) (2) and (4) \(\Leftrightarrow\) (5) by Lemma 0DHQ.
Using that \(A\) is pseudo-coherent relative to \(R\) we use Lemma 067D to see that (2) \(\Leftrightarrow\) (3). However, since \(B \to A\) is surjective, we see directly from Definition 065I that (3) is equivalent with (4).
Relatively perfect modules
This section is the analogue of Section 065E for perfect objects of the derived category. we only define this notion in a limited generality as we are not sure what the correct definition is in general. See Derived Categories of Schemes, Remark 0DI9 for a discussion.
Definition
Let \(R \to A\) be a flat ring map of finite presentation. An object \(K\) of \(D(A)\) is \(R\)-perfect or perfect relative to \(R\) if \(K\) is pseudo-coherent (Definition 064Q) and has finite tor dimension over \(R\) (Definition 0652).
By Lemma 0E1T it would have been the same thing to ask \(K\) to be pseudo-coherent relative to \(R\). Here are some obligatory lemmas.
Lemma
Let \(R \to A\) be a flat ring map of finite presentation. The \(R\)-perfect objects of \(D(A)\) form a saturated10 triangulated strictly full subcategory.
Proof
Lemma
Let \(R \to A\) be a flat ring map of finite presentation. A perfect object of \(D(A)\) is \(R\)-perfect. If \(K, M \in D(A)\) then \(K \otimes_A^\mathbf{L} M\) is \(R\)-perfect if \(K\) is perfect and \(M\) is \(R\)-perfect.
Proof
The first statement follows from the second by taking \(M = A\). The second statement follows from Lemmas 0658, 0B66, and 0DJE.
Lemma
Let \(R \to A\) be a flat ring map of finite presentation. Let \(K \in D(A)\). The following are equivalent
\(K\) is \(R\)-perfect, and
\(K\) is isomorphic to a finite complex of \(R\)-flat, finitely presented \(A\)-modules.
Proof
To prove (2) implies (1) it suffices by Lemma 0DHT to show that an \(R\)-flat, finitely presented \(A\)-module \(M\) defines an \(R\)-perfect object of \(D(A)\). Since \(M\) has finite tor dimension over \(R\), it suffices to show that \(M\) is pseudo-coherent. By Algebra, Lemma 02JO there exists a finite type \(\mathbf{Z}\)-algebra \(R_0 \subset R\) and a flat finite type ring map \(R_0 \to A_0\) and a finite \(A_0\)-module \(M_0\) flat over \(R_0\) such that \(A = A_0 \otimes_{R_0} R\) and \(M = M_0 \otimes_{R_0} R\). By Lemma 066E we see that \(M_0\) is pseudo-coherent \(A_0\)-module. Choose a resolution \(P_0^\bullet \to M_0\) by finite free \(A_0\)-modules \(P_0^n\). Since \(A_0\) is flat over \(R_0\), this is a flat resolution. Since \(M_0\) is flat over \(R_0\) we find that \(P^\bullet = P_0^\bullet \otimes_{R_0} R\) still resolves \(M = M_0 \otimes_{R_0} R\). (You can use Lemma 0661 to see this.) Hence \(P^\bullet\) is a finite free resolution of \(M\) over \(A\) and we conclude that \(M\) is pseudo-coherent.
Assume (1). We can represent \(K\) by a bounded above complex \(P^\bullet\) of finite free \(A\)-modules. Assume that \(K\) viewed as an object of \(D(R)\) has tor amplitude in \([a, b]\). By Lemma 0653 we see that \(\tau_{\geq a}P^\bullet\) is a complex of \(R\)-flat, finitely presented \(A\)-modules representing \(K\).
Lemma
Let \(R \to A\) be a flat ring map of finite presentation. Let \(R \to R'\) be a ring map and set \(A' = A \otimes_R R'\). If \(K \in D(A)\) is \(R\)-perfect, then \(K \otimes_A^\mathbf{L} A'\) is \(R'\)-perfect.
Proof
By Lemma 0650 we see that \(K \otimes_A^\mathbf{L} A'\) is pseudo-coherent. By Lemma 0661 we see that \(K \otimes_A^\mathbf{L} A'\) is equal to \(K \otimes_R^\mathbf{L} R'\) in \(D(R')\). Then we can apply Lemma 066L to see that \(K \otimes_R^\mathbf{L} R'\) in \(D(R')\) has finite tor dimension.
Lemma
Let \(R \to A\) be a flat ring map. Let \(K, L \in D(A)\) with \(K\) pseudo-coherent and \(L\) finite tor dimension over \(R\). We may choose
a bounded above complex \(P^\bullet\) of finite free \(A\)-modules representing \(K\), and
a bounded complex of \(R\)-flat \(A\)-modules \(F^\bullet\) representing \(L\).
Given these choices we have
\(E^\bullet = \Hom^\bullet(P^\bullet, F^\bullet)\) is a bounded below complex of \(R\)-flat \(A\)-modules representing \(R\Hom_A(K, L)\),
for any ring map \(R \to R'\) with \(A' = A \otimes_R R'\) the complex \(E^\bullet \otimes_R R'\) represents \(R\Hom_{A'}(K \otimes_A^\mathbf{L} A', L \otimes_A^\mathbf{L} A')\).
If in addition \(R \to A\) is of finite presentation and \(L\) is \(R\)-perfect, then we may choose \(F^p\) to be finitely presented \(A\)-modules and consequently \(E^n\) will be finitely presented \(A\)-modules as well.
Proof
The existence of \(P^\bullet\) is the definition of a pseudo-coherent complex. We first represent \(L\) by a bounded above complex \(F^\bullet\) of free \(A\)-modules (this is possible because bounded tor dimension in particular implies bounded). Next, say \(L\) viewed as an object of \(D(R)\) has tor amplitude in \([a, b]\). Then, after replacing \(F^\bullet\) by \(\tau_{\geq a}F^\bullet\), we get a complex as in (2). This follows from Lemma 0653.
Proof of (a). Since \(F^\bullet\) is bounded an since \(P^\bullet\) is bounded above, we see that \(E^n = 0\) for \(n \ll 0\) and that \(E^n\) is a finite (!) direct sum \[E^n = \bigoplus\nolimits_{p + q = n} \Hom_A(P^{-q}, F^p)\] and since \(P^{-q}\) is finite free, this is indeed an \(R\)-flat \(A\)-module. The fact that \(E^\bullet\) represents \(R\Hom_A(K, L)\) follows from Lemma 0A66.
Proof of (b). Let \(R \to R'\) be a ring map and \(A' = A \otimes_R R'\). By Lemma 0661 the object \(L \otimes_A^\mathbf{L} A'\) is represented by \(F^\bullet \otimes_R R'\) viewed as a complex of \(A'\)-modules (by flatness of \(F^p\) over \(R\)). Similarly for \(P^\bullet \otimes_R R'\). As above \(R\Hom_{A'}(K \otimes_A^\mathbf{L} A', L \otimes_A^\mathbf{L} A')\) is represented by \[\Hom^\bullet(P^\bullet \otimes_R R', F^\bullet \otimes_R R') = E^\bullet \otimes_R R'\] The equality holds by looking at the terms of the complex individually and using that \(\Hom_{A'}(P^{-q} \otimes_R R', F^p \otimes_R R') = \Hom_A(P^{-q}, F^p) \otimes_R R'\).
Lemma
Let \(R = \colim_{i \in I} R_i\) be a filtered colimit of rings. Let \(0 \in I\) and \(R_0 \to A_0\) be a flat ring map of finite presentation. For \(i \geq 0\) set \(A_i = R_i \otimes_{R_0} A_0\) and set \(A = R \otimes_{R_0} A_0\).
Given an \(R\)-perfect \(K\) in \(D(A)\) there exists an \(i \in I\) and an \(R_i\)-perfect \(K_i\) in \(D(A_i)\) such that \(K \cong K_i \otimes_{A_i}^\mathbf{L} A\) in \(D(A)\).
Given \(K_0, L_0 \in D(A_0)\) with \(K_0\) pseudo-coherent and \(L_0\) finite tor dimension over \(R_0\), then we have \[\Hom_{D(A)}(K_0 \otimes_{A_0}^\mathbf{L} A, L_0 \otimes_{A_0}^\mathbf{L} A) = \colim_{i \geq 0} \Hom_{D(A_i)}(K_0 \otimes_{A_0}^\mathbf{L} A_i, L_0 \otimes_{A_0}^\mathbf{L} A_i)\]
In particular, the triangulated category of \(R\)-perfect complexes over \(A\) is the colimit of the triangulated categories of \(R_i\)-perfect complexes over \(A_i\).
Proof
By Algebra, Lemma 05N7 the category of finitely presented \(A\)-modules is the colimit of the categories of finitely presented \(A_i\)-modules. Given this, Algebra, Lemma 02JO tells us that category of \(R\)-flat, finitely presented \(A\)-modules is the colimit of the categories of \(R_i\)-flat, finitely presented \(A_i\)-modules. Thus the characterization in Lemma 0DHV proves that (1) is true.
To prove (2) we choose \(P_0^\bullet\) representing \(K_0\) and \(F_0^\bullet\) representing \(L_0\) as in Lemma 0E1U. Then \(E_0^\bullet = \Hom^\bullet(P_0^\bullet, F_0^\bullet)\) satisfies \[H^0(E_0^\bullet \otimes_{R_0} R_i) = \Hom_{D(A_i)}(K_0 \otimes_{A_0}^\mathbf{L} A_i, L_0 \otimes_{A_0}^\mathbf{L} A_i)\] and \[H^0(E_0^\bullet \otimes_{R_0} R) = \Hom_{D(A)}(K_0 \otimes_{A_0}^\mathbf{L} A, L_0 \otimes_{A_0}^\mathbf{L} A)\] by the lemma. Thus the result because tensor product commutes with colimits and filtered colimits are exact (Algebra, Lemma 00DB).
Lemma
Let \(R' \to A'\) be a flat ring map of finite presentation. Let \(R' \to R\) be a surjective ring map whose kernel is a nilpotent ideal. Set \(A = A' \otimes_{R'} R\). Let \(K' \in D(A')\) and set \(K = K' \otimes_{A'}^\mathbf{L} A\) in \(D(A)\). If \(K\) is \(R\)-perfect, then \(K'\) is \(R'\)-perfect.
Proof
Observe that \(A' \to A\) has nilpotent kernel and that by flatness of \(R' \to A'\) we have \(K = K' \otimes_{R'}^\mathbf{L} R\) (see Section 065Y). Hence the lemma follows by combining Lemmas 0H76 and 0H75.
Lemma
Let \(R\) be a ring. Let \(A = R[x_1, \ldots, x_d]/I\) be flat and of finite presentation over \(R\). Let \(\mathfrak q \subset A\) be a prime ideal lying over \(\mathfrak p \subset R\). Let \(K \in D(A)\) be pseudo-coherent. Let \(a, b \in \mathbf{Z}\). If \(H^i(K_\mathfrak q \otimes_{R_\mathfrak p}^\mathbf{L} \kappa(\mathfrak p))\) is nonzero only for \(i \in [a, b]\), then \(K_\mathfrak q\) has tor amplitude in \([a - d, b]\) over \(R\).
Proof
By Lemma 0E1T \(K\) is pseudo-coherent as a complex of \(R[x_1, \ldots, x_d]\)-modules. Therefore we may assume \(A = R[x_1, \ldots, x_d]\). Applying Lemma 09PC to \(R_\mathfrak p \to A_\mathfrak q\) and the complex \(K_\mathfrak q\) using our assumption, we find that \(K_\mathfrak q\) is perfect in \(D(A_\mathfrak q)\) with tor amplitude in \([a - d, b]\). Since \(R_\mathfrak p \to A_\mathfrak q\) is flat, we conclude by Lemma 066J.
Lemma
Let \(R \to A\) be a ring map which is flat and of finite presentation. Let \(K \in D(A)\) be pseudo-coherent. The following are equivalent
\(K\) is \(R\)-perfect, and
\(K\) is bounded below and for every prime ideal \(\mathfrak p \subset R\) the object \(K \otimes_R^\mathbf{L} \kappa(\mathfrak p)\) is bounded below.
Proof
Observe that (1) implies (2) as an \(R\)-perfect complex has bounded tor dimension as a complex of \(R\)-modules by definition. Let us prove the other implication.
Write \(A = R[x_1, \ldots, x_d]/I\). Denote \(L\) in \(D(R[x_1, \ldots, x_d])\) the restriction of \(K\). By Lemma 0E1T we see that \(L\) is pseudo-coherent. Since \(L\) and \(K\) have the same image in \(D(R)\) we see that \(L\) is \(R\)-perfect if and only if \(K\) is \(R\)-perfect. Also \(L \otimes_R^\mathbf{L} \kappa(\mathfrak p)\) and \(K \otimes_R^\mathbf{L} \kappa(\mathfrak p)\) are the same objects of \(D(\kappa(\mathfrak p))\). This reduces us to the case \(A = R[x_1, \ldots, x_d]\).
Say \(A = R[x_1, \ldots, x_d]\) and \(K\) satisfies (2). Let \(\mathfrak q \subset A\) be a prime lying over a prime \(\mathfrak p \subset R\). By Lemma 09PC applied to \(R_\mathfrak p \to A_\mathfrak q\) and the complex \(K_\mathfrak q\) using our assumption, we find that \(K_\mathfrak q\) is perfect in \(D(A_\mathfrak q)\). Since \(K\) is bounded below, we see that \(K\) is perfect in \(D(A)\) by Lemma 068W. This implies that \(K\) is \(R\)-perfect by Lemma 0DHU and the proof is complete.
Two term complexes
In this section we prove some results on two term complexes of modules which will help us understand conditions on the naive cotangent complex.
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\) with \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). The following are equivalent
\(H^{-1}(K) = 0\) and \(H^0(K)\) is a projective module and
\(\Ext^1_R(K, M) = 0\) for every \(R\)-module \(M\).
If \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\), then these are also equivalent to
\(\Ext^1_R(K, M) = 0\) for every finite \(R\)-module \(M\).
Proof
The equivalence of (1) and (2) follows from Lemma 0A5P. If \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\), then \(K\) is pseudo-coherent, see Lemma 066E. Thus the equivalence of (1) and (3) follows from Lemma 0G9A.
Remark
The following two statements follow from Lemma 0G9C, Algebra, Definition 00T2, and Algebra, Proposition 031J.
A ring map \(A \to B\) is smooth if and only if \(A \to B\) is of finite presentation and \(\Ext^1_B(\NL_{B/A}, N) = 0\) for every \(B\)-module \(N\).
A ring map \(A \to B\) is formally smooth if and only if \(\Ext^1_B(\NL_{B/A}, N) = 0\) for every \(B\)-module \(N\).
Lemma
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\) with \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). Then
\(K\) can be represented by a two term complex \(K^{-1} \to K^0\) with \(K^0\) a free module, and
if \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\), then \(K\) can be represented by a two term complex \(K^{-1} \to K^0\) with \(K^0\) a finite free module and \(K^{-1}\) finite.
Proof
Proof of (1). Suppose \(K\) is given by the complex of modules \(M^\bullet\). We may first replace \(M^\bullet\) by \(\tau_{\leq 0}M^\bullet\). Thus we may assume \(M^i = 0\) for \(i > 0\), Next, we may choose a free resolution \(P^\bullet \to M^\bullet\) with \(P^i = 0\) for \(i > 0\), see Derived Categories, Lemma 05T7. Finally, we can set \(K^\bullet = \tau_{\geq -1}P^\bullet\).
Proof of (2). Assume \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\). By Lemma 064U we can choose a quasi-isomorphism \(F^\bullet \to M^\bullet\) with \(F^i = 0\) for \(i > 0\) and \(F^i\) finite free. Then we can set \(K^\bullet = \tau_{\geq -1}F^\bullet\).
Maps in the derived category out of the naive cotangent complex \(\NL_{B/A}\) or \(\NL(\alpha)\) (see Algebra, Section 00S0) are easy to understand by the result of the following lemma.
Lemma
Let \(R\) be a ring. Let \(M^\bullet\) be a complex of modules over \(R\) with \(M^i = 0\) for \(i > 0\) and \(M^0\) a projective \(R\)-module. Let \(K^\bullet\) be a second complex.
Assume \(K^i = 0\) for \(i \leq -2\). Then \(\Hom_{D(R)}(M^\bullet, K^\bullet) = \Hom_{K(R)}(M^\bullet, K^\bullet)\).
Assume \(K^i = 0\) for \(i \not \in [-1, 0]\) and \(K^0\) a projective \(R\)-module. Then for a map of complexes \(a^\bullet : M^\bullet \to K^\bullet\), the following are equivalent
\(a^\bullet\) induces the zero map \(\Ext^1_R(K^\bullet, N) \to \Ext^1_R(M^\bullet, N)\) for all \(R\)-modules \(N\), and
there is a map \(h^0 : M^0 \to K^{-1}\) such that \(a^{-1} + h^0 \circ d^{-1}_M = 0\).
Assume \(K^i = 0\) for \(i \leq -3\). Let \(\alpha \in \Hom_{D(R)}(M^\bullet, K^\bullet)\). If the composition of \(\alpha\) with \(K^\bullet \to K^{-2}[2]\) comes from an \(R\)-module map \(a : M^{-2} \to K^{-2}\) with \(a \circ d_M^{-3} = 0\), then \(\alpha\) can be represented by a map of complexes \(a^\bullet : M^\bullet \to K^\bullet\) with \(a^{-2} = a\).
In (3) for any second map of complexes \((a')^\bullet : M^\bullet \to K^\bullet\) representing \(\alpha\) with \(a = (a')^{-2}\) there exist \(h^i : M^i \to K^{i - 1}\) for \(i = 0, -1\) such that \[h^{-1} \circ d_M^{-2} = 0, \quad (a')^{-1} = a^{-1} + d_K^{-2} \circ h^{-1} + h^0 \circ d_M^{-1},\quad (a')^0 = a^0 + d_K^{-1} \circ h^0\]
Proof
Set \(F^0 = M^0\). Choose a free \(R\)-module \(F^{-1}\) and a surjection \(F^{-1} \to M^{-1}\). Choose a free \(R\)-module \(F^{-2}\) and a surjection \(F^{-2} \to M^{-2} \times_{M^{-1}} F^{-1}\). Continuing in this way we obtain a quasi-isomorphism \(p^\bullet : F^\bullet \to M^\bullet\) which is termwise surjective and with \(F^i\) projective for all \(i\).
Proof of (1). By Derived Categories, Lemma 064B we have \[\Hom_{D(R)}(M^\bullet, K^\bullet) = \Hom_{K(R)}(F^\bullet, K^\bullet)\] If \(K^i = 0\) for \(i \leq -2\), then any morphism of complexes \(F^\bullet \to K^\bullet\) factors through \(p^\bullet\). Similarly, any homotopy \(\{h^i : F^i \to K^{i - 1}\}\) factors through \(p^\bullet\). Thus (1) holds.
Proof of (2). If (2)(b) holds, then \(a^\bullet\) is homotopic to a map of complexes \((a')^\bullet : M^\bullet \to K^\bullet\) which is zero in degree \(-1\). On the other hand, let \(N \to I^\bullet\) be an injective resolution. We have \[\Ext^1_R(K^\bullet, N) = \Hom_{D(R)}(K^\bullet, I^\bullet[1]) = \Hom_{K(R)}(K^\bullet, I^\bullet[1])\] by Derived Categories, Lemma 05TG. Let \(b^\bullet : K^\bullet \to I^\bullet[1]\) be a map of complexes. Since \(K^1 = 0\) the map \(b^0 : K^0 \to I^1\) maps into the kernel of \(I^1 \to I^2\) which is the image of \(I^0 \to I^1\). Since \(K^0\) is projective we can lift \(b^0\) to a map \(h : K^0 \to I^0\). Thus we see that \(b^\bullet\) is homotopic to a map of complexes \((b')^\bullet\) with \((b')^0 = 0\). Since \(K^i = 0\) for \(i \not \in [-1, 0]\) it follows that \((b')^\bullet \circ (a')^\bullet = 0\) as a map of complexes. Hence the map \(\Ext^1_R(K^\bullet, N) \to \Ext^1_R(M^\bullet, N)\) is zero. In this way we see that (2)(b) implies (2)(a). Conversely, assume (2)(a). We see that the canonical element in \(\Ext^1_R(K^\bullet, K^{-1})\) maps to zero in \(\Ext^1_R(M^\bullet, K^{-1})\). Using (1) we see immediately that we get a map \(h^0\) as in (2)(b).
Proof of (3). Choose \(b^\bullet : F^\bullet \to K^\bullet\) representing \(\alpha\). The composition of \(\alpha\) with \(K^\bullet \to K^{-2}[2]\) is represented by \(b^{-2} : F^{-2} \to K^{-2}\). As this is homotopic to \(a \circ p^{-2} : F^{-2} \to M^{-2} \to K^{-2}\), there is a map \(h : F^{-1} \to K^{-2}\) such that \(b^{-2} = a \circ p^{-2} + h \circ d_F^{-2}\). Adjusting \(b^\bullet\) by \(h\) viewed as a homotopy from \(F^\bullet\) to \(K^\bullet\), we find that \(b^{-2} = a \circ p^{-2}\). Hence \(b^{-2}\) factors through \(p^{-2}\). Since \(F^0 = M^0\) the kernel of \(p^{-2}\) surjects onto the kernel of \(p^{-1}\) (for example because the kernel of \(p^\bullet\) is an acyclic complex or by a diagram chase). Hence \(b^{-1}\) necessarily factors through \(p^{-1}\) as well and we see that (3) holds for these factorizations and \(a^0 = b^0\).
Proof of (4) is omitted. Hint: There is a homotopy between \(a^\bullet \circ p^\bullet\) and \((a')^\bullet \circ p^\bullet\) and we argue as before that this homotopy factors through \(p^\bullet\).
Let \(A \to B\) be a finitely presented ring map. Given an ideal \(I \subset B\) we can consider the condition
\(\Ext^1_B(\NL_{B/A}, N)\) is annihilated by \(I\) for all \(B\)-modules \(N\).
This condition is one possible precise mathematical formulation of the notion “the singular locus of \(A \to B\) is scheme theoretically contained in \(V(I)\)”. Please compare with Remark 0G9D and the following lemmas.
Lemma
Let \(R\) be a ring and let \(I \subset R\) be an ideal. Let \(K \in D(R)\). Assume \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). The following are equivalent
\(\Ext^1_R(K, N)\) is annihilated by \(I\) for all \(R\)-modules \(N\),
\(K\) can be represented by a complex \(K^{-1} \to K^0\) with \(K^0\) free such that for any \(a \in I\) the map \(a : K^{-1} \to K^{-1}\) factors through \(d_K^{-1} : K^{-1} \to K^0\),
whenever \(K\) is represented by a two term complex \(K^{-1} \to K^0\) with \(K^0\) projective, then for any \(a \in I\) the map \(a : K^{-1} \to K^{-1}\) factors through \(d_K^{-1} : K^{-1} \to K^0\).
If \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\), then these are also equivalent to
\(\Ext^1_R(K, N)\) is annihilated by \(I\) for every finite \(R\)-module \(N\),
\(K\) can be represented by a complex \(K^{-1} \to K^0\) with \(K^0\) finite free and \(K^{-1}\) finite such that for any \(a \in I\) the map \(a : K^{-1} \to K^{-1}\) factors through \(d_K^{-1} : K^{-1} \to K^0\).
Proof
Assume (1) and let \(K^{-1} \to K^0\) be a two term complex representing \(K\) with \(K^0\) projective. We will use the description of maps in \(D(R)\) out of \(K^\bullet\) given in Lemma 0ALN without further mention. Choosing \(N = K^{-1}\) consider the element \(\xi\) of \(\Ext^1_R(K, N)\) given by \(\text{id}_{K^{-1}} : K^{-1} \to K^{-1}\). Since is annihilated by \(a \in I\) we see that we get the dotted arrow fitting into the following commutative diagram \[\xymatrix{ K^{-1} \ar[d]_a \ar[r]_{d_K^{-1}} & K^0 \ar@{..>}[ld]^h \\ K^{-1} }\] This proves that (3) holds. Part (3) implies (2) in view of Lemma 0G9E part (1). Assume \(K^\bullet\) is as in (2) and \(N\) is an arbitrary \(R\)-module. Any element \(\xi\) of \(\Ext^1_R(K, N)\) is given as the class of a map \(\varphi : K^{-1} \to N\). Then for \(a \in I\) by assumption we may choose a map \(h\) as in the diagram above and we see that \(a\varphi = \varphi \circ a = \varphi \circ h \circ \text{d}_K^{-1}\) which proves that \(a \xi\) is zero in \(\Ext^1_R(K, N)\). Thus (1), (2), and (3) are equivalent.
Assume \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\). Part (3) implies (5) in view of Lemma 0G9E part (2). It is clear that (5) implies (2). Trivially (1) implies (4). Thus to finish the proof it suffices to show that (4) implies any of the other conditions. Let \(K^{-1} \to K^0\) be a complex representing \(K\) with \(K^0\) finite free and \(K^{-1}\) finite as in Lemma 0G9E part (2). The argument given in the proof of (2) \(\Rightarrow\) (1) shows that if \(\Ext^1_R(K, K^{-1})\) is annihilated by \(I\), then (1) holds. In this way we see that (4) implies (1) and the proof is complete.
Lemma
Let \(R\) be a ring. Let \(K\) be an object of \(D(R)\) with \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). Let \(K^{-1} \to K^0\) be a two term complex of \(R\)-modules representing \(K\) such that \(K^0\) is a flat \(R\)-module (for example projective or free). Let \(R \to R'\) be a ring map. Then the complex \(K^\bullet \otimes_R R'\) represents \(\tau_{\geq -1}(K \otimes_R^\mathbf{L} R')\).
Proof
We have a distinguished triangle \[K^0 \to K^\bullet \to K^{-1}[1] \to K^0[1]\] in \(D(R)\). This determines a map of distinguished triangles \[\xymatrix{ K^0 \otimes_R^\mathbf{L} R' \ar[d] \ar[r] & K^\bullet \otimes_R^\mathbf{L} R' \ar[r] \ar[d] & K^{-1} \otimes_R^\mathbf{L} R'[1] \ar[r] \ar[d] & K^0 \otimes_R^\mathbf{L} R'[1] \ar[d] \\ K^0 \otimes_R R' \ar[r] & K^\bullet \otimes_R R' \ar[r] & K^{-1} \otimes_R R'[1] \ar[r] & K^0 \otimes_R R'[1] }\] The left and right vertical arrows are isomorphisms as \(K^0\) is flat. Since \(K^{-1} \otimes_R^\mathbf{L} R' \to K^{-1} \otimes_R R'\) is an isomorphism on cohomology in degree \(0\) we conclude.
Lemma
Let \(I\) be an ideal of a ring \(R\). Let \(K\) be an object of \(D(R)\) with \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). Let \(R \to R'\) be a ring map. If \(K\) satisfies the equivalent conditions (1), (2), and (3) of Lemma 0G9F with respect to \((R, I)\), then \(\tau_{\geq -1}(K \otimes_R^\mathbf{L} R')\) satisfies the equivalent conditions (1), (2), and (3) of Lemma 0G9F with respect to \((R', IR')\)
Proof
We may assume \(K\) is represented by a two term complex \(K^{-1} \to K^0\) with \(K^0\) free such that for any \(a \in I\) the map \(a : K^{-1} \to K^{-1}\) is equal to \(h_a \circ d_K^{-1}\) for some map \(h_a : K^0 \to K^{-1}\). By Lemma 0G9G we see that \(\tau_{\geq -1}(K \otimes_R^\mathbf{L} R')\) is represented by \(K^\bullet \otimes_R R'\). Then of course for every \(a \in I\) we see that \(a \otimes 1 : K^{-1} \otimes_R R' \to K^{-1} \otimes_R R'\) is equal to \((h_a \otimes 1) \circ (d_K^{-1} \otimes 1)\). Since the collection of maps \(K^{-1} \otimes_R R' \to K^{-1} \otimes_R R'\) which factor through \(\text{d}_K^{-1} \otimes 1\) forms an \(R'\)-module we conclude.
Lemma
Let \(R\) be a ring. Let \(\alpha : K \to K'\) be a morphism of \(D(R)\). Assume
\(H^i(K) = H^i(K') = 0\) for \(i \not \in \{-1, 0\}\)
\(H^0(\alpha)\) is an isomorphism and \(H^{-1}(\alpha)\) is surjective.
For any \(f \in R\) if \(f : K \to K\) is \(0\), then \(f : K' \to K'\) is \(0\).
Proof
Set \(M = \Ker(H^{-1}(\alpha))\). Then \(\alpha\) fits into a distinguished triangle \[M[1] \to K \to K' \to M[2]\] Since \(K \to K' \xrightarrow{f} K'\) is zero by our assumption, we see that \(f : K' \to K'\) factors over a map \(M[2] \to K'\). However \(\Hom(M[2], K') = 0\) for example by Derived Categories, Lemma 06XS.
Lemma
Let \(I\) be an ideal of a ring \(R\). Let \(\alpha : K \to K'\) be a morphism of \(D(R)\). Assume
\(H^i(K) = H^i(K') = 0\) for \(i \not \in \{-1, 0\}\)
\(H^0(\alpha)\) is an isomorphism and \(H^{-1}(\alpha)\) is surjective.
If \(K\) satisfies the equivalent conditions (1), (2), and (3) of Lemma 0G9F, then \(K'\) does too.
Proof
Set \(M = \Ker(H^{-1}(\alpha))\). Then \(\alpha\) fits into a distinguished triangle \[M[1] \to K \to K' \to M[2]\] For any \(R\)-module \(N\) this determines an exact sequence \[\Ext^0_R(M[1], N) \to \Ext^1_R(K', N) \to \Ext^1_R(K, N)\] Since \(\Ext^0_R(M[1], N) = \Ext^{-1}_R(M, N) = 0\) we see that \(\Ext^1_R(K', N)\) is a submodule of \(\Ext^1_R(K, N)\). Hence if \(\Ext^1_R(K, N)\) is annihilated by \(I\) so is \(\Ext^1_R(K', N)\).
Lemma
Let \(R\) be ring and let \(I \subset R\) be an ideal. Let \(K \in D(R)\) with \(H^i(K) = 0\) for \(i \not \in \{-1, 0\}\). The following are equivalent
there exists a \(c \geq 0\) such that the equivalent conditions (1), (2), (3) of Lemma 0G9F hold for \(K\) and the ideal \(I^c\),
there exists a \(c \geq 0\) such that (a) \(I^c\) annihilates \(H^{-1}(K)\) and (b) \(H^0(K)\) is an \(I^c\)-projective module (see Section 0G8Z).
If \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\), then these are also equivalent to
there exists a \(c \geq 0\) such that the equivalent conditions (4), (5) of Lemma 0G9F hold for \(K\) and the ideal \(I^c\),
\(H^{-1}(K)\) is \(I\)-power torsion and there exist \(f_1, \ldots, f_s \in R\) with \(V(f_1, \ldots, f_s) \subset V(I)\) such that the localizations \(H^0(K)_{f_i}\) are projective \(R_{f_i}\)-modules,
\(H^{-1}(K)\) is \(I\)-power torsion and there exist \(f_1, \ldots, f_s \in I\) with \(V(f_1, \ldots, f_s) = V(I)\) such that the localizations \(H^0(K)_{f_i}\) are projective \(R_{f_i}\)-modules, and
\(H^{-1}(K)\) is \(I\)-power torsion and for any \(f_1, \ldots, f_s \in I\) with \(V(f_1, \ldots, f_s) = V(I)\) the localizations \(H^0(K)_{f_i}\) are projective \(R_{f_i}\)-modules.
Proof
The distinguished triangle \(H^{-1}(K)[1] \to K \to H^0(K)[0] \to H^{-1}(K)[2]\) determines an exact sequence \[0 \to \Ext^1_R(H^0(K), N) \to \Ext^1_R(K, N) \to \Hom_R(H^{-1}(K), N) \to \Ext^2_R(H^0(K), N)\] Thus (2) implies that \(I^{2c}\) annihilates \(\Ext^1_R(K, N)\) for every \(R\)-module \(N\). Assuming (1) we immediately see that \(H^0(K)\) is \(I^c\)-projective. On the other hand, we may choose an injective map \(H^{-1}(K) \to N\) for some injective \(R\)-module \(N\). Then this map is the image of an element of \(\Ext^1_R(K, N)\) by the vanishing of the \(\Ext^2\) in the sequence and we conclude \(H^{-1}(K)\) is annihilated by \(I^c\).
Assume \(R\) is Noetherian and \(H^i(K)\) is a finite \(R\)-module for \(i = -1, 0\). By Lemma 0G9F we see that (3) is equivalent to (1) and (2). Also, if (3) holds then for \(f \in I\) the multiplication by \(f\) on \(H^0(K)\) factors through a projective module, which implies that \(H^0(K)_f\) is a summand of a projective \(R_f\)-module and hence itself a projective \(R_f\)-module. Thus the equivalent conditions (1), (2), and (3) imply (6). Of course (6) implies (5) and (5) trivially implies (4).
Assume (4). Since \(H^{-1}(K)\) is a finite \(R\)-module and \(I\)-power torsion we see that \(I^{c_1}\) annihilates \(H^{-1}(K)\) for some \(c_1 \geq 0\). Choose a short exact sequence \[0 \to M \to R^{\oplus r} \to H^0(K) \to 0\] which determines an element \(\xi \in \Ext^1_R(H^0(K), M)\). For any \(f \in I\) we have \(\Ext^1_R(H^0(K), M)_f = \Ext^1_{R_f}(H^0(K)_f, M_f)\) by Lemma 087R. Hence if \(H^0(K)_f\) is projective, then a power of \(f\) annihilates \(\xi\). We conclude that \(\xi\) is annihilated by \((f_1, \ldots, f_s)^{c_2}\) for some \(c_2 \geq 0\). Since \(V(f_1, \ldots, f_s) \subset V(I)\) we have \(\sqrt{I} \subset (f_1, \ldots, f_s)\) (Algebra, Lemma 00E0). Since \(R\) is Noetherian we find \(I^{c_3} \subset (f_1, \ldots, f_s)\) for some \(c_3 \geq 0\) (Algebra, Lemma 00IM). Hence \(I^{c_2c_3}\) annihilates \(\xi\). This in turn says that \(H^0(K)\) is \(I^{c_2c_3}\)-projective (as multiplication by \(a \in I\) which annihilate \(\xi\) factor through \(R^{\oplus r}\)). Hence taking \(c = \max(c_1, c_2c_3)\) we see that (2) holds.
Lemma
Let \(R\) be a ring. Let \(K_j \in D(R)\), \(j = 1, 2, 3\) with \(H^i(K_j) = 0\) for \(i \not \in \{-1, 0\}\). Let \(\varphi : K_1 \to K_2\) and \(\psi : K_2 \to K_3\) be maps in \(D(R)\). If \(H^0(\varphi) = 0\) and \(H^{-1}(\psi) = 0\), then \(\psi \circ \varphi = 0\).
Proof
Apply Derived Categories, Lemma 08Q2 to see that \(\psi \circ \varphi\) factors through \(\tau_{\leq -2}K_2 = 0\).
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\) be given by a two term complex of the form \(R^{\oplus n} \to R^{\oplus n}\). Denote \(A \in \text{Mat}(n \times n, R)\) the matrix of the differential. Then \(\det(A) : K \to K\) is zero in \(D(R)\).
Proof
Omitted. Good exercise.
The naive cotangent complex
In this section we continue the discussion started in Algebra, Section 00S0. We begin with a discussion of base change. The first lemma shows that taking the naive tensor product of the naive cotangent complex with a ring extension isn’t quite as naive as one might think.
Lemma
Let \(R \to S\) and \(S \to S'\) be ring maps. The canonical map \(\NL_{S/R} \otimes_S^\mathbf{L} S' \to \NL_{S/R} \otimes_S S'\) induces an isomorphism \(\tau_{\geq -1}(\NL_{S/R} \otimes_S^\mathbf{L} S') \to \NL_{S/R} \otimes_S S'\) in \(D(S')\). Similarly, given a presentation \(\alpha\) of \(S\) over \(R\) the canonical map \(\NL(\alpha) \otimes_S^\mathbf{L} S' \to \NL(\alpha) \otimes_S S'\) induces an isomorphism \(\tau_{\geq -1}(\NL(\alpha) \otimes_S^\mathbf{L} S') \to \NL(\alpha) \otimes_S S'\) in \(D(S')\).
Proof
Special case of Lemma 0G9G.
Lemma
Let \(R \to S\) and \(R \to R'\) be ring maps. Let \(\alpha : P \to S\) be a presentation of \(S\) over \(R\). Then \(\alpha' : P \otimes_R R' \to S \otimes_R R'\) is a presentation of \(S' = S \otimes_R R'\) over \(R'\). The canonical map \[NL(\alpha) \otimes_S S' \to \NL(\alpha')\] is an isomorphism on \(H^0\) and surjective on \(H^{-1}\). In particular, the canonical map \[\NL_{S/R} \otimes_S S' \to \NL_{S'/R'}\] is an isomorphism on \(H^0\) and surjective on \(H^{-1}\).
Proof
Denote \(I = \Ker(P \to S)\). Denote \(P' = P \otimes_R R'\) and \(I' = \Ker(P' \to S')\). Suppose \(P\) is a polynomial algebra on \(x_j\) for \(j \in J\). The map displayed in the lemma becomes \[\xymatrix{ \bigoplus_{j \in J} S' \text{d}x_j \ar[r] & \bigoplus_{j \in J} S' \text{d}x_j \\ I/I^2 \otimes_S S' \ar[r] \ar[u] & I'/(I')^2 \ar[u] }\] where the left column is \(\NL(\alpha) \otimes_S S'\) and the right column is \(\NL(\alpha')\). By right exactness of tensor product we see that \(I \otimes_R R' \to I'\) is surjective. Hence the bottom arrow is a surjection. This proves the first statement of the lemma. The statement for \(\NL_{S/R} \otimes_S S' \to \NL_{S'/R'}\) follows as these complexes are homotopic to \(\NL(\alpha) \otimes_S S'\) and \(\NL(\alpha')\).
Lemma
Consider a cocartesian diagram of rings \[\xymatrix{ B \ar[r] & B' \\ A \ar[r] \ar[u] & A' \ar[u] }\] If \(B\) is flat over \(A\), then the canonical map \(\NL_{B/A} \otimes_B B' \to \NL_{B'/A'}\) is a quasi-isomorphism. If in addition \(\NL_{B/A}\) has tor-amplitude in \([-1, 0]\) then \(\NL_{B/A} \otimes_B^\mathbf{L} B' \to \NL_{B'/A'}\) is a quasi-isomorphism too.
Proof
Choose a presentation \(\alpha : P \to B\) as in Algebra, Section 00S0. Let \(I = \Ker(\alpha)\). Set \(P' = P \otimes_A A'\) and denote \(\alpha' : P' \to B'\) the corresponding presentation of \(B'\) over \(A'\). As \(B\) is flat over \(A\) we see that \(I' = \Ker(\alpha')\) is equal to \(I \otimes_A A'\). Hence \[I'/(I')^2 = \Coker(I^2 \otimes_A A' \to I \otimes_A A') = I/I^2 \otimes_A A' = I/I^2 \otimes_B B'\] We have \(\Omega_{P'/A'} = \Omega_{P/A} \otimes_A A'\) because both sides have the same basis. It follows that \(\Omega_{P'/A'} \otimes_{P'} B' = \Omega_{P/A} \otimes_P B \otimes_B B'\). This proves that \(\NL(\alpha) \otimes_B B' \to \NL(\alpha')\) is an isomorphism of complexes and hence the first statement holds.
We have \[\NL(\alpha) = I/I^2 \longrightarrow \Omega_{P/A} \otimes_P B\] as a complex of \(B\)-modules with \(I/I^2\) placed in degree \(-1\). Since the term in degree \(0\) is free, this complex has tor-amplitude in \([-1, 0]\) if and only if \(I/I^2\) is a flat \(B\)-module, see Lemma 0653. If this holds, then \(\NL(\alpha) \otimes_B^\mathbf{L} B' = \NL(\alpha) \otimes_B B'\) and we get the second statement.
Lemma
Let \(A \to B\) be a local complete intersection as in Definition 07D0. Then \(\NL_{B/A}\) is a perfect object of \(D(B)\) with tor amplitude in \([-1, 0]\).
Proof
Write \(B = A[x_1, \ldots, x_n]/I\). Then \(\NL_{B/A}\) is represented by the complex \[I/I^2 \longrightarrow \bigoplus B \text{d}x_i\] of \(B\)-modules with \(I/I^2\) placed in degree \(-1\). Since the term in degree \(0\) is finite free, this complex has tor-amplitude in \([-1, 0]\) if and only if \(I/I^2\) is a flat \(B\)-module, see Lemma 0653. By definition \(I\) is a Koszul regular ideal and hence a quasi-regular ideal, see Section 07CU. Thus \(I/I^2\) is a finite projective \(B\)-module (Lemma 08RK) and we conclude both that \(\NL_{B/A}\) is perfect and that it has tor amplitude in \([-1, 0]\).
Lemma
Consider a cocartesian diagram of rings \[\xymatrix{ B \ar[r] & B' \\ A \ar[r] \ar[u] & A' \ar[u] }\] If \(A \to B\) and \(A' \to B'\) are local complete intersections as in Definition 07D0, then the kernel of \(H^{-1}(\NL_{B/A} \otimes_B B') \to H^{-1}(\NL_{B'/A'})\) is a finite projective \(B'\)-module.
Proof
By Lemma 0FV0 the complexes \(\NL_{B/A}\) and \(\NL_{B'/A'}\) are perfect of tor-amplitude in \([-1, 0]\). Combining Lemmas 0FUY, 066W, and 066L we have \(\NL_{B/A} \otimes_B B' = \NL_{B/A} \otimes_B^\mathbf{L} B'\) and this complex is also perfect of tor-amplitude in \([-1, 0]\). Choose a distinguished triangle \[C \to \NL_{B/A} \otimes_B B' \to \NL_{B'/A'} \to C[1]\] in \(D(B')\). By Lemmas 066R and 0655 we conclude that \(C\) is perfect with tor-amplitude in \([-1, 1]\). By Lemma 0FUZ the complex \(C\) has only one nonzero cohomology module, namely the module of the lemma sitting in degree \(-1\). This module is of finite presentation (Lemma 064T) and flat (Lemma 066F). Hence it is finite projective by Algebra, Lemma 00NX.
Rlim of abelian groups
We briefly discuss \(R\lim\) on abelian groups. In this section we will denote \(\textit{Ab}(\mathbf{N})\) the abelian category of inverse systems of abelian groups. The notation is compatible with the notation for sheaves of abelian groups on a site, as an inverse system of abelian groups is the same thing as a sheaf of groups on the category \(\mathbf{N}\) (with a unique morphism \(i \to j\) if \(i \leq j\)), see Remark 091A. Many of the arguments in this section duplicate the arguments used to construct the cohomological machinery for sheaves of abelian groups on sites.
Lemma
The functor \(\lim : \textit{Ab}(\mathbf{N}) \to \textit{Ab}\) has a right derived functor [08U4]\[\begin{equation} R\lim : D(\textit{Ab}(\mathbf{N})) \longrightarrow D(\textit{Ab}) \end{equation}\] As usual we set \(R^p\lim(K) = H^p(R\lim(K))\). Moreover, we have
for any \((A_n)\) in \(\textit{Ab}(\mathbf{N})\) we have \(R^p\lim A_n = 0\) for \(p > 1\),
the object \(R\lim A_n\) of \(D(\textit{Ab})\) is represented by the complex \[\prod A_n \to \prod A_n,\quad (x_n) \mapsto (x_n - f_{n + 1}(x_{n + 1}))\] sitting in degrees \(0\) and \(1\),
if \((A_n)\) is ML, then \(R^1\lim A_n = 0\), i.e., \((A_n)\) is right acyclic for \(\lim\),
every \(K^\bullet \in D(\textit{Ab}(\mathbf{N}))\) is quasi-isomorphic to a complex whose terms are right acyclic for \(\lim\), and
if each \(K^p = (K^p_n)\) is right acyclic for \(\lim\), i.e., if \(R^1\lim_n K^p_n = 0\), then \(R\lim K\) is represented by the complex whose term in degree \(p\) is \(\lim_n K_n^p\).
Proof
Let \((A_n)\) be an arbitrary inverse system. Let \((B_n)\) be the inverse system with \[B_n = A_n \oplus A_{n - 1} \oplus \ldots \oplus A_1\] and transition maps given by projections. Let \(A_n \to B_n\) be given by \((1, f_n, f_{n - 1} \circ f_n, \ldots, f_2 \circ \ldots \circ f_n)\) where \(f_i : A_i \to A_{i - 1}\) are the transition maps. In this way we see that every inverse system is a subobject of a ML system (Homology, Section 02MY). It follows from Derived Categories, Lemma 05T8 using Homology, Lemma 02N1 that every ML system is right acyclic for \(\lim\), i.e., (3) holds. This already implies that \(R\lim\) is defined on \(D^+(\textit{Ab}(\mathbf{N}))\), see Derived Categories, Proposition 05TA. Set \(C_n = A_{n - 1} \oplus \ldots \oplus A_1\) for \(n > 1\) and \(C_1 = 0\) with transition maps given by projections as well. Then there is a short exact sequence of inverse systems \(0 \to (A_n) \to (B_n) \to (C_n) \to 0\) where \(B_n \to C_n\) is given by \((x_i) \mapsto (x_i - f_{i + 1}(x_{i + 1}))\). Since \((C_n)\) is ML as well, we conclude that (2) holds (by proposition reference above) which also implies (1). Finally, this implies by Derived Categories, Lemma 07K7 that \(R\lim\) is in fact defined on all of \(D(\textit{Ab}(\mathbf{N}))\). In fact, the proof of Derived Categories, Lemma 07K7 proceeds by proving assertions (4) and (5).
Lemma
Let \[0 \to (A_i) \to (B_i) \to (C_i) \to 0\] be a short exact sequence of inverse systems of abelian groups. Then there is an associated \(6\) term exact sequence \(0 \to \lim A_i \to \lim B_i \to \lim C_i \to R^1\lim A_i \to R^1\lim B_i \to R^1\lim C_i \to 0\).
Proof
Follows from the vanishing in Lemma 07KW.
Here is the “correct” formulation of Homology, Lemma 070E.
Lemma
Let \[(A^{-2}_n \to A^{-1}_n \to A^0_n \to A^1_n)\] be an inverse system of complexes of abelian groups and denote \(A^{-2} \to A^{-1} \to A^0 \to A^1\) its limit. Denote \((H_n^{-1})\), \((H_n^0)\) the inverse systems of cohomologies, and denote \(H^{-1}\), \(H^0\) the cohomologies of \(A^{-2} \to A^{-1} \to A^0 \to A^1\). If
\((A^{-2}_n)\) and \((A^{-1}_n)\) have vanishing \(R^1\lim\),
\((H^{-1}_n)\) has vanishing \(R^1\lim\),
then \(H^0 = \lim H_n^0\).
Proof
Let \(K \in D(\textit{Ab}(\mathbf{N}))\) be the object represented by the system of complexes whose \(n\)th constituent is the complex \(A^{-2}_n \to A^{-1}_n \to A^0_n \to A^1_n\). We will compute \(H^0(R\lim K)\) using both spectral sequences11 of Derived Categories, Lemma 015J. The first has \(E_1\)-page \[\begin{matrix} 0 & 0 & R^1\lim A^0_n & R^1\lim A^1_n \\ A^{-2} & A^{-1} & A^0 & A^1 \end{matrix}\] with horizontal differentials and all higher differentials are zero. The second has \(E_2\) page \[\begin{matrix} R^1\lim H^{-2}_n & 0 & R^1\lim H^0_n & R^1 \lim H^1_n \\ \lim H^{-2}_n & \lim H^{-1}_n & \lim H^0_n & \lim H^1_n \end{matrix}\] and degenerates at this point. The result follows.
Lemma
Let \((A_n)\) and \((B_n)\) be inverse systems of abelian groups. A morphism of pro-systems \(\varphi : (A_n) \to (B_n)\) determines maps \(\lim A_n \to \lim B_n\) and \(R^1\lim A_n \to R^1\lim B_n\). These maps are isomorphisms if \(\varphi\) is a pro-isomorphism.
Proof
Please see Categories, Example 0G2W for a discussion of morphisms of pro-systems. The map \(\varphi\) is given by \(1 \leq m_1 < m_2 < m_3 < \ldots\) and maps \(\varphi_n : A_{m_n} \to B_n\) compatible with transition maps. Set \(\varphi', \varphi'' : \prod A_n \to \prod B_n\) equal to \(\varphi'((a_n)) = (\varphi_n(a_{m_n}))\) and \[\varphi''((a_n)) = (\varphi_n(a_{m_n} + f(a_{m_n + 1}) + \ldots + f(a_{m_{n + 1} - 1})))\] where each occurence of \(f\) denotes a suitable transition map of the inverse system \((A_n)\). Then the diagram \[\xymatrix{ \prod A_n \ar[r]_\delta \ar[d]^{\varphi'} & \prod A_n \ar[d]^{\varphi''} \\ \prod B_n \ar[r]^\delta & \prod B_n }\] where the horizontal arrows are as in Lemma 07KW is commutative. In this way we obtain the desired maps. The construction is functorial in the sense that if we’re given an inverse system \((C_n)\) and \(1 \leq m'_1 < m'_2 < m'_3 < \ldots\) and maps \(\psi_n : B_{m'_n} \to C_n\) compatible with transition maps, then \((\psi \circ \varphi)' = \psi' \circ \varphi'\) and \((\psi \circ \varphi)'' = \psi'' \circ \varphi''\) where the composition \(\psi \circ \varphi\) refers to the integers \(1 \leq m_{m'_1} < m_{m'_2} < \ldots\) and the maps \(\psi_n \circ \varphi_{m'_n} : A_{m_{m'_n}} \to C_n\). We will show that if \(B_n = A_n\) and \(\varphi_n\) is the transition map, then the resulting maps are the identity maps. This will both prove that the construction is independent of the choice of the representative \((m_n, \varphi_n)\) of \(\varphi\) and the final statement of the lemma.
Thus we let \(B_n = A_n\) and \(\varphi_n\) be the transition map. Let \((a_n) \in \lim A_n\) be an element. Then \(\varphi'((a_n))\) is the element which has in degree \(n\) the image of \(a_{m_n}\) which is equal to \(a_n\). This proves the statement for \(\lim A_n\). Let \(\xi \in R^1\lim A_n\) be the class of the element \((a_n)\) in \(\prod A_n\). Consider the element \((b_n)\) with \[b_i = a_i + f(a_{i + 1}) + \ldots + f(a_{m_n - 1})\] if \(m_{n - 1} < i < m_n\) and \(0\) if \(i = m_n\) for some \(n\). Then \((a'_n) = (a_n) - \delta((b_n))\) defines the same class in \(R^1\lim A_n\) and a computation shows that \(a'_i = 0\) unless \(i = m_n\) for some \(n\). Thus we may and do assume \(a_i = 0\) unless \(i = m_n\) for some \(n\). Note that \[(0, \ldots, -f(a_{m_j}), 0, \ldots, 0, a_{m_j}, 0, \ldots)\] with nonzero entries in spots \(j\) and \(m_j\), is the image under \(\delta\) of \[c_j = (0, \ldots, 0, f(a_{m_j}), f(a_{m_j}), \ldots, a_{m_j}, 0, \ldots)\] with nonzero entries in spots \(j + 1, \ldots, m_j\). The sum \(c = \sum c_j\) makes sense in \(\prod A_n\). Recalling that \(\varphi''((a_n)) = (a_{m_1}, a_{m_2}, \ldots)\) we see that \[(a_n) - \varphi''((a_n)) = \delta(c)\] and the proof is complete.
Lemma
Let \(\mathcal{D}\) be a triangulated category. Let \((K_n)\) be an inverse system of objects of \(\mathcal{D}\). Let \(K\) be a derived limit of the system \((K_n)\). Then for every \(L\) in \(\mathcal{D}\) we have a short exact sequence \[0 \to R^1\lim \Hom_\mathcal{D}(L, K_n[-1]) \to \Hom_\mathcal{D}(L, K) \to \lim \Hom_\mathcal{D}(L, K_n) \to 0\]
Proof
This follows from Derived Categories, Definition 08TC and Lemma 0149, and the description of \(\lim\) and \(R^1\lim\) in Lemma 07KW above.
Lemma
Let \(\mathcal{D}\) be a triangulated category. Let \((K_n)\) and \((M_n)\) be inverse systems of objects of \(\mathcal{D}\) with derived limits \(K\) and \(M\). Let \(a : (K_n) \to (M_n)\) be a pro-isomorphism of pro-objects. Then \(a\) can be used to produce a (non-canonical) isomorphism \(K \to M\).
Proof
We obtain an arrow \(K \to M\) fitting into a morphism \[\xymatrix{ K \ar[r] \ar[d] & \prod K_n \ar[r] \ar[d]^{a'} & \prod K_n \ar[d]^{a''} \\ M \ar[r] & \prod M_n \ar[r] & \prod M_n }\] of defining distinguished triangles by Derived Categories, Remark 0H9J. Thus, for every object \(L\) of \(\mathcal{D}\) we obtain a map of short exact sequences \[\xymatrix{ 0 \ar[r] & R^1\lim \Hom_\mathcal{D}(L, K_n[-1]) \ar[r] \ar[d] & \Hom_\mathcal{D}(L, K) \ar[r] \ar[d] & \lim \Hom_\mathcal{D}(L, K_n) \ar[r] \ar[d] & 0 \\ 0 \ar[r] & R^1\lim \Hom_\mathcal{D}(L, M_n[-1]) \ar[r] & \Hom_\mathcal{D}(L, M) \ar[r] & \lim \Hom_\mathcal{D}(L, M_n) \ar[r] & 0 }\] see Lemma 0919 and its proof. By Lemma 0H9K the left and right vertical arrows are isomorphisms12. Thus the middle arrow is an isomorphism. By the Yoneda lemma we see that the map \(K \to M\) is an isomorphism.
Lemma
Let \(\mathcal{D}\) be a triangulated category. Let \((K_n)\) be a system of objects of \(\mathcal{D}\). Let \(K\) be a derived colimit of the system \((K_n)\). Then for every \(L\) in \(\mathcal{D}\) we have a short exact sequence \[0 \to R^1\lim \Hom_\mathcal{D}(K_n, L[-1]) \to \Hom_\mathcal{D}(K, L) \to \lim \Hom_\mathcal{D}(K_n, L) \to 0\]
Proof
This follows from Derived Categories, Definition 090Z and Lemma 0149, and the description of \(\lim\) and \(R^1\lim\) in Lemma 07KW above.
Remark
Consider the category \(\mathbf{N}\) whose objects are natural numbers and whose morphisms are unique arrows \(i \to j\) if \(j \geq i\). Endow \(\mathbf{N}\) with the chaotic topology (Sites, Example 07GE) so that a sheaf \(\mathcal{F}\) is the same thing as an inverse system \[\mathcal{F}_1 \leftarrow \mathcal{F}_2 \leftarrow \mathcal{F}_3 \leftarrow \ldots\] of sets over \(\mathbf{N}\). Note that \(\Gamma(\mathbf{N}, \mathcal{F}) = \lim \mathcal{F}_n\). For an inverse system of abelian groups \(\mathcal{F}_n\) we have \[R^p\lim \mathcal{F}_n = H^p(\mathbf{N}, \mathcal{F})\] because both sides are the higher right derived functors of \(\mathcal{F} \mapsto \lim \mathcal{F}_n = H^0(\mathbf{N}, \mathcal{F})\). Thus the existence of \(R\lim\) also follows from the general material in Cohomology on Sites, Sections 01FT and 07A5.
The products in the following lemma can be seen as termwise products of complexes or as products in the derived category \(D(\textit{Ab})\), see Derived Categories, Lemma 07KC.
Lemma
Let \(K = (K_n^\bullet)\) be an object of \(D(\textit{Ab}(\mathbf{N}))\). There exists a canonical distinguished triangle \[R\lim K \to \prod\nolimits_n K_n^\bullet \to \prod\nolimits_n K_n^\bullet \to R\lim K[1]\] in \(D(\textit{Ab})\). In other words, \(R\lim K\) is a derived limit of the inverse system \((K_n^\bullet)\) of \(D(\textit{Ab})\), see Derived Categories, Definition 08TC.
Proof
Suppose that for each \(p\) the inverse system \((K_n^p)\) is right acyclic for \(\lim\). By Lemma 07KW this gives a short exact sequence \[0 \to \lim_n K^p_n \to \prod\nolimits_n K^p_n \to \prod\nolimits_n K^p_n \to 0\] for each \(p\). Since the complex consisting of \(\lim_n K^p_n\) computes \(R\lim K\) by Lemma 07KW we see that the lemma holds in this case.
Next, assume \(K = (K_n^\bullet)\) is general. By Lemma 07KW there is a quasi-isomorphism \(K \to L\) in \(D(\textit{Ab}(\mathbf{N}))\) such that \((L_n^p)\) is acyclic for each \(p\). Then \(\prod K_n^\bullet\) is quasi-isomorphic to \(\prod L_n^\bullet\) as products are exact in \(\textit{Ab}\), whence the result for \(L\) (proved above) implies the result for \(K\).
Lemma
With notation as in Lemma 07KX the long exact cohomology sequence associated to the distinguished triangle breaks up into short exact sequences \[0 \to R^1\lim_n H^{p - 1}(K_n^\bullet) \to H^p(R\lim K) \to \lim_n H^p(K_n^\bullet) \to 0\]
Proof
The long exact sequence of the distinguished triangle is \[\ldots \to H^p(R\lim K) \to \prod\nolimits_n H^p(K_n^\bullet) \to \prod\nolimits_n H^p(K_n^\bullet) \to H^{p + 1}(R\lim K) \to \ldots\] The map in the middle has kernel \(\lim_n H^p(K_n^\bullet)\) by its explicit description given in the lemma. The cokernel of this map is \(R^1\lim_n H^p(K_n^\bullet)\) by Lemma 07KW.
Warning. An object of \(D(\textit{Ab}(\mathbf{N}))\) is a complex of inverse systems of abelian groups. You can also think of this as an inverse system \((K_n^\bullet)\) of complexes. However, this is not the same thing as an inverse system of objects of \(D(\textit{Ab})\); the following lemma and remark explain the difference.
Lemma
Let \((K_n)\) be an inverse system of objects of \(D(\textit{Ab})\). Then there exists an object \(M = (M_n^\bullet)\) of \(D(\textit{Ab}(\mathbf{N}))\) and isomorphisms \(M_n^\bullet \to K_n\) in \(D(\textit{Ab})\) such that the diagrams \[\xymatrix{ M_{n + 1}^\bullet \ar[d] \ar[r] & M_n^\bullet \ar[d] \\ K_{n + 1} \ar[r] & K_n }\] commute in \(D(\textit{Ab})\).
Proof
Namely, let \(M_1^\bullet\) be a complex of abelian groups representing \(K_1\). Suppose we have constructed \(M_e^\bullet \to M_{e - 1}^\bullet \to \ldots \to M_1^\bullet\) and maps \(\psi_i : M_i^\bullet \to K_i\) such that the diagrams in the statement of the lemma commute for all \(n < e\). Then we consider the diagram \[\xymatrix{ & M_n^\bullet \ar[d]^{\psi_n} \\ K_{n + 1} \ar[r] & K_n }\] in \(D(\textit{Ab})\). By the definition of morphisms in \(D(\textit{Ab})\) we can find a complex \(M_{n + 1}^\bullet\) of abelian groups, an isomorphism \(M_{n + 1}^\bullet \to K_{n + 1}\) in \(D(\textit{Ab})\), and a morphism of complexes \(M_{n + 1}^\bullet \to M_n^\bullet\) representing the composition \[K_{n + 1} \to K_n \xrightarrow{\psi_n^{-1}} M_n^\bullet\] in \(D(\textit{Ab})\). Thus the lemma holds by induction.
Remark
Let \((K_n)\) be an inverse system of objects of \(D(\textit{Ab})\). Let \(K = R\lim K_n\) be a derived limit of this system (see Derived Categories, Section 08TB). Such a derived limit exists because \(D(\textit{Ab})\) has countable products (Derived Categories, Lemma 07KC). By Lemma 0CQ9 we can also lift \((K_n)\) to an object \(M\) of \(D(\textit{Ab}(\mathbf{N}))\). Then \(K \cong R\lim M\) where \(R\lim\) is the functor (08U4) because \(R\lim M\) is also a derived limit of the system \((K_n)\) by Lemma 07KX. Thus, although there may be many isomorphism classes of lifts \(M\) of the system \((K_n)\), the isomorphism type of \(R\lim M\) is independent of the choice because it is isomorphic to the derived limit \(K = R\lim K_n\) of the system. Thus we may apply results on \(R\lim\) proved in this section to derived limits. For example, for every \(p \in \mathbf{Z}\) there is a canonical short exact sequence \[0 \to R^1\lim H^{p - 1}(K_n) \to H^p(K) \to \lim H^p(K_n) \to 0\] because we may apply Lemma 07KY to \(M\). This can also be seen directly, without invoking the existence of \(M\), by applying the argument of the proof of Lemma 07KY to the (defining) distinguished triangle \(K \to \prod K_n \to \prod K_n \to K[1]\).
Lemma
Let \(E \to D\) be a morphism of \(D(\textit{Ab}(\mathbf{N}))\). Let \((E_n)\), resp. \((D_n)\) be the system of objects of \(D(\textit{Ab})\) associated to \(E\), resp. \(D\). If \((E_n) \to (D_n)\) is an isomorphism of pro-objects, then \(R\lim E \to R\lim D\) is an isomorphism in \(D(\textit{Ab})\).
Proof
The assumption in particular implies that the pro-objects \(H^p(E_n)\) and \(H^p(D_n)\) are isomorphic. The result follows from the short exact sequences of Lemma 07KY and Lemma 0H9K.
Lemma
Let \((A_n)\) be an inverse system of abelian groups. The following are equivalent
\((A_n)\) is Mittag-Leffler,
\(R^1\lim A_n = 0\) and the same holds for \(\bigoplus_{i \in \mathbf{N}} (A_n)\).
Proof
Set \(B = \bigoplus_{i \in \mathbf{N}} (A_n)\) and hence \(B = (B_n)\) with \(B_n = \bigoplus_{i \in \mathbf{N}} A_n\). If \((A_n)\) is ML, then \(B\) is ML and hence \(R^1\lim A_n = 0\) and \(R^1\lim B_n = 0\) by Lemma 07KW.
Conversely, assume \((A_n)\) is not ML. Then we can pick an \(m\) and a sequence of integers \(m < m_1 < m_2 < \ldots\) and elements \(x_i \in A_{m_i}\) whose image \(y_i \in A_m\) is not in the image of \(A_{m_i + 1} \to A_m\). We will use the elements \(x_i\) and \(y_i\) to show that \(R^1\lim B_n \not = 0\) in two ways. This will finish the proof of the lemma.
First proof. Set \(C = (C_n)\) with \(C_n = \prod_{i \in \mathbf{N}} A_n\). There is a canonical injective map \(B_n \to C_n\) with cokernel \(Q_n\). Set \(Q = (Q_n)\). We may and do think of elements \(q_n\) of \(Q_n\) as sequences of elements \(q_n = (q_{n, 1}, q_{n, 2}, \ldots)\) with \(q_{n, i} \in A_n\) modulo sequences whose tail is zero (in other words, we identify sequences which differ in finitely many places). We have a short exact sequence of inverse systems \[0 \to (B_n) \to (C_n) \to (Q_n) \to 0\] Consider the element \(q_n \in Q_n\) given by \[q_{n, i} = \left\{ \begin{matrix} \text{image of }x_i &\text{if}& m_i \geq n \\ 0 & \text{else} \end{matrix} \right.\] Then it is clear that \(q_{n + 1}\) maps to \(q_n\). Hence we obtain \(q = (q_n) \in \lim Q_n\). On the other hand, we claim that \(q\) is not in the image of \(\lim C_n \to \lim Q_n\). Namely, say that \(c = (c_n)\) maps to \(q\). Then we can write \(c_n = (c_{n, i})\) and since \(c_{n', i} \mapsto c_{n, i}\) for \(n' \geq n\), we see that \(c_{n, i} \in \Im(C_{n'} \to C_n)\) for all \(n, i, n' \geq n\). In particular, the image of \(c_{m, i}\) in \(A_m\) is in \(\Im(A_{m_i + 1} \to A_m)\) whence cannot be equal to \(y_i\). Thus \(c_m\) and \(q_m = (y_1, y_2, y_3, \ldots)\) differ in infinitely many spots, which is a contradiction. Considering the long exact cohomology sequence \[0 \to \lim B_n \to \lim C_n \to \lim Q_n \to R^1\lim B_n\] we conclude that the last group is nonzero as desired.
Second proof. For \(n' \geq n\) we denote \(A_{n, n'} = \Im(A_{n'} \to A_n)\). Then we have \(y_i \in A_m\), \(y_i \not \in A_{m, m_i + 1}\). Let \(\xi = (\xi_n) \in \prod B_n\) be the element with \(\xi_n = 0\) unless \(n = m_i\) and \(\xi_{m_i} = (0, \ldots, 0, x_i, 0, \ldots)\) with \(x_i\) placed in the \(i\)th summand. We claim that \(\xi\) is not in the image of the map \(\prod B_n \to \prod B_n\) of Lemma 07KW. This shows that \(R^1\lim B_n\) is nonzero and finishes the proof. Namely, suppose that \(\xi\) is the image of \(\eta = (z_1, z_2, \ldots)\) with \(z_n = \sum z_{n, i} \in \bigoplus_i A_n\). Observe that \(x_i = z_{m_i, i} \bmod A_{m_i, m_i + 1}\). Then \(z_{m_i - 1, i}\) is the image of \(z_{m_i, i}\) under \(A_{m_i} \to A_{m_i - 1}\), and so on, and we conclude that \(z_{m, i}\) is the image of \(z_{m_i, i}\) under \(A_{m_i} \to A_m\). We conclude that \(z_{m, i}\) is congruent to \(y_i\) modulo \(A_{m, m_i + 1}\). In particular \(z_{m, i} \not = 0\). This is impossible as \(\sum z_{m, i} \in \bigoplus_i A_m\) hence only a finite number of \(z_{m, i}\) can be nonzero.
Lemma
Let \[0 \to (A_i) \to (B_i) \to (C_i) \to 0\] be a short exact sequence of inverse systems of abelian groups. If \((A_i)\) and \((C_i)\) are ML, then so is \((B_i)\).
Proof
This follows from Lemma 0CQA, the fact that taking infinite direct sums is exact, and the long exact sequence of cohomology associated to \(R\lim\).
Lemma
Let \((A_n)\) be an inverse system of abelian groups. The following are equivalent
\((A_n)\) is zero as a pro-object,
\(\lim A_n = 0\) and \(R^1\lim A_n = 0\) and the same holds for \(\bigoplus_{i \in \mathbf{N}} (A_n)\).
Proof
It follows from Lemma 0H9K that (1) implies (2). Assume (2). Then \((A_n)\) is ML by Lemma 0CQA. For \(m \geq n\) let \(A_{n, m} = \Im(A_m \to A_n)\) so that \(A_n = A_{n, n} \supset A_{n, n + 1} \supset \ldots\). Note that \((A_n)\) is zero as a pro-object if and only if for every \(n\) there is an \(m \geq n\) such that \(A_{n, m} = 0\). Note that \((A_n)\) is ML if and only if for every \(n\) there is an \(m_n \geq n\) such that \(A_{n, m} = A_{n, m + 1} = \ldots\). In the ML case it is clear that \(\lim A_n = 0\) implies that \(A_{n, m_n} = 0\) because the maps \(A_{n + 1, m_{n + 1}} \to A_{n, m}\) are surjective. This finishes the proof.
Rlim of modules
We briefly discuss \(R\lim\) on modules. Many of the arguments in this section duplicate the arguments used to construct the cohomological machinery for modules on ringed sites.
Let \((A_n)\) be an inverse system of rings. We will denote \(\textit{Mod}(\mathbf{N}, (A_n))\) the category of inverse systems \((M_n)\) of abelian groups such that each \(M_n\) is given the structure of a \(A_n\)-module and the transition maps \(M_{n + 1} \to M_n\) are \(A_{n + 1}\)-module maps. This is an abelian category. Set \(A = \lim A_n\). Given an object \((M_n)\) of \(\textit{Mod}(\mathbf{N}, (A_n))\) the limit \(\lim M_n\) is an \(A\)-module.
Lemma
In the situation above. The functor \(\lim : \textit{Mod}(\mathbf{N}, (A_n)) \to \text{Mod}_A\) has a right derived functor \[R\lim : D(\textit{Mod}(\mathbf{N}, (A_n))) \longrightarrow D(A)\] As usual we set \(R^p\lim(K) = H^p(R\lim(K))\). Moreover, we have
for any \((M_n)\) in \(\textit{Mod}(\mathbf{N}, (A_n))\) we have \(R^p\lim M_n = 0\) for \(p > 1\),
the object \(R\lim M_n\) of \(D(\text{Mod}_A)\) is represented by the complex \[\prod M_n \to \prod M_n,\quad (x_n) \mapsto (x_n - f_{n + 1}(x_{n + 1}))\] sitting in degrees \(0\) and \(1\),
if \((M_n)\) is ML, then \(R^1\lim M_n = 0\), i.e., \((M_n)\) is right acyclic for \(\lim\),
every \(K^\bullet \in D(\textit{Mod}(\mathbf{N}, (A_n)))\) is quasi-isomorphic to a complex whose terms are right acyclic for \(\lim\), and
if each \(K^p = (K^p_n)\) is right acyclic for \(\lim\), i.e., of \(R^1\lim_n K^p_n = 0\), then \(R\lim K\) is represented by the complex whose term in degree \(p\) is \(\lim_n K_n^p\).
Proof
The proof of this is word for word the same as the proof of Lemma 07KW.
Remark
This remark is a continuation of Remark 091A. A sheaf of rings on \(\mathbf{N}\) is just an inverse system of rings \((A_n)\). A sheaf of modules over \((A_n)\) is exactly the same thing as an object of the category \(\textit{Mod}(\mathbf{N}, (A_n))\) defined above. The derived functor \(R\lim\) of Lemma 091D is simply \(R\Gamma(\mathbf{N}, -)\) from the derived category of modules to the derived category of modules over the global sections of the structure sheaf. It is true in general that cohomology of groups and modules agree, see Cohomology on Sites, Lemma 03FD.
The products in the following lemma can be seen as termwise products of complexes or as products in the derived category \(D(A)\), see Derived Categories, Lemma 07KC.
Lemma
Let \(K = (K_n^\bullet)\) be an object of \(D(\textit{Mod}(\mathbf{N}, (A_n)))\). There exists a canonical distinguished triangle \[R\lim K \to \prod\nolimits_n K_n^\bullet \to \prod\nolimits_n K_n^\bullet \to R\lim K[1]\] in \(D(A)\). In other words, \(R\lim K\) is a derived limit of the inverse system \((K_n^\bullet)\) of \(D(A)\), see Derived Categories, Definition 08TC.
Proof
The proof is exactly the same as the proof of Lemma 07KX using Lemma 091D in stead of Lemma 07KW.
Lemma
With notation as in Lemma 0CQD the long exact cohomology sequence associated to the distinguished triangle breaks up into short exact sequences \[0 \to R^1\lim_n H^{p - 1}(K_n^\bullet) \to H^p(R\lim K) \to \lim_n H^p(K_n^\bullet) \to 0\] of \(A\)-modules.
Proof
The proof is exactly the same as the proof of Lemma 07KY using Lemma 091D in stead of Lemma 07KW.
Warning. As in the case of abelian groups an object \(M = (M_n^\bullet)\) of \(D(\textit{Mod}(\mathbf{N}, (A_n)))\) is an inverse system of complexes of modules, which is not the same thing as an inverse system of objects in the derived categories. In the following lemma we show how an inverse system of objects in derived categories always lifts to an object of \(D(\textit{Mod}(\mathbf{N}, (A_n)))\).
Lemma
Let \((A_n)\) be an inverse system of rings. Suppose that we are given
for every \(n\) an object \(K_n\) of \(D(A_n)\), and
for every \(n\) a map \(\varphi_n : K_{n + 1} \to K_n\) of \(D(A_{n + 1})\) where we think of \(K_n\) as an object of \(D(A_{n + 1})\) by restriction via \(A_{n + 1} \to A_n\).
There exists an object \(M = (M_n^\bullet) \in D(\textit{Mod}(\mathbf{N}, (A_n)))\) and isomorphisms \(\psi_n : M_n^\bullet \to K_n\) in \(D(A_n)\) such that the diagrams \[\xymatrix{ M_{n + 1}^\bullet \ar[d]_{\psi_{n + 1}} \ar[r] & M_n^\bullet \ar[d]^{\psi_n} \\ K_{n + 1} \ar[r]^{\varphi_n} & K_n }\] commute in \(D(A_{n + 1})\).
Proof
We write out the proof in detail. For an \(A_n\)-module \(T\) we write \(T_{A_{n + 1}}\) for the same module viewd as an \(A_{n + 1}\)-module. Suppose that \(K_n^\bullet\) is a complex of \(A_n\)-modules representing \(K_n\). Then \(K_{n, A_{n + 1}}^\bullet\) is the same complex, but viewed as a complex of \(A_{n + 1}\)-modules. By the construction of the derived category, the map \(\psi_n\) can be given as \[\psi_n = \tau_n \circ \sigma_n^{-1}\] where \(\sigma_n : L_{n + 1}^\bullet \to K_{n + 1}^\bullet\) is a quasi-isomorphism of complexes of \(A_{n + 1}\)-modules and \(\tau_n : L_{n + 1}^\bullet \to K_{n, A_{n + 1}}^\bullet\) is a map of complexes of \(A_{n + 1}\)-modules.
Now we construct the complexes \(M_n^\bullet\) by induction. As base case we let \(M_1^\bullet = K_1^\bullet\). Suppose we have already constructed \(M_e^\bullet \to M_{e - 1}^\bullet \to \ldots \to M_1^\bullet\) and maps of complexes \(\psi_i : M_i^\bullet \to K_i^\bullet\) such that the diagrams \[\xymatrix{ M_{n + 1}^\bullet \ar[d]_{\psi_{n + 1}} \ar[rr] & & M_{n, A_{n + 1}}^\bullet \ar[d]^{\psi_{n, A_{n + 1}}} \\ K_{n + 1}^\bullet & L_{n + 1}^\bullet \ar[l]_{\sigma_n} \ar[r]^{\tau_n} & K_{n, A_{n + 1}}^\bullet }\] above commute in \(D(A_{n + 1})\) for all \(n < e\). Then we consider the diagram \[\xymatrix{ & & M_{e, A_{e + 1}}^\bullet \ar[d]^{\psi_{e, A_{e + 1}}} \\ K_{e + 1}^\bullet & L_{e + 1}^\bullet \ar[r]^{\tau_e} \ar[l]_{\sigma_e} & K_{e, A_{e + 1}}^\bullet }\] in \(D(A_{e + 1})\). Because \(\psi_e\) is a quasi-isomorphism, we see that \(\psi_{e, A_{e + 1}}\) is a quasi-isomorphism too. By the definition of morphisms in \(D(A_{e + 1})\) we can find a quasi-isomorphism \(\psi_{e + 1} : M_{e + 1}^\bullet \to K_{e + 1}^\bullet\) of complexes of \(A_{e + 1}\)-modules such that there exists a morphism of complexes \(M_{e + 1}^\bullet \to M_{e, A_{e + 1}}^\bullet\) of \(A_{e + 1}\)-modules representing the composition \(\psi_{e, A_{e + 1}}^{-1} \circ \tau_e \circ \sigma_e^{-1}\) in \(D(A_{e + 1})\). Thus the lemma holds by induction.
Remark
With assumptions as in Lemma 091I. A priori there are many isomorphism classes of objects \(M\) of \(D(\textit{Mod}(\mathbf{N}, (A_n)))\) which give rise to the system \((K_n, \varphi_n)\) of the lemma. For each such \(M\) we can consider the complex \(R\lim M \in D(A)\) where \(A = \lim A_n\). By Lemma 0CQD we see that \(R\lim M\) is a derived limit of the inverse system \((K_n)\) of \(D(A)\). Hence we see that the isomorphism class of \(R\lim M\) in \(D(A)\) is independent of the choices made in constructing \(M\). In particular, we may apply results on \(R\lim\) proved in this section to derived limits of inverse systems in \(D(A)\). For example, for every \(p \in \mathbf{Z}\) there is a canonical short exact sequence \[0 \to R^1\lim H^{p - 1}(K_n) \to H^p(R\lim K_n) \to \lim H^p(K_n) \to 0\] because we may apply Lemma 0CQD to \(M\). This can also been seen directly, without invoking the existence of \(M\), by applying the argument of the proof of Lemma 0CQD to the (defining) distinguished triangle \(R\lim K_n \to \prod K_n \to \prod K_n \to (R\lim K_n)[1]\) of the derived limit.
Lemma
Let \((A_n)\) be an inverse system of rings. Every \(K \in D(\textit{Mod}(\mathbf{N}, (A_n)))\) can be represented by a system of complexes \((M_n^\bullet)\) such that all the transition maps \(M_{n + 1}^\bullet \to M_n^\bullet\) are surjective.
Proof
Let \(K\) be represented by the system \((K_n^\bullet)\). Set \(M_1^\bullet = K_1^\bullet\). Suppose we have constructed surjective maps of complexes \(M_n^\bullet \to M_{n - 1}^\bullet \to \ldots \to M_1^\bullet\) and homotopy equivalences \(\psi_e : K_e^\bullet \to M_e^\bullet\) such that the diagrams \[\xymatrix{ K_{e + 1}^\bullet \ar[d] \ar[r] & K_e^\bullet \ar[d] \\ M_{e + 1}^\bullet \ar[r] & M_e^\bullet }\] commute for all \(e < n\). Then we consider the diagram \[\xymatrix{ K_{n + 1}^\bullet \ar[r] & K_n^\bullet \ar[d] \\ & M_n^\bullet }\] By Derived Categories, Lemma 0642 we can factor the composition \(K_{n + 1}^\bullet \to M_n^\bullet\) as \(K_{n + 1}^\bullet \to M_{n + 1}^\bullet \to M_n^\bullet\) such that the first arrow is a homotopy equivalence and the second a termwise split surjection. The lemma follows from this and induction.
Lemma
Let \((A_n)\) be an inverse system of rings. Every \(K \in D(\textit{Mod}(\mathbf{N}, (A_n)))\) can be represented by a system of complexes \((K_n^\bullet)\) such that each \(K_n^\bullet\) is K-flat.
Proof
First use Lemma 091F to represent \(K\) by a system of complexes \((M_n^\bullet)\) such that all the transition maps \(M_{n + 1}^\bullet \to M_n^\bullet\) are surjective. Next, let \(K_1^\bullet \to M_1^\bullet\) be a quasi-isomorphism with \(K_1^\bullet\) a K-flat complex of \(A_1\)-modules (Lemma 06Y4). Suppose we have constructed \(K_n^\bullet \to K_{n - 1}^\bullet \to \ldots \to K_1^\bullet\) and maps of complexes \(\psi_e : K_e^\bullet \to M_e^\bullet\) such that \[\xymatrix{ K_{e + 1}^\bullet \ar[d] \ar[r] & K_e^\bullet \ar[d] \\ M_{e + 1}^\bullet \ar[r] & M_e^\bullet }\] commutes for all \(e < n\). Then we consider the diagram \[\xymatrix{ C^\bullet \ar@{..>}[d] \ar@{..>}[r] & K_n^\bullet \ar[d]^{\psi_n} \\ M_{n + 1}^\bullet \ar[r]^{\varphi_n} & M_n^\bullet }\] in \(D(A_{n + 1})\). As \(M_{n + 1}^\bullet \to M_n^\bullet\) is termwise surjective, the complex \(C^\bullet\) fitting into the left upper corner with terms \[C^p = M_{n + 1}^p \times_{M_n^p} K_n^p\] is quasi-isomorphic to \(M_{n + 1}^\bullet\) (details omitted). Choose a quasi-isomorphism \(K_{n + 1}^\bullet \to C^\bullet\) with \(K_{n +1}^\bullet\) K-flat. Thus the lemma holds by induction.
Lemma
Let \((A_n)\) be an inverse system of rings. Given \(K, L \in D(\textit{Mod}(\mathbf{N}, (A_n)))\) there is a canonical derived tensor product \(K \otimes^\mathbf{L} L\) in \(D(\mathbf{N}, (A_n))\) compatible with the maps to \(D(A_n)\). The construction is symmetric in \(K\) and \(L\) and an exact functor of triangulated categories in each variable.
Proof
Choose a representative \((K_n^\bullet)\) for \(K\) such that each \(K_n^\bullet\) is a K-flat complex (Lemma 091G). Then you can define \(K \otimes^\mathbf{L} L\) as the object represented by the system of complexes \[(\text{Tot}(K_n^\bullet \otimes_{A_n} L_n^\bullet))\] for any choice of representative \((L_n^\bullet)\) for \(L\). This is well defined in both variables by Lemmas 06Y0 and 064L. Compatibility with the map to \(D(A_n)\) is clear. Exactness follows exactly as in Lemma 064J.
Remark
Let \(A\) be a ring. Let \((E_n)\) be an inverse system of objects of \(D(A)\). We’ve seen above that a derived limit \(R\lim E_n\) exists. Thus for every object \(K\) of \(D(A)\) also the derived limit \(R\lim( K \otimes_A^\mathbf{L} E_n )\) exists. It turns out that we can construct these derived limits functorially in \(K\) and obtain an exact functor \[R\lim(- \otimes_A^\mathbf{L} E_n) : D(A) \longrightarrow D(A)\] of triangulated categories. Namely, we first lift \((E_n)\) to an object \(E\) of \(D(\mathbf{N}, A)\), see Lemma 091I. (The functor will depend on the choice of this lift.) Next, observe that there is a “diagonal” or “constant” functor \[\Delta : D(A) \longrightarrow D(\mathbf{N}, A)\] mapping the complex \(K^\bullet\) to the constant inverse system of complexes with value \(K^\bullet\). Then we simply define \[R\lim(K \otimes_A^\mathbf{L} E_n) = R\lim(\Delta(K)\otimes^\mathbf{L} E)\] where on the right hand side we use the functor \(R\lim\) of Lemma 091D and the functor \(- \otimes^\mathbf{L} -\) of Lemma 091H.
Lemma
Let \(A\) be a ring. Let \(E \to D \to F \to E[1]\) be a distinguished triangle of \(D(\mathbf{N}, A)\). Let \((E_n)\), resp. \((D_n)\), resp. \((F_n)\) be the system of objects of \(D(A)\) associated to \(E\), resp. \(D\), resp. \(F\). Then for every \(K \in D(A)\) there is a canonical distinguished triangle \[R\lim (K \otimes^\mathbf{L}_A E_n) \to R\lim (K \otimes^\mathbf{L}_A D_n) \to R\lim (K \otimes^\mathbf{L}_A F_n) \to R\lim (K \otimes^\mathbf{L}_A E_n)[1]\] in \(D(A)\) with notation as in Remark 091J.
Proof
This is clear from the construction in Remark 091J and the fact that \(\Delta : D(A) \to D(\mathbf{N}, A)\), \(- \otimes^\mathbf{L} -\), and \(R\lim\) are exact functors of triangulated categories.
Lemma
Let \(A\) be a ring. Let \(E \to D\) be a morphism of \(D(\mathbf{N}, A)\). Let \((E_n)\), resp. \((D_n)\) be the system of objects of \(D(A)\) associated to \(E\), resp. \(D\). If \((E_n) \to (D_n)\) is an isomorphism of pro-objects, then for every \(K \in D(A)\) the corresponding map \[R\lim (K \otimes^\mathbf{L}_A E_n) \longrightarrow R\lim (K \otimes^\mathbf{L}_A D_n)\] in \(D(A)\) is an isomorphism (notation as in Remark 091J).
Proof
Follows from the definitions and Lemma 091B.
Torsion modules
In this section “torsion modules” will refer to modules supported on a given closed subset \(V(I)\) of an affine scheme \(\Spec(R)\). This is different, but analogous to, the notion of a torsion module over a domain (Definition 0536).
Definition
Let \(R\) be a ring. Let \(M\) be an \(R\)-module.
Let \(I \subset R\) be an ideal. We say \(M\) is an \(I\)-power torsion module if for every \(m \in M\) there exists an \(n > 0\) such that \(I^n m = 0\).
Let \(f \in R\). We say \(M\) is an \(f\)-power torsion module if for each \(m \in M\), there exists an \(n > 0\) such that \(f^n m = 0\).
Thus an \(f\)-power torsion module is the same thing as an \(I\)-power torsion module for \(I = (f)\). We will use the notation \[M[I^n] = \{m \in M \mid I^nm = 0\}\] and \[M[I^\infty] = \bigcup M[I^n]\] for an \(R\)-module \(M\). Thus \(M\) is \(I\)-power torsion if and only if \(M = M[I^\infty]\) if and only if \(M = \bigcup M[I^n]\).
Lemma
Let \(R\) be a ring. Let \(I\) be an ideal of \(R\). Let \(M\) be an \(I\)-power torsion module. Then \(M\) admits a resolution \[\ldots \to K_2 \to K_1 \to K_0 \to M \to 0\] with each \(K_i\) a direct sum of copies of \(R/I^n\) for \(n\) variable.
Proof
There is a canonical surjection \[\oplus_{m \in M} R/I^{n_m} \to M \to 0\] where \(n_m\) is the smallest positive integer such that \(I^{n_m} \cdot m = 0\). The kernel of the preceding surjection is also an \(I\)-power torsion module. Proceeding inductively, we construct the desired resolution of \(M\).
Lemma
Let \(R\) be a ring. Let \(I\) be an ideal of \(R\). For any \(R\)-module \(M\) set \(M[I^n] = \{m \in M \mid I^nm = 0\}\). If \(I\) is finitely generated then the following are equivalent
\(M[I] = 0\),
\(M[I^n] = 0\) for all \(n \geq 1\), and
if \(I = (f_1, \ldots, f_t)\), then the map \(M \to \bigoplus M_{f_i}\) is injective.
Proof
This follows from Algebra, Lemma 0565.
Lemma
Let \(R\) be a ring. Let \(I\) be a finitely generated ideal of \(R\).
For any \(R\)-module \(M\) we have \((M/M[I^\infty])[I] = 0\).
An extension of \(I\)-power torsion modules is \(I\)-power torsion.
Proof
Let \(m \in M\). If \(m\) maps to an element of \((M/M[I^\infty])[I]\) then \(Im \subset M[I^\infty]\). Write \(I = (f_1, \ldots, f_t)\). Then we see that \(f_i m \in M[I^\infty]\), i.e., \(I^{n_i}f_i m = 0\) for some \(n_i > 0\). Thus we see that \(I^Nm = 0\) with \(N = \sum n_i + 2\). Hence \(m\) maps to zero in \((M/M[I^\infty])\) which proves the first statement of the lemma.
For the second, suppose that \(0 \to M' \to M \to M'' \to 0\) is a short exact sequence of modules with \(M'\) and \(M''\) both \(I\)-power torsion modules. Then \(M[I^\infty] \supset M'\) and hence \(M/M[I^\infty]\) is a quotient of \(M''\) and therefore \(I\)-power torsion. Combined with the first statement and Lemma 05EA this implies that it is zero.
Lemma
Let \(I\) be a finitely generated ideal of a ring \(R\). The \(I\)-power torsion modules form a Serre subcategory of the abelian category \(\text{Mod}_R\), see Homology, Definition 02MO.
Proof
It is clear that a submodule and a quotient module of an \(I\)-power torsion module is \(I\)-power torsion. Moreover, the extension of two \(I\)-power torsion modules is \(I\)-power torsion by Lemma 05EB. Hence the statement of the lemma by Homology, Lemma 02MP.
Lemma
Let \(R\) be a ring and let \(I \subset R\) be a finitely generated ideal. The subcategory \(I^\infty\text{-torsion} \subset \text{Mod}_R\) depends only on the closed subset \(Z = V(I) \subset \Spec(R)\). In fact, an \(R\)-module \(M\) is \(I\)-power torsion if and only if its support is contained in \(Z\).
Proof
Let \(M\) be an \(R\)-module. Let \(x \in M\). If \(x \in M[I^\infty]\), then \(x\) maps to zero in \(M_f\) for all \(f \in I\). Hence \(x\) maps to zero in \(M_\mathfrak p\) for all \(\mathfrak p \not \supset I\). Conversely, if \(x\) maps to zero in \(M_\mathfrak p\) for all \(\mathfrak p \not \supset I\), then \(x\) maps to zero in \(M_f\) for all \(f \in I\). Hence if \(I = (f_1, \ldots, f_r)\), then \(f_i^{n_i}x = 0\) for some \(n_i \geq 1\). It follows that \(x \in M[I^{\sum n_i}]\). Thus \(M[I^\infty]\) is the kernel of \(M \to \prod_{\mathfrak p \not \in Z} M_\mathfrak p\). The second statement of the lemma follows and it implies the first.
The next two lemmas should probably go somewhere else.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(K\) be an object of \(D(R)\) such that \(H^i(K \otimes_R^\mathbf{L} R/I) = 0\) for \(i > 0\). Then
\(H^i(K \otimes_R^\mathbf{L} R/I^n) = 0\) for all \(n \geq 1\) and \(i > 0\),
\(H^i(K \otimes_R^\mathbf{L} N) = 0\) for any \(I\)-power torsion \(R\)-module \(N\) and \(i > 0\), and
for any \(M \in D^b(R)\) whose cohomology modules \(H^i(M)\) are \(I\)-power torsion and \(0\) for \(i > 0\) we have \(H^i(K \otimes_R^\mathbf{L} M) = 0\) for \(i > 0\).
Proof
Proof of (2). We can write \(N = \bigcup N[I^n]\). We have \(K \otimes_R^\mathbf{L} N = \text{hocolim}_n K \otimes_R^\mathbf{L} N[I^n]\) as tensor products commute with colimits (details omitted; hint: represent \(K\) by a K-flat complex and compute directly). Hence we may assume \(N\) is annihilated by \(I^n\). Consider the \(R\)-algebra \(R' = R/I^n \oplus N\) where \(N\) is an ideal of square zero. It suffices to show that the object \(K' = K \otimes_R^\mathbf{L} R'\) of \(D(R')\) has vanishing cohomology in positive degrees. We have a surjection \(R' \to R/I\) of \(R\)-algebras whose kernel \(J\) is nilpotent (any product of \(n\) elements in the kernel is zero). We have \[K \otimes_R^\mathbf{L} R/I = (K \otimes_R^\mathbf{L} R') \otimes_{R'}^\mathbf{L} R/I = K' \otimes_{R'}^\mathbf{L} R/I\] By assumption the complex \(K \otimes_R^\mathbf{L} R/I\) has tor-amplitude in \([-\infty, 0]\). Thus the conclusion by Lemma 0H75.
Part (1) follows trivially from part (2). Part (3) follows from part (2), induction on the number of nonzero cohomology modules of \(M\), and the distinguished triangles of truncation from Derived Categories, Remark 08J5. Details omitted.
Lemma
Let \(R\) be a ring. Let \(I \subset R\) be an ideal. Let \(K\) be an object of \(D(R)\) such that \(K \otimes_R^\mathbf{L} R/I = 0\) in \(D(R)\). Then
\(K \otimes_R^\mathbf{L} R/I^n = 0\) for all \(n \geq 1\),
\(K \otimes_R^\mathbf{L} N = 0\) for any \(I\)-power torsion \(R\)-module \(N\),
\(K \otimes_R^\mathbf{L} M = 0\) for any \(M \in D^b(R)\) whose cohomology modules are \(I\)-power torsion.
Proof
Consequence of Lemma 0H82 (some details omitted).
Lemma
Let \(R \to R'\) be a ring map. Let \(I \subset R\) be an ideal such that \(R/I^n \to R'/I^nR'\) is an isomorphism for \(n > 0\). For any \(I\)-power torsion \(R\)-module \(M\) the map \(M \to M \otimes_R R'\) is an isomorphism. For example, if \(I\) is finitely generated and \(R^\wedge\) is the completion of \(R\) with respect to \(I\), then we have \(M \cong M \otimes_R R^\wedge\).
Proof
If \(M\) is annihilated by \(I^n\), then \[M \otimes_R R' \cong M \otimes_{R/I^n} R'/I^n R' \cong M \otimes_{R/I^n} R/I^n \cong M.\] If \(M\) is \(I\)-power torsion, then \(M = \bigcup M[I^n]\). Since tensor products commute with direct limits (Algebra, Lemma 00DD), we obtain the desired isomorphism. The last statement is a special case of the first statement by Algebra, Lemma 05GG.
Formal glueing of module categories
Fix a Noetherian scheme \(X\), and a closed subscheme \(Z\) with complement \(U\). Our goal is to explain how coherent sheaves on \(X\) can be constructed (uniquely) from coherent sheaves on the formal completion of \(X\) along \(Z\), and those on \(U\) with a suitable compatibility on the overlap. We first do this using only commutative algebra (this section) and later we explain this in the setting of algebraic spaces (Pushouts of Spaces, Section 0AEP).
Here are some references treating some of the material in this section: [ArtinII, Section 2], [Ferrand-Raynaud, Appendix], [Beauville-Laszlo], [MB], and [dJ-crystalline, Section 4.6].
Lemma
Let \(\varphi : R \to S\) be a ring map. Let \(I \subset R\) be an ideal. The following are equivalent
\(\varphi\) is flat and \(R/I \to S/IS\) is faithfully flat,
\(\varphi\) is flat, and the map \(\Spec(S/IS) \to \Spec(R/I)\) is surjective.
\(\varphi\) is flat, and the base change functor \(M \mapsto M \otimes_R S\) is faithful on modules annihilated by \(I\), and
\(\varphi\) is flat, and the base change functor \(M \mapsto M \otimes_R S\) is faithful on \(I\)-power torsion modules.
Proof
If \(R \to S\) is flat, then \(R/I^n \to S/I^nS\) is flat for every \(n\), see Algebra, Lemma 00HI. Hence (1) and (2) are equivalent by Algebra, Lemma 00HQ. The equivalence of (1) with (3) follows by identifying \(I\)-torsion \(R\)-modules with \(R/I\)-modules, using that \[M \otimes_R S = M \otimes_{R/I} S/IS\] for \(R\)-modules \(M\) annihilated by \(I\), and Algebra, Lemma 00HO. The implication (4) \(\Rightarrow\) (3) is immediate. Assume (3). We have seen above that \(R/I^n \to S/I^nS\) is flat, and by assumption it induces a surjection on spectra, as \(\Spec(R/I^n) = \Spec(R/I)\) and similarly for \(S\). Hence the base change functor is faithful on modules annihilated by \(I^n\). Since any \(I\)-power torsion module \(M\) is the union \(M = \bigcup M_n\) where \(M_n\) is annihilated by \(I^n\) we see that the base change functor is faithful on the category of all \(I\)-power torsion modules (as tensor product commutes with colimits).
Lemma
Assume \((\varphi : R \to S, I)\) satisfies the equivalent conditions of Lemma 05E7. The following are equivalent
for any \(I\)-power torsion module \(M\), the natural map \(M \to M \otimes_R S\) is an isomorphism, and
\(R/I \to S/IS\) is an isomorphism.
Proof
The implication (1) \(\Rightarrow\) (2) is immediate. Assume (2). First assume that \(M\) is annihilated by \(I\). In this case, \(M\) is an \(R/I\)-module. Hence, we have an isomorphism \[M \otimes_R S = M \otimes_{R/I} S/IS = M \otimes_{R/I} R/I = M\] proving the claim. Next we prove by induction that \(M \to M \otimes_R S\) is an isomorphism for any module \(M\) is annihilated by \(I^n\). Assume the induction hypothesis holds for \(n\) and assume \(M\) is annihilated by \(I^{n + 1}\). Then we have a short exact sequence \[0 \to I^nM \to M \to M/I^nM \to 0\] and as \(R \to S\) is flat this gives rise to a short exact sequence \[0 \to I^nM \otimes_R S \to M \otimes_R S \to M/I^nM \otimes_R S \to 0\] Using that the canonical map is an isomorphism for \(M' = I^nM\) and \(M'' = M/I^nM\) (by induction hypothesis) we conclude the same thing is true for \(M\). Finally, suppose that \(M\) is a general \(I\)-power torsion module. Then \(M = \bigcup M_n\) where \(M_n\) is annihilated by \(I^n\) and we conclude using that tensor products commute with colimits.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and \(I \subset R\) is a finitely generated ideal such that \(R/I \to S/IS\) is an isomorphism. Then
for any \(R\)-module \(M\) the map \(M \to M \otimes_R S\) induces an isomorphism \(M[I^\infty] \to (M \otimes_R S)[(IS)^\infty]\) of \(I\)-power torsion submodules,
the natural map \[\Hom_R(M, N) \longrightarrow \Hom_S(M \otimes_R S, N \otimes_R S)\] is an isomorphism if either \(M\) or \(N\) is \(I\)-power torsion, and
the base change functor \(M \mapsto M \otimes_R S\) defines an equivalence of categories between \(I\)-power torsion modules and \(IS\)-power torsion modules.
Proof
Note that the equivalent conditions of both Lemma 05E7 and Lemma 05E9 are satisfied. We will use these without further mention. We first prove (1). Let \(M\) be any \(R\)-module. Set \(M' = M/M[I^\infty]\) and consider the exact sequence \[0 \to M[I^\infty] \to M \to M' \to 0\] As \(M[I^\infty] = M[I^\infty] \otimes_R S\) we see that it suffices to show that \((M' \otimes_R S)[(IS)^\infty] = 0\). Write \(I = (f_1, \ldots, f_t)\). By Lemma 05EB we see that \(M'[I^\infty] = 0\). Hence for every \(n > 0\) the map \[M' \longrightarrow \bigoplus\nolimits_{i = 1, \ldots t} M', \quad x \longmapsto (f_1^n x, \ldots, f_t^n x)\] is injective. As \(S\) is flat over \(R\) also the corresponding map \(M' \otimes_R S \to \bigoplus_{i = 1, \ldots t} M' \otimes_R S\) is injective. This means that \((M' \otimes_R S)[I^n] = 0\) as desired.
Next we prove (2). If \(N\) is \(I\)-power torsion, then \(N \otimes_R S = N\) and the displayed map of (2) is an isomorphism by Algebra, Lemma 05DQ. If \(M\) is \(I\)-power torsion, then the image of any map \(M \to N\) factors through \(N[I^\infty]\) and the image of any map \(M \otimes_R S \to N \otimes_R S\) factors through \((N \otimes_R S)[(IS)^\infty]\). Hence in this case part (1) guarantees that we may replace \(N\) by \(N[I^\infty]\) and the result follows from the case where \(N\) is \(I\)-power torsion we just discussed.
Next we prove (3). The functor is fully faithful by (2). For essential surjectivity, we simply note that for any \(IS\)-power torsion \(S\)-module \(N\), the natural map \(N \otimes_R S \to N\) is an isomorphism.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and \(I \subset R\) is a finitely generated ideal such that \(R/I \to S/IS\) is an isomorphism. For any \(f_1, \ldots, f_r \in R\) such that \(V(f_1, \ldots, f_r) = V(I)\)
the map of Koszul complexes \(K(R, f_1, \ldots, f_r) \to K(S, f_1, \ldots, f_r)\) is a quasi-isomorphism, and
The map of extended alternating Čech complexes \[\xymatrix{ R \to \prod_{i_0} R_{f_{i_0}} \to \prod_{i_0 < i_1} R_{f_{i_0}f_{i_1}} \to \ldots \to R_{f_1\ldots f_r} \ar[d] \\ S \to \prod_{i_0} S_{f_{i_0}} \to \prod_{i_0 < i_1} S_{f_{i_0}f_{i_1}} \to \ldots \to S_{f_1\ldots f_r} }\] is a quasi-isomorphism.
Proof
In both cases we have a complex \(K_\bullet\) of \(R\) modules and we want to show that \(K_\bullet \to K_\bullet \otimes_R S\) is a quasi-isomorphism. By Lemma 05E9 and the flatness of \(R \to S\) this will hold as soon as all homology groups of \(K\) are \(I\)-power torsion. This is true for the Koszul complex by Lemma 0663 and for the extended alternating Čech complex by Lemma 0G6K.
Lemma
Let \(R\) be a ring. Let \(I = (f_1, \ldots, f_n)\) be a finitely generated ideal of \(R\). Let \(M\) be the \(R\)-module generated by elements \(e_1, \ldots, e_n\) subject to the relations \(f_i e_j - f_j e_i = 0\). There exists a short exact sequence \[0 \to K \to M \to I \to 0\] such that \(K\) is annihilated by \(I\).
Proof
This is just a truncation of the Koszul complex. The map \(M \to I\) is determined by the rule \(e_i \mapsto f_i\). If \(m = \sum a_i e_i\) is in the kernel of \(M \to I\), i.e., \(\sum a_i f_i = 0\), then \(f_j m = \sum f_j a_i e_i = (\sum f_i a_i) e_j = 0\).
Lemma
Let \(R\) be a ring. Let \(I = (f_1, \ldots, f_n)\) be a finitely generated ideal of \(R\). For any \(R\)-module \(N\) set \[H_1(N, f_\bullet) = \frac{\{(x_1, \ldots, x_n) \in N^{\oplus n} \mid f_i x_j = f_j x_i \}} {\{f_1x, \ldots, f_nx) \mid x \in N\}}\] For any \(R\)-module \(N\) there exists a canonical short exact sequence \[0 \to \Ext_R(R/I, N) \to H_1(N, f_\bullet) \to \Hom_R(K, N)\] where \(K\) is as in Lemma 05ED.
Proof
The notation above indicates the \(\Ext\)-groups in \(\text{Mod}_R\) as defined in Homology, Section 010I. These are denoted \(\Ext_R(M, N)\). Using the long exact sequence of Homology, Lemma 05E2 associated to the short exact sequence \(0 \to I \to R \to R/I \to 0\) and the fact that \(\Ext_R(R, N) = 0\) we see that \[\Ext_R(R/I, N) = \Coker(N \longrightarrow \Hom(I, N))\] Using the short exact sequence of Lemma 05ED we see that we get a complex \[N \to \Hom(M, N) \to \Hom_R(K, N)\] whose homology in the middle is canonically isomorphic to \(\Ext_R(R/I, N)\). The proof of the lemma is now complete as the cokernel of the first map is canonically isomorphic to \(H_1(N, f_\bullet)\).
Lemma
Let \(R\) be a ring. Let \(I = (f_1, \ldots, f_n)\) be a finitely generated ideal of \(R\). For any \(R\)-module \(N\) the Koszul homology group \(H_1(N, f_\bullet)\) defined in Lemma 05EE is annihilated by \(I\).
Proof
Let \((x_1, \ldots, x_n) \in N^{\oplus n}\) with \(f_i x_j = f_j x_i\). Then we have \(f_i(x_1, \ldots, x_n) = (f_i x_i, \ldots, f_i x_n)\). In other words \(f_i\) annihilates \(H_1(N, f_\bullet)\).
We can improve on the full faithfulness of Lemma 05EC by showing that \(\Ext\)-groups whose source is \(I\)-power torsion are insensitive to passing to \(S\) as well. See Dualizing Complexes, Lemma 05EH for a derived version of the following lemma.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and \(I \subset R\) is a finitely generated ideal such that \(R/I \to S/IS\) is an isomorphism. Let \(M\), \(N\) be \(R\)-modules. Assume \(M\) is \(I\)-power torsion. Given an short exact sequence \[0 \to N \otimes_R S \to \tilde E \to M \otimes_R S \to 0\] there exists a commutative diagram \[\xymatrix{ 0 \ar[r] & N \ar[r] \ar[d] & E \ar[r] \ar[d] & M \ar[r] \ar[d] & 0 \\ 0 \ar[r] & N \otimes_R S \ar[r] & \tilde E \ar[r] & M \otimes_R S \ar[r] & 0 }\] with exact rows.
Proof
As \(M\) is \(I\)-power torsion we see that \(M \otimes_R S = M\), see Lemma 05E9. We will use this identification without further mention. As \(R \to S\) is flat, the base change functor is exact and we obtain a functorial map of \(\Ext\)-groups \[\Ext_R(M, N) \longrightarrow \Ext_S(M \otimes_R S, N \otimes_R S),\] see Homology, Lemma 010O. The claim of the lemma is that this map is surjective when \(M\) is \(I\)-power torsion. In fact we will show that it is an isomorphism. By Lemma 05E8 we can find a surjection \(M' \to M\) with \(M'\) a direct sum of modules of the form \(R/I^n\). Using the long exact sequence of Homology, Lemma 05E2 and using Lemma 05EC we see that it suffices to prove the lemma for \(M'\). Using compatibility of \(\Ext\) with direct sums (details omitted) we reduce to the case where \(M = R/I^n\) for some \(n\).
Let \(f_1, \ldots, f_t\) be generators for \(I^n\). By Lemma 05EE we have a commutative diagram \[\xymatrix{ 0 \ar[r] & \Ext_R(R/I^n, N) \ar[r] \ar[d] & H_1(N, f_\bullet) \ar[r] \ar[d] & \Hom_R(K, N) \ar[d] \\ 0 \ar[r] & \Ext_S(S/I^nS, N \otimes S) \ar[r] & H_1(N \otimes S, f_\bullet) \ar[r] & \Hom_S(K \otimes S, N \otimes S) }\] with exact rows where \(K\) is as in Lemma 05ED. Hence it suffices to prove that the two right vertical arrows are isomorphisms. Since \(K\) is annihilated by \(I^n\) we see that \(\Hom_R(K, N) = \Hom_S(K \otimes_R S, N \otimes_R S)\) by Lemma 05EC. As \(R \to S\) is flat we have \(H_1(N, f_\bullet) \otimes_R S = H_1(N \otimes_R S, f_\bullet)\). As \(H_1(N, f_\bullet)\) is annihilated by \(I^n\), see Lemma 05EF we have \(H_1(N, f_\bullet) \otimes_R S = H_1(N, f_\bullet)\) by Lemma 05E9.
Let \(R \to S\) be a ring map. Let \(f_1, \ldots, f_t \in R\) and \(I = (f_1, \ldots, f_t)\). Then for any \(R\)-module \(M\) we can define a complex [05EJ]\[\begin{equation} 0 \to M \xrightarrow{\alpha} M \otimes_R S \times \prod M_{f_i} \xrightarrow{\beta} \prod (M \otimes_R S)_{f_i} \times \prod M_{f_if_j} \end{equation}\] where \(\alpha(m) = (m \otimes 1, m/1, \ldots, m/1)\) and \[\beta(m', m_1, \ldots, m_t) = ((m'/1 - m_1 \otimes 1, \ldots, m'/1 - m_t \otimes 1), (m_1 - m_2, \ldots, m_{t - 1} - m_t).\] We would like to know when this complex is exact.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and \(I = (f_1, \ldots, f_t) \subset R\) is an ideal such that \(R/I \to S/IS\) is an isomorphism. Let \(M\) be an \(R\)-module. Then the complex (05EJ) is exact.
Proof
First proof. Denote \(\check{\mathcal{C}}_R \to \check{\mathcal{C}}_S\) the quasi-isomorphism of extended alternating Čech complexes of Lemma 091M. Since these complexes are bounded with flat terms, we see that \(M \otimes_R \check{\mathcal{C}}_R \to M \otimes_R \check{\mathcal{C}}_S\) is a quasi-isomorphism too (Lemmas 064K and 064L). Now the complex (05EJ) is a truncation of the cone of the map \(M \otimes_R \check{\mathcal{C}}_R \to M \otimes_R \check{\mathcal{C}}_S\) and we win.
Second computational proof. Let \(m \in M\). If \(\alpha(m) = 0\), then \(m \in M[I^\infty]\), see Lemma 05EA. Pick \(n\) such that \(I^n m = 0\) and consider the map \(\varphi : R/I^n \to M\). If \(m \otimes 1 = 0\), then \(\varphi \otimes 1_S = 0\), hence \(\varphi = 0\) (see Lemma 05EC) hence \(m = 0\). In this way we see that \(\alpha\) is injective.
Let \((m', m'_1, \ldots, m'_t) \in \Ker(\beta)\). Write \(m'_i = m_i/f_i^n\) for some \(n > 0\) and \(m_i \in M\). We may, after possibly enlarging \(n\) assume that \(f_i^n m' = m_i \otimes 1\) in \(M \otimes_R S\) and \(f_j^nm_i - f_i^nm_j = 0\) in \(M\). In particular we see that \((m_1, \ldots, m_t)\) defines an element \(\xi\) of \(H_1(M, (f_1^n, \ldots, f_t^n))\). Since \(H_1(M, (f_1^n, \ldots, f_t^n))\) is annihilated by \(I^{tn + 1}\) (see Lemma 05EF) and since \(R \to S\) is flat we see that \[H_1(M, (f_1^n, \ldots, f_t^n)) = H_1(M, (f_1^n, \ldots, f_t^n)) \otimes_R S = H_1(M \otimes_R S, (f_1^n, \ldots, f_t^n))\] by Lemma 05E9 The existence of \(m'\) implies that \(\xi\) maps to zero in the last group, i.e., the element \(\xi\) is zero. Thus there exists an \(m \in M\) such that \(m_i = f_i^n m\). Then \((m', m'_1, \ldots, m'_t) - \alpha(m) = (m'', 0, \ldots, 0)\) for some \(m'' \in (M \otimes_R S)[(IS)^\infty]\). By Lemma 05EC we conclude that \(m'' \in M[I^\infty]\) and we win.
Remark
In this remark we define a category of glueing data. Let \(R \to S\) be a ring map. Let \(f_1, \ldots, f_t \in R\) and \(I = (f_1, \ldots, f_t)\). Consider the category \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) as the category whose
objects are systems \((M', M_i, \alpha_i, \alpha_{ij})\), where \(M'\) is an \(S\)-module, \(M_i\) is an \(R_{f_i}\)-module, \(\alpha_i : (M')_{f_i} \to M_i \otimes_R S\) is an isomorphism, and \(\alpha_{ij} : (M_i)_{f_j} \to (M_j)_{f_i}\) are isomorphisms such that
\(\alpha_{ij} \circ \alpha_i = \alpha_j\) as maps \((M')_{f_if_j} \to (M_j)_{f_i}\), and
\(\alpha_{jk} \circ \alpha_{ij} = \alpha_{ik}\) as maps \((M_i)_{f_jf_k} \to (M_k)_{f_if_j}\) (cocycle condition).
morphisms \((M', M_i, \alpha_i, \alpha_{ij}) \to (N', N_i, \beta_i, \beta_{ij})\) are given by maps \(\varphi' : M' \to N'\) and \(\varphi_i : M_i \to N_i\) compatible with the given maps \(\alpha_i, \beta_i, \alpha_{ij}, \beta_{ij}\).
There is a canonical functor \[\text{Can} : \text{Mod}_R \longrightarrow \text{Glue}(R \to S, f_1, \ldots, f_t), \quad M \longmapsto (M \otimes_R S, M_{f_i}, \text{can}_i, \text{can}_{ij})\] where \(\text{can}_i : (M \otimes_R S)_{f_i} \to M_{f_i} \otimes_R S\) and \(\text{can}_{ij} : (M_{f_i})_{f_j} \to (M_{f_j})_{f_i}\) are the canonical isomorphisms. For any object \(\mathbf{M} = (M', M_i, \alpha_i, \alpha_{ij})\) of the category \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) we define \[H^0(\mathbf{M}) = \{(m', m_i) \mid \alpha_i(m') = m_i \otimes 1, \alpha_{ij}(m_i) = m_j\}\] in other words defined by the exact sequence \[0 \to H^0(\mathbf{M}) \to M' \times \prod M_i \to \prod M'_{f_i} \times \prod (M_i)_{f_j}\] similar to (05EJ). We think of \(H^0(\mathbf{M})\) as an \(R\)-module. Thus we also get a functor \[H^0 : \text{Glue}(R \to S, f_1, \ldots, f_t) \longrightarrow \text{Mod}_R\] Our next goal is to show that the functors \(\text{Can}\) and \(H^0\) are sometimes quasi-inverse to each other.
Lemma
In Remark 05EL the functor \(H^0 : \text{Glue}(R \to S, f_1, \ldots, f_t) \to \text{Mod}_R\) is a right adjoint to the functor \(\text{Can} : \text{Mod}_R \to \text{Glue}(R \to S, f_1, \ldots, f_t)\).
Proof
Let \(\mathbf{M} = (M', M_i, \alpha_i, \alpha_{ij})\) be an object of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\). For any \(R\)-module \(N\) there is a map \[\Hom_{\text{Glue}(R \to S, f_1, \ldots, f_t)}(\text{Can}(N), \mathbf{M}) \to \Hom_R(N, H^0(\mathbf{M}))\] sending \(\psi\) to \(H^0(\psi)\) composed with the obvious map \(N \to H^0(\text{Can}(N))\). By construction the displayed map is an isomorphism for \(N = R\) (even if \(R \to H^0(\text{Can}(R))\) is not an isomorphism in general). The category \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) has direct sums and cokernels. The functor \(\text{Can}\) commutes with direct sums and cokernels. From these observations we find the displayed map is bijective by writing \(N\) as a cokernel of a map between free \(R\)-modules. We omit the details.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and \(I = (f_1, \ldots, f_t) \subset R\) is an ideal such that \(R/I \to S/IS\) is an isomorphism. Then the functor \(H^0\) is a left quasi-inverse to the functor \(\text{Can}\) of Remark 05EL.
Proof
This is a reformulation of Lemma 05EK.
Lemma
Assume \(\varphi : R \to S\) is a flat ring map and let \(I = (f_1, \ldots, f_t) \subset R\) be an ideal. Then \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) is an abelian category, and the functor \(\text{Can}\) is exact and commutes with arbitrary colimits.
Proof
Given a morphism \((\varphi', \varphi_i) : (M', M_i, \alpha_i, \alpha_{ij}) \to (N', N_i, \beta_i, \beta_{ij})\) of the category \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) we see that its kernel exists and is equal to the object \((\Ker(\varphi'), \Ker(\varphi_i), \alpha_i, \alpha_{ij})\) and its cokernel exists and is equal to the object \((\Coker(\varphi'), \Coker(\varphi_i), \beta_i, \beta_{ij})\). This works because \(R \to S\) is flat, hence taking kernels/cokernels commutes with \(- \otimes_R S\). Details omitted. The exactness follows from the \(R\)-flatness of \(R_{f_i}\) and \(S\), while commuting with colimits follows as tensor products commute with colimits.
Lemma
Let \(\varphi : R \to S\) be a flat ring map and \((f_1, \ldots, f_t) = R\). Then \(\text{Can}\) and \(H^0\) are quasi-inverse equivalences of categories \[\text{Mod}_R = \text{Glue}(R \to S, f_1, \ldots, f_t)\]
Proof
Consider an object \(\mathbf{M} = (M', M_i, \alpha_i, \alpha_{ij})\) of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\). By Algebra, Lemma 00EQ there exists a unique module \(M\) and isomorphisms \(M_{f_i} \to M_i\) which recover the glueing data \(\alpha_{ij}\). Then both \(M'\) and \(M \otimes_R S\) are \(S\)-modules which recover the modules \(M_i \otimes_R S\) upon localizing at \(f_i\). Whence there is a canonical isomorphism \(M \otimes_R S \to M'\). This shows that \(\mathbf{M}\) is in the essential image of \(\text{Can}\). Combined with Lemma 05EM the lemma follows.
Lemma
Let \(\varphi : R \to S\) be a flat ring map and \(I = (f_1, \ldots, f_t)\) and ideal. Let \(R \to R'\) be a flat ring map, and set \(S' = S \otimes_R R'\). Then we obtain a commutative diagram of categories and functors \[\xymatrix{ \text{Mod}_R \ar[r]_-{\text{Can}} \ar[d]_{-\otimes_R R'} & \text{Glue}(R \to S, f_1, \ldots, f_t) \ar[r]_-{H^0} \ar[d]^{-\otimes_R R'} & \text{Mod}_R \ar[d]^{-\otimes_R R'} \\ \text{Mod}_{R'} \ar[r]^-{\text{Can}} & \text{Glue}(R' \to S', f_1, \ldots, f_t) \ar[r]^-{H^0} & \text{Mod}_{R'} }\]
Proof
Omitted.
Proposition
Assume \(\varphi : R \to S\) is a flat ring map and \(I = (f_1, \ldots, f_t) \subset R\) is an ideal such that \(R/I \to S/IS\) is an isomorphism. Then \(\text{Can}\) and \(H^0\) are quasi-inverse equivalences of categories \[\text{Mod}_R = \text{Glue}(R \to S, f_1, \ldots, f_t)\]
Proof
We have already seen that \(H^0 \circ \text{Can}\) is isomorphic to the identity functor, see Lemma 05EM. Consider an object \(\mathbf{M} = (M', M_i, \alpha_i, \alpha_{ij})\) of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\). We get a natural morphism \[\Psi : (H^0(\mathbf{M}) \otimes_R S, H^0(\mathbf{M})_{f_i}, \text{can}_i, \text{can}_{ij}) \longrightarrow (M', M_i, \alpha_i, \alpha_{ij}).\] Namely, by definition \(H^0(\mathbf{M})\) comes equipped with compatible \(R\)-module maps \(H^0(\mathbf{M}) \to M'\) and \(H^0(\mathbf{M}) \to M_i\). We have to show that this map is an isomorphism.
Pick an index \(i\) and set \(R' = R_{f_i}\). Combining Lemmas 05EQ and 05EP we see that \(\Psi \otimes_R R'\) is an isomorphism. Hence the kernel, resp. cokernel of \(\Psi\) is a system of the form \((K, 0, 0, 0)\), resp. \((Q, 0, 0, 0)\). Note that \(H^0((K, 0, 0, 0)) = K\), that \(H^0\) is left exact, and that by construction \(H^0(\Psi)\) is bijective. Hence we see \(K = 0\), i.e., the kernel of \(\Psi\) is zero.
The conclusion of the above is that we obtain a short exact sequence \[0 \to H^0(\mathbf{M}) \otimes_R S \to M' \to Q \to 0\] and that \(M_i = H^0(\mathbf{M})_{f_i}\). Note that we may think of \(Q\) as an \(R\)-module which is \(I\)-power torsion so that \(Q = Q \otimes_R S\). By Lemma 05EG we see that there exists a commutative diagram \[\xymatrix{ 0 \ar[r] & H^0(\mathbf{M}) \ar[r] \ar[d] & E \ar[r] \ar[d] & Q \ar[r] \ar[d] & 0 \\ 0 \ar[r] & H^0(\mathbf{M}) \otimes_R S \ar[r] & M' \ar[r] & Q \ar[r] & 0 }\] with exact rows. This clearly determines an isomorphism \(\text{Can}(E) \to (M', M_i, \alpha_i, \alpha_{ij})\) in the category \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) and we win. (Of course, a posteriori we have \(Q = 0\).)
Lemma
Let \(\varphi : R \to S\) be a flat ring map and let \(I \subset R\) be a finitely generated ideal such that \(R/I \to S/IS\) is an isomorphism.
Given an \(R\)-module \(N\), an \(S\)-module \(M'\) and an \(S\)-module map \(\varphi : M' \to N \otimes_R S\) whose kernel and cokernel are \(I\)-power torsion, there exists an \(R\)-module map \(\psi : M \to N\) and an isomorphism \(M \otimes_R S = M'\) compatible with \(\varphi\) and \(\psi\).
Given an \(R\)-module \(M\), an \(S\)-module \(N'\) and an \(S\)-module map \(\varphi : M \otimes_R S \to N'\) whose kernel and cokernel are \(I\)-power torsion, there exists an \(R\)-module map \(\psi : M \to N\) and an isomorphism \(N \otimes_R S = N'\) compatible with \(\varphi\) and \(\psi\).
In both cases we have \(\Ker(\varphi) \cong \Ker(\psi)\) and \(\Coker(\varphi) \cong \Coker(\psi)\).
Proof
Proof of (1). Say \(I = (f_1, \ldots, f_t)\). It is clear that the localization \(\varphi_{f_i}\) is an isomorphism. Thus we see that \((M', N_{f_i}, \varphi_{f_i}, can_{ij})\) is an object of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\), see Remark 05EL. By Proposition 05ER we conclude that there exists an \(R\)-module \(M\) such that \(M' = M \otimes_R S\) and \(N_{f_i} = M_{f_i}\) compatibly with the isomorphisms \(\varphi_{f_i}\) and \(can_{ij}\). There is a morphism \[(M \otimes_R S, M_{f_i}, can_i, can_{ij}) = (M', N_{f_i}, \varphi_{f_i}, can_{ij}) \to (N \otimes_R S, N_{f_i}, can_i, can_{ij})\] of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) which uses \(\varphi\) in the first component. This corresponds to an \(R\)-module map \(\psi : M \to N\) (by the equivalence of categories of Proposition 05ER). The composition of the base change of \(M \to N\) with the isomorphism \(M' \cong M \otimes_R S\) is \(\varphi\), in other words \(M \to N\) is compatible with \(\varphi\).
Proof of (2). This is just the dual of the argument above. Namely, the localization \(\varphi_{f_i}\) is an isomorphism. Thus we see that \((N', M_{f_i}, \varphi_{f_i}^{-1}, can_{ij})\) is an object of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\), see Remark 05EL. By Proposition 05ER we conclude that there exists an \(R\)-module \(N\) such that \(N' = N \otimes_R S\) and \(N_{f_i} = M_{f_i}\) compatibly with the isomorphisms \(\varphi_{f_i}^{-1}\) and \(can_{ij}\). There is a morphism \[(M \otimes_R S, M_{f_i}, can_i, can_{ij}) \to (N', M_{f_i}, \varphi_{f_i}, can_{ij}) = (N \otimes_R S, N_{f_i}, can_i, can_{ij})\] of \(\text{Glue}(R \to S, f_1, \ldots, f_t)\) which uses \(\varphi\) in the first component. This corresponds to an \(R\)-module map \(\psi : M \to N\) (by the equivalence of categories of Proposition 05ER). The composition of the base change of \(M \to N\) with the isomorphism \(N' \cong N \otimes_R S\) is \(\varphi\), in other words \(M \to N\) is compatible with \(\varphi\).
The final statement follows for example from Lemma 05EC.
Next, we specialize Proposition 05ER to get something more usable. Namely, if \(I = (f)\) is a principal ideal then the objects of \(\text{Glue}(R \to S, f)\) are simply triples \((M', M_1, \alpha_1)\) and there is no cocycle condition to check!
Theorem
Let \(R\) be a ring, and let \(f \in R\). Let \(\varphi : R \to S\) be a flat ring map inducing an isomorphism \(R/fR \to S/fS\). Then the functor \[\text{Mod}_R \longrightarrow \text{Mod}_S \times_{\text{Mod}_{S_f}} \text{Mod}_{R_f}, \quad M \longmapsto (M \otimes_R S, M_f, \text{can})\] is an equivalence.
Proof
The category appearing on the right side of the arrow is the category of triples \((M', M_1, \alpha_1)\) where \(M'\) is an \(S\)-module, \(M_1\) is a \(R_f\)-module, and \(\alpha_1 : M'_f \to M_1 \otimes_R S\) is a \(S_f\)-isomorphism, see Categories, Example 003R. Hence this theorem is a special case of Proposition 05ER.
A useful special case of Theorem 05ES is when \(R\) is Noetherian, and \(S\) is a completion of \(R\) at an element \(f\). The completion \(R \to S\) is flat, and the functor \(M \mapsto M \otimes_R S\) can be identified with the \(f\)-adic completion functor when \(M\) is finitely generated. To state this more precisely, let \(\text{Mod}^{fg}_R\) denote the category of finitely generated \(R\)-modules.
Proposition
Let \(R\) be a Noetherian ring. Let \(f \in R\) be an element. Let \(R^\wedge\) be the \(f\)-adic completion of \(R\). Then the functor \(M \mapsto (M^\wedge, M_f, \text{can})\) defines an equivalence \[\text{Mod}^{fg}_R \longrightarrow \text{Mod}^{fg}_{R^\wedge} \times_{\text{Mod}^{fg}_{(R^\wedge)_f}} \text{Mod}^{fg}_{R_f}\]
Proof
The ring map \(R \to R^\wedge\) is flat by Algebra, Lemma 00MB. It is clear that \(R/fR = R^\wedge/fR^\wedge\). By Algebra, Lemma 00MA the completion of a finite \(R\)-module \(M\) is equal to \(M \otimes_R R^\wedge\). Hence the displayed functor of the proposition is equal to the functor occurring in Theorem 05ES. In particular it is fully faithful. Let \((M_1, M_2, \psi)\) be an object of the right hand side. By Theorem 05ES there exists an \(R\)-module \(M\) such that \(M_1 = M \otimes_R R^\wedge\) and \(M_2 = M_f\). As \(R \to R^\wedge \times R_f\) is faithfully flat we conclude from Algebra, Lemma 00EO that \(M\) is finitely generated, i.e., \(M \in \text{Mod}^{fg}_R\). This proves the proposition.
Remark
The equivalences of Proposition 05ER, Theorem 05ES, and Proposition 05ET preserve properties of modules. For example if \(M\) corresponds to \(\mathbf{M} = (M', M_i, \alpha_i, \alpha_{ij})\) then \(M\) is finite, or finitely presented, or flat, or projective over \(R\) if and only if \(M'\) and \(M_i\) have the corresponding property over \(S\) and \(R_{f_i}\). This follows from the fact that \(R \to S \times \prod R_{f_i}\) is faithfully flat and descend and ascent of these properties along faithfully flat maps, see Algebra, Lemma 03C4 and Theorem 05A9. These functors also preserve the \(\otimes\)-structures on either side. Thus, it defines equivalences of various categories built out of the pair \((\text{Mod}_R, \otimes)\), such as the category of algebras.
Remark
Given a differential manifold \(X\) with a compact closed submanifold \(Z\) having complement \(U\), specifying a sheaf on \(X\) is the same as specifying a sheaf on \(U\), a sheaf on an unspecified tubular neighbourhood \(T\) of \(Z\) in \(X\), and an isomorphism between the two resulting sheaves along \(T \cap U\). Tubular neighbourhoods do not exist in algebraic geometry as such, but results such as Proposition 05ER, Theorem 05ES, and Proposition 05ET allow us to work with formal neighbourhoods instead.
The Beauville-Laszlo theorem
Let \(R\) be a ring and let \(f\) be an element of \(R\). Denote \(R^\wedge = \lim R/f^n R\) the \(f\)-adic completion of \(R\). In this section we discuss and slightly generalize a theorem of Beauville and Laszlo, see [Beauville-Laszlo]. The theorem asserts that under suitable conditions, a module over \(R\) can be constructed by “glueing together” modules over \(R^\wedge\) and \(R_f\) along an isomorphism between the base extensions to \((R^\wedge)_f\).
In [Beauville-Laszlo] it is assumed that \(f\) is a nonzerodivisor on both \(R\) and \(M\). In fact, one only needs to assume that \[R[f^\infty] \longrightarrow R^\wedge[f^\infty]\] is bijective and that \[M[f^\infty] \longrightarrow M \otimes_R R^\wedge\] is injective. This optimization was partly inspired by an alternate approach to glueing introduced in [Kedlaya-Liu-I, §1.3] for use in the theory of nonarchimedean analytic spaces.
In fact, we will establish the Beauville-Laszlo theorem in the more general setting of a ring map \[R \longrightarrow R'\] which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for every \(n > 0\) and an isomorphism \(R[f^\infty] \to R'[f^\infty]\). This is better suited for globalizing and does not formally follow from the case when \(R'\) is the completion of \(R\) because, for instance, the condition that \(R[f^\infty] \to R'[f^\infty]\) is a bijection does not imply that \(R[f^\infty] \to R^\wedge[f^\infty]\) is a bijection.
The theorem of Beauville and Laszlo as proved in this section can be viewed as a non-flat version of Theorem 05ES and in the case where \(R' = R^\wedge\) can be viewed as a non-Noetherian version of Proposition 05ET. For a comparison with flat descent, please see Remark 0BNP.
One can establish even stronger results (without imposing restrictions on \(M\) for example) but for this one must work at the level of derived categories. See [Bhatt-Algebraize, §5] for more details.
Lemma
Let \(R\) be a ring and let \(f \in R\). For every positive integer \(n\) the map \(R/f^nR \to R^\wedge/f^n R^\wedge\) is an isomorphism.
Proof
This is a special case of Algebra, Lemma 05GG.
We will use the notation introduced in Section 0ALX. Thus for an \(R\)-module \(M\), we denote \(M[f^n]\) the submodule of \(M\) annihilated by \(f^n\) and we put \[M[f^\infty] = \bigcup\nolimits_{n = 1}^\infty M[f^n] = \Ker(M \to M_f).\] If \(M = M[f^\infty]\), we say that \(M\) is an \(f\)-power torsion module.
Lemma
Let \(R\) be a ring, let \(f \in R\), and let \(R \to R'\) be a ring map which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). The \(R\)-module \(R' \oplus R_f\) is faithful: for every nonzero \(R\)-module \(M\), the module \(M \otimes_R (R' \oplus R_f)\) is also nonzero. For example, if \(M\) is nonzero, then \(M \otimes_R (R^\wedge \oplus R_f)\) is nonzero.
However, the map \(M \to M \otimes_R (R' \oplus R_f)\) need not be injective; see Example 0BNU.
Proof
If \(M \neq 0\) but \(M \otimes_R R_f = 0\), then \(M\) is \(f\)-power torsion. By Lemma 0BNK we find that \(M \otimes_R R' \cong M \neq 0\). The last statement is a special case of the first statement by Lemma 0BNJ.
Lemma
Let \(R\) be a ring, let \(f \in R\), and let \(R \to R'\) be a ring map which induces an isomorphism \(R/fR \to R'/fR'\). The map \(\Spec(R') \amalg \Spec(R_f) \to \Spec(R)\) is surjective. For example, the map \(\Spec(R^\wedge) \amalg \Spec(R_f) \to \Spec(R)\) is surjective.
Proof
Recall that \(\Spec(R) = V(f) \amalg D(f)\) where \(V(f) = \Spec(R/fR)\) and \(D(f) = \Spec(R_f)\), see Algebra, Section 00DY and especially Lemmas 00E5 and 00E4. Thus the lemma follows as the map \(R \to R/fR\) factors through \(R'\). The last statement is a special case of the first statement by Lemma 0BNJ.
Lemma
Let \(R\) be a ring, let \(f \in R\), and let \(R \to R'\) be a ring map which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). An \(R\)-module \(M\) is finitely generated if and only if the (\(R' \oplus R_f\))-module \(M \otimes_R (R' \oplus R_f)\) is finitely generated. For example, if \(M \otimes_R (R^\wedge \oplus R_f)\) is finitely generated as a module over \(R^\wedge \oplus R_f\), then \(M\) is a finitely generated \(R\)-module.
Proof
The ‘only if’ is clear, so we assume that \(M \otimes_R (R' \oplus R_f)\) is finitely generated. In this case, by writing each generator as a sum of simple tensors, \(M \otimes_R (R' \oplus R_f)\) admits a finite generating set consisting of elements of \(M\). That is, there exists a morphism from a finite free \(R\)-module to \(M\) whose cokernel is killed by tensoring with \(R' \oplus R_f\); we may thus deduce \(M\) is finite generated by applying Lemma 0BNL to this cokernel. The last statement is a special case of the first statement by Lemma 0BNJ.
Remark
While \(R \to R_f\) is always flat, \(R \to R^\wedge\) is typically not flat unless \(R\) is Noetherian (see Algebra, Lemma 00MB and the discussion in Examples, Section 0AL8). Consequently, we cannot in general apply faithfully flat descent as discussed in Descent, Section 023F to the morphism \(R \to R^\wedge \oplus R_f\). Moreover, even in the Noetherian case, the usual definition of a descent datum for this morphism refers to the ring \(R^\wedge \otimes_R R^\wedge\), which we will avoid considering in this section.
Glueing pairs. Let \(R \to R'\) be a ring map that induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). Consider the sequence [0F1Q]\[\begin{equation} 0 \to R \to R' \oplus R_f \to R'_f \to 0, \end{equation}\] in which the map on the right is the difference between the two canonical homomorphisms. If this sequence is exact, then we say that \((R \to R', f)\) is a glueing pair. We will say that \((R, f)\) is a glueing pair if \((R \to R^\wedge, f)\) is a glueing pair; this makes sense by Lemma 0BNJ. Thus \((R, f)\) is a glueing pair if and only if the sequence [0BNQ]\[\begin{equation} 0 \to R \to R^\wedge \oplus R_f \to (R^\wedge)_f \to 0, \end{equation}\] is exact.
Lemma
Let \(R\) be a ring, let \(f \in R\), and let \(R \to R'\) be a ring map which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). The sequence (0F1Q) is
exact on the right,
exact on the left if and only if \(R[f^\infty] \to R'[f^\infty]\) is injective, and
exact in the middle if and only if \(R[f^\infty] \to R'[f^\infty]\) is surjective.
In particular, \((R \to R', f)\) is a glueing pair if and only if \(R[f^\infty] \to R'[f^\infty]\) is bijective. For example, \((R, f)\) is a glueing pair if and only if \(R[f^\infty] \to R^\wedge[f^\infty]\) is bijective.
Proof
Let \(x \in R'_f\). Write \(x = x'/f^n\) with \(x' \in R'\). Write \(x' = x'' + f^n y\) with \(x'' \in R\) and \(y \in R'\). Then we see that \((y, -x''/f^n)\) maps to \(x\). Thus (1) holds.
Part (2) follows from the fact that \(\Ker(R \to R_f) = R[f^\infty]\).
If the sequence is exact in the middle, then elements of the form \((x, 0)\) with \(x \in R'[f^\infty]\) are in the image of the first arrow. This implies that \(R[f^\infty] \to R'[f^\infty]\) is surjective. Conversely, assume that \(R[f^\infty] \to R'[f^\infty]\) is surjective. Let \((x, y)\) be an element in the middle which maps to zero on the right. Write \(y = y'/f^n\) for some \(y' \in R\). Then we see that \(f^n x - y'\) is annihilated by some power of \(f\) in \(R'\). By assumption we can write \(f^nx - y' = z\) for some \(z \in R[f^\infty]\). Then \(y = y''/f^n\) where \(y'' = y' + z\) is in the kernel of \(R \to R/f^nR\). Hence we see that \(y\) can be represented as \(y'''/1\) for some \(y''' \in R\). Then \(x - y'''\) is in \(R'[f^\infty]\). Thus \(x - y''' = z' \in R[f^\infty]\). Then \((x, y'''/1) = (y''' + z', (y''' + z')/1)\) as desired.
The last statement of the lemma is a special case of the penultimate statement by Lemma 0BNJ.
Remark
Suppose that \(f\) is a nonzerodivisor. Then Algebra, Lemma 0BNG shows that \(f\) is a nonzerodivisor in \(R^\wedge\). Hence \((R, f)\) is a glueing pair.
Remark
If \(R \to R^\wedge\) is flat, then for each positive integer \(n\) tensoring the sequence \(0 \to R[f^n] \to R \to R\) with \(R^\wedge\) gives the sequence \(0 \to R[f^n] \otimes_R R^\wedge \to R^\wedge \to R^\wedge\). Combined with Lemma 0BNK we conclude that \(R[f^n] \to R^\wedge[f^n]\) is an isomorphism. Thus \((R, f)\) is a glueing pair. This holds in particular if \(R\) is Noetherian, see Algebra, Lemma 00MB.
Example
Let \(k\) be a field and put \[R = k[f, T_1, T_2, \ldots]/(fT_1, fT_2 - T_1, fT_3 - T_2, \ldots).\] Then \((R, f)\) is not a glueing pair because the map \(R[f^\infty] \to R^\wedge[f^\infty]\) is not injective as the image of \(T_1\) is \(f\)-divisible in \(R^\wedge\). For \[R = k[f, T_1, T_2, \ldots]/(fT_1, f^2T_2, \ldots),\] the map \(R[f^\infty] \to R^\wedge[f^\infty]\) is not surjective as the element \(T_1 + fT_2 + f^2 T_3 + \ldots\) is not in the image. In particular, by Remark 0BNT, these are both examples where \(R \to R^\wedge\) is not flat.
Glueable modules. Let \(R \to R'\) be a ring map which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). For any \(R\)-module \(M\), we may tensor (0F1Q) with \(M\) to obtain a sequence [0F1R]\[\begin{equation} 0 \to M \to (M \otimes_R R') \oplus (M \otimes_R R_f) \to M \otimes_R R'_f \to 0 \end{equation}\] Observe that \(M \otimes_R R_f = M_f\) and that \(M \otimes_R R'_f = (M \otimes_R R')_f\). If this sequence is exact, we say that \(M\) is glueable for \((R \to R', f)\). If \(R\) is a ring and \(f \in R\), then we say an \(R\)-module is glueable if \(M\) is glueable for \((R \to R^\wedge, f)\). Thus \(M\) is glueable if and only if the sequence [0BNV]\[\begin{equation} 0 \to M \to (M \otimes_R R^\wedge) \oplus (M \otimes_R R_f) \to M \otimes_R (R^\wedge)_f \to 0 \end{equation}\] is exact.
Lemma
Let \(R\) be a ring, let \(f \in R\), and let \(R \to R'\) be a ring map which induces isomorphisms \(R/f^nR \to R'/f^nR'\) for \(n > 0\). The sequence (0F1R) is
exact on the right,
exact on the left if and only if \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is injective, and
exact in the middle if and only if \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is surjective.
Thus \(M\) is glueable for \((R \to R', f)\) if and only if \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is bijective. If \((R \to R', f)\) is a glueing pair, then \(M\) is glueable for \((R \to R', f)\) if and only if \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is injective. For example, if \((R, f)\) is a glueing pair, then \(M\) is glueable if and only if \(M[f^\infty] \to (M \otimes_R R^\wedge)[f^\infty]\) is injective.
Proof
We will use the results of Lemma 0BNR without further mention. The functor \(M \otimes_R -\) is right exact (Algebra, Lemma 00DF) hence we get (1).
The kernel of \(M \to M \otimes_R R_f = M_f\) is \(M[f^\infty]\). Thus (2) follows.
If the sequence is exact in the middle, then elements of the form \((x, 0)\) with \(x \in (M \otimes_R R')[f^\infty]\) are in the image of the first arrow. This implies that \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is surjective. Conversely, assume that \(M[f^\infty] \to (M \otimes_R R')[f^\infty]\) is surjective. Let \((x, y)\) be an element in the middle which maps to zero on the right. Write \(y = y'/f^n\) for some \(y' \in M\). Then we see that \(f^n x - y'\) is annihilated by some power of \(f\) in \(M \otimes_R R'\). By assumption we can write \(f^nx - y' = z\) for some \(z \in M[f^\infty]\). Then \(y = y''/f^n\) where \(y'' = y' + z\) is in the kernel of \(M \to M/f^nM\). Hence we see that \(y\) can be represented as \(y'''/1\) for some \(y''' \in M\). Then \(x - y'''\) is in \((M \otimes_R R')[f^\infty]\). Thus \(x - y''' = z' \in M[f^\infty]\). Then \((x, y'''/1) = (y''' + z', (y''' + z')/1)\) as desired.
If \((R \to R', f)\) is a glueing pair, then (0F1R) is exact in the middle for any \(M\) by Algebra, Lemma 00DF. This gives the penultimate statement of the lemma. The final statement of the lemma follows from this and the fact that \((R, f)\) is a glueing pair if and only if \((R \to R^\wedge, f)\) is a glueing pair.
Remark
Let \((R \to R', f)\) be a glueing pair and let \(M\) be an \(R\)-module. Here are some observations which can be used to determine whether \(M\) is glueable for \((R \to R', f)\).
By Lemma 0BNW we see that \(M\) is glueable for \((R \to R^\wedge, f)\) if and only if \(M[f^\infty] \to M \otimes_R R^\wedge\) is injective. This holds if \(M[f] \to M^\wedge\) is injective, i.e., when \(M[f] \cap \bigcap_{n = 1}^\infty f^n M = 0\).
If \(\text{Tor}_1^R(M, R'_f) = 0\), then \(M\) is glueable for \((R \to R', f)\) (use Algebra, Lemma 00M0). This is equivalent to saying that \(\text{Tor}_1^R(M, R')\) is \(f\)-power torsion. In particular, any flat \(R\)-module is glueable for \((R \to R', f)\).
If \(R \to R'\) is flat, then \(\text{Tor}_1^R(M, R') = 0\) for every \(R\)-module so every \(R\)-module is glueable for \((R \to R', f)\). This holds in particular when \(R\) is Noetherian and \(R' = R^\wedge\), see Algebra, Lemma 00MB
Example
Let \(R\) be the ring of germs at \(0\) of \(C^\infty\) functions on \(\mathbf{R}\). Let \(f \in R\) be the function \(f(x) = x\). Then \(f\) is a nonzerodivisor in \(R\), so \((R, f)\) is a glueing pair and \(R^\wedge \cong \mathbf{R}[[x]]\). Let \(\varphi \in R\) be the function \(\varphi(x) = \text{exp}(-1/x^2)\). Then \(\varphi\) has zero Taylor series, so \(\varphi \in \Ker(R \to R^\wedge)\). Since \(\varphi(x) \neq 0\) for \(x \neq 0\), we see that \(\varphi\) is a nonzerodivisor in \(R\). The function \(\varphi/f\) also has zero Taylor series, so its image in \(M = R/\varphi R\) is a nonzero element of \(M[f]\) which maps to zero in \(M \otimes_R R^\wedge = R^\wedge/\varphi R^\wedge = R^\wedge\). Hence \(M\) is not glueable.
We next make some calculations of Tor groups.
Lemma
Let \((R \to R', f)\) be a glueing pair. Then \(\text{Tor}^R_1(R', f^n R) = 0\) for each \(n > 0\).
Proof
From the exact sequence \(0 \to R[f^n] \to R \to f^n R \to 0\) we see that it suffices to check that \(R[f^n] \otimes_R R' \to R'\) is injective. By Lemma 0BNK we have \(R[f^n] \otimes_R R' = R[f^n]\) and by Lemma 0BNR we see that \(R[f^n] \to R'\) is injective as \((R \to R', f)\) is a glueing pair.
Lemma
Let \((R \to R',f)\) be a glueing pair. Then \(\text{Tor}^R_1(R', R/R[f^\infty]) = 0\).
Proof
We have \(R/R[f^\infty] = \colim R/R[f^n] = \colim f^nR\). As formation of Tor groups commutes with filtered colimits (Algebra, Lemma 0BNF) we may apply Lemma 0BNZ.
Lemma
Let \((R \to R', f)\) be a glueing pair. For every \(R\)-module \(M\), we have \(\text{Tor}^R_1(R', \Coker(M \to M_f)) = 0\).
Proof
Set \(\overline{M} = M/M[f^\infty]\). Then \(\Coker(M \to M_f) \cong \Coker(\overline{M} \to \overline{M}_f)\) hence we may and do assume that \(f\) is a nonzerodivisor on \(M\). In this case \(M \subset M_f\) and \(M_f/M = \colim M/f^nM\) where the transition maps are given by multiplication by \(f\). Since formation of Tor groups commutes with colimits (Algebra, Lemma 0BNF) it suffices to show that \(\text{Tor}^R_1(R', M/f^n M) = 0\).
We first treat the case \(M = R/R[f^\infty]\). By Lemma 0BNR we have \(M \otimes_R R' = R'/R'[f^\infty]\). From the short exact sequence \(0 \to M \to M \to M/f^nM \to 0\) we obtain the exact sequence \[\xymatrix{ \text{Tor}_1^R(R', R/R[f^\infty]) \ar[r] & \text{Tor}_1^R(R', M/f^n M) \ar[r] & R'/R'[f^\infty] \ar[dll]_{f^n} \\ R'/R'[f^\infty] \ar[r] & (R'/R'[f^\infty])/(f^n (R'/R'[f^\infty])) \ar[r] & 0 }\] by Algebra, Lemma 00M0. Here the diagonal arrow is injective. Since the first group \(\text{Tor}_1^R(R', R/R[f^\infty])\) is zero by Lemma 0BP0, we deduce that \(\text{Tor}_1^R(R', M/f^nM) = 0\) as desired.
To treat the general case, choose a surjection \(F \to M\) with \(F\) a free \(R/R[f^\infty]\)-module, and form an exact sequence \[0 \to N \to F/f^n F \to M/f^n M \to 0.\] By Lemma 0BNK this sequence remains unchanged, and hence exact, upon tensoring with \(R'\). Since \(\text{Tor}^R_1(R', F/f^n F) = 0\) by the previous paragraph, we deduce that \(\text{Tor}^R_1(R', M/f^n M) = 0\) as desired.
Let \((R \to R', f)\) be a glueing pair. This means that \(R/f^nR \to R'/f^nR'\) is an isomorphism for \(n > 0\) and the sequence \[0 \to R \to R' \oplus R_f \to R_f' \to 0\] is exact. Consider the category \(\text{Glue}(R \to R', f)\) introduced in Remark 05EL. We will call an object \((M', M_1, \alpha_1)\) of \(\text{Glue}(R \to R', f)\) a glueing datum. It consists of an \(R'\)-module \(M'\), an \(R_f\)-module \(M_1\), and an isomorphism \(\alpha_1 : (M')_f \to M_1 \otimes_R R'\). There is an obvious functor \[\text{Can} : \text{Mod}_R \longrightarrow \text{Glue}(R \to R', f),\quad M \longmapsto (M \otimes_R R', M_f, \text{can}),\] and there is a functor \[H^0 : \text{Glue}(R \to R', f) \longrightarrow \text{Mod}_R,\quad (M', M_1, \alpha_1) \longmapsto \Ker(M' \oplus M_1 \to (M')_f)\] in the reverse direction, see Remark 05EL for the precise definition.
Theorem
Let \((R \to R',f)\) be a glueing pair. The functor \(\text{Can} : \text{Mod}_R \longrightarrow \text{Glue}(R \to R', f)\) determines an equivalence of the category of \(R\)-modules glueable for \((R \to R', f)\) and the category \(\text{Glue}(R \to R', f)\) of glueing data.
Proof
Let \((M', M_1, \alpha_1)\) be a glueing datum. We will show that \(M = H^0((M', M_1, \alpha_1))\) is a glueable module for \((R \to R', f)\) and that \((M', M_1, \alpha_1) \cong \text{Can}(M)\).
We first check that the map \(\text{d} : M' \oplus M_1 \to (M')_f\) used in the definition of the functor \(H^0\) is surjective. Observe that \((x, y) \in M' \oplus M_1\) maps to \(\text{d}(x, y) = x/1 - \alpha_1^{-1}(y \otimes 1)\) in \((M')_f\). If \(z \in (M')_f\), then we can write \(\alpha_1(z) = \sum y_i \otimes g_i\) with \(g_i \in R'\) and \(y_i \in M_1\). Write \(\alpha_1^{-1}(y_i \otimes 1) = y_i'/f^n\) for some \(y'_i \in M'\) and \(n \geq 0\) (we can pick the same \(n\) for all \(i\)). Write \(g_i = a_i + f^n b_i\) with \(a_i \in R\) and \(b_i \in R'\). Then with \(y = \sum a_i y_i \in M_1\) and \(x = \sum b_i y'_i \in M'\) we have \(\text{d}(x, -y) = z\) as desired.
Since \(M = H^0((M', M_1, \alpha_1)) = \Ker(\text{d})\) we obtain an exact sequence of \(R\)-modules [0BP3]\[\begin{equation} 0 \to M \to M' \oplus M_1 \to (M')_f \to 0. \end{equation}\] We will prove that the maps \(M \to M'\) and \(M \to M_1\) induce isomorphisms \(M \otimes_R R' \to M'\) and \(M \otimes_R R_f \to M_1\). This will imply that \(M\) is glueable for \((R \to R', f)\) and \(\text{Can}(M) \cong (M', M_1, \alpha_1)\) as desired.
Since \(f\) is a nonzerodivisor on \(M_1\), we have \(M[f^\infty] \cong M'[f^\infty]\). This yields an exact sequence [0BP4]\[\begin{equation} 0 \to M/M[f^\infty] \to M_1 \to (M')_f/M' \to 0. \end{equation}\] Since \(R \to R_f\) is flat, we may tensor this exact sequence with \(R_f\) to deduce that \(M \otimes_R R_f = (M/M[f^\infty]) \otimes_R R_f \to M_1\) is an isomorphism.
By Lemma 0BP1 we have \(\text{Tor}_1^R(R', \Coker(M' \to (M')_f)) = 0\). The sequence (0BP4) thus remains exact upon tensoring over \(R\) with \(R'\). Using \(\alpha_1\) and Lemma 0BNK the resulting exact sequence can be written as [0BP5]\[\begin{equation} 0 \to (M/M[f^\infty]) \otimes_R R' \to (M')_f \to (M')_f/M' \to 0 \end{equation}\] This yields an isomorphism \((M/M[f^\infty]) \otimes_R R' \cong M'/M'[f^\infty]\). This implies that in the diagram \[\xymatrix{ & M[f^\infty] \otimes_R R' \ar[r] \ar[d] & M \otimes_R R' \ar[r] \ar[d] & (M/M[f^\infty]) \otimes_R R' \ar[r] \ar[d] & 0 \\ 0 \ar[r] & M'[f^\infty] \ar[r] & M' \ar[r] & M'/M'[f^\infty] \ar[r] & 0, }\] the third vertical arrow is an isomorphism. Since the rows are exact and the first vertical arrow is an isomorphism by Lemma 0BNK and \(M[f^\infty] = M'[f^\infty]\), the five lemma implies that \(M \otimes_R R' \to M'\) is an isomorphism.
The above shows that \(\text{Can}\) is essentially surjective and that the functor \(H^0\) maps into the category of glueable modules. Due to the exactness of (0F1R) for glueable modules we have \(H^0 \circ \text{Can} = \text{id}\) on the category of glueable modules. This implies \(\text{Can}\) is fully faithful by Lemma 0H77 combined with Categories, Lemma 07RB. This finishes the proof.
Remark
Let \((R \to R', f)\) be a glueing pair. Let \(M\) be an \(R\)-module that is not necessarily glueable for \((R \to R', f)\). Setting \(M' = M \otimes_R R'\) and \(M_1 = M_f\) we obtain the glueing datum \(\text{Can}(M) = (M', M_1, \text{can})\). Then \(\tilde M = H^0(M', M_1, \text{can})\) is an \(R\)-module that is glueable for \((R \to R', f)\) and the canonical map \(M \to \tilde M\) gives isomorphisms \(M \otimes_R R' \to \tilde M \otimes_R R'\) and \(M_f \to \tilde M_f\), see Theorem 0BP2. From the exactness of the sequences \[M \to (M \otimes_R R' )\oplus M_f \to M \otimes_R (R')_f \to 0\] and \[0 \to \tilde M \to (\tilde M \otimes_R R') \oplus \tilde M_f \to \tilde M \otimes_R (R')_f \to 0\] we conclude that the map \(M \to \tilde M\) is surjective.
Recall that flat \(R\)-modules over a glueing pair \((R \to R', f)\) are glueable (Remark 0BNX). Hence the following lemma shows that Theorem 0BP2 determines an equivalence between the category of flat \(R\)-modules and the category of glueing data \((M', M_1, \alpha_1)\) where \(M'\) and \(M_1\) are flat over \(R'\) and \(R_f\).
Lemma
Let \((R \to R', f)\) be a glueing pair. Let \(M\) be an \(R\)-module which is not necessarily glueable for \((R \to R', f)\). Then \(M\) is flat over \(R\) if and only if \(M \otimes_R R'\) is flat over \(R'\) and \(M_f\) is flat over \(R_f\).
Proof
One direction of the lemma follows from Algebra, Lemma 00HI. For the other direction, assume \(M \otimes_R R'\) is flat over \(R'\) and \(M_f\) is flat over \(R_f\). Let \(\tilde M\) be as in Remark 0BP9. If \(\tilde M\) is flat over \(R\), then applying Algebra, Lemma 00HL to the short exact sequence \(0 \to \Ker(M \to \tilde M) \to M \to \tilde M \to 0\) we find that \(\Ker(M \to \tilde M) \otimes_R (R' \oplus R_f)\) is zero. Hence \(M = \tilde M\) by Lemma 0BNL and we conclude. In other words, we may replace \(M\) by \(\tilde M\) and assume \(M\) is glueable for \((R \to R', f)\). Let \(N\) be a second \(R\)-module. It suffices to prove that \(\text{Tor}_1^R(M, N) = 0\), see Algebra, Lemma 00M5.
The long exact sequence of Tors associated to the short exact sequence \(0 \to R \to R' \oplus R_f \to (R')_f \to 0\) and \(N\) gives an exact sequence \[0 \to \text{Tor}_1^R(R', N) \to \text{Tor}_1^R((R')_f, N)\] and isomorphisms \(\text{Tor}_i^R(R', N) = \text{Tor}_i^R((R')_f, N)\) for \(i \geq 2\). Since \(\text{Tor}_i^R((R')_f, N) = \text{Tor}_i^R(R', N)_f\) we conclude that \(f\) is a nonzerodivisor on \(\text{Tor}_1^R(R', N)\) and invertible on \(\text{Tor}_i^R(R', N)\) for \(i \geq 2\). Since \(M \otimes_R R'\) is flat over \(R'\) we have \[\text{Tor}_i^R(M \otimes_R R', N) = (M \otimes_R R') \otimes_{R'} \text{Tor}_i^R(R', N)\] by the spectral sequence of Example 068F. Writing \(M \otimes_R R'\) as a filtered colimit of finite free \(R'\)-modules (Algebra, Theorem 058G) we conclude that \(f\) is a nonzerodivisor on \(\text{Tor}_1^R(M \otimes_R R', N)\) and invertible on \(\text{Tor}_i^R(M \otimes_R R', N)\). Next, we consider the exact sequence \(0 \to M \to M \otimes_R R' \oplus M_f \to M \otimes_R (R')_f \to 0\) coming from the fact that \(M\) is glueable and the associated long exact sequence of \(\text{Tor}\). The relevant part is \[\xymatrix{ \text{Tor}_1^R(M, N) \ar[r] & \text{Tor}_1^R(M \otimes_R R', N) \ar[r] & \text{Tor}_1^R(M \otimes_R (R')_f, N) \\ & \text{Tor}_2^R(M \otimes_R R', N) \ar[r] & \text{Tor}_2^R(M \otimes_R (R')_f, N) \ar[llu] }\] We conclude that \(\text{Tor}_1^R(M, N) = 0\) by our remarks above on the action on \(f\) on \(\text{Tor}_i^R(M \otimes_R R', N)\).
Observe that we have seen the result of the following lemma for “finitely generated” in Lemma 0BNN.
Lemma
Let \((R \to R', f)\) be a glueing pair. Let \(M\) be an \(R\)-module which is not necessarily glueable for \((R \to R', f)\). Then \(M\) is a finite projective \(R\)-module if and only if \(M \otimes_R R'\) is finite projective over \(R'\) and \(M_f\) is finite projective over \(R_f\).
Proof
Assume that \(M \otimes_R R'\) is a finite projective module over \(R'\) and that \(M_f\) is a finite projective module over \(R_f\). Our task is to prove that \(M\) is finite projective over \(R\). We will use Algebra, Lemma 00NX without further mention. By Lemma 0BP7 we see that \(M\) is flat. By Lemma 0BNN we see that \(M\) is finite. Choose a short exact sequence \(0 \to K \to R^{\oplus n} \to M \to 0\). Since a finite projective module is of finite presentation and since the sequence remains exact after tensoring with \(R'\) (by Algebra, Lemma 00HL) and \(R_f\), we conclude that \(K \otimes_R R'\) and \(K_f\) are finite modules. Using the lemma above we conclude that \(K\) is finitely generated. Hence \(M\) is finitely presented and hence finite projective.
Remark
In [Beauville-Laszlo] it is assumed that \(f\) is a nonzerodivisor in \(R\) and \(R' = R^\wedge\), which gives a glueing pair by Lemma 0BNR. Even in this setting Theorem 0BP2 says something new: the results of [Beauville-Laszlo] only apply to modules on which \(f\) is a nonzerodivisor (and hence glueable in our sense, see Lemma 0BNW). Lemma 0BP6 also provides a slight extension of the results of [Beauville-Laszlo]: not only can we allow \(M\) to have nonzero \(f\)-power torsion, we do not even require it to be glueable.
Derived Completion
Some references for the material in this section are [Dwyer-Greenlees], [Greenlees-May], [PSY], [dag12] (especially Chapter 4). Our exposition follows [BS]. The analogue (or “dual”) of this section for torsion modules is Dualizing Complexes, Section 0952. The relationship between the derived category of complexes with torsion cohomology and derived complete complexes can be found in Dualizing Complexes, Section 0A6V.
Let \(K \in D(A)\). Let \(f \in A\). We denote \(T(K, f)\) a derived limit of the system \[\ldots \to K \xrightarrow{f} K \xrightarrow{f} K\] in \(D(A)\).
Lemma
Let \(A\) be a ring. Let \(f \in A\). Let \(K \in D(A)\). The following are equivalent
\(\Ext^n_A(A_f, K) = 0\) for all \(n\),
\(\Hom_{D(A)}(E, K) = 0\) for all \(E\) in \(D(A_f)\),
\(T(K, f) = 0\),
for every \(p \in \mathbf{Z}\) we have \(T(H^p(K), f) = 0\),
for every \(p \in \mathbf{Z}\) we have \(\Hom_A(A_f, H^p(K)) = 0\) and \(\Ext^1_A(A_f, H^p(K)) = 0\),
\(R\Hom_A(A_f, K) = 0\),
the map \(\prod_{n \geq 0} K \to \prod_{n \geq 0} K\), \((x_0, x_1, \ldots) \mapsto (x_0 - fx_1, x_1 - fx_2, \ldots)\) is an isomorphism in \(D(A)\), and
add more here.
Proof
It is clear that (2) implies (1) and that (1) is equivalent to (6). Assume (1). Let \(I^\bullet\) be a K-injective complex of \(A\)-modules representing \(K\). Condition (1) signifies that \(\Hom_A(A_f, I^\bullet)\) is acyclic. Let \(M^\bullet\) be a complex of \(A_f\)-modules representing \(E\). Then \[\Hom_{D(A)}(E, K) = \Hom_{K(A)}(M^\bullet, I^\bullet) = \Hom_{K(A_f)}(M^\bullet, \Hom_A(A_f, I^\bullet))\] by Algebra, Lemma 08YP. As \(\Hom_A(A_f, I^\bullet)\) is a K-injective complex of \(A_f\)-modules by Lemma 0917 the fact that it is acyclic implies that it is homotopy equivalent to zero (Derived Categories, Lemma 070I). Thus we get (2).
A free resolution of the \(A\)-module \(A_f\) is given by \[0 \to \bigoplus\nolimits_{n \in \mathbf{N}} A \to \bigoplus\nolimits_{n \in \mathbf{N}} A \to A_f \to 0\] where the first map sends the \((a_0, a_1, a_2, \ldots)\) to \((a_0, a_1 - fa_0, a_2 - fa_1, \ldots)\) and the second map sends \((a_0, a_1, a_2, \ldots)\) to \(a_0 + a_1/f + a_2/f^2 + \ldots\). Applying \(\Hom_A(-, I^\bullet)\) we get \[0 \to \Hom_A(A_f, I^\bullet) \to \prod I^\bullet \to \prod I^\bullet \to 0\] Since \(\prod I^\bullet\) represents \(\prod_{n \geq 0} K\) this proves the equivalence of (1) and (7). On the other hand, by construction of derived limits in Derived Categories, Section 08TB the displayed exact sequence shows the object \(T(K, f)\) is a representative of \(R\Hom_A(A_f, K)\) in \(D(A)\). Thus the equivalence of (1) and (3).
There is a spectral sequence \[E_2^{p, q} = \Ext^p_A(A_f, H^q(K)) \Rightarrow \Ext^{p + q}_A(A_f, K)\] See Equation (0AVH)13. This spectral sequence degenerates at \(E_2\) because \(A_f\) has a length \(1\) resolution by projective \(A\)-modules (see above) hence the \(E_2\)-page has only 2 nonzero columns. Thus we obtain short exact sequences \[0 \to \Ext^1_A(A_f, H^{p - 1}(K)) \to \Ext^p_A(A_f, K) \to \Hom_A(A_f, H^p(K)) \to 0\] This proves (4) and (5) are equivalent to (1).
Lemma
Let \(A\) be a ring. Let \(K \in D(A)\). The set \(I\) of \(f \in A\) such that \(T(K, f) = 0\) is a radical ideal of \(A\).
Proof
We will use the results of Lemma 091P without further mention. If \(f \in I\), and \(g \in A\), then \(A_{gf}\) is an \(A_f\)-module hence \(\Ext^n_A(A_{gf}, K) = 0\) for all \(n\), hence \(gf \in I\). Suppose \(f, g \in I\). Then there is a short exact sequence \[0 \to A_{f + g} \to A_{f(f + g)} \oplus A_{g(f + g)} \to A_{gf(f + g)} \to 0\] because \(f, g\) generate the unit ideal in \(A_{f + g}\). This follows from Algebra, Lemma 00EJ and the easy fact that the last arrow is surjective. From the long exact sequence of \(\Ext\) and the vanishing of \(\Ext^n_A(A_{f(f + g)}, K)\), \(\Ext^n_A(A_{g(f + g)}, K)\), and \(\Ext^n_A(A_{gf(f + g)}, K)\) for all \(n\) we deduce the vanishing of \(\Ext^n_A(A_{f + g}, K)\) for all \(n\). Finally, if \(f^n \in I\) for some \(n > 0\), then \(f \in I\) because \(T(K, f) = T(K, f^n)\) or because \(A_f \cong A_{f^n}\).
Lemma
Let \(A\) be a ring. Let \(I \subset A\) be an ideal. Let \(M\) be an \(A\)-module.
If \(M\) is \(I\)-adically complete, then \(T(M, f) = 0\) for all \(f \in I\).
Conversely, if \(T(M, f) = 0\) for all \(f \in I\) and \(I\) is finitely generated, then \(M \to \lim M/I^nM\) is surjective.
Proof
Proof of (1). Assume \(M\) is \(I\)-adically complete. By Lemma 091P it suffices to prove \(\Ext^1_A(A_f, M) = 0\) and \(\Hom_A(A_f, M) = 0\). Since \(M = \lim M/I^nM\) and since \(\Hom_A(A_f, M/I^nM) = 0\) it follows that \(\Hom_A(A_f, M) = 0\). Suppose we have an extension \[0 \to M \to E \to A_f \to 0\] For \(n \geq 0\) pick \(e_n \in E\) mapping to \(1/f^n\). Set \(\delta_n = fe_{n + 1} - e_n \in M\) for \(n \geq 0\). Replace \(e_n\) by \[e'_n = e_n + \delta_n + f\delta_{n + 1} + f^2 \delta_{n + 2} + \ldots\] The infinite sum exists as \(M\) is complete with respect to \(I\) and \(f \in I\). A simple calculation shows that \(fe'_{n + 1} = e'_n\). Thus we get a splitting of the extension by mapping \(1/f^n\) to \(e'_n\).
Proof of (2). Assume that \(I = (f_1, \ldots, f_r)\) and that \(T(M, f_i) = 0\) for \(i = 1, \ldots, r\). By Algebra, Lemma 090S we may assume \(I = (f)\) and \(T(M, f) = 0\). Let \(x_n \in M\) for \(n \geq 0\). Consider the extension \[0 \to M \to E \to A_f \to 0\] given by \[E = M \oplus \bigoplus Ae_n\Big/\langle x_n - fe_{n + 1} + e_n\rangle\] mapping \(e_n\) to \(1/f^n\) in \(A_f\) (see above). By assumption and Lemma 091P this extension is split, hence we obtain an element \(x + e_0\) which generates a copy of \(A_f\) in \(E\). Then \[x + e_0 = x - x_0 + fe_1 = x - x_0 - f x_1 + f^2 e_2 = \ldots\] Since \(M/f^nM = E/f^nE\) by the snake lemma, we see that \(x = x_0 + fx_1 + \ldots + f^{n - 1}x_{n - 1}\) modulo \(f^nM\). In other words, the map \(M \to \lim M/f^nM\) is surjective as desired.
Motivated by the results above we make the following definition.
Definition
Let \(A\) be a ring. Let \(K \in D(A)\). Let \(I \subset A\) be an ideal. We say \(K\) is derived complete with respect to \(I\) if for every \(f \in I\) we have \(T(K, f) = 0\). If \(M\) is an \(A\)-module, then we say \(M\) is derived complete with respect to \(I\) if \(M[0] \in D(A)\) is derived complete with respect to \(I\).
The full subcategory \(D_{comp}(A) = D_{comp}(A, I) \subset D(A)\) consisting of derived complete objects is a strictly full, saturated triangulated subcategory, see Derived Categories, Definitions 05QM and 05RB. By Lemma 091Q the subcategory \(D_{comp}(A, I)\) depends only on the radical \(\sqrt{I}\) of \(I\), in other words it depends only on the closed subset \(Z = V(I)\) of \(\Spec(A)\). The subcategory \(D_{comp}(A, I)\) is preserved under products and homotopy limits in \(D(A)\). But it is not preserved under countable direct sums in general. We will often simply say \(M\) is a derived complete module if the choice of the ideal \(I\) is clear from the context.
Proposition
Let \(I \subset A\) be a finitely generated ideal of a ring \(A\). Let \(M\) be an \(A\)-module. The following are equivalent
\(M\) is \(I\)-adically complete, and
\(M\) is derived complete with respect to \(I\) and \(\bigcap I^nM = 0\).
Proof
This is clear from the results of Lemma 091R.
The next lemma shows that the category \(\mathcal{C}\) of derived complete modules is abelian. It turns out that \(\mathcal{C}\) is not a Grothendieck abelian category, see Examples, Section 0ARC.
Lemma
Let \(I\) be an ideal of a ring \(A\).
The derived complete \(A\)-modules form a weak Serre subcategory \(\mathcal{C}\) of \(\text{Mod}_A\).
\(D_\mathcal{C}(A) \subset D(A)\) is the full subcategory of derived complete objects.
Proof
Part (2) is immediate from Lemma 091P and the definitions. For part (1), suppose that \(M \to N\) is a map of derived complete modules. Denote \(K = (M \to N)\) the corresponding object of \(D(A)\). Pick \(f \in I\). Then \(\Ext_A^n(A_f, K)\) is zero for all \(n\) because \(\Ext_A^n(A_f, M)\) and \(\Ext_A^n(A_f, N)\) are zero for all \(n\). Hence \(K\) is derived complete. By (2) we see that \(\Ker(M \to N)\) and \(\Coker(M \to N)\) are objects of \(\mathcal{C}\). Finally, suppose that \(0 \to M_1 \to M_2 \to M_3 \to 0\) is a short exact sequence of \(A\)-modules and \(M_1\), \(M_3\) are derived complete. Then it follows from the long exact sequence of \(\Ext\)’s that \(M_2\) is derived complete. Thus \(\mathcal{C}\) is a weak Serre subcategory by Homology, Lemma 0754.
We will generalize the following lemma in Lemma 0G1U.
Lemma
Let \(I\) be a finitely generated ideal of a ring \(A\). Let \(M\) be a derived complete \(A\)-module. If \(M/IM = 0\), then \(M = 0\).
Proof
Assume that \(M/IM\) is zero. Let \(I = (f_1, \ldots, f_r)\). Let \(i < r\) be the largest integer such that \(N = M/(f_1, \ldots, f_i)M\) is nonzero. If \(i\) does not exist, then \(M = 0\) which is what we want to show. Then \(N\) is derived complete as a cokernel of a map between derived complete modules, see Lemma 091U. By our choice of \(i\) we have that \(f_{i + 1} : N \to N\) is surjective. Hence \[\lim (\ldots \to N \xrightarrow{f_{i + 1}} N \xrightarrow{f_{i + 1}} N)\] is nonzero, contradicting the derived completeness of \(N\).
If the ring is \(I\)-adically complete, then one obtains an ample supply of derived complete complexes.
Lemma
Let \(A\) be a ring and \(I \subset A\) an ideal. If \(A\) is derived complete (eg. \(I\)-adically complete) then any pseudo-coherent object of \(D(A)\) is derived complete.
Proof
(Lemma 091R explains the parenthetical statement of the lemma.) Let \(K\) be a pseudo-coherent object of \(D(A)\). By definition this means \(K\) is represented by a bounded above complex \(K^\bullet\) of finite free \(A\)-modules. Since \(A\) is derived complete it follows that \(H^n(K)\) is derived complete for all \(n\), by part (1) of Lemma 091U. This in turn implies that \(K\) is derived complete by part (2) of the same lemma.
Lemma
Let \(A\) be a ring. Let \(f, g \in A\). Then for \(K \in D(A)\) we have \(R\Hom_A(A_f, R\Hom_A(A_g, K)) = R\Hom_A(A_{fg}, K)\).
Proof
This follows from Lemma 0A65.
Lemma
Let \(I\) be a finitely generated ideal of a ring \(A\). The inclusion functor \(D_{comp}(A, I) \to D(A)\) has a left adjoint, i.e., given any object \(K\) of \(D(A)\) there exists a map \(K \to K^\wedge\) of \(K\) into a derived complete object of \(D(A)\) such that the map \[\Hom_{D(A)}(K^\wedge, E) \longrightarrow \Hom_{D(A)}(K, E)\] is bijective whenever \(E\) is a derived complete object of \(D(A)\). In fact, if \(I\) is generated by \(f_1, \ldots, f_r \in A\), then we have \[K^\wedge = R\Hom\left((A \to \prod\nolimits_{i_0} A_{f_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{f_{i_0}f_{i_1}} \to \ldots \to A_{f_1\ldots f_r}), K\right)\] functorially in \(K\).
Proof
Define \(K^\wedge\) by the last displayed formula of the lemma. There is a map of complexes \[(A \to \prod\nolimits_{i_0} A_{f_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{f_{i_0}f_{i_1}} \to \ldots \to A_{f_1\ldots f_r}) \longrightarrow A\] which induces a map \(K \to K^\wedge\). It suffices to prove that \(K^\wedge\) is derived complete and that \(K \to K^\wedge\) is an isomorphism if \(K\) is derived complete14.
Let \(f \in A\). By Lemma 0A6C the object \(R\Hom_A(A_f, K^\wedge)\) is equal to \[R\Hom\left((A_f \to \prod\nolimits_{i_0} A_{ff_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{ff_{i_0}f_{i_1}} \to \ldots \to A_{ff_1\ldots f_r}), K\right)\] If \(f \in I\), then \(f_1, \ldots, f_r\) generate the unit ideal in \(A_f\), hence the extended alternating Čech complex \[A_f \to \prod\nolimits_{i_0} A_{ff_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{ff_{i_0}f_{i_1}} \to \ldots \to A_{ff_1\ldots f_r}\] is zero in \(D(A)\) by Lemma 0G6K. (In fact, if \(f = f_i\) for some \(i\), then this complex is homotopic to zero by Lemma 0G6J; this is the only case we need.) Hence \(R\Hom_A(A_f, K^\wedge) = 0\) and we conclude that \(K^\wedge\) is derived complete by Lemma 091P.
Conversely, if \(K\) is derived complete, then \(R\Hom_A(A_f, K)\) is zero for all \(f = f_{i_0} \ldots f_{i_p}\), \(p \geq 0\). Thus \(K \to K^\wedge\) is an isomorphism in \(D(A)\).
Remark
Let \(A\) be a ring and let \(I \subset A\) be a finitely generated ideal. The left adjoint to the inclusion functor \(D_{comp}(A, I) \to D(A)\) which exists by Lemma 091V is called the derived completion. To indicate this we will say “let \(K^\wedge\) be the derived completion of \(K\)”. Please keep in mind that the unit of the adjunction is a functorial map \(K \to K^\wedge\).
Lemma
Let \(A\) be a ring and let \(I \subset A\) be a finitely generated ideal. Let \(K^\bullet\) be a complex of \(A\)-modules such that \(f : K^\bullet \to K^\bullet\) is an isomorphism for some \(f \in I\), i.e., \(K^\bullet\) is a complex of \(A_f\)-modules. Then the derived completion of \(K^\bullet\) is zero.
Proof
Indeed, in this case the \(R\Hom_A(K, L)\) is zero for any derived complete complex \(L\), see Lemma 091P. Hence \(K^\wedge\) is zero by the universal property in Lemma 091V.
Lemma
Let \(A\) be a ring and let \(I \subset A\) be a finitely generated ideal. Let \(K, L \in D(A)\). Then \[R\Hom_A(K, L)^\wedge = R\Hom_A(K, L^\wedge) = R\Hom_A(K^\wedge, L^\wedge)\]
Proof
By Lemma 091V we know that derived completion is given by \(R\Hom_A(C, -)\) for some \(C \in D(A)\). Then \[\begin{align*} R\Hom_A(C, R\Hom_A(K, L)) & = R\Hom_A(C \otimes_A^\mathbf{L} K, L) \\ & = R\Hom_A(K, R\Hom_A(C, L)) \end{align*}\] by Lemma 0A65. This proves the first equation. The map \(K \to K^\wedge\) induces a map \[R\Hom_A(K^\wedge, L^\wedge) \to R\Hom_A(K, L^\wedge)\] which is an isomorphism in \(D(A)\) by definition of the derived completion as the left adjoint to the inclusion functor.
Lemma
Let \(A\) be a ring and let \(I \subset A\) be an ideal. Let \((K_n)\) be an inverse system of objects of \(D(A)\) such that for all \(f \in I\) and \(n\) there exists an \(e = e(n, f)\) such that \(f^e\) is zero on \(K_n\). Then for \(K \in D(A)\) the object \(K' = R\lim (K \otimes_A^\mathbf{L} K_n)\) is derived complete with respect to \(I\).
Proof
Since the category of derived complete objects is preserved under \(R\lim\) it suffices to show that each \(K \otimes_A^\mathbf{L} K_n\) is derived complete. By assumption for all \(f \in I\) there is an \(e\) such that \(f^e\) is zero on \(K \otimes_A^\mathbf{L} K_n\). Of course this implies that \(T(K \otimes_A^\mathbf{L} K_n, f) = 0\) and we win.
Situation
Let \(A\) be a ring. Let \(I = (f_1, \ldots, f_r) \subset A\). Let \(K_n^\bullet = K_\bullet(A, f_1^n, \ldots, f_r^n)\) be the Koszul complex on \(f_1^n, \ldots, f_r^n\) viewed as a cochain complex in degrees \(-r, -r + 1, \ldots, 0\). Using the functoriality of Lemma 0624 we obtain an inverse system \[\ldots \to K_3^\bullet \to K_2^\bullet \to K_1^\bullet\] compatible with the inverse system \(H^0(K_n^\bullet) = A/(f_1^n, \ldots, f_r^n)\) and compatible with the maps \(A \to K_n^\bullet\).
A key feature of the discussion below will use that for \(m > n\) the map \[K_m^{-p} = \wedge^p(A^{\oplus r}) \to \wedge^p(A^{\oplus r}) = K_n^{-p}\] is given by multiplication by \(f_{i_1}^{m - n} \ldots f_{i_p}^{m - n}\) on the basis element \(e_{i_1} \wedge \ldots \wedge e_{i_p}\).
Lemma
In Situation 0BKC. For \(K \in D(A)\) the object \(K' = R\lim (K \otimes_A^\mathbf{L} K_n^\bullet)\) is derived complete with respect to \(I\).
Proof
This is a special case of Lemma 091W because \(f_i^n\) acts by an endomorphism of \(K_n^\bullet\) which is homotopic to zero by Lemma 0663.
Lemma
In Situation 0BKC. Let \(K \in D(A)\). The following are equivalent
\(K\) is derived complete with respect to \(I\), and
the canonical map \(K \to R\lim (K \otimes_A^\mathbf{L} K_n^\bullet)\) is an isomorphism of \(D(A)\).
Proof
If (2) holds, then \(K\) is derived complete with respect to \(I\) by Lemma 091Y. Conversely, assume that \(K\) is derived complete with respect to \(I\). Consider the filtrations \[K_n^\bullet \supset \sigma_{\geq -r + 1}K_n^\bullet \supset \sigma_{\geq -r + 2}K_n^\bullet \supset \ldots \supset \sigma_{\geq -1}K_n^\bullet \supset \sigma_{\geq 0}K_n^\bullet = A\] by stupid truncations (Homology, Section 0118). Because the construction \(R\lim(K \otimes E)\) is exact in the second variable (Lemma 091K) we see that it suffices to show \[R\lim \left( K \otimes_A^\mathbf{L} (\sigma_{\geq p}K_n^\bullet/ \sigma_{\geq p + 1}K_n^\bullet) \right) = 0\] for \(p < 0\). The explicit description of the Koszul complexes above shows that \[R\lim \left( K \otimes_A^\mathbf{L} (\sigma_{\geq p}K_n^\bullet/ \sigma_{\geq p + 1}K_n^\bullet) \right) = \bigoplus\nolimits_{i_1, \ldots, i_{-p}} T(K, f_{i_1}\ldots f_{i_{-p}})\] which is zero for \(p < 0\) by assumption on \(K\).
Lemma
In Situation 0BKC. The functor which sends \(K \in D(A)\) to the derived limit \(K' = R\lim( K \otimes_A^\mathbf{L} K_n^\bullet )\) is the left adjoint to the inclusion functor \(D_{comp}(A) \to D(A)\) constructed in Lemma 091V.
Proof
The assignment \(K \leadsto K'\) is a functor and \(K'\) is derived complete with respect to \(I\) by Lemma 091Y. By a formal argument (omitted) we see that it suffices to show \(K \to K'\) is an isomorphism if \(K\) is derived complete with respect to \(I\). This is Lemma 091Z.
Proof
Denote \(K \mapsto K^\wedge\) the adjoint constructed in Lemma 091V. By that lemma we have \[K^\wedge = R\Hom\left((A \to \prod\nolimits_{i_0} A_{f_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{f_{i_0}f_{i_1}} \to \ldots \to A_{f_1\ldots f_r}), K\right)\] In Lemma 0913 we have seen that the extended alternating Čech complex \[A \to \prod\nolimits_{i_0} A_{f_{i_0}} \to \prod\nolimits_{i_0 < i_1} A_{f_{i_0}f_{i_1}} \to \ldots \to A_{f_1\ldots f_r}\] is a colimit of the Koszul complexes \(K^n = K(A, f_1^n, \ldots, f_r^n)\) sitting in degrees \(0, \ldots, r\). Note that \(K^n\) is a finite chain complex of finite free \(A\)-modules with dual (as in Lemma 07VI) \(R\Hom_A(K^n, A) = K_n\) where \(K_n\) is the Koszul cochain complex sitting in degrees \(-r, \ldots, 0\) (as usual). Thus it suffices to show that \[R\Hom_A(\text{hocolim} K^n, K) = R\lim (K \otimes_A^\mathbf{L} K_n)\] This follows from Lemma 0BKB.
Lemma
Let \(I = (f_1, \ldots, f_r)\) be a finitely generated ideal of a ring \(A\). Let \(K\) be a derived complete object of \(D(A)\). The following are equivalent
\(H^i(K) = 0\) for \(i > 0\),
\(H^i(K \otimes_A^\mathbf{L} A/I) = 0\) for \(i > 0\),
\(H^i(K \otimes_A^\mathbf{L} K_1^\bullet) = 0\) for \(i > 0\) where \(K_1^\bullet\) is as in Situation 0BKC.
Proof
The implication (1) \(\Rightarrow\) (2) is always true. The implication (2) \(\Rightarrow\) (3) follows from Lemma 0H82. Assume (3). For \(s = 0, \ldots, r\) consider the complex \[K(s) = K \otimes_A^\mathbf{L} K_\bullet(A, f_1, \ldots, f_s)\] where \(K_\bullet(A, f_1, \ldots, f_s)\) is the Koszul complex placed in cohomological degrees \(-s, \ldots, 0\). We have a distinguished triangles \[K(s - 1) \xrightarrow{f_s} K(s - 1) \to K(s) \to K(s - 1)[1]\] by Lemma 0629. Since \(K(0) = K\) is derived complete, it follows by ascending induction that each \(K(s)\) is a derived complete object of \(D(A)\). By descending induction we’ll show that \(K(s)\) has no nonzero cohomology in degrees \(> 0\). Namely, this holds for \(s = r\) by our assumption (3). If it holds for \(K(s)\) and \(s > 0\), then \(f_s : H^i(K(s - 1)) \to H^i(K(s - 1))\) is surjective for \(i > 0\) and we conclude that \(H^i(K(s - 1))\) vanishes from Lemmas 091U and 09B9.
Lemma
Let \(I\) be a finitely generated ideal of a ring \(A\). Let \(K\) be a derived complete object of \(D(A)\). If \(K \otimes_A^\mathbf{L} A/I = 0\), then \(K = 0\).
Proof
This is a special case of Lemma 0H83 but we’ll also give a direct proof. Choose generators \(f_1, \ldots, f_r\) of \(I\). Denote \(K_n\) the Koszul complex on \(f_1^n, \ldots, f_r^n\) over \(A\). Recall that \(K_n\) is bounded and that the cohomology modules of \(K_n\) are annihilated by \(f_1^n, \ldots, f_r^n\) and hence by \(I^{nr}\). By Lemma 0G1T we see that \(K \otimes_A^\mathbf{L} K_n = 0\). Since \(K\) is derived complete by Lemma 0920 we have \(K = R\lim K \otimes_A^\mathbf{L} K_n = 0\) as desired.
As an application of the relationship with the Koszul complex we obtain that derived completion has finite cohomological dimension.
Lemma
Let \(A\) be a ring and let \(I \subset A\) be an ideal which can be generated by \(r\) elements. Then derived completion has finite cohomological dimension:
Let \(K \to L\) be a morphism in \(D(A)\) such that \(H^i(K) \to H^i(L)\) is an isomorphism for \(i \geq 1\) and surjective for \(i = 0\). Then \(H^i(K^\wedge) \to H^i(L^\wedge)\) is an isomorphism for \(i \geq 1\) and surjective for \(i = 0\).
Let \(K \to L\) be a morphism of \(D(A)\) such that \(H^i(K) \to H^i(L)\) is an isomorphism for \(i \leq -1\) and injective for \(i = 0\). Then \(H^i(K^\wedge) \to H^i(L^\wedge)\) is an isomorphism for \(i \leq -r - 1\) and injective for \(i = -r\).
Proof
Say \(I\) is generated by \(f_1, \ldots, f_r\). For any \(K \in D(A)\) by Lemma 0920 we have \(K^\wedge = R\lim K \otimes_A^\mathbf{L} K_n\) where \(K_n\) is the Koszul complex on \(f_1^n, \ldots, f_r^n\) and hence we obtain a short exact sequence \[0 \to R^1\lim H^{i - 1}(K \otimes_A^\mathbf{L} K_n) \to H^i(K^\wedge) \to \lim H^i(K \otimes_A^\mathbf{L} K_n) \to 0\] by Lemma 0CQE.
Proof of (1). Pick a distinguished triangle \(K \to L \to C \to K[1]\). Then \(H^i(C) = 0\) for \(i \geq 0\). Since \(K_n\) is sitting in degrees \(\leq 0\) we see that \(H^i(C \otimes_A^\mathbf{L} K_n) = 0\) for \(i \geq 0\) and that \(H^{-1}(C \otimes_A^\mathbf{L} K_n) = H^{-1}(C) \otimes_A A/(f_1^n, \ldots, f_r^n)\) is a system with surjective transition maps. The displayed equation above shows that \(H^i(C^\wedge) = 0\) for \(i \geq 0\). Applying the distinguished triangle \(K^\wedge \to L^\wedge \to C^\wedge \to K^\wedge[1]\) we get (1).
Proof of (2). Pick a distinguished triangle \(K \to L \to C \to K[1]\). Then \(H^i(C) = 0\) for \(i < 0\). Since \(K_n\) is sitting in degrees \(-r, \ldots, 0\) we see that \(H^i(C \otimes_A^\mathbf{L} K_n) = 0\) for \(i < -r\). The displayed equation above shows that \(H^i(C^\wedge) = 0\) for \(i < -r\). Applying the distinguished triangle \(K^\wedge \to L^\wedge \to C^\wedge \to K^\wedge[1]\) we get (2).
Lemma
Let \(A\) be a ring and let \(I \subset A\) be a finitely generated ideal. Let \(K^\bullet\) be a filtered complex of \(A\)-modules. There exists a canonical spectral sequence \((E_r, \text{d}_r)_{r \geq 1}\) of bigraded derived complete \(A\)-modules with \(d_r\) of bidegree \((r, -r + 1)\) and with \[E_1^{p, q} = H^{p + q}((\text{gr}^pK^\bullet)^\wedge)\] If the filtration on each \(K^n\) is finite, then the spectral sequence is bounded and converges to \(H^*((K^\bullet)^\wedge)\).
Proof
By Lemma 091V we know that derived completion is given by \(R\Hom_A(C, -)\) for some \(C \in D^b(A)\). By Lemmas 0AAJ and 0A5P we see that \(C\) has finite projective dimension. Thus we may choose a bounded complex of projective modules \(P^\bullet\) representing \(C\). Then \[M^\bullet = \Hom^\bullet(P^\bullet, K^\bullet)\] is a complex of \(A\)-modules representing \((K^\bullet)^\wedge\). It comes with a filtration given by \(F^pM^\bullet = \Hom^\bullet(P^\bullet, F^pK^\bullet)\). We see that \(F^pM^\bullet\) represents \((F^pK^\bullet)^\wedge\) and hence \(\text{gr}^pM^\bullet\) represents \((\text{gr}K^\bullet)^\wedge\). Thus we find our spectral sequence by taking the spectral sequence of the filtered complex \(M^\bullet\), see Homology, Section 012K. If the filtration on each \(K^n\) is finite, then the filtration on each \(M^n\) is finite because \(P^\bullet\) is a bounded complex. Hence the final statement follows from Homology, Lemma 012W.
Example
Let \(A\) be a ring and let \(I \subset A\) be a finitely generated ideal. Let \(K^\bullet\) be a complex of \(A\)-modules. We can apply Lemma 0BKD with \(F^pK^\bullet = \tau_{\leq -p}K^\bullet\). Then we get a bounded spectral sequence \[E_1^{p, q} = H^{p + q}(H^{-p}(K^\bullet)^\wedge[p]) = H^{2p + q}(H^{-p}(K^\bullet)^\wedge)\] converging to \(H^{p + q}((K^\bullet)^\wedge)\). After renumbering \(p = -j\) and \(q = i + 2j\) we find that for any \(K \in D(A)\) there is a bounded spectral sequence \((E'_r, d'_r)_{r \geq 2}\) of bigraded derived complete modules with \(d'_r\) of bidegree \((r, -r + 1)\), with \[(E'_2)^{i, j} = H^i(H^j(K)^\wedge)\] and converging to \(H^{i + j}(K^\wedge)\).
Lemma
Let \(A \to B\) be a ring map. Let \(I \subset A\) be an ideal. The inverse image of \(D_{comp}(A, I)\) under the restriction functor \(D(B) \to D(A)\) is \(D_{comp}(B, IB)\).
Proof
Using Lemma 091Q we see that \(L \in D(B)\) is in \(D_{comp}(B, IB)\) if and only if \(T(L, f)\) is zero for every local section \(f \in I\). Observe that the cohomology of \(T(L, f)\) is computed in the category of abelian groups, so it doesn’t matter whether we think of \(f\) as an element of \(A\) or take the image of \(f\) in \(B\). The lemma follows immediately from this and the definition of derived complete objects.
Lemma
Let \(A \to B\) be a ring map. Let \(I \subset A\) be a finitely generated ideal. If \(A \to B\) is flat and \(A/I \cong B/IB\), then the restriction functor \(D(B) \to D(A)\) induces an equivalence \(D_{comp}(B, IB) \to D_{comp}(A, I)\).
Proof
Choose generators \(f_1, \ldots, f_r\) of \(I\). Denote \(\check{\mathcal{C}}^\bullet_A \to \check{\mathcal{C}}^\bullet_B\) the quasi-isomorphism of extended alternating Čech complexes of Lemma 091M. Let \(K \in D_{comp}(A, I)\). Let \(I^\bullet\) be a K-injective complex of \(A\)-modules representing \(K\). Since \(\Ext^n_A(A_f, K)\) and \(\Ext^n_A(B_f, K)\) are zero for all \(f \in I\) and \(n \in \mathbf{Z}\) (Lemma 091P) we conclude that \(\check{\mathcal{C}}^\bullet_A \to A\) and \(\check{\mathcal{C}}^\bullet_B \to B\) induce quasi-isomorphisms \[I^\bullet = \Hom_A(A, I^\bullet) \longrightarrow \text{Tot}(\Hom_A(\check{\mathcal{C}}^\bullet_A, I^\bullet))\] and \[\Hom_A(B, I^\bullet) \longrightarrow \text{Tot}(\Hom_A(\check{\mathcal{C}}^\bullet_B, I^\bullet))\] Some details omitted. Since \(\check{\mathcal{C}}^\bullet_A \to \check{\mathcal{C}}^\bullet_B\) is a quasi-isomorphism and \(I^\bullet\) is K-injective we conclude that \(\Hom_A(B, I^\bullet) \to I^\bullet\) is a quasi-isomorphism. As the complex \(\Hom_A(B, I^\bullet)\) is a complex of \(B\)-modules we conclude that \(K\) is in the image of the restriction map, i.e., the functor is essentially surjective
In fact, the argument shows that \(F : D_{comp}(A, I) \to D_{comp}(B, IB)\), \(K \mapsto \Hom_A(B, I^\bullet)\) is a left inverse to restriction. Finally, suppose that \(L \in D_{comp}(B, IB)\). Represent \(L\) by a K-injective complex \(J^\bullet\) of \(B\)-modules. Then \(J^\bullet\) is also K-injective as a complex of \(A\)-modules (Lemma 0915) hence \(F(\text{restriction of }L) = \Hom_A(B, J^\bullet)\). There is a map \(J^\bullet \to \Hom_A(B, J^\bullet)\) of complexes of \(B\)-modules, whose composition with \(\Hom_A(B, J^\bullet) \to J^\bullet\) is the identity. We conclude that \(F\) is also a right inverse to restriction and the proof is finished.
The category of derived complete modules
Let \(A\) be a ring and let \(I\) be an ideal. Denote \(\mathcal{C}\) the category of derived complete modules, see Definition 091S. In this section we discuss some properties of this category. In Examples, Section 0ARC we show that \(\mathcal{C}\) isn’t a Grothendieck abelian category in general.
By Lemma 091U the category \(\mathcal{C}\) is abelian and the inclusion functor \(\mathcal{C} \to \text{Mod}_A\) is exact.
Since \(D_{comp}(A) \subset D(A)\) is closed under products (see discussion following Definition 091S) and since products in \(D(A)\) are computed on the level of complexes, we see that \(\mathcal{C}\) has products which agree with products in \(\text{Mod}_A\). Thus \(\mathcal{C}\) in fact has arbitrary limits and the inclusion functor \(\mathcal{C} \to \text{Mod}_A\) commutes with them, see Categories, Lemma 002N.
Assume \(I\) is finitely generated. Let \({}^\wedge : D(A) \to D(A)\) denote the derived completion functor of Lemma 091V. Let us show the functor \[\text{Mod}_A \longrightarrow \mathcal{C},\quad M \longmapsto H^0(M^\wedge)\] is a left adjoint to the inclusion functor \(\mathcal{C} \to \text{Mod}_A\). Note that \(H^i(M^\wedge) = 0\) for \(i > 0\) for example by Lemma 0AAJ. Hence, if \(N\) is a derived complete \(A\)-module, then we have \[\begin{align*} \Hom_\mathcal{C}(H^0(M^\wedge), N) & = \Hom_{D_{comp}(A)}(M^\wedge, N)\\ & = \Hom_{D(A)}(M, N) \\ & = \Hom_A(M, N) \end{align*}\] as desired.
Let \(T\) be a preordered set and let \(t \mapsto M_t\) be a system of derived complete \(A\)-modules, i.e., a system over \(T\) in \(\mathcal{C}\), see Categories, Section 002Z. Denote \(\colim_{t \in T} M_t\) the colimit of the system in \(\text{Mod}_A\). It follows formally from the above that \[H^0((\colim_{t \in T} M_t)^\wedge)\] is the colimit of the system in \(\mathcal{C}\). In this way we see that \(\mathcal{C}\) has all colimits. In general the inclusion functor \(\mathcal{C} \to \text{Mod}_A\) will not commute with colimits, see Examples, Section 0ARC.
Lemma
Let \(A\) be a ring and let \(I \subset A\) be an ideal. The category \(\mathcal{C}\) of derived complete modules is abelian, has arbitrary limits, and the inclusion functor \(F : \mathcal{C} \to \text{Mod}_A\) is exact and commutes with limits. If \(I\) is finitely generated, then \(\mathcal{C}\) has arbitrary colimits and \(F\) has a left adjoint
Proof
This summarizes the discussion above.
Derived completion for a principal ideal
In this section we discuss what happens with derived completion when the ideal is generated by a single element.
Lemma
Let \(A\) be a ring. Let \(f \in A\). If there exists an integer \(c \geq 1\) such that \(A[f^c] = A[f^{c + 1}] = A[f^{c + 2}] = \ldots\) (for example if \(A\) is Noetherian), then for all \(n \geq 1\) there exist maps \[(A \xrightarrow{f^n} A) \longrightarrow A/(f^n), \quad\text{and}\quad A/(f^{n + c}) \longrightarrow (A \xrightarrow{f^n} A)\] in \(D(A)\) inducing an isomorphism of the pro-objects \(\{A/(f^n)\}\) and \(\{(f^n : A \to A)\}\) in \(D(A)\).
Proof
The first displayed arrow is obvious. We can define the second arrow of the lemma by the diagram \[\xymatrix{ A/A[f^c] \ar[r]_-{f^{n + c}} \ar[d]_{f^c} & A \ar[d]^1 \\ A \ar[r]^{f^n} & A }\] Since the top horizontal arrow is injective the complex in the top row is quasi-isomorphic to \(A/f^{n + c}A\). We omit the calculation of compositions needed to show the statement on pro objects.
Lemma
Let \(A\) be a ring and \(f \in A\). Set \(I = (f)\). In this situation we have the naive derived completion \(K \mapsto K' = R\lim (K \otimes_A^\mathbf{L} A/f^nA)\) and the derived completion \[K \mapsto K^\wedge = R\lim (K \otimes_A^\mathbf{L} (A \xrightarrow{f^n} A))\] of Lemma 0920. The natural transformation of functors \(K^\wedge \to K'\) is an isomorphism if and only if the \(f\)-power torsion of \(A\) is bounded.
Proof
If the \(f\)-power torsion is bounded, then the pro-objects \(\{(f^n : A \to A)\}\) and \(\{A/f^nA\}\) are isomorphic by Lemma 091X. Hence the functors are isomorphic by Lemma 091B. Conversely, we see from Lemma 091K that the condition is exactly that \[R\lim (K \otimes_A^\mathbf{L} A[f^n])\] is zero for all \(K \in D(A)\). Here the maps of the system \((A[f^n])\) are given by multiplication by \(f\). Taking \(K = A\) and \(K = \bigoplus_{i \in \mathbf{N}} A\) we see from Lemma 091C this implies \((A[f^n])\) is zero as a pro-object, i.e., \(f^{n - 1}A[f^n] = 0\) for some \(n\), i.e., \(A[f^{n - 1}] = A[f^n]\), i.e., the \(f\)-power torsion is bounded.
Example
Let \(A\) be a ring. Let \(f \in A\) be a nonzerodivisor. An example to keep in mind is \(A = \mathbf{Z}_p\) and \(f = p\). Let \(M\) be an \(A\)-module. Claim: \(M\) is derived complete with respect to \(f\) if and only if there exists a short exact sequence \[0 \to K \to L \to M \to 0\] where \(K, L\) are \(f\)-adically complete modules whose \(f\)-torsion is zero. Namely, if there is a such a short exact sequence, then \[M \otimes_A^\mathbf{L} (A \xrightarrow{f^n} A) = (K/f^nK \to L/f^nL)\] because \(f\) is a nonzerodivisor on \(K\) and \(L\) and we conclude that \(R\lim (M \otimes_A^\mathbf{L} (A \xrightarrow{f^n} A))\) is quasi-isomorphic to \(K \to L\), i.e., \(M\). This shows that \(M\) is derived complete by Lemma 091Z. Conversely, suppose that \(M\) is derived complete. Choose a surjection \(F \to M\) where \(F\) is a free \(A\)-module. Since \(f\) is a nonzerodivisor on \(F\) the derived completion of \(F\) is \(L = \lim F/f^nF\). Note that \(L\) is \(f\)-torsion free: if \((x_n)\) with \(x_n \in F\) represents an element \(\xi\) of \(L\) and \(f\xi = 0\), then \(x_n = x_{n + 1} + f^nz_n\) and \(fx_n = f^ny_n\) for some \(z_n, y_n \in F\). Then \(f^n y_n = fx_n = fx_{n + 1} + f^{n + 1}z_n = f^{n + 1}y_{n + 1} + f^{n + 1}z_n\) and since \(f\) is a nonzerodivisor on \(F\) we see that \(y_n \in fF\) which implies that \(x_n \in f^nF\), i.e., \(\xi = 0\). Since \(L\) is the derived completion, the universal property gives a map \(L \to M\) factoring \(F \to M\). Let \(K = \Ker(L \to M)\) be the kernel. Again \(K\) is \(f\)-torsion free, hence the derived completion of \(K\) is \(\lim K/f^nK\). On the other hand, both \(M\) and \(L\) are derived complete, hence \(K\) is too by Lemma 091U. It follows that \(K = \lim K/f^nK\) and the claim is proved.
Example
Let \(p\) be a prime number. Consider the map \(\mathbf{Z}_p[x] \to \mathbf{Z}_p[y]\) of polynomial algebras sending \(x\) to \(py\). Consider the cokernel \(M = \Coker(\mathbf{Z}_p[x]^\wedge \to \mathbf{Z}_p[y]^\wedge)\) of the induced map on (ordinary) \(p\)-adic completions. Then \(M\) is a derived complete \(\mathbf{Z}_p\)-module by Proposition 091T and Lemma 091U; see also discussion in Example 09AT. However, \(M\) is not \(p\)-adically complete as \(1 + py + p^2 y^2 + \ldots\) maps to a nonzero element of \(M\) which is contained in \(\bigcap p^nM\).
Example
Let \(A\) be a ring and let \(f \in A\). Denote \(K \mapsto K^\wedge\) the derived completion with respect to \((f)\). Let \(M\) be an \(A\)-module. Using that \[M^\wedge = R\lim (M \xrightarrow{f^n} M)\] by Lemma 0920 and using Lemma 0CQE we obtain \[H^{-1}(M^\wedge) = \lim M[f^n] = T_f(M)\] the \(f\)-adic Tate module of \(M\). Here the maps \(M[f^n] \to M[f^{n - 1}]\) are given by multiplication by \(f\). Then there is a short exact sequence \[0 \to R^1\lim M[f^n] \to H^0(M^\wedge) \to \lim M/f^n M \to 0\] describing \(H^0(M^\wedge)\). We have \(H^1(M^\wedge) = R^1\lim M/f^nM = 0\) as the transition maps are surjective (Lemma 091D). All the other cohomologies of \(M^\wedge\) are zero for trivial reasons. Finally, for \(K \in D(A)\) and \(p \in \mathbf{Z}\) there is a short exact sequence \[0 \to H^0(H^p(K)^\wedge) \to H^p(K^\wedge) \to T_f(H^{p + 1}(K)) \to 0\] This follows from the spectral sequence of Example 0BKE because it degenerates at \(E_2\) (as only \(i = -1, 0\) give nonzero terms); the next lemma gives more information.
Lemma
Let \(A\) be a ring and let \(f \in A\). Let \(K\) be an object of \(D(A)\). Denote \(K_n = K \otimes_A^\mathbf{L} (A \xrightarrow{f^n} A)\). For all \(p \in \mathbf{Z}\) there is a commutative diagram \[\xymatrix{ & 0 & 0 \\ 0 \ar[r] & \widehat{H^p(K)} \ar[r] \ar[u] & \lim H^p(K_n) \ar[r] \ar[u] & T_f(H^{p + 1}(K)) \ar[r] & 0 \\ 0 \ar[r] & H^0(H^p(K)^\wedge) \ar[r] \ar[u] & H^p(K^\wedge) \ar[r] \ar[u] & T_f(H^{p + 1}(K)) \ar[r] \ar@{=}[u] & 0 \\ & R^1\lim H^p(K)[f^n] \ar[u] \ar[r]^\cong & R^1\lim H^{p - 1}(K_n) \ar[u] \\ & 0 \ar[u] & 0 \ar[u] }\] with exact rows and columns where \(\widehat{H^p(K)} = \lim H^p(K)/f^nH^p(K)\) is the usual \(f\)-adic completion. The left vertical short exact sequence and the middle horizontal short exact sequence are taken from Example 0BKG The middle vertical short exact sequence is the one from Lemma 0CQE.
Proof
To construct the top horizontal short exact sequence, observe that we have the following inverse system short exact sequences \[0 \to H^p(K)/f^nH^p(K) \to H^p(K_n) \to H^{p + 1}(K)[f^n] \to 0\] coming from the construction of \(K_n\) as a shift of the cone on \(f^n : K \to K\). Taking the inverse limit of these we obtain the top horizontal short exact sequence, see Homology, Lemma 02N1.
Let us prove that we have a commutative diagram as in the lemma. We consider the map \(L = \tau_{\leq p}K \to K\). Setting \(L_n = L \otimes_A^\mathbf{L} (A \xrightarrow{f^n} A)\) we obtain a map \((L_n) \to (K_n)\) of inverse systems which induces a map of short exact sequences \[\xymatrix{ 0 & 0 \\ \lim H^p(L_n) \ar[r] \ar[u] & \lim H^p(K_n) \ar[u] \\ H^p(L^\wedge) \ar[r] \ar[u] & H^p(K^\wedge) \ar[u] \\ R^1\lim H^{p - 1}(L_n) \ar[r] \ar[u] & R^1\lim H^{p - 1}(K_n) \ar[u] \\ 0 \ar[u] & 0 \ar[u] }\] Since \(H^i(L) = 0\) for \(i > p\) and \(H^p(L) = H^p(K)\), a computation using the references in the statement of the lemma shows that \(H^p(L^\wedge) = H^0(H^p(K)^\wedge)\) and that \(H^p(L_n) = H^p(K)/f^nH^p(K)\). On the other hand, we have \(H^{p - 1}(L_n) = H^{p - 1}(K_n)\) and hence we see that we get the isomorphism as indicated in the statement of the lemma since we already know the kernel of \(H^0(H^p(K)^\wedge) \to \widehat{H^p(K)}\) is equal to \(R^1\lim H^p(K)[f^n]\). We omit the verification that the rightmost square in the diagram commutes if we define the top row by the construction in the first paragraph of the proof.
Remark
With notation as in Lemma 0H32 we also see that the inverse system \(H^p(K_n)\) has ML if and only if the inverse system \(H^{p + 1}(K)[f^n]\) has ML. This follows from the inverse system of short exact sequences \(0 \to H^p(K)/f^nH^p(K) \to H^p(K_n) \to H^{p + 1}(K)[f^n] \to 0\) (see proof of the lemma) combined with Homology, Lemma 02N1 and Lemma 0CQB.
Lemma
Let \(I\) be a finitely generated ideal in a ring \(A\). Let \(M\) be a derived complete \(A\)-module. If \(M\) is an \(I\)-power torsion module, then \(I^nM = 0\) for some \(n\).
Proof
Say \(I = (f_1, \ldots, f_r)\). It suffices to show that for each \(i\) there is an \(n_i\) such that \(f_i^{n_i}M = 0\). Hence we may assume that \(I = (f)\) is a principal ideal. Let \(B = \mathbf{Z}[x] \to A\) be the ring map sending \(x\) to \(f\). By Lemma 0924 we see that \(M\) is derived complete as a \(B\)-module with respect to the ideal \((x)\). After replacing \(A\) by \(B\), we may assume that \(f\) is a nonzerodivisor in \(A\).
Assume \(I = (f)\) with \(f \in A\) a nonzerodivisor. According to Example 09AT there exists a short exact sequence \[0 \to K \xrightarrow{u} L \to M \to 0\] where \(K\) and \(L\) are \(I\)-adically complete \(A\)-modules whose \(f\)-torsion is zero15. Consider \(K\) and \(L\) as topological modules with the \(I\)-adic topology. Then \(u\) is continuous. Let \[L_n = \{x \in L \mid f^n x \in u(K)\}\] Since \(M\) is \(f\)-power torsion we see that \(L = \bigcup L_n\). Let \(N_n\) be the closure of \(L_n\) in \(L\). By Lemma 0CQV we see that \(N_n\) is open in \(L\) for some \(n\). Fix such an \(n\). Since \(f^{n + m} : L \to L\) is a continuous open map, and since \(f^{n + m} L_n \subset u(f^m K)\) we conclude that the closure of \(u(f^mK)\) is open for all \(m \geq 1\). Thus by Lemma 0CQW we conclude that \(u\) is open. Hence \(f^tL \subset \Im(u)\) for some \(t\) and we conclude that \(f^t\) annihilates \(M\) as desired.
Lemma
Let \(f \in A\) be an element of a ring. Set \(J = \bigcap f^nA\). Let \(M\) be an \(A\)-module derived complete with respect to \(f\). Then \(JM' = 0\) where \(M' = \Ker(M \to \lim M/f^nM)\). In particular, if \(A\) is derived complete then \(J\) is an ideal of square zero.
Proof
Take \(x \in M'\) and \(g \in J\). For every \(n \geq 1\) we may write \(x = f^n x_n\). Since \(g\) is in \(f^nA\) we see that the element \(y_n = gx_n\) in \(M'\) is independent of the choice of \(x_n\). In particular, we may take \(x_n = fx_{n + 1}\) and we find that \(y_n = fy_{n + 1}\). Thus we obtain a map \(A_f \to M\) sending \(1/f^n\) to \(y_n\). This map has to be zero as \(M\) is derived complete (Lemma 091P) and hence \(y_n = 0\) for all \(n\). Since \(gx = gfx_1 = fy_1\) this completes the proof.
Lemma
Let \(A\) be a ring derived complete with respect to an ideal \(I\). Then \((A, I)\) is a henselian pair.
Proof
Let \(f \in I\). By Lemma 0G1S it suffices to show that \((A, fA)\) is a henselian pair. Observe that \(A\) is derived complete with respect to \(fA\) (follows immediately from Definition 091S). By Lemma 091R the map from \(A\) to the \(f\)-adic completion \(A'\) of \(A\) is surjective. By Lemma 0ALJ the pair \((A', fA')\) is henselian. Thus it suffices to show that \((A, \bigcap f^nA)\) is a henselian pair, see Lemma 0DYD. This follows from Lemmas 0G3G and 0ALI.
Lemma
Let \(A\) be a ring derived complete with respect to an ideal \(I\). Set \(J = \bigcap I^n\). If \(I\) can be generated by \(r\) elements then \(J^N = 0\) where \(N = 2^r\).
Proof
When \(r = 1\) this is Lemma 0G3G. Say \(I = (f_1, \ldots, f_r)\) with \(r > 1\). By Lemma 091U the ring \(A_t = A/f_r^tA\) is derived complete with respect to \(I\) and hence a fortiori derived complete with respect to \(I_t = (f_1, \ldots, f_{r - 1})A_t\). Observe that \(A \to A_t\) sends \(J\) into \(J_t = \bigcap I_t^n\). By induction \(J_t^{N/2} = 0\) with \(N = 2^r\). The ideal \(\bigcap \Ker(A \to A_t) = \bigcap f_r^t A\) has square zero by the case \(r = 1\). This finishes the proof.
Lemma
Let \(A\) be a reduced ring derived complete with respect to a finitely generated ideal \(I\). Then \(A\) is \(I\)-adically complete.
Proof
Derived completion for Noetherian rings
Let \(A\) be a ring and let \(I \subset A\) be an ideal. For any \(K \in D(A)\) we can consider the derived limit \[K' = R\lim (K \otimes_A^\mathbf{L} A/I^n)\] This is a functor in \(K\), see Remark 091J. The system of maps \(A \to A/I^n\) induces a map \(K \to K'\) and \(K'\) is derived complete with respect to \(I\) (Lemma 091W). This “naive” derived completion construction does not agree with the adjoint of Lemma 091V in general. For example, if \(A = \mathbf{Z}_p \oplus \mathbf{Q}_p/\mathbf{Z}_p\) with the second summand an ideal of square zero, \(K = A[0]\), and \(I = (p)\), then the naive derived completion gives \(\mathbf{Z}_p[0]\), but the construction of Lemma 091V gives \(K^\wedge \cong \mathbf{Z}_p[1] \oplus \mathbf{Z}_p[0]\) (computation omitted). Lemma 0923 characterizes when the two functors agree in the case \(I\) is generated by a single element.
The main goal of this section is to show that the naive derived completion is equal to derived completion if \(A\) is Noetherian.
Lemma
In Situation 0BKC. If \(A\) is Noetherian, then the pro-objects \(\{K_n^\bullet\}\) and \(\{A/(f_1^n, \ldots, f_r^n)\}\) of \(D(A)\) are isomorphic16.
Proof
We have an inverse system of distinguished triangles \[\tau_{\leq -1}K_n^\bullet \to K_n^\bullet \to A/(f_1^n, \ldots, f_r^n) \to (\tau_{\leq -1}K_n^\bullet)[1]\] See Derived Categories, Remark 08J5. By Derived Categories, Lemma 0G3C it suffices to show that the inverse system \(\tau_{\leq -1}K_n^\bullet\) is pro-zero. Recall that \(K_n^\bullet\) has nonzero terms only in degrees \(i\) with \(-r \leq i \leq 0\). Thus by Derived Categories, Lemma 0G3B it suffices to show that \(H^p(K_n^\bullet)\) is pro-zero for \(p \leq -1\). In other words, for every \(n \in \mathbf{N}\) we have to show there exists an \(m \geq n\) such that \(H^p(K_m^\bullet) \to H^p(K_n^\bullet)\) is zero. Since \(A\) is Noetherian, we see that \[H^p(K_n^\bullet) = \frac{\Ker(K_n^p \to K_n^{p + 1})}{\Im(K_n^{p - 1} \to K_n^p)}\] is a finite \(A\)-module. Moreover, the map \(K_m^p \to K_n^p\) is given by a diagonal matrix whose entries are in the ideal \((f_1^{m - n}, \ldots, f_r^{m - n})\) as \(p < 0\). Note that \(H^p(K_n^\bullet)\) is annihilated by \(J = (f_1^n, \ldots, f_r^n)\), see Lemma 0663. Now \((f_1^{m - n}, \ldots, f_r^{m - n}) \subset J^t\) for \(m - n \geq tn\). Thus by Algebra, Lemma 00IN (Artin-Rees) applied to the ideal \(J\) and the module \(M = K_n^p\) with submodule \(N = \Ker(K_n^p \to K_n^{p + 1})\) for \(m\) large enough the image of \(K_m^p \to K_n^p\) intersected with \(\Ker(K_n^p \to K_n^{p + 1})\) is contained in \(J \Ker(K_n^p \to K_n^{p + 1})\). For such \(m\) we get the zero map.
Proposition
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. The functor which sends \(K \in D(A)\) to the derived limit \(K' = R\lim( K \otimes_A^\mathbf{L} A/I^n )\) is the left adjoint to the inclusion functor \(D_{comp}(A) \to D(A)\) constructed in Lemma 091V.
Proof
Say \((f_1, \ldots, f_r) = I\) and let \(K_n^\bullet\) be the Koszul complex with respect to \(f_1^n, \ldots, f_r^n\). By Lemma 0920 it suffices to prove that \[R\lim (K \otimes_A^\mathbf{L} K_n^\bullet) = R\lim (K \otimes_A^\mathbf{L} A/(f_1^n, \ldots, f_r^n) ) = R\lim (K \otimes_A^\mathbf{L} A/I^n ).\] By Lemma 0921 the pro-objects \(\{K_n^\bullet\}\) and \(\{A/(f_1^n, \ldots, f_r^n)\}\) of \(D(A)\) are isomorphic. It is clear that the pro-objects \(\{A/(f_1^n, \ldots, f_r^n)\}\) and \(\{A/I^n\}\) are isomorphic. Thus the map from left to right is an isomorphism by Lemma 091L.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(M\) be an \(A\)-module with derived completion \(M^\wedge\). Then there are short exact sequences \[0 \to R^1\lim \text{Tor}_{i + 1}^A(M, A/I^n) \to H^{-i}(M^\wedge) \to \lim \text{Tor}_i^A(M, A/I^n) \to 0\] A similar result holds for \(M \in D^-(A)\).
Proof
As an application of the proposition above we identify the derived completion in the Noetherian case for pseudo-coherent complexes.
Lemma
Let \(A\) be a Noetherian ring and \(I \subset A\) an ideal. Let \(K\) be an object of \(D(A)\) such that \(H^n(K)\) is a finite \(A\)-module for all \(n \in \mathbf{Z}\). Then the cohomology modules \(H^n(K^\wedge)\) of the derived completion are the \(I\)-adic completions of the cohomology modules \(H^n(K)\).
Proof
The complex \(\tau_{\leq m}K\) is pseudo-coherent for all \(m\) by Lemma 066E. Thus \(\tau_{\leq m}K\) is represented by a bounded above complex \(P^\bullet\) of finite free \(A\)-modules. Then \(\tau_{\leq m}K \otimes_A^\mathbf{L} A/I^n = P^\bullet/I^nP^\bullet\). Hence \((\tau_{\leq m}K)^\wedge = R\lim P^\bullet/I^nP^\bullet\) (Proposition 0922) and since the \(R\lim\) is just given by termwise \(\lim\) (Lemma 091D) and since \(I\)-adic completion is an exact functor on finite \(A\)-modules (Algebra, Lemma 00MB) we conclude the result holds for \(\tau_{\leq m}K\). Hence the result holds for \(K\) as derived completion has finite cohomological dimension, see Lemma 0AAJ.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(M\) be a derived complete \(A\)-module. If \(M/IM\) is a finite \(A/I\)-module, then \(M = \lim M/I^nM\) and \(M\) is a finite \(A^\wedge\)-module.
Proof
Assume \(M/IM\) is finite. Pick \(x_1, \ldots, x_t \in M\) which map to generators of \(M/IM\). We obtain a map \(A^{\oplus t} \to M\) mapping the \(i\)th basis vector to \(x_i\). By Proposition 0922 the derived completion of \(A\) is \(A^\wedge = \lim A/I^n\). As \(M\) is derived complete, we see that our map factors through a map \(q : (A^\wedge)^{\oplus t} \to M\). The module \(\Coker(q)\) is zero by Lemma 09B9. Thus \(M\) is a finite \(A^\wedge\)-module. Since \(A^\wedge\) is Noetherian and complete with respect to \(IA^\wedge\), it follows that \(M\) is \(I\)-adically complete (use Algebra, Lemmas 05GH, 031B, and 00IN).
Lemma
Let \(I\) be an ideal in a Noetherian ring \(A\).
If \(M\) is a finite \(A\)-module and \(N\) is a flat \(A\)-module, then the derived \(I\)-adic completion of \(M \otimes_A N\) is the usual \(I\)-adic completion of \(M \otimes_A N\).
If \(M\) is a finite \(A\)-module and \(f \in A\), then the derived \(I\)-adic completion of \(M_f\) is the usual \(I\)-adic completion of \(M_f\).
Proof
For an \(A\)-module \(M\) denote \(M^\wedge\) the derived completion and \(\lim M/I^nM\) the usual completion. Assume \(M\) is finite. The system \(\text{Tor}^A_i(M, A/I^n)\) is pro-zero for \(i > 0\), see Lemma 0911. Since \(\text{Tor}_i^A(M \otimes_A N, A/I^n) = \text{Tor}_i^A(M, A/I^n) \otimes_A N\) as \(N\) is flat, the same is true for the system \(\text{Tor}^A_i(M \otimes_A N, A/I^n)\). By Lemma 0EET we conclude \(R\lim (M \otimes_A N) \otimes_A^\mathbf{L} A/I^n\) only has cohomology in degree \(0\) given by the usual completion \(\lim M \otimes_A N/ I^n(M \otimes_A N)\). This proves (1). Part (2) follows from (1) and the fact that \(M_f = M \otimes_A A_f\).
Lemma
Let \(I\) be an ideal in a Noetherian ring \(A\). Let \({}^\wedge\) denote derived completion with respect to \(I\). Let \(K \in D^-(A)\).
If \(M\) is a finite \(A\)-module, then \((K \otimes_A^\mathbf{L} M)^\wedge = K^\wedge \otimes_A^\mathbf{L} M\).
If \(L \in D(A)\) is pseudo-coherent, then \((K \otimes_A^\mathbf{L} L)^\wedge = K^\wedge \otimes_A^\mathbf{L} L\).
Proof
Let \(L\) be as in (2). We may represent \(K\) by a bounded above complex \(P^\bullet\) of free \(A\)-modules. We may represent \(L\) by a bounded above complex \(F^\bullet\) of finite free \(A\)-modules. Since \(\text{Tot}(P^\bullet \otimes_A F^\bullet)\) represents \(K \otimes_A^\mathbf{L} L\) we see that \((K \otimes_A^\mathbf{L} L)^\wedge\) is represented by \[\text{Tot}((P^\bullet)^\wedge \otimes_A F^\bullet)\] where \((P^\bullet)^\wedge\) is the complex whose terms are the usual (equivalently, derived) completions \((P^n)^\wedge\), see for example Proposition 0922 and Lemma 0EEU. This proves (2). Part (1) is a special case of (2).
An operator introduced by Berthelot and Ogus
In this section we discuss a construction introduced in [Berthelot-Ogus, Section 8] and generalized in [BMS, Section 6]. We urge the reader to look at the original papers discussing this notion.
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. If \(M\) is a \(A\)-module then by Lemma 05EA following are equivalent
\(f\) is a nonzerodivisor on \(M\),
\(M[f] = 0\),
\(M[f^n] = 0\) for all \(n \geq 1\), and
the map \(M \to M_f\) is injective.
If these equivalent conditions hold, then (in this section) we will say \(M\) is \(f\)-torsion free. If so, then we denote \(f^iM \subset M_f\) the submodule consisting of elements of the form \(f^ix\) with \(x \in M\). Of course \(f^iM\) is isomorphic to \(M\) as an \(A\)-module. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules with differentials \(d^i : M^i \to M^{i + 1}\). In this case we define \(\eta_fM^\bullet\) to be the complex with terms \[(\eta_fM)^i = \{x \in f^iM^i \mid d^i(x) \in f^{i + 1}M^{i + 1}\}\] and differential induced by \(d^i\). Observe that \(\eta_fM^\bullet\) is another complex of \(f\)-torsion free \(A\)-modules. If \(a^\bullet : M^\bullet \to N^\bullet\) is a map of complexes of \(f\)-torsion free \(A\)-modules, then we obtain a map of complexes \[\eta_fa^\bullet : \eta_fM^\bullet \longrightarrow \eta_fN^\bullet\] induced by the maps \(f^iM^i \to f^iN^i\). The reader checks that we obtain an endo-functor on the category of complexes of \(f\)-torsion free \(A\)-modules. If \(a^\bullet, b^\bullet : M^\bullet \to N^\bullet\) are two maps of complexes of \(f\)-torsion free \(A\)-modules and \(h = \{h^i : M^i \to N^{i - 1}\}\) is a homotopy between \(a^\bullet\) and \(b^\bullet\), then we define \(\eta_fh\) to be the family of maps \((\eta_fh)^i : (\eta_fM)^i \to (\eta_fN)^{i - 1}\) which sends \(x\) to \(h^i(x)\); this makes sense as \(x \in f^iM^i\) implies \(h^i(x) \in f^iN^{i - 1}\) which is certainly contained in \((\eta_fN)^{i - 1}\). The reader checks that \(\eta_fh\) is a homotopy between \(\eta_fa^\bullet\) and \(\eta_fb^\bullet\). All in all we see that we obtain a functor \[\eta_f : K(f\text{-torsion free }A\text{-modules}) \longrightarrow K(f\text{-torsion free }A\text{-modules})\] on the homotopy category (Derived Categories, Section 05RN) of the additive category of \(f\)-torsion free \(A\)-modules. There is no sense in which \(\eta_f\) is an exact functor of triangulated categories, see Example 0GSN.
Example
Let \(A\) be a ring. Let \(f \in A\) be a nonzerodivisor. Consider the functor \(\eta_f : K(f\text{-torsion free }A\text{-modules}) \to K(f\text{-torsion free }A\text{-modules})\). Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. Multiplication by \(f\) defines an isomorphism \(\eta_f(M^\bullet[1]) \to (\eta_fM^\bullet)[1]\), so in this sense \(\eta_f\) is compatible with shifts. However, consider the diagram \[\xymatrix{ A \ar[r]_f & A \ar[r]_1 & A \ar[r] & 0 \\ 0 \ar[r] \ar[u] & 0 \ar[r] \ar[u] & A \ar[r]^{-1} \ar[u]^f & A \ar[u] }\] Think of each column as a complex of \(f\)-torsion free \(A\)-modules with the module on top in degree \(1\) and the module under it in degree \(0\). Then this diagram provides us with a distinguished triangle in \(K(f\text{-torsion free }A\text{-modules})\) with triangulated structure as given in Derived Categories, Section 014P. Namely the third complex is the cone of the map between the first two complexes. However, applying \(\eta_f\) to each column we obtain \[\xymatrix{ fA \ar[r]_f & fA \ar[r]_1 & fA \ar[r] & 0 \\ 0 \ar[r] \ar[u] & 0 \ar[r] \ar[u] & A \ar[r]^{-1} \ar[u]^f & A \ar[u] }\] However, the third complex is acyclic and even homotopic to zero. Hence if this were a distinguished triangle, then the first arrow would have to be an isomorphism in the homotopy category, which is not true unless \(f\) is a unit.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. There is a canonical isomorphism \[f^i : H^i(M^\bullet)/H^i(M^\bullet)[f] \longrightarrow H^i(\eta_fM^\bullet)\] given by multiplication by \(f^i\).
Proof
Observe that \(\Ker(d^i : (\eta_fM)^i \to (\eta_fM)^{i + 1})\) is equal to \(\Ker(d^i : f^iM^i \to f^iM^{i + 1}) = f^i\Ker(d^i : M^i \to M^{i + 1})\). Thus we get a surjection \(f^i : H^i(M^\bullet) \to H^i(\eta_fM^\bullet)\) by sending the class of \(z \in \Ker(d^i : M^i \to M^{i + 1})\) to the class of \(f^iz\). If we obtain the zero class in \(H^i(\eta_fM^\bullet)\) then we see that \(f^i z = d^{i - 1}(f^{i - 1}y)\) for some \(y \in M^{i - 1}\). Since \(f\) is a nonzerodivisor on all the modules involved, this means \(f z = d^{i - 1}(y)\) which exactly means that the class of \(z\) is \(f\)-torsion as desired.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. If \(M^\bullet \to N^\bullet\) is a quasi-isomorphism of complexes of \(f\)-torsion free \(A\)-modules, then the induced map \(\eta_fM^\bullet \to \eta_fN^\bullet\) is a quasi-isomorphism too.
Proof
This is true because the isomorphisms of Lemma 0F7P are compatible with maps of complexes.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. There is an additive functor17 \(L\eta_f : D(A) \to D(A)\) such that if \(M \in D(A)\) is represented by a complex \(M^\bullet\) of \(f\)-torsion free \(A\)-modules, then \(L\eta_fM = \eta_fM^\bullet\) and similarly for morphisms.
Proof
Denote \(\mathcal{T} \subset \text{Mod}_A\) the full subcategory of \(f\)-torsion free \(A\)-modules. We have a corresponding inclusion \[K(\mathcal{T}) \quad\subset\quad K(\text{Mod}_A) = K(A)\] of \(K(\mathcal{T})\) as a full triangulated subcategory of \(K(A)\). Let \(S \subset \text{Arrows}(K(\mathcal{T}))\) be the quasi-isomorphisms. We will apply Derived Categories, Lemma 0GSL to show that the map \[S^{-1}K(\mathcal{T}) \longrightarrow D(A)\] is an equivalence of triangulated categories. The lemma shows that it suffices to prove: given a complex \(M^\bullet\) of \(A\)-modules, there exists a quasi-isomorphism \(K^\bullet \to M^\bullet\) with \(K^\bullet\) a complex of \(f\)-torsion free modules. By Lemma 06Y4 we can find a quasi-isomorphism \(K^\bullet \to M^\bullet\) such that the complex \(K^\bullet\) is K-flat (we won’t use this) and consists of flat \(A\)-modules \(K^i\). In particular, \(f\) is a nonzerodivisor on \(K^i\) for all \(i\) as desired.
With these preliminaries out of the way we can define \(L\eta_f\). Namely, by the discussion at the start of this section we have already a well defined functor \[K(\mathcal{T}) \xrightarrow{\eta_f} K(\mathcal{T}) \to K(A) \to D(A)\] which according to Lemma 0F7Q sends quasi-isomorphisms to quasi-isomorphisms. Hence this functor factors over \(S^{-1}K(\mathcal{T}) = D(A)\) by Categories, Lemma 04VG.
Remark
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. For every \(i\) set \(\overline{M}^i = M^i/fM^i\). Denote \(B^i \subset Z^i \subset \overline{M}^i\) the boundaries and cocycles for the differentials on the complex \(\overline{M}^\bullet = M^\bullet \otimes_A A/fA\). We claim that there exists a commutative diagram \[\xymatrix{ 0 \ar[r] & B^{i + 1} \ar[r] \ar@{=}[d] & B^{i + 1} \oplus B^i \ar[r] \ar[d]^{s, s'} & B^i \ar[r] \ar[d] & 0 \\ 0 \ar[r] & B^{i + 1} \ar[r]^-s & (\eta_fM)^i /f(\eta_fM)^i \ar[r]^-t & Z^i \ar[r] & 0 }\] with exact rows. Here are the constructions of the maps
If \(x \in (\eta_fM)^i\) then \(x = f^ix'\) with \(d^i(x') = 0\) in \(\overline{M}^{i + 1}\). Hence we can define the map \(t\) by sending \(x\) to the class of \(x'\).
If \(y \in M^{i + 1}\) has class \(\overline{y}\) in \(B^{i + 1} \subset \overline{M}^{i + 1}\) then we can write \(y = fy' + d^i(x)\) for \(y' \in M^{i + 1}\) and \(x \in M^i\). Hence we can define the map \(s\) sending \(\overline{y}\) to the class of \(f^{i + 1}x\) in \((\eta_fM)^i /f(\eta_fM)^i\); we omit the verification that this is well defined.
If \(x \in M^i\) has class \(\overline{x}\) in \(B^i \subset \overline{M}^i\) then we can write \(x = fx' + d^{i - 1}(z)\) for \(x' \in M^i\) and \(z \in M^{i - 1}\). We define the map \(s'\) by sending \(\overline{x}\) to the class of \(f^i d^{i - 1}(z)\) in \((\eta_fM)^i/f(\eta_fM)^i\). This is well defined because if \(fx' + d^{i - 1}(z) = 0\), then \(f^ix'\) is in \((\eta_fM)^i\) and consequently \(f^id^{i - 1}(z)\) is in \(f(\eta_fM)^i\).
We omit the verification that the lower row in the displayed diagram is a short exact sequence of modules. It is immediately clear from these constructions that we have commutative diagrams \[\xymatrix{ B^{i + 1} \oplus B^i \ar[d]^{s, s'} \ar[r] & B^{i + 2} \oplus B^{i + 1} \ar[d]^{s, s'} \\ (\eta_fM)^i /f(\eta_fM)^i \ar[r] & (\eta_fM)^{i + 1} /f(\eta_fM)^{i + 1} }\] where the upper horizontal arrow is given by the identification of the summands \(B^{i + 1}\) in source and target. In other words, we have found an acyclic subcomplex of \(\eta_fM^\bullet / f(\eta_fM^\bullet) = \eta_fM^\bullet \otimes_A A/fA\) and the quotient by this subcomplex is a complex whose terms \(Z^i/B^i\) are the cohomology modules of the complex \(\overline{M}^\bullet = M^\bullet \otimes_A A/fA\).
To explain the phenomenon observed in Remark 0F7S in a more canonical manner, we are going to construct the Bockstein operators. Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. For every \(i \in \mathbf{Z}\) there is a commutative diagram (with tensor products over \(A\)) \[\xymatrix{ 0 \ar[r] & M^\bullet \otimes f^{i + 1}A \ar[r] \ar[d] & M^\bullet \otimes f^iA \ar[r] \ar[d] & M^\bullet \otimes f^iA/f^{i + 1}A \ar[r] \ar@{=}[d] & 0 \\ 0 \ar[r] & M^\bullet \otimes f^{i + 1}A/f^{i + 2}A \ar[r] & M^\bullet \otimes f^iA/f^{i + 2}A \ar[r] & M^\bullet \otimes f^iA/f^{i + 1}A \ar[r] & 0 }\] whose rows are short exact sequences of complexes. Of course these short exact sequences for different \(i\) are all isomorphic to each other by suitably multiplying with powers of \(f\). The long exact sequence of cohomology of the bottom sequence in particular determines the Bockstein operator \[\beta = \beta^i : H^i(M^\bullet \otimes f^iA/f^{i + 1}A) \to H^{i + 1}(M^\bullet \otimes f^{i + 1}A/f^{i + 2}A)\] for all \(i \in \mathbf{Z}\). For later use we record here that by the commutative diagram above there is a factorization [0GSP]\[\begin{equation} \vcenter{ \xymatrix{ H^i(M^\bullet \otimes f^iA/f^{i + 1}A) \ar[r]_\delta \ar[rd]_\beta & H^{i + 1}(M^\bullet \otimes f^{i + 1}A) \ar[d] \\ & H^{i + 1}(M^\bullet \otimes f^{i + 1}A/f^{i + 2}A) } } \end{equation}\] of the Bockstein operator where \(\delta\) is the boundary operator coming from the top row in the commutative diagram above. Let us show that we obtain a complex [0GSQ]\[\begin{equation} H^\bullet(M^\bullet/f) = \left[ \begin{matrix} \ldots \\ \downarrow \\ H^{i - 1}(M^\bullet \otimes f^{i - 1}A/f^iA) \\ \downarrow \beta \\ H^i(M^\bullet \otimes f^iA/f^{i + 1}A) \\ \downarrow \beta \\ H^{i + 1}(M^\bullet \otimes f^{i + 1}A/f^{i + 2}A) \\ \downarrow \\ \ldots \end{matrix} \right] \end{equation}\] i.e., that \(\beta \circ \beta = 0\)18. Namely, using the factorization (0GSP) we see that it suffices to show that \[H^{i + 1}(M^\bullet \otimes f^{i + 1}A) \to H^{i + 1}(M^\bullet \otimes f^{i + 1}A/f^{i + 2}A) \xrightarrow{\beta^{i + 1}} H^{i + 2}(M^\bullet \otimes f^{i + 2}A/f^{i + 3}A)\] is zero. This is true because the kernel of \(\beta^{i + 1}\) consists of the cohomology classes which can be lifted to \(H^{i + 1}(M^\bullet \otimes f^{i + 1}A/f^{i + 3}A)\) and those in the image of the first map certainly can!
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. There is a canonical map of complexes \[\eta_fM^\bullet \otimes_A A/fA \longrightarrow H^\bullet(M^\bullet/f)\] which is a quasi-isomorphism where the right hand side is the complex (0GSQ).
Proof
Let \(x \in (\eta_fM)^i\). Then \(x = f^ix' \in f^iM\) and \(d^i(x) = f^{i + 1}y \in f^{i + 1}M^{i + 1}\). Thus \(d^i\) maps \(x' \otimes f^i\) to zero in \(M^{i + 1} \otimes f^iA/f^{i + 1}A\). All tensor products are over \(A\) in this proof. Hence we may map \(x\) to the class of \(x' \otimes f^i\) in \(H^i(M^\bullet \otimes f^iA/f^{i + 1}A)\). It is clear that this rule defines a map \[(\eta_fM)^i \otimes A/fA \longrightarrow H^i(M^\bullet \otimes f^iA/f^{i + 1}A)\] of \(A/fA\)-modules. Observe that in the situation above, we may view \(x' \otimes f^i\) as an element of \(M^i \otimes f^iA/f^{i + 2}A\) with differential \(d^i(x' \otimes f^i) = y \otimes f^{i + 1}\). By the construction of \(\beta\) above we find that \(\beta(x' \otimes f^i) = y \otimes f^{i + 1}\) and we conclude that our maps are compatible with differentials, i.e., we have a map of complexes.
To finish the proof, we observe that the construction given in the previous paragraph agrees with the maps \((\eta_fM)^i \otimes A/fA \to Z^i/B^i\) discussed in Remark 0F7S. Since we have seen that the kernel of these maps is an acyclic subcomplex of \(\eta_fM^\bullet \otimes A/fA\), the lemma is proved.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. For \(i \in \mathbf{Z}\) the following are equivalent
\(\Ker(d^i \bmod f^2)\) surjects onto \(\Ker(d^i \bmod f)\),
\(\beta : H^i(M^\bullet \otimes_A f^iA/f^{i + 1}A) \to H^{i + 1}(M^\bullet \otimes_A f^{i + 1}A/f^{i + 2}A)\) is zero.
These equivalent conditions are implied by the condition \(H^{i + 1}(M^\bullet)[f] = 0\).
Proof
The equivalence of (1) and (2) follows from the definition of \(\beta\) as the boundary map on cohomology of a short exact sequence of complexes isomorphic to the short exact sequence of complexes \(0 \to fM^\bullet/f^2M^\bullet \to M^\bullet/f^2M^\bullet \to M^\bullet/fM^\bullet \to 0\). If \(\beta \not = 0\), then \(H^{i + 1}(M^\bullet)[f] \not = 0\) because of the factorization (0GSP).
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. If \(\Ker(d^i \bmod f^2)\) surjects onto \(\Ker(d^i \bmod f)\), then the canonical map \[(1, d^i) : (\eta_fM)^i / f(\eta_fM)^i \longrightarrow f^iM^i/f^{i + 1}M^i \oplus f^{i + 1}M^{i + 1}/f^{i + 2}M^{i + 1}\] identifies the left hand side with a direct sum of submodules of the right hand side.
Proof
With notation as in Remark 0F7S we define a map \(t^{-1} : Z^i \to (\eta_fM)^i / f(\eta_fM)^i\). Namely, for \(x \in M^i\) with \(d^i(x) = f^2y\) we send the class of \(x\) in \(Z^i\) to the class of \(f^ix\) in \((\eta_fM)^i / f(\eta_fM)^i\). We omit the verification that this is well defined; the assumption of the lemma exactly signifies that the domain of this operation is all of \(Z^i\). Then \(t \circ t^{-1} = \text{id}_{Z^i}\). Hence \(t^{-1}\) defines a splitting of the short exact sequence in Remark 0F7S and the resulting direct sum decomposition \[(\eta_fM)^i / f(\eta_fM)^i = Z^i \oplus B^{i + 1}\] is compatible with the map displayed in the lemma.
Lemma
Let \(A\) be a ring and let \(f, g \in A\) be nonzerodivisors. Let \(M^\bullet\) be a complex of \(A\)-modules such that \(fg\) is a nonzerodivisor on all \(M^i\). Then \(\eta_f\eta_gM^\bullet = \eta_{fg}M^\bullet\).
Proof
The statement means that in degree \(i\) we obtain the same submodule of the localization \(M^i_{fg} = (M^i_g)_f\). We omit the details.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(A \to B\) be a flat ring map and let \(g \in B\) the image of \(f\). Let \(M^\bullet\) be a complex of \(f\)-torsion free \(A\)-modules. Then \(g\) is a nonzerodivisor, \(M^\bullet \otimes_A B\) is a complex of \(g\)-torsion free modules, and \(\eta_fM^\bullet \otimes_A B = \eta_g(M^\bullet \otimes_A B)\).
Proof
Omitted.
Perfect complexes and the eta operator
In this section we do some algebra to prepare for our version of Macpherson’s graph construction, see More on Flatness, Section 0F8Z. We will use the \(\eta_f\) operator introduced in Section 0F7N.
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. For each \(i\) let \(r_i\) be the rank of \(M^i\) and set \[I_i(M^\bullet, f) = \text{ideal generated by the } r_i \times r_i\text{-minors of } (f, d^i) : M^i \to M^i \oplus M^{i + 1}\] Observe that \(f^{r_i} \in I_i(M^\bullet, f)\).
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) and \(N^\bullet\) be two bounded complexes of finite free \(A\)-modules representing the same object of \(D(A)\). Then \[f^m I_i(M^\bullet, f) = f^n I_i(N^\bullet, f)\] as ideals of \(A\) for integers \(n, m \geq 0\) such that \[m + \sum\nolimits_{j \geq i} (-1)^{j - i}rk(M^j) = n + \sum\nolimits_{j \geq i} (-1)^{j - i}rk(N^j)\]
Proof
It suffices to prove the equality after localization at every prime ideal of \(A\). Thus by Lemma 0F9V and an induction argument we omit we may assume \(N^\bullet = M^\bullet \oplus Q^\bullet\) for some trivial complex \(Q^\bullet\), i.e., \[Q^\bullet = \ldots \to 0 \to A \xrightarrow{1} A \to 0 \to \ldots\] where \(A\) is placed in degree \(j\) and \(j + 1\). If \(j \not = i - 1, i, i + 1\) then we clearly have equality \(I_i(M^\bullet, f) = I_i(N^\bullet, f)\) and \(m = n\) and we have the desired equality. If \(j = i + 1\) then the maps \[(f, d^i) : M^i \to M^i \oplus M^{i + 1} \quad\text{and}\quad (f, d^i, 0) : M^i \to M^i \oplus M^{i + 1} \oplus A\] have the same nonzero minors hence in this case we also have \(I_i(M^\bullet, f) = I_i(N^\bullet, f)\) and \(m = n\). If \(j = i\), then \(I_i(M^\bullet, f)\) is the ideal generated by the \(r_i \times r_i\)-minors of \[(f, d^i) : M^i \to M^i \oplus M^{i + 1}\] and \(I_i(N^\bullet, f)\) is the ideal generated by the \((r_i + 1) \times (r_i + 1)\)-minors of \[(f \oplus f, d^i \oplus 1) : (M^i \oplus A) \to (M^i \oplus A) \oplus (M^{i + 1} \oplus A)\] With suitable choice of coordinates we see that the matrix of the second map is in block form \[T = \left( \begin{matrix} T_1 & 0 \\ 0 & T_2 \end{matrix} \right), \quad T_1 = \text{matrix of first map}, \quad T_2 = \left( \begin{matrix} f \\ 1 \end{matrix} \right)\] With notation as in Lemma 07Z7 we have \(I_0(T_2) = A\), \(I_1(T_2) = A\), \(I_p(T_2) = 0\) for \(p \geq 2\) and hence \(I_{r_i + 1}(T) = I_{r_i + 1}(T_1) + I_{r_i}(T_1) = I_{r_i}(T_1)\) which means that \(I_i(M^\bullet, f) = I_i(N^\bullet, f)\). We also have \(m = n\) so this finishes the case \(j = i\). Finally, say \(j = i - 1\). Then we see that \(m = n + 1\), thus we have to show that \(fI_i(M^\bullet, f) = I_i(N^\bullet, f)\). In this case \(I_i(M^\bullet, f)\) is the ideal generated by the \(r_i \times r_i\)-minors of \[(f, d^i) : M^i \to M^i \oplus M^{i + 1}\] and \(I_i(N^\bullet, f)\) is the ideal generated by the \((r_i + 1) \times (r_i + 1)\)-minors of \[(f \oplus f, d^i) : (M^i \oplus A) \to (M^i \oplus A) \oplus M^{i + 1}\] With suitable choice of coordinates we see that the matrix of the second map is in block form \[T = \left( \begin{matrix} T_1 & 0 \\ 0 & T_2 \end{matrix} \right), \quad T_1 = \text{matrix of first map}, \quad T_2 = \left( \begin{matrix} f \end{matrix} \right)\] Arguing as above we find that indeed \(fI_i(M^\bullet, f) = I_i(N^\bullet, f)\).
Lemma
Let \(f \in A\) be a nonzerodivisor of a ring \(A\). Let \(u \in A\) be a unit. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Then \(I_i(M^\bullet, f) = I_i(M^\bullet, uf)\).
Proof
Omitted.
Lemma
Let \(A \to B\) be a ring map. Let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Assume \(f\) maps to a nonzerodivisor \(g\) in \(B\). Then \(I_i(M^\bullet, f)B = I_i(M^\bullet \otimes_A B, g)\).
Proof
The minors of \((f, d^i) : M^i \to M^i \oplus M^{i + 1}\) map to the corresponding minors of \((g, d^i) : M^i \otimes_A B \to M^i \otimes_A B \oplus M^{i + 1} \otimes_A B\).
Lemma
Let \(A\) be a ring, let \(\mathfrak p \subset A\) be a prime ideal, and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. If \(H^i(M^\bullet)_\mathfrak p\) is free for all \(i\), then \(I_i(M^\bullet, f)_\mathfrak p\) is a principal ideal and in fact generated by a power of \(f\) for all \(i\).
Proof
We may assume \(A\) is local with maximal ideal \(\mathfrak p\) by Lemma 0GSU. We may also replace \(M^\bullet\) with a quasi-isomorphic complex by Lemma 0GSS. By our assumption on the freeness of cohomology modules we see that \(M^\bullet\) is quasi-isomorphic to the complex whose term in degree \(i\) is \(H^i(M^\bullet)\) with vanishing differentials, see for example Derived Categories, Lemma 0GM4. In other words, we may assume the differentials in the complex \(M^\bullet\) are all zero. In this case it is clear that \(I_i(M^\bullet, f) = (f^{r_i})\) is principal.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Assume \(I_i(M^\bullet, f)\) is a principal ideal. Then \((\eta_fM)^i\) is locally free of rank \(r_i\) and the map \((1, d^i) : (\eta_fM)^i \to f^iM^i \oplus f^{i + 1}M^{i + 1}\) is the inclusion of a direct summand.
Proof
Choose a generator \(g\) for \(I_i(M^\bullet, f)\). Since \(f^{r_i} \in I_i(M^\bullet, f)\) we see that \(g\) divides a power of \(f\). In particular \(g\) is a nonzerodivisor in \(A\). The \(r_i \times r_i\)-minors of the map \((f, d^i) : M^i \to M^i \oplus M^{i + 1}\) generate the ideal \(I_i(M^\bullet, f)\) and the \((r_i + 1) \times (r_i + 1)\)-minors of \((f, d^i)\) are zero: we may check this after localizing at \(f\) where the rank of the map is equal to \(r_i\). Consider the surjection \[M^i \oplus M^{i + 1} \longrightarrow Q = \Coker(f, d^i)/g\text{-torsion}\] By Lemma 0F7M the module \(Q\) is finite locally free of rank \(r_{i + 1}\). Hence \(Q\) is \(f\)-torsion free and we conclude the cokernel of \((f, d^i)\) modulo \(f\)-power torsion is \(Q\) as well.
Consider the complex of finite free \(A\)-modules \[0 \to f^{i + 1}M^i \xrightarrow{1, d^i} f^iM^i \oplus f^{i + 1}M^{i + 1} \xrightarrow{d^i, -1} f^iM^{i + 1} \to 0\] which becomes split exact after localizing at \(f\). The map \((1, d^i) : f^{i + 1}M^i \to f^iM^i \oplus f^{i + 1}M^{i + 1}\) is isomorphic to the map \((f, d^i) : M^i \to M^i \oplus M^{i + 1}\) we studied above. Hence the image \[Q' = \Im(f^iM^i \oplus f^{i + 1}M^{i + 1} \xrightarrow{d^i, -1} f^iM^{i + 1})\] is isomorphic to \(Q\) in particular projective. On the other hand, by construction of \(\eta_f\) in Section 0F7N the image of the injective map \((1, d^i) : (\eta_fM)^i \to f^iM^i \oplus f^{i + 1}M^{i + 1}\) is the kernel of \((d^i, -1)\). We conclude that we obtain an isomorphism \((\eta_fM)^i \oplus Q' = f^iM^i \oplus f^{i + 1}M^{i + 1}\) and we see that indeed \((\eta_fM)^i\) is finite locally free of rank \(r_i\) and that \((1, d^i)\) is the inclusion of a direct summand.
Lemma
Let \(A \to B\) be a ring map. Let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Assume \(f\) maps to a nonzerodivisor \(g\) in \(B\) and \(I_i(M^\bullet, f)\) is a principal ideal for all \(i \in \mathbf{Z}\). Then there is a canonical isomorphism \(\eta_fM^\bullet \otimes_A B = \eta_g(M^\bullet \otimes_A B)\).
Proof
Set \(N^i = M^i \otimes_A B\). Observe that \(f^iM^i \otimes_A B = g^iN^i\) as submodules of \((N^i)_g\). The maps \[(\eta_fM)^i \otimes_A B \to g^iN^i \oplus g^{i + 1}N^{i + 1} \quad\text{and}\quad (\eta_gN)^i \to g^iN^i \oplus g^{i + 1}N^{i + 1}\] are inclusions of direct summands by Lemma 0F7W. Since their images agree after localizing at \(g\) we conclude.
Lemma
Let \(A\) be a ring. Let \(M\), \(N_1\), \(N_2\) be finite projective \(A\)-modules. Let \(s : M \to N_1 \oplus N_2\) be a split injection. There exists a finitely generated ideal \(J \subset A\) with the following property: a ring map \(A \to B\) factors through \(A/J\) if and only if \(s \otimes \text{id}_B\) identifies \(M \otimes_A B\) with a direct sum of submodules of \(N_1 \otimes_A B \oplus N_2 \otimes_A B\).
Proof
Choose a splitting \(\pi : N_1 \oplus N_2 \to M\) of \(s\). Denote \(q_i : N_1 \oplus N_2 \to N_1 \oplus N_2\) the projector onto \(N_i\). Set \(p_i = \pi \circ q_i \circ s\). Observe that \(p_1 + p_2 = \text{id}_M\). We claim \(M\) is a direct sum of submodules of \(N_1 \oplus N_2\) if and only if \(p_1\) and \(p_2\) are orthogonal projectors. Thus \(J\) is the smallest ideal of \(A\) such that \(p_1 \circ p_1 - p_1\), \(p_2 \circ p_2 - p_2\), \(p_1 \circ p_2\), and \(p_2 \circ p_1\) are contained in \(J \otimes_A \text{End}_A(M)\). Some details omitted.
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Assume the ideals \(I_i(M^\bullet, f)\) are principal for all \(i \in \mathbf{Z}\). Then the maps \[(1, d^i) : (\eta_fM)^i / f(\eta_fM)^i \longrightarrow f^iM^i/f^{i + 1}M^i \oplus f^{i + 1}M^{i + 1}/f^{i + 2}M^{i + 1}\] are split injections by Lemma 0F7W. Denote \(J_i(M^\bullet, f) \subset A/fA\) the finitely generated ideal of Lemma 0F80 corresponding to the split injection \((1, d^i)\) displayed above.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) and \(N^\bullet\) be two bounded complexes of finite free \(A\)-modules representing the same object in \(D(A)\). Assume \(I_i(M^\bullet, f)\) is a principal ideal for all \(i \in \mathbf{Z}\). Then \(J_i(M^\bullet, f) = J_i(N^\bullet, f)\) as ideals in \(A/fA\).
Proof
Observe that the fact that \(I_i(M^\bullet, f)\) is a principal ideal implies that \(I_i(N^\bullet, f)\) is a principal ideal by Lemma 0GSS and hence the statement makes sense. As in the proof of Lemma 0GSS we may assume \(N^\bullet = M^\bullet \oplus Q^\bullet\) for some trivial complex \(Q^\bullet\), i.e., \[Q^\bullet = \ldots \to 0 \to A \xrightarrow{1} A \to 0 \to \ldots\] where \(A\) is placed in degree \(j\) and \(j + 1\). Since \(\eta_f\) is compatible with direct sums, we see that the map \[(1, d^i) : (\eta_fN)^i / f(\eta_fN)^i \longrightarrow f^iN^i/f^{i + 1}N^i \oplus f^{i + 1}N^{i + 1}/f^{i + 2}N^{i + 1}\] is the direct sum of the corresponding map for \(M^\bullet\) and for \(Q^\bullet\). By the universal property defining the ideals in question, we conclude that \(J_i(N^\bullet, f) = J_i(M^\bullet, f) + J_i(Q^\bullet, f)\). Hence it suffices to show that \(J_i(Q^\bullet, f) = 0\) for all \(i\). This is a computation that we omit.
Lemma
Let \(A\) be a ring and let \(f \in A\) be a nonzerodivisor. Let \(M^\bullet\) be a bounded complex of finite free \(A\)-modules. Assume \(I_i(M^\bullet, f)\) is a principal ideal for all \(i \in \mathbf{Z}\). Consider the ideal \(J(M^\bullet, f) = \sum_i J_i(M^\bullet, f)\) of \(A/fA\). Consider the set of prime ideals \[\begin{align*} E & = \{f \in \mathfrak p \subset A \mid \Ker(d^i \bmod f^2)_\mathfrak p \text{ surjects onto } \Ker(d^i \bmod f)_\mathfrak p \text{ for all }i \in \mathbf{Z}\} \\ & = \{f \in \mathfrak p \subset A \mid \text{the localizations }\beta_\mathfrak p \text{ of the Bockstein operators are zero}\} \end{align*}\] Then we have
\(J(M^\bullet, f)\) is finitely generated,
\(A/fA \to C = (A/fA)/J(M^\bullet, f)\) is surjective of finite presentation,
\(J(M^\bullet, f)_\mathfrak p = 0\) for \(\mathfrak p \in E\),
if \(f \in \mathfrak p\) and \(H^i(M^\bullet)_\mathfrak p\) is free for all \(i \in \mathbf{Z}\), then \(\mathfrak p \in E\), and
the cohomology modules of \(\eta_f M^\bullet \otimes_A C\) are finite locally free \(C\)-modules.
Proof
The equality in the definition of \(E\) follows from Lemma 0F7Y and in addition the final statement of that lemma implies part (4).
Part (1) is true because the ideals \(J_i(M^\bullet, f)\) are finitely generated and because \(M^\bullet\) is bounded and hence \(J_i(M^\bullet, f)\) is zero for almost all \(i\). Part (2) is just a reformulation of part (1).
Proof of (3). By Lemma 0F7W we find that \((\eta_fM)^i\) is finite locally free of rank \(r_i\) for all \(i\). Consider the map \[(1, d^i) : (\eta_fM)^i / f(\eta_fM)^i \longrightarrow f^iM^i/f^{i + 1}M^i \oplus f^{i + 1}M^{i + 1}/f^{i + 2}M^{i + 1}\] Pick \(\mathfrak p \in E\). By Lemma 0F7Z and the local freeness of the modules \((\eta_fM)^i\) we may write \[\left((\eta_fM)^i / f(\eta_fM)^i\right)_\mathfrak p = (A/fA)_\mathfrak p^{\oplus m_i} \oplus (A/fA)_\mathfrak p^{\oplus n_i}\] compatible with the arrow \((1, d^i)\) above. By the universal property of the ideal \(J_i(M^\bullet, f)\) we conclude that \(J_i(M^\bullet, f)_\mathfrak p = 0\). Hence \(J(M^\bullet, f)_\mathfrak p = 0\) for \(\mathfrak p \in E\).
Proof of (5). Observe that the differential on \(\eta_fM^\bullet\) fits into a commutative diagram \[\xymatrix{ (\eta_fM)^i \ar[d] \ar[r] & f^iM^i \oplus f^{i + 1}M^{i + 1} \ar[d]^{\left( \begin{matrix} 0 & 1 \\ 0 & 0 \end{matrix} \right)} \\ (\eta_fM)^{i + 1} \ar[r] & f^{i + 1}M^{i + 1} \oplus f^{i + 2}M^{i + 2} }\] By construction, after tensoring with \(C\), the modules on the left are direct sums of direct summands of the summands on the right. Picture \[\xymatrix{ (\eta_fM)^i \otimes_A C \ar[d] \ar@{=}[r] & K^i \oplus L^i \ar[r] \ar[d] & f^iM^i \otimes_A C \oplus f^{i + 1}M^{i + 1} \otimes_A C \ar[d]^{\left( \begin{matrix} 0 & 1 \\ 0 & 0 \end{matrix} \right)} \\ (\eta_fM)^{i + 1} \otimes_A C \ar@{=}[r] & K^{i + 1} \oplus L^{i + 1} \ar[r] & f^{i + 1}M^{i + 1} \otimes_A C \oplus f^{i + 2}M^{i + 2} \otimes_A C }\] where the horizontal arrows are compatible with direct sum decompositions as well as inclusions of direct summands. It follows that the differential identifies \(L^i\) with a direct summand of \(K^{i + 1}\) and we conclude that the cohomology of \(\eta_fM^\bullet \otimes_A C\) in degree \(i\) is the module \(K^i/L^{i - 1}\) which is finite projective as desired.
Taking limits of complexes
In this section we discuss what happens when we have a “formal deformation” of a complex and we take its limit. We will consider two cases
we have a limit \(A = \lim A_n\) of an inverse system of rings whose transition maps are surjective with locally nilpotent kernels and objects \(K_n \in D(A_n)\) which fit together in the sense that \(K_n = K_{n + 1} \otimes_{A_{n + 1}}^\mathbf{L} A_n\), or
we have a ring \(A\), an ideal \(I\), and objects \(K_n \in D(A/I^n)\) which fit together in the sense that \(K_n = K_{n + 1} \otimes_{A/I^{n + 1}}^\mathbf{L} A/I^n\).
Under additional hypotheses we can show that \(K = R\lim K_n\) reproduces the system in the sense that \(K_n = K \otimes_A^\mathbf{L} A_n\) or \(K_n = K \otimes_A^\mathbf{L} A/I^n\).
Lemma
Let \(A = \lim A_n\) be a limit of an inverse system \((A_n)\) of rings. Suppose given \(K_n \in D(A_n)\) and maps \(K_{n + 1} \to K_n\) in \(D(A_{n + 1})\). Assume
the transition maps \(A_{n + 1} \to A_n\) are surjective with locally nilpotent kernels,
either all \(K_n\) are pseudo-coherent or there exists an integer \(n_0\) such that \(K_{n_0}\) is pseudo-coherent and the kernels of \(A_{n + 1} \to A_n\) are nilpotent ideals for \(n \geq n_0\),
the maps induce isomorphisms \(K_{n + 1} \otimes_{A_{n + 1}}^\mathbf{L} A_n \to K_n\).
Then \(K = R\lim K_n\) is a pseudo-coherent object of \(D(A)\) and \(K \otimes_A^\mathbf{L} A_n \to K_n\) is an isomorphism for all \(n\).
Proof
By assumption we can find a bounded above complex of finite free \(A_1\)-modules \(P_1^\bullet\) representing \(K_1\), see Definition 064Q. By Lemma 0H76 we conclude that \(K_n\) is pseudo-coherent for all \(n\). Then by Lemma 0BCB we can, by induction on \(n > 1\), find complexes \(P_n^\bullet\) of finite free \(A_n\)-modules representing \(K_n\) and maps \(P_n^\bullet \to P_{n - 1}^\bullet\) representing the maps \(K_n \to K_{n - 1}\) inducing isomorphisms (!) of complexes \(P_n^\bullet \otimes_{A_n} A_{n - 1} \to P_{n - 1}^\bullet\). Thus \(K = R\lim K_n\) is represented by \(P^\bullet = \lim P_n^\bullet\), see Lemma 091D and Remark 07KZ. Since \(P_n^i\) is a finite free \(A_n\)-module for each \(n\) and \(A = \lim A_n\) we see that \(P^i\) is finite free of the same rank as \(P_1^i\) for each \(i\). This means that \(K\) is pseudo-coherent. It also follows that \(K \otimes_A^\mathbf{L} A_n\) is represented by \(P^\bullet \otimes_A A_n = P_n^\bullet\) which proves the final assertion.
Lemma
Let \(A\) be a ring and \(I \subset A\) an ideal. Suppose given \(K_n \in D(A/I^n)\) and maps \(K_{n + 1} \to K_n\) in \(D(A/I^{n + 1})\). Assume
\(A\) is \(I\)-adically complete,
\(K_1\) is pseudo-coherent, and
the maps induce isomorphisms \(K_{n + 1} \otimes_{A/I^{n + 1}}^\mathbf{L} A/I^n \to K_n\).
Then \(K = R\lim K_n\) is a pseudo-coherent, derived complete object of \(D(A)\) and \(K \otimes_A^\mathbf{L} A/I^n \to K_n\) is an isomorphism for all \(n\).
Proof
We already know that \(K\) is pseudo-coherent and that \(K \otimes_A^\mathbf{L} A/I^n \to K_n\) is an isomorphism for all \(n\), see Lemma 0CQF. Finally, \(K\) is derived complete by Lemma 091W.
Lemma
Let \(A = \lim A_n\) be a limit of an inverse system \((A_n)\) of rings. Suppose given \(K_n \in D(A_n)\) and maps \(K_{n + 1} \to K_n\) in \(D(A_{n + 1})\). Assume
the transition maps \(A_{n + 1} \to A_n\) are surjective with locally nilpotent kernels,
either all \(K_n\) are perfect or there exists an integer \(n_0\) such that \(K_{n_0}\) is perfect and the kernels \(A_{n + 1} \to A_n\) are nilpotent for \(n \geq n_0\), and
the maps induce isomorphisms \(K_{n + 1} \otimes_{A_{n + 1}}^\mathbf{L} A_n \to K_n\).
Then \(K = R\lim K_n\) is a perfect object of \(D(A)\) and \(K \otimes_A^\mathbf{L} A_n \to K_n\) is an isomorphism for all \(n\).
Proof
We already know that \(K\) is pseudo-coherent and that \(K \otimes_A^\mathbf{L} A_n \to K_n\) is an isomorphism for all \(n\) by Lemma 0CQF. Consider a surjective map \(A \to \kappa\) whose kernel is a maximal ideal \(\mathfrak m\). Any element of \(A\) which maps to a unit in \(A_1\) is a unit in \(A\) by Algebra, Lemma 0AMG and hence \(\Ker(A \to A_1)\) is contained in the Jacobson radical of \(A\) by Algebra, Lemma 0AME. Hence \(A \to \kappa\) factors as \(A \to A_1 \to \kappa\). Hence \[K \otimes_A^\mathbf{L} \kappa = K \otimes_A^\mathbf{L} A_1 \otimes_{A_1}^\mathbf{L} \kappa = K_1 \otimes_{A_1}^\mathbf{L} \kappa\] Note that \(K_1\) is perfect by assumption (2). Hence \(K_1\) has finite tor dimension by Lemma 0658. Thus there exist \(a, b \in \mathbf{Z}\) such that \(H^i(K \otimes_A^\mathbf{L} \kappa) = 0\) for all \(i \not \in [a, b]\). By Lemma 068V we conclude that \(K\) is perfect.
Lemma
Let \(A\) be a ring and \(I \subset A\) an ideal. Suppose given \(K_n \in D(A/I^n)\) and maps \(K_{n + 1} \to K_n\) in \(D(A/I^{n + 1})\). Assume
\(A\) is \(I\)-adically complete,
\(K_1\) is a perfect object, and
the maps induce isomorphisms \(K_{n + 1} \otimes_{A/I^{n + 1}}^\mathbf{L} A/I^n \to K_n\).
Then \(K = R\lim K_n\) is a perfect, derived complete object of \(D(A)\) and \(K \otimes_A^\mathbf{L} A/I^n \to K_n\) is an isomorphism for all \(n\).
Proof
We do not know if the following lemma holds for unbounded complexes.
Lemma
Let \(A\) be a ring and \(I \subset A\) an ideal. Suppose given \(K_n \in D(A/I^n)\) and maps \(K_{n + 1} \to K_n\) in \(D(A/I^{n + 1})\). If
\(A\) is Noetherian,
\(K_1\) is bounded above, and
the maps induce isomorphisms \(K_{n + 1} \otimes_{A/I^{n + 1}}^\mathbf{L} A/I^n \to K_n\),
then \(K = R\lim K_n\) is a derived complete object of \(D^-(A)\) and \(K \otimes_A^\mathbf{L} A/I^n \to K_n\) is an isomorphism for all \(n\).
Proof
The object \(K\) of \(D(A)\) is derived complete by Lemma 091W.
Suppose that \(H^i(K_1) = 0\) for \(i > b\). Then we can find a complex of free \(A/I\)-modules \(P_1^\bullet\) representing \(K_1\) with \(P_1^i = 0\) for \(i > b\). By Lemma 09AR we can, by induction on \(n > 1\), find complexes \(P_n^\bullet\) of free \(A/I^n\)-modules representing \(K_n\) and maps \(P_n^\bullet \to P_{n - 1}^\bullet\) representing the maps \(K_n \to K_{n - 1}\) inducing isomorphisms (!) of complexes \(P_n^\bullet/I^{n - 1}P_n^\bullet \to P_{n - 1}^\bullet\).
Thus we have arrived at the situation where \(R\lim K_n\) is represented by \(P^\bullet = \lim P_n^\bullet\), see Lemma 091D and Remark 07KZ. The complexes \(P_n^\bullet\) are uniformly bounded above complexes of flat \(A/I^n\)-modules and the transition maps are termwise surjective. Then \(P^\bullet\) is a bounded above complex of flat \(A\)-modules by Lemma 0912. It follows that \(K \otimes_A^\mathbf{L} A/I^t\) is represented by \(P^\bullet \otimes_A A/I^t\). We have \(P^\bullet \otimes_A A/I^t = \lim P_n^\bullet \otimes_A A/I^t\) termwise by Lemma 0912. The transition maps \(P_{n + 1}^\bullet \otimes_A A/I^t \to P_n^\bullet \otimes_A A/I^t\) are isomorphisms for \(n \geq t\) by our choice of \(P_n^\bullet\), hence we have \(\lim P_n^\bullet \otimes_A A/I^t = P_t^\bullet \otimes_A A/I^t = P_t^\bullet\). Since \(P_t^\bullet\) represents \(K_t\), we see that \(K \otimes_A^\mathbf{L} A/I^t \to K_t\) is an isomorphism.
Here is a different type of result.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(K \in D(A)\). Set \(K_n = K \otimes_A^\mathbf{L} A/I^n\). Assume for all \(i \in \mathbf{Z}\) we have
\(H^i(K)\) is a finite \(A\)-module, and
the system \(H^i(K_n)\) satisfies Mittag-Leffler.
Then \(\lim H^i(K)/I^nH^i(K)\) is equal to \(\lim H^i(K_n)\) for all \(i \in \mathbf{Z}\).
Proof
Recall that \(K^\wedge = R\lim K_n\) is the derived completion of \(K\), see Proposition 0922. By Lemma 0A06 we have \(H^i(K^\wedge) = \lim H^i(K)/I^nH^i(K)\). By Lemma 0CQE we get short exact sequences \[0 \to R^1\lim H^{i - 1}(K_n) \to H^i(K^\wedge) \to \lim H^i(K_n) \to 0\] The Mittag-Leffler condition guarantees that the left terms are zero (Lemma 091D) and we conclude the lemma is true.
Some evaluation maps
In this section we prove that certain canonical maps of \(R\Hom\)’s are isomorphisms for suitable types of complexes.
Lemma
Let \(R\) be a ring. Let \(K, L, M\) be objects of \(D(R)\). the map \[R\Hom_R(L, M) \otimes_R^\mathbf{L} K \longrightarrow R\Hom_R(R\Hom_R(K, L), M)\] of Lemma 0A67 is an isomorphism in the following two cases
\(K\) perfect, or
\(K\) is pseudo-coherent, \(L \in D^+(R)\), and \(M\) finite injective dimension.
Proof
Choose a K-injective complex \(I^\bullet\) representing \(M\), a K-injective complex \(J^\bullet\) representing \(L\), and a bounded above complex of finite projective modules \(K^\bullet\) representing \(K\). Consider the map of complexes \[\text{Tot}(\Hom^\bullet(J^\bullet, I^\bullet) \otimes_R K^\bullet) \longrightarrow \Hom^\bullet(\Hom^\bullet(K^\bullet, J^\bullet), I^\bullet)\] of Lemma 0A60. Note that \[\left(\prod\nolimits_{p + r = t} \Hom_R(J^{-r}, I^p)\right) \otimes_R K^s = \prod\nolimits_{p + r = t} \Hom_R(J^{-r}, I^p) \otimes_R K^s\] because \(K^s\) is finite projective. The map is given by the maps \[c_{p, r, s} : \Hom_R(J^{-r}, I^p) \otimes_R K^s \longrightarrow \Hom_R(\Hom_R(K^s, J^{-r}), I^p)\] which are isomorphisms as \(K^s\) is finite projective. For every element \(\alpha = (\alpha^{p, r, s})\) of degree \(n\) of the left hand side, there are only finitely many values of \(s\) such that \(\alpha^{p, r, s}\) is nonzero (for some \(p, r\) with \(n = p + r + s\)). Hence our map is an isomorphism if the same vanishing condition is forced on the elements \(\beta = (\beta^{p, r, s})\) of the right hand side. If \(K^\bullet\) is a bounded complex of finite projective modules, this is clear. On the other hand, if we can choose \(I^\bullet\) bounded and \(J^\bullet\) bounded below, then \(\beta^{p, r, s}\) is zero for \(p\) outside a fixed range, for \(s \gg 0\), and for \(r \gg 0\). Hence among solutions of \(n = p + r + s\) with \(\beta^{p, r, s}\) nonzero only a finite number of \(s\) values occur.
Lemma
Let \(R\) be a ring. Let \(K, L, M\) be objects of \(D(R)\). the map \[R\Hom_R(L, M) \otimes_R^\mathbf{L} K \longrightarrow R\Hom_R(R\Hom_R(K, L), M)\] of Lemma 0A67 is an isomorphism if the following three conditions are satisfied
\(L, M\) have finite injective dimension,
\(R\Hom_R(L, M)\) has finite tor dimension,
for every \(n \in \mathbf{Z}\) the truncation \(\tau_{\leq n}K\) is pseudo-coherent
Proof
Pick an integer \(n\) and consider the distinguished triangle \[\tau_{\leq n}K \to K \to \tau_{\geq n + 1}K \to \tau_{\leq n}K[1]\] see Derived Categories, Remark 08J5. By assumption (3) and Lemma 0A68 the map is an isomorphism for \(\tau_{\leq n}K\). Hence it suffices to show that both \[R\Hom_R(L, M) \otimes_R^\mathbf{L} \tau_{\geq n + 1}K \quad\text{and}\quad R\Hom_R(R\Hom_R(\tau_{\geq n + 1}K, L), M)\] have vanishing cohomology in degrees \(\leq n - c\) for some \(c\). This follows immediately from assumptions (2) and (1).
Lemma
Let \(R\) be a ring. Let \(K, L, M\) be objects of \(D(R)\). The map \[K \otimes_R^\mathbf{L} R\Hom_R(M, L) \longrightarrow R\Hom_R(M, K \otimes_R^\mathbf{L} L)\] of Lemma 0BYN is an isomorphism in the following cases
\(M\) perfect, or
\(K\) is perfect, or
\(M\) is pseudo-coherent, \(L \in D^+(R)\), and \(K\) has tor amplitude in \([a, \infty]\).
Proof
Proof in case \(M\) is perfect. Note that both sides of the arrow transform distinguished triangles in \(M\) into distinguished triangles and commute with direct sums. Hence it suffices to check it holds when \(M = R[n]\), see Derived Categories, Remark 0ATH and Lemma 0ATI. In this case the result is obvious.
Proof in case \(K\) is perfect. Same argument as in the previous case.
Proof in case (3). We may represent \(K\) and \(L\) by bounded below complexes of \(R\)-modules \(K^\bullet\) and \(L^\bullet\). We may assume that \(K^\bullet\) is a K-flat complex consisting of flat \(R\)-modules, see Lemma 0BYL. We may represent \(M\) by a bounded above complex \(M^\bullet\) of finite free \(R\)-modules, see Definition 064Q. Then the object on the LHS is represented by \[\text{Tot}(K^\bullet \otimes_R \Hom^\bullet(M^\bullet, L^\bullet))\] and the object on the RHS by \[\Hom^\bullet(M^\bullet, \text{Tot}(K^\bullet \otimes_R L^\bullet))\] This uses Lemma 0A66. Both complexes have in degree \(n\) the module \[\bigoplus\nolimits_{p + q + r = n} K^p \otimes \Hom_R(M^{-r}, L^q) = \bigoplus\nolimits_{p + q + r = n} \Hom_R(M^{-r}, K^p \otimes_R L^q)\] because \(M^{-r}\) is finite free (as well these are finite direct sums). The map defined in Lemma 0BYN comes from the map of complexes defined in Lemma 0BYM which uses the canonical isomorphisms between these modules.
Lemma
Let \(R\) be a ring. Let \(P^\bullet\) be a bounded above complex of projective \(R\)-modules. Let \(K^\bullet\) be a K-flat complex of \(R\)-modules. If \(P^\bullet\) is a perfect object of \(D(R)\), then \(\Hom^\bullet(P^\bullet, K^\bullet)\) is K-flat and represents \(R\Hom_R(P^\bullet, K^\bullet)\).
Proof
The last statement is Lemma 0A66. Since \(P^\bullet\) represents a perfect object, there exists a finite complex of finite projective \(R\)-modules \(F^\bullet\) such that \(P^\bullet\) and \(F^\bullet\) are isomorphic in \(D(R)\), see Definition 0657. Then \(P^\bullet\) and \(F^\bullet\) are homotopy equivalent, see Derived Categories, Lemma 064B. Then \(\Hom^\bullet(P^\bullet, K^\bullet)\) and \(\Hom^\bullet(F^\bullet, K^\bullet)\) are homotopy equivalent. Hence the first is K-flat if and only if the second is (follows from Definition 06XZ and Lemma 064I). It is clear that \[\Hom^\bullet(F^\bullet, K^\bullet) = \text{Tot}(E^\bullet \otimes_R K^\bullet)\] where \(E^\bullet\) is the dual complex to \(F^\bullet\) with terms \(E^n = \Hom_R(F^{-n}, R)\), see Lemma 07VI and its proof. Since \(E^\bullet\) is a bounded complex of projectives we find that it is K-flat by Lemma 064K. Then we conclude by Lemma 0795.
Base change for derived hom
We have already seen some material discussing this in Lemma 087R and in Algebra, Section 087M.
Lemma
Let \(R \to R'\) be a ring map. For \(K \in D(R)\) and \(M \in D(R')\) there is a canonical isomorphism \[R\Hom_R(K, M) = R\Hom_{R'}(K \otimes_R^\mathbf{L} R', M)\]
Proof
Choose a K-injective complex of \(R'\)-modules \(J^\bullet\) representing \(M\). Choose a quasi-isomorphism \(J^\bullet \to I^\bullet\) where \(I^\bullet\) is a K-injective complex of \(R\)-modules. Choose a K-flat complex \(K^\bullet\) of \(R\)-modules representing \(K\). Consider the map \[\Hom^\bullet(K^\bullet \otimes_R R', J^\bullet) \longrightarrow \Hom^\bullet(K^\bullet, I^\bullet)\] The map on degree \(n\) terms is given by the map \[\prod\nolimits_{n = p + q} \Hom_{R'}(K^{-q} \otimes_R R', J^p) \longrightarrow \prod\nolimits_{n = p + q} \Hom_R(K^{-q}, I^p)\] coming from precomposing by \(K^{-q} \to K^{-q} \otimes_R R'\) and postcomposing by \(J^p \to I^p\). To finish the proof it suffices to show that we get isomorphisms on cohomology groups: \[\Hom_{D(R)}(K, M) = \Hom_{D(R')}(K \otimes_R^\mathbf{L} R', M)\] which is true because base change \(- \otimes_R^\mathbf{L} R' : D(R) \to D(R')\) is left adjoint to the restriction functor \(D(R') \to D(R)\) by Lemma 0GMT.
Let \(R \to R'\) be a ring map. There is a base change map [0E1X]\[\begin{equation} R\Hom_R(K, M) \otimes_R^\mathbf{L} R' \longrightarrow R\Hom_{R'}(K \otimes_R^\mathbf{L} R', M \otimes_R^\mathbf{L} R') \end{equation}\] in \(D(R')\) functorial in \(K, M \in D(R)\). Namely, by adjointness of \(- \otimes_R^\mathbf{L} R' : D(R) \to D(R')\) and the restriction functor \(D(R') \to D(R)\), this is the same thing as a map \[R\Hom_R(K, M) \longrightarrow R\Hom_{R'}(K \otimes_R^\mathbf{L} R', M \otimes_R^\mathbf{L} R') = R\Hom_R(K, M \otimes_R^\mathbf{L} R')\] (equality by Lemma 0E1W) for which we can use the canonical map \(M \to M \otimes_R^\mathbf{L} R'\) (unit of the adjunction).
Lemma
Let \(R \to R'\) be a ring map. Let \(K, M \in D(R)\). The map (0E1X) \[R\Hom_R(K, M) \otimes_R^\mathbf{L} R' \longrightarrow R\Hom_{R'}(K \otimes_R^\mathbf{L} R', M \otimes_R^\mathbf{L} R')\] is an isomorphism in \(D(R')\) in the following cases
\(K\) is perfect,
\(R'\) is perfect as an \(R\)-module,
\(R \to R'\) is flat, \(K\) is pseudo-coherent, and \(M \in D^{+}(R)\), or
\(R'\) has finite tor dimension as an \(R\)-module, \(K\) is pseudo-coherent, and \(M \in D^{+}(R)\)
Proof
We may check the map is an isomorphism after applying the restriction functor \(D(R') \to D(R)\). After applying this functor our map becomes the map \[R\Hom_R(K, M) \otimes_R^\mathbf{L} R' \longrightarrow R\Hom_R(K, M \otimes_R^\mathbf{L} R')\] of Lemma 0BYN. See discussion above the lemma to match the left and right hand sides; in particular, this uses Lemma 0E1W. Thus we conclude by Lemma 0ATK.
Systems of modules
Let \(I\) be an ideal of a Noetherian ring \(A\). In this section we add to our knowledge of the relationship between finite modules over \(A\) and systems of finite \(A/I^n\)-modules.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(K \xrightarrow{\alpha} L \xrightarrow{\beta} M\) be a complex of finite \(A\)-modules. Set \(H = \Ker(\beta)/\Im(\alpha)\). For \(n \geq 0\) let \[K/I^nK \xrightarrow{\alpha_n} L/I^nL \xrightarrow{\beta_n} M/I^nM\] be the induced complex. Set \(H_n = \Ker(\beta_n)/\Im(\alpha_n)\). Then there are canonical \(A\)-module maps giving a commutative diagram \[\xymatrix{ & & & H \ar[lld] \ar[ld] \ar[d] \\ \ldots \ar[r] & H_3 \ar[r] & H_2 \ar[r] & H_1 }\] Moreover, there exists a \(c > 0\) and canonical \(A\)-module maps \(H_n \to H/I^{n - c}H\) for \(n \geq c\) such that the compositions \[H/I^n H \to H_n \to H/I^{n - c}H \quad\text{and}\quad H_n \to H/I^{n - c}H \to H_{n - c}\] are the canonical ones. Moreover, we have
\((H_n)\) and \((H/I^nH)\) are isomorphic as pro-objects of \(\text{Mod}_A\),
\(\lim H_n = \lim H/I^n H\),
the inverse system \((H_n)\) is Mittag-Leffler,
the image of \(H_{n + c} \to H_n\) is equal to the image of \(H \to H_n\),
the composition \(I^cH_n \to H_n \to H/I^{n - c}H \to H_n/I^{n - c}H_n\) is the inclusion \(I^cH_n \to H_n\) followed by the quotient map \(H_n \to H_n/I^{n - c}H_n\), and
the kernel and cokernel of \(H/I^nH \to H_n\) is annihilated by \(I^c\).
Proof
Observe that \(H_n = \beta^{-1}(I^nM)/\Im(\alpha) + I^nL\). For \(n \geq 2\) we have \(\beta^{-1}(I^nM) \subset \beta^{-1}(I^{n - 1}M)\) and \(\Im(\alpha) + I^nL \subset \Im(\alpha) + I^{n - 1}L\). Thus we obtain our canonical map \(H_n \to H_{n - 1}\). Similarly, we have \(\Ker(\beta) \subset \beta^{-1}(I^nM)\) and \(\Im(\alpha) \subset \Im(\alpha) + I^nL\) which produces the canonical map \(H \to H_n\). We omit the verification that the diagram commutes.
By Artin-Rees we may choose \(c_1, c_2 \geq 0\) such that \(\beta^{-1}(I^nM) \subset \Ker(\beta) + I^{n - c_1}L\) for \(n \geq c_1\) and \(\Ker(\beta) \cap I^nL \subset I^{n - c_2}\Ker(\beta)\) for \(n \geq c_2\), see Algebra, Lemmas 00IO and 00IN. Set \(c = c_1 + c_2\).
Let \(n \geq c\). We define \(\psi_n : H_n \to H/I^{n - c}H\) as follows. Say \(x \in H_n\). Choose \(y \in \beta^{-1}(I^nM)\) representing \(x\). Write \(y = z + w\) with \(z \in \Ker(\beta)\) and \(w \in I^{n - c_1}L\) (this is possible by our choice of \(c_1\)). We set \(\psi_n(x)\) equal to the class of \(z\) in \(H/I^{n - c}H\). To see this is well defined, suppose we have a second set of choices \(y', z', w'\) as above for \(x\) with obvious notation. Then \(y' - y \in \Im(\alpha) + I^nL\), say \(y' - y = \alpha(v) + u\) with \(v \in K\) and \(u \in I^nL\). Thus \[y' = z' + w' = \alpha(v) + u + z + w \Rightarrow z' = z + \alpha(v) + u + w - w'\] Since \(\beta(z' - z - \alpha(v)) = 0\) we find that \(u + w - w' \in \Ker(\beta) \cap I^{n - c_1}L\) which is contained in \(I^{n - c_1 - c_2}\Ker(\beta) = I^{n - c}\Ker(\beta)\) by our choice of \(c_2\). Thus \(z'\) and \(z\) have the same image in \(H/I^{n - c}H\) as desired.
The composition \(H/I^n H \to H_n \to H/I^{n - c}H\) is the canonical map because if \(z \in \Ker(\beta)\) represents an element \(x\) in \(H/I^nH = \Ker(\beta)/\Im(\alpha) + I^n\Ker(\beta)\) then it is clear from the above that \(x\) maps to the class of \(z\) in \(H/I^{n - c}H\) under the maps constructed above.
Let us consider the composition \(H_n \to H/I^{n - c}H \to H_{n - c}\). Given \(x, y, z, w\) as in the construction of \(\psi_n\) above, we see that \(x\) is mapped to the class of \(z\) in \(H_{n - c}\). On the other hand, the canonical map \(H_n \to H_{n - c}\) from the first paragraph of the proof sends \(x\) to the class of \(y\). Thus we have to show that \(y - z \in \Im(\alpha) + I^{n - c}L\) which is the case because \(y - z = w \in I^{n - c_1}L \subset I^{n - c}L\).
Statements (1) – (4) are formal consequences of what we just proved. Namely, (1) follows from the existence of the maps and the definition of morphisms of pro-objects in Categories, Remark 05PX. Part (2) holds because isomorphic pro-objects have isomorphic limits. Part (3) is immediate from part (4). Part (4) follows from the factorization \(H_{n + c} \to H/I^nH \to H_n\) of the canonical map \(H_{n + c} \to H_n\).
Proof of part (5). Let \(x \in I^cH_n\). Write \(x = \sum f_i x_i\) with \(x_i \in H_n\) and \(f_i \in I^c\). Choose \(y_i, z_i, w_i\) as in the construction of \(\psi_n\) for \(x_i\). Then for the computation of \(\psi_n\) of \(x\) we may choose \(y = \sum f_iy_i\), \(z = \sum f_i z_i\) and \(w = \sum f_i w_i\) and we see that \(\psi_n(x)\) is given by the class of \(z\). The image of this in \(H_n/I^{n - c}H_n\) is equal to the class of \(y\) as \(w = \sum f_i w_i\) is in \(I^nL\). This proves (5).
Proof of part (6). Let \(y \in \Ker(\beta)\) whose class is \(x\) in \(H\). If \(x\) maps to zero in \(H_n\), then \(y \in I^nL + \Im(\alpha)\). Hence \(y - \alpha(v) \in \Ker(\beta) \cap I^nL\) for some \(v \in K\). Then \(y - \alpha(v) \in I^{n - c_2}\Ker(\beta)\) and hence the class of \(y\) in \(H/I^nH\) is annihilated by \(I^{c_2}\). Finally, let \(x \in H_n\) be the class of \(y \in \beta^{-1}(I^nM)\). Then we write \(y = z + w\) with \(z \in \Ker(\beta)\) and \(w \in I^{n - c_1}L\) as above. Clearly, if \(f \in I^{c_1}\) then \(fx\) is the class of \(fy + fw \equiv fy\) modulo \(\Im(\alpha) + I^nL\) and hence \(fx\) is the image of the class of \(fy\) in \(H\) as desired.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(K \in D(A)\) be pseudo-coherent. Set \(K_n = K \otimes_A^\mathbf{L} A/I^n\). Then for all \(i \in \mathbf{Z}\) the system \(H^i(K_n)\) satisfies Mittag-Leffler and \(\lim H^i(K)/I^nH^i(K)\) is equal to \(\lim H^i(K_n)\).
Proof
We may represent \(K\) by a bounded above complex \(P^\bullet\) of finite free \(A\)-modules. Then \(K_n\) is represented by \(P^\bullet/I^nP^\bullet\). Hence the Mittag-Leffler property by Lemma 0EGU. The final statement follows then from Lemma 0EGS.
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M^\bullet\) be a bounded complex of finite \(A\)-modules. The inverse system of maps \[M^\bullet \otimes_A^\mathbf{L} A/I^n \longrightarrow M^\bullet/I^nM^\bullet\] defines an isomorphism of pro-objects of \(D(A)\).
Proof
Say \(I = (f_1, \ldots, f_r)\). Let \(K_n \in D(A)\) be the object represented by the Koszul complex on \(f_1^n, \ldots, f_r^n\). Recall that we have maps \(K_n \to A/I^n\) which induce a pro-isomorphism of inverse systems, see Lemma 0921. Hence it suffices to show that \[M^\bullet \otimes_A^\mathbf{L} K_n \longrightarrow M^\bullet/I^nM^\bullet\] defines an isomorphism of pro-objects of \(D(A)\). Since \(K_n\) is represented by a complex of finite free \(A\)-modules sitting in degrees \(-r, \ldots, 0\) there exist \(a, b \in \mathbf{Z}\) such that the source and target of the displayed arrow have vanishing cohomology in degrees outside \([a, b]\) for all \(n\). Thus we may apply Derived Categories, Lemma 0G3D and we find that it suffices to show that the maps \[H^i(M^\bullet \otimes_A^\mathbf{L} A/I^n) \to H^i(M^\bullet/I^nM^\bullet)\] define isomorphisms of pro-systems of \(A\)-modules for any \(i \in \mathbf{Z}\). To see this choose a quasi-isomorphism \(P^\bullet \to M^\bullet\) where \(P^\bullet\) is a bounded above complex of finite free \(A\)-modules. The arrows above are given by the maps \[H^i(P^\bullet/I^nP^\bullet) \to H^i(M^\bullet/I^nM^\bullet)\] These define an isomorphism of pro-systems by Lemma 0EGU. Namely, the lemma shows both are isomorphic to the pro-system \(H^i/I^nH^i\) with \(H^i = H^i(M^\bullet) = H^i(P^\bullet)\).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\), \(N\) be finite \(A\)-modules. Set \(M_n = M/I^nM\) and \(N_n = N/I^nN\). Then
the systems \((\Hom_A(M_n, N_n))\) and \((\text{Isom}_A(M_n, N_n))\) are Mittag-Leffler,
there exists a \(c \geq 0\) such that the kernels and cokernels of \[\Hom_A(M, N)/I^n\Hom_A(M, N) \to \Hom_A(M_n, N_n)\] are killed by \(I^c\) for all \(n\),
we have \(\lim \Hom_A(M_n, N_n) =\Hom_A(M, N)^\wedge = \Hom_{A^\wedge}(M^\wedge, N^\wedge)\)
\(\lim \text{Isom}_A(M_n, N_n) = \text{Isom}_{A^\wedge}(M^\wedge, N^\wedge)\).
Here \({}^\wedge\) denotes usual \(I\)-adic completion.
Proof
Note that \(\Hom_A(M_n, N_n) = \Hom_A(M, N_n)\). Choose a presentation \[A^{\oplus t} \to A^{\oplus s} \to M \to 0\] Applying the left exact functor \(\Hom_A(-, N)\) we obtain a complex \[0 \xrightarrow{\alpha} N^{\oplus s} \xrightarrow{\beta} N^{\oplus t}\] whose cohomology in the middle is \(\Hom_A(M, N)\) and such that for \(n \geq 0\) the cohomology of \[0 \xrightarrow{\alpha_n} N_n^{\oplus s} \xrightarrow{\beta_n} N_n^{\oplus t}\] is \(\Hom_A(M_n, N_n)\). Let \(c \geq 0\) be as in Lemma 0EGU for this \(A\), \(I\), \(\alpha\), and \(\beta\). By part (3) of the lemma we deduce the Mittag-Leffler property for \((\Hom_A(M_n, N_n))\). The kernel and cokernel of the maps \(\Hom_A(M, N)/I^n\Hom_A(M, N) \to \Hom_A(M_n, N_n)\) are killed by \(I^c\) by part (6) of the lemma. We find that \(\lim \Hom_A(M_n, N_n) = \Hom_A(M, N)^\wedge\) by part (2) of the lemma. The equality \[\Hom_{A^\wedge}(M^\wedge, N^\wedge) = \lim \Hom_A(M_n, N_n)\] follows formally from the fact that \(M^\wedge = \lim M_n\) and \(M_n = M^\wedge/I^nM^\wedge\) and the corresponding facts for \(N\), see Algebra, Lemma 031C.
The result for isomorphisms follows from the case of homomorphisms applied to both \((\Hom(M_n, N_n))\) and \((\Hom(N_n, M_n))\) and the following fact: for \(n > m > 0\), if we have maps \(\alpha : M_n \to N_n\) and \(\beta : N_n \to M_n\) which induce an isomorphisms \(M_m \to N_m\) and \(N_m \to M_m\), then \(\alpha\) and \(\beta\) are isomorphisms. Namely, then \(\alpha \circ \beta\) is surjective by Nakayama’s lemma (Algebra, Lemma 00DV) hence \(\alpha \circ \beta\) is an isomorphism by Algebra, Lemma 05G8.
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\), \(N\) be finite \(A\)-modules. Set \(M_n = M/I^nM\) and \(N_n = N/I^nN\). If \(M_n \cong N_n\) for all \(n\), then \(M^\wedge \cong N^\wedge\) as \(A^\wedge\)-modules.
Proof
By Lemma 09BB the system \((\text{Isom}_A(M_n, N_n))\) is Mittag-Leffler. By assumption each of the sets \(\text{Isom}_A(M_n, N_n)\) is nonempty. Hence \(\lim \text{Isom}_A(M_n, N_n)\) is nonempty. Since \(\lim \text{Isom}_A(M_n, N_n) = \text{Isom}_{A^\wedge}(M^\wedge, N^\wedge)\) we obtain an isomorphism.
Remark
Let \(I\) be an ideal of a Noetherian ring \(A\). Set \(A_n = A/I^n\) for \(n \geq 1\). Consider the following category:
An object is a sequence \(\{E_n\}_{n \geq 1}\) where \(E_n\) is a finite \(A_n\)-module.
A morphism \(\{E_n\} \to \{E'_n\}\) is given by maps \[\varphi_n : I^cE_n \longrightarrow E'_n/E'_n[I^c] \quad\text{for }n \geq c\] where \(E'_n[I^c]\) is the torsion submodule (Section 0ALX) up to equivalence: we say \((c, \varphi_n)\) is the same as \((c + 1, \overline{\varphi}_n)\) where \(\overline{\varphi}_n : I^{c + 1}E_n \longrightarrow E'_n/E'_n[I^{c + 1}]\) is the induced map.
Composition of \((c, \varphi_n) : \{E_n\} \to \{E'_n\}\) and \((c', \varphi'_n) : \{E'_n\} \to \{E''_n\}\) is defined by the obvious compositions \[I^{c + c'}E_n \to I^{c'}E'_n/E'_n[I^{c}] \to E''_n/E''_n[I^{c + c'}]\] for \(n \geq c + c'\). We omit the verification that this is a category.
Lemma
A morphism \((c, \varphi_n)\) of the category of Remark 0EGW is an isomorphism if and only if there exists a \(c' \geq 0\) such that \(\Ker(\varphi_n)\) and \(\Coker(\varphi_n)\) are \(I^{c'}\)-torsion for all \(n \gg 0\).
Proof
We may and do assume \(c' \geq c\) and that the \(\Ker(\varphi_n)\) and \(\Coker(\varphi_n)\) are \(I^{c'}\)-torsion for all \(n\). For \(n \geq c'\) and \(x \in I^{c'}E'_n\) we can choose \(y \in I^cE_n\) with \(x = \varphi_n(y) \bmod E'_n[I^c]\) as \(\Coker(\varphi_n)\) is annihilated by \(I^{c'}\). Set \(\psi_n(x)\) equal to the class of \(y\) in \(E_n/E_n[I^{c'}]\). For a different choice \(y' \in I^cE_n\) with \(x = \varphi_n(y') \bmod E'_n[I^c]\) the difference \(y - y'\) maps to zero in \(E'_n/E'_n[I^c]\) and hence is annihilated by \(I^{c'}\) in \(I^cE_n\). Thus the maps \(\psi_n : I^{c'}E'_n \to E_n/E_n[I^{c'}]\) are well defined. We omit the verification that \((c', \psi_n)\) is the inverse of \((c, \varphi_n)\) in the category.
Lemma
Let \(I\) be an ideal of the Noetherian ring \(A\). Let \(M\) and \(N\) be finite \(A\)-modules. Write \(A_n = A/I^n\), \(M_n = M/I^nM\), and \(N_n = N/I^nN\). For every \(i \geq 0\) the objects \[\{\Ext^i_A(M, N)/I^n\Ext^i_A(M, N)\}_{n \geq 1} \quad\text{and}\quad \{\Ext^i_{A_n}(M_n, N_n)\}_{n \geq 1}\] are isomorphic in the category \(\mathcal{C}\) of Remark 0EGW.
Proof
Choose a short exact sequence \[0 \to K \to A^{\oplus r} \to M \to 0\] and set \(K_n = K/I^nK\). For \(n \geq 1\) define \(K(n) = \Ker(A_n^{\oplus r} \to M_n)\) so that we have exact sequences \[0 \to K(n) \to A_n^{\oplus r} \to M_n \to 0\] and surjections \(K_n \to K(n)\). In fact, by Lemma 0EGU there is a \(c \geq 0\) and maps \(K(n) \to K_n/I^{n - c}K_n\) which are “almost inverse”. Since \(I^{n - c}K_n \subset K_n[I^c]\) these maps witness the fact that the systems \(\{K(n)\}_{n \geq 1}\) and \(\{K_n\}_{n \geq 1}\) are isomorphic in \(\mathcal{C}\).
We claim the systems \[\{\Ext^i_{A_n}(K(n), N_n)\}_{n \geq 1} \quad\text{and}\quad \{\Ext^i_{A_n}(K_n, N_n)\}_{n \geq 1}\] are isomorphic in the category \(\mathcal{C}\). Namely, the surjective maps \(K_n \to K(n)\) have kernels annihilated by \(I^c\) and therefore determine maps \[\Ext^i_{A_n}(K(n), N_n) \to \Ext^i_{A_n}(K_n, N_n)\] whose kernel and cokernel are annihilated by \(I^c\). Hence the claim by Lemma 0EGX.
For \(i \geq 2\) we have isomorphisms \[\Ext^{i - 1}_A(K, N) = \Ext^i_A(M, N) \quad\text{and}\quad \Ext^{i - 1}_{A_n}(K(n), N_n) = \Ext^i_{A_n}(M_n, N_n)\] In this way we see that it suffices to prove the lemma for \(i = 0, 1\).
For \(i = 0, 1\) we consider the commutative diagram \[\xymatrix{ 0 \ar[r] & \Hom(M, N) \ar[r] \ar[dd] & N^{\oplus r} \ar[r]_-\varphi \ar[dd] & \Hom(K, N) \ar[r] \ar[d] & \Ext^1(M, N) \ar[r] & 0 \\ & & & \Hom(K_n, N_n) \\ 0 \ar[r] & \Hom(M_n, N_n) \ar[r] & N_n^{\oplus r} \ar[r] & \Hom(K(n), N_n) \ar[r] \ar[u] & \Ext^1(M_n, N_n) \ar[r] & 0 }\] By Lemma 09BB we see that the kernel and cokernel of \(\Hom(M, N)/I^n \Hom(M, N) \to \Hom(M_n, N_n)\) and \(\Hom(K, N)/I^n \Hom(K, N) \to \Hom(K_n, N_n)\) and are \(I^c\)-torsion for some \(c \geq 0\) independent of \(n\). Above we have seen the cokernel of the injective maps \(\Hom(K(n), N_n) \to \Hom(K_n, N_n)\) are annihilated by \(I^c\) after possibly increasing \(c\). For such a \(c\) we obtain maps \(\delta_n : I^c\Hom(K, N)/I^n\Hom(K, N) \to \Hom(K(n), N_n)\) fitting into the diagram (precise formulation omitted). The kernel and cokernel of \(\delta_n\) are annihilated by \(I^c\) after possibly increasing \(c\) since we know that the same thing is true for \(\Hom(K, N)/I^n \Hom(K, N) \to \Hom(K_n, N_n)\) and \(\Hom(K(n), N_n) \to \Hom(K_n, N_n)\). Then we can use commutativity of the solid diagram \[\xymatrix{ \varphi^{-1}(I^c\Hom(K, N)) \ar[r]_-\varphi \ar[d] & I^c\Hom(K, N)/I^n\Hom(K, N) \ar[r] \ar[d]^{\delta_n} & I^c\Ext^1(M, N)/I^n\Ext^1(M, N) \ar[r] \ar@{..>}[d] & 0 \\ N_n^{\oplus r} \ar[r] & \Hom(K(n), N_n) \ar[r] & \Ext^1(M_n, N_n) \ar[r] & 0 }\] to define the dotted arrow. A straightforward diagram chase (omitted) shows that the kernel and cokernel of the dotted arrow are annihilated by \(I^c\) after possibly increasing \(c\) one final time.
Remark
The awkwardness in the statement of Lemma 0EGY is partly due to the fact that there are no obvious maps between the modules \(\Ext^i_{A_n}(M_n, N_n)\) for varying \(n\). What we may conclude from the lemma is that there exists a \(c \geq 0\) such that for \(m \gg n \gg 0\) there are (canonical) maps \[I^c\Ext^i_{A_m}(M_m, N_m)/I^n\Ext^i_{A_m}(M_m, N_m) \to \Ext^i_{A_n}(M_n, N_n)/\Ext^i_{A_n}(M_n, N_n)[I^c]\] whose kernel and cokernel are annihilated by \(I^c\). This is the (weak) sense in which we get a system of modules.
Example
Let \(k\) be a field. Let \(A = k[[x, y]]/(xy)\). By abuse of notation we denote \(x\) and \(y\) the images of \(x\) and \(y\) in \(A\). Let \(I = (x)\). Let \(M = A/(y)\). There is a free resolution \[\ldots \to A \xrightarrow{y} A \xrightarrow{x} A \xrightarrow{y} A \to M \to 0\] We conclude that \[\Ext^2_A(M, N) = N[y]/xN\] where \(N[y] = \Ker(y : N \to N)\). We denote \(A_n = A/I^n\), \(M_n = M/I^nM\), and \(N_n = N/I^nN\). For each \(n\) we have a free resolution \[\ldots \to A_n^{\oplus 2} \xrightarrow{y, x^{n - 1}} A_n \xrightarrow{x} A_n \xrightarrow{y} A_n \to M_n \to 0\] We conclude that \[\Ext^2_{A_n}(M_n, N_n) = (N_n[y] \cap N_n[x^{n - 1}])/xN_n\] where \(N_n[y] = \Ker(y : N_n \to N_n)\) and \(N_n[x^{n - 1}] = \Ker(x^{n - 1} : N_n \to N_n)\). Take \(N = A/(y)\). Then we see that \[\Ext^2_A(M, N) = N[y]/xN = N/xN \cong k\] but \[\Ext^2_{A_n}(M_n, N_n) = (N_n[y] \cap N_n[x^{n - 1}])/xN_n = N_n[x^{n - 1}]/xN_n = 0\] for all \(n\) because \(N_n = k[x]/(x^n)\) and the sequence \[N_n \xrightarrow{x} N_n \xrightarrow{x^{n - 1}} N_n\] is exact. Thus ignoring some kind of \(I\)-power torsion is necessary to get a result as in Lemma 0EGY.
Lemma
Let \(A \to B\) be a flat homomorphism of Noetherian rings. Let \(I \subset A\) be an ideal. Let \(M, N\) be finite \(B\)-modules. Set \(B_n = B/I^nB\), \(M_n = M/I^nM\), \(N_n = N/I^nN\). If \(M\) is flat over \(A\), then we have \[\lim \Ext^i_B(M, N)/I^n \Ext^i_B(M, N) = \lim \Ext^i_{B_n}(M_n, N_n)\] for all \(i \in \mathbf{Z}\).
Proof
Choose a resolution \[\ldots \to P_2 \to P_1 \to P_0 \to M \to 0\] by finite free \(B\)-modules \(P_i\). Set \(P_{i, n} = P_i/I^nP_i\). Since \(M\) and \(B\) are flat over \(A\), the sequence \[\ldots \to P_{2, n} \to P_{1, n} \to P_{0, n} \to M_n \to 0\] is exact. We see that on the one hand the complex \[\Hom_B(P_0, N) \to \Hom_B(P_1, N) \to \Hom_B(P_2, N) \to \ldots\] computes the modules \(\Ext^i_B(M, N)\) and on the other hand the complex \[\Hom_{B_n}(P_{0, n}, N_n) \to \Hom_{B_n}(P_{1, n}, N_n) \to \Hom_{B_n}(P_{2, n}, N_n) \to \ldots\] computes the modules \(\Ext^i_{B_n}(M_n, N_n)\). Since \[\Hom_{B_n}(P_{i, n}, N_n) = \Hom_B(P_i, N)/I^n \Hom_B(P_i, N)\] we obtain the result from Lemma 0EGU part (2).
Systems of modules, bis
Let \(I\) be an ideal of a Noetherian ring \(A\). In Section 0EGT we considered what happens when considering systems of the form \(M/I^nM\) for finite \(A\)-modules \(M\). In this section we consider the systems \(I^nM\) instead.
Lemma
Let \(I\) be an ideal of a Noetherian ring \(A\). Let \(K \xrightarrow{\alpha} L \xrightarrow{\beta} M\) be a complex of finite \(A\)-modules. Set \(H = \Ker(\beta)/\Im(\alpha)\). For \(n \geq 0\) let \[I^nK \xrightarrow{\alpha_n} I^nL \xrightarrow{\beta_n} I^nM\] be the induced complex. Set \(H_n = \Ker(\beta_n)/\Im(\alpha_n)\). Then there are canonical \(A\)-module maps \[\ldots \to H_3 \to H_2 \to H_1 \to H\] There exists a \(c > 0\) such that for \(n \geq c\) the image of \(H_n \to H\) is contained in \(I^{n - c}H\) and there is a canonical \(A\)-module map \(I^nH \to H_{n - c}\) such that the compositions \[I^n H \to H_{n - c} \to I^{n - 2c}H \quad\text{and}\quad H_n \to I^{n - c}H \to H_{n - 2c}\] are the canonical ones. In particular, the inverse systems \((H_n)\) and \((I^nH)\) are isomorphic as pro-objects of \(\text{Mod}_A\).
Proof
We have \(H_n = \Ker(\beta) \cap I^nL/\alpha(I^nK)\). Since \(\Ker(\beta) \cap I^nL \subset \Ker(\beta) \cap I^{n - 1}L\) and \(\alpha(I^nK) \subset \alpha(I^{n - 1}K)\) we get the maps \(H_n \to H_{n - 1}\). Similarly for the map \(H_1 \to H\).
By Artin-Rees we may choose \(c_1, c_2 \geq 0\) such that \(\Im(\alpha) \cap I^nL \subset \alpha(I^{n - c_1}K)\) for \(n \geq c_1\) and \(\Ker(\beta) \cap I^nL \subset I^{n - c_2}\Ker(\beta)\) for \(n \geq c_2\), see Algebra, Lemmas 00IO and 00IN. Set \(c = c_1 + c_2\).
It follows immediately from our choice of \(c \geq c_2\) that for \(n \geq c\) the image of \(H_n \to H\) is contained in \(I^{n - c}H\).
Let \(n \geq c\). We define \(\psi_n : I^nH \to H_{n - c}\) as follows. Say \(x \in I^nH\). Choose \(y \in I^n\Ker(\beta)\) representing \(x\). We set \(\psi_n(x)\) equal to the class of \(y\) in \(H_{n - c}\). To see this is well defined, suppose we have a second choice \(y'\) as above for \(x\). Then \(y' - y \in \Im(\alpha)\). By our choice of \(c \geq c_1\) we conclude that \(y' - y \in \alpha(I^{n - c}K)\) which implies that \(y\) and \(y'\) represent the same element of \(H_{n - c}\). Thus \(\psi_n\) is well defined.
The statements on the compositions \(I^n H \to H_{n - c} \to I^{n - 2c}H\) and \(H_n \to I^{n - c}H \to H_{n - 2c}\) follow immediately from our definitions.
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\), \(N\) be \(A\)-modules with \(M\) finite. For each \(p > 0\) there exists a \(c \geq 0\) such that for \(n \geq c\) the map \(\Ext_A^p(M, N) \to \Ext_A^p(I^nM, N)\) factors through \(\Ext^p_A(I^nM, I^{n - c}N) \to \Ext_A^p(I^nM, N)\).
Proof
For \(p = 0\), if \(\varphi : M \to N\) is an \(A\)-linear map, then \(\varphi(\sum f_i m_i) = \sum f_i \varphi(m_i)\) for \(f_i \in A\) and \(m_i \in M\). Hence \(\varphi\) induces a map \(I^nM \to I^nN\) for all \(n\) and the result is true with \(c = 0\).
Choose a short exact sequence \(0 \to K \to A^{\oplus t} \to M \to 0\). For each \(n\) we pick a short exact sequence \(0 \to L_n \to A^{\oplus s_n} \to I^nM \to 0\). It is clear that we can construct a map of short exact sequences \[\xymatrix{ 0 \ar[r] & L_n \ar[r] \ar[d] & A^{\oplus s_n} \ar[r] \ar[d] & I^nM \ar[r] \ar[d] & 0 \\ 0 \ar[r] & K \ar[r] & A^{\oplus t} \ar[r] & M \ar[r] & 0 }\] such that \(A^{\oplus s_n} \to A^{\oplus t}\) has image in \((I^n)^{\oplus t}\). By Artin-Rees (Algebra, Lemma 00IN) there exists a \(c \geq 0\) such that \(L_n \to K\) factors through \(I^{n - c}K\) if \(n \geq c\).
For \(p = 1\) our choices above induce a solid commutative diagram \[\xymatrix{ \Hom_A(A^{\oplus s_n}, N) \ar[r] & \Hom_A(L_n, N) \ar[r] & \Ext_A^1(I^nM, N) \ar[r] & 0 \\ \Hom_A((I^n)^{\oplus t}, I^{n - c}N) \ar[r] \ar[u] & \Hom_A(K \cap (I^n)^{\oplus t}, I^{n - c}N) \ar[r] \ar[u] & \Ext_A^1(I^nM, I^{n - c}N) \ar[u] \\ \Hom_A(A^{\oplus t}, N) \ar[r] \ar[u] & \Hom_A(K, N) \ar[r] \ar[u] & \Ext_A^1(M, N) \ar@{..>}[u] \ar[r] & 0 }\] whose horizontal arrows are exact. The lower middle vertical arrow arises because \(K \cap (I^n)^{\oplus t} \subset I^{n - c}K\) and hence any \(A\)-linear map \(K \to N\) induces an \(A\)-linear map \((I^n)^{\oplus t} \to I^{n - c}N\) by the argument of the first paragraph. Thus we obtain the dotted arrow as desired.
For \(p > 1\) we obtain a commutative diagram \[\xymatrix{ \Ext^{p - 1}_A(I^{n - c}K, N) \ar[r] & \Ext^{p - 1}_A(L_n, N) \ar[r] & \Ext_A^p(I^nM, N) \\ \Ext^{p - 1}_A(K, N) \ar[rr] \ar[u] & & \Ext_A^p(M, N) \ar[u] }\] whose bottom horizontal arrow is an isomorphism. By induction on \(p\) the left vertical map factors through \(\Ext^{p - 1}_A(I^{n - c}K, I^{n - c - c'}N)\) for some \(c' \geq 0\) and all \(n \geq c + c'\). Using the composition \(\Ext^{p - 1}_A(I^{n - c}K, I^{n - c - c'}N) \to \Ext^{p - 1}_A(L_n, I^{n - c - c'}N) \to \Ext^p_A(I^nM, I^{n - c - c'}N)\) we obtain the desired factorization (for \(n \geq c + c'\) and with \(c\) replaced by \(c + c'\)).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\), \(N\) be \(A\)-modules with \(M\) finite and \(N\) annihilated by a power of \(I\). For each \(p > 0\) there exists an \(n\) such that the map \(\Ext_A^p(M, N) \to \Ext_A^p(I^nM, N)\) is zero.
Proof
Immediate consequence of Lemma 0G3L and the fact that \(I^mN = 0\) for some \(m > 0\).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(K \in D(A)\) be pseudo-coherent and let \(M\) be a finite \(A\)-module. For each \(p \in \mathbf{Z}\) there exists an \(c\) such that the image of \(\Ext_A^p(K, I^nM) \to \Ext_A^p(K, M)\) is contained in \(I^{n - c}\Ext_A^p(K, M)\) for \(n \geq c\).
Proof
Choose a bounded above complex \(P^\bullet\) of finite free \(A\)-modules representing \(K\). Then \(\Ext_A^p(K, M)\) is the cohomology of \[\Hom_A(P^{-p + 1}, M) \xrightarrow{a} \Hom_A(P^{-p}, M) \xrightarrow{b} \Hom_A(P^{-p - 1}, M)\] and \(\Ext_A^p(K, I^nM)\) is computed by replacing these finite \(A\)-modules by \(I^n\) times themselves. Thus the result by Lemma 0G3K (and much more is true).
In Situation 0BKC we define complexes \(I_n^\bullet\) such that we have distinguished triangles \[I_n^\bullet \to A \to K_n^\bullet \to I_n^\bullet[1]\] in the triangulated category \(K(A)\) of complexes of \(A\)-modules up to homotopy. Namely, we set \(I_n^\bullet = \sigma_{\leq -1}K_n^\bullet[-1]\). We have termwise split short exact sequences of complexes \[0 \to A \to K_n^\bullet \to I_n^\bullet[1] \to 0\] defining distinguished triangles by definition of the triangulated structure on \(K(A)\). Their rotations determine the desired distinguished triangles above. Note that \(I_n^0 = A^{\oplus r} \to A\) is given by multiplication by \(f_i^n\) on the \(i\)th factor. Hence \(I_n^\bullet \to A\) factors as \[I_n^\bullet \to (f_1^n, \ldots, f_r^n) \to A\] In fact, there is a short exact sequence \[0 \to H^{-1}(K_n^\bullet) \to H^0(I_n^\bullet) \to (f_1^n, \ldots, f_r^n) \to 0\] and for every \(i < 0\) we have \(H^i(I_n^\bullet) = H^{i - 1}(K_n^\bullet)\). The maps \(K_{n + 1}^\bullet \to K_n^\bullet\) induce maps \(I_{n + 1}^\bullet \to I_n^\bullet\) and we obtain a commutative diagram \[\xymatrix{ \ldots \ar[r] & I_3^\bullet \ar[d] \ar[r] & I_2^\bullet \ar[d] \ar[r] & I_1^\bullet \ar[d] \\ \ldots \ar[r] & (f_1^3, \ldots, f_r^3) \ar[r] & (f_1^2, \ldots, f_r^2) \ar[r] & (f_1, \ldots, f_r) }\] in \(K(A)\).
Lemma
In Situation 0BKC assume \(A\) is Noetherian. With notation as above, the inverse system \((I^n)\) is pro-isomorphic in \(D(A)\) to the inverse system \((I_n^\bullet)\).
Proof
It is elementary to show that the inverse system \(I^n\) is pro-isomorphic to the inverse system \((f_1^n, \ldots, f_r^n)\) in the category of \(A\)-modules. Consider the inverse system of distinguished triangles \[I_n^\bullet \to (f_1^n, \ldots, f_r^n) \to C_n^\bullet \to I_n^\bullet[1]\] where \(C_n^\bullet\) is the cone of the first arrow. By Derived Categories, Lemma 0G3C it suffices to show that the inverse system \(C_n^\bullet\) is pro-zero. The complex \(I_n^\bullet\) has nonzero terms only in degrees \(i\) with \(-r + 1 \leq i \leq 0\) hence \(C_n^\bullet\) is bounded similarly. Thus by Derived Categories, Lemma 0G3B it suffices to show that \(H^p(C_n^\bullet)\) is pro-zero. By the discussion above we have \(H^p(C_n^\bullet) = H^p(K_n^\bullet)\) for \(p \leq -1\) and \(H^p(C_n^\bullet) = 0\) for \(p \geq 0\). The fact that the inverse systems \(H^p(K_n^\bullet)\) are pro-zero was shown in the proof of Lemma 0921 (and this is where the assumption that \(A\) is Noetherian is used).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M^\bullet\) be a bounded complex of finite \(A\)-modules. The inverse system of maps \[I^n \otimes_A^\mathbf{L} M^\bullet \longrightarrow I^nM^\bullet\] defines an isomorphism of pro-objects of \(D(A)\).
Proof
Choose generators \(f_1, \ldots, f_r \in I\) of \(I\). The inverse system \(I^n\) is pro-isomorphic to the inverse system \((f_1^n, \ldots, f_r^n)\) in the category of \(A\)-modules. With notation as in Lemma 0G3M we find that it suffices to prove the inverse system of maps \[I_n^\bullet \otimes_A^\mathbf{L} M^\bullet \longrightarrow (f_1^n, \ldots, f_r^n)M^\bullet\] defines an isomorphism of pro-objects of \(D(A)\). Say we have \(a \leq b\) such that \(M^i = 0\) if \(i \not \in [a, b]\). Then source and target of the arrows above have cohomology only in degrees \([-r + a, b]\). Thus it suffices to show that for any \(p \in \mathbf{Z}\) the inverse system of maps \[H^p(I_n^\bullet \otimes_A^\mathbf{L} M^\bullet) \longrightarrow H^p((f_1^n, \ldots, f_r^n)M^\bullet)\] defines an isomorphism of pro-objects of \(A\)-modules, see Derived Categories, Lemma 0G3D. Using the pro-isomorphism between \(I_n^\bullet \otimes_A^\mathbf{L} M^\bullet\) and \(I^n \otimes_A^\mathbf{L} M^\bullet\) and the pro-isomorphism between \((f_1^n, \ldots, f_r^n)M^\bullet\) and \(I^nM^\bullet\) this is equivalent to showing that the inverse system of maps \[H^p(I^n \otimes_A^\mathbf{L} M^\bullet) \longrightarrow H^p(I^nM^\bullet)\] defines an isomorphism of pro-objects of \(A\)-modules Choose a bounded above complex of finite free \(A\)-modules \(P^\bullet\) and a quasi-isomorphism \(P^\bullet \to M^\bullet\). Then it suffices to show that the inverse system of maps \[H^p(I^nP^\bullet) \longrightarrow H^p(I^nM^\bullet)\] is a pro-isomorphism. This follows from Lemma 0G3K as \(H^p(P^\bullet) = H^p(M^\bullet)\).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\) be a finite \(A\)-module. There exists an integer \(n > 0\) such that \(I^nM \to M\) factors through the map \(I \otimes_A^\mathbf{L} M \to M\) in \(D(A)\).
Proof
This follows from Lemma 0G3N. It can also be seen directly as follows. Consider the distinguished triangle \[I \otimes_A^\mathbf{L} M \to M \to A/I \otimes_A^\mathbf{L} M \to I \otimes_A^\mathbf{L} M[1]\] By the axioms of a triangulated category it suffices to prove that \(I^nM \to A/I \otimes_A^\mathbf{L} M\) is zero in \(D(A)\) for some \(n\). Choose generators \(f_1, \ldots, f_r\) of \(I\) and let \(K = K_\bullet(A, f_1, \ldots, f_r)\) be the Koszul complex and consider the factorization \(A \to K \to A/I\) of the quotient map. Then we see that it suffices to show that \(I^nM \to K \otimes_A M\) is zero in \(D(A)\) for some \(n > 0\). Suppose that we have found an \(n > 0\) such that \(I^nM \to K \otimes_A M\) factors through \(\tau_{\geq t}(K \otimes_A M)\) in \(D(A)\). Then the obstruction to factoring through \(\tau_{\geq t + 1}(K \otimes_A M)\) is an element in \(\Ext^t(I^nM, H_t(K \otimes_A M))\). The finite \(A\)-module \(H_t(K \otimes_A M)\) is annihilated by \(I\). Then by Lemma 0927 we can after increasing \(n\) assume this obstruction element is zero. Repeating this a finite number of times we find \(n\) such that \(I^nM \to K \otimes_A M\) factors through \(0 = \tau_{\geq r + 1}(K \otimes_A M)\) in \(D(A)\) and we win.
Miscellany
Some results which do not fit anywhere else.
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(K \in D(A)\) be pseudo-coherent. Let \(a \in \mathbf{Z}\). Assume that for every finite \(A\)-module \(M\) the modules \(\Ext^i_A(K, M)\) are \(I\)-power torsion for \(i \geq a\). Then for \(i \geq a\) and \(M\) finite the system \(\Ext^i_A(K, M/I^nM)\) is essentially constant with value \[\Ext^i_A(K, M) = \lim \Ext^i_A(K, M/I^nM)\]
Proof
Let \(M\) be a finite \(A\)-module. Since \(K\) is pseudo-coherent we see that \(\Ext^i_A(K, M)\) is a finite \(A\)-module. Thus for \(i \geq a\) it is annihilated by \(I^t\) for some \(t \geq 0\). By Lemma 0DYI we see that the image of \(\Ext^i_A(K, I^nM) \to \Ext^i_A(K, M)\) is zero for some \(n > 0\). The short exact sequence \(0 \to I^nM \to M \to M/I^n M \to 0\) gives a long exact sequence \[\Ext^i_A(K, I^nM) \to \Ext^i_A(K, M) \to \Ext^i_A(K, M/I^nM) \to \Ext^{i + 1}_A(K, I^nM)\] The systems \(\Ext^i_A(K, I^nM)\) and \(\Ext^{i + 1}_A(K, I^nM)\) are essentially constant with value \(0\) by what we just said (applied to the finite \(A\)-modules \(I^mM\)). A diagram chase shows \(\Ext^i_A(K, M/I^nM)\) is essentially constant with value \(\Ext^i_A(K, M)\).
Lemma
Let \(A\) be a Noetherian ring. Let \(I \subset A\) be an ideal. Let \(M\) be a finite \(A\)-module. Let \(N\) be an \(A\)-module annihilated by \(I\). There exists an integer \(n > 0\) such that \(\text{Tor}^A_p(I^nM, N) \to \text{Tor}^A_p(M, N)\) is zero for all \(p \geq 0\).
Proof
By Lemma 0928 we can factor \(I^nM \to M\) as \(I^nM \to M \otimes_A^\mathbf{L} I \to M\). We claim the composition \[I^nM \otimes_A^\mathbf{L} N \to (M \otimes_A^\mathbf{L} I) \otimes_A^\mathbf{L} N \to M \otimes_A^\mathbf{L} N\] is zero. Namely, the diagram \[\xymatrix{ (M \otimes_A^\mathbf{L} I) \otimes_A^\mathbf{L} N \ar[rr] \ar[rd] & & M \otimes_A^\mathbf{L} (I \otimes_A^\mathbf{L} N) \ar[ld] \\ & M \otimes_A^\mathbf{L} N }\] commutes (details omitted) and the map \(I \otimes_A^\mathbf{L} N \to N\) is zero as \(N\) is annihilated by \(I\).
Lemma
Let \(R\) be a ring. Let \(K \in D(R)\) be pseudo-coherent. Let \((M_n)\) be an inverse system of \(R\)-modules. Then \(R\lim K \otimes_R^\mathbf{L} M_n = K \otimes_R^\mathbf{L} R\lim M_n\).
Proof
Consider the defining distinguished triangle \[R\lim M_n \to \prod M_n \to \prod M_n \to R\lim M_n[1]\] and apply Lemma 0CYB.
Lemma
Let \(R\) be a Noetherian local ring. Let \(I \subset R\) be an ideal and let \(E\) be a nonzero finite module over \(R/I\). If \(R/I\) has finite projective dimension and \(E\) has finite projective dimension over \(R/I\), then \(E\) has finite projective dimension over \(R\) and \[\text{pd}_R(E) = \text{pd}_R(R/I) + \text{pd}_{R/I}(E)\]
Proof
We will use that, for a finite module, having finite projective dimension over \(R\), resp. \(R/I\) is the same as being a perfect module, see discussion following Definition 0657. We see that \(E\) has finite projective dimension over \(R\) by Lemma 066U. Thus we can apply Auslander-Buchsbaum (Algebra, Proposition 090V) to see that \[\text{pd}_R(E) + \text{depth}(E) = \text{depth}(R),\quad \text{pd}_{R/I}(E) + \text{depth}(E) = \text{depth}(R/I),\] and \[\text{pd}_R(R/I) + \text{depth}(R/I) = \text{depth}(R)\] Note that in the first equation we take the depth of \(E\) as an \(R\)-module and in the second as an \(R/I\)-module. However these depths are the same (this is trivial but also follows from Algebra, Lemma 0AUK). This concludes the proof.
Lemma
Let \(A \to B\) be a ring map. There exists a cardinal \(\kappa = \kappa(A \to B)\) with the following property: Let \(M^\bullet\), resp. \(N^\bullet\) be a complex of \(A\)-modules, resp. \(B\)-modules. Let \(a : M^\bullet \to N^\bullet\) be a map of complexes of \(A\)-modules which induces an isomorphism \(M^\bullet \otimes_A^\mathbf{L} B \to N^\bullet\) in \(D(B)\). Let \(M_1^\bullet \subset M^\bullet\), resp. \(N_1^\bullet \subset N^\bullet\) be a subcomplex of \(A\)-modules, resp. \(B\)-modules such that \(a(M_1^\bullet) \subset N_1^\bullet\). Then there exist subcomplexes \[M_1^\bullet \subset M_2^\bullet \subset M^\bullet \quad\text{and}\quad N_1^\bullet \subset N_2^\bullet \subset N^\bullet\] such that \(a(M_2^\bullet) \subset N_2^\bullet\) with the following properties:
\(\Ker(H^i(M_1^\bullet \otimes_A^\mathbf{L} B) \to H^i(N_1^\bullet))\) maps to zero in \(H^i(M_2^\bullet \otimes_A^\mathbf{L} B)\),
\(\Im(H^i(N_1^\bullet) \to H^i(N_2^\bullet))\) is contained in \(\Im(H^i(M_2^\bullet \otimes_A^\mathbf{L} B) \to H^i(N_2^\bullet))\),
\(|\bigcup M_2^i \cup \bigcup N_2^i| \leq \max(\kappa, |\bigcup M_1^i \cup \bigcup N_1^i|)\).
Proof
Let \(\kappa = \max(|A|, |B|, \aleph_0)\). Set \(|M^\bullet| = |\bigcup M^i|\) and similarly for other complexes. With this notation we have \[\max(\kappa, |\bigcup M_1^i \cup \bigcup N_1^i|) = \max(\kappa, |M_1^\bullet|, |N_1^\bullet|)\] for the quantity used in the statement of the lemma. We are going to use this and other observations coming from arithmetic of cardinals without further mention.
First, let us show that there are plenty of “small” subcomplexes. For every pair of collections \(E = \{E^i\}\) and \(F = \{F^i\}\) of finite subsets \(E^i \subset M^i\), \(i \in \mathbf{Z}\) and \(F^i \subset N^i\), \(i \in \mathbf{Z}\) we can let \[M_1^\bullet \subset M_1(E, F)^\bullet \subset M^\bullet \quad\text{and}\quad N_1^\bullet \subset N_1(E, F)^\bullet \subset N^\bullet\] be the smallest subcomplexes of \(A\) and \(B\)-modules such that \(a(M_1(E, F)^\bullet) \subset N_1(E, F)^\bullet\) and such that \(E^i \subset M_1(E, F)^i\) and \(F^i \subset N_1(E, F)^i\). Then it is easy to see that \[|M_1(E, F)^\bullet| \leq \max(\kappa, |M_1^\bullet|) \quad\text{and}\quad |N_1(E, F)^\bullet| \leq \max(\kappa, |M_1^\bullet|, |N_1^\bullet|)\] Details omitted. It is clear that we have \[M^\bullet = \colim_{(E, F)} M_1(E, F)^\bullet \quad\text{and}\quad N^\bullet = \colim_{(E, F)} N_1(E, F)^\bullet\] and the colimits are (termwise) filtered colimits.
There exists a resolution \(\ldots \to F^{-1} \to F^0 \to B\) by free \(A\)-modules \(F^i\) with \(|F^i| \leq \kappa\) (details omitted). The cohomology modules of \(M_1^\bullet \otimes_A^\mathbf{L} B\) are computed by \(\text{Tot}(M_1^\bullet \otimes_A F^\bullet)\). It follows that \(|H^i(M_1^\bullet \otimes_A^\mathbf{L} B)| \leq \max(\kappa, |M_1^\bullet|)\).
Let \(i \in \mathbf{Z}\) and let \(\xi \in H^i(M_1^\bullet \otimes_A^\mathbf{L} B)\) be an element which maps to zero in \(H^i(N_1^\bullet)\). Then \(\xi\) maps to zero in \(H^i(N^\bullet)\) and hence \(\xi\) maps to zero in \(H^i(M^\bullet \otimes_A^\mathbf{L} B)\). Since derived tensor product commutes with filtered colimits, we can find finite collections \(E_\xi\) and \(F_\xi\) as above such that \(\xi\) maps to zero in \(H^i(M_1(E_\xi, F_\xi)^\bullet \otimes_A^\mathbf{L} B)\).
Let \(i \in \mathbf{Z}\) and let \(\eta \in H^i(N_1^\bullet)\). Then the image of \(\eta\) in \(H^i(N^\bullet)\) is in the image of \(H^i(M^\bullet \otimes_A^\mathbf{L} B) \to H^i(N^\bullet)\). Hence as before, we can find finite collections \(E_\eta\) and \(F_\eta\) as above such that \(\eta\) maps to an element of \(H^i(N_1(E_\eta, F_\eta)^\bullet)\) which is in the image of the map \(H^i(M_1(E_\eta, F_\eta)^\bullet \otimes_A^\mathbf{L} B) \to H^i(N_1(E_\eta, F_\eta)^\bullet)\).
Now we simply define \[M_2^\bullet = \sum\nolimits_\xi M_1(E_\xi, F_\xi)^\bullet + \sum\nolimits_\eta M_1(E_\eta, F_\eta)^\bullet\] where the sum is over \(\xi\) and \(\eta\) as in the previous two paragraphs and the sum is taken inside \(M^\bullet\). Similarly we set \[N_2^\bullet = \sum\nolimits_\xi N_1(E_\xi, F_\xi)^\bullet + \sum\nolimits_\eta N_1(E_\eta, F_\eta)^\bullet\] where the sum is taken inside \(N^\bullet\). By construction we will have properties (1) and (2) with these choices. The bound (3) also follows as the set of \(\xi\) and \(\eta\) has cardinality at most \(\max(\kappa, |M_1^\bullet|, |N_1^\bullet|)\).
Lemma
Let \(R\) be a ring and \(f \in R\). Let \(M \in D(R)\) and let \(C\) be the cone of \(f : M \to M\). If \(H^i(M)_f = 0\) for \(i < 0\) and \(H^i(C) = 0\) for \(i < -1\), then \(H^i(M) = 0\) for \(i < 0\).
Proof
Denote \(M_f = M \otimes_R^\mathbf{L} R_f\) and choose a distinguished triangle \(F \to M \to M_f\). By assumption \(H^i(M_f) = H^i(M)_f = 0\) for \(i < 0\). Thus it suffices to show that \(H^i(F) = 0\) for \(i < 0\). Note that \(H^i(F)\) is \(f\)-power torsion for all \(i \in \mathbf{Z}\). On the other hand, since \(f : M_f \to M_f\) is an isomorphism, we see that \(C\) is isomorphic to the cone of \(f : F \to F\) (use Derived Categories, Proposition 05R0). Now, if \(H^i(F) \not = 0\), then the kernel of \(f : H^i(F) \to H^i(F)\) is nonzero, which implies that \(H^{i - 1}(C)\) is nonzero. Our assumption implies this cannot happen if \(i - 1 < -1\) which finishes the proof.
Lemma
Let \(R\) be a ring and \(f \in R\). Let \(M \in D(R)\). Assume
\(H^i(M) = 0\) for \(i > 0\),
\(M \otimes_R^\mathbf{L} R_f\) is isomorphic to a flat \(R_f\)-module placed in degree \(0\),
\(M \otimes_R^\mathbf{L} R/fR\) is isomorphic to a flat \(R/fR\)-module placed in degree \(0\).
Then \(M\) is isomorphic to a flat \(R\)-module placed in degree \(0\).
Proof
Let \(K\) be an \(R\)-module. It suffices to show that \(N = M \otimes_R^\mathbf{L} K\) is isomorphic to an \(R\)-module placed in degree \(0\), see Section 0651. By assumption (1) we see that \(N\) only has nonvanishing cohomology in degrees \(\leq 0\). Thus, by Lemma 0H84, it suffices to show that \(H^i(N)_f = 0\) for \(i < 0\) and that \(H^i(C) = 0\) for \(i < -1\) where \(C\) is the cone of \(f : N \to N\). For the first, we note that \[H^i(N)_f = H^i(N \otimes_R^\mathbf{L} R_f) = H^i(M \otimes_R^\mathbf{L} K \otimes_R^\mathbf{L} R_f ) = H^i((M \otimes_R^\mathbf{L} R_f) \otimes_{R_f}^\mathbf{L} (K \otimes_R^\mathbf{L} R_f))\] By our assumption (2) this is zero, except if \(i = 0\) and then one gets \(H^0(M)_f \otimes_R K\). For the second, let \(C'\) be the cone of \(f : K \to K\). Then \[C = M \otimes_R^\mathbf{L} C'\] Now \(C'\) only has nonzero cohomology in degrees \(0\) and \(-1\) equal to \(K/fK\) and \(K[f]\); in other words, there is a distinguished triangle \[(K[f])[1] \to C' \to K/fK\] On the other hand, for any \(R\)-module \(K'\) annihilated by \(f\) we have \[M \otimes_R^\mathbf{L} K' = M \otimes_R^\mathbf{L} R/fR \otimes_{R/fR}^\mathbf{L} K'\] By our assumption (3) this is equal to the module \(H^0(M \otimes_R^\mathbf{L} R/fR) \otimes_{R/fR} K'\) placed in degree \(0\). Combining the above we conclude that \(C\) only has nonzero cohomology in degrees \(0\) and \(-1\) and the proof is complete.
Tricks with double complexes
This section continues the discussion in Homology, Section 0E1P.
Lemma
Let \(A_0^\bullet \to A_1^\bullet \to A_2^\bullet \to \ldots\) be a complex of complexes of abelian groups. Assume \(H^{-p}(A_p^\bullet) = 0\) for all \(p \geq 0\). Set \(A^{p, q} = A_p^q\) and view \(A^{\bullet, \bullet}\) as a double complex. Then \(H^0(\text{Tot}_\pi(A^{\bullet, \bullet})) = 0\).
Proof
Denote \(f_p : A_p^\bullet \to A_{p + 1}^\bullet\) the given maps of complexes. Recall that the differential on \(\text{Tot}_\pi(A^{\bullet, \bullet})\) is given by \[\prod\nolimits_{p + q = n} (f^q_p + (-1)^p\text{d}^q_{A_p^\bullet})\] on elements in degree \(n\). Let \(\xi \in H^0(\text{Tot}_\pi(A^{\bullet, \bullet}))\) be a cohomology class. We will show \(\xi\) is zero. Represent \(\xi\) as the class of an cocycle \(x = (x_p) \in \prod A^{p, -p}\). Since \(\text{d}(x) = 0\) we find that \(\text{d}_{A_0^\bullet}(x_0) = 0\). Since \(H^0(A_0^\bullet) = 0\) there exists a \(y_{-1} \in A^{0, -1}\) with \(\text{d}_{A_0^\bullet}(y_{-1}) = x_0\). Then we see that \(\text{d}_{A_1^\bullet}(x_1 + f_0(y_{-1})) = 0\). Since \(H^{-1}(A_1^\bullet) = 0\) we can find a \(y_{-2} \in A^{1, -2}\) such that \(-\text{d}_{A_1^\bullet}(y_{-2}) = x_1 + f_0(y_{-1})\). By induction we can find \(y_{-p - 1} \in A^{p, -p - 1}\) such that \[(-1)^p\text{d}_{A_p^\bullet}(y_{-p - 1}) = x_p + f_{p - 1}(y_{-p})\] This implies that \(\text{d}(y) = x\) where \(y = (y_{-p - 1})\).
Lemma
Let \[(A_0^\bullet \to A_1^\bullet \to A_2^\bullet \to \ldots) \longrightarrow (B_0^\bullet \to B_1^\bullet \to B_2^\bullet \to \ldots)\] be a map between two complexes of complexes of abelian groups. Set \(A^{p, q} = A_p^q\), \(B^{p, q} = B_p^q\) to obtain double complexes. Let \(\text{Tot}_\pi(A^{\bullet, \bullet})\) and \(\text{Tot}_\pi(B^{\bullet, \bullet})\) be the product total complexes associated to the double complexes. If each \(A_p^\bullet \to B_p^\bullet\) is a quasi-isomorphism, then \(\text{Tot}_\pi(A^{\bullet, \bullet}) \to \text{Tot}_\pi(B^{\bullet, \bullet})\) is a quasi-isomorphism.
Proof
Recall that \(\text{Tot}_\pi(A^{\bullet, \bullet})\) in degree \(n\) is given by \(\prod_{p + q = n} A^{p, q} = \prod_{p + 1 = n} A^q_p\). Let \(C_p^\bullet\) be the cone on the map \(A_p^\bullet \to B_p^\bullet\), see Derived Categories, Section 014D. By the functoriality of the cone construction we obtain a complex of complexes \[C_0^\bullet \to C_1^\bullet \to C_2^\bullet \to \ldots\] Then we see \(\text{Tot}_\pi(C^{\bullet, \bullet})\) in degree \(n\) is given by \[\prod_{p + q = n} C^{p, q} = \prod_{p + q = n} C^q_p = \prod_{p + q = n} (B^q_p \oplus A^{q + 1}_p) = \prod_{p + q = n} B^q_p \oplus \prod_{p + q = n} A^{q + 1}_p\] We conclude that \(\text{Tot}_\pi(C^{\bullet, \bullet})\) is the cone of the map \(\text{Tot}_\pi(A^{\bullet, \bullet}) \to \text{Tot}_\pi(B^{\bullet, \bullet})\) (We omit the verification that the differentials agree.) Thus it suffices to show \(\text{Tot}_\pi(A^{\bullet, \bullet})\) is acyclic if each \(A_p^\bullet\) is acyclic. This follows from Lemma 0H0Q.
Weakly étale ring maps
Most of the results in this section are from the paper [Olivier-AF] by Olivier. See also the related paper [Ferrand-epi].
Definition
A ring \(A\) is called absolutely flat if every \(A\)-module is flat over \(A\). A ring map \(A \to B\) is weakly étale or absolutely flat if both \(A \to B\) and \(B \otimes_A B \to B\) are flat.
Absolutely flat rings are sometimes called von Neumann regular rings (often in the setting of noncommutative rings). A localization is a weakly étale ring map. An étale ring map is weakly étale. Here is a simple, yet key property.
Lemma
Let \(A \to B\) be a ring map such that \(B \otimes_A B \to B\) is flat. Let \(N\) be a \(B\)-module. If \(N\) is flat as an \(A\)-module, then \(N\) is flat as a \(B\)-module.
Proof
Assume \(N\) is a flat as an \(A\)-module. Then the functor \[\text{Mod}_B \longrightarrow \text{Mod}_{B \otimes_A B},\quad N' \mapsto N \otimes_A N'\] is exact. As \(B \otimes_A B \to B\) is flat we conclude that the functor \[\text{Mod}_B \longrightarrow \text{Mod}_B,\quad N' \mapsto (N \otimes_A N') \otimes_{B \otimes_A B} B = N \otimes_B N'\] is exact, hence \(N\) is flat over \(B\).
Definition
Let \(A\) be a ring. Let \(d \geq 0\) be an integer. We say that \(A\) has weak dimension \(\leq d\) if every \(A\)-module has tor dimension \(\leq d\).
Lemma
Let \(A \to B\) be a weakly étale ring map. If \(A\) has weak dimension at most \(d\), then so does \(B\).
Proof
Let \(N\) be a \(B\)-module. If \(d = 0\), then \(N\) is flat as an \(A\)-module, hence flat as a \(B\)-module by Lemma 092C. Assume \(d > 0\). Choose a resolution \(F_\bullet \to N\) by free \(B\)-modules. Our assumption implies that \(K = \Im(F_d \to F_{d - 1})\) is \(A\)-flat, see Lemma 0653. Hence it is \(B\)-flat by Lemma 092C. Thus \(0 \to K \to F_{d - 1} \to \ldots \to F_0 \to N \to 0\) is a flat resolution of length \(d\) and we see that \(N\) has tor dimension at most \(d\).
Lemma
Let \(A\) be a ring. The following are equivalent
\(A\) has weak dimension \(\leq 0\),
\(A\) is absolutely flat,
\(A\) is reduced and every prime is maximal, and
every local ring of \(A\) is a field.
Proof
The equivalence of (1) and (2) is immediate.
Proof of (1) \(\Rightarrow\) (3). Assume \(A\) is absolutely flat. Then every ideal of \(A\) is pure, see Algebra, Definition 04PR. Hence every finitely generated ideal is generated by an idempotent by Algebra, Lemma 05KK. If \(f \in A\), then \((f) = (e)\) for some idempotent \(e \in A\) and \(D(f) = D(e)\) is open and closed (Algebra, Lemma 00EC). This implies every prime ideal of \(A\) is maximal by Algebra, Lemma 04MG. Moreover, if \(f\) is nilpotent, then \(e = 0\) hence \(f = 0\). Thus \(A\) is reduced.
Proof of (3) \(\Rightarrow\) (4). If \(A\) is reduced and every prime is maximal, then every local ring is a field by Algebra, Lemma 00EU.
Proof of (4) \(\Rightarrow\) (2). If every local ring of \(A\) is a field, then every \(A\)-module is flat by Algebra, Lemma 00HT.
Lemma
A product of fields is an absolutely flat ring.
Proof
Let \(K_i\) be a family of fields. If \(f = (f_i) \in \prod K_i\), then the ideal generated by \(f\) is the same as the ideal generated by the idempotent \(e = (e_i)\) with \(e_i = 0, 1\) according to whether \(f_i\) is \(0\) or not. Thus \(D(f) = D(e)\) is open and closed and we conclude by Lemma 092F and Algebra, Lemma 04MG.
Lemma
Let \(A \to B\) and \(A \to A'\) be ring maps. Let \(B' = B \otimes_A A'\) be the base change of \(B\).
If \(B \otimes_A B \to B\) is flat, then \(B' \otimes_{A'} B' \to B'\) is flat.
If \(A \to B\) is weakly étale, then \(A' \to B'\) is weakly étale.
Proof
Assume \(B \otimes_A B \to B\) is flat. The ring map \(B' \otimes_{A'} B' \to B'\) is the base change of \(B \otimes_A B \to B\) by \(A \to A'\). Hence it is flat by Algebra, Lemma 00HI. This proves (1). Part (2) follows from (1) and the fact (just used) that the base change of a flat ring map is flat.
Lemma
Let \(A \to B\) be a ring map such that \(B \otimes_A B \to B\) is flat.
If \(A\) is an absolutely flat ring, then so is \(B\).
If \(A\) is reduced and \(A \to B\) is weakly étale, then \(B\) is reduced.
Proof
Part (1) follows immediately from Lemma 092C and the definitions. If \(A\) is reduced, then there exists an injection \(A \to A' = \prod_{\mathfrak p \subset A\text{ minimal}} A_\mathfrak p\) of \(A\) into an absolutely flat ring (Algebra, Lemma 00EW and Lemma 092G). If \(A \to B\) is flat, then the induced map \(B \to B' = B \otimes_A A'\) is injective too. By Lemma 092H the ring map \(A' \to B'\) is weakly étale. By part (1) we see that \(B'\) is absolutely flat. By Lemma 092F the ring \(B'\) is reduced. Hence \(B\) is reduced.
Lemma
Let \(A \to B\) and \(B \to C\) be ring maps.
If \(B \otimes_A B \to B\) and \(C \otimes_B C \to C\) are flat, then \(C \otimes_A C \to C\) is flat.
If \(A \to B\) and \(B \to C\) are weakly étale, then \(A \to C\) is weakly étale.
Proof
Part (1) follows from the factorization \[C \otimes_A C \longrightarrow C \otimes_B C \longrightarrow C\] of the multiplication map, the fact that \[C \otimes_B C = (C \otimes_A C) \otimes_{B \otimes_A B} B,\] the fact that a base change of a flat map is flat, and the fact that the composition of flat ring maps is flat. See Algebra, Lemmas 00HI and 00HC. Part (2) follows from (1) and the fact (just used) that the composition of flat ring maps is flat.
Lemma
Let \(A \to B \to C\) be ring maps.
If \(B \to C\) is faithfully flat and \(C \otimes_A C \to C\) is flat, then \(B \otimes_A B \to B\) is flat.
If \(B \to C\) is faithfully flat and \(A \to C\) is weakly étale, then \(A \to B\) is weakly étale.
Proof
Assume \(B \to C\) is faithfully flat and \(C \otimes_A C \to C\) is flat. Consider the commutative diagram \[\xymatrix{ C \otimes_A C \ar[r] & C \\ B \otimes_A B \ar[r] \ar[u] & B \ar[u] }\] The vertical arrows are flat, the top horizontal arrow is flat. Hence \(C\) is flat as a \(B \otimes_A B\)-module. The map \(B \to C\) is faithfully flat and \(C = B \otimes_B C\). Hence \(B\) is flat as a \(B \otimes_A B\)-module by Algebra, Lemma 0584. This proves (1). Part (2) follows from (1) and the fact that \(A \to B\) is flat if \(A \to C\) is flat and \(B \to C\) is faithfully flat (Algebra, Lemma 0584).
Lemma
Let \(A\) be a ring. Let \(B \to C\) be an \(A\)-algebra map of weakly étale \(A\)-algebras. Then \(B \to C\) is weakly étale.
Proof
The ring map \(B \to C\) is flat by Lemma 092C. The ring map \(C \otimes_A C \to C \otimes_B C\) is surjective, hence an epimorphism. Thus Lemma 092C implies, that since \(C\) is flat over \(C \otimes_A C\) also \(C\) is flat over \(C \otimes_B C\).
Lemma
Let \(A \to B\) be a ring map such that \(B \otimes_A B \to B\) is flat. Then \(\Omega_{B/A} = 0\), i.e., \(B\) is formally unramified over \(A\).
Proof
Let \(I \subset B \otimes_A B\) be the kernel of the flat surjective map \(B \otimes_A B \to B\). Then \(I\) is a pure ideal (Algebra, Definition 04PR), so \(I^2 = I\) (Algebra, Lemma 04PS). Since \(\Omega_{B/A} = I/I^2\) (Algebra, Lemma 00RW) we obtain the vanishing. This means \(B\) is formally unramified over \(A\) by Algebra, Lemma 00UO.
Lemma
Let \(A \to B\) be a ring map such that \(B \otimes_A B \to B\) is flat.
If \(A \to B\) is of finite type, then \(A \to B\) is unramified.
If \(A \to B\) is of finite presentation and flat, then \(A \to B\) is étale.
In particular a weakly étale ring map of finite presentation is étale.
Proof
Part (1) follows from Lemma 092M and Algebra, Definition 00UT. Part (2) follows from part (1) and Algebra, Lemma 08WD.
Lemma
Let \(A \to B\) be a ring map. Then \(A \to B\) is weakly étale in each of the following cases
\(B = S^{-1}A\) is a localization of \(A\),
\(A \to B\) is étale,
\(B\) is a filtered colimit of weakly étale \(A\)-algebras.
Proof
An étale ring map is flat and the map \(B \otimes_A B \to B\) is also étale as a map between étale \(A\)-algebras (Algebra, Lemma 00U7). This proves (2).
Let \(B_i\) be a directed system of weakly étale \(A\)-algebras. Then \(B = \colim B_i\) is flat over \(A\) by Algebra, Lemma 05UT. Note that the transition maps \(B_i \to B_{i'}\) are flat by Lemma 092L. Hence \(B\) is flat over \(B_i\) for each \(i\), and we see that \(B\) is flat over \(B_i \otimes_A B_i\) by Algebra, Lemma 00HC. Thus \(B\) is flat over \(B \otimes_A B = \colim B_i \otimes_A B_i\) by Algebra, Lemma 05UU.
Part (1) can be proved directly, but also follows by combining (2) and (3).
Lemma
Let \(L/K\) be an extension of fields. If \(L \otimes_K L \to L\) is flat, then \(L\) is an algebraic separable extension of \(K\).
Proof
By Lemma 092K we see that any subfield \(K \subset L' \subset L\) the map \(L' \otimes_K L' \to L'\) is flat. Thus we may assume \(L\) is a finitely generated field extension of \(K\). In this case the fact that \(L/K\) is formally unramified (Lemma 092M) implies that \(L/K\) is finite separable, see Algebra, Lemma 090W.
Lemma
Let \(B\) be an algebra over a field \(K\). The following are equivalent
\(B \otimes_K B \to B\) is flat,
\(K \to B\) is weakly étale, and
\(B\) is a filtered colimit of étale \(K\)-algebras.
Moreover, every finitely generated \(K\)-subalgebra of \(B\) is étale over \(K\).
Proof
Parts (1) and (2) are equivalent because every \(K\)-algebra is flat over \(K\). Part (3) implies (1) and (2) by Lemma 092N
Assume (1) and (2) hold. We will prove (3) and the finite statement of the lemma. A field is absolutely flat ring, hence \(B\) is a absolutely flat ring by Lemma 092I. Hence \(B\) is reduced and every local ring is a field, see Lemma 092F.
Let \(\mathfrak q \subset B\) be a prime. The ring map \(B \to B_\mathfrak q\) is weakly étale, hence \(B_\mathfrak q\) is weakly étale over \(K\) (Lemma 092J). Thus \(B_\mathfrak q\) is a separable algebraic extension of \(K\) by Lemma 092P.
Let \(K \subset A \subset B\) be a finitely generated \(K\)-sub algebra. We will show that \(A\) is étale over \(K\) which will finish the proof of the lemma. Then every minimal prime \(\mathfrak p \subset A\) is the image of a prime \(\mathfrak q\) of \(B\), see Algebra, Lemma 00FK. Thus \(\kappa(\mathfrak p)\) as a subfield of \(B_\mathfrak q = \kappa(\mathfrak q)\) is separable algebraic over \(K\). Hence every generic point of \(\Spec(A)\) is closed (Algebra, Lemma 00GA). Thus \(\dim(A) = 0\). Then \(A\) is the product of its local rings, e.g., by Algebra, Proposition 00KJ. Moreover, since \(A\) is reduced, all local rings are equal to their residue fields which are finite separable over \(K\). This means that \(A\) is étale over \(K\) by Algebra, Lemma 00U3 and finishes the proof.
Lemma
Let \(A \to B\) be a ring map. If \(A \to B\) is weakly étale, then \(A \to B\) induces separable algebraic residue field extensions.
Proof
Let \(\mathfrak p\) be a prime of \(A\). Then \(\kappa(\mathfrak p) \to B \otimes_A \kappa(\mathfrak p)\) is weakly étale by Lemma 092H. Hence \(B \otimes_A \kappa(\mathfrak p)\) is a filtered colimit of étale \(\kappa(\mathfrak p)\)-algebras by Lemma 092Q. Hence for \(\mathfrak q \subset B\) lying over \(\mathfrak p\) the extension \(\kappa(\mathfrak q)/\kappa(\mathfrak p)\) is a filtered colimit of finite separable extensions by Algebra, Lemma 00U3.
Lemma
Let \(A\) be a ring. The following are equivalent
\(A\) has weak dimension \(\leq 1\),
every ideal of \(A\) is flat,
every finitely generated ideal of \(A\) is flat,
every submodule of a flat \(A\)-module is flat, and
every local ring of \(A\) is a valuation ring.
Proof
If \(A\) has weak dimension \(\leq 1\), then the resolution \(0 \to I \to A \to A/I \to 0\) shows that every ideal \(I\) is flat by Lemma 0653. Hence (1) \(\Rightarrow\) (2).
Assume (4). Let \(M\) be an \(A\)-module. Choose a surjection \(F \to M\) where \(F\) is a free \(A\)-module. Then \(\Ker(F \to M)\) is flat by assumption, and we see that \(M\) has tor dimension \(\leq 1\) by Lemma 066F. Hence (4) \(\Rightarrow\) (1).
Every ideal is the union of the finitely generated ideals contained in it. Hence (3) implies (2) by Algebra, Lemma 05UT. Thus (3) \(\Leftrightarrow\) (2).
Assume (2). Suppose that \(N \subset M\) with \(M\) a flat \(A\)-module. We will prove that \(N\) is flat. We can write \(M = \colim M_i\) with each \(M_i\) finite free, see Algebra, Theorem 058G. Setting \(N_i \subset M_i\) the inverse image of \(N\) we see that \(N = \colim N_i\). By Algebra, Lemma 05UT. it suffices to prove \(N_i\) is flat and we reduce to the case \(M = R^{\oplus n}\). In this case the module \(N\) has a finite filtration by the submodules \(R^{\oplus j} \cap N\) whose subquotients are ideals. By (2) these ideals are flat and hence \(N\) is flat by Algebra, Lemma 00HM. Thus (2) \(\Rightarrow\) (4).
Assume \(A\) satisfies (1) and let \(\mathfrak p \subset A\) be a prime ideal. By Lemmas 092N and 092E we see that \(A_\mathfrak p\) satisfies (1). We will show \(A\) is a valuation ring if \(A\) is a local ring satisfying (3). Let \(f \in \mathfrak m\) be a nonzero element. Then \((f)\) is a flat nonzero module generated by one element. Hence it is a free \(A\)-module by Algebra, Lemma 00NZ. It follows that \(f\) is a nonzerodivisor and \(A\) is a domain. If \(I \subset A\) is a finitely generated ideal, then we similarly see that \(I\) is a finite free \(A\)-module, hence (by considering the rank) free of rank \(1\) and \(I\) is a principal ideal. Thus \(A\) is a valuation ring by Algebra, Lemma 090Q. Thus (1) \(\Rightarrow\) (5).
Assume (5). Let \(I \subset A\) be a finitely generated ideal. Then \(I_\mathfrak p \subset A_\mathfrak p\) is a finitely generated ideal in a valuation ring, hence principal (Algebra, Lemma 090Q), hence flat. Thus \(I\) is flat by Algebra, Lemma 00HT. Thus (5) \(\Rightarrow\) (3). This finishes the proof of the lemma.
Lemma
Let \(J\) be a set. For each \(j \in J\) let \(A_j\) be a valuation ring with fraction field \(K_j\). Set \(A = \prod A_j\) and \(K = \prod K_j\). Then \(A\) has weak dimension at most \(1\) and \(A \to K\) is a localization.
Proof
Let \(I \subset A\) be a finitely generated ideal. By Lemma 092S it suffices to show that \(I\) is a flat \(A\)-module. Let \(I_j \subset A_j\) be the image of \(I\). Observe that \(I_j = I \otimes_A A_j\), hence \(I \to \prod I_j\) is surjective by Algebra, Proposition 059J. Thus \(I = \prod I_j\). Since \(A_j\) is a valuation ring, the ideal \(I_j\) is generated by a single element (Algebra, Lemma 090Q). Say \(I_j = (f_j)\). Then \(I\) is generated by the element \(f = (f_j)\). Let \(e \in A\) be the idempotent which has a \(0\) or \(1\) in \(A_j\) depending on whether \(f_j\) is \(0\) or not. Then \(f = g e\) for some nonzerodivisor \(g \in A\): take \(g = (g_j)\) with \(g_j = 1\) if \(f_j = 0\) and \(g_j = f_j\) else. Thus \(I \cong (e)\) as a module. We conclude \(I\) is flat as \((e)\) is a direct summand of \(A\). The final statement is true because \(K = S^{-1}A\) where \(S = \prod (A_j \setminus \{0\})\).
Lemma
Let \(A\) be a normal domain with fraction field \(K\). There exists a cartesian diagram \[\xymatrix{ A \ar[d] \ar[r] & K \ar[d] \\ V \ar[r] & L }\] of rings where \(V\) has weak dimension at most \(1\) and \(V \to L\) is a flat, injective, epimorphism of rings.
Proof
For every \(x \in K\), \(x \not \in A\) pick \(V_x \subset K\) as in Algebra, Lemma 090P. Set \(V = \prod_{x \in K \setminus A} V_x\) and \(L = \prod_{x \in K \setminus A} K\). The ring \(V\) has weak dimension at most \(1\) by Lemma 092T which also shows that \(V \to L\) is a localization. A localization is flat and an epimorphism, see Algebra, Lemmas 00HT and 04VS.
Lemma
Let \(A\) be a ring of weak dimension at most \(1\). If \(A \to B\) is a flat, injective, epimorphism of rings, then \(A\) is integrally closed in \(B\).
Proof
Let \(x \in B\) be integral over \(A\). Let \(A' = A[x] \subset B\). Then \(A'\) is a finite ring extension of \(A\) by Algebra, Lemma 02JJ. To show \(A = A'\) it suffices to show \(A \to A'\) is an epimorphism by Algebra, Lemma 04VT. Note that \(A'\) is flat over \(A\) by assumption on \(A\) and the fact that \(B\) is flat over \(A\) (Lemma 092S). Hence the composition \[A' \otimes_A A' \to B \otimes_A A' \to B \otimes_A B \to B\] is injective, i.e., \(A' \otimes_A A' \cong A'\) and the lemma is proved.
Lemma
Let \(A\) be a normal domain with fraction field \(K\). Let \(A \to B\) be weakly étale. Then \(B\) is integrally closed in \(B \otimes_A K\).
Proof
Choose a diagram as in Lemma 092U. As \(A \to B\) is flat, the base change gives a cartesian diagram \[\xymatrix{ B \ar[d] \ar[r] & B \otimes_A K \ar[d] \\ B \otimes_A V \ar[r] & B \otimes_A L }\] of rings. Note that \(V \to B \otimes_A V\) is weakly étale (Lemma 092H), hence \(B \otimes_A V\) has weak dimension at most \(1\) by Lemma 092E. Note that \(B \otimes_A V \to B \otimes_A L\) is a flat, injective, epimorphism of rings as a flat base change of such (Algebra, Lemmas 00HI and 04VQ). By Lemma 092V we see that \(B \otimes_A V\) is integrally closed in \(B \otimes_A L\). It follows from the cartesian property of the diagram that \(B\) is integrally closed in \(B \otimes_A K\).
Lemma
Let \(A \to B\) be a ring homomorphism. Assume
\(A\) is a henselian local ring,
\(A \to B\) is integral,
\(B\) is a domain.
Then \(B\) is a henselian local ring and \(A \to B\) is a local homomorphism. If \(A\) is strictly henselian, then \(B\) is a strictly henselian local ring and the extension \(\kappa(\mathfrak m_B)/\kappa(\mathfrak m_A)\) of residue fields is purely inseparable.
Proof
Write \(B\) as a filtered colimit \(B = \colim B_i\) of finite \(A\)-sub algebras. If we prove the results for each \(B_i\), then the result follows for \(B\). See Algebra, Lemma 04GI. If \(A \to B\) is finite, then \(B\) is a product of local henselian rings by Algebra, Lemma 04GH. Since \(B\) is a domain we see that \(B\) is a local ring. The maximal ideal of \(B\) lies over the maximal ideal of \(A\) by going up for \(A \to B\) (Algebra, Lemma 00GU). If \(A\) is strictly henselian, then the field extension \(\kappa(\mathfrak m_B)/\kappa(\mathfrak m_A)\) being algebraic, has to be purely inseparable. Of course, then \(\kappa(\mathfrak m_B)\) is separably algebraically closed and \(B\) is strictly henselian.
Theorem
Let \(A \to B\) be a local homomorphism of local rings. If \(A\) is strictly henselian and \(A \to B\) is weakly étale, then \(A = B\).
Proof
We will show that for all \(\mathfrak p \subset A\) there is a unique prime \(\mathfrak q \subset B\) lying over \(\mathfrak p\) and \(\kappa(\mathfrak p) = \kappa(\mathfrak q)\). This implies that \(B \otimes_A B \to B\) is bijective on spectra as well as surjective and flat. Hence it is an isomorphism for example by the description of pure ideals in Algebra, Lemma 04PU. Hence \(A \to B\) is a faithfully flat epimorphism of rings. We get \(A = B\) by Algebra, Lemma 04VU.
Note that the fibre ring \(B \otimes_A \kappa(\mathfrak p)\) is a colimit of étale extensions of \(\kappa(\mathfrak p)\) by Lemmas 092H and 092Q. Hence, if there exists more than one prime lying over \(\mathfrak p\) or if \(\kappa(\mathfrak p) \not = \kappa(\mathfrak q)\) for some \(\mathfrak q\), then \(B \otimes_A L\) has a nontrivial idempotent for some (separable) algebraic field extension \(L/\kappa(\mathfrak p)\).
Let \(L/\kappa(\mathfrak p)\) be an algebraic field extension. Let \(A' \subset L\) be the integral closure of \(A/\mathfrak p\) in \(L\). By Lemma 092X we see that \(A'\) is a strictly henselian local ring whose residue field is a purely inseparable extension of the residue field of \(A\). Thus \(B \otimes_A A'\) is a local ring by Algebra, Lemma 092Y. On the other hand, \(B \otimes_A A'\) is integrally closed in \(B \otimes_A L\) by Lemma 092W. Since \(B \otimes_A A'\) is local, it follows that the ring \(B \otimes_A L\) does not have nontrivial idempotents which is what we wanted to prove.
Weakly étale algebras over fields
If \(K\) is a field, then an algebra \(B\) is weakly étale over \(K\) if and only if it is a filtered colimit of étale \(K\)-algebras. This is Lemma 092Q.
Lemma
Let \(K\) be a field. If \(B\) is weakly étale over \(K\), then
\(B\) is reduced,
\(B\) is integral over \(K\),
any finitely generated \(K\)-subalgebra of \(B\) is a finite product of finite separable extensions of \(K\),
\(B\) is a field if and only if \(B\) does not have nontrivial idempotents and in this case it is a separable algebraic extension of \(K\),
any sub or quotient \(K\)-algebra of \(B\) is weakly étale over \(K\),
if \(B'\) is weakly étale over \(K\), then \(B \otimes_K B'\) is weakly étale over \(K\).
Proof
Part (1) follows from Lemma 092I but of course it follows from part (3) as well. Part (3) follows from Lemma 092Q and the fact that étale \(K\)-algebras are finite products of finite separable extensions of \(K\), see Algebra, Lemma 00U3. Part (3) implies (2). Part (4) follows from (3) as a product of fields is a field if and only if it has no nontrivial idempotents.
If \(S \subset B\) is a subalgebra, then it is the filtered colimit of its finitely generated subalgebras which are all étale over \(K\) by the above and hence \(S\) is weakly étale over \(K\) by Lemma 092Q. If \(B \to Q\) is a quotient algebra, then \(Q\) is the filtered colimit of \(K\)-algebra quotients of finite products \(\prod_{i \in I} L_i\) of finite separable extensions \(L_i/K\). Such a quotient is of the form \(\prod_{i \in J} L_i\) for some subset \(J \subset I\) and hence the result holds for quotients by the same reasoning.
The statement on tensor products follows in a similar manner or by combining Lemmas 092H and 092J.
Lemma
Let \(K\) be a field. Let \(A\) be a \(K\)-algebra. There exists a maximal weakly étale \(K\)-subalgebra \(B_{max} \subset A\).
Proof
Let \(B_1, B_2 \subset A\) be weakly étale \(K\)-subalgebras. Then \(B_1 \otimes_K B_2\) is weakly étale over \(K\) and so is the image of \(B_1 \otimes_K B_2 \to A\) (Lemma 0CKR). Thus the collection \(\mathcal{B}\) of weakly étale \(K\)-subalgebras \(B \subset A\) is directed and the colimit \(B_{max} = \colim_{B \in \mathcal{B}} B\) is a weakly étale \(K\)-algebra by Lemma 092N. Hence the image of \(B_{max} \to A\) is weakly étale over \(K\) (previous lemma cited). It follows that this image is in \(\mathcal{B}\) and hence \(\mathcal{B}\) has a maximal element (and the image is the same as \(B_{max}\)).
Lemma
Let \(K\) be a field. For a \(K\)-algebra \(A\) denote \(B_{max}(A)\) the maximal weakly étale \(K\)-subalgebra of \(A\) as in Lemma 0CKS. Then
any \(K\)-algebra map \(A' \to A\) induces a \(K\)-algebra map \(B_{max}(A') \to B_{max}(A)\),
if \(A' \subset A\), then \(B_{max}(A') = B_{max}(A) \cap A'\),
if \(A = \colim A_i\) is a filtered colimit, then \(B_{max}(A) = \colim B_{max}(A_i)\),
the map \(B_{max}(A) \to B_{max}(A_{red})\) is an isomorphism,
\(B_{max}(A_1 \times \ldots \times A_n) = B_{max}(A_1) \times \ldots \times B_{max}(A_n)\),
if \(A\) has no nontrivial idempotents, then \(B_{max}(A)\) is a field and a separable algebraic extension of \(K\),
add more here.
Proof
Proof of (1). This is true because the image of \(B_{max}(A') \to A\) is weakly étale over \(K\) by Lemma 0CKR.
Proof of (2). By (1) we have \(B_{max}(A') \subset B_{max}(A)\). Conversely, \(B_{max}(A) \cap A'\) is a weakly étale \(K\)-algebra by Lemma 0CKR and hence contained in \(B_{max}(A')\).
Proof of (3). By (1) there is a map \(\colim B_{max}(A_i) \to A\) which is injective because the system is filtered and \(B_{max}(A_i) \subset A_i\). The colimit \(\colim B_{max}(A_i)\) is weakly étale over \(K\) by Lemma 092N. Hence we get an injective map \(\colim B_{max}(A_i) \to B_{max}(A)\). Suppose that \(a \in B_{max}(A)\). Then \(a\) generates a finitely presented \(K\)-subalgebra \(B \subset B_{max}(A)\). By Algebra, Lemma 00QO there is an \(i\) and a \(K\)-algebra map \(f : B \to A_i\) lifting the given map \(B \to A\). Since \(B\) is weakly étale by Lemma 0CKR, we see that \(f(B) \subset B_{max}(A_i)\) and we conclude that \(a\) is in the image of \(\colim B_{max}(A_i) \to B_{max}(A)\).
Proof of (4). Write \(B_{max}(A_{red}) = \colim B_i\) as a filtered colimit of étale \(K\)-algebras (Lemma 092Q). By Algebra, Lemma 07K4 for each \(i\) there is a \(K\)-algebra map \(f_i : B_i \to A\) lifting the given map \(B_i \to A_{red}\). It follows that the canonical map \(B_{max}(A_{red}) \to B_{max}(A)\) is surjective. The kernel consists of nilpotent elements and hence is zero as \(B_{max}(A_{red})\) is reduced (Lemma 0CKR).
Proof of (5). Omitted.
Proof of (6). Follows from Lemma 0CKR part (4).
Lemma
Let \(L/K\) be an extension of fields. Let \(A\) be a \(K\)-algebra. Let \(B \subset A\) be the maximal weakly étale \(K\)-subalgebra of \(A\) as in Lemma 0CKS. Then \(B \otimes_K L\) is the maximal weakly étale \(L\)-subalgebra of \(A \otimes_K L\).
Proof
For an algebra \(A\) over \(K\) we write \(B_{max}(A/K)\) for the maximal weakly étale \(K\)-subalgebra of \(A\). Similarly we write \(B_{max}(A'/L)\) for the maximal weakly étale \(L\)-subalgebra of \(A'\) if \(A'\) is an \(L\)-algebra. Since \(B_{max}(A/K) \otimes_K L\) is weakly étale over \(L\) (Lemma 092H) and since \(B_{max}(A/K) \otimes_K L \subset A \otimes_K L\) we obtain a canonical injective map \[B_{max}(A/K) \otimes_K L \to B_{max}((A \otimes_K L)/L)\] The lemma states that this map is an isomorphism.
To prove the lemma for \(L\) and our \(K\)-algebra \(A\), it suffices to prove the lemma for any field extension \(L'\) of \(L\). Namely, we have the factorization \[B_{max}(A/K) \otimes_K L' \to B_{max}((A \otimes_K L)/L) \otimes_L L' \to B_{max}((A \otimes_K L')/L')\] hence the composition cannot be surjective without \(B_{max}(A/K) \otimes_K L \to B_{max}((A \otimes_K L)/L)\) being surjective. Thus we may assume \(L\) is algebraically closed.
Reduction to finite type \(K\)-algebra. We may write \(A\) is the filtered colimit of its finite type \(K\)-subalgebras. Using Lemma 0CKT we see that it suffices to prove the lemma for finite type \(K\)-algebras.
Assume \(A\) is a finite type \(K\)-algebra. Since the kernel of \(A \to A_{red}\) is nilpotent, the same is true for \(A \otimes_K L \to A_{red} \otimes_K L\). Then \[B_{max}((A \otimes_K L)/L) \to B_{max}((A_{red} \otimes_K L)/L)\] is injective because the kernel is nilpotent and the weakly étale \(L\)-algebra \(B_{max}((A \otimes_K L)/L)\) is reduced (Lemma 0CKR). Since \(B_{max}(A/K) = B_{max}(A_{red}/K)\) by Lemma 0CKT we conclude that it suffices to prove the lemma for \(A_{red}\).
Assume \(A\) is a reduced finite type \(K\)-algebra. Let \(Q = Q(A)\) be the total quotient ring of \(A\). Then \(A \subset Q\) and \(A \otimes_K L \subset Q \otimes_K L\) and hence \[B_{max}(A/K) = A \cap B_{max}(Q/K)\] and \[B_{max}((A \otimes_K L)/L) = (A \otimes_K L) \cap B_{max}((Q \otimes_K L)/L)\] by Lemma 0CKT. Since \(-\otimes_K L\) is an exact functor, it follows that if we prove the result for \(Q\), then the result follows for \(A\). Since \(Q\) is a finite product of fields (Algebra, Lemmas 02LX, 00EU, 00FR, and 00FN) and since \(B_{max}\) commutes with products (Lemma 0CKT) it suffices to prove the lemma when \(A\) is a field.
Assume \(A\) is a field. We reduce to \(A\) being finitely generated over \(K\) by the argument in the third paragraph of the proof. (In fact the way we reduced to the case of a field produces a finitely generated field extension of \(K\).)
Assume \(A\) is a finitely generated field extension of \(K\). Then \(K' = B_{max}(A/K)\) is a field separable algebraic over \(K\) by Lemma 0CKT part (6). Hence \(K'\) is a finite separable field extension of \(K\) and \(A\) is geometrically irreducible over \(K'\) by Algebra, Lemma 037Q. Since \(L\) is algebraically closed and \(K'/K\) finite separable we see that \[K' \otimes_K L \to \prod\nolimits_{\sigma \in \Hom_K(K', L)} L,\quad \alpha \otimes \beta \mapsto (\sigma(\alpha)\beta)_\sigma\] is an isomorphism (Fields, Lemma 0CKN). We conclude \[A \otimes_K L = A \otimes_{K'} (K' \otimes_K L) = \prod\nolimits_{\sigma \in \Hom_K(K', L)} A \otimes_{K', \sigma} L\] Since \(A\) is geometrically irreducible over \(K'\) we see that \(A \otimes_{K', \sigma} L\) has a unique minimal prime. Since \(L\) is algebraically closed it follows that \(B_{max}((A \otimes_{K', \sigma} L)/L) = L\) because this \(L\)-algebra is a field algebraic over \(L\) by Lemma 0CKT part (6). It follows that the maximal weakly étale \(K' \otimes_K L\)-subalgebra of \(A \otimes_K L\) is \(K' \otimes_K L\) because we can decompose these subalgebras into products as above. Hence the inclusion \(K' \otimes_K L \subset B_{max}((A \otimes_K L)/L)\) is an equality: the ring map \(K' \otimes_K L \to B_{max}((A \otimes_K L)/L)\) is weakly étale by Lemma 092L.
Local irreducibility
The following definition seems to be the generally accepted one. To parse it, observe that if \(A \subset B\) is an integral extension of local domains, then \(A \to B\) is a local ring homomorphism by going up (Algebra, Lemma 00GU).
Definition
Let \(A\) be a local ring. We say \(A\) is unibranch if the reduction \(A_{red}\) is a domain and if the integral closure \(A'\) of \(A_{red}\) in its field of fractions is local. We say \(A\) is geometrically unibranch if \(A\) is unibranch and moreover the residue field of \(A'\) is purely inseparable over the residue field of \(A\).
Let \(A\) be a local ring. Here is an equivalent formulation
\(A\) is unibranch if \(A\) has a unique minimal prime \(\mathfrak p\) and the integral closure of \(A/\mathfrak p\) in its fraction field is a local ring, and
\(A\) is geometrically unibranch if \(A\) has a unique minimal prime \(\mathfrak p\) and the integral closure of \(A/\mathfrak p\) in its fraction field is a local ring whose residue field is purely inseparable over the residue field of \(A\).
A local ring which is normal is geometrically unibranch (follows from Definition 0BPZ and Algebra, Definition 00GV). Lemmas 0BQ0 and 06DM suggest that being (geometrically) unibranch is a reasonable property to look at.
Lemma
Let \(A\) be a local ring. Assume \(A\) has finitely many minimal prime ideals. Let \(A'\) be the integral closure of \(A\) in the total ring of fractions of \(A_{red}\). Let \(A^h\) be the henselization of \(A\). Consider the maps \[\Spec(A') \leftarrow \Spec((A')^h) \rightarrow \Spec(A^h)\] where \((A')^h = A' \otimes_A A^h\). Then
the left arrow is bijective on maximal ideals,
the right arrow is bijective on minimal primes,
every minimal prime of \((A')^h\) is contained in a unique maximal ideal and every maximal ideal contains exactly one minimal prime.
Proof
Let \(I \subset A\) be the ideal of nilpotents. We have \((A/I)^h = A^h/IA^h\) by (Algebra, Lemma 05WQ). The spectra of \(A\), \(A^h\), \(A'\), and \((A')^h\) are the same as the spectra of \(A/I\), \(A^h/IA^h\), \(A'\), and \((A')^h = A' \otimes_{A/I} A^h/IA^h\). Thus we may replace \(A\) by \(A_{red} = A/I\) and assume \(A\) is reduced. Then \(A \subset A'\) which we will use below without further mention.
Proof of (1). As \(A'\) is integral over \(A\) we see that \((A')^h\) is integral over \(A^h\). By going up (Algebra, Lemma 00GU) every maximal ideal of \(A'\), resp. \((A')^h\) lies over the maximal ideal \(\mathfrak m\), resp. \(\mathfrak m^h\) of \(A\), resp. \(A^h\). Thus (1) follows from the isomorphism \[(A')^h \otimes_{A^h} \kappa^h = A' \otimes_A A^h \otimes_{A^h} \kappa^h = A' \otimes_A \kappa\] because the residue field extension \(\kappa^h/\kappa\) induced by \(A \to A^h\) is trivial. We will use below that the displayed ring is integral over a field hence spectrum of this ring is a profinite space, see Algebra, Lemmas 00GS and 04MG.
Proof of (3). The ring \(A'\) is a normal ring and in fact a finite product of normal domains, see Algebra, Lemma 030C. Since \(A^h\) is a filtered colimit of étale \(A\)-algebras, \((A')^h\) is filtered colimit of étale \(A'\)-algebras hence \((A')^h\) is a normal ring by Algebra, Lemmas 033C and 037D. Thus every local ring of \((A')^h\) is a normal domain and we see that every maximal ideal contains a unique minimal prime. By Lemma 09XK applied to \(A^h \to (A')^h\) we see that \(((A')^h, \mathfrak m(A')^h)\) is a henselian pair. If \(\mathfrak q \subset (A')^h\) is a minimal prime (or any prime), then the intersection of \(V(\mathfrak q)\) with \(V(\mathfrak m (A')^h)\) is connected by Lemma 09Y6 Since \(V(\mathfrak m (A')^h) = \Spec((A')^h \otimes \kappa^h)\) is a profinite space by we see there is a unique maximal ideal containing \(\mathfrak q\).
Proof of (2). The minimal primes of \(A'\) are exactly the primes lying over a minimal prime of \(A\) (by construction). Since \(A' \to (A')^h\) is flat by going down (Algebra, Lemma 00HS) every minimal prime of \((A')^h\) lies over a minimal prime of \(A'\). Conversely, any prime of \((A')^h\) lying over a minimal prime of \(A'\) is minimal because \((A')^h\) is a filtered colimit of étale hence quasi-finite algebras over \(A'\) (small detail omitted). We conclude that the minimal primes of \((A')^h\) are exactly the primes which lie over a minimal prime of \(A\). Similarly, the minimal primes of \(A^h\) are exactly the primes lying over minimal primes of \(A\). By construction we have \(A' \otimes_A Q(A) = Q(A)\) where \(Q(A)\) is the total fraction ring of our reduced local ring \(A\). Of course \(Q(A)\) is the finite product of residue fields of the minimal primes of \(A\). It follows that \[(A')^h \otimes_A Q(A) = A^h \otimes_A A' \otimes_A Q(A) = A^h \otimes_A Q(A)\] Our discussion above shows the spectrum of the ring on the left is the set of minimal primes of \((A')^h\) and the spectrum of the ring on the right is the is the set of minimal primes of \(A^h\). This finishes the proof.
Lemma
Let \(A\) be a local ring. Let \(A^h\) be the henselization of \(A\). The following are equivalent
\(A\) is unibranch, and
\(A^h\) has a unique minimal prime.
Proof
This follows from Lemma 0C24 but we will also give a direct proof. Denote \(\mathfrak m\) the maximal ideal of the ring \(A\). Recall that the residue field \(\kappa = A/\mathfrak m\) is the same as the residue field of \(A^h\).
Assume (2). Let \(\mathfrak p^h\) be the unique minimal prime of \(A^h\). The flatness of \(A \to A^h\) implies that \(\mathfrak p = A \cap \mathfrak p^h\) is the unique minimal prime of \(A\) (by going down, see Algebra, Lemma 00HS). Also, since \(A^h/\mathfrak pA^h = (A/\mathfrak p)^h\) (see Algebra, Lemma 05WQ) is reduced by Lemma 06DH we see that \(\mathfrak p^h = \mathfrak pA^h\). Let \(A'\) be the integral closure of \(A/\mathfrak p\) in its fraction field. We have to show that \(A'\) is local. Since \(A \to A'\) is integral, every maximal ideal of \(A'\) lies over \(\mathfrak m\) (by going up for integral ring maps, see Algebra, Lemma 00GU). If \(A'\) is not local, then we can find distinct maximal ideals \(\mathfrak m_1\), \(\mathfrak m_2\). Choose elements \(f_1, f_2 \in A'\) with \(f_i \in \mathfrak m_i\) and \(f_i \not \in \mathfrak m_{3 - i}\). We find a finite subalgebra \(B = A/\mathfrak p[f_1, f_2] \subset A'\) with distinct maximal ideals \(B \cap \mathfrak m_i\), \(i = 1, 2\). Note that the inclusions \[A/\mathfrak p \subset B \subset \kappa(\mathfrak p)\] give, on tensoring with the flat ring map \(A \to A^h\) the inclusions \[A^h/\mathfrak p^h \subset B \otimes_A A^h \subset \kappa(\mathfrak p) \otimes_A A^h \subset \kappa(\mathfrak p^h)\] the last inclusion because \(\kappa(\mathfrak p) \otimes_A A^h = \kappa(\mathfrak p) \otimes_{A/\mathfrak p} A^h/\mathfrak p^h\) is a localization of the domain \(A^h/\mathfrak p^h\). Note that \(B \otimes_A \kappa\) has at least two maximal ideals because \(B/\mathfrak mB\) has two maximal ideals. Hence, as \(A^h\) is henselian we see that \(B \otimes_A A^h\) is a product of \(\geq 2\) local rings, see Algebra, Lemma 04GJ. But we’ve just seen that \(B \otimes_A A^h\) is a subring of a domain and we get a contradiction.
Assume (1). Let \(\mathfrak p \subset A\) be the unique minimal prime and let \(A'\) be the integral closure of \(A/\mathfrak p\) in its fraction field. Let \(A \to B\) be a local map of local rings inducing an isomorphism of residue fields which is a localization of an étale \(A\)-algebra. In particular \(\mathfrak m_B\) is the unique prime containing \(\mathfrak m B\). Then \(B' = A' \otimes_A B\) is integral over \(B\) and the assumption that \(A \to A'\) is local implies that \(B'\) is local (Algebra, Lemma 092Y). On the other hand, \(A' \to B'\) is the localization of an étale ring map, hence \(B'\) is normal, see Algebra, Lemma 033C. Thus \(B'\) is a (local) normal domain. Finally, we have \[B/\mathfrak pB \subset B \otimes_A \kappa(\mathfrak p) = B' \otimes_{A'} (\text{fraction field of }A') \subset \text{fraction field of }B'\] Hence \(B/\mathfrak pB\) is a domain, which implies that \(B\) has a unique minimal prime (since by flatness of \(A \to B\) these all have to lie over \(\mathfrak p\)). Since \(A^h\) is a filtered colimit of the local rings \(B\) it follows that \(A^h\) has a unique minimal prime. Namely, if \(fg = 0\) in \(A^h\) for some non-nilpotent elements \(f, g\), then we can find a \(B\) as above containing both \(f\) and \(g\) which leads to a contradiction.
Lemma
Let \((A, \mathfrak m, \kappa)\) be a local ring. Assume \(A\) has finitely many minimal prime ideals. Let \(A'\) be the integral closure of \(A\) in the total ring of fractions of \(A_{red}\). Choose an algebraic closure \(\overline{\kappa}\) of \(\kappa\) and denote \(\kappa^{sep} \subset \overline{\kappa}\) the separable algebraic closure of \(\kappa\). Let \(A^{sh}\) be the strict henselization of \(A\) with respect to \(\kappa^{sep}\). Consider the maps \[\Spec(A') \xleftarrow{c} \Spec((A')^{sh}) \xrightarrow{e} \Spec(A^{sh})\] where \((A')^{sh} = A' \otimes_A A^{sh}\). Then
for \(\mathfrak m' \subset A'\) maximal the residue field \(\kappa'\) is algebraic over \(\kappa\) and the fibre of \(c\) over \(\mathfrak m'\) can be canonically identified with \(\Hom_\kappa(\kappa', \overline{\kappa})\),
the right arrow is bijective on minimal primes,
every minimal prime of \((A')^{sh}\) is contained in a unique maximal ideal and every maximal ideal contains a unique minimal prime.
Proof
The proof is almost exactly the same as for Lemma 0C24. Let \(I \subset A\) be the ideal of nilpotents. We have \((A/I)^{sh} = A^{sh}/IA^{sh}\) by (Algebra, Lemma 05WQ). The spectra of \(A\), \(A^{sh}\), \(A'\), and \((A')^{sh}\) are the same as the spectra of \(A/I\), \(A^{sh}/IA^{sh}\), \(A'\), and \((A')^{sh} = A' \otimes_{A/I} A^{sh}/IA^{sh}\). Thus we may replace \(A\) by \(A_{red} = A/I\) and assume \(A\) is reduced. Then \(A \subset A'\) which we will use below without further mention.
Proof of (1). The field extension \(\kappa'/\kappa\) is algebraic because \(A'\) is integral over \(A\). Since \(A'\) is integral over \(A\), we see that \((A')^{sh}\) is integral over \(A^{sh}\). By going up (Algebra, Lemma 00GU) every maximal ideal of \(A'\), resp. \((A')^{sh}\) lies over the maximal ideal \(\mathfrak m\), resp. \(\mathfrak m^{sh}\) of \(A\), resp. \(A^{sh}\). We have \[(A')^{sh} \otimes_{A^{sh}} \kappa^{sep} = A' \otimes_A A^{sh} \otimes_{A^{sh}} \kappa^{sep} = (A' \otimes_A \kappa) \otimes_{\kappa} \kappa^{sep}\] because the residue field of \(A^{sh}\) is \(\kappa^{sep}\). Thus the fibre of \(c\) over \(\mathfrak m'\) is the spectrum of \(\kappa' \otimes_\kappa \kappa^{sep}\). We conclude (1) is true because there is a bijection \[\Hom_\kappa(\kappa', \overline{\kappa}) \to \Spec(\kappa' \otimes_\kappa \kappa^{sep}),\quad \sigma \mapsto \Ker( \sigma \otimes 1 : \kappa' \otimes_\kappa \kappa^{sep} \to \overline{\kappa} )\] We will use below that the displayed ring is integral over a field hence spectrum of this ring is a profinite space, see Algebra, Lemmas 00GS and 04MG.
Proof of (3). The ring \(A'\) is a normal ring and in fact a finite product of normal domains, see Algebra, Lemma 030C. Since \(A^{sh}\) is a filtered colimit of étale \(A\)-algebras, \((A')^{sh}\) is filtered colimit of étale \(A'\)-algebras hence \((A')^{sh}\) is a normal ring by Algebra, Lemmas 033C and 037D. Thus every local ring of \((A')^{sh}\) is a normal domain and we see that every maximal ideal contains a unique minimal prime. By Lemma 09XK applied to \(A^{sh} \to (A')^{sh}\) to see that \(((A')^{sh}, \mathfrak m(A')^{sh})\) is a henselian pair. If \(\mathfrak q \subset (A')^{sh}\) is a minimal prime (or any prime), then the intersection of \(V(\mathfrak q)\) with \(V(\mathfrak m (A')^{sh})\) is connected by Lemma 09Y6 Since \(V(\mathfrak m (A')^{sh}) = \Spec((A')^{sh} \otimes \kappa^{sep})\) is a profinite space by we see there is a unique maximal ideal containing \(\mathfrak q\).
Proof of (2). The minimal primes of \(A'\) are exactly the primes lying over a minimal prime of \(A\) (by construction). Since \(A' \to (A')^{sh}\) is flat by going down (Algebra, Lemma 00HS) every minimal prime of \((A')^{sh}\) lies over a minimal prime of \(A'\). Conversely, any prime of \((A')^{sh}\) lying over a minimal prime of \(A'\) is minimal because \((A')^{sh}\) is a filtered colimit of étale hence quasi-finite algebras over \(A'\) (small detail omitted). We conclude that the minimal primes of \((A')^{sh}\) are exactly the primes which lie over a minimal prime of \(A\). Similarly, the minimal primes of \(A^{sh}\) are exactly the primes lying over minimal primes of \(A\). By construction we have \(A' \otimes_A Q(A) = Q(A)\) where \(Q(A)\) is the total fraction ring of our reduced local ring \(A\). Of course \(Q(A)\) is the finite product of residue fields of the minimal primes of \(A\). It follows that \[(A')^{sh} \otimes_A Q(A) = A^{sh} \otimes_A A' \otimes_A Q(A) = A^{sh} \otimes_A Q(A)\] Our discussion above shows the spectrum of the ring on the left is the set of minimal primes of \((A')^{sh}\) and the spectrum of the ring on the right is the is the set of minimal primes of \(A^{sh}\). This finishes the proof.
Lemma
Let \(A\) be a local ring. Let \(A^{sh}\) be a strict henselization of \(A\). The following are equivalent
\(A\) is geometrically unibranch, and
\(A^{sh}\) has a unique minimal prime.
Proof
This follows from Lemma 0C25 but we will also give a direct proof; this direct proof is almost exactly the same as the direct proof of Lemma 0BQ0. Denote \(\mathfrak m\) the maximal ideal of the ring \(A\). Denote \(\kappa\), \(\kappa^{sh}\) the residue field of \(A\), \(A^{sh}\).
Assume (2). Let \(\mathfrak p^{sh}\) be the unique minimal prime of \(A^{sh}\). The flatness of \(A \to A^{sh}\) implies that \(\mathfrak p = A \cap \mathfrak p^{sh}\) is the unique minimal prime of \(A\) (by going down, see Algebra, Lemma 00HS). Also, since \(A^{sh}/\mathfrak pA^{sh} = (A/\mathfrak p)^{sh}\) (see Algebra, Lemma 05WS) is reduced by Lemma 06DH we see that \(\mathfrak p^{sh} = \mathfrak pA^{sh}\). Let \(A'\) be the integral closure of \(A/\mathfrak p\) in its fraction field. We have to show that \(A'\) is local and that its residue field is purely inseparable over \(\kappa\). Since \(A \to A'\) is integral, every maximal ideal of \(A'\) lies over \(\mathfrak m\) (by going up for integral ring maps, see Algebra, Lemma 00GU). If \(A'\) is not local, then we can find distinct maximal ideals \(\mathfrak m_1\), \(\mathfrak m_2\). Choosing elements \(f_1, f_2 \in A'\) with \(f_i \in \mathfrak m_i, f_i \not \in \mathfrak m_{3 - i}\) we find a finite subalgebra \(B = A[f_1, f_2] \subset A'\) with distinct maximal ideals \(B \cap \mathfrak m_i\), \(i = 1, 2\). If \(A'\) is local with maximal ideal \(\mathfrak m'\), but \(A/\mathfrak m \subset A'/\mathfrak m'\) is not purely inseparable, then we can find \(f \in A'\) whose image in \(A'/\mathfrak m'\) generates a finite, not purely inseparable extension of \(A/\mathfrak m\) and we find a finite local subalgebra \(B = A[f] \subset A'\) whose residue field is not a purely inseparable extension of \(A/\mathfrak m\). Note that the inclusions \[A/\mathfrak p \subset B \subset \kappa(\mathfrak p)\] give, on tensoring with the flat ring map \(A \to A^{sh}\) the inclusions \[A^{sh}/\mathfrak p^{sh} \subset B \otimes_A A^{sh} \subset \kappa(\mathfrak p) \otimes_A A^{sh} \subset \kappa(\mathfrak p^{sh})\] the last inclusion because \(\kappa(\mathfrak p) \otimes_A A^{sh} = \kappa(\mathfrak p) \otimes_{A/\mathfrak p} A^{sh}/\mathfrak p^{sh}\) is a localization of the domain \(A^{sh}/\mathfrak p^{sh}\). Note that \(B \otimes_A \kappa^{sh}\) has at least two maximal ideals because \(B/\mathfrak mB\) either has two maximal ideals or one whose residue field is not purely inseparable over \(\kappa\), and because \(\kappa^{sh}\) is separably algebraically closed. Hence, as \(A^{sh}\) is strictly henselian we see that \(B \otimes_A A^{sh}\) is a product of \(\geq 2\) local rings, see Algebra, Lemma 06DD. But we’ve just seen that \(B \otimes_A A^{sh}\) is a subring of a domain and we get a contradiction.
Assume (1). Let \(\mathfrak p \subset A\) be the unique minimal prime and let \(A'\) be the integral closure of \(A/\mathfrak p\) in its fraction field. Let \(A \to B\) be a local map of local rings which is a localization of an étale \(A\)-algebra. In particular \(\mathfrak m_B\) is the unique prime containing \(\mathfrak m_AB\). Then \(B' = A' \otimes_A B\) is integral over \(B\) and the assumption that \(A \to A'\) is local with purely inseparable residue field extension implies that \(B'\) is local (Algebra, Lemma 092Y). On the other hand, \(A' \to B'\) is the localization of an étale ring map, hence \(B'\) is normal, see Algebra, Lemma 033C. Thus \(B'\) is a (local) normal domain. Finally, we have \[B/\mathfrak pB \subset B \otimes_A \kappa(\mathfrak p) = B' \otimes_{A'} (\text{fraction field of }A') \subset \text{fraction field of }B'\] Hence \(B/\mathfrak pB\) is a domain, which implies that \(B\) has a unique minimal prime (since by flatness of \(A \to B\) these all have to lie over \(\mathfrak p\)). Since \(A^{sh}\) is a filtered colimit of the local rings \(B\) it follows that \(A^{sh}\) has a unique minimal prime. Namely, if \(fg = 0\) in \(A^{sh}\) for some non-nilpotent elements \(f, g\), then we can find a \(B\) as above containing both \(f\) and \(g\) which leads to a contradiction.
Definition
Let \(A\) be a local ring with henselization \(A^h\) and strict henselization \(A^{sh}\). The number of branches of \(A\) is the number of minimal primes of \(A^h\) if finite and \(\infty\) otherwise. The number of geometric branches of \(A\) is the number of minimal primes of \(A^{sh}\) if finite and \(\infty\) otherwise.
We spell out the relationship with Definition 0BPZ.
Lemma
Let \((A, \mathfrak m, \kappa)\) be a local ring.
If \(A\) has infinitely many minimal prime ideals, then the number of (geometric) branches of \(A\) is \(\infty\).
The number of branches of \(A\) is \(1\) if and only if \(A\) is unibranch.
The number of geometric branches of \(A\) is \(1\) if and only if \(A\) is geometrically unibranch.
Assume \(A\) has finitely many minimal primes and let \(A'\) be the integral closure of \(A\) in the total ring of fractions of \(A_{red}\). Then
the number of branches of \(A\) is the number of maximal ideals \(\mathfrak m'\) of \(A'\),
to get the number of geometric branches of \(A\) we have to count each maximal ideal \(\mathfrak m'\) of \(A'\) with multiplicity given by the separable degree of \(\kappa(\mathfrak m')/\kappa\).
Proof
This lemma follows immediately from the definitions, Lemma 0C24, Lemma 0C25, and Fields, Lemma 09HJ.
Lemma
Let \(A \to B\) be a local homomorphism of local rings which is the localization of a smooth ring map.
The number of geometric branches of \(A\) is equal to the number of geometric branches of \(B\).
If \(A \to B\) induces a purely inseparable extension of residue fields, then the number of branches of \(A\) is the number of branches of \(B\).
Proof
We will use that smooth ring maps are flat (Algebra, Lemma 00TA), that localizations are flat (Algebra, Lemma 00HT), that compositions of flat ring maps are flat (Algebra, Lemma 00HC), that base change of a flat ring map is flat (Algebra, Lemma 00HI), that flat local homomorphisms are faithfully flat (Algebra, Lemma 00HR), that (strict) henselization is flat (Lemma 07QM), and Going down for flat ring maps (Algebra, Lemma 00HS).
Proof of (2). Let \(A^h\), \(B^h\) be the henselizations of \(A\), \(B\). Then \(B^h\) is the henselization of \(A^h \otimes_A B\) at the unique maximal ideal lying over \(\mathfrak m_B\), see Algebra, Lemma 08HU. Thus we may and do assume \(A\) is henselian. Since \(A \to B \to B^h\) is flat, every minimal prime of \(B^h\) lies over a minimal prime of \(A\) and since \(A \to B^h\) is faithfully flat, every minimal prime of \(A\) does lie under a minimal prime of \(B^h\); in both cases use going down for flat ring maps. Therefore it suffices to show that given a minimal prime \(\mathfrak p \subset A\), there is at most one minimal prime of \(B^h\) lying over \(\mathfrak p\). After replacing \(A\) by \(A/\mathfrak p\) and \(B\) by \(B/\mathfrak p B\) we may assume that \(A\) is a domain; the \(A\) is still henselian by Algebra, Lemma 05WQ. By Lemma 0BQ0 we see that the integral closure \(A'\) of \(A\) in its field of fractions is a local domain. Of course \(A'\) is a normal domain. By Algebra, Lemma 033C we see that \(A' \otimes_A B^h\) is a normal ring (the lemma just gives it for \(A' \otimes_A B\), to go up to \(A' \otimes_A B^h\) use that \(B^h\) is a colimit of étale \(B\)-algebras and use Algebra, Lemma 037D). By Algebra, Lemma 092Y we see that \(A' \otimes_A B^h\) is local (this is where we use the assumption on the residue fields of \(A\) and \(B\)). Hence \(A' \otimes_A B^h\) is a local normal ring, hence a local domain. Since \(B^h \subset A' \otimes_A B^h\) by flatness of \(A \to B^h\) we conclude that \(B^h\) is a domain as desired.
Proof of (1). Let \(A^{sh}\), \(B^{sh}\) be strict henselizations of \(A\), \(B\). Then \(B^{sh}\) is a strict henselization of \(A^{sh} \otimes_A B\) at a maximal ideal lying over \(\mathfrak m_B\) and \(\mathfrak m_{A^{sh}}\), see Algebra, Lemma 08HV. Thus we may and do assume \(A\) is strictly henselian. Since \(A \to B \to B^{sh}\) is flat, every minimal prime of \(B^{sh}\) lies over a minimal prime of \(A\) and since \(A \to B^{sh}\) is faithfully flat, every minimal prime of \(A\) does lie under a minimal prime of \(B^{sh}\); in both cases use going down for flat ring maps. Therefore it suffices to show that given a minimal prime \(\mathfrak p \subset A\), there is at most one minimal prime of \(B^{sh}\) lying over \(\mathfrak p\). After replacing \(A\) by \(A/\mathfrak p\) and \(B\) by \(B/\mathfrak p B\) we may assume that \(A\) is a domain; then \(A\) is still strictly henselian by Algebra, Lemma 05WS. By Lemma 06DM we see that the integral closure \(A'\) of \(A\) in its field of fractions is a local domain whose residue field is a purely inseparable extension of the residue field of \(A\). Of course \(A'\) is a normal domain. By Algebra, Lemma 033C we see that \(A' \otimes_A B^{sh}\) is a normal ring (the lemma just gives it for \(A' \otimes_A B\), to go up to \(A' \otimes_A B^{sh}\) use that \(B^{sh}\) is a colimit of étale \(B\)-algebras and use Algebra, Lemma 037D). By Algebra, Lemma 092Y we see that \(A' \otimes_A B^{sh}\) is local (since \(A \subset A'\) induces a purely inseparable residue field extension). Hence \(A' \otimes_A B^{sh}\) is a local normal ring, hence a local domain. Since \(B^{sh} \subset A' \otimes_A B^{sh}\) by flatness of \(A \to B^{sh}\) we conclude that \(B^{sh}\) is a domain as desired.
Miscellaneous on branches
Some results related to branches of local rings as defined in Section 06DT.
Lemma
Let \(A\) and \(B\) be domains and let \(A \to B\) be a ring map. Assume \(A \to B\) has additionally at least one of the following properties
it is the localization of an étale ring map,
it is flat and the localization of an unramified ring map,
it is flat and the localization of a quasi-finite ring map,
it is flat and the localization of an integral ring map,
it is flat and there are no nontrivial specializations between points of fibres of \(\Spec(B) \to \Spec(A)\),
\(\Spec(B) \to \Spec(A)\) maps the generic point to the generic point and there are no nontrivial specializations between points of fibres, or
exactly one point of \(\Spec(B)\) is mapped to the generic point of \(\Spec(A)\).
Then \(A \cap J\) is nonzero for every nonzero ideal \(J\) of \(B\).
Proof
Proof in case (7). Let \(K\), resp. \(L\) be the fraction field of \(A\), resp. \(B\). By Algebra, Lemma 0CAN we see that the unique point of \(\Spec(B)\) which maps to the generic point \((0) \in \Spec(A)\) is \((0) \in \Spec(B)\). We conclude that \(B \otimes_A K\) is a ring with a unique prime ideal whose residue field is \(L\) (in fact it is equal to \(L\) but we do not need this). Choose \(b \in J\) nonzero. Then \(b\) maps to a unit of \(L\). Hence \(b\) maps to a unit of \(B \otimes_A K\) (Algebra, Lemma 0B7C). Since \(B \otimes_A K = \colim_{f \in A \setminus \{0\}} B_f\) we see that \(b\) maps to a unit of \(B_f\) for some \(f \in A\) nonzero. This means that \(b b' = f^n\) for some \(b' \in B\) and \(n \geq 1\). Thus \(f^n \in A \cap J\) as desired.
In the rest of the proof, we show that each of the other assumptions imply (7). Under assumptions (1) – (5), the ring map \(A \to B\) is flat and hence \(A \to B\) is injective (since flat local homomorphisms are faithfully flat by Algebra, Lemma 00HR). Hence the generic point of \(\Spec(B)\) maps to the generic point of \(\Spec(A)\). Now, if there are no nontrivial specializations between points of fibres of \(\Spec(B) \to \Spec(A)\), then of course this generic point of \(\Spec(B)\) has to be the unique point mapping to the generic point of \(\Spec(A)\). So (6) implies (7). Finally, to finish we show that in cases (1) – (5) there are no nontrivial specializations between the points of fibres of \(\Spec(B) \to \Spec(A)\). Namely, see Algebra, Lemma 00GT for the integral case, Algebra, Definition 00PL for the quasi-finite case, and use that unramified and étale ring maps are quasi-finite (Algebra, Lemmas 02UR and 00U5).
Lemma
Let \(A \to B\) be a ring map. Let \(\mathfrak q \subset B\) be a prime ideal lying over the prime \(\mathfrak p \subset A\). Assume
\(A\) is a domain,
\(A_\mathfrak p\) is geometrically unibranch,
\(A \to B\) is unramified at \(\mathfrak q\), and
\(A_\mathfrak p \to B_\mathfrak q\) is injective.
Then there exists a \(g \in B\), \(g \not \in \mathfrak q\) such that \(B_g\) is étale over \(A\).
Proof
By Algebra, Proposition 0395 after replacing \(B\) by a principal localization, we can find a standard étale ring map \(A \to B'\) and a surjection \(B' \to B\). Denote \(\mathfrak q' \subset B'\) the inverse image of \(\mathfrak q\). We will show that \(B' \to B\) is injective after possibly replacing \(B'\) by a principal localization.
In this paragraph we reduce to the case that \(B'\) is a domain. Since \(A\) is a domain, the ring \(B'\) is reduced, see Algebra, Lemma 07QK. Let \(K\) be the fraction field of \(A\). Then \(B' \otimes_A K\) is étale over a field, hence is a finite product of fields, see Algebra, Lemma 00U3. Since \(A \to B'\) is étale (hence flat) the minimal primes of \(B'\) are lie over \((0) \subset A\) (by going down for flat ring maps). We conclude that \(B'\) has finitely many minimal primes, say \(\mathfrak r_1, \ldots, \mathfrak r_r \subset B'\). Since \(A_\mathfrak p\) is geometrically unibranch and \(A \to B'\) étale, the ring \(B'_{\mathfrak q'}\) is a domain, see Lemmas 0DQ1 and 0C37. Hence \(\mathfrak q' \supset \mathfrak r_i\) for exactly one \(i = i_0\). Choose \(g' \in B'\), \(g' \not \in \mathfrak r_{i_0}\) but \(g' \in \mathfrak r_i\) for \(i \not = i_0\), see Algebra, Lemma 00DS. After replacing \(B'\) and \(B\) by \(B'_{g'}\) and \(B_{g'}\) we obtain that \(B'\) is a domain.
Assume \(B'\) is a domain, in particular \(B' \subset B'_{\mathfrak q'}\). If \(B' \to B\) is not injective, then \(J = \Ker(B'_{\mathfrak q'} \to B_\mathfrak q)\) is nonzero. By Lemma 0GS5 applied to \(A_\mathfrak p \to B'_{\mathfrak q'}\) we find a nonzero element \(a \in A_\mathfrak p\) mapping to zero in \(B_\mathfrak q\) contradicting assumption (4). This finishes the proof.
Lemma
Let \((A, \mathfrak m)\) be a geometrically unibranch local domain. Let \(A \to B\) be an injective local homomorphism of local rings, which is essentially of finite type. If \(\mathfrak m B\) is the maximal ideal of \(B\) and the induced extension of residue fields is separable, then \(A \to B\) is the localization of an étale ring map.
Proof
We may write \(B = C_\mathfrak q\) where \(A \to C\) is a finite type ring map and \(\mathfrak q \subset C\) is a prime ideal lying over \(\mathfrak m\). By Algebra, Lemma 02FM the ring map \(A \to C\) is unramified at \(\mathfrak q\). By Algebra, Proposition 0395 after replacing \(C\) by a principal localization, we can find a standard étale ring map \(A \to C'\) and a surjection \(C' \to C\). Denote \(\mathfrak q' \subset C'\) the inverse image of \(\mathfrak q\) and set \(B' = C'_{\mathfrak q'}\). Then \(B' \to B\) is surjective. It suffices to show that \(B' \to B\) is also injective.
Since \(A\) is a domain, the rings \(C'\) and \(B'\) are reduced, see Algebra, Lemma 07QK. Since \(A\) is geometrically unibranch, the ring \(B'\) is a domain, see by Lemmas 0DQ1 and 0C37. If \(B' \to B\) is not injective, then \(A \cap \Ker(B' \to B)\) is nonzero by Lemma 0GS5 which contradicts the assumption that \(A \to B\) is injective.
Lemma
Let \(k\) be an algebraically closed field. Let \(A\), \(B\) be strictly henselian local \(k\)-algebras with residue field equal to \(k\). Let \(C\) be the strict henselization of \(A \otimes_k B\) at the maximal ideal \(\mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B\). Then the minimal primes of \(C\) correspond \(1\)-to-\(1\) to pairs of minimal primes of \(A\) and \(B\).
Proof
First note that a minimal prime \(\mathfrak r\) of \(C\) maps to a minimal prime \(\mathfrak p\) in \(A\) and to a minimal prime \(\mathfrak q\) of \(B\) because the ring maps \(A \to C\) and \(B \to C\) are flat (by going down for flat ring map Algebra, Lemma 00HS). Hence it suffices to show that the strict henselization of \((A/\mathfrak p \otimes_k B/\mathfrak q)_{ \mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B}\) has a unique minimal prime ideal. By Algebra, Lemma 05WS the rings \(A/\mathfrak p\), \(B/\mathfrak q\) are strictly henselian. Hence we may assume that \(A\) and \(B\) are strictly henselian local domains and our goal is to show that \(C\) has a unique minimal prime. By Lemma 06DM the integral closure \(A'\) of \(A\) in its fraction field is a normal local domain with residue field \(k\). Similarly for the integral closure \(B'\) of \(B\) into its fraction field. By Algebra, Lemma 06DF we see that \(A' \otimes_k B'\) is a normal ring. Hence its localization \[R = (A' \otimes_k B')_{ \mathfrak m_{A'} \otimes_k B' + A' \otimes_k \mathfrak m_{B'}}\] is a normal local domain. Note that \(A \otimes_k B \to A' \otimes_k B'\) is integral (hence gong up holds – Algebra, Lemma 00GU) and that \(\mathfrak m_{A'} \otimes_k B' + A' \otimes_k \mathfrak m_{B'}\) is the unique maximal ideal of \(A' \otimes_k B'\) lying over \(\mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B\). Hence we see that \[R = (A' \otimes_k B')_{ \mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B}\] by Algebra, Lemma 00EA. It follows that \[(A \otimes_k B)_{ \mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B} \longrightarrow R\] is integral. We conclude that \(R\) is the integral closure of \((A \otimes_k B)_{ \mathfrak m_A \otimes_k B + A \otimes_k \mathfrak m_B}\) in its fraction field, and by Lemma 06DM once again we conclude that \(C\) has a unique minimal prime ideal.
Branches of the completion
Let \((A, \mathfrak m)\) be a Noetherian local ring. Consider the maps \(A \to A^h \to A^\wedge\). In general the map \(A^h \to A^\wedge\) need not induce a bijection on minimal primes, see Examples, Section 0AL7. In other words, the number of branches of \(A\) (as defined in Definition 0C26) may be different from the number of branches of \(A^\wedge\). However, under some conditions the number of branches is the same, for example if the dimension of \(A\) is \(1\).
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring.
The map \(A^h \to A^\wedge\) defines a surjective map from minimal primes of \(A^\wedge\) to minimal primes of \(A^h\).
The number of branches of \(A\) is at most the number of branches of \(A^\wedge\).
The number of geometric branches of \(A\) is at most the number of geometric branches of \(A^\wedge\).
Proof
By Lemma 06LJ the map \(A^h \to A^\wedge\) is flat and injective. Combining going down (Algebra, Lemma 00HS) and Algebra, Lemma 00FK we see that part (1) holds. Part (2) follows from this, Definition 0C26, and the fact that \(A^\wedge\) is henselian (Algebra, Lemma 04GM). By Lemma 06LJ we have \((A^\wedge)^{sh} = A^{sh} \otimes_{A^h} A^\wedge\). Thus we can repeat the arguments above using the flat injective map \(A^{sh} \to (A^\wedge)^{sh}\) to prove (3).
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring. The number of branches of \(A\) is the same as the number of branches of \(A^\wedge\) if and only if \(\sqrt{\mathfrak qA^\wedge}\) is prime for every minimal prime \(\mathfrak q \subset A^h\) of the henselization.
Proof
Follows from Lemma 0C28 and the fact that there are only a finite number of branches for both \(A\) and \(A^\wedge\) by Algebra, Lemma 00FR and the fact that \(A^h\) and \(A^\wedge\) are Noetherian (Lemma 06LJ).
A simple glueing lemma.
Lemma
Let \(A\) be a ring and let \(I\) be a finitely generated ideal. Let \(A \to C\) be a ring map such that for all \(f \in I\) the ring map \(A_f \to C_f\) is localization at an idempotent. Then there exists a surjection \(A \to C'\) such that \(A_f \to (C \times C')_f\) is an isomorphism for all \(f \in I\).
Proof
Choose generators \(f_1, \ldots, f_r\) of \(I\). Write \[C_{f_i} = (A_{f_i})_{e_i}\] for some idempotent \(e_i \in A_{f_i}\). Write \(e_i = a_i/f_i^n\) for some \(a_i \in A\) and \(n \geq 0\); we may use the same \(n\) for all \(i = 1, \ldots, r\). After replacing \(a_i\) by \(f_i^ma_i\) and \(n\) by \(n + m\) for a suitable \(m \gg 0\), we may assume \(a_i^2 = f_i^n a_i\) for all \(i\). Since \(e_i\) maps to \(1\) in \(C_{f_if_j} = (A_{f_if_j})_{e_j} = A_{f_if_ja_j}\) we see that \[(f_if_ja_j)^N(f_j^n a_i - f_i^na_j) = 0\] for some \(N\) (we can pick the same \(N\) for all pairs \(i, j\)). Using \(a_j^2 = f_j^na_j\) this gives \[f_i^{N + n} f_j^{N + nN} a_j = f_i^N f_j^{N + n} a_ia_j^N\] After increasing \(n\) to \(n + N + nN\) and replacing \(a_i\) by \(f_i^{N + nN}a_i\) we see that \(f_i^n a_j\) is in the ideal of \(a_i\) for all pairs \(i, j\). Let \(C' = A/(a_1, \ldots, a_r)\). Then \[C'_{f_i} = A_{f_i}/(a_i) = A_{f_i}/(e_i)\] because \(a_j\) is in the ideal generated by \(a_i\) after inverting \(f_i\). Since for an idempotent \(e\) of a ring \(B\) we have \(B = B_e \times B/(e)\) we see that the conclusion of the lemma holds for \(f\) equal to one of \(f_1, \ldots, f_r\). Using glueing of functions, in the form of Algebra, Lemma 00EO, we conclude that the result holds for all \(f \in I\). Namely, for \(f \in I\) the elements \(f_1, \ldots, f_r\) generate the unit ideal in \(A_f\) so \(A_f \to (C \times C')_f\) is an isomorphism if and only if this is the case after localizing at \(f_1, \ldots, f_r\).
Lemma 0ALR can be used to construct finite type extensions from given finite type extensions of the formal completion. We will generalize this lemma in Algebraization of Formal Spaces, Lemma 0AKG.
Lemma
Let \(A\) be a Noetherian ring and \(I\) an ideal. Let \(B\) be a finite type \(A\)-algebra. Let \(B^\wedge \to C\) be a surjective ring map with kernel \(J\) where \(B^\wedge\) is the \(I\)-adic completion. If \(J/J^2\) is annihilated by \(I^c\) for some \(c \geq 0\), then \(C\) is isomorphic to the completion of a finite type \(A\)-algebra.
Proof
Let \(f \in I\). Since \(B^\wedge\) is Noetherian (Algebra, Lemma 0316), we see that \(J\) is a finitely generated ideal. Hence we conclude from Algebra, Lemma 00EH that \[C_f = ((B^\wedge)_f)_e\] for some idempotent \(e \in (B^\wedge)_f\). By Lemma 0C2A we can find a surjection \(B^\wedge \to C'\) such that \(B^\wedge \to C \times C'\) becomes an isomorphism after inverting any \(f \in I\). Observe that \(C \times C'\) is a finite \(B^\wedge\)-algebra.
Choose generators \(f_1, \ldots, f_r \in I\). Denote \(\alpha_i : (C \times C')_{f_i} \to B_{f_i} \otimes_B B^\wedge\) the inverse of the isomorphism of \((B^\wedge)_{f_i}\)-algebras we obtained above. Denote \(\alpha_{ij} : (B_{f_i})_{f_j} \to (B_{f_j})_{f_i}\) the obvious \(B\)-algebra isomorphism. Consider the object \[(C \times C', B_{f_i}, \alpha_i, \alpha_{ij})\] of the category \(\text{Glue}(B \to B^\wedge, f_1, \ldots, f_r)\) introduced in Remark 05EL. We omit the verification of conditions (1)(a) and (1)(b). Since \(B \to B^\wedge\) is a flat map (Algebra, Lemma 00MB) inducing an isomorphism \(B/IB \to B^\wedge/IB^\wedge\) we may apply Proposition 05ER and Remark 05EU. We conclude that \(C \times C'\) is isomorphic to \(D \otimes_B B^\wedge\) for some finite \(B\)-algebra \(D\). Then \(D/ID \cong C/IC \times C'/IC'\). Let \(\overline{e} \in D/ID\) be the idempotent corresponding to the factor \(C/IC\). By Lemma 07M4 there exists an étale ring map \(B \to B'\) which induces an isomorphism \(B/IB \to B'/IB'\) such that \(D' = D \otimes_B B'\) contains an idempotent \(e\) lifting \(\overline{e}\). Since \(C \times C'\) is \(I\)-adically complete the pair \((C \times C', IC \times IC')\) is henselian (Lemma 0ALJ). Thus we can factor the map \(B \to C \times C'\) through \(B'\) (see Lemma 09XI). Doing so we may replace \(B\) by \(B'\) and \(D\) by \(D'\) (because the \(I\)-adic completetions are unaltered by this). Then we find that \(D = D_e \times D_{1 - e} = D/(1 - e) \times D/(e)\) is a product of finite type \(A\)-algebras and the completion of the first part is \(C\) and the completion of the second part is \(C'\).
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring with henselization \(A^h\). Let \(\mathfrak q \subset A^\wedge\) be a minimal prime with \(\dim(A^\wedge/\mathfrak q) = 1\). Then there exists a minimal prime \(\mathfrak q^h\) of \(A^h\) such that \(\mathfrak q = \sqrt{\mathfrak q^hA^\wedge}\).
Proof
Since the completion of \(A\) and \(A^h\) are the same, we may assume that \(A\) is henselian (Lemma 06LJ). We will apply Lemma 0ALR to \(A^\wedge \to A^\wedge/J\) where \(J = \Ker(A^\wedge \to (A^\wedge)_{\mathfrak q})\). Since \(\dim((A^\wedge)_\mathfrak q) = 0\) we see that \(\mathfrak q^n \subset J\) for some \(n\). Hence \(J/J^2\) is annihilated by \(\mathfrak q^n\). On the other hand \((J/J^2)_\mathfrak q = 0\) because \(J_\mathfrak q = 0\). Hence \(\mathfrak m\) is the only associated prime of \(J/J^2\) and we find that a power of \(\mathfrak m\) annihilates \(J/J^2\). Thus the lemma applies and we find that \(A^\wedge/J = C^\wedge\) for some finite type \(A\)-algebra \(C\).
Then \(C/\mathfrak m C = A/\mathfrak m\) because \(A^\wedge/J\) has the same property. Hence \(\mathfrak m_C = \mathfrak m C\) is a maximal ideal and \(A \to C\) is unramified at \(\mathfrak m_C\) (Algebra, Lemma 02FM). After replacing \(C\) by a principal localization we may assume that \(C\) is a quotient of an étale \(A\)-algebra \(B\), see Algebra, Proposition 0395. However, since the residue field extension of \(A \to C_{\mathfrak m_C}\) is trivial and \(A\) is henselian, we conclude that \(B = A\) again after a localization. Thus \(C = A/I\) for some ideal \(I \subset A\) and it follows that \(J = IA^\wedge\) (because completion is exact in our situation by Algebra, Lemma 00MB) and \(I = J \cap A\) (by flatness of \(A \to A^\wedge\)). Since \(\mathfrak q^n \subset J \subset \mathfrak q\) we see that \(\mathfrak p = \mathfrak q \cap A\) satisfies \(\mathfrak p^n \subset I \subset \mathfrak p\). Then \(\sqrt{\mathfrak p A^\wedge} = \mathfrak q\) and the proof is complete.
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring. The punctured spectrum of \(A^\wedge\) is disconnected if and only if the punctured spectrum of \(A^h\) is disconnected.
Proof
Since the completion of \(A\) and \(A^h\) are the same, we may assume that \(A\) is henselian (Lemma 06LJ).
Since \(A \to A^\wedge\) is faithfully flat (see reference just given) the map from the punctured spectrum of \(A^\wedge\) to the punctured spectrum of \(A\) is surjective (see Algebra, Lemma 00HQ). Hence if the punctured spectrum of \(A\) is disconnected, then the same is true for \(A^\wedge\).
Assume the punctured spectrum of \(A^\wedge\) is disconnected. This means that \[\Spec(A^\wedge) \setminus \{\mathfrak m^\wedge\} = Z \amalg Z'\] with \(Z\) and \(Z'\) closed. Let \(\overline{Z}, \overline{Z}' \subset \Spec(A^\wedge)\) be the closures. Say \(\overline{Z} = V(J)\), \(\overline{Z}' = V(J')\) for some ideals \(J, J' \subset A^\wedge\). Then \(V(J + J') = \{\mathfrak m^\wedge\}\) and \(V(JJ') = \Spec(A^\wedge)\). The first equality means that \(\mathfrak m^\wedge = \sqrt{J + J'}\) which implies \((\mathfrak m^\wedge)^e \subset J + J'\) for some \(e \geq 1\). The second equality implies every element of \(JJ'\) is nilpotent hence \((JJ')^n = 0\) for some \(n \geq 1\). Combined this means that \(J^n/J^{2n}\) is annihilated by \(J^n\) and \((J')^n\) and hence by \((\mathfrak m^\wedge)^{2en}\). Thus we may apply Lemma 0ALR to see that there is a finite type \(A\)-algebra \(C\) and an isomorphism \(A^\wedge/J^n = C^\wedge\).
The rest of the proof is exactly the same as the second part of the proof of Lemma 0C2B; of course that lemma is a special case of this one! We have \(C/\mathfrak m C = A/\mathfrak m\) because \(A^\wedge/J^n\) has the same property. Hence \(\mathfrak m_C = \mathfrak m C\) is a maximal ideal and \(A \to C\) is unramified at \(\mathfrak m_C\) (Algebra, Lemma 02FM). After replacing \(C\) by a principal localization we may assume that \(C\) is a quotient of an étale \(A\)-algebra \(B\), see Algebra, Proposition 0395. However, since the residue field extension of \(A \to C_{\mathfrak m_C}\) is trivial and \(A\) is henselian, we conclude that \(B = A\) again after a localization. Thus \(C = A/I\) for some ideal \(I \subset A\) and it follows that \(J^n = IA^\wedge\) (because completion is exact in our situation by Algebra, Lemma 00MB) and \(I = J^n \cap A\) (by flatness of \(A \to A^\wedge\)). By symmetry \(I' = (J')^n \cap A\) satisfies \((J')^n = I'A^\wedge\). Then \(\mathfrak m^e \subset I + I'\) and \(II' = 0\) and we conclude that \(V(I)\) and \(V(I')\) are closed subschemes which give the desired disjoint union decomposition of the punctured spectrum of \(A\).
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring of dimension \(1\). Then the number of (geometric) branches of \(A\) and \(A^\wedge\) is the same.
Proof
To see this for the number of branches, combine Lemmas 0C28, 0C29, and 0C2B and use that the dimension of \(A^\wedge\) is one, see Lemma 07NV. To see this is true for the number of geometric branches we use the result for branches, the fact that the dimension does not change under strict henselization (Lemma 06LK), and the fact that \((A^{sh})^\wedge = ((A^\wedge)^{sh})^\wedge\) by Lemma 06LJ.
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring. If the formal fibres of \(A\) are geometrically normal (for example if \(A\) is excellent or quasi-excellent), then \(A\) is Nagata and the number of (geometric) branches of \(A\) and \(A^\wedge\) is the same.
Proof
Since a normal ring is reduced, we see that \(A\) is Nagata by Lemma 0BJ0. In the rest of the proof we will use Lemma 0BIX, Proposition 0BIV, and Lemma 0BIU. This tells us that \(A\) is a P-ring where \(P(k \to R) =\)“\(R\) is geometrically normal over \(k\)” and the same is true for any (essentially of) finite type \(A\)-algebra.
Let \(\mathfrak q \subset A\) be a minimal prime. Then \(A^\wedge/\mathfrak q A^\wedge = (A/\mathfrak q)^\wedge\) and \(A^h/\mathfrak qA^h = (A/\mathfrak q)^h\) (Algebra, Lemma 05WQ). Hence the number of branches of \(A\) is the sum of the number of branches of the rings \(A/\mathfrak q\) and similarly for \(A^\wedge\). In this way we reduce to the case that \(A\) is a domain.
Assume \(A\) is a domain. Let \(A'\) be the integral closure of \(A\) in the fraction field \(K\) of \(A\). Since \(A\) is Nagata, we see that \(A \to A'\) is finite. Recall that the number of branches of \(A\) is the number of maximal ideals \(\mathfrak m'\) of \(A'\) (Lemma 0C24). Also, recall that \[(A')^\wedge = A' \otimes_A A^\wedge = \prod\nolimits_{\mathfrak m' \subset A'} (A'_{\mathfrak m'})^\wedge\] by Algebra, Lemma 07N9. Because \(A'_{\mathfrak m'}\) is a local ring whose formal fibres are geometrically normal, we see that \((A'_{\mathfrak m'})^\wedge\) is normal (Lemma 0C23). Hence the minimal primes of \(A' \otimes_A A^\wedge\) are in \(1\)-to-\(1\) correspondence with the factors in the decomposition above. By flatness of \(A \to A^\wedge\) we have \[A^\wedge \subset A' \otimes_A A^\wedge \subset K \otimes_A A^\wedge\] Since the left and the right ring have the same set of minimal primes, the same is true for the ring in the middle (small detail omitted) and this finishes the proof.
To see this is true for the number of geometric branches we use the result for branches, the fact that the formal fibres of \(A^{sh}\) are geometrically normal (Lemmas 0BIX and 0C36) and the fact that \((A^{sh})^\wedge = ((A^\wedge)^{sh})^\wedge\) by Lemma 06LJ.
Formally catenary rings
In this section we prove a theorem of Ratliff [Ratliff] that a Noetherian local ring is universally catenary if and only if it is formally catenary.
Definition
A Noetherian local ring \(A\) is formally catenary if for every minimal prime \(\mathfrak p \subset A\) the spectrum of \(A^\wedge/\mathfrak p A^\wedge\) is equidimensional.
Let \(A\) be a Noetherian local ring which is formally catenary. By Ratliff’s result (Proposition 0AW6) we see that any quotient of \(A\) is also formally catenary (because the class of universally catenary rings is stable under quotients). We conclude that the spectrum of \(A^\wedge/\mathfrak p A^\wedge\) is equidimensional for every prime ideal \(\mathfrak p\) of \(A\).
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring which is not formally catenary. Then \(A\) is not universally catenary.
Proof
By assumption there exists a minimal prime \(\mathfrak p \subset A\) such that the spectrum of \(A^\wedge /\mathfrak p A^\wedge\) is not equidimensional. After replacing \(A\) by \(A/\mathfrak p\) we may assume that \(A\) is a domain and that the spectrum of \(A^\wedge\) is not equidimensional. Let \(\mathfrak q\) be a minimal prime of \(A^\wedge\) such that \(d = \dim(A^\wedge/\mathfrak q)\) is minimal and hence \(0 < d < \dim(A)\). We prove the lemma by induction on \(d\).
The case \(d = 1\). In this case \(\dim(A^\wedge_\mathfrak q) = 0\). Hence \(A^\wedge_\mathfrak q\) is Artinian local and we see that for some \(n > 0\) the ideal \(J = \mathfrak q^n\) maps to zero in \(A^\wedge_\mathfrak q\). It follows that \(\mathfrak m\) is the only associated prime of \(J/J^2\), whence \(\mathfrak m^m\) annihilates \(J/J^2\) for some \(m > 0\). Thus we can use Lemma 0ALR to find \(A \to B\) of finite type such that \(B^\wedge \cong A^\wedge/J\). It follows that \(\mathfrak m_B = \sqrt{\mathfrak mB}\) is a maximal ideal with the same residue field as \(\mathfrak m\) and \(B^\wedge\) is the \(\mathfrak m_B\)-adic completion (Algebra, Lemma 0394). Then \[\dim(B_{\mathfrak m_B}) = \dim(B^\wedge) = 1 = d.\] Since we have the factorization \(A \to B \to A^\wedge/J\) the inverse image of \(\mathfrak q/J\) is a prime \(\mathfrak q' \subset \mathfrak m_B\) lying over \((0)\) in \(A\). Thus, if \(A\) were universally catenary, the dimension formula (Algebra, Lemma 02IJ) would give \[\begin{align*} \dim(B_{\mathfrak m_B}) & \geq \dim((B/\mathfrak q')_{\mathfrak m_B}) \\ & = \dim(A) + \text{trdeg}_A(B/\mathfrak q') - \text{trdeg}_{\kappa(\mathfrak m)}(\kappa(\mathfrak m_B)) \\ & = \dim(A) + \text{trdeg}_A(B/\mathfrak q') \end{align*}\] This contradiction finishes the argument in case \(d = 1\).
Assume \(d > 1\). Let \(Z \subset \Spec(A^\wedge)\) be the union of the irreducible components distinct from \(V(\mathfrak q)\). Let \(\mathfrak r_1, \ldots, \mathfrak r_m \subset A^\wedge\) be the prime ideals corresponding to irreducible components of \(V(\mathfrak q) \cap Z\) of dimension \(> 0\). Choose \(f \in \mathfrak m\), \(f \not \in A \cap \mathfrak r_j\) using prime avoidance (Algebra, Lemma 00DS). Then \(\dim(A/fA) = \dim(A) - 1\) and there is some irreducible component of \(V(\mathfrak q, f)\) of dimension \(d - 1\). Thus \(A/fA\) is not formally catenary and the invariant \(d\) has decreased. By induction \(A/fA\) is not universally catenary, hence \(A\) is not universally catenary.
Lemma
Let \(A \to B\) be a flat local ring map of local Noetherian rings. Assume \(B\) is catenary and is \(\Spec(B)\) equidimensional. Then
\(\Spec(B/\mathfrak p B)\) is equidimensional for all \(\mathfrak p \subset A\) and
\(A\) is catenary and \(\Spec(A)\) is equidimensional.
Proof
Let \(\mathfrak p \subset A\) be a prime ideal. Let \(\mathfrak q \subset B\) be a prime minimal over \(\mathfrak pB\). Then \(\mathfrak q \cap A = \mathfrak p\) by going down for \(A \to B\) (Algebra, Lemma 00HS). Hence \(A_\mathfrak p \to B_\mathfrak q\) is a flat local ring map with special fibre of dimension \(0\) and hence \[\dim(A_\mathfrak p) = \dim(B_\mathfrak q) = \dim(B) - \dim(B/\mathfrak q)\] (Algebra, Lemma 00ON). The second equality because \(\Spec(B)\) is equidimensional and \(B\) is catenary. Thus \(\dim(B/\mathfrak q)\) is independent of the choice of \(\mathfrak q\) and we conclude that \(\Spec(B/\mathfrak p B)\) is equidimensional of dimension \(\dim(B) - \dim(A_\mathfrak p)\). On the other hand, we have \(\dim(B/\mathfrak p B) = \dim(A/\mathfrak p) + \dim(B/\mathfrak m_A B)\) and \(\dim(B) = \dim(A) + \dim(B/\mathfrak m_A B)\) by flatness (see lemma cited above) and we get \[\dim(A_\mathfrak p) = \dim(A) - \dim(A/\mathfrak p)\] for all \(\mathfrak p\) in \(A\). Applying this to all minimal primes in \(A\) we see that \(A\) is equidimensional. If \(\mathfrak p \subset \mathfrak p'\) is a strict inclusion with no primes in between, then we may apply the above to the prime \(\mathfrak p'/\mathfrak p\) in \(A/\mathfrak p\) because \(A/\mathfrak p \to B/\mathfrak p B\) is flat and \(\Spec(B/\mathfrak p B)\) is equidimensional, to get \[1 = \dim((A/\mathfrak p)_{\mathfrak p'}) = \dim(A/\mathfrak p) - \dim(A/\mathfrak p')\] Thus \(\mathfrak p \mapsto \dim(A/\mathfrak p)\) is a dimension function and we conclude that \(A\) is catenary.
Lemma
Let \(A\) be a formally catenary Noetherian local ring. Then \(A\) is universally catenary.
Proof
We may replace \(A\) by \(A/\mathfrak p\) where \(\mathfrak p\) is a minimal prime of \(A\), see Algebra, Lemma 0AUP. Thus we may assume that the spectrum of \(A^\wedge\) is equidimensional. It suffices to show that every local ring essentially of finite type over \(A\) is catenary (see for example Algebra, Lemma 0AUN). Hence it suffices to show that \(A[x_1, \ldots, x_n]_\mathfrak m\) is catenary where \(\mathfrak m \subset A[x_1, \ldots, x_n]\) is a maximal ideal lying over \(\mathfrak m_A\), see Algebra, Lemma 0AUG (and Algebra, Lemmas 00NK and 00NJ). Let \(\mathfrak m' \subset A^\wedge[x_1, \ldots, x_n]\) be the unique maximal ideal lying over \(\mathfrak m\). Then \[A[x_1, \ldots, x_n]_\mathfrak m \to A^\wedge[x_1, \ldots, x_n]_{\mathfrak m'}\] is local and flat (Algebra, Lemma 00MB). Hence it suffices to show that the ring on the right hand side catenary with equidimensional spectrum, see Lemma 0AW4. It is catenary because complete local rings are universally catenary (Algebra, Remark 032C). Pick any minimal prime \(\mathfrak q\) of \(A^\wedge[x_1, \ldots, x_n]_{\mathfrak m'}\). Then \(\mathfrak q = \mathfrak p A^\wedge[x_1, \ldots, x_n]_{\mathfrak m'}\) for some minimal prime \(\mathfrak p\) of \(A^\wedge\) (small detail omitted). Hence \[\dim(A^\wedge[x_1, \ldots, x_n]_{\mathfrak m'}/\mathfrak q) = \dim(A^\wedge/\mathfrak p) + n = \dim(A^\wedge) + n\] the first equality by Algebra, Lemma 00ON and the second because the spectrum of \(A^\wedge\) is equidimensional. This finishes the proof.
Proposition
A Noetherian local ring is universally catenary if and only if it is formally catenary.
Proof
Lemma
Let \((A, \mathfrak m)\) be a Noetherian local ring with geometrically normal formal fibres. Then
\(A^h\) is universally catenary, and
if \(A\) is unibranch (for example normal), then \(A\) is universally catenary.
Proof
By Lemma 0C2E the number of branches of \(A\) and \(A^\wedge\) are the same, hence Lemma 0C29 applies. Then for any minimal prime \(\mathfrak q \subset A^h\) we see that \(A^\wedge/\mathfrak q A^\wedge\) has a unique minimal prime. Thus \(A^h\) is formally catenary (by definition) and hence universally catenary by Proposition 0AW6. If \(A\) is unibranch, then \(A^h\) has a unique minimal prime, hence \(A^\wedge\) has a unique minimal prime, hence \(A\) is formally catenary and we conclude in the same way.
Group actions and integral closure
This section is in some sense a continuation of Algebra, Section 037E. More material of a similar kind can be found in Fundamental Groups, Section 0BSN
Lemma
Let \(\varphi : A \to B\) be a surjection of rings. Let \(G\) be a finite group of order \(n\) acting on \(\varphi : A \to B\). If \(b \in B^G\), then there exists a monic polynomial \(P \in A^G[T]\) which maps to \((T - b)^n\) in \(B^G[T]\).
Proof
Choose \(a \in A\) lifting \(b\) and set \(P = \prod_{\sigma \in G} (T - \sigma(a))\).
Lemma
Let \(R\) be a ring. Let \(G\) be a finite group acting on \(R\). Let \(I \subset R\) be an ideal such that \(\sigma(I) \subset I\) for all \(\sigma \in G\). Then \(R^G/I^G \subset (R/I)^G\) is an integral extension of rings which induces a homeomorphism on spectra and purely inseparable extensions of residue fields.
Proof
Since \(I^G = R^G \cap I\) it is clear that the map is injective. Lemma 0BRF shows that Algebra, Lemma 09EF applies.
Lemma
Let \(G\) be a finite group of order \(n\) acting on a ring \(R\). Let \(J \subset R^G\) be an ideal. For \(x \in JR\) we have \(\prod_{\sigma \in G} (T - \sigma(x)) = T^n + a_1 T^{n - 1} + \ldots + a_n\) with \(a_i \in J\).
Proof
Observe that the polynomial is indeed monic and has coefficients in \(R^G\). We can write \(x = f_1 b_1 + \ldots + f_m b_m\) with \(f_j \in J\) and \(b_j \in R\). Thus, arguing by induction on \(m\), we may assume that \(x = y - fb\) with \(f \in J\), \(b \in R\), and \(y \in JR\) such that the result holds for \(y\). Then we see that \[\prod\nolimits_{\sigma \in G} (T - \sigma(x)) = \prod\nolimits_{\sigma \in G} (T - \sigma(y) + f\sigma(b)) = \prod\nolimits_{\sigma \in G} (T - \sigma(y)) + \sum_{i = 1, \ldots, n} f^i a_i\] where we have \[a_i = \sum\nolimits_{S \subset G,\ |S| = i} \prod\nolimits_{\sigma \in S} \sigma(b) \prod\nolimits_{\sigma \not \in S} (T - \sigma(y))\] A computation we omit shows that \(a_i \in R^G\) (hint: the given expression is symmetric). Thus the polynomial of the statement of the lemma for \(x\) is congruent modulo \(J\) to the polynomial for \(y\) and this proves the induction step.
Lemma
Let \(R\) be a ring. Let \(G\) be a finite group of order \(n\) acting on \(R\). Let \(J \subset R^G\) be an ideal. Then \(R^G/J \to (R/JR)^G\) is ring map such that
for \(b \in (R/JR)^G\) there is a monic polynomial \(P \in R^G/J[T]\) whose image in \((R/JR)^G[T]\) is \((T - b)^n\),
for \(a \in \Ker(R^G/J \to (R/JR)^G)\) we have \((T - a)^n = T^n\) in \(R^G/J[T]\).
In particular, \(R^G/J \to (R/JR)^G\) is an integral ring map which induces homeomorphisms on spectra and purely inseparable extensions of residue fields.
Proof
Part (1) follow from Lemma 0BRF with \(I = JR\). If \(a\) is as in part (2), then \(a\) is the image of \(x \in R^G \cap JR\). Hence \((T - x)^n = \prod_{\sigma \in G} (T - \sigma(x))\) is congruent to \(T^n\) modulo \(J\) by Lemma 0H34. This proves part (2). To see the final statement we may apply Algebra, Lemma 09EF.
Remark
In Lemma 0H35 we see that the map \(R^G/J \to (R/JR)^G\) is an isomorphism if \(n\) is invertible in \(R\).
Lemma
Let \(R\) be a ring. Let \(G\) be a finite group of order \(n\) acting on \(R\). Let \(A\) be an \(R^G\)-algebra.
for \(b \in (A \otimes_{R^G} R)^G\) there exists a monic polynomial \(P \in A[T]\) whose image in \((A \otimes_{R^G} R)^G[T]\) is \((T - b)^n\),
for \(a \in \Ker(A \to (A \otimes_{R^G} R)^G)\) we have \((T - a)^n = T^n\) in \(A[T]\).
Proof
Choose a surjection \(E \to A\) where \(E\) is a polynomial algebra over \(R^G\). Then \((E \otimes_{R^G} R)^G = E\) because \(E\) is free as an \(R^G\)-module. Denote \(J = \Ker(E \to A)\). Since tensor product is right exact we see that \(A \otimes_{R^G} R\) is the quotient of \(E \otimes_{R^G} R\) by the ideal generated by \(J\). In this way we see that our lemma is a special case of Lemma 0H35.
Lemma
Let \(R\) be a ring. Let \(G\) be a finite group acting on \(R\). Let \(R^G \to A\) be a ring map. The map \[A \to (A \otimes_{R^G} R)^G\] is an isomorphism if \(R^G \to A\) is flat. In general the map is integral, induces a homeomorphism on spectra, and induces purely inseparable residue field extensions.
Proof
To see the first statement consider the exact sequence \(0 \to R^G \to R \to \bigoplus_{\sigma \in G} R\) where the second map sends \(x\) to \((\sigma(x) - x)_{\sigma \in G}\). Tensoring with \(A\) the sequence remains exact if \(R^G \to A\) is flat. Thus \(A\) is the \(G\)-invariants in \(A \otimes_{R^G} R\).
The second statement follows from Lemma 0BRG and Algebra, Lemma 09EF.
Lemma
Let \(G\) be a finite group acting on a ring \(R\). For any two primes \(\mathfrak q, \mathfrak q' \subset R\) lying over the same prime in \(R^G\) there exists a \(\sigma \in G\) with \(\sigma(\mathfrak q) = \mathfrak q'\).
Proof
The extension \(R^G \subset R\) is integral because every \(x \in R\) is a root of the monic polynomial \(\prod_{\sigma \in G}(T - \sigma(x))\) in \(R^G[T]\). Thus there are no inclusion relations among the primes lying over a given prime \(\mathfrak p\) (Algebra, Lemma 00GT). If the lemma is wrong, then we can choose \(x \in \mathfrak q'\), \(x \not \in \sigma(\mathfrak q)\) for all \(\sigma \in G\). See Algebra, Lemma 00DS. Then \(y = \prod_{\sigma \in G} \sigma(x)\) is in \(R^G\) and in \(\mathfrak p = R^G \cap \mathfrak q'\). On the other hand, \(x \not \in \sigma(\mathfrak q)\) for all \(\sigma\) means \(\sigma(x) \not \in \mathfrak q\) for all \(\sigma\). Hence \(y \not \in \mathfrak q\) as \(\mathfrak q\) is a prime ideal. This is impossible as \(y \in \mathfrak p \subset \mathfrak q\).
Lemma
Let \(G\) be a finite group acting on a ring \(R\). Let \(\mathfrak q \subset R\) be a prime lying over \(\mathfrak p \subset R^G\). Then \(\kappa(\mathfrak q)/\kappa(\mathfrak p)\) is an algebraic normal extension and the map \[D = \{\sigma \in G \mid \sigma(\mathfrak q) = \mathfrak q\} \longrightarrow \text{Aut}(\kappa(\mathfrak q)/\kappa(\mathfrak p))\] is surjective19.
Proof
With \(A = (R^G)_\mathfrak p\) and \(B = A \otimes_{R^G} R\) we see that \(A = B^G\) as localization is flat, see Lemma 0BRH. Observe that \(\mathfrak pA\) and \(\mathfrak qB\) are prime ideals, \(D\) is the stabilizer of \(\mathfrak qB\), and \(\kappa(\mathfrak p) = \kappa(\mathfrak pA)\) and \(\kappa(\mathfrak q) = \kappa(\mathfrak qB)\). Thus we may replace \(R\) by \(B\) and assume that \(\mathfrak p\) is a maximal ideal. Since \(R^G \subset R\) is an integral ring extension, we find that the maximal ideals of \(R\) are exactly the primes lying over \(\mathfrak p\) (follows from Algebra, Lemmas 00GT and 00GU). By Lemma 0BRI there are finitely many of them \(\mathfrak q = \mathfrak q_1, \mathfrak q_2, \ldots, \mathfrak q_m\) and they form a single orbit for \(G\). By the Chinese remainder theorem (Algebra, Lemma 00DT) the map \(R \to \prod_{j = 1, \ldots, m} R/\mathfrak q_j\) is surjective.
First we prove that the extension is normal. Pick an element \(\alpha \in \kappa(\mathfrak q)\). We have to show that the minimal polynomial \(P\) of \(\alpha\) over \(\kappa(\mathfrak p)\) splits completely. By the above we can choose \(a \in \mathfrak q_2 \cap \ldots \cap \mathfrak q_m\) mapping to \(\alpha\) in \(\kappa(\mathfrak q)\). Consider the polynomial \(Q = \prod_{\sigma \in G} (T - \sigma(a))\) in \(R^G[T]\). The image of \(Q\) in \(R[T]\) splits completely into linear factors, hence the same is true for its image in \(\kappa(\mathfrak q)[T]\). Since \(P\) divides the image of \(Q\) in \(\kappa(\mathfrak p)[T]\) we conclude that \(P\) splits completely into linear factors over \(\kappa(\mathfrak q)\) as desired.
Since \(\kappa(\mathfrak q)/\kappa(\mathfrak p)\) is normal we may assume \(\kappa(\mathfrak q) = \kappa_1 \otimes_{\kappa(\mathfrak p)} \kappa_2\) with \(\kappa_1/\kappa(\mathfrak p)\) purely inseparable and \(\kappa_2/\kappa(\mathfrak p)\) Galois, see Fields, Lemma 030M. Pick \(\alpha \in \kappa_2\) which generates \(\kappa_2\) over \(\kappa(\mathfrak p)\) if it is finite and a subfield of degree \(> |G|\) if it is infinite (to get a contradiction). This is possible by Fields, Lemma 030N. Pick \(a\), \(P\), and \(Q\) as in the previous paragraph. If \(\alpha' \in \kappa_2\) is a Galois conjugate of \(\alpha\) over \(\kappa(\mathfrak p)\), then the fact that \(P\) divides the image of \(Q\) in \(\kappa(\mathfrak p)[T]\) shows there exists a \(\sigma \in G\) such that \(\sigma(a)\) maps to \(\alpha'\). By our choice of \(a\) (vanishing at other maximal ideals) this implies \(\sigma \in D\) and that the image of \(\sigma\) in \(\text{Aut}(\kappa(\mathfrak q)/\kappa(\mathfrak p))\) maps \(\alpha\) to \(\alpha'\). Hence the surjectivity or the desired absurdity in case \(\alpha\) has degree \(> |G|\) over \(\kappa(\mathfrak p)\).
Lemma
Let \(A\) be a normal domain with fraction field \(K\). Let \(L/K\) be a (possibly infinite) Galois extension. Let \(G = \text{Gal}(L/K)\) and let \(B\) be the integral closure of \(A\) in \(L\).
For any two primes \(\mathfrak q, \mathfrak q' \subset B\) lying over the same prime in \(A\) there exists a \(\sigma \in G\) with \(\sigma(\mathfrak q) = \mathfrak q'\).
Let \(\mathfrak q \subset B\) be a prime lying over \(\mathfrak p \subset A\). Then \(\kappa(\mathfrak q)/\kappa(\mathfrak p)\) is an algebraic normal extension and the map \[D = \{\sigma \in G \mid \sigma(\mathfrak q) = \mathfrak q\} \longrightarrow \text{Aut}(\kappa(\mathfrak q)/\kappa(\mathfrak p))\] is surjective.
Proof
Proof of (1). Consider pairs \((M, \sigma)\) where \(K \subset M \subset L\) is a subfield such that \(M/K\) is Galois, \(\sigma \in \text{Gal}(M/K)\) with \(\sigma(\mathfrak q \cap M) = \mathfrak q' \cap M\). We say \((M', \sigma') \geq (M, \sigma)\) if and only if \(M \subset M'\) and \(\sigma'|_M = \sigma\). Observe that \((K, \text{id}_K)\) is such a pair as \(A = K \cap B\) since \(A\) is a normal domain. The collection of these pairs satisfies the hypotheses of Zorn’s lemma, hence there exists a maximal pair \((M, \sigma)\). If \(M \not = L\), then we can find \(M \subset M' \subset L\) with \(M'/M\) nontrivial and finite and \(M'/K\) Galois (Fields, Lemma 0BMG). Choose \(\sigma' \in \text{Gal}(M'/K)\) whose restriction to \(M\) is \(\sigma\) (Fields, Lemma 0BMK). Then the primes \(\sigma'(\mathfrak q \cap M')\) and \(\mathfrak q' \cap M'\) restrict to the same prime of \(B \cap M\). Since \(B \cap M = (B \cap M')^{\text{Gal}(M'/M)}\) we can use Lemma 0BRI to find \(\tau \in \text{Gal}(M'/M)\) with \(\tau(\sigma'(\mathfrak q \cap M')) = \mathfrak q' \cap M'\). Hence \((M', \tau \circ \sigma') > (M, \sigma)\) contradicting the maximality of \((M, \sigma)\).
Part (2) is proved in exactly the same manner as part (1). We write out the details. Pick \(\overline{\sigma} \in \text{Aut}(\kappa(\mathfrak q)/\kappa(\mathfrak p))\). Consider pairs \((M, \sigma)\) where \(K \subset M \subset L\) is a subfield such that \(M/K\) is Galois, \(\sigma \in \text{Gal}(M/K)\) with \(\sigma(\mathfrak q \cap M) = \mathfrak q \cap M\) and \[\xymatrix{ \kappa(\mathfrak q \cap M) \ar[r] \ar[d]_\sigma & \kappa(\mathfrak q) \ar[d]_{\overline{\sigma}} \\ \kappa(\mathfrak q \cap M) \ar[r] & \kappa(\mathfrak q) }\] commutes. We say \((M', \sigma') \geq (M, \sigma)\) if and only if \(M \subset M'\) and \(\sigma'|_M = \sigma\). As above \((K, \text{id}_K)\) is such a pair. The collection of these pairs satisfies the hypotheses of Zorn’s lemma, hence there exists a maximal pair \((M, \sigma)\). If \(M \not = L\), then we can find \(M \subset M' \subset L\) with \(M'/M\) finite and \(M'/K\) Galois (Fields, Lemma 0BMG). Choose \(\sigma' \in \text{Gal}(M'/K)\) whose restriction to \(M\) is \(\sigma\) (Fields, Lemma 0BMK). Then the primes \(\sigma'(\mathfrak q \cap M')\) and \(\mathfrak q \cap M'\) restrict to the same prime of \(B \cap M\). Adjusting the choice of \(\sigma'\) as in the first paragraph, we may assume that \(\sigma'(\mathfrak q \cap M') = \mathfrak q \cap M'\). Then \(\sigma'\) and \(\overline{\sigma}\) define maps \(\kappa(\mathfrak q \cap M') \to \kappa(\mathfrak q)\) which agree on \(\kappa(\mathfrak q \cap M)\). Since \(B \cap M = (B \cap M')^{\text{Gal}(M'/M)}\) we can use Lemma 0BRJ to find \(\tau \in \text{Gal}(M'/M)\) with \(\tau(\mathfrak q \cap M') = \mathfrak q \cap M'\) such that \(\tau \circ \sigma'\) and \(\overline{\sigma}\) induce the same map on \(\kappa(\mathfrak q \cap M')\). There is a small detail here in that the lemma first guarantees that \(\kappa(\mathfrak q \cap M')/\kappa(\mathfrak q \cap M)\) is normal, which then tells us that the difference between the maps is an automorphism of this extension (Fields, Lemma 0BR4), to which we can apply the lemma to get \(\tau\). Hence \((M', \tau \circ \sigma') > (M, \sigma)\) contradicting the maximality of \((M, \sigma)\).
Lemma
Let \(A\) be a normal domain with fraction field \(K\). Let \(M/L/K\) be a tower of (possibly infinite) Galois extensions of \(K\). Let \(H = \text{Gal}(M/K)\) and \(G = \text{Gal}(L/K)\) and let \(C\) and \(B\) be the integral closure of \(A\) in \(M\) and \(L\). Let \(\mathfrak r \subset C\) and \(\mathfrak q = B \cap \mathfrak r\). Set \(D_\mathfrak r = \{\tau \in H \mid \tau(\mathfrak r) = \mathfrak r\}\) and \(I_\mathfrak r = \{\tau \in D_\mathfrak r \mid \tau \bmod \mathfrak r = \text{id}_{\kappa(\mathfrak r)}\}\) and similarly for \(D_\mathfrak q\) and \(I_\mathfrak q\). Under the map \(H \to G\) the induced maps \(D_\mathfrak r \to D_\mathfrak q\) and \(I_\mathfrak r \to I_\mathfrak q\) are surjective.
Proof
Let \(\sigma \in D_\mathfrak q\). Pick \(\tau \in H\) mapping to \(\sigma\). This is possible by Fields, Lemma 0BMK. Then \(\tau(\mathfrak r)\) and \(\mathfrak r\) both lie over \(\mathfrak q\). Hence by Lemma 0BRK there exists a \(\sigma' \in \text{Gal}(M/L)\) with \(\sigma'(\tau(\mathfrak r)) = \mathfrak r\). Hence \(\sigma'\tau \in D_\mathfrak r\) maps to \(\sigma\). The case of inertia groups is proved in exactly the same way using surjectivity onto automorphism groups.
Extensions of discrete valuation rings
In this section and the next few we use the following definitions.
Definition
We say that \(A \to B\) or \(A \subset B\) is an extension of discrete valuation rings if \(A\) and \(B\) are discrete valuation rings and \(A \to B\) is injective and local. In particular, if \(\pi_A\) and \(\pi_B\) are uniformizers of \(A\) and \(B\), then \(\pi_A = u \pi_B^e\) for some \(e \geq 1\) and unit \(u\) of \(B\). The integer \(e\) does not depend on the choice of the uniformizers as it is also the unique integer \(\geq 1\) such that \[\mathfrak m_A B = \mathfrak m_B^e\] The integer \(e\) is called the ramification index of \(B\) over \(A\). We say that \(B\) is weakly unramified over \(A\) if \(e = 1\). If the extension of residue fields \(\kappa_A = A/\mathfrak m_A \subset \kappa_B = B/\mathfrak m_B\) is finite, then we set \(f = [\kappa_B : \kappa_A]\) and we call it the residual degree or residue degree of the extension \(A \subset B\).
Note that we do not require the extension of fraction fields to be finite.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). If the extension \(L/K\) is finite, then the residue field extension is finite and we have \(ef \leq [L : K]\).
Proof
Finiteness of the residue field extension is Algebra, Lemma 031F. The inequality follows from Algebra, Lemmas 00PE and 02M0.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings. Then the ramification indices of \(B/A\) and \(C/B\) multiply to give the ramification index of \(C/A\). In a formula \(e_{C/A} = e_{B/A} e_{C/B}\). Similarly for the residual degrees in case they are finite.
Proof
This is immediate from the definitions and Fields, Lemma 09G9.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings inducing the field extension \(K \subset L\). If the characteristic of \(K\) is \(p > 0\) and \(L\) is purely inseparable over \(K\), then the ramification index \(e\) is a power of \(p\).
Proof
Write \(\pi_A = u \pi_B^e\) for some \(u \in B^*\). On the other hand, we have \(\pi_B^q \in K\) for some \(p\)-power \(q\). Write \(\pi_B^q = v \pi_A^k\) for some \(v \in A^*\) and \(k \in \mathbf{Z}\). Then \(\pi_A^q = u^q \pi_B^{qe} = u^q v^e \pi_A^{ke}\). Taking valuations in \(B\) we conclude that \(ke = q\).
In the following lemma we discuss what it means for an extension \(A \subset B\) of discrete valuation rings to be “unramified”, i.e., have ramification index \(1\) and separable (possibly nonalgebraic) extension of residue fields. However, we cannot use the term “unramified” itself because there already exists a notion of an unramified ring map, see Algebra, Section 00US.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings. The following are equivalent
\(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology, and
\(A \to B\) is weakly unramified and \(\kappa_B/\kappa_A\) is a separable field extension.
Proof
This follows from Proposition 07NQ and Algebra, Proposition 0322.
Remark
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite separable field extension. Let \(B \subset L\) be the integral closure of \(A\) in \(L\). Picture: \[\xymatrix{ B \ar[r] & L \\ A \ar[u] \ar[r] & K \ar[u] }\] By Algebra, Lemma 032L the ring extension \(A \subset B\) is finite, hence \(B\) is Noetherian. By Algebra, Lemma 00OK the dimension of \(B\) is \(1\), hence \(B\) is a Dedekind domain, see Algebra, Lemma 034X. Let \(\mathfrak m_1, \ldots, \mathfrak m_n\) be the maximal ideals of \(B\) (i.e., the primes lying over \(\mathfrak m_A\)). We obtain extensions of discrete valuation rings \[A \subset B_{\mathfrak m_i}\] and hence ramification indices \(e_i\) and residue degrees \(f_i\). We have \[[L : K] = \sum\nolimits_{i = 1, \ldots, n} e_i f_i\] by Algebra, Lemma 02MJ applied to a uniformizer in \(A\). We observe that \(n = 1\) if \(A\) is henselian (by Algebra, Lemma 04GH), e.g. if \(A\) is complete.
Definition
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite separable extension. With \(B\) and \(\mathfrak m_i\), \(i = 1, \ldots, n\) as in Remark 09E8 we say the extension \(L/K\) is
unramified with respect to \(A\) if \(e_i = 1\) and the extension \(\kappa(\mathfrak m_i)/\kappa_A\) is separable for all \(i\),
tamely ramified with respect to \(A\) if either the characteristic of \(\kappa_A\) is \(0\) or the characteristic of \(\kappa_A\) is \(p > 0\), the field extensions \(\kappa(\mathfrak m_i)/\kappa_A\) are separable, and the ramification indices \(e_i\) are prime to \(p\), and
totally ramified with respect to \(A\) if \(n = 1\) and the residue field extension \(\kappa(\mathfrak m_1)/\kappa_A\) is trivial.
If the discrete valuation ring \(A\) is clear from context, then we sometimes say \(L/K\) is unramified, totally ramified, or tamely ramified for short.
For unramified extensions we have the following basic lemma.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\).
If \(M/L/K\) are finite separable extensions and \(M\) is unramified with respect to \(A\), then \(L\) is unramified with respect to \(A\).
If \(L/K\) is a finite separable extension which is unramified with respect to \(A\), then there exists a Galois extension \(M/K\) containing \(L\) which is unramified with respect to \(A\).
If \(L_1/K\), \(L_2/K\) are finite separable extensions which are unramified with respect to \(A\), then there exists a a finite separable extension \(L/K\) which is unramified with respect to \(A\) containing \(L_1\) and \(L_2\).
Proof
We will use the results of the discussion in Remark 09E8 without further mention.
Proof of (1). Let \(C/B/A\) be the integral closures of \(A\) in \(M/L/K\). Since \(C\) is a finite ring extension of \(B\), we see that \(\Spec(C) \to \Spec(B)\) is surjective. Hence for ever maximal ideal \(\mathfrak m \subset B\) there is a maximal ideal \(\mathfrak m' \subset C\) lying over \(\mathfrak m\). By the multiplicativity of ramification indices (Lemma 0BRL) and the assumption, we conclude that the ramification index of \(B_\mathfrak m\) over \(A\) is \(1\). Since \(\kappa(\mathfrak m')/\kappa_A\) is finite separable, the same is true for \(\kappa(\mathfrak m)/\kappa_A\).
Proof of (2). Let \(M\) be the normal closure of \(L\) over \(K\), see Fields, Definition 0BMF. Then \(M/K\) is Galois by Fields, Lemma 0EXM. On the other hand, there is a surjection \[L \otimes_K \ldots \otimes_K L \longrightarrow M\] of \(K\)-algebras, see Fields, Lemma 0EXL. Let \(B\) be the integral closure of \(A\) in \(L\) as in Remark 09E8. The condition that \(L\) is unramified with respect to \(A\) exactly means that \(A \to B\) is an étale ring map, see Algebra, Lemma 00U6. By permanence properties of étale ring maps we see that \[B \otimes_A \ldots \otimes_A B\] is étale over \(A\), see Algebra, Lemma 00U2. Hence the displayed ring is a product of Dedekind domains, see Lemma 0AP2. We conclude that \(M\) is the fraction field of a Dedekind domain finite étale over \(A\). This means that \(M\) is unramified with respect to \(A\) as desired.
Proof of (3). Let \(B_i \subset L_i\) be the integral closure of \(A\). Argue in the same manner as above to show that \(B_1 \otimes_A B_2\) is finite étale over \(A\). Details omitted.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(M/L/K\) be finite separable extensions. Let \(B\) be the integral closure of \(A\) in \(L\). If \(L/K\) is unramified with respect to \(A\) and \(M/L\) is unramified with respect to \(B_\mathfrak m\) for every maximal ideal \(\mathfrak m\) of \(B\), then \(M/K\) is unramified with respect to \(A\).
Proof
Let \(C\) be the integral closure of \(A\) in \(M\). Every maximal ideal \(\mathfrak m'\) of \(C\) lies over a maximal ideal \(\mathfrak m\) of \(B\). Then the lemma follows from the multiplicativity of ramification indices (Lemma 0BRL) and the fact that we have the tower \(\kappa(\mathfrak m')/\kappa(\mathfrak m)/\kappa_A\) of finite extensions of fields.
Galois extensions and ramification
In the case of Galois extensions, we can elaborate on the discussion in Section 0EXQ.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension with Galois group \(G\). Then \(G\) acts on the ring \(B\) of Remark 09E8 and acts transitively on the set of maximal ideals of \(B\).
Proof
Observe that \(A = B^G\) as \(A\) is integrally closed in \(K\) and \(K = L^G\). Hence this lemma is a special case of Lemma 0BRI.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension. Then there are \(e \geq 1\) and \(f \geq 1\) such that \(e_i = e\) and \(f_i = f\) for all \(i\) (notation as in Remark 09E8). In particular \([L : K] = n e f\).
Proof
Immediate consequence of Lemma 09EA and the definitions.
Definition
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension with Galois group \(G\). Let \(B\) be the integral closure of \(A\) in \(L\). Let \(\mathfrak m \subset B\) be a maximal ideal.
The decomposition group of \(\mathfrak m\) is the subgroup \(D = \{\sigma \in G \mid \sigma(\mathfrak m) = \mathfrak m\}\).
The inertia group of \(\mathfrak m\) is the kernel \(I\) of the map \(D \to \text{Aut}(\kappa(\mathfrak m)/\kappa_A)\).
Note that the field \(\kappa(\mathfrak m)\) may be inseparable over \(\kappa_A\). In particular the field extension \(\kappa(\mathfrak m)/\kappa_A\) need not be Galois. If \(\kappa_A\) is perfect, then it is.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\) and residue field \(\kappa\). Let \(L/K\) be a finite Galois extension with Galois group \(G\). Let \(B\) be the integral closure of \(A\) in \(L\). Let \(\mathfrak m\) be a maximal ideal of \(B\). Then
the field extension \(\kappa(\mathfrak m)/\kappa\) is normal, and
\(D \to \text{Aut}(\kappa(\mathfrak m)/\kappa)\) is surjective.
If for some (equivalently all) maximal ideal(s) \(\mathfrak m \subset B\) the field extension \(\kappa(\mathfrak m)/\kappa\) is separable, then
\(\kappa(\mathfrak m)/\kappa\) is Galois, and
\(D \to \text{Gal}(\kappa(\mathfrak m)/\kappa)\) is surjective.
Here \(D \subset G\) is the decomposition group of \(\mathfrak m\).
Proof
Observe that \(A = B^G\) as \(A\) is integrally closed in \(K\) and \(K = L^G\). Thus parts (1) and (2) follow from Lemma 0BRJ. The “equivalently all” part of the lemma follows from Lemma 09EA. Assume \(\kappa(\mathfrak m)/\kappa\) is separable. Then parts (3) and (4) follow immediately from (1) and (2).
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension with Galois group \(G\). Let \(B\) be the integral closure of \(A\) in \(L\). Let \(\mathfrak m \subset B\) be a maximal ideal. The inertia group \(I\) of \(\mathfrak m\) sits in a canonical exact sequence \[1 \to P \to I \to I_t \to 1\] such that
if \(D\) is the decomposition group we have \[P = \{\sigma \in I \mid \sigma|_{\mathfrak m/\mathfrak m^2} = \text{id}_{\mathfrak m/\mathfrak m^2}\} = \{\sigma \in D \mid \sigma \text{ acts trivially on }\text{Gr}_\mathfrak m(B)\}\]
\(P\) is a normal subgroup of \(D\),
\(P\) is a \(p\)-group if the characteristic of \(\kappa_A\) is \(p > 0\) and \(P = \{1\}\) if the characteristic of \(\kappa_A\) is zero,
\(I_t\) is cyclic of order the prime to \(p\) part of the integer \(e\),
there is a canonical isomorphism \(\theta : I_t \to \mu_e(\kappa(\mathfrak m))\), and
\(P = \{\sigma \in D \mid \sigma|_{B/\mathfrak m^2} = \text{id}_{B/\mathfrak m^2}\}\) if \(\kappa(m)\) is separable over the residue field of \(A\).
Here \(e\) is the integer of Lemma 09EB.
Proof
Recall that \(|G| = [L : K] = nef\), see Lemma 09EB. Since \(G\) acts transitively on the set \(\{\mathfrak m_1, \ldots, \mathfrak m_n\}\) of maximal ideals of \(B\) (Lemma 09EA) and since \(D\) is the stabilizer of an element we see that \(|D| = ef\). By Lemma 09ED we have \[ef = |D| = |I| \cdot |\text{Aut}(\kappa(\mathfrak m)/\kappa)|\] where \(\kappa\) is the residue field of \(A\). As \(\kappa(\mathfrak m)\) is normal over \(\kappa\) the order of \(\text{Aut}(\kappa(\mathfrak m)/\kappa)\) differs from \(f\) by a power of \(p\) (see Fields, Lemma 09HS and discussion following Fields, Definition 030L). Hence the prime to \(p\) part of \(|I|\) is equal to the prime to \(p\) part of \(e\).
Set \(C = B_\mathfrak m\). Then \(I\) acts on \(C\) over \(A\) and trivially on the residue field of \(C\). Let \(\pi_A \in A\) and \(\pi_C \in C\) be uniformizers. Write \(\pi_A = u \pi_C^e\) for some unit \(u\) in \(C\). For \(\sigma \in I\) write \(\sigma(\pi_C) = \theta_\sigma \pi_C\) for some unit \(\theta_\sigma\) in \(C\). Then we have \[\pi_A = \sigma(\pi_A) = \sigma(u) (\theta_\sigma \pi_C)^e = \sigma(u) \theta_\sigma^e \pi_C^e = \frac{\sigma(u)}{u} \theta_\sigma^e \pi_A\] Since \(\sigma(u) \equiv u \bmod \mathfrak m_C\) as \(\sigma \in I\) we see that the image \(\overline{\theta}_\sigma\) of \(\theta_\sigma\) in \(\kappa_C = \kappa(\mathfrak m)\) is an \(e\)th root of unity. We obtain a map [0BU3]\[\begin{equation} \theta : I \longrightarrow \mu_e(\kappa(\mathfrak m)),\quad \sigma \mapsto \overline{\theta}_\sigma \end{equation}\] We claim that \(\theta\) is a homomorphism of groups and independent of the choice of uniformizer \(\pi_C\). Namely, if \(\tau\) is a second element of \(I\), then \(\tau(\sigma(\pi_C)) = \tau(\theta_\sigma \pi_C) = \tau(\theta_\sigma) \theta_\tau \pi_C\), hence \(\theta_{\tau \sigma} = \tau(\theta_\sigma) \theta_\tau\) and since \(\tau \in I\) we conclude that \(\overline{\theta}_{\tau \sigma} = \overline{\theta}_\sigma \overline{\theta}_\tau\). If \(\pi'_C\) is a second uniformizer, then we see that \(\pi'_C = w \pi_C\) for some unit \(w\) of \(C\) and \(\sigma(\pi'_C) = w^{-1}\sigma(w)\theta_\sigma \pi'_C\), hence \(\theta'_\sigma = w^{-1}\sigma(w)\theta_\sigma\), hence \(\theta'_\sigma\) and \(\theta_\sigma\) map to the same element of the residue field as before.
Since \(\kappa(\mathfrak m)\) has characteristic \(p\), the group \(\mu_e(\kappa(\mathfrak m))\) is cyclic of order at most the prime to \(p\) part of \(e\) (see Fields, Section 09HW).
Let \(P = \Ker(\theta)\). By construction the elements of \(P\) are exactly the elements of \(I\) which act trivially on \(\mathfrak m/\mathfrak m^2 = \text{Gr}_\mathfrak m^1(B)\), i.e., \(P = \{\sigma \in I \mid \sigma|_{\mathfrak m/\mathfrak m^2} = \text{id}_{\mathfrak m/\mathfrak m^2}\}\). Also \(I\) consists of the elements of \(D\) which act trivially on \(\kappa(\mathfrak m) = B/\mathfrak m\). Since the graded ring \(\text{Gr}_\mathfrak m(B)\) is generated by \(\text{Gr}^1_\mathfrak m(B) = \mathfrak m/\mathfrak m^2\) over \(\text{Gr}^0_\mathfrak m(B) = B/\mathfrak m\), we conclude that \(P = \{\sigma \in D \mid \sigma \text{ acts trivially on } \text{Gr}_\mathfrak m(B)\}\). Thus (1) is true. This implies (2) as \(P\) is the kernel of the homomorphism \(D \to \text{Aut}(\text{Gr}_\mathfrak m(B))\). If we can prove (3), then parts (4) and (5) will follow as \(I_t\) will be isomorphic to \(\mu_e(\kappa(\mathfrak m))\) as the arguments above show that \(|I_t| \geq |\mu_e(\kappa(\mathfrak m))|\).
Thus it suffices to prove that the kernel \(P\) of \(\theta\) is a \(p\)-group. Let \(\sigma\) be a nontrivial element of the kernel. Then \(\sigma - \text{id}\) sends \(\mathfrak m_C^i\) into \(\mathfrak m_C^{i + 1}\) for all \(i\). Let \(m\) be the order of \(\sigma\). Pick \(c \in C\) such that \(\sigma(c) \not = c\). Then \(\sigma(c) - c \in \mathfrak m_C^i\), \(\sigma(c) - c \not \in \mathfrak m_C^{i + 1}\) for some \(i\) and we have \[\begin{align*} 0 & = \sigma^m(c) - c \\ & = \sigma^m(c) - \sigma^{m - 1}(c) + \ldots + \sigma(c) - c \\ & = \sum\nolimits_{j = 0, \ldots, m - 1} \sigma^j(\sigma(c) - c) \\ & \equiv m(\sigma(c) - c) \bmod \mathfrak m_C^{i + 1} \end{align*}\] It follows that \(p | m\) (or \(m = 0\) if \(p = 1\)). Thus every element of the kernel of \(\theta\) has order divisible by \(p\), i.e., \(\Ker(\theta)\) is a \(p\)-group.
Proof of (6). Assume \(\kappa(\mathfrak m)/\kappa\) is separable. In this case \(f = |D/I|\) by Lemma 09ED and \(I\) has order \(e\). If \(e = 1\), then \(P = I = \{\text{id}\}\) and the result is true. If \(e > 1\), then \(B/\mathfrak m^2 = C/\pi_C^2C\) is a \(\kappa\)-algebra and a small extension of \(\kappa(\mathfrak m)\). Because \(\kappa(\mathfrak m)\) is formally étale over \(\kappa\) (Algebra, Lemma 090W) there is a unique \(\kappa\)-algebra map \(\kappa(\mathfrak m) \to B/\mathfrak m^2\) right inverse to \(B/\mathfrak m^2 \to \kappa(\mathfrak m)\). In other words, there is a \(D\)-equivariant isomorphism \(B/\mathfrak m^2 = \mathfrak m/\mathfrak m^2 \oplus \kappa(\mathfrak m)\) of \(\kappa\)-algebras. This immediately shows that \(P = \{\sigma \in D \mid \sigma|_{B/\mathfrak m^2} = \text{id}_{B/\mathfrak m^2}\}\) in this case.
Example
The equality in (6) of Lemma 09EE is false without the assumption on \(\kappa(\mathfrak m)\). A counterexample is the extension of \(A = Z_2[t]_{(2)}\) given by \(B = A[x]/(x^2 - t)\). Namely then \(\sigma(x) = -x = x - 2x\) and \(2x\) is not in \(\mathfrak m^2\).
Definition
With assumptions and notation as in Lemma 09EE.
The wild inertia group of \(\mathfrak m\) is the subgroup \(P\).
The tame inertia group of \(\mathfrak m\) is the quotient \(I \to I_t\).
We denote \(\theta : I \to \mu_e(\kappa(\mathfrak m))\) the surjective map (0BU3) whose kernel is \(P\) and which induces the isomorphism \(I_t \to \mu_e(\kappa(\mathfrak m))\).
Lemma
With assumptions and notation as in Lemma 09EE. The inertia character \(\theta : I \to \mu_e(\kappa(\mathfrak m))\) satisfies the following property \[\theta(\tau \sigma \tau^{-1}) = \tau(\theta(\sigma))\] for \(\tau \in D\) and \(\sigma \in I\).
Proof
The formula makes sense as \(I\) is a normal subgroup of \(D\) and as \(\tau\) acts on \(\kappa(\mathfrak m)\) via the map \(D \to \text{Aut}(\kappa(\mathfrak m))\) discussed in Lemma 09ED for example. Recall the construction of \(\theta\). Choose a uniformizer \(\pi\) of \(B_\mathfrak m\) and for \(\sigma \in I\) write \(\sigma(\pi) = \theta_\sigma \pi\). Then \(\theta(\sigma)\) is the image \(\overline{\theta}_\sigma\) of \(\theta_\sigma\) in the residue field. For any \(\tau \in D\) we can write \(\tau(\pi) = \theta_\tau \pi\) for some unit \(\theta_\tau\). Then \(\theta_{\tau^{-1}} = \tau^{-1}(\theta_\tau^{-1})\). We compute \[\begin{align*} \theta_{\tau \sigma \tau^{-1}} & = \tau(\sigma(\tau^{-1}(\pi)))/\pi \\ & = \tau(\sigma(\tau^{-1}(\theta_\tau^{-1}) \pi))/\pi \\ & = \tau(\sigma(\tau^{-1}(\theta_\tau^{-1})) \theta_\sigma \pi)/\pi \\ & = \tau(\sigma(\tau^{-1}(\theta_\tau^{-1}))) \tau(\theta_\sigma) \theta_\tau \end{align*}\] However, since \(\sigma\) acts trivially modulo \(\pi\) we see that the product \(\tau(\sigma(\tau^{-1}(\theta_\tau^{-1}))) \theta_\tau\) maps to \(1\) in the residue field. This proves the lemma.
We will generalize the following lemma in Fundamental Groups, Lemma 0BSU.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension. Let \(\mathfrak m \subset B\) be a maximal ideal of the integral closure of \(A\) in \(L\). Let \(I \subset G\) be the inertia group of \(\mathfrak m\). Then \(B^I\) is the integral closure of \(A\) in \(L^I\) and \(A \to (B^I)_{B^I \cap \mathfrak m}\) is étale.
Proof
Write \(B' = B^I\). It follows from the definitions that \(B' = B^I\) is the integral closure of \(A\) in \(L^I\). Write \(\mathfrak m' = B^I \cap \mathfrak m = B' \cap \mathfrak m \subset B'\). By Lemma 0BRI the maximal ideal \(\mathfrak m\) is the unique prime ideal of \(B\) lying over \(\mathfrak m'\). As \(I\) acts trivially on \(\kappa(\mathfrak m)\) we see from Lemma 09EG that the extension \(\kappa(\mathfrak m)/\kappa(\mathfrak m')\) is purely inseparable (perhaps an easier alternative is to apply the result of Lemma 0BRJ). Since \(D/I\) acts faithfully on \(\kappa(\mathfrak m')\), we conclude that \(D/I\) acts faithfully on \(\kappa(\mathfrak m)\). Of course the elements of the residue field \(\kappa\) of \(A\) are fixed by this action. By Galois theory we see that \([\kappa(\mathfrak m') : \kappa] \geq |D/I|\), see Fields, Lemma 09I3.
Let \(\pi\) be the uniformizer of \(A\). Since \(\text{Norm}_{L/K}(\pi) = \pi^{[L : K]}\) we see from Algebra, Lemma 02MJ that \[|G| = [L : K] = [L : K]\ \text{ord}_A(\pi) = |G/D|\ [\kappa(\mathfrak m) : \kappa]\ \text{ord}_{B_\mathfrak m}(\pi)\] as there are \(n = |G/D|\) maximal ideals of \(B\) which are all conjugate under \(G\), see Remark 09E8 and Lemma 09EA. Applying the same reasoning to the finite extension the finite extension \(L/L^I\) of degree \(|I|\) we find \[|I|\ \text{ord}_{B'_{\mathfrak m'}}(\pi) = [\kappa(\mathfrak m) : \kappa(\mathfrak m')]\ \text{ord}_{B_\mathfrak m}(\pi)\] We conclude that \[\text{ord}_{B'_{\mathfrak m'}}(\pi) = \frac{|D/I|}{[\kappa(\mathfrak m') : \kappa]}\] Since the left hand side is a positive integer and since the right hand side is \(\leq 1\) by the above, we conclude that we have equality, \(\text{ord}_{B'_{\mathfrak m'}}(\pi) = 1\) and \(\kappa(\mathfrak m')/\kappa\) has degree \(|D/I|\). Thus \(\pi B'_{\mathfrak m'} = \mathfrak m' B_\mathfrak m'\) and \(\kappa(\mathfrak m')\) is Galois over \(\kappa\) with Galois group \(D/I\), in particular separable, see Fields, Lemma 09I1. By Algebra, Lemma 00U6 we find that \(A \to B'_{\mathfrak m'}\) is étale as desired.
Remark
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(L/K\) be a finite Galois extension. Let \(\mathfrak m \subset B\) be a maximal ideal of the integral closure of \(A\) in \(L\). Let \[P \subset I \subset D \subset G\] be the wild inertia, inertia, decomposition group of \(\mathfrak m\). Consider the diagram \[\xymatrix{ \mathfrak m \ar@{-}[d] \ar@{-}[r] & \mathfrak m^P \ar@{-}[d] \ar@{-}[r] & \mathfrak m^I \ar@{-}[d] \ar@{-}[r] & \mathfrak m^D \ar@{-}[d] \ar@{-}[r] & A \cap \mathfrak m \ar@{-}[d] \\ B & B^P \ar[l] & B^I \ar[l] & B^D \ar[l] & A \ar[l] }\] Observe that \(B^P, B^I, B^D\) are the integral closures of \(A\) in the fields \(L^P\), \(L^I\), \(L^D\). Thus we also see that \(B^P\) is the integral closure of \(B^I\) in \(L^P\) and so on. Observe that \(\mathfrak m^P = \mathfrak m \cap B^P\), \(\mathfrak m^I = \mathfrak m \cap B^I\), and \(\mathfrak m^D = \mathfrak m \cap B^D\). Hence the top line of the diagram corresponds to the images of \(\mathfrak m \in \Spec(B)\) under the induced maps of spectra. Having said all of this we have the following
the extension \(L^I/L^D\) is Galois with group \(D/I\),
the extension \(L^P/L^I\) is Galois with group \(I_t = I/P\),
the extension \(L^P/L^D\) is Galois with group \(D/P\),
\(\mathfrak m^I\) is the unique prime of \(B^I\) lying over \(\mathfrak m^D\),
\(\mathfrak m^P\) is the unique prime of \(B^P\) lying over \(\mathfrak m^I\),
\(\mathfrak m\) is the unique prime of \(B\) lying over \(\mathfrak m^P\),
\(\mathfrak m^P\) is the unique prime of \(B^P\) lying over \(\mathfrak m^D\),
\(\mathfrak m\) is the unique prime of \(B\) lying over \(\mathfrak m^I\),
\(\mathfrak m\) is the unique prime of \(B\) lying over \(\mathfrak m^D\),
\(A \to B^D_{\mathfrak m^D}\) is étale and induces a trivial residue field extension,
\(B^D_{\mathfrak m^D} \to B^I_{\mathfrak m^I}\) is étale and induces a Galois extension of residue fields with Galois group \(D/I\),
\(A \to B^I_{\mathfrak m^I}\) is étale,
\(B^I_{\mathfrak m^I} \to B^P_{\mathfrak m^P}\) has ramification index \(|I/P|\) prime to \(p\) and induces a trivial residue field extension,
\(B^D_{\mathfrak m^D} \to B^P_{\mathfrak m^P}\) has ramification index \(|I/P|\) prime to \(p\) and induces a separable residue field extension,
\(A \to B^P_{\mathfrak m^P}\) has ramification index \(|I/P|\) prime to \(p\) and induces a separable residue field extension.
Statements (1), (2), and (3) are immediate from Galois theory (Fields, Section 09DU) and Lemma 09EE. Statements (4) – (9) are clear from Lemma 09EA. Part (12) is Lemma 09EH. Since we have the factorization \(A \to B^D_{\mathfrak m^D} \to B^I_{\mathfrak m^I}\) we obtain the étaleness in (10) and (11) as a consequence. The residue field extension in (10) must be trivial because it is separable and \(D/I\) maps onto \(\text{Aut}(\kappa(\mathfrak m)/\kappa_A)\) as shown in Lemma 09ED. The same argument provides the proof of the statement on residue fields in (11). To see (13), (14), and (15) it suffices to prove (13). By the above, the extension \(L^P/L^I\) is Galois with a cyclic Galois group of order prime to \(p\), the prime \(\mathfrak m^P\) is the unique prime lying over \(\mathfrak m^I\) and the action of \(I/P\) on the residue field is trivial. Thus we can apply Lemma 09EE to this extension and the discrete valuation ring \(B^I_{\mathfrak m^I}\) to see that (13) holds.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(M/L/K\) be a tower with \(M/K\) and \(L/K\) finite Galois. Let \(C\), \(B\) be the integral closure of \(A\) in \(M\), \(L\). Let \(\mathfrak m' \subset C\) be a maximal ideal and set \(\mathfrak m = \mathfrak m' \cap B\). Let \[P \subset I \subset D \subset \text{Gal}(L/K) \quad\text{and}\quad P' \subset I' \subset D' \subset \text{Gal}(M/K)\] be the wild inertia, inertia, decomposition group of \(\mathfrak m\) and \(\mathfrak m'\). Then the canonical surjection \(\text{Gal}(M/K) \to \text{Gal}(L/K)\) induces surjections \(P' \to P\), \(I' \to I\), and \(D' \to D\). Moreover these fit into commutative diagrams \[\vcenter{ \xymatrix{ D' \ar[r] \ar[d] & \text{Aut}(\kappa(\mathfrak m')/\kappa_A) \ar[d] \\ D \ar[r] & \text{Aut}(\kappa(\mathfrak m)/\kappa_A) } } \quad\text{and}\quad \vcenter{ \xymatrix{ I' \ar[r]_-{\theta'} \ar[d] & \mu_{e'}(\kappa(\mathfrak m')) \ar[d]^{(-)^{e'/e}} \\ I \ar[r]^-\theta & \mu_e(\kappa(\mathfrak m)) } }\] where \(e'\) and \(e\) are the ramification indices of \(A \to C_{\mathfrak m'}\) and \(A \to B_\mathfrak m\).
Proof
The fact that under the map \(\text{Gal}(M/K) \to \text{Gal}(L/K)\) the groups \(P', I', D'\) map into \(P, I, D\) is immediate from the definitions of these groups. The commutativity of the first diagram is clear (observe that since \(\kappa(\mathfrak m)/\kappa_A\) is normal every automorphism of \(\kappa(\mathfrak m')\) over \(\kappa_A\) indeed induces an automorphism of \(\kappa(\mathfrak m)\) over \(\kappa_A\) and hence we obtain the right vertical arrow in the first diagram, see Lemma 09ED and Fields, Lemma 0BME).
The maps \(I' \to I\) and \(D' \to D\) are surjective by Lemma 0BSX. The surjectivity of \(P' \to P\) follows as \(P'\) and \(P\) are p-Sylow subgroups of \(I'\) and \(I\).
To see the commutativity of the second diagram we choose a uniformizer \(\pi'\) of \(C_{\mathfrak m'}\) and a uniformizer \(\pi\) of \(B_\mathfrak m\). Then \(\pi = c' (\pi')^{e'/e}\) for some unit \(c'\) of \(C_{\mathfrak m'}\). For \(\sigma' \in I'\) the image \(\sigma \in I\) is simply the restriction of \(\sigma'\) to \(L\). Write \(\sigma'(\pi') = c \pi'\) for a unit \(c \in C_{\mathfrak m'}\) and write \(\sigma(\pi) = b \pi\) for a unit \(b\) of \(B_\mathfrak m\). Then \(\sigma'(\pi) = b \pi\) and we obtain \[b \pi = \sigma'(\pi) = \sigma'(c' (\pi')^{e'/e}) = \sigma'(c') c^{e'/e} (\pi')^{e'/e} = \frac{\sigma'(c')}{c'} c^{e'/e} \pi\] As \(\sigma' \in I'\) we see that \(b\) and \(c^{e'/e}\) have the same image in the residue field which proves what we want.
Remark
In order to use the inertia character \(\theta : I \to \mu_e(\kappa(\mathfrak m))\) for infinite Galois extensions, it is convenient to scale it. Let \(A, K, L, B, \mathfrak m, G, P, I, D, e, \theta\) be as in Lemma 09EE and Definition 0BU4. Then \(e = q |I_t|\) with \(q\) is a power of the characteristic \(p\) of \(\kappa(\mathfrak m)\) if positive or \(1\) if zero. Note that \(\mu_e(\kappa(\mathfrak m)) = \mu_{|I_t|}(\kappa(\mathfrak m))\) because the characteristic of \(\kappa(\mathfrak m)\) is \(p\). Consider the map \[\theta_{can} = q\theta : I \longrightarrow \mu_{|I_t|}(\kappa(\mathfrak m))\] This map induces an isomorphism \(\theta_{can} : I_t \to \mu_{|I_t|}(\kappa(\mathfrak m))\). We have \(\theta_{can}(\tau \sigma \tau^{-1}) = \tau(\theta_{can}(\sigma))\) for \(\tau \in D\) and \(\sigma \in I\) by Lemma 0BU5. Finally, if \(M/L\) is an extension such that \(M/K\) is Galois and \(\mathfrak m'\) is a prime of the integral closure of \(A\) in \(M\) lying over \(\mathfrak m\), then we get the commutative diagram \[\xymatrix{ I' \ar[r]_-{\theta'_{can}} \ar[d] & \mu_{|I'_t|}(\kappa(\mathfrak m')) \ar[d]^{(-)^{|I'_t|/|I_t|}} \\ I \ar[r]^-{\theta_{can}} & \mu_{|I_t|}(\kappa(\mathfrak m)) }\] by Lemma 0BU7.
Krasner’s lemma
Here is Krasner’s lemma in the case of discretely valued fields.
Lemma
Let \(A\) be a complete local domain of dimension \(1\). Let \(P(t) \in A[t]\) be a polynomial with coefficients in \(A\). Let \(\alpha \in A\) be a root of \(P\) but not a root of the derivative \(P' = \text{d}P/\text{d}t\). For every \(c \geq 0\) there exists an integer \(n\) such that for any \(Q \in A[t]\) whose coefficients are in \(\mathfrak m_A^n\) the polynomial \(P + Q\) has a root \(\beta \in A\) with \(\beta - \alpha \in \mathfrak m_A^c\).
Proof
Choose a nonzero \(\pi \in \mathfrak m\). Since the dimension of \(A\) is \(1\) we have \(\mathfrak m = \sqrt{(\pi)}\). By assumption we may write \(P'(\alpha)^{-1} = \pi^{-m} a\) for some \(m \geq 0\) and \(a \in A\). We may and do assume that \(c \geq m + 1\). Pick \(n\) such that \(\mathfrak m_A^n \subset (\pi^{c + m})\). Pick any \(Q\) as in the statement. For later use we observe that we can write \[P(x + y) = P(x) + P'(x)y + R(x, y)y^2\] for some \(R(x, y) \in A[x, y]\). We will show by induction that we can find a sequence \(\alpha_m, \alpha_{m + 1}, \alpha_{m + 2}, \ldots\) such that
\(\alpha_k \equiv \alpha \bmod \pi^c\),
\(\alpha_{k + 1} - \alpha_k \in (\pi^k)\), and
\((P + Q)(\alpha_k) \in (\pi^{m + k})\).
Setting \(\beta = \lim \alpha_k\) will finish the proof.
Base case. Since the coefficients of \(Q\) are in \((\pi^{c + m})\) we have \((P + Q)(\alpha) \in (\pi^{c + m})\). Hence \(\alpha_m = \alpha\) works. This choice guarantees that \(\alpha_k \equiv \alpha \bmod \pi^c\) for all \(k \geq m\).
Induction step. Given \(\alpha_k\) we write \(\alpha_{k + 1} = \alpha_k + \delta\) for some \(\delta \in (\pi^k)\). Then we have \[(P + Q)(\alpha_{k + 1}) = P(\alpha_k + \delta) + Q(\alpha_k + \delta)\] Because the coefficients of \(Q\) are in \((\pi^{c + m})\) we see that \(Q(\alpha_k + \delta) \equiv Q(\alpha_k) \bmod \pi^{c + m + k}\). On the other hand we have \[P(\alpha_k + \delta) = P(\alpha_k) + P'(\alpha_k)\delta + R(\alpha_k, \delta)\delta^2\] Note that \(P'(\alpha_k) \equiv P'(\alpha) \bmod (\pi^{m + 1})\) as \(\alpha_k \equiv \alpha \bmod \pi^{m + 1}\). Hence we obtain \[P(\alpha_k + \delta) \equiv P(\alpha_k) + P'(\alpha) \delta \bmod \pi^{k + m + 1}\] Recombining the two terms we see that \[(P + Q)(\alpha_{k + 1}) \equiv (P + Q)(\alpha_k) + P'(\alpha) \delta \bmod \pi^{k + m + 1}\] Thus a solution is to take \(\delta = -P'(\alpha)^{-1} (P + Q)(\alpha_k) = - \pi^{-m} a (P + Q)(\alpha_k)\) which is contained in \((\pi^k)\) by induction assumption.
Lemma
Let \(A\) be a discrete valuation ring with field of fractions \(K\). Let \(A^\wedge\) be the completion of \(A\) with fraction field \(K^\wedge\). If \(M/K^\wedge\) is a finite separable extension, then there exists a finite separable extension \(L/K\) such that \(M = K^\wedge \otimes_K L\).
Proof
Note that \(A^\wedge\) is a discrete valuation ring too (by Lemmas 07NY and 07NV). In particular \(A^\wedge\) is a domain. The proof will work more generally for Noetherian local rings \(A\) such that \(A^\wedge\) is a local domain of dimension \(1\).
Let \(\theta \in M\) be an element that generates \(M\) over \(K^\wedge\). (Theorem of the primitive element.) Let \(P(t) \in K^\wedge[t]\) be the minimal polynomial of \(\theta\) over \(K^\wedge\). Let \(\pi \in \mathfrak m_A\) be a nonzero element. After replacing \(\theta\) by \(\pi^n\theta\) we may assume that the coefficients of \(P(t)\) are in \(A^\wedge\). Let \(B = A^\wedge[\theta] = A^\wedge[t]/(P(t))\). Note that \(B\) is a complete local domain of dimension \(1\) because it is finite over \(A\) and contained in \(M\). Since \(M\) is separable over \(K\) the element \(\theta\) is not a root of the derivative of \(P\). For any integer \(n\) we can find a monic polynomial \(P_1 \in A[t]\) such that \(P - P_1\) has coefficients in \(\pi^nA^\wedge[t]\). By Krasner’s lemma (Lemma 09EI) we see that \(P_1\) has a root \(\beta\) in \(B\) for \(n\) sufficiently large. Moreover, we may assume (if \(n\) is chosen large enough) that \(\theta - \beta \in \pi B\). Consider the map \(\Phi : A^\wedge[t]/(P_1) \to B\) of \(A^\wedge\)-algebras which maps \(t\) to \(\beta\). Since \(B = \pi B + \sum_{i < \deg(P)} A^\wedge \theta^i\), the map \(\Phi\) is surjective by Nakayama’s lemma. As \(\deg(P_1) = \deg(P)\) it follows that \(\Phi\) is an isomorphism. We conclude that the ring extension \(L = K[t]/(P_1(t))\) satisfies \(K^\wedge \otimes_K L \cong M\). This implies that \(L\) is a field and the proof is complete.
Definition
Let \(A\) be a discrete valuation ring. We say \(A\) has mixed characteristic if the characteristic of the residue field of \(A\) is \(p > 0\) and the characteristic of the fraction field of \(A\) is \(0\). In this case we obtain an extension of discrete valuation rings \(\mathbf{Z}_{(p)} \subset A\) and the absolute ramification index of \(A\) is the ramification index of this extension.
Abhyankar’s lemma and tame ramification
In this section we prove what we think is the most general version of Abhyankar’s lemma for discrete valuation rings. After doing so, we apply this to prove some results about tamely ramified extensions of the fraction field of a discrete valuation ring.
Remark
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Let \(K_1/K\) be a finite extension of fields. Let \(A_1 \subset K_1\) be the integral closure of \(A\) in \(K_1\). On the other hand, let \(L_1 = (L \otimes_K K_1)_{red}\). Then \(L_1\) is a nonempty finite product of finite field extensions of \(L\). Let \(B_1\) be the integral closure of \(B\) in \(L_1\). We obtain compatible commutative diagrams \[\vcenter{ \xymatrix{ L \ar[r] & L_1 \\ K \ar[u] \ar[r] & K_1 \ar[u] } } \quad\text{and}\quad \vcenter{ \xymatrix{ B \ar[r] & B_1 \\ A \ar[u] \ar[r] & A_1 \ar[u] } }\] In this situation we have the following
By Algebra, Lemma 09IG the ring \(A_1\) is a Dedekind domain and \(B_1\) is a finite product of Dedekind domains.
Note that \(L \otimes_K K_1 = (B \otimes_A A_1)_\pi\) where \(\pi \in A\) is a uniformizer and that \(\pi\) is a nonzerodivisor on \(B \otimes_A A_1\). Thus the ring map \(B \otimes_A A_1 \to B_1\) is integral with kernel consisting of nilpotent elements. Hence \(\Spec(B_1) \to \Spec(B \otimes_A A_1)\) is surjective on spectra (Algebra, Lemma 00GQ). The map \(\Spec(B \otimes_A A_1) \to \Spec(A_1)\) is surjective as \(A_1/\mathfrak m_A A_1 \to B/\mathfrak m_AB \otimes_{\kappa_A} A_1/\mathfrak m_A A_1\) is an injective ring map with \(A_1/\mathfrak m_A A_1\) Artinian. We conclude that \(\Spec(B_1) \to \Spec(A_1)\) is surjective.
Let \(\mathfrak m_i\), \(i = 1, \ldots n\) with \(n \geq 1\) be the maximal ideals of \(A_1\). For each \(i = 1, \ldots, n\) let \(\mathfrak m_{ij}\), \(j = 1, \ldots, m_i\) with \(m_i \geq 1\) be the maximal ideals of \(B_1\) lying over \(\mathfrak m_i\). We obtain diagrams \[\xymatrix{ B \ar[r] & (B_1)_{\mathfrak m_{ij}} \\ A \ar[u] \ar[r] & (A_1)_{\mathfrak m_i} \ar[u] }\] of extensions of discrete valuation rings.
If \(A\) is henselian (for example complete), then \(A_1\) is a discrete valuation ring, i.e., \(n = 1\). Namely, \(A_1\) is a union of finite extensions of \(A\) which are domains, hence local by Algebra, Lemma 04GH.
If \(B\) is henselian (for example complete), then \(B_1\) is a product of discrete valuation rings, i.e., \(m_i = 1\) for \(i = 1, \ldots, n\).
If \(K \subset K_1\) is purely inseparable, then \(A_1\) and \(B_1\) are both discrete valuation rings, i.e., \(n = 1\) and \(m_1 = 1\). This is true because for every \(b \in B_1\) a \(p\)-power power of \(b\) is in \(B\), hence \(B_1\) can only have one maximal ideal.
If \(K \subset K_1\) is finite separable, then \(L_1 = L \otimes_K K_1\) and is a finite product of finite separable extensions too. Hence \(A \subset A_1\) and \(B \subset B_1\) are finite by Algebra, Lemma 032L.
If \(A\) is Nagata, then \(A \subset A_1\) is finite.
If \(B\) is Nagata, then \(B \subset B_1\) is finite.
Lemma
Let \(A\) be a discrete valuation ring with uniformizer \(\pi\). Let \(n \geq 2\). Let \(K_1 = K[\pi^{1/n}]\). Then
\([K_1 : K] = n\),
the integral closure \(A_1\) of \(A\) in \(K_1\) is the ring \(A[\pi^{1/n}]\),
\(A_1\) is a discrete valuation ring,
the ramification index of \(A_1\) over \(A\) is \(n\),
\(K_1\) is totally ramified with respect to \(A\), and
if \(n\) is prime to the residue characteristic of \(A\), then \(K_1/K\) is tamely ramified and any subextension of \(K_1/K\) is generated by \(\pi^{1/d}\) for some divisor \(d\) of \(n\).
Proof
Consider the ring \(A' = A[x]/(x^n - \pi)\) and denote \(\pi'\) the image of \(x\) in \(A_1\). As \(A'\) is finite free of rank \(n\) as an \(A\)-module, the element \(\pi\) is a nonzerodivisor in \(A'\), and hence so is \(\pi'\). Moreover, \(A'/\pi'A'\) is isomorphic to \(A/\pi A\) which is a field. Hence \(A'\) is a discrete valuation ring with uniformizer \(\pi'\). It follows that the fraction field \(K'\) of \(A'\) is an extension of degree \(n\). Clearly, \(K'\) is obtained from \(K\) by adjoint an \(n\)th root \(\pi' = \pi^{1/n}\) of \(\pi\), i.e., \(K' \cong K_1\). Since \(A'\) is normal, we see that \(A'\) is the integral closure of \(A\) in \(K'\), i.e., \(A' = A_1\). This proves (1) – (5).
The assertion that \(K_1/K\) is tamely ramified if \(n\) is invertible in the residue field is immediate from Definition 09E9 and (3) and (4). To prove the very last statement, it suffices to show that every \(n\)th root of unity \(\zeta \in K'\) is contained in \(K\), see Fields, Lemma 0EXP. It is clear that \(\zeta \in (A')^*\). We can write \(\zeta = a + b\) with \(a \in A\) and \(b \in A \pi' \oplus \ldots \oplus A (\pi')^{n - 1}\). If \(b\) is nonzero, then we can write \(b = u(\pi')^t\) where \(t \geq 1\) and \(b\) a unit of \(A'\). Then \[1 = \zeta^n \equiv a^n + n a^{n - 1} u (\pi')^t \bmod (\pi')^{t + 1}\] Since \(n\) and \(a\) are units of \(A\), this would imply that \(b = u(\pi')^t\) is congruent to \((1 - a^n)/(n a^{n - 1}) \in A\) modulo \((\pi')^{t + 1}\). The direct sum decomposition \(A' = A \oplus \ldots \oplus A(\pi')^{n - 1}\) shows that this is impossible because the image of multiplication by \((\pi')^{t + 1}\) is a corresponding direct sum whose intersection with \(A(\pi')^t\) is \((\pi A)(\pi')^t\). Thus \(b = 0\) and \(\zeta \in A\) as desired.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Assume that \(A \to B\) is formally smooth in the \(\mathfrak m_B\)-adic topology. Then for any finite extension \(K_1/K\) we have \(L_1 = L \otimes_K K_1\), \(B_1 = B \otimes_A A_1\), and each extension \((A_1)_{\mathfrak m_i} \subset (B_1)_{\mathfrak m_{ij}}\) (see Remark 09EM) is formally smooth in the \(\mathfrak m_{ij}\)-adic topology.
Proof
We will use the equivalence of Lemma 09E7 without further mention. Let \(\pi \in A\) and \(\pi_i \in (A_1)_{\mathfrak m_i}\) be uniformizers. As \(\kappa_A \subset \kappa_B\) is separable, the ring \[(B \otimes_A (A_1)_{\mathfrak m_i})/\pi_i (B \otimes_A (A_1)_{\mathfrak m_i}) = B/\pi B \otimes_{A/\pi A} (A_1)_{\mathfrak m_i}/\pi_i (A_1)_{\mathfrak m_i}\] is a product of fields each separable over \(\kappa_{\mathfrak m_i}\). Hence the element \(\pi_i\) in \(B \otimes_A (A_1)_{\mathfrak m_i}\) is a nonzerodivisor and the quotient by this element is a product of fields. It follows that \(B \otimes_A A_1\) is a Dedekind domain in particular reduced. Thus \(B \otimes_A A_1 \subset B_1\) is an equality.
The following lemma is our version of Abhyankar’s lemma for discrete valuation rings. Observe that \(\kappa_B/\kappa_A\) is not assumed to be an algebraic extension of fields.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings. Assume that either the residue characteristic of \(A\) is \(0\) or it is \(p\), the ramification index \(e\) is prime to \(p\), and \(\kappa_B/\kappa_A\) is a separable field extension. Let \(K_1/K\) be a finite extension. Using the notation of Remark 09EM assume \(e\) divides the ramification index of \(A \subset (A_1)_{\mathfrak m_i}\) for some \(i\). Then \((A_1)_{\mathfrak m_i} \subset (B_1)_{\mathfrak m_{ij}}\) is formally smooth in the \(\mathfrak m_{ij}\)-adic topology for all \(j = 1, \ldots, m_i\).
Proof
Let \(\pi \in A\) be a uniformizer. Let \(\pi_1\) be a uniformizer of \((A_1)_{\mathfrak m_i}\). Write \(\pi = u \pi_1^{e_1}\) with \(u\) a unit of \((A_1)_{\mathfrak m_i}\) and \(e_1\) the ramification index of \(A \subset (A_1)_{\mathfrak m_i}\).
Claim: we may assume that \(u\) is an \(e\)th power in \(K_1\). Namely, let \(K_2\) be an extension of \(K_1\) obtained by adjoining a root of \(x^e = u\); thus \(K_2\) is a factor of \(K_1[x]/(x^e - u)\). Then \(K_2/K_1\) is a finite separable extension (by our assumption on \(e\)) and hence \(A_1 \subset A_2\) is finite. Since \((A_1)_{\mathfrak m_i} \to (A_1)_{\mathfrak m_i}[x]/(x^e - u)\) is finite étale (as \(e\) is prime to the residue characteristic and \(u\) a unit) we conclude that \((A_2)_{\mathfrak m_i}\) is a factor of a finite étale extension of \((A_1)_{\mathfrak m_i}\) hence finite étale over \((A_1)_{\mathfrak m_i}\) itself. The same reasoning shows that \(B_1 \subset B_2\) induces finite étale extensions \((B_1)_{\mathfrak m_{ij}} \subset (B_2)_{\mathfrak m_{ij}}\). Pick a maximal ideal \(\mathfrak m'_{ij} \subset B_2\) lying over \(\mathfrak m_{ij} \subset B_1\) (of course there may be more than one) and consider \[\xymatrix{ (B_1)_{\mathfrak m_{ij}} \ar[r] & (B_2)_{\mathfrak m'_{ij}} \\ (A_1)_{\mathfrak m_i} \ar[u] \ar[r] & (A_2)_{\mathfrak m'_i} \ar[u] }\] where \(\mathfrak m'_i \subset A_2\) is the image. Now the horizontal arrows have ramification index \(1\) and induce finite separable residue field extensions. Thus, using the equivalence of Lemma 09E7, we see that it suffices to show that the right vertical arrow is formally smooth in the \(\mathfrak m'_{ij}\)-adic topology. Since \(u\) has a \(e\)th root in \(K_2\) we obtain the claim.
Assume \(u\) has an \(e\)th root in \(K_1\). Since \(e | e_1\) and since \(u\) has a \(e\)th root in \(K_1\) we see that \(\pi = \theta^e\) for some \(\theta \in K_1\). Let \(K'_1 = K[\theta] \subset K_1\) be the subfield generated by \(\theta\). By Lemma 09EV the integral closure \(A'_1\) of \(A\) in \(K[\theta]\) is the discrete valuation ring \(A'_1 = A[\theta]\) which has ramification index \(e\) over \(A\). If we can prove the lemma for the extension \(K'_1/K\), then we conclude by Lemma 09EQ applied to the diagram \[\xymatrix{ (B'_1)_{B'_1 \cap \mathfrak m_{ij}} \ar[r] & (B_1)_{\mathfrak m_{ij}} \\ A'_1 \ar[u] \ar[r] & (A_1)_{\mathfrak m_i} \ar[u] }\] for all \(j = 1, \ldots, m_i\). This reduces us to the case discussed in the next paragraph.
Assume \(K_1 = K[\pi^{1/e}]\) and set \(\theta = \pi^{1/e}\). Let \(\pi_B\) be a uniformizer for \(B\) and write \(\pi = w \pi_B^e\) for some unit \(w\) of \(B\). Then we see that \(L_1 = L \otimes_K K_1\) is obtained by adjoining \(\theta / \pi_B\) which is an \(e\)th root of the unit \(w\). Thus \(B \subset B_1\) is finite étale. Thus for any maximal ideal \(\mathfrak m \subset B_1\) consider the commutative diagram \[\xymatrix{ B \ar[r]_1 & (B_1)_{\mathfrak m} \\ A \ar[u]^e \ar[r]^e & A_1 \ar[u]_{e_\mathfrak m} }\] Here the numbers along the arrows are the ramification indices. By multiplicativity of ramification indices (Lemma 0BRL) we conclude \(e_\mathfrak m = 1\). Looking at the residue field extensions we find that \(\kappa(\mathfrak m)\) is a finite separable extension of \(\kappa_B\) which is separable over \(\kappa_A\). Therefore \(\kappa(\mathfrak m)\) is separable over \(\kappa_A\) which is equal to the residue field of \(A_1\) and we win by Lemma 09E7.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(M/L/K\) be finite separable extensions. Let \(B\) be the integral closure of \(A\) in \(L\). If \(L/K\) is tamely ramified with respect to \(A\) and \(M/L\) is tamely ramified with respect to \(B_\mathfrak m\) for every maximal ideal \(\mathfrak m\) of \(B\), then \(M/K\) is tamely ramified with respect to \(A\).
Proof
Let \(C\) be the integral closure of \(A\) in \(M\). Every maximal ideal \(\mathfrak m'\) of \(C\) lies over a maximal ideal \(\mathfrak m\) of \(B\). Then the lemma follows from the multiplicativity of ramification indices (Lemma 0BRL) and the fact that we have the tower \(\kappa(\mathfrak m')/\kappa(\mathfrak m)/\kappa_A\) of finite extensions of fields.
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). If \(M/L/K\) are finite separable extensions and \(M\) is tamely ramified with respect to \(A\), then \(L\) is tamely ramified with respect to \(A\).
Proof
We will use the results of the discussion in Remark 09E8 without further mention. Let \(C/B/A\) be the integral closures of \(A\) in \(M/L/K\). Since \(C\) is a finite ring extension of \(B\), we see that \(\Spec(C) \to \Spec(B)\) is surjective. Hence for ever maximal ideal \(\mathfrak m \subset B\) there is a maximal ideal \(\mathfrak m' \subset C\) lying over \(\mathfrak m\). By the multiplicativity of ramification indices (Lemma 0BRL) and the assumption, we conclude that the ramification index of \(B_\mathfrak m\) over \(A\) is prime to the residue characteristic. Since \(\kappa(\mathfrak m')/\kappa_A\) is finite separable, the same is true for \(\kappa(\mathfrak m)/\kappa_A\).
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\). Let \(\pi \in A\) be a uniformizer. Let \(L/K\) be a finite separable extension. The following are equivalent
\(L\) is tamely ramified with respect to \(A\),
there exists an \(e \geq 1\) invertible in \(\kappa_A\) and an extension \(L'/K' = K[\pi^{1/e}]\) unramified with respect to \(A' = A[\pi^{1/e}]\) such that \(L\) is contained in \(L'\), and
there exists an \(e_0 \geq 1\) invertible in \(\kappa_A\) such that for every \(d \geq 1\) invertible in \(\kappa_A\) (2) holds with \(e = de_0\).
Proof
Observe that \(A'\) is a discrete valuation ring with fraction field \(K'\), see Lemma 09EV. Of course the ramification index of \(A'\) over \(A\) is \(e\). Thus if (2) holds, then \(L'\) is tamely ramified with respect to \(A\) by Lemma 0EXU. Hence \(L\) is tamely ramified with respect to \(A\) by Lemma 0EXV.
The implication (3) \(\Rightarrow\) (2) is immediate.
Assume that (1) holds. Let \(B\) be the integral closure of \(A\) in \(L\) and let \(\mathfrak m_1, \ldots, \mathfrak m_n\) be its maximal ideals. Denote \(e_i\) the ramification index of \(A \to B_{\mathfrak m_i}\). Let \(e_0\) be the least common multiple of \(e_1, \ldots, e_r\). This is invertible in \(\kappa_A\) by our assumption (1). Let \(e = de_0\) as in (3). Set \(A' = A[\pi^{1/e}]\). Then \(A \to A'\) is an extension of discrete valuation rings with fraction field \(K' = K[\pi^{1/e}]\), see Lemma 09EV. Choose a product decomposition \[L \otimes_K K' = \prod L'_j\] where \(L'_j\) are fields. Let \(B'_j\) be the integral closure of \(A\) in \(L'_j\). Let \(\mathfrak m_{ijk}\) be the maximal ideals of \(B'_j\) lying over \(\mathfrak m_i\). Observe that \((B'_j)_{\mathfrak m_i}\) is the integral closure of \(B_{\mathfrak m_i}\) in \(L'_j\). By Abhyankar’s lemma (Lemma 0BRM) applied to \(A \subset B_{\mathfrak m_i}\) and the extension \(K'/K\) we see that \(A' \to (B'_j)_{\mathfrak m_{ijk}}\) is formally smooth in the \(\mathfrak m_{ijk}\)-adic topology. This implies that the ramification index is \(1\) and that the residue field extension is separable (Lemma 09E7). In this way we see that \(L'_j\) is unramified with respect to \(A'\). This finishes the proof: we take \(L' = L'_j\) for some \(j\).
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\).
If \(L/K\) is a finite separable extension which is tamely ramified with respect to \(A\), then there exists a Galois extension \(M/K\) containing \(L\) which is tamely ramified with respect to \(A\).
If \(L_1/K\), \(L_2/K\) are finite separable extensions which are tamely ramified with respect to \(A\), then there exists a finite separable extension \(L/K\) which is tamely ramified with respect to \(A\) containing \(L_1\) and \(L_2\).
Proof
Proof of (2). Choose a uniformizer \(\pi \in A\). We can choose an integer \(e\) invertible in \(\kappa_A\) and extensions \(L_i'/K' = K[\pi^{1/e}]\) unramified with respect to \(A' = A[\pi^{1/e}]\) with \(L'_i/L_i\) as extensions of \(K\), see Lemma 0EXW. By Lemma 0EXR we can find an extension \(L'/K'\) which is unramified with respect to \(A'\) such that \(L'_i/K\) is isomorphic to a subextension of \(L'/K'\) for \(i = 1, 2\). This finishes the proof of (3) as \(L'/K\) is tamely ramified (use same lemma as above).
Proof of (1). We may first replace \(L\) by a larger extension and assume that \(L\) is an extension of \(K' = K[\pi^{1/e}]\) unramified with respect to \(A' = A[\pi^{1/e}]\) where \(e\) is invertible in \(\kappa_A\), see Lemma 0EXW. Let \(M\) be the normal closure of \(L\) over \(K\), see Fields, Definition 0BMF. Then \(M/K\) is Galois by Fields, Lemma 0EXM. On the other hand, there is a surjection \[L \otimes_K \ldots \otimes_K L \longrightarrow M\] of \(K\)-algebras, see Fields, Lemma 0EXL. Let \(B\) be the integral closure of \(A\) in \(L\) as in Remark 09E8. The condition that \(L\) is unramified with respect to \(A' = A[\pi^{1/e}]\) exactly means that \(A' \to B\) is an étale ring map, see Algebra, Lemma 00U6. Claim: \[K' \otimes_K \ldots \otimes_K K' = \prod K'_i\] is a product of field extensions \(K'_i/K\) tamely ramified with respect to \(A\). Then if \(A'_i\) is the integral closure of \(A\) in \(K'_i\) we see that \[\prod A'_i \otimes_{(A' \otimes_A \ldots \otimes_A A')} (B \otimes_A \ldots \otimes_A B)\] is finite étale over \(\prod A'_i\) and hence a product of Dedekind domains (Lemma 0AP2). We conclude that \(M\) is the fraction field of one of these Dedekind domains which is finite étale over \(A'_i\) for some \(i\). It follows that \(M/K'_i\) is unramified with respect to every maximal ideal of \(A'_i\) and hence \(M/K\) is tamely ramified by Lemma 0EXU.
It remains the prove the claim. For this we write \(A' = A[x]/(x^e - \pi)\) and we see that \[A' \otimes_A \ldots \otimes_A A' = A'[x_1, \ldots, x_r]/(x_1^e - \pi, \ldots, x_r^e - \pi)\] The normalization of this ring certainly contains the elements \(y_i = x_i/x_1\) for \(i = 2, \ldots, r\) subject to the relations \(y_i^e - 1 = 0\) and we obtain \[A[x_1, y_2, \ldots, y_r]/(x_1^e - \pi, y_2^e - 1, \ldots, y_r^e - 1) = A'[y_2, \ldots, y_r]/(y_2^e - 1, \ldots, y_r^e - 1)\] This ring is finite étale over \(A'\) because \(e\) is invertible in \(A'\). Hence it is a product of Dedekind domains each unramified over \(A'\) as desired (see references given above in case of confusion).
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings. Denote \(L/K\) the corresponding extension of fraction fields. Let \(K'/K\) be a finite separable extension. Then \[K' \otimes_K L = \prod L'_i\] is a finite product of fields and the following is true
If \(K'\) is unramified with respect to \(A\), then each \(L'_i\) is unramified with respect to \(B\).
If \(K'\) is tamely ramified with respect to \(A\), then each \(L'_i\) is tamely ramified with respect to \(B\).
Proof
The algebra \(K' \otimes_K L\) is a finite product of fields as it is a finite étale algebra over \(L\). Let \(A'\) be the integral closure of \(A\) in \(K'\).
In case (1) the ring map \(A \to A'\) is finite étale. Hence \(B' = B \otimes_A A'\) is finite étale over \(B\) and is a finite product of Dedekind domains (Lemma 0AP2). Hence \(B'\) is the integral closure of \(B\) in \(K' \otimes_K L\). It follows immediately that each \(L'_i\) is unramified with respect to \(B\).
Choose a uniformizer \(\pi \in A\). To prove (2) we may replace \(K'\) by a larger extension tame ramified with respect to \(A\) (details omitted; hint: use Lemma 0EXV). Thus by Lemma 0EXW we may assume there exists some \(e \geq 1\) invertible in \(\kappa_A\) such that \(K'\) contains \(K[\pi^{1/e}]\) and such that \(K'\) is unramified with respect to \(A[\pi^{1/e}]\). Choose a product decomposition \[K[\pi^{1/e}] \otimes_K L = \prod L_{e, j}\] For every \(i\) there exists a \(j_i\) such that \(L'_i/L_{e, j_i}\) is a finite separable extension. Let \(B_{e, j}\) be the integral closure of \(B\) in \(L_{e, j}\). By (1) applied to \(K'/K[\pi^{1/e}]\) and \(A[\pi^{1/e}] \subset (B_{e, j_i})_\mathfrak m\) we see that \(L'_i\) is unramified with respect to \((B_{e, j_i})_\mathfrak m\) for every maximal ideal \(\mathfrak m \subset B_{e, j_i}\). Hence the proof will be complete if we can show that \(L_{e, j}\) is tamely ramified with respect to \(B\), see Lemma 0EXU.
Choose a uniformizer \(\theta\) in \(B\). Write \(\pi = u \theta^t\) where \(u\) is a unit of \(B\) and \(t \geq 1\). Then we have \[A[\pi^{1/e}] \otimes_A B = B[x]/(x^e - u \theta^t) \subset B[y, z]/(y^{e'} - \theta, z^e - u)\] where \(e' = e/\gcd(e, t)\). The map sends \(x\) to \(z y^{t/\gcd(e, t)}\). Since the right hand side is a product of Dedekind domains each tamely ramified over \(B\) the proof is complete (details omitted).
Eliminating ramification
In this section we discuss a result of Helmut Epp, see [Epp]. We strongly encourage the reader to read the original. Our approach is slightly different as we try to handle the mixed and equicharacteristic cases by the same method. For related results, see also [Ponomarev], [Ponomarev-Abhyankar], [Kuhlmann], and [ZK].
Let \(A \subset B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). The goal in this section is to find a finite extension \(K_1/K\) such that with \[\vcenter{ \xymatrix{ L \ar[r] & L_1 \\ K \ar[u] \ar[r] & K_1 \ar[u] } } \quad\text{and}\quad \vcenter{ \xymatrix{ B \ar[r] & B_1 \ar[r] & (B_1)_{\mathfrak m_{ij}} \\ A \ar[u] \ar[r] & A_1 \ar[r] \ar[u] & (A_1)_{\mathfrak m_i} \ar[u] } }\] as in Remark 09EM the extensions \((A_1)_{\mathfrak m_i} \subset (B_1)_{\mathfrak m_{ij}}\) are all weakly unramified or even formally smooth in the relevant adic topologies. The simplest (but nontrivial) example of this is Abhyankar’s lemma, see Lemma 0BRM.
Definition
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\).
We say a finite field extension \(K_1/K\) is a weak solution for \(A \subset B\) if all the extensions \((A_1)_{\mathfrak m_i} \subset (B_1)_{\mathfrak m_{ij}}\) of Remark 09EM are weakly unramified.
We say a finite field extension \(K_1/K\) is a solution for \(A \subset B\) if each extension \((A_1)_{\mathfrak m_i} \subset (B_1)_{\mathfrak m_{ij}}\) of Remark 09EM is formally smooth in the \(\mathfrak m_{ij}\)-adic topology.
We say a solution \(K_1/K\) is a separable solution if \(K_1/K\) is separable.
In general (weak) solutions do not exist; there is an example in [Epp]. Under a mild hypothesis on the residue field extension, we will prove the existence of weak solutions in Theorem 09F9 following [Epp]. In the next section, we will deduce the existence of solutions and sometimes separable solutions in geometrically meaningful cases, see Proposition 09II and Lemma 0BRP. However, the following example shows that in general one needs inseparable extensions to get even a weak solution.
Example
Let \(k\) be a perfect field of characteristic \(p > 0\). Let \(A = k[[x]]\) and \(K = k((x))\). Let \(B = A[x^{1/p}]\). Any weak solution \(K_1/K\) for \(A \to B\) is inseparable (and any finite inseparable extension of \(K\) is a solution). We omit the proof.
Solutions are stable under further extensions, see Lemma 0GLR. This may not be true for weak solutions. Weak solutions are in some sense stable under totally ramified extensions, see Lemma 09ER.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Assume that \(A \to B\) is weakly unramified. Then for any finite separable extension \(K_1/K\) totally ramified with respect to \(A\) we have that \(L_1 = L \otimes_K K_1\) is a field, \(A_1\) and \(B_1 = B \otimes_A A_1\) are discrete valuation rings, and the extension \(A_1 \subset B_1\) (see Remark 09EM) is weakly unramified.
Proof
Let \(\pi \in A\) and \(\pi_1 \in A_1\) be uniformizers. As \(K_1/K\) is totally ramified with respect to \(A\) we have \(\pi_1^e = u_1 \pi\) for some unit \(u_1\) in \(A_1\). Hence \(A_1\) is generated by \(\pi_1\) over \(A\) and the minimal polynomial \(P(t)\) of \(\pi_1\) over \(K\) has the form \[P(t) = t^e + a_{e - 1} t^{e - 1} + \ldots + a_0\] with \(a_i \in (\pi)\) and \(a_0 = u\pi\) for some unit \(u\) of \(A\). Note that \(e = [K_1 : K]\) as well. Since \(A \to B\) is weakly unramified we see that \(\pi\) is a uniformizer of \(B\) and hence \(B_1 = B[t]/(P(t))\) is a discrete valuation ring with uniformizer the class of \(t\). Thus the lemma is clear.
Lemma
Let \(A \to B \to C\) be extensions of discrete valuation rings with fraction fields \(K \subset L \subset M\). Let \(K_1/K\) be a finite extension.
If \(K_1\) is a (weak) solution for \(A \to C\), then \(K_1\) is a (weak) solution for \(A \to B\).
If \(K_1\) is a (weak) solution for \(A \to B\) and \(L_1 = (L \otimes_K K_1)_{red}\) is a product of fields which are (weak) solutions for \(B \to C\), then \(K_1\) is a (weak) solution for \(A \to C\).
Proof
Let \(L_1 = (L \otimes_K K_1)_{red}\) and \(M_1 = (M \otimes_K K_1)_{red}\) and let \(B_1 \subset L_1\) and \(C_1 \subset M_1\) be the integral closure of \(B\) and \(C\). Note that \(M_1 = (M \otimes_L L_1)_{red}\) and that \(L_1\) is a (nonempty) finite product of finite extensions of \(L\). Hence the ring map \(B_1 \to C_1\) is a finite product of ring maps of the form discussed in Remark 09EM. In particular, the map \(\Spec(C_1) \to \Spec(B_1)\) is surjective. Choose a maximal ideal \(\mathfrak m \subset C_1\) and consider the extensions of discrete valuation rings \[(A_1)_{A_1 \cap \mathfrak m} \to (B_1)_{B_1 \cap \mathfrak m} \to (C_1)_\mathfrak m\] If the composition is weakly unramified, so is the map \((A_1)_{A_1 \cap \mathfrak m} \to (B_1)_{B_1 \cap \mathfrak m}\). If the residue field extension \(\kappa_{A_1 \cap \mathfrak m} \to \kappa_\mathfrak m\) is separable, so is the subextension \(\kappa_{A_1 \cap \mathfrak m} \to \kappa_{B_1 \cap \mathfrak m}\). Taking into account Lemma 09E7 this proves (1). A similar argument works for (2).
Lemma
Let \(A \to B\) be an extension of discrete valuation rings. There exists a commutative diagram \[\xymatrix{ B \ar[r] & B' \\ A \ar[r] \ar[u] & A' \ar[u] }\] of extensions of discrete valuation rings such that
the extensions \(K'/K\) and \(L'/L\) of fraction fields are separable algebraic,
the residue fields of \(A'\) and \(B'\) are separable algebraic closures of the residue fields of \(A\) and \(B\), and
if a solution, weak solution, or separable solution exists for \(A' \to B'\), then a solution, weak solution, or separable solution exists for \(A \to B\).
Proof
By Algebra, Lemma 09E0 there exists an extension \(A \subset A'\) which is a filtered colimit of finite étale extensions such that the residue field of \(A'\) is a separable algebraic closure of the residue field of \(A\). Then \(A \subset A'\) is an extension of discrete valuation rings such that the induced extension \(K'/K\) of fraction fields is separable algebraic.
Let \(B \subset B'\) be a strict henselization of \(B\). Then \(B \subset B'\) is an extension of discrete valuation rings whose fraction field extension is separable algebraic. By Algebra, Lemma 04GT there exists a commutative diagram as in the statement of the lemma. Parts (1) and (2) of the lemma are clear.
Let \(K'_1/K'\) be a (weak) solution for \(A' \to B'\). Since \(A'\) is a colimit, we can find a finite étale extension \(A \subset A_1'\) and a finite extension \(K_1\) of the fraction field \(F\) of \(A_1'\) such that \(K'_1 = K' \otimes_F K_1\). As \(A \subset A_1'\) is finite étale and \(B'\) strictly henselian, it follows that \(B' \otimes_A A_1'\) is a finite product of rings isomorphic to \(B'\). Hence \[L' \otimes_K K_1 = L' \otimes_K F \otimes_F K_1\] is a finite product of rings isomorphic to \(L' \otimes_{K'} K'_1\). Thus we see that \(K_1/K\) is a (weak) solution for \(A \to B'\). Hence it is also a (weak) solution for \(A \to B\) by Lemma 09ES.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Let \(K_1/K\) be a normal extension. Say \(G = \text{Aut}(K_1/K)\). Then \(G\) acts on the rings \(K_1\), \(L_1\), \(A_1\) and \(B_1\) of Remark 09EM and acts transitively on the set of maximal ideals of \(B_1\).
Proof
Everything is clear apart from the last assertion. If there are two or more orbits of the action, then we can find an element \(b \in B_1\) which vanishes at all the maximal ideals of one orbit and has residue \(1\) at all the maximal ideals in another orbit. Then \(b' = \prod_{\sigma \in G} \sigma(b)\) is a \(G\)-invariant element of \(B_1 \subset L_1 = (L \otimes_K K_1)_{red}\) which is in some maximal ideals of \(B_1\) but not in all maximal ideals of \(B_1\). Lifting it to an element of \(L \otimes_K K_1\) and raising to a high power we obtain a \(G\)-invariant element \(b''\) of \(L \otimes_K K_1\) mapping to \((b')^N\) for some \(N > 0\); in fact, we only need to do this in case the characteristic is \(p > 0\) and in this case raising to a suitably large \(p\)-power \(q\) defines a canonical map \((L \otimes_K K_1)_{red} \to L \otimes_K K_1\). Since \(K = (K_1)^G\) we conclude that \(b'' \in L\). Since \(b''\) maps to an element of \(B_1\) we see that \(b'' \in B\) (as \(B\) is normal). Then on the one hand it must be true that \(b'' \in \mathfrak m_B\) as \(b'\) is in some maximal ideal of \(B_1\) and on the other hand it must be true that \(b'' \not \in \mathfrak m_B\) as \(b'\) is not in all maximal ideals of \(B_1\). This contradiction finishes the proof of the lemma.
Lemma
Let \(A\) be a discrete valuation ring with uniformizer \(\pi\). If the residue characteristic of \(A\) is \(p > 0\), then for every \(n > 1\) and \(p\)-power \(q\) there exists a degree \(q\) separable extension \(L/K\) totally ramified with respect to \(A\) such that the integral closure \(B\) of \(A\) in \(L\) has ramification index \(q\) and a uniformizer \(\pi_B\) such that \(\pi_B^q = \pi + \pi^n b\) and \(\pi_B^q = \pi + (\pi_B)^{nq}b'\) for some \(b, b' \in B\).
Proof
If the characteristic of \(K\) is zero, then we can take the extension given by \(\pi_B^q = \pi\), see Lemma 09EV. If the characteristic of \(K\) is \(p > 0\), then we can take the extension of \(K\) given by \(z^q - \pi^n z = \pi^{1 - q}\). Namely, then we see that \(y^q - \pi^{n + q - 1} y = \pi\) where \(y = \pi z\). Taking \(\pi_B = y\) we obtain the desired result.
Lemma
Let \(A\) be a discrete valuation ring. Assume the reside field \(\kappa_A\) has characteristic \(p > 0\) and that \(a \in A\) is an element whose residue class in \(\kappa_A\) is not a \(p\)th power. Then \(a\) is not a \(p\)th power in \(K\) and the integral closure of \(A\) in \(K[a^{1/p}]\) is the ring \(A[a^{1/p}]\) which is a discrete valuation ring weakly unramified over \(A\).
Proof
This lemma proves itself.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings with fractions fields \(K \subset L \subset M\). Let \(\pi \in A\) be a uniformizer. Assume
\(B\) is a Nagata ring,
\(A \subset B\) is weakly unramified,
\(M\) is a degree \(p\) purely inseparable extension of \(L\).
Then either
\(A \to C\) is weakly unramified, or
\(C = B[\pi^{1/p}]\), or
there exists a degree \(p\) separable extension \(K_1/K\) totally ramified with respect to \(A\) such that \(L_1 = L \otimes_K K_1\) and \(M_1 = M \otimes_K K_1\) are fields and the maps of integral closures \(A_1 \to B_1 \to C_1\) are weakly unramified extensions of discrete valuation rings.
Proof
Let \(e\) be the ramification index of \(C\) over \(B\). If \(e = 1\), then we are done. If not, then \(e = p\) by Lemmas 09E5 and 09E6. This in turn implies that the residue fields of \(B\) and \(C\) agree. Choose a uniformizer \(\pi_C\) of \(C\). Write \(\pi_C^p = u \pi\) for some unit \(u\) of \(C\). Since \(\pi_C^p \in L\), we see that \(u \in B^*\). Also \(M = L[\pi_C]\).
Suppose there exists an integer \(m \geq 0\) such that \[u = \sum\nolimits_{0 \leq i < m} b_i^p \pi^i + b \pi^m\] with \(b_i \in B\) and with \(b \in B\) an element whose image in \(\kappa_B\) is not a \(p\)th power. Choose an extension \(K_1/K\) as in Lemma 09EW with \(n = m + 2\) and denote \(\pi'\) the uniformizer of the integral closure \(A_1\) of \(A\) in \(K_1\) such that \(\pi = (\pi')^p + (\pi')^{np} a\) for some \(a \in A_1\). Let \(B_1\) be the integral closure of \(B\) in \(L \otimes_K K_1\). Observe that \(A_1 \to B_1\) is weakly unramified by Lemma 09ER. In \(B_1\) we have \[u \pi = \left(\sum\nolimits_{0 \leq i < m} b_i (\pi')^{i + 1}\right)^p + b (\pi')^{(m + 1)p} + (\pi')^{np} b_1\] for some \(b_1 \in B_1\) (computation omitted). We conclude that \(M_1\) is obtained from \(L_1\) by adjoining a \(p\)th root of \[b + (\pi')^{n - m - 1} b_1\] Since the residue field of \(B_1\) equals the residue field of \(B\) we see from Lemma 09EX that \(M_1/L_1\) has degree \(p\) and the integral closure \(C_1\) of \(B_1\) is weakly unramified over \(B_1\). Thus we conclude in this case.
If there does not exist an integer \(m\) as in the preceding paragraph, then \(u\) is a \(p\)th power in the \(\pi\)-adic completion of \(B_1\). Since \(B\) is Nagata, this means that \(u\) is a \(p\)th power in \(B_1\) by Algebra, Lemma 09E2. Whence the second case of the statement of the lemma holds.
Lemma
Let \(A\) be a local ring annihilated by a prime \(p\) whose maximal ideal is nilpotent. There exists a ring map \(\sigma : \kappa_A \to A\) which is a section to the residue map \(A \to \kappa_A\). If \(A \to A'\) is a local homomorphism of local rings, then we can choose a similar ring map \(\sigma' : \kappa_{A'} \to A'\) compatible with \(\sigma\) provided that the extension \(\kappa_{A'}/\kappa_A\) is separable.
Proof
Separable extensions are formally smooth by Algebra, Proposition 0322. Thus the existence of \(\sigma\) follows from the fact that \(\mathbf{F}_p \to \kappa_A\) is separable. Similarly for the existence of \(\sigma'\) compatible with \(\sigma\).
Lemma
Let \(A\) be a discrete valuation ring with fraction field \(K\) of characteristic \(p > 0\). Let \(\xi \in K\). Let \(L\) be an extension of \(K\) obtained by adjoining a root of \(z^p - z = \xi\). Then \(L/K\) is Galois and one of the following happens
\(L = K\),
\(L/K\) is unramified with respect to \(A\) of degree \(p\),
\(L/K\) is totally ramified with respect to \(A\) with ramification index \(p\), and
the integral closure \(B\) of \(A\) in \(L\) is a discrete valuation ring, \(A \subset B\) is weakly unramified, and \(A \to B\) induces a purely inseparable residue field extension of degree \(p\).
Let \(\pi\) be a uniformizer of \(A\). We have the following implications:
If \(\xi \in A\), then we are in case (1) or (2).
If \(\xi = \pi^{-n}a\) where \(n > 0\) is not divisible by \(p\) and \(a\) is a unit in \(A\), then we are in case (3)
If \(\xi = \pi^{-n} a\) where \(n > 0\) is divisible by \(p\) and the image of \(a\) in \(\kappa_A\) is not a \(p\)th power, then we are in case (4).
Proof
The extension is Galois of order dividing \(p\) by the discussion in Fields, Section 09I7. It immediately follows from the discussion in Section 09E3 that we are in one of the cases (1) – (4) listed in the lemma.
Case (A). Here we see that \(A \to A[x]/(x^p - x - \xi)\) is a finite étale ring extension. Hence we are in cases (1) or (2).
Case (B). Write \(\xi = \pi^{-n}a\) where \(p\) does not divide \(n\). Let \(B \subset L\) be the integral closure of \(A\) in \(L\). If \(C = B_\mathfrak m\) for some maximal ideal \(\mathfrak m\), then it is clear that \(p \text{ord}_C(z) = -n \text{ord}_C(\pi)\). In particular \(A \subset C\) has ramification index divisible by \(p\). It follows that it is \(p\) and that \(B = C\).
Case (C). Set \(k = n/p\). Then we can rewrite the equation as \[(\pi^kz)^p - \pi^{n - k} (\pi^kz) = a\] Since \(A[y]/(y^p - \pi^{n - k}y - a)\) is a discrete valuation ring weakly unramified over \(A\), the lemma follows.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings with fractions fields \(K \subset L \subset M\). Assume
\(A \subset B\) weakly unramified,
the characteristic of \(K\) is \(p\),
\(M\) is a degree \(p\) Galois extension of \(L\), and
\(\kappa_A = \bigcap_{n \geq 1} \kappa_B^{p^n}\).
Then there exists a finite Galois extension \(K_1/K\) totally ramified with respect to \(A\) which is a weak solution for \(A \to C\).
Proof
Since the characteristic of \(L\) is \(p\) we know that \(M\) is an Artin-Schreier extension of \(L\) (Fields, Lemma 09DY). Thus we may pick \(z \in M\), \(z \not \in L\) such that \(\xi = z^p - z \in L\). Choose \(n \geq 0\) such that \(\pi^n\xi \in B\). We pick \(z\) such that \(n\) is minimal. If \(n = 0\), then \(M/L\) is unramified with respect to \(B\) (Lemma 09F0) and we are done. Thus we have \(n > 0\).
Assumption (4) implies that \(\kappa_A\) is perfect. Thus we may choose compatible ring maps \(\overline{\sigma} : \kappa_A \to A/\pi^n A\) and \(\overline{\sigma} : \kappa_B \to B/\pi^n B\) as in Lemma 09EZ. We lift the second of these to a map of sets \(\sigma : \kappa_B \to B\)20. Then we can write \[\xi = \sum\nolimits_{i = n, \ldots, 1} \sigma(\lambda_i) \pi^{-i} + b\] for some \(\lambda_i \in \kappa_B\) and \(b \in B\). Let \[I = \{i \in \{n, \ldots, 1\} \mid \lambda_i \in \kappa_A\}\] and \[J = \{j \in \{n, \ldots, 1\} \mid \lambda_i \not \in \kappa_A\}\] We will argue by induction on the size of the finite set \(J\).
The case \(J = \emptyset\). Here for all \(i \in \{n, \ldots, 1\}\) we have \(\sigma(\lambda_i) = a_i + \pi^n b_i\) for some \(a_i \in A\) and \(b_i \in B\) by our choice of \(\sigma\). Thus \(\xi = \pi^{-n} a + b\) for some \(a \in A\) and \(b \in B\). If \(p | n\), then we write \(a = a_0^p + \pi a_1\) for some \(a_0, a_1 \in A\) (as the residue field of \(A\) is perfect). We compute \[(z - \pi^{-n/p}a_0)^p - (z - \pi^{-n/p}a_0) = \pi^{-(n - 1)}(a_1 + \pi^{n - 1 - n/p}a_0) + b'\] for some \(b' \in B\). This would contradict the minimality of \(n\). Thus \(p\) does not divide \(n\). Consider the degree \(p\) extension \(K_1\) of \(K\) given by \(w^p - w = \pi^{-n}a\). By Lemma 09F0 this extension is Galois and totally ramified with respect to \(A\). Thus \(L_1 = L \otimes_K K_1\) is a field and \(A_1 \subset B_1\) is weakly unramified (Lemma 09ER). By Lemma 09F0 the ring \(M_1 = M \otimes_K K_1\) is either a product of \(p\) copies of \(L_1\) (in which case we are done) or a field extension of \(L_1\) of degree \(p\). Moreover, in the second case, either \(C_1\) is weakly unramified over \(B_1\) (in which case we are done) or \(M_1/L_1\) is degree \(p\), Galois, and totally ramified with respect to \(B_1\). In this last case the extension \(M_1/L_1\) is generated by the element \(z - w\) and \[(z - w)^p - (z - w) = z^p - z - (w^p - w) = b\] with \(b \in B\) (see above). Thus by Lemma 09F0 once more the extension \(M_1/L_1\) is unramified with respect to \(B_1\) and we conclude that \(K_1\) is a weak solution for \(A \to C\). From now on we assume \(J \not = \emptyset\).
Suppose that \(j', j \in J\) such that \(j' = p^r j\) for some \(r > 0\). Then we change our choice of \(z\) into \[z' = z - (\sigma(\lambda_j) \pi^{-j} + \sigma(\lambda_j^p) \pi^{-pj} + \ldots + \sigma(\lambda_j^{p^{r - 1}}) \pi^{-p^{r - 1}j})\] Then \(\xi\) changes into \(\xi' = (z')^p - (z')\) as follows \[\xi' = \xi - \sigma(\lambda_j) \pi^{-j} + \sigma(\lambda_j^{p^r}) \pi^{-j'} + \text{something in }B\] Writing \(\xi' = \sum\nolimits_{i = n, \ldots, 1} \sigma(\lambda'_i) \pi^{-i} + b'\) as before we find that \(\lambda'_i = \lambda_i\) for \(i \not = j, j'\) and \(\lambda'_j = 0\). Thus the set \(J\) has gotten smaller. By induction on the size of \(J\) we may assume no such pair \(j, j'\) exists. (Please observe that in this procedure we may get thrown back into the case that \(J = \emptyset\) we treated above.)
For \(j \in J\) write \(\lambda_j = \mu_j^{p^{r_j}}\) for some \(r_j \geq 0\) and \(\mu_j \in \kappa_B\) which is not a \(p\)th power. This is possible by our assumption (4). Let \(j \in J\) be the unique index such that \(j p^{-r_j}\) is maximal. (The index is unique by the result of the preceding paragraph.) Choose \(r > \max(r_j + 1)\) and such that \(j p^{r - r_j} > n\) for \(j \in J\). Choose a separable extension \(K_1/K\) totally ramified with respect to \(A\) of degree \(p^r\) such that the corresponding discrete valuation ring \(A_1 \subset K_1\) has uniformizer \(\pi'\) with \((\pi')^{p^r} = \pi + \pi^{n + 1}a\) for some \(a \in A_1\) (Lemma 09EW). Observe that \(L_1 = L \otimes_K K_1\) is a field and that \(L_1/L\) is totally ramified with respect to \(B\) (Lemma 09ER). Computing in the integral closure \(B_1\) we get \[\xi = \sum\nolimits_{i \in I} \sigma(\lambda_i) (\pi')^{-i p^r} + \sum\nolimits_{j \in J} \sigma(\mu_j)^{p^{r_j}} (\pi')^{-j p^r} + b_1\] for some \(b_1 \in B_1\). Note that \(\sigma(\lambda_i)\) for \(i \in I\) is a \(q\)th power modulo \(\pi^n\), i.e., modulo \((\pi')^{n p^r}\). Hence we can rewrite the above as \[\xi = \sum\nolimits_{i \in I} x_i^{p^r} (\pi')^{-i p^r} + \sum\nolimits_{j \in J} \sigma(\mu_j)^{p^{r_j}} (\pi')^{- j p^r} + b_1\] As in the previous paragraph we change our choice of \(z\) into \[\begin{align*} z' & = z \\ & - \sum\nolimits_{i \in I} \left(x_i (\pi')^{-i} + \ldots + x_i^{p^{r - 1}} (\pi')^{-i p^{r - 1}}\right) \\ & - \sum\nolimits_{j \in J} \left( \sigma(\mu_j) (\pi')^{- j p^{r - r_j}} + \ldots + \sigma(\mu_j)^{p^{r_j - 1}} (\pi')^{- j p^{r - 1}} \right) \end{align*}\] to obtain \[(z')^p - z' = \sum\nolimits_{i \in I} x_i (\pi')^{-i} + \sum\nolimits_{j \in J} \sigma(\mu_j) (\pi')^{- j p^{r - r_j}} + b_1'\] for some \(b'_1 \in B_1\). Since there is a unique \(j\) such that \(j p^{r - r_j}\) is maximal and since \(j p^{r - r_j}\) is bigger than \(i \in I\) and divisible by \(p\), we see that \(M_1 / L_1\) falls into case (C) of Lemma 09F0. This finishes the proof.
Lemma
Let \(A\) be a ring which contains a primitive \(p\)th root of unity \(\zeta\). Set \(w = 1 - \zeta\). Then \[P(z) = \frac{(1 + wz)^p - 1}{w^p} = z^p - z + \sum\nolimits_{0 < i < p} a_i z^i\] is an element of \(A[z]\) and in fact \(a_i \in (w)\). Moreover, we have \[P(z_1 + z_2 + w z_1 z_2) = P(z_1) + P(z_2) + w^p P(z_1) P(z_2)\] in the polynomial ring \(A[z_1, z_2]\).
Proof
It suffices to prove this when \[A = \mathbf{Z}[\zeta] = \mathbf{Z}[x]/(x^{p - 1} + \ldots + x + 1)\] is the ring of integers of the cyclotomic field. The polynomial identity \(t^p - 1 = (t - 1)(t - \zeta) \ldots (t - \zeta^{p - 1})\) (which is proved by looking at the roots on both sides) shows that \(t^{p - 1} + \ldots + t + 1 = (t - \zeta) \ldots (t - \zeta^{p - 1})\). Substituting \(t = 1\) we obtain \(p = (1 - \zeta)(1 - \zeta^2) \ldots (1 - \zeta^{p - 1})\). The maximal ideal \((p, w) = (w)\) is the unique prime ideal of \(A\) lying over \(p\) (as fields of characteristic \(p\) do not have nontrivial \(p\)th roots of \(1\)). It follows that \(p = u w^{p - 1}\) for some unit \(u\). This implies that \[a_i = \frac{1}{p} {p \choose i} u w^{i - 1}\] for \(p > i > 1\) and \(- 1 + a_1 = pw/w^p = u\). Since \(P(-1) = 0\) we see that \(0 = (-1)^p - u\) modulo \((w)\). Hence \(a_1 \in (w)\) and the proof if the first part is done. The second part follows from a direct computation we omit.
Lemma
Let \(A\) be a discrete valuation ring of mixed characteristic \((0, p)\) which contains a primitive \(p\)th root of \(1\). Let \(P(t) \in A[t]\) be the polynomial of Lemma 09F2. Let \(\xi \in K\). Let \(L\) be an extension of \(K\) obtained by adjoining a root of \(P(z) = \xi\). Then \(L/K\) is Galois and one of the following happens
\(L = K\),
\(L/K\) is unramified with respect to \(A\) of degree \(p\),
\(L/K\) is totally ramified with respect to \(A\) with ramification index \(p\), and
the integral closure \(B\) of \(A\) in \(L\) is a discrete valuation ring, \(A \subset B\) is weakly unramified, and \(A \to B\) induces a purely inseparable residue field extension of degree \(p\).
Let \(\pi\) be a uniformizer of \(A\). We have the following implications:
If \(\xi \in A\), then we are in case (1) or (2).
If \(\xi = \pi^{-n}a\) where \(n > 0\) is not divisible by \(p\) and \(a\) is a unit in \(A\), then we are in case (3)
If \(\xi = \pi^{-n} a\) where \(n > 0\) is divisible by \(p\) and the image of \(a\) in \(\kappa_A\) is not a \(p\)th power, then we are in case (4).
Proof
Adjoining a root of \(P(z) = \xi\) is the same thing as adjoining a root of \(y^p = w^p(1 + \xi)\). Since \(K\) contains a primitive \(p\)th root of \(1\) the extension is Galois of order dividing \(p\) by the discussion in Fields, Section 09I6. It immediately follows from the discussion in Section 09E3 that we are in one of the cases (1) – (4) listed in the lemma.
Case (A). Here we see that \(A \to A[x]/(P(x) - \xi)\) is a finite étale ring extension. Hence we are in cases (1) or (2).
Case (B). Write \(\xi = \pi^{-n}a\) where \(p\) does not divide \(n\). Let \(B \subset L\) be the integral closure of \(A\) in \(L\). If \(C = B_\mathfrak m\) for some maximal ideal \(\mathfrak m\), then it is clear that \(p \text{ord}_C(z) = -n \text{ord}_C(\pi)\). In particular \(A \subset C\) has ramification index divisible by \(p\). It follows that it is \(p\) and that \(B = C\).
Case (C). Set \(k = n/p\). Then we can rewrite the equation as \[(\pi^kz)^p - \pi^{n - k} (\pi^kz) + \sum a_i \pi^{n - ik} (\pi^kz)^i = a\] Since \(A[y]/(y^p - \pi^{n - k}y - \sum a_i \pi^{n - ik} y^i - a)\) is a discrete valuation ring weakly unramified over \(A\), the lemma follows.
Let \(A\) be a discrete valuation ring of mixed characteristic \((0, p)\) containing a primitive \(p\)th root of \(1\). Let \(w \in A\) and \(P(t) \in A[t]\) be as in Lemma 09F2. Let \(L\) be a finite extension of \(K\). We say \(L/K\) is a degree \(p\) extension of finite level if \(L\) is a degree \(p\) extension of \(K\) obtained by adjoining a root of the equation \(P(z) = \xi\) where \(\xi \in K\) is an element with \(w^p \xi \in \mathfrak m_A\).
This definition is relevant to the discussion in this section due to the following straightforward lemma.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings with fractions fields \(K \subset L \subset M\). Assume that
\(A\) has mixed characteristic \((0, p)\),
\(A \subset B\) is weakly unramified,
\(B\) contains a primitive \(p\)th root of \(1\), and
\(M/L\) is Galois of degree \(p\).
Then there exists a finite Galois extension \(K_1/K\) totally ramified with respect to \(A\) which is either a weak solution for \(A \to C\) or is such that \(M_1/L_1\) is a degree \(p\) extension of finite level.
Proof
Let \(\pi \in A\) be a uniformizer. By Kummer theory (Fields, Lemma 09DX) \(M\) is obtained from \(L\) by adjoining the root of \(y^p = b\) for some \(b \in L\).
If \(\text{ord}_B(b)\) is prime to \(p\), then we choose a degree \(p\) separable extension \(K_1/K\) totally ramified with respect to \(A\) (for example using Lemma 09EW). Let \(A_1\) be the integral closure of \(A\) in \(K_1\). By Lemma 09ER the integral closure \(B_1\) of \(B\) in \(L_1 = L \otimes_K K_1\) is a discrete valuation ring weakly unramified over \(A_1\). If \(K_1/K\) is not a weak solution for \(A \to C\), then the integral closure \(C_1\) of \(C\) in \(M_1 = M \otimes_K K_1\) is a discrete valuation ring and \(B_1 \to C_1\) has ramification index \(p\). In this case, the field \(M_1\) is obtained from \(L_1\) by adjoining the \(p\)th root of \(b\) with \(\text{ord}_{B_1}(b)\) divisible by \(p\). Replacing \(A\) by \(A_1\), etc we may assume that \(b = \pi^n u\) where \(u \in B\) is a unit and \(n\) is divisible by \(p\). Of course, in this case the extension \(M\) is obtained from \(L\) by adjoining the \(p\)th root of a unit.
Suppose \(M\) is obtained from \(L\) by adjoining the root of \(y^p = u\) for some unit \(u\) of \(B\). If the residue class of \(u\) in \(\kappa_B\) is not a \(p\)th power, then \(B \subset C\) is weakly unramified (Lemma 09EX) and we are done. Otherwise, we can replace our choice of \(y\) by \(y/v\) where \(v^p\) and \(u\) have the same image in \(\kappa_B\). After such a replacement we have \[y^p = 1 + \pi b\] for some \(b \in B\). Then we see that \(P(z) = \pi b/ w^p\) where \(z = (y - 1)/w\). Thus we see that the extension is a degree \(p\) extension of finite level with \(\xi = \pi b / w^p\).
Let \(A\) be a discrete valuation ring of mixed characteristic \((0, p)\) containing a primitive \(p\)th root of \(1\). Let \(w \in A\) and \(P(t) \in A[t]\) be as in Lemma 09F2. Let \(L\) be a degree \(p\) extension of \(K\) of finite level. Choose \(z \in L\) generating \(L\) over \(K\) with \(\xi = P(z) \in K\). Choose a uniformizer \(\pi\) for \(A\) and write \(w = u \pi^{e_1}\) for some integer \(e_1 = \text{ord}_A(w)\) and unit \(u \in A\). Finally, pick \(n \geq 0\) such that \[\pi^n \xi \in A\] The level of \(L/K\) is the smallest value of the quantity \(n/e_1\) taking over all \(z\) generating \(L/K\) with \(\xi = P(z) \in K\).
We make a couple of remarks. Since the extension is of finite level we know that we can choose \(z\) such that \(n < pe_1\). Thus the level is a rational number contained in \([0, p)\). If the level is zero then \(L/K\) is unramified with respect to \(A\) by Lemma 09F3. Our next goal is to lower the level.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings with fractions fields \(K \subset L \subset M\). Assume
\(A\) has mixed characteristic \((0, p)\),
\(A \subset B\) weakly unramified,
\(B\) contains a primitive \(p\)th root of \(1\),
\(M/L\) is a degree \(p\) extension of finite level \(l > 0\),
\(\kappa_A = \bigcap_{n \geq 1} \kappa_B^{p^n}\).
Then there exists a finite separable extension \(K_1\) of \(K\) totally ramified with respect to \(A\) such that either \(K_1\) is a weak solution for \(A \to C\), or the extension \(M_1/L_1\) is a degree \(p\) extension of finite level \(\leq \max(0, l - 1, 2l - p)\).
Proof
Let \(\pi \in A\) be a uniformizer. Let \(w \in B\) and \(P \in B[t]\) be as in Lemma 09F2 (for \(B\)). Set \(e_1 = \text{ord}_B(w)\), so that \(w\) and \(\pi^{e_1}\) are associates in \(B\). Pick \(z \in M\) generating \(M\) over \(L\) with \(\xi = P(z) \in K\) and \(n\) such that \(\pi^n\xi \in B\) as in the definition of the level of \(M\) over \(L\), i.e., \(l = n/e_1\).
The proof of this lemma is completely similar to the proof of Lemma 09F1. To explain what is going on, observe that [09F6]\[\begin{equation} P(z) \equiv z^p - z \bmod \pi^{-n + e_1}B \end{equation}\] for any \(z \in L\) such that \(\pi^{-n} P(z) \in B\) (use that \(z\) has valuation at worst \(-n/p\) and the shape of the polynomial \(P\)). Moreover, we have [09F7]\[\begin{equation} \xi_1 + \xi_2 + w^p \xi_1 \xi_2 \equiv \xi_1 + \xi_2 \bmod \pi^{-2n + pe_1}B \end{equation}\] for \(\xi_1, \xi_2 \in \pi^{-n}B\). Finally, observe that \(n - e_1 = (l - 1)/e_1\) and \(-2n + pe_1 = -(2l - p)e_1\). Write \(m = n - e_1 \max(0, l - 1, 2l - p)\). The above shows that doing calculations in \(\pi^{-n}B / \pi^{-n + m}B\) the polynomial \(P\) behaves exactly as the polynomial \(z^p - z\). This explains why the lemma is true but we also give the details below.
Assumption (4) implies that \(\kappa_A\) is perfect. Observe that \(m \leq e_1\) and hence \(A/\pi^m\) is annihilated by \(w\) and hence \(p\). Thus we may choose compatible ring maps \(\overline{\sigma} : \kappa_A \to A/\pi^mA\) and \(\overline{\sigma} : \kappa_B \to B/\pi^mB\) as in Lemma 09EZ. We lift the second of these to a map of sets \(\sigma : \kappa_B \to B\). Then we can write \[\xi = \sum\nolimits_{i = n, \ldots, n - m + 1} \sigma(\lambda_i) \pi^{-i} + \pi^{-n + m)} b\] for some \(\lambda_i \in \kappa_B\) and \(b \in B\). Let \[I = \{i \in \{n, \ldots, n - m + 1\} \mid \lambda_i \in \kappa_A\}\] and \[J = \{j \in \{n, \ldots, n - m + 1\} \mid \lambda_i \not \in \kappa_A\}\] We will argue by induction on the size of the finite set \(J\).
The case \(J = \emptyset\). Here for all \(i \in \{n, \ldots, n - m + 1\}\) we have \(\sigma(\lambda_i) = a_i + \pi^{n - m}b_i\) for some \(a_i \in A\) and \(b_i \in B\) by our choice of \(\overline{\sigma}\). Thus \(\xi = \pi^{-n} a + \pi^{-n + m} b\) for some \(a \in A\) and \(b \in B\). If \(p | n\), then we write \(a = a_0^p + \pi a_1\) for some \(a_0, a_1 \in A\) (as the residue field of \(A\) is perfect). Set \(z_1 = - \pi^{-n/p} a_0\). Note that \(P(z_1) \in \pi^{-n}B\) and that \(z + z_1 + w z z_1\) is an element generating \(M\) over \(L\) (note that \(wz_1 \not = -1\) as \(n < pe_1\)). Moreover, by Lemma 09F2 we have \[P(z + z_1 + w z z_1) = P(z) + P(z_1) + w^p P(z) P(z_1) \in K\] and by equations (09F6) and (09F7) we have \[P(z) + P(z_1) + w^p P(z) P(z_1) \equiv \xi + z_1^p - z_1 \bmod \pi^{-n + m}B\] for some \(b' \in B\). This contradict the minimality of \(n\)! Thus \(p\) does not divide \(n\). Consider the degree \(p\) extension \(K_1\) of \(K\) given by \(P(y) = -\pi^{-n}a\). By Lemma 09F3 this extension is separable and totally ramified with respect to \(A\). Thus \(L_1 = L \otimes_K K_1\) is a field and \(A_1 \subset B_1\) is weakly unramified (Lemma 09ER). By Lemma 09F3 the ring \(M_1 = M \otimes_K K_1\) is either a product of \(p\) copies of \(L_1\) (in which case we are done) or a field extension of \(L_1\) of degree \(p\). Moreover, in the second case, either \(C_1\) is weakly unramified over \(B_1\) (in which case we are done) or \(M_1/L_1\) is degree \(p\), Galois, totally ramified with respect to \(B_1\). In this last case the extension \(M_1/L_1\) is generated by the element \(z + y + wzy\) and we see that \(P(z + y + wzy) \in L_1\) and \[\begin{align*} P(z + y + wzy) & = P(z) + P(y) + w^p P(z) P(y) \\ & \equiv \xi - \pi^{-n}a \bmod \pi^{-n + m}B_1 \\ & \equiv 0 \bmod \pi^{-n + m}B_1 \end{align*}\] in exactly the same manner as above. By our choice of \(m\) this means exactly that \(M_1/L_1\) has level at most \(\max(0, l - 1, 2l - p)\). From now on we assume that \(J \not = \emptyset\).
Suppose that \(j', j \in J\) such that \(j' = p^r j\) for some \(r > 0\). Then we set \[z_1 = - \sigma(\lambda_j) \pi^{-j} - \sigma(\lambda_j^p) \pi^{-pj} - \ldots - \sigma(\lambda_j^{p^{r - 1}}) \pi^{-p^{r - 1}j}\] and we change \(z\) into \(z' = z + z_1 + wzz_1\). Observe that \(z' \in M\) generates \(M\) over \(L\) and that we have \(\xi' = P(z') = P(z) + P(z_1) + wP(z)P(z_1) \in L\) with \[\xi' \equiv \xi - \sigma(\lambda_j) \pi^{-j} + \sigma(\lambda_j^{p^r}) \pi^{-j'} \bmod \pi^{-n + m}B\] by using equations (09F6) and (09F7) as above. Writing \[\xi' = \sum\nolimits_{i = n, \ldots, n - m + 1} \sigma(\lambda'_i) \pi^{-i} + \pi^{-n + m}b'\] as before we find that \(\lambda'_i = \lambda_i\) for \(i \not = j, j'\) and \(\lambda'_j = 0\). Thus the set \(J\) has gotten smaller. By induction on the size of \(J\) we may assume there is no pair \(j, j'\) of \(J\) such that \(j'/j\) is a power of \(p\). (Please observe that in this procedure we may get thrown back into the case that \(J = \emptyset\) we treated above.)
For \(j \in J\) write \(\lambda_j = \mu_j^{p^{r_j}}\) for some \(r_j \geq 0\) and \(\mu_j \in \kappa_B\) which is not a \(p\)th power. This is possible by our assumption (4). Let \(j \in J\) be the unique index such that \(j p^{-r_j}\) is maximal. (The index is unique by the result of the preceding paragraph.) Choose \(r > \max(r_j + 1)\) and such that \(j p^{r - r_j} > n\) for \(j \in J\). Let \(K_1/K\) be the extension of degree \(p^r\), totally ramified with respect to \(A\), defined by \((\pi')^{p^r} = \pi\). Observe that \(\pi'\) is the uniformizer of the corresponding discrete valuation ring \(A_1 \subset K_1\). Observe that \(L_1 = L \otimes_K K_1\) is a field and \(L_1/L\) is totally ramified with respect to \(B\) (Lemma 09ER). Computing in the integral closure \(B_1\) we get \[\xi = \sum\nolimits_{i \in I} \sigma(\lambda_i) (\pi')^{-i p^r} + \sum\nolimits_{j \in J} \sigma(\mu_j)^{p^{r_j}} (\pi')^{-j p^r} + \pi^{-n + m} b_1\] for some \(b_1 \in B_1\). Note that \(\sigma(\lambda_i)\) for \(i \in I\) is a \(q\)th power modulo \(\pi^m\), i.e., modulo \((\pi')^{m p^r}\). Hence we can rewrite the above as \[\xi = \sum\nolimits_{i \in I} x_i^{p^r} (\pi')^{-i p^r} + \sum\nolimits_{j \in J} \sigma(\mu_j)^{p^{r_j}} (\pi')^{- j p^r} + \pi^{-n + m}b_1\] Similar to our choice in the previous paragraph we set \[\begin{align*} z_1 & - \sum\nolimits_{i \in I} \left(x_i (\pi')^{-i} + \ldots + x_i^{p^{r - 1}} (\pi')^{-i p^{r - 1}}\right) \\ & - \sum\nolimits_{j \in J} \left( \sigma(\mu_j) (\pi')^{- j p^{r - r_j}} + \ldots + \sigma(\mu_j)^{p^{r_j - 1}} (\pi')^{- j p^{r - 1}} \right) \end{align*}\] and we change our choice of \(z\) into \(z' = z + z_1 + wzz_1\). Then \(z'\) generates \(M_1\) over \(L_1\) and \(\xi' = P(z') = P(z) + P(z_1) + w^p P(z) P(z_1) \in L_1\) and a calculation shows that \[\xi' \equiv \sum\nolimits_{i \in I} x_i (\pi')^{-i} + \sum\nolimits_{j \in J} \sigma(\mu_j) (\pi')^{- j p^{r - r_j}} + (\pi')^{(-n + m)p^r}b'_1\] for some \(b'_1 \in B_1\). There is a unique \(j\) such that \(j p^{r - r_j}\) is maximal and \(j p^{r - r_j}\) is bigger than \(i \in I\). If \(j p^{r - r_j} \leq (n - m)p^r\) then the level of the extension \(M_1/L_1\) is less than \(\max(0, l - 1, 2l - p)\). If not, then, as \(p\) divides \(j p^{r - r_j}\), we see that \(M_1 / L_1\) falls into case (C) of Lemma 09F3. This finishes the proof.
Lemma
Let \(A \subset B \subset C\) be extensions of discrete valuation rings with fraction fields \(K \subset L \subset M\). Assume
the residue field \(k\) of \(A\) is algebraically closed of characteristic \(p > 0\),
\(A\) and \(B\) are complete,
\(A \to B\) is weakly unramified,
\(M\) is a finite extension of \(L\),
\(k = \bigcap\nolimits_{n \geq 1} \kappa_B^{p^n}\)
Then there exists a finite extension \(K_1/K\) which is a weak solution for \(A \to C\).
Proof
Let \(M'\) be any finite extension of \(L\) and consider the integral closure \(C'\) of \(B\) in \(M'\). Then \(C'\) is finite over \(B\) as \(B\) is Nagata by Algebra, Lemma 032W. Moreover, \(C'\) is a discrete valuation ring, see discussion in Remark 09EM. Moreover \(C'\) is complete as a \(B\)-module, hence complete as a discrete valuation ring, see Algebra, Section 00M9. It follows in particular that \(C\) is the integral closure of \(B\) in \(M\) (by definition of valuation rings as maximal for the relation of domination).
Let \(M \subset M'\) be a finite extension and let \(C' \subset M'\) be the integral closure of \(B\) as above. By Lemma 09ES it suffices to prove the result for \(A \to B \to C'\). Hence we may assume that \(M/L\) is normal, see Fields, Lemma 09DT.
If \(M / L\) is normal, we can find a chain of finite extensions \[L = L^0 \subset L^1 \subset L^2 \subset \ldots \subset L^r = M\] such that each extension \(L^{j + 1}/L^j\) is either:
purely inseparable of degree \(p\),
totally ramified with respect to \(B^j\) and Galois of degree \(p\),
totally ramified with respect to \(B^j\) and Galois cyclic of order prime to \(p\),
Galois and unramified with respect to \(B^j\).
Here \(B^j\) is the integral closure of \(B\) in \(L^j\). Namely, since \(M/L\) is normal we can write it as a compositum of a Galois extension and a purely inseparable extension (Fields, Lemma 030M). For the purely inseparable extension the existence of the filtration is clear. In the Galois case, note that \(G\) is “the” decomposition group and let \(I \subset G\) be the inertia group. Then on the one hand \(I\) is solvable by Lemma 09EE and on the other hand the extension \(M^I/L\) is unramified with respect to \(B\) by Lemma 09EH. This proves we have a filtration as stated.
We are going to argue by induction on the integer \(r\). Suppose that we can find a finite extension \(K_1/K\) which is a weak solution for \(A \to B^1\) where \(B^1\) is the integral closure of \(B\) in \(L^1\). Let \(K'_1\) be the normal closure of \(K_1/K\) (Fields, Lemma 09DT). Since \(A\) is complete and the residue field of \(A\) is algebraically closed we see that \(K'_1/K_1\) is separable and totally ramified with respect to \(A_1\) (some details omitted). Hence \(K'_1/K\) is a weak solution for \(A \to B^1\) as well by Lemma 09ER. In other words, we may and do assume that \(K_1\) is a normal extension of \(K\). Having done so we consider the sequence \[L^0_1 = (L^0 \otimes_K K_1)_{red} \subset L^1_1 = (L^1 \otimes_K K_1)_{red} \subset \ldots \subset L^r_1 = (L^r \otimes_K K_1)_{red}\] and the corresponding integral closures \(B^i_1\). Note that \(C_1 = B^r_1\) is a product of discrete valuation rings which are transitively permuted by \(G = \text{Aut}(K_1/K)\) by Lemma 09EU. In particular all the extensions of discrete valuation rings \(A_1 \to (C_1)_\mathfrak m\) are isomorphic and a weak solution for one will be a weak solution for all of them. We can apply the induction hypothesis to the sequence \[A_1 \to (B^1_1)_{B^1_1 \cap \mathfrak m} \to (B^2_1)_{B^2_1 \cap \mathfrak m} \to \ldots \to (B^r_1)_{B^r_1 \cap \mathfrak m} = (C_1)_\mathfrak m\] to get a weak solution \(K_2/K_1\) for \(A_1 \to (C_1)_\mathfrak m\). The extension \(K_2/K\) will then be a weak solution for \(A \to C\) by what we said before. Note that the induction hypothesis applies: the ring map \(A_1 \to (B^1_1)_{B^1_1 \cap \mathfrak m}\) is weakly unramified by our choice of \(K_1\) and the sequence of fraction field extensions each still have one of the properties (a), (b), (c), or (d) listed above. Moreover, observe that for any finite extension \(\kappa_B \subset \kappa\) we still have \(k = \bigcap \kappa^{p^n}\).
Thus everything boils down to finding a weak solution for \(A \subset C\) when the field extension \(M/L\) satisfies one of the properties (a), (b), (c), or (d).
Case (d). This case is trivial as here \(B \to C\) is unramified already.
Case (c). Say \(M/L\) is cyclic of order \(n\) prime to \(p\). Because \(M/L\) is totally ramified with respect to \(B\) we see that the ramification index of \(B \subset C\) is \(n\) and hence the ramification index of \(A \subset C\) is \(n\) as well. Choose a uniformizer \(\pi \in A\) and set \(K_1 = K[\pi^{1/n}]\). Then \(K_1/K\) is a solution for \(A \subset C\) by Abhyankar’s lemma (Lemma 0BRM).
Case (b). We divide this case into the mixed characteristic case and the equicharacteristic case. In the equicharacteristic case this is Lemma 09F1. In the mixed characteristic case, we first replace \(K\) by a finite extension to get to the situation where \(M/L\) is a degree \(p\) extension of finite level using Lemma 09F4. Then the level is a rational number \(l \in [0, p)\), see discussion preceding Lemma 09F5. If the level is \(0\), then \(B \to C\) is weakly unramified and we’re done. If not, then we can replacing the field \(K\) by a finite extension to obtain a new situation with level \(l' \leq \max(0, l - 1, 2l - p)\) by Lemma 09F5. If \(l = p - \epsilon\) for \(\epsilon < 1\) then we see that \(l' \leq p - 2\epsilon\). Hence after a finite number of replacements we obtain a case with level \(\leq p - 1\). Then after at most \(p - 1\) more such replacements we reach the situation where the level is zero.
Case (a) is Lemma 09EY. This is the only case where we possibly need a purely inseparable extension of \(K\), namely, in case (2) of the statement of the lemma we win by adjoining a \(p\)th power of the element \(\pi\). This finishes the proof of the lemma.
At this point we have collected all the lemmas we need to prove the main result of this section.
Theorem
Let \(A \subset B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). If the characteristic of \(\kappa_A\) is \(p > 0\), assume that every element of \[\bigcap\nolimits_{n \geq 1} \kappa_B^{p^n}\] is separable algebraic over \(\kappa_A\). Then there exists a finite extension \(K_1/K\) which is a weak solution for \(A \to B\) as defined in Definition 09EN.
Proof
If the characteristic of \(\kappa_A\) is zero or if the residue characteristic is \(p\), the ramification index is prime to \(p\), and the residue field extension is separable, then this follows from Abhyankar’s lemma (Lemma 0BRM). Namely, suppose the ramification index is \(e\). Choose a uniformizer \(\pi \in A\). Let \(K_1/K\) be the extension obtained by adjoining an \(e\)th root of \(\pi\). By Lemma 09EV we see that the integral closure \(A_1\) of \(A\) in \(K_1\) is a discrete valuation ring with ramification index over \(A\). Thus \(A_1 \to (B_1)_\mathfrak m\) is formally smooth in the \(\mathfrak m\)-adic topology for all maximal ideals \(\mathfrak m\) of \(B_1\) by Lemma 0BRM and a fortiori these are weakly unramified extensions of discrete valuation rings.
From now on we let \(p\) be a prime number and we assume that \(\kappa_A\) has characteristic \(p\). We first apply Lemma 09ET to reduce to the case that \(A\) and \(B\) have separably closed residue fields. Since \(\kappa_A\) and \(\kappa_B\) are replaced by their separable algebraic closures by this procedure we see that we obtain \[\kappa_A \supset \bigcap\nolimits_{n \geq 1} \kappa_B^{p^n}\] from the condition of the theorem.
Let \(\pi \in A\) be a uniformizer. Let \(A^\wedge\) and \(B^\wedge\) be the completions of \(A\) and \(B\). We have a commutative diagram \[\xymatrix{ B \ar[r] & B^\wedge \\ A \ar[u] \ar[r] & A^\wedge \ar[u] }\] of extensions of discrete valuation rings. Let \(K^\wedge\) be the fraction field of \(A^\wedge\). Suppose that we can find a finite extension \(M/K^\wedge\) which is (a) a weak solution for \(A^\wedge \to B^\wedge\) and (b) a compositum of a separable extension and an extension obtained by adjoining a \(p\)-power root of \(\pi\). Then by Lemma 09EJ we can find a finite extension \(K_1/K\) such that \(K^\wedge \otimes_K K_1 = M\). Let \(A_1\), resp. \(A_1^\wedge\) be the integral closure of \(A\), resp. \(A^\wedge\) in \(K_1\), resp. \(M\). Since \(A \to A^\wedge\) is formally smooth in the \(\mathfrak m^\wedge\)-adic topology (Lemma 09E7) we see that \(A_1 \to A_1^\wedge\) is formally smooth in the \(\mathfrak m_1^\wedge\)-adic topology (Lemma 09EQ and \(A_1\) and \(A_1^\wedge\) are discrete valuation rings by discussion in Remark 09EM). We conclude from Lemma 09ES part (2) that \(K_1/K\) is a weak solution for \(A \to B^\wedge\). Applying Lemma 09ES part (1) we see that \(K_1/K\) is a weak solution for \(A \to B\).
Thus we may assume \(A\) and \(B\) are complete discrete valuation rings with separably closed residue fields of characteristic \(p\) and with \(\kappa_A \supset \bigcap\nolimits_{n \geq 1} \kappa_B^{p^n}\). We are also given a uniformizer \(\pi \in A\) and we have to find a weak solution for \(A \to B\) which is a compositum of a separable extension and a field obtained by taking \(p\)-power roots of \(\pi\). Note that the second condition is automatic if \(A\) has mixed characteristic.
Set \(k = \bigcap\nolimits_{n \geq 1} \kappa_B^{p^n}\). Observe that \(k\) is an algebraically closed field of characteristic \(p\). If \(A\) has mixed characteristic let \(\Lambda\) be a Cohen ring for \(k\) and in the equicharacteristic case set \(\Lambda = k[[t]]\). We can choose a ring map \(\Lambda \to A\) which maps \(t\) to \(\pi\) in the equicharacteristic case. In the equicharacteristic case this follows from the Cohen structure theorem (Algebra, Theorem 032A) and in the mixed characteristic case this follows as \(\mathbf{Z}_p \to \Lambda\) is formally smooth in the adic topology (Lemmas 09E7 and 07NJ). Applying Lemma 09ES we see that it suffices to prove the existence of a weak solution for \(\Lambda \to B\) which in the equicharacteristic \(p\) case is a compositum of a separable extension and a field obtained by taking \(p\)-power roots of \(t\). However, since \(\Lambda = k[[t]]\) in the equicharacteristic case and any extension of \(k((t))\) is such a compositum, we can now drop this requirement!
Thus we arrive at the situation where \(A\) and \(B\) are complete, the residue field \(k\) of \(A\) is algebraically closed of characteristic \(p > 0\), we have \(k = \bigcap \kappa_B^{p^n}\), and in the mixed characteristic case \(p\) is a uniformizer of \(A\) (i.e., \(A\) is a Cohen ring for \(k\)). If \(A\) has mixed characteristic choose a Cohen ring \(\Lambda\) for \(\kappa_B\) and in the equicharacteristic case set \(\Lambda = \kappa_B[[t]]\). Arguing as above we may choose a ring map \(A \to \Lambda\) lifting \(k \to \kappa_B\) and mapping a uniformizer to a uniformizer. Since \(k \subset \kappa_B\) is separable the ring map \(A \to \Lambda\) is formally smooth in the adic topology (Lemma 09E7). Hence we can find a ring map \(\Lambda \to B\) such that the composition \(A \to \Lambda \to B\) is the given ring map \(A \to B\) (see Lemma 07NJ). Since \(\Lambda\) and \(B\) are complete discrete valuation rings with the same residue field, \(B\) is finite over \(\Lambda\) (Algebra, Lemma 031D). This reduces us to the special case discussed in Lemma 09F8.
Eliminating ramification, II
In this section we use the results of Section 09EL to obtain (separable) solutions in some cases.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). If \(K_1/K\) is a solution for \(A \subset B\), then for any finite extension \(K_2/K_1\) the extension \(K_2/K\) is a solution for \(A \subset B\).
Proof
This follows from Lemma 09EQ. Details omitted.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings. If \(B\) is Nagata and the extension \(L/K\) of fraction fields is separable, then \(A\) is Nagata.
Proof
A discrete valuation ring is Nagata if and only if it is N-2. Let \(K_1/K\) be a finite purely inseparable field extension. We have to show that the integral closure \(A_1\) of \(A\) in \(K_1\) is finite over \(A\), see Algebra, Lemma 032N. Since \(L/K\) is separable and \(K_1/K\) is purely inseparable, the algebra \(L \otimes_K K_1\) is a field (by Algebra, Lemmas 030U and 0BRD). Let \(B_1\) be the integral closure of \(B\) in \(L \otimes_K K_1\). Since \(B\) is Nagata, \(B_1\) is finite over \(B\). Since \(B \otimes_A A_1 \subset B_1\) and \(B\) is Noetherian, we see that \(B \otimes_A A_1\) is finite over \(B\). As \(A \to B\) is faithfully flat, this implies \(A_1\) is finite over \(A\), see Algebra, Lemma 03C4.
Lemma
Let \(A' \subset A\) be an extension of rings. Let \(f \in A'\). Assume that (a) \(A\) is finite over \(A'\), (b) \(f\) is a nonzerodivisor on \(A\), and (c) \(A'_f = A_f\). Then there exists an integer \(n_0 > 0\) such that for all \(n \geq n_0\) the following is true: given a ring \(B'\), a nonzerodivisor \(g \in B'\), and an isomorphism \(\varphi' : A'/f^n A' \to B'/g^n B'\) with \(\varphi'(f) \equiv g\), there is a finite extension \(B' \subset B\) and an isomorphism \(\varphi : A/fA \to B/gB\) compatible with \(\varphi'\).
Proof
Since \(A\) is finite over \(A'\) and since \(A'_f = A_f\) we can Cchoose \(t > 0\) such that \(f^t A \subset A'\). Set \(n_0 = 2t\). Given \(n, B', g, \varphi'\) as in the statement of the lemma, denote \(N \subset B'\) the set of elements \(b \in B'\) such that \(b \bmod g^nB' \in \varphi'(f^tA)\). Set \(B = g^{-t}N\). As \(f^tA' \subset f^tA\) and \(\varphi'\) sends \(f\) to \(g\) we have \(g^tB' \subset N\), hence \(B' \subset B\). Since \(f^tA \cdot f^tA \subset f^t \cdot f^tA\) and \(\varphi'\) sends \(f\) to \(g\), we see that \(N \cdot N \subset g^t N\). Hence we obtain a multiplication on \(B\) extending the multiplication of \(B'\). We have an isomorphism of \(A'/f^nA'\)-modules \[A/f^tA' \xrightarrow{f^t} f^tA/f^nA' \xrightarrow{\varphi'} g^tB/g^nB' \xrightarrow{g^{-t}} B/g^tB'\] where the module structures on the right are defined using \(\varphi'\). Since \(A/f^tA'\) is a finite \(A'\)-module, we conclude that \(B/g^tB'\) is a finite \(B'\)-module and hence we see that \(B' \to B\) is finite. Finally, we leave it to the reader to see that the displayed isomorphism of modules sends \(fA\) into \(gB\) and induces an isomorphism of rings \(\varphi : A/fA \to B/gB\) compatible with \(\varphi'\) (it even induces an isomorphism \(A/f^tA \to B/g^tB\) but we don’t need this).
Remark
The construction in Lemma 0GLT satisfies the following “functoriality”. Suppose we have a commutative diagram \[\xymatrix{ A'_2 \ar[r] & A_2 \\ A'_1 \ar[r] \ar[u] & A_1 \ar[u] }\] with injective horizontal arrows. Suppose given an element \(f \in A'_1\) such that \((A'_1 \subset A_1, f)\) and \((A'_2 \subset A_2, f)\) satisfy properties (a), (b), (c) of Lemma 0GLT. Let \(n_{0, 1}\) and \(n_{0, 2}\) be the integers found in the lemma for these two situations. Finally, let \(B'_1 \to B'_2\) be a ring map, let \(g \in B'_1\) be a nonzerodivisor on \(B_1\) and \(B_2\), let \(n \geq \max(n_{0, 1}, n_{0, 2})\), and let a commutative diagram \[\xymatrix{ A'_2/f^nA'_2 \ar[r]_{\varphi'_2} & B'_2/g^nB'_2 \\ A'_1/f^nA'_1 \ar[r]^{\varphi'_1} \ar[u] & B'_2/g^nB'_2 \ar[u] }\] be given whose horizontal arrows are isomorphisms and where \(\varphi'_1(f) \equiv g\). Then we obtain commutative diagrams \[\vcenter{ \xymatrix{ B'_2 \ar[r] & B_2 \\ B'_1 \ar[r] \ar[u] & B_1 \ar[u] } } \quad\text{and}\quad \vcenter{ \xymatrix{ A_2/fA_2 \ar[r]_{\varphi_2} & B_2/gB_2 \\ A_1/fA_1 \ar[r]^{\varphi_1} \ar[u] & B_2/gB_2 \ar[u] } }\] where \((B'_1 \subset B_1, \varphi_1)\) and \((B'_2 \subset B_2, \varphi_2)\) are constructed as in the proof of Lemma 0GLT. We omit the detailed verification.
Lemma
Let \(p\) be a prime number. Let \(A \subset B\) be an extension of discrete valuation rings with fraction field extension \(L/K\). Let \(K_2/K_1/K\) be a tower of finite field extensions. Assume
\(K\) has characteristic \(p\),
\(L/K\) is separable,
\(B\) is Nagata,
\(K_2\) is a solution for \(A \subset B\),
\(K_2/K_1\) is purely inseparable of degree \(p\).
Then there exists a separable extension \(K_3/K_1\) which is a solution for \(A \subset B\).
Proof
Let us use notation as in Remark 09EM; we will use all the observations made there. Since \(L/K\) is separable, the algebra \(L_1 = L \otimes_K K_1\) is reduced (Algebra, Lemma 030U). Since \(B\) is Nagata, the ring extension \(B \subset B_1\) is finite where \(B_1\) is the integral closure of \(B\) in \(L_1\) and \(B_1\) is a Nagata ring. Similarly, the ring \(A\) is Nagata by Lemma 0GLS hence \(A \subset A_1\) is finite and \(A_1\) is a Nagata ring too. Moreover, the same assertions are true for \(K_2\), i.e., \(L_2 = L \otimes_K K_2\) is reduced, the ring extensions \(A_1 \subset A_2\) and \(B_1 \subset B_2\) are finite where \(A_2\), resp. \(B_2\) is the integral closure of \(A\), resp. \(B\) in \(K_2\), resp. \(L_2\).
Let \(\pi \in A\) be a uniformizer. Observe that \(\pi\) is a nonzerodivisor on \(K_1\), \(K_2\), \(A_1\), \(A_2\), \(L_1\), \(L_2\), \(B_1\), and \(B_2\) and we have \(K_1 = (A_1)_\pi\), \(K_2 = (A_2)_\pi\), \(L_1 = (B_1)_\pi\), and \(L_2 = (B_2)_\pi\). We may write \(K_2 = K_1(\alpha)\) where \(\alpha^p = a_1 \in K_1\), see Fields, Lemma 09HI. After multiplying \(\alpha\) by a power of \(\pi\) we may and do assume \(a_1 \in A_1\). For the rest of the proof it is convenient to write \(K_2 = K_1[x]/(x^p - a_1)\) and \(L_2 = L_1[x]/(x^p - a_1)\). Consider the extensions of rings \[A'_2 = A_1[x]/(x^p - a_1) \subset A_2 \quad\text{and}\quad B'_2 = B_1[x]/(x^p - a_1) \subset B_2\] We may apply Lemma 0GLT to \(A'_2 \subset A_2\) and \(f = \pi^2\) and to \(B'_2 \subset B_2\) and \(f = \pi^2\). Choose an integer \(n\) large enough which works for both of these.
Consider the algebras \[K_3 = K_1[x]/(x^p - \pi^{2n} x - a_1) \quad\text{and}\quad L_3 = L_1[x]/(x^p - \pi^{2n} x - a_1)\] Observe that \(K_3/K_1\) and \(L_3/L_1\) are finite étale algebra extensions of degree \(p\). Consider the subrings \[A'_3 = A_1[x]/(x^p - \pi^n x - a_1) \quad\text{and}\quad B'_3 = B_1[x]/(x^p - \pi^n x - a_1)\] of \(K_3 = (A'_2)_\pi\) and \(L_3 = (B'_3)_\pi\). We are going to construct a commutative diagram \[\xymatrix{ B'_2/\pi^{2n} B'_2 \ar[r]_{\psi'} & B'_3/\pi^{2n} B'_3 \\ A'_2/\pi^{2n} A'_2 \ar[r]^{\varphi'} \ar[u] & A'_3/\pi^{2n} A'_3 \ar[u] }\] Namely, \(\varphi'\) is the unique \(A_1\)-algebra isomorphism sending the class of \(x\) to the class of \(x\). Similarly, \(\psi'\) is the unique \(B_1\)-algebra isomorphism sending the class of \(x\) to the class of \(x\). By our choice of \(n\) we obtain, via Lemma 0GLT and Remark 0GLU finite ring extensions \(A'_3 \subset A_3\) and \(B'_3 \subset B_3\) such that \(A'_3 \to B'_3\) extends to a ring map \(A_3 \to B_3\) and a commutative diagram \[\xymatrix{ B_2/\pi^2 B_2 \ar[r]_\psi & B_3/\pi^2B_3 \\ A_2/\pi^2 A_2 \ar[r]^\varphi \ar[u] & A_3/\pi^2A_3 \ar[u] }\] with all the properties asserted in the references mentioned above (in particular \(\varphi\) and \(\psi\) are isomorphisms).
With all of this data in hand, we can finish the proof. Namely, we first observe that \(A_3\) and \(B_3\) are finite products of Dedekind domains with \(\pi\) contained in all of the maximal ideals. Namely, if \(\mathfrak p \subset A_3\) is a maximal ideal, then \(\pi \in \mathfrak p\) as \(A \to A_3\) is finite. Then \(\mathfrak p/\pi^2 A_3\) corresponds via \(\varphi\) to a maximal ideal in \(A_2 / \pi^2A_2\) which is principal as \(A_2\) is a finite product of Dedekind domains. We conclude that \(\mathfrak p/\pi^2 A_3\) is principal and hence by Nakayama we see that \(\mathfrak p (A_3)_\mathfrak p\) is principal. The same argument works for \(B_3\). We conclude that \(A_3\) is the integral closure of \(A\) in \(K_3\) and that \(B_3\) is the integral closure of \(B\) in \(L_3\). Let \(\mathfrak q \subset B_3\) be a maximal ideal lying over \(\mathfrak p \subset A_3\). To finish the proof we have to show that \((A_3)_\mathfrak p \to (B_3)_\mathfrak q\) is formally smooth in the \(\mathfrak q\)-adic topology. By the criterion of Lemma 09E7 it suffices to show that \(\mathfrak p (B_3)_\mathfrak q = \mathfrak q (B_3)_\mathfrak q\) and that the field extension \(\kappa(\mathfrak q)/\kappa(\mathfrak p)\) is separable. This is true because we may check both assertions by looking at the ring map \(A_3/\pi^2 A_3 \to B_3/\pi^2 B_3\) and this is isomorphic to the ring map \(A_2/\pi^2 A_2 \to B_2/\pi^2 B_2\) where the corresponding statement holds by our assumption that \(K_2\) is a solution for \(A \subset B\). Some details omitted.
Lemma
Let \(A \subset B\) be an extension of discrete valuation rings. Assume
the extension \(L/K\) of fraction fields is separable,
\(B\) is Nagata, and
there exists a solution for \(A \subset B\).
Then there exists a separable solution for \(A \subset B\).
Proof
The lemma is trivial if the characteristic of \(K\) is zero; thus we may and do assume that the characteristic of \(K\) is \(p > 0\).
Let \(K_2/K\) be a solution for \(A \to B\). We will use induction on the inseparable degree \([K_2 : K]_i\) (Fields, Definition 030L) of \(K_2/K\). If \([K_2 : K]_i = 1\), then \(K_2\) is separable over \(K\) and we are done. If not, then there exists a subfield \(K_2/K_1/K\) such that \(K_2/K_1\) is purely inseparable of degree \(p\) (Fields, Lemmas 030K and 09HI). By Lemma 0GLV there exists a separable extension \(K_3/K_1\) which is a solution for \(A \subset B\). Then \([K_3 : K]_i = [K_1 : K]_i = [K_2 : K]_i/p\) (Fields, Lemma 09HK) is smaller and we conclude by induction.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Assume \(B\) is essentially of finite type over \(A\). Let \(K'/K\) be an algebraic extension of fields such that the integral closure \(A'\) of \(A\) in \(K'\) is Noetherian. Then the integral closure \(B'\) of \(B\) in \(L' = (L \otimes_K K')_{red}\) is Noetherian as well. Moreover, the map \(\Spec(B') \to \Spec(A')\) is surjective and the corresponding residue field extensions are finitely generated field extensions.
Proof
Let \(A \to C\) be a finite type ring map such that \(B\) is a localization of \(C\) at a prime \(\mathfrak p\). Then \(C' = C \otimes_A A'\) is a finite type \(A'\)-algebra, in particular Noetherian. Since \(A \to A'\) is integral, so is \(C \to C'\). Thus \(B = C_\mathfrak p \subset C'_\mathfrak p\) is integral too. It follows that the dimension of \(C'_\mathfrak p\) is \(1\) (Algebra, Lemma 00OK). Of course \(C'_\mathfrak p\) is Noetherian. Let \(\mathfrak q_1, \ldots, \mathfrak q_n\) be the minimal primes of \(C'_\mathfrak p\). Let \(B'_i\) be the integral closure of \(B = C_\mathfrak p\), or equivalently by the above of \(C'_\mathfrak p\) in the field of fractions of \(C'_{\mathfrak p'}/\mathfrak q_i\). It follows from Krull-Akizuki (Algebra, Lemma 00PG applied to the finitely many localizations of \(C'_\mathfrak p\) at its maximal ideals) that each \(B'_i\) is Noetherian. Moreover the residue field extensions in \(C'_\mathfrak p \to B'_i\) are finite by Algebra, Lemma 031F. Finally, we observe that \(B' = \prod B'_i\) is the integral closure of \(B\) in \(L' = (L \otimes_K K')_{red}\).
Proposition
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). If \(B\) is essentially of finite type over \(A\), then there exists a finite extension \(K_1/K\) which is a solution for \(A \to B\) as defined in Definition 09EN.
Proof
Observe that a weak solution is a solution if the residue field of \(A\) is perfect, see Lemma 09E7. Thus the proposition follows immediately from Theorem 09F9 if the residue characteristic of \(A\) is \(0\) (and in fact we do not need the assumption that \(A \to B\) is essentially of finite type). If the residue characteristic of \(A\) is \(p > 0\) we will also deduce it from Epp’s theorem.
Let \(x_i \in A\), \(i \in I\) be a set of elements mapping to a \(p\)-base of the residue field \(\kappa\) of \(A\). Set \[A' = \bigcup\nolimits_{n \geq 1} A[t_{i, n}]/(t_{i, n}^{p^n} - x_i)\] where the transition maps send \(t_{i, n + 1}\) to \(t_{i, n}^p\). Observe that \(A'\) is a filtered colimit of weakly unramified finite extensions of discrete valuation rings over \(A\). Thus \(A'\) is a discrete valuation ring and \(A \to A'\) is weakly unramified. By construction the residue field \(\kappa' = A'/\mathfrak m_A A'\) is the perfection of \(\kappa\).
Let \(K'\) be the fraction field of \(A'\). We may apply Lemma 09IH to the extension \(K'/K\). Thus \(B'\) is a finite product of Dedekind domains. Let \(\mathfrak m_1, \ldots, \mathfrak m_n\) be the maximal ideals of \(B'\). Using Epp’s theorem (Theorem 09F9) we find a weak solution \(K'_i/K'\) for each of the extensions \(A' \subset B'_{\mathfrak m_i}\). Since the residue field of \(A'\) is perfect, these are actually solutions. Let \(K'_1/K'\) be a finite extension which contains each \(K'_i\). Then \(K'_1/K'\) is still a solution for each \(A' \subset B'_{\mathfrak m_i}\) by Lemma 0GLR.
Let \(A'_1\) be the integral closure of \(A\) in \(K'_1\). Note that \(A'_1\) is a Dedekind domain by the discussion in Remark 09EM applied to \(K' \subset K'_1\). Thus Lemma 09IH applies to \(K'_1/K\). Therefore the integral closure \(B'_1\) of \(B\) in \(L'_1 = (L \otimes_K K'_1)_{red}\) is a Dedekind domain and because \(K'_1/K'\) is a solution for each \(A' \subset B'_{\mathfrak m_i}\) we see that \((A'_1)_{A'_1 \cap \mathfrak m} \to (B'_1)_{\mathfrak m}\) is formally smooth in the \(\mathfrak m\)-adic topology for each maximal ideal \(\mathfrak m \subset B'_1\).
By construction, the field \(K'_1\) is a filtered colimit of finite extensions of \(K\). Say \(K'_1 = \colim_{i \in I} K_i\). For each \(i\) let \(A_i\), resp. \(B_i\) be the integral closure of \(A\), resp. \(B\) in \(K_i\), resp. \(L_i = (L \otimes_K K_i)_{red}\). Then it is clear that \[A'_1 = \colim A_i\quad\text{and}\quad B'_1 = \colim B_i\] Since the ring maps \(A_i \to A'_1\) and \(B_i \to B'_1\) are injective integral ring maps and since \(A'_1\) and \(B'_1\) have finite spectra, we see that for all \(i\) large enough the ring maps \(A_i \to A'_1\) and \(B_i \to B'_1\) are bijective on spectra. Once this is true, for all \(i\) large enough the maps \(A_i \to A'_1\) and \(B_i \to B'_1\) will be weakly unramified (once the uniformizer is in the image). It follows from multiplicativity of ramification indices that \(A_i \to B_i\) induces weakly unramified maps on all localizations at maximal ideals of \(B_i\) for such \(i\). Increasing \(i\) a bit more we see that \[B_i \otimes_{A_i} A'_1 \longrightarrow B'_1\] induces surjective maps on residue fields (because the residue fields of \(B'_1\) are finitely generated over those of \(A'_1\) by Lemma 09IH). Picture of residue fields at maximal ideals lying under a chosen maximal ideal of \(B'_1\): \[\xymatrix{ \kappa_{B_i} \ar[r] & \kappa_{B_{i'}} \ar[r] & & \ldots & \kappa_{B'_1} \\ \kappa_{A_i} \ar[r] \ar[u] & \kappa_{A_{i'}} \ar[r] \ar[u] & & \ldots & \kappa_{A'_1} \ar[u] }\] Thus \(\kappa_{B_i}\) is a finitely generated extension of \(\kappa_{A_i}\) such that the compositum of \(\kappa_{B_i}\) and \(\kappa_{A'_1}\) in \(\kappa_{B'_1}\) is separable over \(\kappa_{A'_1}\). Then that happens already at a finite stage: for example, say \(\kappa_{B'_1}\) is finite separable over \(\kappa_{A'_1}(x_1, \ldots, x_n)\), then just increase \(i\) such that \(x_1, \ldots, x_n\) are in \(\kappa_{B_i}\) and such that all generators satisfy separable polynomial equations over \(\kappa_{A_i}(x_1, \ldots, x_n)\). This means that \(A_i \to (B_i)_\mathfrak m\) is formally smooth in the \(\mathfrak m\)-adic topology for all maximal ideals \(\mathfrak m\) of \(B_i\) and the proof is complete.
Lemma
Let \(A \to B\) be an extension of discrete valuation rings with fraction fields \(K \subset L\). Assume
\(B\) is essentially of finite type over \(A\),
either \(A\) or \(B\) is a Nagata ring, and
\(L/K\) is separable.
Then there exists a separable solution for \(A \to B\) (Definition 09EN).
Proof
Observe that if \(A\) is Nagata, then so is \(B\) (Algebra, Lemma 032U and Proposition 0334). Thus the lemma follows on combining Proposition 09II and Lemma 0BRN.
Picard groups of rings
We first define invertible modules as follows.
Definition
Let \(R\) be a ring. An \(R\)-module \(M\) is invertible if the functor \[\text{Mod}_R \longrightarrow \text{Mod}_R,\quad N \longmapsto M \otimes_R N\] is an equivalence of categories. An invertible \(R\)-module is said to be trivial if it is isomorphic to \(R\) as an \(R\)-module.
Lemma
Let \(R\) be a ring. Let \(M\) be an \(R\)-module. Equivalent are
\(M\) is finite locally free module of rank \(1\),
\(M\) is invertible, and
there exists an \(R\)-module \(N\) such that \(M \otimes_R N \cong R\).
Moreover, in this case the module \(N\) in (3) is isomorphic to \(\Hom_R(M, R)\).
Proof
Assume (1). Consider the module \(N = \Hom_R(M, R)\) and the evaluation map \(M \otimes_R N = M \otimes_R \Hom_R(M, R) \to R\). If \(f \in R\) such that \(M_f \cong R_f\), then the evaluation map becomes an isomorphism after localization at \(f\) (details omitted). Thus we see the evaluation map is an isomorphism by Algebra, Lemma 00EO. Thus (1) \(\Rightarrow\) (3).
Assume (3). Then the functor \(K \mapsto K \otimes_R N\) is a quasi-inverse to the functor \(K \mapsto K \otimes_R M\). Thus (3) \(\Rightarrow\) (2). Conversely, if (2) holds, then \(K \mapsto K \otimes_R M\) is essentially surjective and we see that (3) holds.
Assume the equivalent conditions (2) and (3) hold. Denote \(\psi : M \otimes_R N \to R\) the isomorphism from (3). Choose an element \(\xi = \sum_{i = 1, \ldots, n} x_i \otimes y_i\) such that \(\psi(\xi) = 1\). Consider the isomorphisms \[M \to M \otimes_R M \otimes_R N \to M\] where the first arrow sends \(x\) to \(\sum x_i \otimes x \otimes y_i\) and the second arrow sends \(x \otimes x' \otimes y\) to \(\psi(x' \otimes y)x\). We conclude that \(x \mapsto \sum \psi(x \otimes y_i)x_i\) is an automorphism of \(M\). This automorphism factors as \[M \to R^{\oplus n} \to M\] where the first arrow is given by \(x \mapsto (\psi(x \otimes y_1), \ldots, \psi(x \otimes y_n))\) and the second arrow by \((a_1, \ldots, a_n) \mapsto \sum a_i x_i\). In this way we conclude that \(M\) is a direct summand of a finite free \(R\)-module. This means that \(M\) is finite locally free (Algebra, Lemma 00NX). Since the same is true for \(N\) by symmetry and since \(M \otimes_R N \cong R\), we see that \(M\) and \(N\) both have to have rank \(1\).
The set of isomorphism classes of these modules is often called the class group or Picard group of \(R\). The group structure is determined by assigning to the isomorphism classes of the invertible modules \(L\) and \(L'\) the isomorphism class of \(L \otimes_R L'\). The inverse of an invertible module \(L\) is the module \[L^{\otimes -1} = \Hom_R(L, R),\] because as seen in the proof of Lemma 0B8I the evaluation map \(L \otimes_R L^{\otimes -1} \to R\) is an isomorphism. Let us denote the Picard group of \(R\) by \(\Pic(R)\).
Lemma
Let \(R\) be a UFD. Then \(\Pic(R)\) is trivial.
Proof
Let \(L\) be an invertible \(R\)-module. By Lemma 0B8I we see that \(L\) is a finite locally free \(R\)-module. In particular \(L\) is torsion free and finite over \(R\). Pick a nonzero element \(\varphi \in \Hom_R(L, R)\) of the dual invertible module. Then \(I = \varphi(L) \subset R\) is an ideal which is an invertible module. Pick a nonzero \(f \in I\) and let \[f = u p_1^{e_1} \ldots p_r^{e_r}\] be the factorization into prime elements with \(p_i\) pairwise distinct. Since \(L\) is finite locally free there exist \(a_i \in R\), \(a_i \not \in (p_i)\) such that \(I_{a_i} = (g_i)\) for some \(g_i \in R_{a_i}\). Then \(p_i\) is still a prime element of the UFD \(R_{a_i}\) and we can write \(g_i = p_i^{c_i} g'_i\) for some \(g'_i \in R_{a_i}\) not divisible by \(p_i\). Since \(f \in I_{a_i}\) we see that \(e_i \geq c_i\). We claim that \(I\) is generated by \(h = p_1^{c_1} \ldots p_r^{c_r}\) which finishes the proof.
To prove the claim it suffices to show that \(I_a\) is generated by \(h\) for any \(a \in R\) such that \(I_a\) is a principal ideal (Algebra, Lemma 00EO). Say \(I_a = (g)\). Let \(J \subset \{1, \ldots, r\}\) be the set of \(i\) such that \(p_i\) is a nonunit (and hence a prime element) in \(R_a\). Because \(f \in I_a = (g)\) we find the prime factorization \(g = v \prod_{i \in J} p_j^{b_j}\) with \(v\) a unit and \(b_j \leq e_j\). For each \(j \in J\) we have \(I_{aa_j} = g R_{aa_j} = g_j R_{aa_j}\), in other words \(g\) and \(g_j\) map to associates in \(R_{aa_j}\). By uniqueness of factorization this implies that \(b_j = c_j\) and the proof is complete.
Determinants
Let \(R\) be a ring. Let \(M\) be a finite projective \(R\)-module. There exists a product decomposition \(R = R_0 \times \ldots \times R_t\) such that in the corresponding decomposition \(M = M_0 \times \ldots \times M_t\) of \(M\) we have that \(M_i\) is finite locally free of rank \(i\) over \(R_i\). This follows from Algebra, Lemma 00NX (to see that the rank is locally constant) and Algebra, Lemmas 00EE and 00EM (to decompose \(R\) into a product). In this situation we define \[\det(M) = \wedge^0_{R_0}(M_0) \times \ldots \times \wedge^t_{R_t}(M_t)\] as an \(R\)-module. This is a finite locally free module of rank \(1\) as each term is finite locally free of rank \(1\). If \(\varphi : M \to N\) is an isomorphism of finite projective \(R\)-modules, then we obtain a canonical isomorphism \[\det(\varphi) : \det(M) \longrightarrow \det(N)\] of locally free modules of rank \(1\). More generally, if for all primes \(\mathfrak p\) of \(R\) the ranks of the free modules \(M_\mathfrak p\) and \(N_\mathfrak p\) are the same, then any \(R\)-module homomorphism \(\varphi : M \to N\) induces an \(R\)-module map \(\det(\varphi) : \det(M) \to \det(N)\). Finally, if \(M = N\) then \(\det(\varphi) : \det(M) \to \det(M)\) is an endomorphism of an invertible \(R\)-module. Since \(R = \Hom_R(L, L)\) for an invertible \(R\)-module we may and do view \(\det(\varphi)\) as an element of \(R\). In this way we obtain the determinant \[\det : \Hom_R(M, M) \longrightarrow R\] which is a multiplicative map.
Remark
Let \(R\) be a ring. Let \(M\) be a finite projective \(R\)-module. Then we can consider the graded commutative \(R\)-algebra exterior algebra \(\wedge^*_R(M)\) on \(M\) over \(R\). A formula for \(\det(M)\) is that \(\det(M) \subset \wedge^*_R(M)\) is the annihilator of \(M \subset \wedge^*_R(M)\). This is sometimes useful as it does not refer to the decomposition of \(R\) into a product. Of course, to prove this satisfies the desired properties one has to either decompose \(R\) into a product (as above), or one has to look at the localizations at primes of \(R\).
Next, we consider what happens to the determinant give a short exact sequence of finite projective modules.
Lemma
Let \(R\) be a ring. Let \[0 \to M' \to M \to M'' \to 0\] be a short exact sequence of finite projective \(R\)-modules. Then there is a canonical isomorphism \[\gamma : \det(M') \otimes \det(M'') \longrightarrow \det(M)\]
Proof
First proof. Decompose \(R\) into a product of rings \(R_{ij}\) such that \(M' = \prod M'_{ij}\) and \(M'' = \prod M''_{ij}\) where \(M'_{ij}\) has rank \(i\) and \(M''_{ij}\) has rank \(j\). Of course then \(M = \prod M_{ij}\) and \(M_{ij}\) has rank \(i + j\). This reduces us to the case where \(M'\) and \(M''\) have constant rank say \(i\) and \(j\). In this case we have to construct a canonical map \[\wedge^i(M') \otimes \wedge^j(M'') \longrightarrow \wedge^{i + j}(M)\] To do this choose \(m'_1, \ldots, m'_i\) in \(M'\) and \(m''_1, \ldots, m''_j\) in \(M''\). Denote \(m_1, \ldots, m_i \in M\) the images of \(m'_1, \ldots, m'_i\) and denote \(m_{i + 1}, \ldots , m_{i + j} \in M\) elements mapping to \(m''_1, \ldots, m''_j\) in \(M''\). Our rule will be that \[m'_1 \wedge \ldots \wedge m'_i \otimes m''_1 \wedge \ldots \wedge m''_j \longmapsto m_1 \wedge \ldots \wedge m_{i + j}\] We omit the detailed proof that this is well defined and an isomorphism.
Proof
We will use the description of \(\det(M)\), \(\det(M')\), and \(\det(M'')\) given in Remark 0FJA. Consider the \(R\)-algebra maps \(\wedge^*_R(M') \to \wedge^*_R(M)\) and \(\wedge^*_R(M) \to \wedge^*_R(M'')\). The first is injective and the second is surjective. Take an element \(x' \in \det(M') \subset \wedge^*_R(M')\) and an element \(x'' \in \det(M'') \subset \wedge^*_R(M'')\). Choose an element \(y'' \in \wedge^*(M)\) mapping to \(x''\) and set \[\gamma(x' \otimes x'') = x' \wedge y'' \in \det(M) \subset \wedge^*_R(M)\] The reader verifies easily by looking at localizations at primes that this well defined and an isomorphism. Moreover, this construction gives the same map as the construction given in the first proof.
Lemma
Let \(R\) be a ring. Let \[\xymatrix{ 0 \ar[r] & M' \ar[r] \ar[d]^u & M \ar[r] \ar[d]^v & M'' \ar[r] \ar[d]^w & 0 \\ 0 \ar[r] & K' \ar[r] & K \ar[r] & K'' \ar[r] & 0 }\] be a commutative diagram of finite projective \(R\)-modules whose vertical arrows are isomorphisms. Then we get a commutative diagram of isomorphisms \[\xymatrix{ \det(M') \otimes \det(M'') \ar[r]_-\gamma \ar[d]_{\det(u) \otimes \det(w)} & \det(M) \ar[d]^{\det(v)} \\ \det(K') \otimes \det(K'') \ar[r]^-\gamma & \det(K) }\] where the horizontal arrows are the ones constructed in Lemma 0FJB.
Proof
Omitted. Hint: use the second construction of the maps \(\gamma\) in Lemma 0FJB.
Lemma
Let \(R\) be a ring. Let \[K \subset L \subset M\] be \(R\)-modules such that \(K\), \(L/K\), and \(M/L\) are finite projective \(R\)-modules. Then the diagram \[\xymatrix{ \det(K) \otimes \det(L/K) \otimes \det(M/L) \ar[r] \ar[d] & \det(L) \otimes \det(M/L) \ar[d] \\ \det(K) \otimes \det(M/K) \ar[r] & \det(M) }\] commutes where the maps are those of Lemma 0FJB.
Proof
Omitted. Hint: after localizing at a prime of \(R\) we can assume \(K \subset L \subset M\) is isomorphic to \(R^{\oplus a} \subset R^{\oplus a + b} \subset R^{\oplus a + b + c}\) and in this case the result is an evident computation.
Lemma
Let \(R\) be a ring. Let \(M'\) and \(M''\) be two finite projective \(R\)-modules. Then the diagram \[\xymatrix{ \det(M') \otimes \det(M'') \ar[r] \ar[d]_{\epsilon \cdot (\text{switch tensors})} & \det(M' \oplus M'') \ar[d]^{\det(\text{switch summands})} \\ \det(M'') \otimes \det(M') \ar[r] & \det(M'' \oplus M') }\] commutes where \(\epsilon = \det( -\text{id}_{M' \otimes M''}) \in R^*\) and the horizontal arrows are those of Lemma 0FJB.
Proof
Omitted.
Lemma
Let \(R\) be a ring. Let \(M\), \(N\) be finite projective \(R\)-modules. Let \(a : M \to N\) and \(b : N \to M\) be \(R\)-linear maps. Then \[\det(\text{id} + a \circ b) = \det(\text{id} + b \circ a)\] as elements of \(R\).
Proof
It suffices to prove the assertion after replacing \(R\) by a localization at a prime ideal. Thus we may assume \(R\) is local and \(M\) and \(N\) are finite free. In this case we have to prove the equality \[\det(I_n + AB) = \det(I_m + BA)\] of usual determinants of matrices where \(A\) has size \(n \times m\) and \(B\) has size \(m \times n\). This reduces to the case of the ring \(R = \mathbf{Z}[a_{ij}, b_{ji}; 1 \leq i \leq n, 1 \leq j \leq m]\) where \(a_{ij}\) and \(b_{ij}\) are variables and the entries of the matrices \(A\) and \(B\). Taking the fraction field, this reduces to the case of a field of characteristic zero. In characteristic zero there is a universal polynomial expressing the determinant of a matrix of size \(\leq N\) in the traces of the powers of said matrix. Hence it suffices to prove \[\text{Trace}((I_n + AB)^k) = \text{Trace}((I_m + BA)^k)\] for all \(k \geq 1\). Expanding we see that it suffices to prove \(\text{Trace}((AB)^k) = \text{Trace}((BA)^k)\) for all \(k \geq 0\). For \(k = 1\) this is the well known fact that \(\text{Trace}(AB) = \text{Trace}(BA)\). For \(k > 1\) it follows from this by writing \((AB)^k = A(BA)^{k - 1}B\) and \((BA)^k = (BA)^{k - 1} A B\).
Recall that we have defined in Algebra, Section 00JC a group \(K_0(R)\) as the free group on isomorphism classes of finite projective \(R\)-modules modulo the relations \([M'] + [M''] = [M' \oplus M'']\).
Lemma
Let \(R\) be a ring. There is a map \[\det : K_0(R) \longrightarrow \Pic(R)\] which maps \([M]\) to the class of the invertible module \(\wedge^n(M)\) if \(M\) is a finite locally free module of rank \(n\).
Proof
This follows immediately from the constructions above and in particular Lemma 0FJB to see that the relations are mapped to \(0\).
Perfect complexes and K-groups
We quickly show that the zeroth K-group of the derived category of perfect complexes of a ring \(R\) is the same as \(K_0(R)\) defined in Algebra, Section 00JC.
Lemma
Let \(R\) be a ring. There is a map \[c : \text{perfect complexes over }R \longrightarrow K_0(R)\] with the following properties
\(c(K[n]) = (-1)^nc(K)\) for a perfect complex \(K\),
if \(K \to L \to M \to K[1]\) is a distinguished triangle of perfect complexes, then \(c(L) = c(K) + c(M)\),
if \(K\) is represented by a finite complex \(M^\bullet\) consisting of finite projective modules, then \(c(K) = \sum (-1)^i[M_i]\).
Proof
Let \(K\) be a perfect object of \(D(R)\). By definition we can represent \(K\) by a finite complex \(M^\bullet\) of finite projective \(R\)-modules. We define \(c\) by setting \[c(K) = \sum (-1)^n[M^n]\] in \(K_0(R)\). Of course we have to show that this is well defined, but once it is well defined, then (1) and (3) are immediate. For the moment we view the map \(c\) as defined on complexes of finite projective \(R\)-modules.
Suppose that \(L^\bullet \to M^\bullet\) is a surjective map of finite complexes of finite projective \(R\)-modules. Let \(K^\bullet\) be the kernel. Then we obtain short exact sequences of \(R\)-modules \[0 \to K^n \to L^n \to M^n \to 0\] which are split because \(M^n\) is projective. Hence \(K^\bullet\) is also a finite complex of finite projective \(R\)-modules and \(c(L^\bullet) = c(K^\bullet) + c(M^\bullet)\) in \(K_0(R)\).
Suppose given finite complex \(M^\bullet\) of finite projective \(R\)-modules which is acyclic. Say \(M^n = 0\) for \(n \not \in [a, b]\). Then we can break \(M^\bullet\) into short exact sequences \[\begin{matrix} 0 \to M^a \to M^{a + 1} \to N^{a + 1} \to 0, \\ 0 \to N^{a + 1} \to M^{a + 2} \to N^{a + 3} \to 0, \\ \ldots \\ 0 \to N^{b - 3} \to M^{b - 2} \to N^{b - 2} \to 0, \\ 0 \to N^{b - 2} \to M^{b - 1} \to M^b \to 0 \end{matrix}\] Arguing by descending induction we see that \(N^{b - 2}, \ldots, N^{a + 1}\) are finite projective \(R\)-modules, the sequences are split exact, and \[c(M^\bullet) = \sum (-1)[M^n] = \sum (-1)^n([N^{n - 1}] + [N^n]) = 0\] Thus our construction gives zero on acyclic complexes.
It follows formally from the results of the preceding two paragraphs that \(c\) is well defined and satisfies (2). Namely, suppose the finite complexes \(M^\bullet\) and \(L^\bullet\) of finite projective \(R\)-modules represent the same object of \(D(R)\). Then we can represent the isomorphism by a map \(f : M^\bullet \to L^\bullet\) of complexes, see Derived Categories, Lemma 064B. We obtain a short exact sequence of complexes \[0 \to L^\bullet \to C(f)^\bullet \to K^\bullet[1] \to 0\] see Derived Categories, Definition 014E. Since \(f\) is a quasi-isomorphism, the cone \(C(f)^\bullet\) is acyclic (this follows for example from the discussion in Derived Categories, Section 014Z). Hence \[0 = c(C(f)^\bullet) = c(L^\bullet) + c(K^\bullet[1]) = c(L^\bullet) - c(K^\bullet)\] as desired. We omit the proof of (2) which is similar.
The following lemma shows that \(K_0(R)\) is equal to \(K_0(D_{perf}(R))\).
Lemma
Let \(R\) be a ring. Let \(D_{perf}(R)\) be the derived category of perfect objects, see Lemma 0ATI. The map \(c\) of Lemma 0AFY gives an isomorphism \(K_0(D_{perf}(R)) = K_0(R)\).
Proof
It follows from the definition of \(K_0(D_{perf}(R))\) (Derived Categories, Definition 0FCN) that \(c\) induces a homomorphism \(K_0(D_{perf}(R)) \to K_0(R)\).
Given a finite projective module \(M\) over \(R\) let us denote \(M[0]\) the perfect complex over \(R\) which has \(M\) sitting in degree \(0\) and zero in other degrees. Given a short exact sequence \(0 \to M \to M' \to M'' \to 0\) of finite projective modules we obtain a distinguished triangle \(M[0] \to M'[0] \to M''[0] \to M[1]\), see Derived Categories, Section 014Z. This shows that we obtain a map \(K_0(R) \to K_0(D_{perf}(R))\) by sending \([M]\) to \([M[0]]\) with apologies for the horrendous notation.
It is clear that \(K_0(R) \to K_0(D_{perf}(R)) \to K_0(R)\) is the identity. On the other hand, if \(M^\bullet\) is a bounded complex of finite projective \(R\)-modules, then the the existence of the distinguished triangles of “stupid truncations” (see Homology, Section 0118) \[\sigma_{\geq n}M^\bullet \to \sigma_{\geq n - 1}M^\bullet \to M^{n - 1}[-n + 1] \to (\sigma_{\geq n}M^\bullet)[1]\] and induction show that \[[M^\bullet] = \sum (-1)^i[M^i[0]]\] in \(K_0(D_{perf}(R))\) (with again apologies for the notation). Hence the map \(K_0(R) \to K_0(D_{perf}(R))\) is surjective which finishes the proof.
Determinants of endomorphisms of finite length modules
Let \((R, \mathfrak m, \kappa)\) be a local ring. Consider the category of pairs \((M, \varphi)\) consisting of a finite length \(R\)-module and an endomorphism \(\varphi : M \to M\). This category is abelian and every object is Artinian as well as Noetherian. See Homology, Section 0FCD for definitions.
If \((M, \varphi)\) is a simple object of this category, then \(M\) is annihilated by \(\mathfrak m\) since otherwise \((\mathfrak m M, \varphi|_{\mathfrak m M})\) would be a nontrivial subobject. Also \(\dim_\kappa(M) = \text{length}_R(M)\) is finite. Thus we may define the determinant and the trace \[\det\nolimits_\kappa(\varphi),\quad \text{Trace}_\kappa(\varphi)\] as elements of \(\kappa\) using linear algebra. Similarly for the characteristic polynomial of \(\varphi\) in this case.
By Homology, Lemma 0FCJ for an arbitrary object \((M, \varphi)\) of our category we have a finite filtration \[0 \subset M_1 \subset \ldots \subset M_n = M\] by submodules stable under \(\varphi\) such that \((M_i/M_{i - 1}, \varphi_i)\) is a simple object of the category where \(\varphi_i : M_i/M_{i - 1} \to M_i/M_{i - 1}\) is the induced map. We define the determinant of \((M, \varphi)\) over \(\kappa\) as \[\det\nolimits_\kappa(\varphi) = \prod \det\nolimits_\kappa(\varphi_i)\] with \(\det_\kappa(\varphi_i)\) as defined in the previous paragraph. We define the trace of \((M, \varphi)\) over \(\kappa\) as \[\text{Trace}_\kappa(\varphi) = \sum \text{Trace}_\kappa(\varphi_i)\] with \(\text{Trace}_\kappa(\varphi_i)\) as defined in the previous paragraph. We can similarly define the characteristic polynomial of \(\varphi\) over \(\kappa\) as the product of the characteristic polynomials of \(\varphi_i\) as defined in the previous paragraph. By Jordan-Hölder (Homology, Lemma 0FCK) this is well defined.
Lemma
Let \((R, \mathfrak m, \kappa)\) be a local ring. Let \(0 \to (M, \varphi) \to (M', \varphi') \to (M'', \varphi'') \to 0\) be a short exact sequence in the category discussed above. Then \[\det\nolimits_\kappa(\varphi') = \det\nolimits_\kappa(\varphi)\det\nolimits_\kappa(\varphi''),\quad \text{Trace}_\kappa(\varphi') = \text{Trace}_\kappa(\varphi) + \text{Trace}_\kappa(\varphi'')\] Also, the characteristic polynomial of \(\varphi'\) over \(\kappa\) is the product of the characteristic polynomials of \(\varphi\) and \(\varphi''\).
Proof
Left as an exercise.
Lemma
Let \((R, \mathfrak m, \kappa) \to (R', \mathfrak m', \kappa')\) be a local homomorphism of local rings. Assume that \(\kappa'/\kappa\) is a finite extension. Let \(u \in R'\). Then for any finite length \(R'\)-module \(M'\) we have \[\det\nolimits_\kappa(u : M' \to M') = \text{Norm}_{\kappa'/\kappa}(u \bmod \mathfrak m')^m\] where \(m = \text{length}_{R'}(M')\).
Proof
Observe that the statement makes sense as \(\text{length}_R(M') = \text{length}_{R'}(M') [\kappa' : \kappa]\). If \(M' = \kappa'\), then the equality holds by definition of the norm as the determinant of the linear operator given by multiplication by \(u\). In general one reduces to this case by choosing a suitable filtration and using the multiplicativity of Lemma 0GSY. Some details omitted.
Lemma
Let \((R, \mathfrak m, \kappa) \to (R', \mathfrak m', \kappa')\) be a flat local homomorphism of local rings such that \(m = \text{length}_{R'}(R'/\mathfrak mR') < \infty\). For any \((M, \varphi)\) as above, the element \(\det_\kappa(\varphi)^m\) maps to \(\det_{\kappa'}(\varphi \otimes 1 : M \otimes_R R' \to M \otimes_R R')\) in \(\kappa'\).
Proof
The flatness of \(R \to R'\) assures us that short exact sequences as in Lemma 0GSY base change to short exact sequences over \(R'\). Hence by the multiplicativity of Lemma 0GSY we may assume that \((M, \varphi)\) is a simple object of our category (see introduction to this section). In the simple case \(M\) is annihilated by \(\mathfrak m\). Choose a filtration \[0 \subset I_1 \subset I_2 \subset \ldots \subset I_{m - 1} \subset R'/\mathfrak mR'\] whose successive quotients are isomorphic to \(\kappa'\) as \(R'\)-modules. Then we obtain the filtration \[0 \subset M \otimes_\kappa I_1 \subset M \otimes_\kappa I_2 \subset \ldots \subset M \otimes_\kappa I_{m - 1} \subset M \otimes_\kappa R'/\mathfrak mR' = M \otimes_R R'\] whose successive quotients are isomorphic to \(M \otimes_\kappa \kappa'\). Also, these submodules are invariant under \(\varphi \otimes 1\). By Lemma 0GSY we find \[\det\nolimits_{\kappa'}(\varphi \otimes 1 : M \otimes_R R' \to M \otimes_R R') = \det\nolimits_{\kappa'}(\varphi \otimes 1 : M \otimes_\kappa \kappa' \to M \otimes_\kappa \kappa')^m = \det\nolimits_\kappa(\varphi)^m\] The last equality holds by the compatibility of determinants of linear maps with field extensions. This proves the lemma.
A regular local ring is a UFD
We prove the result mentioned in the section title.
Lemma
Let \(R\) be a regular local ring. Let \(f \in R\). Then \(\Pic(R_f) = 0\).
Proof
Let \(L\) be an invertible \(R_f\)-module. In particular \(L\) is a finite \(R_f\)-module. There exists a finite \(R\)-module \(M\) such that \(M_f \cong L\), see Algebra, Lemma 05N5. By Algebra, Proposition 00O7 we see that \(M\) has a finite free resolution \(F_\bullet\) over \(R\). It follows that \(L\) is quasi-isomorphic to a finite complex of free \(R_f\)-modules. Hence by Lemma 0AFY we see that \([L_f] = n[R_f]\) in \(K_0(R_f)\) for some \(n \in \mathbf{Z}\). Applying the map of Lemma 0AFX we see that \(L\) is trivial.
Lemma
A regular local ring is a UFD.
Proof
Recall that a regular local ring is a domain, see Algebra, Lemma 00NP. We will prove the unique factorization property by induction on the dimension of the regular local ring \(R\). If \(\dim(R) = 0\), then \(R\) is a field and in particular a UFD. Assume \(\dim(R) > 0\). Let \(x \in \mathfrak m\), \(x \not \in \mathfrak m^2\). Then \(R/(x)\) is regular by Algebra, Lemma 00NQ, hence a domain by Algebra, Lemma 00NP, hence \(x\) is a prime element. Let \(\mathfrak p \subset R\) be a height \(1\) prime. We have to show that \(\mathfrak p\) is principal, see Algebra, Lemma 0AFT. We may assume \(x \not \in \mathfrak p\), since if \(x \in \mathfrak p\), then \(\mathfrak p = (x)\) and we are done. For every nonmaximal prime \(\mathfrak q \subset R\) the local ring \(R_\mathfrak q\) is a regular local ring, see Algebra, Lemma 0AFS. By induction we see that \(\mathfrak pR_\mathfrak q\) is principal. In particular, the \(R_x\)-module \(\mathfrak p_x = \mathfrak pR_x \subset R_x\) is a finitely presented \(R_x\)-module whose localization at any prime is free of rank \(1\). By Algebra, Lemma 00NX we see that \(\mathfrak p_x\) is an invertible \(R_x\)-module. By Lemma 0AFZ we see that \(\mathfrak p_x = (y)\) for some \(y \in R_x\). We can write \(y = x^e f\) for some \(f \in \mathfrak p\) and \(e \in \mathbf{Z}\). Factor \(f = a_1 \ldots a_r\) into irreducible elements of \(R\) (Algebra, Lemma 034R). Since \(\mathfrak p\) is prime, we see that \(a_i \in \mathfrak p\) for some \(i\). Since \(\mathfrak p_x = (y)\) is prime and \(a_i | y\) in \(R_x\), it follows that \(\mathfrak p_x\) is generated by \(a_i\) in \(R_x\), i.e., the image of \(a_i\) in \(R_x\) is prime. As \(x\) is a prime element, we find that \(a_i\) is prime in \(R\) by Algebra, Lemma 0AFU. Since \((a_i) \subset \mathfrak p\) and \(\mathfrak p\) has height \(1\) we conclude that \((a_i) = \mathfrak p\) as desired.
Lemma
Let \(R\) be a valuation ring with fraction field \(K\) and residue field \(\kappa\). Let \(R \to A\) be a homomorphism of rings such that
\(A\) is local and \(R \to A\) is local,
\(A\) is flat and essentially of finite type over \(R\),
\(A \otimes_R \kappa\) regular.
Then \(\Pic(A \otimes_R K) = 0\).
Proof
Let \(L\) be an invertible \(A \otimes_R K\)-module. In particular \(L\) is a finite module. There exists a finite \(A\)-module \(M\) such that \(M \otimes_R K \cong L\), see Algebra, Lemma 05N5. We may assume \(M\) is torsion free as an \(R\)-module. Thus \(M\) is flat as an \(R\)-module (Lemma 0539). From Lemma 053E we deduce that \(M\) is of finite presentation as an \(A\)-module and \(A\) is essentially of finite presentation as an \(R\)-algebra. By Lemma 0DHV we see that \(M\) is perfect relative to \(R\), in particular \(M\) is pseudo-coherent as an \(A\)-module. By Lemma 09PC we see that \(M\) is perfect, hence \(M\) has a finite free resolution \(F_\bullet\) over \(A\). It follows that \(L\) is quasi-isomorphic to a finite complex of free \(A \otimes_R K\)-modules. Hence by Lemma 0AFY we see that \([L] = n[A \otimes_R K]\) in \(K_0(A \otimes_R K)\) for some \(n \in \mathbf{Z}\). Applying the map of Lemma 0AFX we see that \(L\) is trivial.
Determinants of complexes
In Section 0FJG we have seen how to a perfect complex \(K\) over a ring \(R\) there is associated an isomorphism class of invertible \(R\)-modules, i.e., an element of \(\Pic(R)\). In fact, analogously to Section 0FJ9 it turns out there is a functor \[\det : \left\{ \begin{matrix} \text{category of perfect complexes} \\ \text{morphisms are isomorphisms} \end{matrix} \right\} \longrightarrow \left\{ \begin{matrix} \text{category of invertible modules} \\ \text{morphisms are isomorphisms} \end{matrix} \right\}\] Moreover, given an object \((L, F)\) of the filtered derived category \(DF(R)\) of \(R\) whose filtration is finite and whose graded parts are perfect complexes, there is a canonical isomorphism \(\det(\text{gr}L) \to \det(L)\). See [determinant] for the original exposition. We will add this material later (insert future reference).
For the moment we will present an ad hoc construction in the case of perfect objects \(L\) in \(D(R)\) of tor-amplitude in \([-1, 0]\). Such an object may be represented by a complex \[L^\bullet = \ldots \to 0 \to L^{-1} \to L^0 \to 0 \to \ldots\] with \(L^{-1}\) and \(L^0\) finite projective \(R\)-modules, see Lemma 0658. In this case we set \[\det(L^\bullet) = \det(L^0) \otimes_R \det(L^{-1})^{\otimes -1} = \Hom_R(\det(L^{-1}), \det(L^0))\] Let us say a complex of this form has rank \(0\) if \(L^{-1}_\mathfrak p\) and \(L^0_\mathfrak p\) have the same rank for all primes of \(R\). If \(L^\bullet\) has rank \(0\), then we have seen in Section 0FJ9 that there is a canonical element \[\delta(L^\bullet) \in \det(L^\bullet)\] which is simply the determininant of \(d : L^{-1} \to L^0\). Note that \(\delta(L^\bullet)\) is a trivialization of \(\det(L^\bullet)\) if and only if \(L^\bullet\) is acyclic.
Consider a map of complexes \(a^\bullet : K^\bullet \to L^\bullet\) such that
\(a^\bullet\) is a quasi-isomorphism,
\(a^n : K^n \to L^n\) is surjective for all \(n\),
\(K^n\), \(L^n\) are finite projective \(R\)-modules, nonzero only for \(n \in \{-1, 0\}\).
In this situation we will construct an isomorphism \[\det(a^\bullet) : \det(K^\bullet) \longrightarrow \det(L^\bullet)\] Using the exact sequences \(0 \to \Ker(a^i) \to K^i \to L^i \to 0\) we obtain isomorphisms \[\gamma^i : \det(\Ker(a^i)) \otimes \det(L^i) \to \det(K^i)\] for \(i = -1, 0\) by Lemma 0FJB. Since \(a^\bullet\) is a quasi-isomorphism the complex \(\Ker(a^\bullet)\) is acyclic and has rank \(0\). Hence the canonical element \(\delta(\Ker(a^\bullet))\) is a trivialization of the invertible \(R\)-module \(\det(\Ker(a^\bullet))\), see above. We define \(\det(a^\bullet) : \det(K^\bullet) \to \det(L^\bullet)\) as the unique isomorphism such that the diagram \[\xymatrix{ \det(K^\bullet) \ar[rr]_{\det(a^\bullet)} & & \det(L^\bullet) \ar[ld]^{\delta(\Ker(a^\bullet))} \\ & \det(L^\bullet) \otimes \det(\Ker(a^\bullet)) \ar[lu]^{\gamma^0 \otimes (\gamma^{-1})^{\otimes -1}} }\] commutes.
Lemma
Let \(R\) be a ring. Let \(a^\bullet : K^\bullet \to L^\bullet\) be a map of complexes of \(R\)-modules satisfying (1), (2), (3) above. If \(L^\bullet\) has rank \(0\), then \(\det(a^\bullet)\) maps the canonical element \(\delta(K^\bullet)\) to \(\delta(L^\bullet)\).
Proof
Write \(M^i = \Ker(a^i)\). Thus we have a map of short exact sequences \[\xymatrix{ 0 \ar[r] & M^{-1} \ar[r] \ar[d]_{d_M} & K^{-1} \ar[r] \ar[d]_{d_K} & L^{-1} \ar[r] \ar[d]_{d_L} & 0 \\ 0 \ar[r] & M^0 \ar[r] & K^0 \ar[r] & L^0 \ar[r] & 0 }\] By Lemma 0FJC we know that \(\det(d_K)\) corresponds to \(\det(d_M) \otimes \det(d_L)\) as maps. Unwinding the definitions this gives the required equality.
Lemma
Let \(R\) be a ring. Let \(a^\bullet : K^\bullet \to L^\bullet\) be a map of complexes of \(R\)-modules satisfying (1), (2), (3) above. Let \(h : K^0 \to L^{-1}\) be a map such that \(b^0 = a^0 + d \circ h\) and \(b^{-1} = a^{-1} + h \circ d\) are surjective. Then \(\det(a^\bullet) = \det(b^\bullet)\) as maps \(\det(K^\bullet) \to \det(L^\bullet)\).
Proof
Suppose there exists a map \(\tilde h : K^0 \to K^{-1}\) such that \(h = a^{-1} \circ \tilde h\) and such that \(k^0 = \text{id} + d \circ \tilde h : K^0 \to K^0\) and \(k^1 = \text{id} + \tilde h \circ d : K^{-1} \to K^{-1}\) are isomorphisms. Then we obtain a commutative diagram \[\xymatrix{ 0 \ar[r] & \Ker(b^\bullet) \ar[r] \ar[d]_{c^\bullet} & K^\bullet \ar[r]_{b^\bullet} \ar[d]_{k^\bullet} & L^\bullet \ar[r] \ar[d]^{\text{id}} & 0 \\ 0 \ar[r] & \Ker(a^\bullet) \ar[r] & K^\bullet \ar[r]^{a^\bullet} & L^\bullet \ar[r] & 0 }\] of complexes, where \(c^\bullet\) is the induced isomorphism of kernels. Using Lemma 0FJC we see that \[\xymatrix{ \det(\Ker(b^i)) \otimes \det(L^i) \ar[r] \ar[d]_{\det(c^i) \otimes 1} & \det(K^i) \ar[d]^{\det(k^i)} \\ \det(\Ker(a^i)) \otimes \det(L^i) \ar[r] & \det(K^i) }\] commutes. Since \(\det(c^\bullet)\) maps the canonical trivialization of \(\det(\Ker(a^\bullet))\) to the canonical trivializatio of \(\Ker(b^\bullet)\) (Lemma 0FJJ) we see that we conclude if (and only if) \[\det(k^0) = \det(k^{-1})\] as elements of \(R\) which follows from Lemma 0FJF.
Suppose there exists a direct summand \(U \subset K^{-1}\) such that both \(a^{-1}|_U : U \to L^{-1}\) and \(b^{-1}|_U : U \to L^{-1}\) are isomorphisms. Define \(\tilde h\) as the composition of \(h\) with the inverse of \(a^{-1}|_U\). We claim that \(\tilde h\) is a map as in the first paragraph of the proof. Namely, we have \(h = a^{-1} \circ \tilde h\) by construction. To show that \(k^{-1} : K^{-1} \to K^{-1}\) is an isomorphism it suffices to show that it is surjective (Algebra, Lemma 05G8). Let \(u \in U\). We may choose \(u' \in U\) such that \(b^{-1}(u') = a^{-1}(u)\). Then \(u = k^{-1}(u')\). Namely, both \(u\) and \(k^{-1}(u')\) are in \(U\) and \(a^{-1}(u) = a^{-1}(k^{-1}(u'))\) by a calculation21 Since \(a^{-1}|_U\) is an isomorphism we get the equality. Thus \(U \subset \Im(k^{-1})\). On the other hand, if \(x \in \Ker(a^{-1})\) then \(x = k^{-1}(x) \bmod U\). Since \(K^{-1} = \Ker(a^{-1}) + U\) we conclude \(k^{-1}\) is surjective. Finally, we show that \(k^0 : K^0 \to K^0\) is surjective. First, since \(a^0 \circ k^0 = b^0\) we see that \(a^0 \circ k^0\) is surjective. If \(x \in \Ker(a^0)\), then \(x = d(y)\) for some \(y \in \Ker(a^{-1})\). We may write \(y = k^{-1}(z)\) for some \(z \in K^{-1}\) by the above. Then \(x = k^0(d(z))\) and we conclude.
Final step of the proof. It suffices to find \(U\) as in the preceding paragraph, but this may not always be possible. However, in order to show equality of two maps of \(R\)-modules, it suffices to do so after localization at primes of \(R\). Hence we may assume \(R\) is local. Then we get the following problem: suppose \[\alpha, \beta : R^{\oplus n} \longrightarrow R^{\oplus m}\] are two surjective \(R\)-linear maps. Find a direct summand \(U \subset R^{\oplus n}\) such that both \(\alpha|_U\) and \(\beta|_U\) are isomorphisms. If \(R\) is a field, this is possible by linear algebra. In general, one takes a solution over the residue field and lifts this to a solution over the local ring \(R\). Some details omitted.
Lemma
Let \(R\) be a ring. Let \(a^\bullet : K^\bullet \to L^\bullet\) and \(b^\bullet : L^\bullet \to M^\bullet\) be maps of complexes of \(R\)-modules satisfying (1), (2), (3) above. Then we have \(\det(b^\bullet) \circ \det(a^\bullet) = \det(b^\bullet \circ a^\bullet)\) as maps \(\det(M^\bullet) \to \det(K^\bullet)\).
Proof
Omitted. Hints: Straightforward from Lemmas 0FJB, 0FJC, and 0FJD.
Lemma
Let \(R\) be a ring. The constructions above determine a functor \[\det : \left\{ \begin{matrix} \text{category of perfect complexes} \\ \text{with tor amplitude in }[-1, 0] \\ \text{morphisms are isomorphisms} \end{matrix} \right\} \longrightarrow \left\{ \begin{matrix} \text{category of invertible modules} \\ \text{morphisms are isomorphisms} \end{matrix} \right\}\] Moreover, given a rank \(0\) perfect object \(L\) of \(D(R)\) with tor-amplitude in \([-1, 0]\) there is a canonical element \(\delta(L) \in \det(L)\) such that for any isomorphism \(a : L \to K\) in \(D(R)\) we have \(\det(a)(\delta(L)) = \delta(K)\).
Proof
By Lemma 0658 every object of the source category may be represented by a complex \[L^\bullet = \ldots \to 0 \to L^{-1} \to L^0 \to 0 \to \ldots\] with \(L^{-1}\) and \(L^0\) finite projective \(R\)-modules. Let us temporarily call a complex of this type good. By Derived Categories, Lemma 064B morphisms between good complexes in the derived category are homotopy classes of maps of complexes. Thus we may work with good complexes and we can use the determinant \(\det(L^\bullet) = \det(L^0) \otimes \det(L^{-1})^{\otimes -1}\) we investigated above.
Let \(a^\bullet : L^\bullet \to K^\bullet\) be a morphism of good complexes which is an isomorphism in \(D(R)\), i.e., a quasi-isomorphism. We say that \[\xymatrix{ L^\bullet \ar[rr]_{a^\bullet} & & K^\bullet \\ & M^\bullet \ar[lu]^{b^\bullet} \ar[ru]_{c^\bullet} }\] is a good diagram if it commutes up to homotopy and \(b^\bullet\) and \(c^\bullet\) satisfy conditions (1), (2), (3) above. Whenever we have such a diagram it makes sense to define \[\det(a^\bullet) = \det(c^\bullet) \circ \det(b^\bullet)^{-1}\] where \(\det(c^\bullet)\) and \(\det(b^\bullet)\) are the isomorphisms constructed in the text above. We will show that good diagrams always exist and that the resulting map \(\det(a^\bullet)\) is independent of the choice of good diagram.
Existence of good diagrams for a quasi-isomorphism \(a^\bullet : L^\bullet \to K^\bullet\) of good complexes. Choose a surjection \(p : R^{\oplus n} \to K^{-1}\). Then we can consider the new good complex \[M^\bullet = \ldots \to 0 \to L^{-1} \oplus R^{\oplus n} \xrightarrow{d \oplus 1} L^0 \oplus R^{\oplus n} \to 0 \to \ldots\] with the projection map \(b^\bullet : M^\bullet \to L^\bullet\) and the map \(c^\bullet : M^\bullet \to K^\bullet\) using \(a^{-1} \oplus p\) in degree \(-1\) and using \(a^0 \oplus d \circ p\) in degree \(0\). The maps \(b^\bullet : M^\bullet \to L^\bullet\) and \(c^\bullet : M^\bullet \to K^\bullet\) satisfy conditions (1), (2), (3) above and we get a good diagram.
Suppose that we have a good diagram \[\xymatrix{ L^\bullet \ar[rr]_{\text{id}^\bullet} & & L^\bullet \\ & M^\bullet \ar[lu]^{b^\bullet} \ar[ru]_{c^\bullet} }\] Then by Lemma 0FJK we see that \(\det(c^\bullet) = \det(b^\bullet)\). Thus we see that \(\det(\text{id}^\bullet) = \text{id}\) is independent of the choice of good diagram.
Before we prove independence in general, we think about composition. Suppose we have quasi-isomorphisms \(L_1^\bullet \to L_2^\bullet\) and \(L_2^\bullet \to L_3^\bullet\) of good complexes and good diagrams \[\vcenter{ \xymatrix{ L_1^\bullet \ar[rr] & & L_2^\bullet \\ & M_{12}^\bullet \ar[lu] \ar[ru] } } \quad\text{and}\quad \vcenter{ \xymatrix{ L_2^\bullet \ar[rr] & & L_3^\bullet \\ & M_{23}^\bullet \ar[lu] \ar[ru] } }\] We can extend this to a diagram \[\xymatrix{ L_1^\bullet \ar[rr] & & L_2^\bullet \ar[rr] & & L_3^\bullet \\ & M_{12}^\bullet \ar[lu] \ar[ru] & & M_{23}^\bullet \ar[lu] \ar[ru] \\ & & M_{123}^\bullet \ar[lu] \ar[ru] }\] where \(M_{123}^\bullet \to M_{12}^\bullet\) and \(M_{123}^\bullet \to M_{23}^\bullet\) have properties (1), (2), (3) and the square in the diagram commutes: we can just take \(M_{123}^n = M_{12}^n \times_{L_2^n} M_{23}^n\). Then Lemma 0FJL shows that \[\xymatrix{ \det(L_2^\bullet) & \det(M_{23}^\bullet) \ar[l] \\ \det(M_{12}^\bullet) \ar[u] & \det(M_{123}^\bullet) \ar[l] \ar[u] }\] commutes. A diagram chase shows that the composition \(\det(L_1^\bullet) \to \det(L_2^\bullet) \to \det(L_3^\bullet)\) of the maps associated to the two good diagrams using \(M_{12}^\bullet\) and \(M_{23}^\bullet\) is equal to the map associated to the good diagram \[\xymatrix{ L_1^\bullet \ar[rr] & & L_3^\bullet \\ & M_{123}^\bullet \ar[lu] \ar[ru] }\] Thus if we can show that these maps are independent of choices, then the composition law is satisfied too and we obtain our functor.
Independence. Let a quasi-isomorphism \(a^\bullet : L^\bullet \to K^\bullet\) of good complexes be given. Choose an inverse quasi-isomorphism \(b^\bullet : K^\bullet \to L^\bullet\). Setting \(L_1^\bullet = L\), \(L_2^\bullet = K^\bullet\) and \(L_3^\bullet = L^\bullet\) may fix our choice of good diagram for \(b^\bullet\) and consider varying good diagrams for \(a^\bullet\). Then the result of the previous paragraphs is that no matter what choices, the composition always equals the identity map on \(\det(L^\bullet)\). This clearly proves independence of those choices.
The statement on canonical elements follows immediately from Lemma 0FJJ and our construction.
Extensions of valuation rings
This section is the analogue of Section 0EXQ for general valuation rings.
Definition
We say that \(A \to B\) or \(A \subset B\) is an extension of valuation rings if \(A\) and \(B\) are valuation rings and \(A \to B\) is injective and local. Such an extension induces a commutative diagram \[\xymatrix{ A \setminus \{0\} \ar[r] \ar[d]_v & B \setminus \{0\} \ar[d]^v \\ \Gamma_A \ar[r] & \Gamma_B }\] where \(\Gamma_A\) and \(\Gamma_B\) are the value groups. We say that \(B\) is weakly unramified over \(A\) if the lower horizontal arrow is a bijection. If the extension of residue fields \(\kappa_A = A/\mathfrak m_A \subset \kappa_B = B/\mathfrak m_B\) is finite, then we set \(f = [\kappa_B : \kappa_A]\) and we call it the residual degree or residue degree of the extension \(A \subset B\).
Note that \(\Gamma_A \to \Gamma_B\) is injective, because the units of \(A\) are the inverse of the units of \(B\) under the map \(A \to B\). Note also, that we do not require the extension of fraction fields to be finite.
Lemma
Let \(A \subset B\) be an extension of valuation rings with fraction fields \(K \subset L\). If the extension \(L/K\) is finite, then the residue field extension is finite, the index of \(\Gamma_A\) in \(\Gamma_B\) is finite, and \[[\Gamma_B : \Gamma_A] [\kappa_B : \kappa_A] \leq [L : K].\]
Proof
Let \(b_1, \ldots, b_n \in B\) be units whose images in \(\kappa_B\) are linearly independent over \(\kappa_A\). Let \(c_1, \ldots, c_m \in B\) be nonzero elements whose images in \(\Gamma_B/\Gamma_A\) are pairwise distinct. We claim that \(b_i c_j\) are \(K\)-linearly independent in \(L\). Namely, we claim a sum \[\sum a_{ij} b_i c_j\] with \(a_{ij} \in K\) not all zero cannot be zero. Choose \((i_0, j_0)\) with \(v(a_{i_0j_0}b_{i_0}c_{j_0})\) minimal. Replace \(a_{ij}\) by \(a_{ij}/a_{i_0j_0}\), so that \(a_{i_0 j_0} = 1\). Let \[P = \{(i, j) \mid v(a_{ij}b_ic_j) = v(a_{i_0j_0}b_{i_0}c_{j_0}) \}\] By our choice of \(c_1, \ldots, c_m\) we see that \((i, j) \in P\) implies \(j = j_0\). Hence if \((i, j) \in P\), then \(v(a_{ij}) = v(a_{i_0j_0}) = 0\), i.e., \(a_{ij}\) is a unit. By our choice of \(b_1, \ldots, b_n\) we see that \[\sum\nolimits_{(i, j) \in P} a_{ij}b_i\] is a unit in \(B\). Thus the valuation of \(\sum\nolimits_{(i, j) \in P} a_{ij}b_ic_j\) is \(v(c_{j_0}) = v(a_{i_0j_0}b_{i_0}c_{j_0})\). Since the terms with \((i, j) \not \in P\) in the first displayed sum have strictly bigger valuation, we conclude that this sum cannot be zero, thereby proving the lemma.
Lemma
Let \(A\) be a valuation ring with fraction field \(K\) of characteristic \(p > 0\). Let \(L/K\) be a purely inseparable extension. Then the integral closure \(B\) of \(A\) in \(L\) is a valuation ring with fraction field \(L\) and \(A \subset B\) is an extension of valuation rings.
Proof
Omitted. Hints: use Algebra, Lemmas 052K and 00GQ for example.
Lemma
Let \(A \to B\) be a flat local homomorphism of Noetherian local normal domains. Let \(f \in A\) and \(h \in B\) such that \(f = w h^n\) for some \(n > 1\) and some unit \(w\) of \(B\). Assume that for every height \(1\) prime \(\mathfrak p \subset A\) there is a height \(1\) prime \(\mathfrak q \subset B\) lying over \(\mathfrak p\) such that the extension \(A_\mathfrak p \subset B_\mathfrak q\) is weakly unramified. Then \(f = u g^n\) for some \(g \in A\) and unit \(u\) of \(A\).
Proof
The local rings of \(A\) and \(B\) at height \(1\) primes are discrete valuation rings (Algebra, Lemma 00PD). Thus the assumption makes sense (via Definition 09E4). Let \(\mathfrak p_1, \ldots, \mathfrak p_r\) be the primes of \(A\) minimal over \(f\). These have height \(1\) by Algebra, Lemma 00KV. For each \(i\) let \(\mathfrak q_{i, j} \subset B\), \(j = 1, \ldots, r_i\) be the height \(1\) primes of \(B\) lying over \(\mathfrak p_i\). Say we number them so that \(A_{\mathfrak p_i} \to B_{\mathfrak q_{i, 1}}\) is weakly unramified. Since \(f\) maps to an \(n\)th power times a unit in \(B_{\mathfrak q_{i, 1}}\) we see that the valuation \(v_i\) of \(f\) in \(A_{\mathfrak p_i}\) is divisible by \(n\). Say \(v_i = n w_i\) for some \(w_i \geq 0\). Consider the exact sequence \[0 \to I \to A \to \prod\nolimits_{i = 1, \ldots, r} A_{\mathfrak p_i}/\mathfrak p_i^{w_i}A_{\mathfrak p_i}\] defining the ideal \(I\). Applying the exact functor \(- \otimes_A B\) we obtain an exact sequence \[0 \to I \otimes_A B \to B \to \prod\nolimits_{i = 1, \ldots, r} (A_{\mathfrak p_i}/\mathfrak p_i^{w_i}A_{\mathfrak p_i}) \otimes_A B\] Fix \(i\). We claim that the canonical map \[(A_{\mathfrak p_i}/\mathfrak p_i^{w_i}A_{\mathfrak p_i}) \otimes_A B \to \prod\nolimits_{j = 1, \ldots, r_i} B_{\mathfrak q_{i, j}}/\mathfrak q_{i, j}^{e_{i, j}w_i}B_{\mathfrak q_{i, j}}\] is injective. Here \(e_{i, j}\) is the ramification index of \(A_{\mathfrak p_i} \to B_{\mathfrak q_{i, j}}\). The claim asserts that \(\mathfrak p_i^{w_i}B_{\mathfrak p_i}\) is equal to the set of elements \(b\) of \(B_{\mathfrak p_i}\) whose valuation at \(\mathfrak q_{i, j}\) is \(\geq e_{i, j}w_i\). Choose a generator \(a \in A_{\mathfrak p_i}\) of the principal ideal \(\mathfrak p_i^{w_i}\). Then the valuation of \(a\) at \(\mathfrak q_{i, j}\) is equal to \(e_{i, j}w_i\). Hence, as \(B_{\mathfrak p_i}\) is a normal domain whose height one primes are the primes \(\mathfrak q_{i, j}\), \(j = 1, \ldots, r_i\), we see that, for \(b\) as above, we have \(b/a \in B_{\mathfrak p_i}\) by Algebra, Lemma 031T. Thus the claim.
The claim combined with the second exact sequence above determines an exact sequence \[0 \to I \otimes_A B \to B \to \prod\nolimits_{i = 1, \ldots, r} \prod\nolimits_{j = 1, \ldots, r_i} B_{\mathfrak q_{i, j}}/\mathfrak q_{i, j}^{e_{i, j}w_i}B_{\mathfrak q_{i, j}}\] It follows that \(I \otimes_A B\) is the set of elements \(h'\) of \(B\) which have valuation \(\geq e_{i, j}w_i\) at \(\mathfrak q_{i, j}\). Since \(f = wh^n\) in \(B\) we see that \(h\) has valuation \(e_{i, j}w_i\) at \(\mathfrak q_{i, j}\). Thus \(h'/h \in B\) by Algebra, Lemma 031T. It follows that \(I \otimes_A B\) is a free \(B\)-module of rank \(1\) (generated by \(h\)). Therefore \(I\) is a free \(A\)-module of rank \(1\), see Algebra, Lemma 00O1. Let \(g \in I\) be a generator. Then we see that \(g\) and \(h\) differ by a unit in \(B\). Working backwards we conclude that the valuation of \(g\) in \(A_{\mathfrak p_i}\) is \(w_i = v_i/n\). Hence \(g^n\) and \(f\) differ by a unit in \(A\) (by Algebra, Lemma 031T) as desired.
Lemma
Let \(A\) be a valuation ring. Let \(A \to B\) be an étale ring map and let \(\mathfrak m \subset B\) be a prime lying over the maximal ideal of \(A\). Then \(A \subset B_\mathfrak m\) is an extension of valuation rings which is weakly unramified.
Proof
The ring \(A\) has weak dimension \(\leq 1\) by Lemma 092S. Then \(B\) has weak dimension \(\leq 1\) by Lemmas 092E and 092N. hence the local ring \(B_\mathfrak m\) is a valuation ring by Lemma 092S. Since the extension \(A \subset B_\mathfrak m\) induces a finite extension of fraction fields, we see that the \(\Gamma_A\) has finite index in the value group of \(B_{\mathfrak m}\). Thus for every \(h \in B_\mathfrak m\) there exists an \(n > 0\), an element \(f \in A\), and a unit \(w \in B_\mathfrak m\) such that \(f = w h^n\) in \(B_\mathfrak m\). We will show that this implies \(f = ug^n\) for some \(g \in A\) and unit \(u \in A\); this will show that the value groups of \(A\) and \(B_\mathfrak m\) agree, as claimed in the lemma.
Write \(A = \colim A_i\) as the colimit of its local subrings which are essentially of finite type over \(\mathbf{Z}\). Since \(A\) is a normal domain (Algebra, Lemma 00IC), we may assume that each \(A_i\) is normal (here we use that taking normalizations the local rings remain essentially of finite type over \(\mathbf{Z}\) by Algebra, Proposition 0335). For some \(i\) we can find an étale extension \(A_i \to B_i\) such that \(B = A \otimes_{A_i} B_i\), see Algebra, Lemma 00U2. Let \(\mathfrak m_i\) be the intersection of \(B_i\) with \(\mathfrak m\). Then we may apply Lemma 0ASI to the ring map \(A_i \to (B_i)_{\mathfrak m_i}\) to conclude. The hypotheses of the lemma are satisfied because:
\(A_i\) and \((B_i)_{\mathfrak m_i}\) are Noetherian as they are essentially of finite type over \(\mathbf{Z}\),
\(A_i \to (B_i)_{\mathfrak m_i}\) is flat as \(A_i \to B_i\) is étale,
\(B_i\) is normal as \(A_i \to B_i\) is étale, see Algebra, Lemma 033C,
for every height \(1\) prime of \(A_i\) there exists a height \(1\) prime of \((B_i)_{\mathfrak m_i}\) lying over it by Algebra, Lemma 02MA and the fact that \(\Spec((B_i)_{\mathfrak m_i}) \to \Spec(A_i)\) is surjective,
the induced extensions \((A_i)_\mathfrak p \to (B_i)_\mathfrak q\) are unramified for every prime \(\mathfrak q\) lying over a prime \(\mathfrak p\) as \(A_i \to B_i\) is étale.
This concludes the proof of the lemma.
Lemma
Let \(A\) be a valuation ring. Let \(A^h\), resp. \(A^{sh}\) be its henselization, resp. strict henselization. Then \[A \subset A^h \subset A^{sh}\] are extensions of valuation rings which induce bijections on value groups, i.e., which are weakly unramified.
Proof
Write \(A^h = \colim (B_i)_{\mathfrak q_i}\) where \(A \to B_i\) is étale and \(\mathfrak q_i \subset B_i\) is a prime ideal lying over \(\mathfrak m_A\), see Algebra, Lemma 04GV. Then Lemma 0ASJ tells us that \((B_i)_{\mathfrak q_i}\) is a valuation ring and that the induced map \[(A \setminus \{0\})/A^* \longrightarrow ((B_i)_{\mathfrak q_i} \setminus \{0\}) / (B_i)_{\mathfrak q_i}^*\] is bijective. By Algebra, Lemma 0AS4 we conclude that \(A^h\) is a valuation ring. It also follows that \((A \setminus \{0\})/A^* \to (A^h \setminus \{0\})/(A^h)^*\) is bijective. This proves the lemma for the inclusion \(A \subset A^h\). To prove it for \(A \subset A^{sh}\) we can use exactly the same argument except we replace Algebra, Lemma 04GV by Algebra, Lemma 04GW. Since \(A^{sh} = (A^h)^{sh}\) we see that this also proves the assertions of the lemma for the inclusion \(A^h \subset A^{sh}\).
Structure of modules over a PID
We work a little bit more generally (following the papers [Warfield-Purity] and [Warfield-Decomposition] by Warfield) so that the proofs work over valuation rings.
Lemma
Let \(P\) be a module over a ring \(R\). The following are equivalent
\(P\) is a direct summand of a direct sum of modules of the form \(R/fR\), for \(f \in R\) varying.
for every short exact sequence \(0 \to A \to B \to C \to 0\) of \(R\)-modules such that \(fA = A \cap fB\) for all \(f \in R\) the map \(\Hom_R(P, B) \to \Hom_R(P, C)\) is surjective.
Proof
Let \(0 \to A \to B \to C \to 0\) be an exact sequence as in (2). To prove that (1) implies (2) it suffices to prove that \(\Hom_R(R/fR, B) \to \Hom_R(R/fR, C)\) is surjective for every \(f \in R\). Let \(\psi : R/fR \to C\) be a map. Say \(\psi(1)\) is the image of \(b \in B\). Then \(fb \in A\). Hence there exists an \(a \in A\) such that \(fa = fb\). Then \(f(b - a) = 0\) hence we get a morphism \(\varphi : R/fR \to B\) mapping \(1\) to \(b - a\) which lifts \(\psi\).
Conversely, assume that (2) holds. Let \(I\) be the set of pairs \((f, \varphi)\) where \(f \in R\) and \(\varphi : R/fR \to P\). For \(i \in I\) denote \((f_i, \varphi_i)\) the corresponding pair. Consider the map \[B = \bigoplus\nolimits_{i \in I} R/f_iR \longrightarrow P\] which sends the element \(r\) in the summand \(R/f_iR\) to \(\varphi_i(r)\) in \(P\). Let \(A = \Ker(B \to P)\). Then we see that (1) is true if the sequence \[0 \to A \to B \to P \to 0\] is an exact sequence as in (2). To see this suppose \(f \in R\) and \(a \in A\) maps to \(f b\) in \(B\). Write \(b = (r_i)_{i \in I}\) with almost all \(r_i = 0\). Then we see that \[f\sum \varphi_i(r_i) = 0\] in \(P\). Hence there is an \(i_0 \in I\) such that \(f_{i_0} = f\) and \(\varphi_{i_0}(1) = \sum \varphi_i(r_i)\). Let \(x_{i_0} \in R/f_{i_0}R\) be the class of \(1\). Then we see that \[a' = (r_i)_{i \in I} - (0, \ldots, 0, x_{i_0}, 0, \ldots )\] is an element of \(A\) and \(fa' = a\) as desired.
Lemma
Let \(R\) be a nonzero ring. The following are equivalent
For \(a, b \in R\) either \(a\) divides \(b\) or \(b\) divides \(a\).
Every finitely generated ideal is principal and \(R\) is local.
The set of ideals of \(R\) is linearly ordered by inclusion.
This holds in particular if \(R\) is a valuation ring.
Proof
Assume (2) and let \(a, b \in R\). Then \((a, b) = (c)\). If \(c = 0\), then \(a = b = 0\) and \(a\) divides \(b\). Assume \(c \not = 0\). Write \(c = ua + vb\) and \(a = wc\) and \(b = zc\). Then \(c(1 - uw - vz) = 0\). Since \(R\) is local, this implies that \(1 - uw - vz \in \mathfrak m\). Hence either \(w\) or \(z\) is a unit, so either \(a\) divides \(b\) or \(b\) divides \(a\). Thus (2) implies (1).
Assume (1). If \(R\) has two maximal ideals \(\mathfrak m_i\) we can choose \(a \in \mathfrak m_1\) with \(a \not \in \mathfrak m_2\) and \(b \in \mathfrak m_2\) with \(b \not \in \mathfrak m_1\). Then \(a\) does not divide \(b\) and \(b\) does not divide \(a\). Hence \(R\) has a unique maximal ideal and is local. It follows easily from condition (1) and induction that every finitely generated ideal is principal. Thus (1) implies (2).
It is straightforward to prove that (1) and (3) are equivalent. The final statement is Algebra, Lemma 00IB.
Lemma
Let \(R\) be a ring satisfying the equivalent conditions of Lemma 0ASN. Then every finitely presented \(R\)-module is isomorphic to a finite direct sum of modules of the form \(R/fR\).
Proof
Let \(M\) be a finitely presented \(R\)-module. We will use all the equivalent properties of \(R\) from Lemma 0ASN without further mention. Denote \(\mathfrak m \subset R\) the maximal ideal and \(\kappa = R/\mathfrak m\) the residue field. Let \(I \subset R\) be the annihilator of \(M\). Choose a basis \(y_1, \ldots, y_n\) of the finite dimensional \(\kappa\)-vector space \(M/\mathfrak m M\). We will argue by induction on \(n\).
By Nakayama’s lemma any collection of elements \(x_1, \ldots, x_n \in M\) lifting the elements \(y_1, \ldots, y_n\) in \(M/\mathfrak m M\) generate \(M\), see Algebra, Lemma 00DV. This immediately proves the base case \(n = 0\) of the induction.
We claim there exists an index \(i\) such that for any choice of \(x_i \in M\) mapping to \(y_i\) the annihilator of \(x_i\) is \(I\). Namely, if not, then we can choose \(x_1, \ldots, x_n\) such that \(I_i = \text{Ann}(x_i) \not = I\) for all \(i\). But as \(I \subset I_i\) for all \(i\), ideals being totally ordered implies \(I_i\) is strictly bigger than \(I\) for \(i = 1, \ldots, n\), and by total ordering once more we would see that \(\text{Ann}(M) = I_1 \cap \ldots \cap I_n\) is bigger than \(I\) which is a contradiction. After renumbering we may assume that \(y_1\) has the property: for any \(x_1 \in M\) lifting \(y_1\) the annihilator of \(x_1\) is \(I\).
We set \(A = Rx_1 \subset M\). Consider the exact sequence \(0 \to A \to M \to M/A \to 0\). Since \(A\) is finite, we see that \(M/A\) is a finitely presented \(R\)-module (Algebra, Lemma 0519) with fewer generators. Hence \(M/A \cong \bigoplus_{j = 1, \ldots, m} R/f_jR\) by induction. On the other hand, we claim that \(A \to M\) satisfies the property: if \(f \in R\), then \(fA = A \cap fM\). The inclusion \(fA \subset A \cap fM\) is trivial. Conversely, if \(x \in A \cap fM\), then \(x = gx_1 = f y\) for some \(g \in R\) and \(y \in M\). If \(f\) divides \(g\), then \(x \in fA\) as desired. If not, then we can write \(f = hg\) for some \(h \in \mathfrak m\). The element \(x'_1 = x_1 - hy\) has annihilator \(I\) by the previous paragraph. Thus \(g \in I\) and we see that \(x = 0\) as desired. The claim and Lemma 0ASM imply the sequence \(0 \to A \to M \to M/A \to 0\) is split and we find \(M \cong A \oplus \bigoplus_{j = 1, \ldots, m} R/f_jR\). Then \(A = R/I\) is finitely presented (as a summand of \(M\)) and hence \(I\) is finitely generated, hence principal. This finishes the proof.
Lemma
Let \(R\) be a ring such that every local ring of \(R\) at a maximal ideal satisfies the equivalent conditions of Lemma 0ASN. Then every finitely presented \(R\)-module is a summand of a finite direct sum of modules of the form \(R/fR\) for \(f\) in \(R\) varying.
Proof
Let \(M\) be a finitely presented \(R\)-module. We first show that \(M\) is a summand of a direct sum of modules of the form \(R/fR\) and at the end we argue the direct sum can be taken to be finite. Let \[0 \to A \to B \to C \to 0\] be a short exact sequence of \(R\)-modules such that \(fA = A \cap fB\) for all \(f \in R\). By Lemma 0ASM we have to show that \(\Hom_R(M, B) \to \Hom_R(M, C)\) is surjective. It suffices to prove this after localization at maximal ideals \(\mathfrak m\), see Algebra, Lemma 00HN. Note that the localized sequences \(0 \to A_\mathfrak m \to B_\mathfrak m \to C_\mathfrak m \to 0\) satisfy the condition that \(fA_\mathfrak m = A_\mathfrak m \cap fB_\mathfrak m\) for all \(f \in R_\mathfrak m\) (because we can write \(f = uf'\) with \(u \in R_\mathfrak m\) a unit and \(f' \in R\) and because localization is exact). Since \(M\) is finitely presented, we see that \[\Hom_R(M, B)_\mathfrak m = \Hom_{R_\mathfrak m}(M_\mathfrak m, B_\mathfrak m) \quad\text{and}\quad \Hom_R(M, C)_\mathfrak m = \Hom_{R_\mathfrak m}(M_\mathfrak m, C_\mathfrak m)\] by Algebra, Lemma 0583. The module \(M_\mathfrak m\) is a finitely presented \(R_\mathfrak m\)-module. By Lemma 0ASP we see that \(M_\mathfrak m\) is a direct sum of modules of the form \(R_\mathfrak m/fR_\mathfrak m\). Thus we conclude by Lemma 0ASM that the map on localizations is surjective.
At this point we know that \(M\) is a summand of \(\bigoplus_{i \in I} R/f_i R\). Consider the map \(M \to \bigoplus_{i \in I} R/f_i R\). Since \(M\) is a finite \(R\)-module, the image is contained in \(\bigoplus_{i \in I'} R/f_i R\) for some finite subset \(I' \subset I\). This finishes the proof.
Definition
Let \(R\) be a domain.
We say \(R\) is a Bézout domain if every finitely generated ideal of \(R\) is principal.
We say \(R\) is an elementary divisor domain if for all \(n , m \geq 1\) and every \(n \times m\) matrix \(A\), there exist invertible matrices \(U, V\) of size \(n \times n, m \times m\) such that \[U A V = \left( \begin{matrix} f_1 & 0 & 0 & \ldots \\ 0 & f_2 & 0 & \ldots \\ 0 & 0 & f_3 & \ldots \\ \ldots & \ldots & \ldots & \ldots \end{matrix} \right)\] with \(f_1, \ldots, f_{\min(n, m)} \in R\) and \(f_1 | f_2 | \ldots\).
It is apparently still an open question as to whether every Bézout domain \(R\) is an elementary divisor domain (or not). This is equivalent to the question of whether every finitely presented module over \(R\) is a direct sum of cyclic modules. The converse implication is true.
Lemma
An elementary divisor domain is Bézout.
Proof
Let \(a, b \in R\) be nonzero. Consider the \(1 \times 2\) matrix \(A = (a\ b)\). Then we see that \(u(a\ b)V = (f\ 0)\) with \(u \in R\) invertible and \(V = (g_{ij})\) an invertible \(2 \times 2\) matrix. Then \(f = u a g_{11} + u b g_{2 1}\) and \((g_{11}, g_{2 1}) = R\). It follows that \((a, b) = (f)\). An induction argument (omitted) then shows any finitely generated ideal in \(R\) is generated by one element.
Lemma
The localization of a Bézout domain is Bézout. Every local ring of a Bézout domain is a valuation ring. A local domain is Bézout if and only if it is a valuation ring.
Proof
We omit the proof of the statement on localizations. The final statement is Algebra, Lemma 090Q. The second statement follows from the other two.
Lemma
Let \(R\) be a Bézout domain.
Every finite submodule of a free module is finite free.
Every finitely presented \(R\)-module \(M\) is a direct sum of a finite free module and a torsion module \(M_{tors}\) which is a summand of a module of the form \(\bigoplus_{i = 1, \ldots, n} R/f_iR\) with \(f_1, \ldots, f_n \in R\) nonzero.
Proof
Proof of (1). Let \(M \subset F\) be a finite submodule of a free module \(F\). Since \(M\) is finite, we may assume \(F\) is a finite free module (details omitted). Say \(F = R^{\oplus n}\). We argue by induction on \(n\). If \(n = 1\), then \(M\) is a finitely generated ideal, hence principal by our assumption that \(R\) is Bézout. If \(n > 1\), then we consider the image \(I\) of \(M\) under the projection \(R^{\oplus n} \to R\) onto the last summand. If \(I = (0)\), then \(M \subset R^{\oplus n - 1}\) and we are done by induction. If \(I \not = 0\), then \(I = (f) \cong R\). Hence \(M \cong R \oplus \Ker(M \to I)\) and we are done by induction as well.
Let \(M\) be a finitely presented \(R\)-module. Since the localizations of \(R\) at maximal ideals are valuation rings (Lemma 0AST) we may apply Lemma 0ASQ. Thus \(M\) is a summand of a module of the form \(R^{\oplus r} \oplus \bigoplus_{i = 1, \ldots, n} R/f_iR\) with \(f_i \not = 0\). Since taking the torsion submodule is a functor we see that \(M_{tors}\) is a summand of the module \(\bigoplus_{i = 1, \ldots, n} R/f_iR\) and \(M/M_{tors}\) is a summand of \(R^{\oplus r}\). By the first part of the proof we see that \(M/M_{tors}\) is finite free. Hence \(M \cong M_{tors} \oplus M/M_{tors}\) as desired.
Lemma
Let \(R\) be a PID. Every finite \(R\)-module \(M\) is isomorphic to a module of the form \[R^{\oplus r} \oplus \bigoplus\nolimits_{i = 1, \ldots, n} R/f_iR\] for some \(r, n \geq 0\) and \(f_1, \ldots, f_n \in R\) nonzero.
Proof
A PID is a Noetherian Bézout ring. By Lemma 0ASU it suffices to prove the result if \(M\) is torsion. Since \(M\) is finite, this means that the annihilator of \(M\) is nonzero. Say \(fM = 0\) for some \(f \in R\) nonzero. Then we can think of \(M\) as a module over \(R/fR\). Since \(R/fR\) is Noetherian of dimension \(0\) (small detail omitted) we see that \(R/fR = \prod R_j\) is a finite product of Artinian local rings \(R_j\) (Algebra, Proposition 00KJ). Write \(R_j = R/f_jR\) for some \(f_j \in R\). Each \(R_j\), being a local ring and a quotient of a PID, is a generalized valuation ring in the sense of Lemma 0ASN (small detail omitted). Write \(M = \prod M_j\) with \(M_j = e_j M\) where \(e_j \in R/fR\) is the idempotent corresponding to the factor \(R_j\). By Lemma 0ASP we see that \(M_j = \bigoplus_{i = 1, \ldots, n_j} R_j/\overline{f}_{ji}R_j\) for some \(\overline{f}_{ji} \in R_j\). Choose lifts \(f_{ji} \in R\) and choose \(g_{ji} \in R\) with \((g_{ji}) = (f_j, f_{ji})\). Then we conclude that \[M \cong \bigoplus R/g_{ji}R\] as an \(R\)-module which finishes the proof.
One can also prove that a PID is an elementary divisor domain (insert future reference here), by proving lemmas similar to the following.
Lemma
Let \(R\) be a Bézout domain. Let \(n \geq 1\) and \(f_1, \ldots, f_n \in R\) generate the unit ideal. There exists an invertible \(n \times n\) matrix in \(R\) whose first row is \(f_1 \ldots f_n\).
Proof
This follows from Lemma 0ASU but we can also prove it directly as follows. By induction on \(n\). The result holds for \(n = 1\). Assume \(n > 1\). We may assume \(f_1 \not = 0\) after renumbering. Choose \(f \in R\) such that \((f) = (f_1, \ldots, f_{n - 1})\). Let \(A\) be an \((n - 1) \times (n - 1)\) matrix whose first row is \(f_1/f, \ldots, f_{n - 1}/f\). Choose \(a, b \in R\) such that \(af - bf_n = 1\) which is possible because \(1 \in (f_1, \ldots, f_n) = (f, f_n)\). Then a solution is the matrix \[\left( \begin{matrix} f & 0 & \ldots & 0 & f_n \\ 0 & 1 & \ldots & 0 & 0 \\ & & \ldots \\ 0 & 0 & \ldots & 1 & 0 \\ b & 0 & \ldots & 0 & a \end{matrix} \right) \left( \begin{matrix} & & & 0 \\ & A \\ & & & 0 \\ 0 & \ldots & 0 & 1 \end{matrix} \right)\] Observe that the left matrix is invertible because it has determinant \(1\).
Principal radical ideals
In this section we prove that for a catenary Noetherian normal local domain there exists a nontrivial principal radical ideal. This result can be found in [Artin-Lipman].
Lemma
Let \((R,\mathfrak m)\) be a Noetherian local ring of dimension one, and let \(x\in\mathfrak m\) be an element not contained in any minimal prime of \(R\). Then
the function \(P : n \mapsto \text{length}_R(R/x^n R)\) satisfies \(P(n) \leq n P(1)\) for \(n \geq 0\),
if \(x\) is a nonzerodivisor, then \(P(n) = nP(1)\) for \(n \geq 0\).
Proof
Since \(\dim(R) = 1\), we have \(\dim(R/x^n R) = 0\) and so \(\text{length}_R(R/x^n R)\) is finite for each \(n\) (Algebra, Lemma 00L5). To show the lemma we will induct on \(n\). Since \(x^0 R = R\), we have that \(P(0) = \text{length}_R(R/x^0R) = \text{length}_R 0 = 0\). The statement also holds for \(n = 1\). Now let \(n \geq 2\) and suppose the statement holds for \(n - 1\). The following sequence is exact \[R/x^{n-1}R \xrightarrow{x} R/x^nR \to R/xR \to 0\] where \(x\) denotes the multiplication by \(x\) map. Since length is additive (Algebra, Lemma 00IV), we have that \(P(n) \leq P(n - 1) + P(1)\). By induction \(P(n - 1) \leq (n - 1)P(1)\), whence \(P(n) \leq nP(1)\). This proves the induction step.
If \(x\) is a nonzerodivisor, then the displayed exact sequence above is exact on the left also. Hence we get \(P(n) = P(n - 1) + P(1)\) for all \(n \geq 1\).
Lemma
Let \((R, \mathfrak m)\) be a Noetherian local ring of dimension \(1\). Let \(x \in \mathfrak m\) be an element not contained in any minimal prime of \(R\). Let \(t\) be the number of minimal prime ideals of \(R\). Then \(t \leq \text{length}_R(R/xR)\).
Proof
Let \(\mathfrak p_1, \ldots, \mathfrak p_t\) be the minimal prime ideals of \(R\). Set \(R' = R/\sqrt{0} = R/(\bigcap_{i = 1}^t \mathfrak p_i)\). We claim it suffices to prove the lemma for \(R'\). Namely, it is clear that \(R'\) has \(t\) minimal primes too and \(\text{length}_{R'}(R'/xR') = \text{length}_R(R'/xR')\) is less than \(\text{length}_R(R/xR)\) as there is a surjection \(R/xR \to R'/xR'\). Thus we may assume \(R\) is reduced.
Assume \(R\) is reduced with minimal primes \(\mathfrak p_1, \ldots, \mathfrak p_t\). This means there is an exact sequence \[0 \to R \to \prod\nolimits_{i = 1}^t R/\mathfrak p_i \to Q \to 0\] Here \(Q\) is the cokernel of the first map. Write \(M = \prod_{i = 1}^t R/\mathfrak p_i\). Localizing at \(\mathfrak p_j\) we see that \[R_{\mathfrak p_j} \to M_{\mathfrak p_j} = \left(\prod\nolimits_{i=1}^t R/\mathfrak p_i\right)_{\mathfrak p_j} = (R/\mathfrak p_j)_{\mathfrak p_j}\] is surjective. Thus \(Q_{\mathfrak p_j} = 0\) for all \(j\). We conclude that \(\text{Supp}(Q) = \{\mathfrak m\}\) as \(\mathfrak m\) is the only prime of \(R\) different from the \(\mathfrak p_i\). It follows that \(Q\) has finite length (Algebra, Lemma 00L5). Since \(\text{Supp}(Q) = \{\mathfrak m\}\) we can pick an \(n \gg 0\) such that \(x^n\) acts as \(0\) on \(Q\) (Algebra, Lemma 00L6). Now consider the diagram \[\xymatrix{ 0 \ar[r] & R \ar[r] \ar[d]^-{x^n} & M \ar[r] \ar[d]^-{x^n} & Q \ar[r] \ar[d]^-{x^n} & 0 \\ 0 \ar[r] & R \ar[r] & M \ar[r] & Q \ar[r] & 0 }\] where the vertical maps are multiplication by \(x^n\). This is injective on \(R\) and on \(M\) since \(x\) is not contained in any of the \(\mathfrak p_i\). By the snake lemma (Algebra, Lemma 07JW), the following sequence is exact: \[0 \to Q \to R/x^nR \to M/x^nM \to Q \to 0\] Hence we find that \(\text{length}_R(R/x^nR) = \text{length}_R(M/x^nM)\) for large enough \(n\). Writing \(R_i = R/\mathfrak p_i\) we see that \(\text{length}(M/x^nM) = \sum_{i = 1}^t \text{length}_R(R_i/x^nR_i)\). Applying Lemma 0BWS and the fact that \(x\) is a nonzerodivisor on \(R\) and \(R_i\), we conclude that \[n \text{length}_R(R/xR) = \sum\nolimits_{i = 1}^t n \text{length}_{R_i}(R_i/x R_i)\] Since \(\text{length}_{R_i}(R_i/x R_i) \geq 1\) the lemma is proved.
Lemma
Let \((R,\mathfrak m)\) be a Noetherian local ring of dimension \(d > 1\), let \(f \in \mathfrak m\) be an element not contained in any minimal prime ideal of \(R\), and let \(k\in\mathbf{N}\). Then there exist elements \(g_1, \ldots, g_{d - 1} \in \mathfrak m^k\) such that \(f, g_1, \ldots, g_{d - 1}\) is a system of parameters.
Proof
We have \(\dim(R/fR) = d - 1\) by Algebra, Lemma 00KW. Choose a system of parameters \(\overline{g}_1, \ldots, \overline{g}_{d - 1}\) in \(R/fR\) (Algebra, Proposition 00KQ) and take lifts \(g_1, \ldots, g_{d - 1}\) in \(R\). It is straightforward to see that \(f, g_1, \ldots, g_{d - 1}\) is a system of parameters in \(R\). Then \(f, g_1^k, \ldots, g_{d - 1}^k\) is also a system of parameters and the proof is complete.
Lemma
Let \((R,\mathfrak m)\) be a Noetherian local ring of dimension two, and let \(f \in \mathfrak m\) be an element not contained in any minimal prime ideal of \(R\). Then there exist \(g \in \mathfrak m\) and \(N \in \mathbf{N}\) such that
\(f,g\) form a system of parameters for \(R\).
If \(h \in \mathfrak m^N\), then \(f + h, g\) is a system of parameters and \(\text{length}_R (R/(f, g)) = \text{length}_R(R/(f + h, g))\).
Proof
By Lemma 0BWU there exists a \(g \in \mathfrak m\) such that \(f, g\) is a system of parameters for \(R\). Then \(\mathfrak m = \sqrt{(f, g)}\). Thus there exists an \(n\) such that \(\mathfrak m^n \subset (f, g)\), see Algebra, Lemma 00IM. We claim that \(N = n + 1\) works. Namely, let \(h \in \mathfrak m^N\). By our choice of \(N\) we can write \(h = af + bg\) with \(a, b \in \mathfrak m\). Thus \[(f + h, g) = (f + af + bg, g) = ((1 + a)f, g) = (f, g)\] because \(1 + a\) is a unit in \(R\). This proves the equality of lengths and the fact that \(f + h, g\) is a system of parameters.
Lemma
Let \(R\) be a Noetherian local normal domain of dimension \(2\). Let \(\mathfrak p_1, \ldots, \mathfrak p_r\) be pairwise distinct primes of height \(1\). There exists a nonzero element \(f \in \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) such that \(R/fR\) is reduced.
Proof
Let \(f \in \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) be a nonzero element. We will modify \(f\) slightly to obtain an element that generates a radical ideal. The localization \(R_\mathfrak p\) of \(R\) at each height one prime ideal \(\mathfrak p\) is a discrete valuation ring, see Algebra, Lemma 00PD or Algebra, Lemma 031S. We denote by \(\text{ord}_\mathfrak p(f)\) the corresponding valuation of \(f\) in \(R_{\mathfrak p}\). Let \(\mathfrak q_1, \ldots, \mathfrak q_s\) be the distinct height one prime ideals containing \(f\). Write \(\text{ord}_{\mathfrak q_j}(f) = m_j \geq 1\) for each \(j\). Then we define \(\text{div}(f) = \sum_{j = 1}^s m_j\mathfrak q_j\) as a formal linear combination of height one primes with integer coefficients. Note for later use that each of the primes \(\mathfrak p_i\) occurs among the primes \(\mathfrak q_j\). The ring \(R/fR\) is reduced if and only if \(m_j = 1\) for \(j = 1, \ldots, s\). Namely, if \(m_j\) is \(1\) then \((R/fR)_{\mathfrak q_j}\) is reduced and \(R/fR \subset \prod (R/fR)_{\mathfrak q_j}\) as \(\mathfrak q_1, \ldots, \mathfrak q_s\) are the associated primes of \(R/fR\), see Algebra, Lemmas 0311 and 031T.
Choose and fix \(g\) and \(N\) as in Lemma 0BWV. For a nonzero \(y \in R\) denote \(t(y)\) the number of primes minimal over \(y\). Since \(R\) is a normal domain, these primes are height one and correspond \(1\)-to-\(1\) to the minimal primes of \(R/yR\) (Algebra, Lemmas 00KV and 031T). For example \(t(f) = s\) is the number of primes \(\mathfrak q_j\) occurring in \(\text{div}(f)\). Let \(h \in \mathfrak m^N\). By Lemma 0BWT we have \[\begin{align*} t(f + h) & \leq \text{length}_{R/(f + h)}(R/(f + h, g)) \\ & = \text{length}_R(R/(f + h, g)) \\ & = \text{length}_R(R/(f, g)) \end{align*}\] see Algebra, Lemma 00IX for the first equality. Therefore we see that \(t(f + h)\) is bounded independent of \(h \in \mathfrak m^N\).
By the boundedness proved above we may pick \(h \in \mathfrak m^N \cap \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) such that \(t(f + h)\) is maximal among such \(h\). Set \(f' = f + h\). Given \(h' \in \mathfrak m^N \cap \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) we see that the number \(t(f' + h') \leq t(f + h)\). Thus after replacing \(f\) by \(f'\) we may assume that for every \(h \in \mathfrak m^N \cap \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) we have \(t(f + h) \leq s\).
Next, assume that we can find an element \(h \in \mathfrak m^N\) such that for each \(j\) we have \(\text{ord}_{\mathfrak q_j}(h) \geq 1\) and \(\text{ord}_{\mathfrak q_j}(h) = 1 \Leftrightarrow m_j > 1\). Observe that \(h \in \mathfrak m^N \cap \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\). Then \(\text{ord}_{\mathfrak q_j}(f + h) = 1\) for every \(j\) by elementary properties of valuations. Thus \[\text{div}(f + h) = \sum\nolimits_{j = 1}^s \mathfrak q_j + \sum\nolimits_{k = 1}^v e_k \mathfrak r_k\] for some pairwise distinct height one prime ideals \(\mathfrak r_1, \ldots, \mathfrak r_v\) and \(e_k \geq 1\). However, since \(s = t(f) \geq t(f + h)\) we see that \(v = 0\) and we have found the desired element.
Now we will pick \(h\) that satisfies the above criteria. By prime avoidance (Algebra, Lemma 00DS) for each \(1 \leq j \leq s\) we can find an element \(a_j \in \mathfrak q_j\) such that \(a_j \not \in \mathfrak q_{j'}\) for \(j' \not = j\) and \(a_j \not \in \mathfrak q_j^{(2)}\). Here \(\mathfrak q_j^{(2)} = \{x \in R \mid \text{ord}_{\mathfrak q_j}(x) \geq 2\}\) is the second symbolic power of \(\mathfrak q_j\). Then we take \[h = \prod\nolimits_{m_j = 1} a_j^2 \times \prod\nolimits_{m_j > 1} a_j\] Then \(h\) clearly satisfies the conditions on valuations imposed above. If \(h \not \in \mathfrak m^N\), then we multiply by an element of \(\mathfrak m^N\) which is not contained in \(\mathfrak q_j\) for all \(j\).
Lemma
Let \((A, \mathfrak m, \kappa)\) be a Noetherian normal local domain of dimension \(2\). If \(a \in \mathfrak m\) is nonzero, then there exists an element \(c \in A\) such that \(A/cA\) is reduced and such that \(a\) divides \(c^n\) for some \(n\).
Proof
Let \(\text{div}(a) = \sum_{i = 1}^r n_i \mathfrak p_i\) with notation as in the proof of Lemma 0AXH. Choose \(c \in \mathfrak p_1 \cap \ldots \cap \mathfrak p_r\) with \(A/cA\) reduced, see Lemma 0AXH. For \(n \geq \max(n_i)\) we see that \(-\text{div}(a) + \text{div}(c^n)\) is an effective divisor (all coefficients nonnegative). Thus \(c^n/a \in A\) by Algebra, Lemma 031T.
In the rest of this section we prove the result in dimension \(> 2\).
Lemma
Let \((R, \mathfrak m)\) be a Noetherian local ring of dimension \(d\), let \(g_1, \ldots, g_d\) be a system of parameters, and let \(I = (g_1, \ldots, g_d)\). If \(e_I/d!\) is the leading coefficient of the numerical polynomial \(n \mapsto \text{length}_R(R/I^{n+1})\), then \(e_I \leq \text{length}_R(R/I)\).
Proof
The function is a numerical polynomial by Algebra, Proposition 00K8. It has degree \(d\) by Algebra, Proposition 00KQ. If \(d = 0\), then the result is trivial. If \(d = 1\), then the result is Lemma 0BWS. To prove it in general, observe that there is a surjection \[\bigoplus\nolimits_{i_1, \ldots, i_d \geq 0,\ \sum i_j = n} R/I \longrightarrow I^n/I^{n + 1}\] sending the basis element corresponding to \(i_1, \ldots, i_d\) to the class of \(g_1^{i_1} \ldots g_d^{i_d}\) in \(I^n/I^{n + 1}\). Thus we see that \[\text{length}_R(R/I^{n + 1}) - \text{length}_R(R/I^n) \leq \text{length}_R(R/I) {n + d - 1 \choose d - 1}\] Since \(d \geq 2\) the numerical polynomial on the left has degree \(d - 1\) with leading coefficient \(e_I / (d - 1)!\). The polynomial on the right has degree \(d - 1\) and its leading coefficient is \(\text{length}_R(R/I)/ (d - 1)!\). This proves the lemma.
Lemma
Let \((R, \mathfrak m)\) be a Noetherian local ring of dimension \(d\), let \(t\) be the number of minimal prime ideals of \(R\) of dimension \(d\), and let \((g_1,\ldots,g_d)\) be a system of parameters. Then \(t \leq \text{length}_R(R/(g_1,\ldots,g_d))\).
Proof
If \(d = 0\) the lemma is trivial. If \(d = 1\) the lemma is Lemma 0BWT. Thus we may assume \(d > 1\). Let \(\mathfrak p_1, \ldots, \mathfrak p_s\) be the minimal prime ideals of \(R\) where the first \(t\) have dimension \(d\), and denote \(I = (g_1, \ldots, g_d)\). Arguing in exactly the same way as in the proof of Lemma 0BWT we can assume \(R\) is reduced.
Assume \(R\) is reduced with minimal primes \(\mathfrak p_1, \ldots, \mathfrak p_s\). This means there is an exact sequence \[0 \to R \to \prod\nolimits_{i = 1}^s R/\mathfrak p_i \to Q \to 0\] Here \(Q\) is the cokernel of the first map. Write \(M = \prod_{i = 1}^s R/\mathfrak p_i\). Localizing at \(\mathfrak p_j\) we see that \[R_{\mathfrak p_j} \to M_{\mathfrak p_j} = \left(\prod\nolimits_{i=1}^s R/\mathfrak p_i\right)_{\mathfrak p_j} = (R/\mathfrak p_j)_{\mathfrak p_j}\] is surjective. Thus \(Q_{\mathfrak p_j} = 0\) for all \(j\). Therefore no height \(0\) prime of \(R\) is in the support of \(Q\). It follows that the degree of the numerical polynomial \(n \mapsto \text{length}_R(Q/I^nQ)\) equals \(\dim(\text{Supp}(Q)) < d\), see Algebra, Lemma 00L8. By Algebra, Lemma 00KC (which applies as \(R\) does not have finite length) the polynomial \[n \longmapsto \text{length}_R(M/I^nM) - \text{length}_R(R/I^n) - \text{length}_R(Q/I^nQ)\] has degree \(< d\). Since \(M = \prod R/\mathfrak p_i\) and since \(n \to \text{length}_R(R/\mathfrak p_i + I^n)\) is a numerical polynomial of degree exactly(!) \(d\) for \(i = 1, \ldots, t\) (by Algebra, Lemma 00L8) we see that the leading coefficient of \(n \mapsto \text{length}_R(M/I^nM)\) is at least \(t/d!\). Thus we conclude by Lemma 0BWW.
Lemma
Let \((R, \mathfrak m)\) be a Noetherian local ring of dimension \(d\), and let \(f \in \mathfrak m\) be an element not contained in any minimal prime ideal of \(R\). Then there exist elements \(g_1, \ldots, g_{d - 1} \in \mathfrak m\) and \(N \in \mathbf{N}\) such that
\(f, g_1, \ldots, g_{d - 1}\) form a system of parameters for \(R\)
If \(h \in \mathfrak m^N\), then \(f + h, g_1, \ldots, g_{d - 1}\) is a system of parameters and we have \(\text{length}_R R/(f, g_1, \ldots, g_{d-1}) = \text{length}_R R/(f + h, g_1, \ldots, g_{d-1})\).
Proof
By Lemma 0BWU there exist \(g_1, \ldots, g_{d - 1} \in \mathfrak m\) such that \(f, g_1, \ldots, g_{d - 1}\) is a system of parameters for \(R\). Then \(\mathfrak m = \sqrt{(f, g_1, \ldots, g_{d - 1})}\). Thus there exists an \(n\) such that \(\mathfrak m^n \subset (f, g_1, \ldots, g_{d - 1})\), see Algebra, Lemma 00IM. We claim that \(N = n + 1\) works. Namely, let \(h \in \mathfrak m^N\). By our choice of \(N\) we can write \(h = af + \sum b_ig_i\) with \(a, b_i \in \mathfrak m\). Thus \[\begin{align*} (f + h, g_1, \ldots, g_{d - 1}) & = (f + af + \sum b_ig_i, g_1, \ldots, g_{d - 1}) \\ & = ((1 + a)f, g_1, \ldots, g_{d - 1}) \\ & = (f, g_1, \ldots, g_{d - 1}) \end{align*}\] because \(1 + a\) is a unit in \(R\). This proves the equality of lengths and the fact that \(f + h, g_1, \ldots, g_{d - 1}\) is a system of parameters.
Proposition
Let \(R\) be a catenary Noetherian local normal domain. Let \(J \subset R\) be a radical ideal. Then there exists a nonzero element \(f \in J\) such that \(R/fR\) is reduced.
Proof
The proof is the same as that of Lemma 0AXH, using Lemma 0BWX instead of Lemma 0BWT and Lemma 0BWY instead of Lemma 0BWV. We can use Lemma 0BWX because \(R\) is a catenary domain, so every height one prime ideal of \(R\) has dimension \(d - 1\), and hence the spectrum of \(R/(f + h)\) is equidimensional. For the convenience of the reader we write out the details.
Let \(f \in J\) be a nonzero element. We will modify \(f\) slightly to obtain an element that generates a radical ideal. The localization \(R_\mathfrak p\) of \(R\) at each height one prime ideal \(\mathfrak p\) is a discrete valuation ring, see Algebra, Lemma 00PD or Algebra, Lemma 031S. We denote by \(\text{ord}_\mathfrak p(f)\) the corresponding valuation of \(f\) in \(R_{\mathfrak p}\). Let \(\mathfrak q_1, \ldots, \mathfrak q_s\) be the distinct height one prime ideals containing \(f\). Write \(\text{ord}_{\mathfrak q_j}(f) = m_j \geq 1\) for each \(j\). Then we define \(\text{div}(f) = \sum_{j = 1}^s m_j\mathfrak q_j\) as a formal linear combination of height one primes with integer coefficients. The ring \(R/fR\) is reduced if and only if \(m_j = 1\) for \(j = 1, \ldots, s\). Namely, if \(m_j\) is \(1\) then \((R/fR)_{\mathfrak q_j}\) is reduced and \(R/fR \subset \prod (R/fR)_{\mathfrak q_j}\) as \(\mathfrak q_1, \ldots, \mathfrak q_s\) are the associated primes of \(R/fR\), see Algebra, Lemmas 0311 and 031T.
Choose and fix \(g_1, \ldots, g_{d - 1}\) and \(N\) as in Lemma 0BWY. For a nonzero \(y \in R\) denote \(t(y)\) the number of primes minimal over \(y\). Since \(R\) is a normal domain, these primes are height one and correspond \(1\)-to-\(1\) to the minimal primes of \(R/yR\) (Algebra, Lemmas 00KV and 031T). For example \(t(f) = s\) is the number of primes \(\mathfrak q_j\) occurring in \(\text{div}(f)\). Let \(h \in \mathfrak m^N\). Because \(R\) is catenary, for each height one prime \(\mathfrak p\) of \(R\) we have \(\dim(R/\mathfrak p) = d - 1\). Hence by Lemma 0BWX we have \[\begin{align*} t(f + h) & \leq \text{length}_{R/(f + h)}(R/(f + h, g_1, \ldots, g_{d - 1})) \\ & = \text{length}_R(R/(f + h, g_1, \ldots, g_{d - 1})) \\ & = \text{length}_R(R/(f, g_1, \ldots, g_{d - 1})) \end{align*}\] see Algebra, Lemma 00IX for the first equality. Therefore we see that \(t(f + h)\) is bounded independent of \(h \in \mathfrak m^N\).
By the boundedness proved above we may pick \(h \in \mathfrak m^N \cap J\) such that \(t(f + h)\) is maximal among such \(h\). Set \(f' = f + h\). Given \(h' \in \mathfrak m^N \cap J\) we see that the number \(t(f' + h') \leq t(f + h)\). Thus after replacing \(f\) by \(f'\) we may assume that for every \(h \in \mathfrak m^N \cap J\) we have \(t(f + h) \leq s\).
Next, assume that we can find an element \(h \in \mathfrak m^N \cap J\) such that for each \(j\) we have \(\text{ord}_{\mathfrak q_j}(h) \geq 1\) and \(\text{ord}_{\mathfrak q_j}(h) = 1 \Leftrightarrow m_j > 1\). Then \(\text{ord}_{\mathfrak q_j}(f + h) = 1\) for every \(j\) by elementary properties of valuations. Thus \[\text{div}(f + h) = \sum\nolimits_{j = 1}^s \mathfrak q_j + \sum\nolimits_{k = 1}^v e_k \mathfrak r_k\] for some pairwise distinct height one prime ideals \(\mathfrak r_1, \ldots, \mathfrak r_v\) and \(e_k \geq 1\). However, since \(s = t(f) \geq t(f + h)\) we see that \(v = 0\) and we have found the desired element.
Now we will pick \(h\) that satisfies the above criteria. By prime avoidance (Algebra, Lemma 00DS) for each \(1 \leq j \leq s\) we can find an element \(a_j \in \mathfrak q_j \cap J\) such that \(a_j \not \in \mathfrak q_{j'}\) for \(j' \not = j\). Next, we can pick \(b_j \in J \cap \mathfrak q_1 \cap \ldots \cap \mathfrak q_s\) with \(b_j \not \in \mathfrak q_j^{(2)}\). Here \(\mathfrak q_j^{(2)} = \{x \in R \mid \text{ord}_{\mathfrak q_j}(x) \geq 2\}\) is the second symbolic power of \(\mathfrak q_j\). Prime avoidance applies because the ideal \(J' = J \cap \mathfrak q_1 \cap \ldots \cap \mathfrak q_s\) is radical, hence \(R/J'\) is reduced, hence \((R/J')_{\mathfrak q_j}\) is reduced, hence \(J'\) contains an element \(x\) with \(\text{ord}_{\mathfrak q_j}(x) = 1\), hence \(J' \not \subset \mathfrak q_j^{(2)}\). Then the element \[c = \sum\nolimits_{j = 1, \ldots, s} b_j \times \prod\nolimits_{j' \not = j} a_{j'}\] is an element of \(J\) with \(\text{ord}_{\mathfrak q_j}(c) = 1\) for all \(j = 1, \ldots, s\) by elementary properties of valuations. Finally, we let \[h = c \times \prod\nolimits_{m_j = 1} a_j \times y\] where \(y \in \mathfrak m^N\) is an element which is not contained in \(\mathfrak q_j\) for all \(j\).
Invertible objects in the derived category
We characterize invertible objects in the derived category of a ring.
Lemma
Let \(R\) be a ring. The derived category \(D(R)\) of \(R\) is a symmetric monoidal category with tensor product given by derived tensor product and associativity and commutativity constraints as in Section 0FNG.
Proof
Omitted. Hints: The associativity constraint is the isomorphism of Lemma 0BYJ and the commutativity constraint is the isomorphism of Lemma 0BYI. Having said this the commutativity of various diagrams follows from the corresponding result for the category of complexes of \(R\)-modules, see Section 0GWN.
Thus we know what it means for an object of \(D(R)\) to have a (left) dual or to be invertible. Before we can work out what this amounts to we need a simple lemma.
Lemma
Let \(R\) be a ring. Let \(F^\bullet\) be a bounded above complex of free \(R\)-modules. Given pairs \((n_i, f_i)\), \(i = 1, \ldots, N\) with \(n_i \in \mathbf{Z}\) and \(f_i \in F^{n_i}\) there exists a subcomplex \(G^\bullet \subset F^\bullet\) containing all \(f_i\) which is bounded and consists of finite free \(R\)-modules.
Proof
By descending induction on \(a = \min(n_i; i = 1, \ldots, N)\). If \(F^n = 0\) for \(n \geq a\), then the result is true with \(G^\bullet\) equal to the zero complex. In general, after renumbering we may assume there exists \(1 \leq r \leq N\) such that \(n_1 = \ldots = n_r = a\) and \(n_i > a\) for \(i > r\). Choose a basis \(b_j, j \in J\) for \(F^a\). We can choose a finite subset \(J' \subset J\) such that \(f_i \in \bigoplus_{j \in J'} Rb_j\) for \(i = 1, \ldots, r\). Choose a basis \(c_k, k \in K\) for \(F^{a + 1}\). We can choose a finite subset \(K' \subset K\) such that \(\text{d}_F^a(b_j) \in \bigoplus_{k \in K'} Rc_k\) for \(j \in J'\). Then we can apply the induction hypothesis to find a subcomplex \(H^\bullet \subset F^\bullet\) containing \(c_k \in F^{a + 1}\) for \(k \in K'\) and \(f_i \in F^{n_i}\) for \(i > r\). Take \(G^\bullet\) equal to \(H^\bullet\) in degrees \(> a\) and equal to \(\bigoplus_{j \in J'} Rb_j\) in degree \(a\).
Lemma
Let \(R\) be a ring. Let \(M\) be an object of \(D(R)\). The following are equivalent
\(M\) has a left dual in \(D(R)\) as in Categories, Definition 0FFP,
\(M\) is a perfect object of \(D(R)\).
Moreover, in this case the left dual of \(M\) is the object \(M^\vee\) of Lemma 07VI.
Proof
If \(M\) is perfect, then we can represent \(M\) by a bounded complex \(M^\bullet\) of finite projective \(R\)-modules. In this case \(M^\bullet\) has a left dual in the category of complexes by Lemma 0FNK which is a fortiori a left dual in \(D(R)\).
Assume (1). Say \(N\), \(\eta : R \to M \otimes_R^\mathbf{L} N\), and \(\epsilon : M \otimes_R^\mathbf{L} N \to R\) is a left dual as in Categories, Definition 0FFP. Choose a complex \(M^\bullet\) representing \(M\). Choose a K-flat complex \(N^\bullet\) with flat terms representing \(N\), see Lemma 06Y4. Then \(\eta\) is given by a map of complexes \[\eta : R \longrightarrow \text{Tot}(M^\bullet \otimes_R N^\bullet)\] We can write the image of \(1\) as a finite sum \[\eta(1) = \sum\nolimits_n \sum\nolimits_i m_{n, i} \otimes n_{-n, i}\] with \(m_{n, i} \in M^n\) and \(n_{-n, i} \in N^{-n}\). Let \(K^\bullet \subset M^\bullet\) be the subcomplex generated by all the elements \(m_{n, i}\) and \(\text{d}(m_{n, i})\). By our choice of \(N^\bullet\) we find that \(\text{Tot}(K^\bullet \otimes_R N^\bullet) \subset \text{Tot}(M^\bullet \otimes_R N^\bullet)\) and \(\eta(1)\) is in the subcomplex by our choice above. Denote \(K\) the object of \(D(R)\) represented by \(K^\bullet\). Then we see that \(\eta\) factors over a map \(\tilde \eta : R \longrightarrow K \otimes_R^\mathbf{L} N\). Since \((1 \otimes \epsilon) \circ (\eta \otimes 1) = \text{id}_M\) we conclude that the identity on \(M\) factors through \(K\) by the commutative diagram \[\xymatrix{ M \ar[rr]_-{\eta \otimes 1} \ar[rrd]_{\tilde \eta \otimes 1} & & M \otimes_R^\mathbf{L} N \otimes_R^\mathbf{L} M \ar[r]_-{1 \otimes \epsilon} & M \\ & & K \otimes_R^\mathbf{L} N \otimes_R^\mathbf{L} M \ar[u] \ar[r]^-{1 \otimes \epsilon} & K \ar[u] }\] Since \(K\) is bounded above it follows that \(M \in D^-(R)\). Thus we can represent \(M\) by a bounded above complex \(M^\bullet\) of free \(R\)-modules, see for example Derived Categories, Lemma 05T7. Write \(\eta(1) = \sum\nolimits_n \sum\nolimits_i m_{n, i} \otimes n_{-n, i}\) as before. By Lemma 0FNR we can find a subcomplex \(K^\bullet \subset M^\bullet\) containing all the elements \(m_{n, i}\) which is bounded and consists of finite free \(R\)-modules. As above we find that the identity on \(M\) factors through \(K\). Since \(K\) is perfect we conclude \(M\) is perfect too, see Lemma 066S.
Lemma
Let \(R\) be a ring. Let \(M\) be an object of \(D(R)\). The following are equivalent
\(M\) is invertible in \(D(R)\), see Categories, Definition 0FFN, and
for every prime ideal \(\mathfrak p \subset R\) there exists an \(f \in R\), \(f \not \in \mathfrak p\) such that \(M_f \cong R_f[-n]\) for some \(n \in \mathbf{Z}\).
Moreover, in this case
\(M\) is a perfect object of \(D(R)\),
\(M = \bigoplus H^n(M)[-n]\) in \(D(R)\),
each \(H^n(M)\) is a finite projective \(R\)-module,
we can write \(R = \prod_{a \leq n \leq b} R_n\) such that \(H^n(M)\) corresponds to an invertible \(R_n\)-module.
Proof
Assume (2). Consider the object \(R\Hom_R(M, R)\) and the composition map \[R\Hom(M, R) \otimes_R^\mathbf{L} M \to R\] Checking locally we see that this is an isomorphism; we omit the details. Because \(D(R)\) is symmetric monoidal we see that \(M\) is invertible.
Assume (1). Observe that an invertible object of a monoidal category has a left dual, namely, its inverse. Thus \(M\) is perfect by Lemma 0FNS. Consider a prime ideal \(\mathfrak p \subset R\) with residue field \(\kappa\). Then we see that \(M \otimes_R^\mathbf{L} \kappa\) is an invertible object of \(D(\kappa)\). Clearly this implies that \(\dim H^i(M \otimes_R^\mathbf{L} \kappa)\) is nonzero exactly for one \(i\) and equal to \(1\) in that case. By Lemma 0BCD this gives (2).
In the proof above we have seen that (a) holds. Let \(U_n \subset \Spec(R)\) be the union of the opens of the form \(D(f)\) such that \(M_f \cong R_f[-n]\). Clearly, \(U_n \cap U_{n'} = \emptyset\) if \(n \not = n'\). If \(M\) has tor amplitude in \([a, b]\), then \(U_n = \emptyset\) if \(n \not \in [a, b]\). Hence we see that we have a product decomposition \(R = \prod_{a \leq n \leq b} R_n\) as in (d) such that \(U_n\) corresponds to \(\Spec(R_n)\), see Algebra, Lemma 00EM. Since \(D(R) = \prod_{a \leq n \leq b} D(R_n)\) and similarly for the category of modules parts (b), (c), and (d) follow immediately.
Splitting off a free module
The arguments in this section are due to Serre, see [Serre-projective].
Situation
Here \(R\) is a ring and \(M\) is a finitely presented \(R\)-module. Denote \(\Omega \subset \Spec(R)\) the set of closed points with the induced topology. For \(x \in \Omega\) denote \(M(x) = M/xM\) the fibre of \(M\) at \(x\). This is a finite dimensional vector space over the residue field \(\kappa(x)\) at \(x\). Given \(s \in M\) we denote \(s(x)\) the image of \(s\) in \(M(x)\).
Lemma
In Situation 0GV8 let \(x \in \Omega\). There exists a canonical short exact sequence \[0 \to B(x) \to M(x) \to V(x) \to 0\] of \(\kappa(x)\)-vector spaces which has the following property: for \(s_1, \ldots, s_r \in M\) the following are equivalent
there exists an \(f \in R\), \(f \not \in x\) such that the map \(s_1, \ldots, s_r : R^{\oplus r} \to M\) becomes the inclusion of a direct summand after inverting \(f\), and
\(s_1(x), \ldots, s_r(x)\) map to linearly independent elements of \(V(x)\).
Proof
Define \(B(x) \subset M(x)\) as the perpendicular of the image of the map \[\Hom_R(M, R) \to \Hom_{\kappa(x)}(M(x), \kappa(x))\] and set \(V(x) = M(x)/B(x)\). Then any \(R\)-linear map \(\varphi : M \to R\) induces a map \(\overline{\varphi} : V(x) \to \kappa(x)\) and conversely any \(\kappa(x)\)-linear map \(\lambda : V(x) \to \kappa(x)\) is equal to \(\overline{\varphi}\) for some \(\varphi\). Let \(s_1, \ldots, s_r \in M\).
Suppose \(s_1, \ldots, s_r\) map to linearly independent elements of \(V(x)\). Then we can find \(\varphi_1, \ldots, \varphi_r \in \Hom_R(M, R)\) such that \(\varphi_i(s_j)\) maps to \(\delta_{ij}\)22 in \(\kappa(x)\). Hence the matrix of the composition \[R^{\oplus r} \xrightarrow{s_1, \ldots, s_r} M \xrightarrow{\varphi_1, \ldots, \varphi_r} R^{\oplus r}\] has a determinant \(f \in R\) which maps to \(1\) in \(\kappa(x)\) Clearly, this implies that \(s_1, \ldots, s_r : R^{\oplus r} \to M\) is the inclusion of a direct summand after inverting \(f\).
Conversely, suppose that we have an \(f \in R\), \(f \not \in x\) such that \(s_1, \ldots, s_r : R^{\oplus r} \to M\) is the inclusion of a direct summand after inverting \(f\). Hence we can find \(R_f\)-linear maps \(\varphi_i : M_f \to R_f\) such that \(\varphi_i(s_j) = \delta_{ij} \in R_f\). Since \(\Hom_R(M, R)_f = \Hom_{R_f}(M_f, R_f)\) by Algebra, Lemma 0583 we conclude that we can find \(n \geq 0\) and \(\varphi'_i \in \Hom_R(M, R)\) such that \(\varphi'_i(s_j) = f^n\delta_{ij} \in R\). It follows that \(s_1, \ldots, s_r\) map to linearly independent elements of \(V(x)\) as \(\overline{\varphi}'_i(s_j) = f^n\delta_{ij}\).
In Situation 0GV8 given \(s_1, \ldots, s_r \in M\) we denote \(Z(s_1, \ldots, s_r) \subset \Omega\) the set of \(x \in \Omega\) such that \(s_1(x), \ldots, s_r(x)\) map to linearly dependent elements of \(V(x)\). By the lemma this is a closed subset of \(\Omega\).
Lemma
In Situation 0GV8 let \(x_1, \ldots, x_n \in \Omega\) be pairwise distinct. Let \(v_i \in V(x_i)\). Then there exists an \(s \in M\) such that \(s(x_i)\) maps to \(v_i\) for \(i = 1, \ldots, n\).
Proof
Since \(x_i\) is a maximal ideal of \(R\) we may use Algebra, Lemma 00DT to see that \(M(x_1) \oplus \ldots \oplus M(x_n)\) is a quotient of \(M\).
Proposition
In Situation 0GV8 assume \(\Omega\) is a Noetherian topological space. Let \(s_1, \ldots, s_h \in M\). Let \(Z(s_1, \ldots, s_h) \subset F \subset \Omega\) be closed. Let \(x_1, \ldots, x_n \in F\) be pairwise distinct. Let \(v_i \in V(x_i)\). Let \(k \geq 0\) be an integer such that \[(*)\quad h + k \leq \dim_{\kappa(x)} V(x)\text{ for all }x \in \Omega\] Then there exist \(s \in M\) and \(F' \subset \Omega\) closed such that
\(s(x_i)\) maps to \(v_i\),
\(Z(s_1, \ldots, s_h, s) \subset F \cup F'\), and
every irreducible component of \(F'\) has codimension \(\geq k\) in \(\Omega\).
Proof
We note that codimension was defined in Topology, Section 02I0 and that we will use some results on Noetherian topological spaces contained in Topology, Section 0050.
The proof is by induction on \(k\). If \(k = 0\), then we choose \(s \in M\) as in Lemma 0GVA and we choose \(F' = \Omega\).
Assume \(k > 0\). By our induction hypothesis we may choose \(u \in M\) and \(G \subset \Omega\) closed satisfying (a), (b), (c) for \(s_1, \ldots, s_h\), \(F\), \(x_1, \ldots, x_n\), \(v_1, \ldots, v_n\), and \(k - 1\).
Let \(G = G_1 \cup \ldots \cup G_m\) be the decomposition of \(G\) into its irreducible components. If \(G_j \subset F\), then we can remove it from the list. Thus we may assume \(G_j\) is not contained in \(F\) for \(j = 1, \ldots, m\). For \(j = 1, \ldots, m\) choose \(y_j \in G_j\) with \(y_j \not \in F\) and \(y_j \not \in G_{j'}\) for \(j' \not = j\). This is possible as there are no inclusions among the irreducible components of \(G\). Choose \(w_j \in V(y_j)\) not contained in the span of the images of \(s_1(y_j), \ldots, s_h(y_j)\); this is possible because \(h + k \leq \dim V(y_j)\) and \(k > 0\).
Apply the induction hypothesis to the \(h + 1\) sections \(s_1, \ldots, s_h, u\), the closed set \(F \cup G\), the points \(x_1, \ldots, x_n, y_1, \ldots, y_m \in F \cup G\), the elements \(0 \in V(x_i)\) and \(w_j \in V(y_j)\), and the integer \(k - 1\). Note that we have increased \(h\) by \(1\) and decreased \(k\) by \(1\) hence the assumption \((*)\) of the proposition remains valid. This produces \(t \in M\) and \(H \subset \Omega\) closed satisfying (a), (b), (c) for \(s_1, \ldots, s_h, u\), \(F \cup G\), \(x_1, \ldots, x_n, y_1, \ldots, y_m\), \(0, \ldots, 0, w_1, \ldots, w_m\), and \(k - 1\).
Let \(H_1, \ldots, H_p \subset H\) be the irreducible components of \(H\) which are not contained in \(F \cup G\). As before pick \(z_l \in H_l\), \(z_l \not \in F \cup G\) and \(z_l \not \in H_{l'}\) for \(l' \not = l\). Using Algebra, Lemma 00DT we may choose \(f \in R\) such that \(f(y_j) = 1\), \(j = 1, \ldots, m\) and \(f(z_l) = 0\), \(l = 1, \ldots, p\). Claim: the element \(s = u + f t\) works.
First, the value \(s(x_i)\) agrees with \(u(x_i)\) because \(t(x_i) = 0\) and hence we see that \(s(x_i)\) maps to \(v_i\). This proves (a). To finish the proof it suffices to show that every irreducible component \(Z\) of \(Z(s_1, \ldots, s_h, s)\) not contained in \(F\) has codimension \(\geq k\) in \(\Omega\). Namely, then we can set \(F'\) equal to the union of these and we get (b) and (c). We can see that irreducible components \(Z\) of \(Z(s_1, \ldots, s_h, s)\) of codimension \(\leq k - 1\) do not exist as follows:
Observe that \(Z(s_1, \ldots, s_h, s) \subset Z(s_1, \ldots, s_h, u, t) = F \cup H\) as \(s = u + ft\). Hence \(Z \subset H\).
The irreducible components of \(H\) have codimension \(\geq k - 1\). Hence \(Z\) is equal to an irreducible component of \(H\) as \(Z\) has codimension \(\leq k - 1\). Hence \(Z = H_l\) for some \(l \in \{1, \ldots, p\}\) or \(Z = G_j\) for some \(j \in \{1, \ldots, m\}\).
But \(Z = G_j\) is impossible as \(s_1(y_j), \ldots, s_h(y_j)\) map to linearly independent elements of \(V(y_j)\) and \(s(y_j) = u(y_j) + f(y_j) t(y_j) = u(y_j) + t(y_j)\) maps to an element of the form \[\text{linear combination of images of }s_i(y_j) + w_j\] which is linearly independent of the images of \(s_1(y_j), \ldots, s_h(y_j)\) in \(V(y_j)\) by our choice of \(w_j\).
Also \(Z = H_l\) is impossible. Namely, again \(s_1(z_l), \ldots, s_h(z_l)\) map to linearly independent elements of \(V(z_l)\) and \(s(z_l) = u(z_l) + f(z_l) t(z_l) = u(z_l)\) maps to an element of \(V(z_l)\) linearly independent of those as \(z_l \not \in F \cup G\).
This finishes the proof.
Theorem
Let \(R\) be a ring whose max spectrum \(\Omega \subset \Spec(R)\) is a Noetherian topological space of dimension \(d < \infty\). Let \(M\) be a finitely presented \(R\)-module such that for all \(\mathfrak m \in \Omega\) the \(R_\mathfrak m\)-module \(M_\mathfrak m\) has a free direct summand of rank \(> d\). Then \(M \cong R \oplus M'\).
Proof
For \(\mathfrak m \in \Omega\) suppose that \(R_\mathfrak m^{\oplus r}\) is a direct summand of \(M_\mathfrak m\). Then by Algebra, Lemmas 00CR and 05N7 we see that \(R_f^{\oplus r}\) is a direct summand of \(M_f\) for some \(f \in R\), \(f \not \in \mathfrak m\). Hence the assumption means that \(\dim V(x) > d\) for all \(x \in \Omega\) where \(V(x)\) is as in Lemma 0GV9. By Proposition 0GVB applied with \(F = \emptyset\), \(h = 0\) and no \(s_i\), \(n = 0\) and no \(x_i, v_i\), and \(k = d + 1\) we find an \(s \in M\) and \(F' \subset \Omega\) such that every irreducible component of \(F'\) has codimension \(\geq d + 1\) and \(Z(s) \subset F'\). Since \(d = \dim(\Omega)\) this forces \(F' = \emptyset\). Hence \(s : R \to M\) is the inclusion of a direct summand at all maximal ideals. It follows that \(s\) is universally injective, see Algebra, Lemma 05CL. Then \(s\) is split injective by Algebra, Lemma 058L.
Big projective modules are free
In this section we discuss one of the results of [Bass]; we suggest the reader look at the original paper. Our argument will use the slightly simplified proof given in the papers [Akasaki] and [Hinohara].
Lemma
If \(P \oplus Q \cong F\) with \(F\) a nonfinitely generated free module, then \(P \oplus F \cong F\).
Proof
\[F \cong F \oplus F \oplus \ldots \cong P \oplus Q \oplus P \oplus Q \oplus \ldots \cong P \oplus F \oplus F \oplus \ldots \cong P \oplus F\]
Lemma
Let \(R\) be a ring. Let \(P\) be a projective module. There exists a free module \(F\) such that \(P \oplus F\) is free.
Proof
Since \(P\) is projective we see that \(F_0 = P \oplus Q\) is a free module for some module \(Q\). Set \(F = \bigoplus_{n \geq 1} F_0\). Then \(P \oplus F \cong F\) by Lemma 0GVF.
Lemma
Let \(R\) be a ring. Let \(P\) be a projective module. Let \(s \in P\). There exists a finite free module \(F\) and a finite free direct summand \(K \subset F \oplus P\) with \((0, s) \in K\).
Proof
By Lemma 0GVG we can find a (possibly infinite) free module \(F\) such that \(F \oplus P\) is free. Then of course \((0, s)\) is contained in a finite free direct summand \(K \subset F \oplus P\). In turn \(K\) is contained in \(F' \oplus P\) where \(F' \subset F\) is a finite free direct summand.
Lemma
Let \(R\) be a ring with Jacobson radical \(J\) such that \(R/J\) is Noetherian. Let \(P\) be a projective \(R\)-module such that \(P_\mathfrak m\) has infinite rank for all maximal ideals \(\mathfrak m\) of \(R\). Let \(s \in P\) and \(M \subset P\) such that \(Rs + M = P\). Then we can find \(m \in M\) such that \(R(s + m)\) is a free direct summand of \(P\).
Proof
The statement makes sense as \(P_\mathfrak m\) is free by Algebra, Theorem 0593.
Denote \(M' \subset P/JP\) the image of \(M\) and \(s' \in P/JP\) the image of \(s\). Observe that \(R/J s' + M' = P/JP\). Suppose we can find \(m' \in M'\) such that \(R/J(s' + m')\) is a free direct summand of \(P/JP\). Choose \(\varphi' : P/JP \to R/J\) which gives a splitting, i.e., we have \(\varphi'(s' + m') = 1\) in \(R/J\). Then since \(P\) is a projective \(R\)-module we can find a lift \(\varphi : P \to R\) of \(\varphi'\). Choose \(m \in M\) mapping to \(m'\). Then \(\varphi(s + m) \in R\) is congruent to \(1\) modulo \(J\) and hence a unit in \(R\) (Algebra, Lemma 0AME). Whence \(R(s + m)\) is a free direct summand of \(P\). This reduces us to the case discussed in the next paragraph.
Assume \(R\) is Noetherian. Let \(m \in M\) be an element and let \(\varphi_1, \ldots, \varphi_n : P \to R\) be \(R\)-linear maps. Denote \[Z(s + m, \varphi_1, \ldots, \varphi_n) \subset \Spec(R)\] the vanishing locus of \(\varphi_1(s + m), \ldots, \varphi_n(s + m) \in R\).
Suppose \(\mathfrak m\) is a maximal ideal of \(R\) and \(\mathfrak m \in Z(s + m, \varphi_1, \ldots, \varphi_n)\). Set \(K = M \cap \bigcap \Ker(\varphi_i)\). We claim the image of \[K/\mathfrak mK \to P/\mathfrak m P\] has infinite dimension. Namely, the quotient \(P/K\) is a finite \(R\)-module as it is isomorphic to a submodule of \(P/M \oplus R^{\oplus n}\). Thus we see that the kernel of the displayed arrow is a quotient of \(\text{Tor}_1^R(P/K, \kappa(\mathfrak m))\) which is finite by Algebra, Lemma 0AZ4. Combined with the fact that \(P/\mathfrak mP\) has infinite dimension we obtain our claim. Thus we can find a \(t \in K\) which maps to an element \(\overline{t}\) of the vector space \(P/\mathfrak mP\) which is linearly independent of the image \(\overline{s + m}\) of \(s + m\). By linear algebra, we find an \(R\)-linear map \(\overline{\varphi} : P \to \kappa(\mathfrak m)\) such that \(\overline{\varphi}(\overline{t}) = 1\) and \(\overline{\varphi}(\overline{s + m}) = 0\). Since \(P\) is projective, we can find an \(R\)-linear map \(\varphi : P \to R\) lifting \(\overline{\varphi}\). Then we see that the vanishing locus \(Z(s + m + t, \varphi_1, \ldots, \varphi_n, \varphi)\) is contained in \(Z(s + m, \varphi_1, \ldots, \varphi_n)\) but does not contain \(\mathfrak m\), i.e., it is strictly smaller than \(Z(s + m, \varphi_1, \ldots, \varphi_n)\).
Since \(\Spec(R)\) is a Noetherian topological space, we see from the arguments above that we may find \(m \in M\) and \(\varphi_1, \ldots, \varphi_n : P \to R\) such that the closed subset \(Z(s + m, \varphi_1, \ldots, \varphi_n)\) does not contain any closed points of \(\Spec(R)\). Hence \(Z(s + m, \varphi_1, \ldots, \varphi_n) = \emptyset\). Hence we can find \(r_1, \ldots, r_n \in R\) such that \(\sum r_i\varphi_i(s + m) = 1\). Hence \[R \xrightarrow{s + m} P \xrightarrow{\sum r_i \varphi_i} R\] is the desired splitting.
Lemma
Let \(R\) be a ring with Jacobson radical \(J\) such that \(R/J\) is Noetherian. Let \(P\) be a projective \(R\)-module such that \(P_\mathfrak m\) has infinite rank for all maximal ideals \(\mathfrak m\) of \(R\). Let \(s \in P\). Then we can find a finite stably free direct summand \(M \subset P\) such that \(s \in M\).
Proof
By Lemma 0GVH we can find a finite free module \(F\) and a finite free direct summand \(K \subset F \oplus P\) such that \((0, s) \in K\). By induction on the rank of \(F\) we reduce to the case discussed in the next paragraph.
Assume there exists a finite stably free direct summand \(K \subset R \oplus P\) such that \((0, s) \in K\). Choose a complement \(K'\) of \(K\), i.e., such that \(R \oplus P = K \oplus K'\). The projection \(\pi : R \oplus P \to K'\) is surjective, hence by Lemma 0GVI we find a \(p \in P\) such that \(\pi(1, p) \in K'\) generates a free direct summand. Accordingly we write \(K' = R\pi(1, p) \oplus K''\). We see that \[R \oplus P = K \oplus K' = K \oplus R\pi(1, p) \oplus K''\] The projection \(\pi' : P \to K''\) is surjective23 and hence split (as \(K''\) is projective). Thus \(\Ker(\pi') \subset P\) is a direct summand containing \(s\). Finally, by construction we have an isomorphism \[R \oplus \Ker(\pi') \cong K \oplus R\pi(1, p)\] and hence since \(K\) is finite and stably free, so is \(\Ker(\pi')\).
Theorem
Let \(R\) be a ring with Jacobson radical \(J\) such that \(R/J\) is Noetherian. Let \(P\) be a countably generated projective \(R\)-module such that \(P_\mathfrak m\) has infinite rank for all maximal ideals \(\mathfrak m\) of \(R\). Then \(P\) is free.
Proof
We first prove that \(P\) is a countable direct sum of finite stably free modules. Let \(x_1, x_2, \ldots\) be a countable set of generators for \(P\). We inductively construct finite stably free direct summands \(F_1, F_2, \ldots\) of \(P\) such that for all \(n\) we have that \(F_1 \oplus \ldots \oplus F_n\) is a direct summand of \(P\) which contains \(x_1, \ldots, x_n\). Namely, given \(F_1, \ldots, F_n\) with the desired properties, write \[P = F_1 \oplus \ldots \oplus F_n \oplus P'\] and let \(s \in P'\) be the image of \(x_{n + 1}\). By Lemma 0GVJ we can find a finite stably free direct summand \(F_{n + 1} \subset P'\) containing \(s\). Then \(P = \bigoplus_{i = 1}^{\infty} F_i\).
Assume that \(P\) is an infinite direct sum \(P = \bigoplus_{i = 1}^{\infty} F_i\) of nonzero finite stably free modules. The stable freeness of the modules \(F_i\) will be used in the following manner: the rank of each \(F_i\) is constant (and positive). Hence we see that \(P_\mathfrak m\) is free of countably infinite rank for each maximal ideal \(\mathfrak m\) of \(R\). By Lemma 0GVI applied with \(s = 0\) and \(M = P\), we can find a \(t_1 \in P\) such that \(Rt_1\) is a free direct summand of \(P\). Then \(t_1\) is contained in \(F_1 \oplus \ldots \oplus F_{n_1}\) for some \(n_1 > n_0 = 0\). The same reasoning applied to \(\bigoplus_{n > n_1} F_n\) produces an \(n_1 < n_2\) and \(t_2 \in F_{n_1 + 1} \oplus \ldots \oplus F_{n_2}\) which generates a free direct summand. Continuing in this fashion we obtain a free direct summand \[\bigoplus\nolimits_{i \geq 1} t_i : \bigoplus\nolimits_{i \geq 1} R \longrightarrow \bigoplus\nolimits_{i \geq 1} \bigoplus\nolimits_{n_i \geq n > n_{i - 1}} F_n = P\] of infinite rank. Thus we see that \(P \cong Q \oplus F\) for some free \(R\)-module \(F\) of countable rank. Since \(Q\) is countably generated it follows that \(Q \oplus Q' \cong F\) for some module \(Q'\). Then the Eilenberg swindle (Lemma 0GVF) implies that \(Q \oplus F \cong F\) and \(P\) is free.
We include being separated as part of being complete as we’d like to have a unique limits in complete groups. There is a definition of completeness for any topological group, agreeing, modulo the separation issue, with this one in our special case.↩︎
By our conventions this includes separated.↩︎
Thus the \(I\)-adic topology is sometimes called the \(I\)-pre-adic topology.↩︎
It may happen that the \(I\)-adic completion \(M^\wedge\) is not \(I\)-adically complete, even though \(M^\wedge\) is always complete with respect to the limit topology. If \(I\) is finitely generated then the \(I\)-adic topology and the limit topology on \(M^\wedge\) agree, see Algebra, Lemma 05GG and its proof.↩︎
For example, if \(A\) is a G-ring or (quasi-)excellent, see Sections 07GG and 07QS.↩︎
This clashes with what is meant by a pseudo-coherent module in [Bourbaki-CA].↩︎
This is nonstandard notation.↩︎
To use these spectral sequences we have to show that \(\textit{Ab}(\mathbf{N})\) has enough injectives. An inverse system \((I_n)\) of abelian groups is injective if and only if each \(I_n\) is an injective abelian group and the transition maps are split surjections. Every system embeds in one of these. Details omitted.↩︎
If the pro-isomorphism \(a\) is given by maps \(a_n : K_{m_n} \to M_n\) then we obtain a pro-isomorphism on \(\Hom\) using the corresponding maps \(\Hom_\mathcal{D}(L, K_{m_n}) \to \Hom_\mathcal{D}(L, M_n)\). Thus the construction of the arrows on \(\lim\) and \(R^1\lim\) in the proof of Lemma 0H9K agrees with the construction of \(a'\) and \(a''\) in Derived Categories, Remark 0H9J.↩︎
If \(K\) is not bounded below, we can’t use the given reference. In that case, we chooses a complex \(K^\bullet\) representing \(K\) and a free two-term resolution \(0 \to F^{-1} \to F^0 \to A_f \to 0\) (see above). Then we get a double complex with terms \(\Hom_A(F^{-q}, K^p)\) whose totalization computes \(R\Hom(A_f, K)\). The second spectral sequence associated to this double complex (Homology, Section 012X) gives us what we want. Details omitted.↩︎
Namely, if \(E\) is derived complete and \(a : K \to E\) is a map, then the commutative diagram \[\xymatrix{ K \ar[r]_a \ar[d] & E \ar[d]^{\cong} \\ K^\wedge \ar[r]^{a^\wedge} & E^\wedge }\] shows that any map into \(E\) factors through \(K^\wedge\). Choose a distinguished triangle \(K \to K^\wedge \to C\) and apply \({}^\wedge\). We find that \(C^\wedge = 0\). By the above this means that \(\Hom(C, E) = 0\) for \(E\) derived complete and this proves that \(\Hom(K, E) = \Hom(K^\wedge, E)\) as desired.↩︎
For the proof it is enough to show that there exists a sequence \(K \xrightarrow{u} L \to M \to 0\) where \(K\) and \(L\) are \(I\)-adically complete \(A\)-modules. This can be shown by choosing a presentation \(F_1 \to F_0 \to M \to 0\) with \(F_i\) free and then setting \(K\) and \(L\) equal to the \(f\)-adic completions of \(F_1\) and \(F_0\). Namely, as \(f\) is a nonzerodivisor these completions will be the derived completions and the sequence will remain exact.↩︎
In particular, for every \(n\) there exists an \(m \geq n\) such that \(K_m^\bullet \to K_n^\bullet\) factors through the map \(K_m^\bullet \to A/(f_1^m, \ldots, f_r^m)\).↩︎
Beware that this functor isn’t exact, i.e., does not transform distinguished triangles into distinguished triangles. See Example 0GSN.↩︎
An alternative is to argue that \(\beta\) occurs as the differential for the spectral sequence for the complex \((M^\bullet)_f\) filtered by the subcomplexes \(f^iM^\bullet\). Yet another argument, which proves something stronger, is to first consider the case \(M^\bullet = A\). Here the short exact sequences \(0 \to f^{i + 1}A/f^{i + 2}A \to f^iA/f^{i + 2}A \to f^iA/f^{i + 1}A \to 0\) define maps \(\beta^i : f^iA/f^{i + 1}A \to f^{i + 1}A/f^{i + 2}A[1]\) in \(D(A)\). Then one computes (arguing similarly to the text) that the composition \(f^iA/f^{i + 1}A \to f^{i + 1}A/f^{i + 2}A[1] \to f^{i + 2}A/f^{i + 3}A[2]\) is zero in \(D(A)\). Since \(M^\bullet \otimes f^iA/f^{i + 1}A = M^\bullet \otimes^\mathbf{L} f^iA/f^{i + 1}A\) by our assumption on \(M^\bullet\) having \(f\)-torsion free terms, we conclude the composition \[(M^\bullet \otimes f^iA/f^{i + 1}A) \to (M^\bullet \otimes f^{i + 1}A/f^{i + 2}A)[1] \to (M^\bullet \otimes f^{i + 2}A/f^{i + 3}A)[2]\] in \(D(A)\) is zero as well.↩︎
Recall that we use the notation \(\text{Gal}\) only in the case of Galois extensions.↩︎
If \(B\) is complete, then we can choose \(\sigma\) to be a ring map. If \(A\) is also complete and \(\sigma\) is a ring map, then \(\sigma\) maps \(\kappa_A\) into \(A\).↩︎
\(a^{-1}(k^{-1}(u')) = a^{-1}(u') + a^{-1}(\tilde h(d(u'))) = a^{-1}(u') + h(d(u')) = b^{-1}(u') = a^{-1}(u)\)↩︎
Kronecker delta.↩︎
Namely, if \(k'' \in K''\) then \(k''\) viewed as an element of \(K'\) can be written as \(k'' = \lambda \pi(1, 0) + \pi(0, q)\) for some \(\lambda \in R\) and \(q \in P\). This means \(k'' = \lambda \pi(1, p) + \pi(0, q - \lambda p)\). This in turn means that \(q - \lambda p\) maps to \(k''\) by the composition \(P \to R \oplus P \xrightarrow{\pi} K' \to K''\) since \(K' \to K''\) annihilates \(\pi(1, p)\).↩︎