Divided power envelopes of weakly regular ideals and their graded rings
Written by Claude Opus 5.5 (Anthropic) on 23 October 2026. The question was posed by Claude while writing this course and solved by GPT-6 Astra (OpenAI) in ChatGPT's Pro mode; the solution was refereed by Claude Opus 5.5 in a separate session, and this lesson was written from it and refereed and revised by Claude Opus 5.5 in two further sessions. Public domain (CC0).
Let \(I\) be an ideal of a ring \(A\), and let \(B=A/I\). The divided power envelope \((D,J,\gamma)\) of \((A,I)\) carries a filtration by the ideals \(J^{[n]}\). These are generated by the products of divided powers of total weight at least \(n\). The divided powers induce a homomorphism of graded \(B\)-algebras \[ c\colon\Gamma_B(I/I^2)\to\operatorname{gr}D. \] Illusie asked whether \(c\) is an isomorphism for every weakly regular ideal. This lesson constructs \(c\) for rings of any topos and proves the following.
- \(c\) is surjective for every ideal, and it is an isomorphism in weights \(0\) and \(1\).
- If \(I\) is weakly regular, \(c\otimes\mathbb Q\) is an isomorphism, in every topos.
- In a topos with enough points, \(c\) is an isomorphism exactly when its stalks are. Its stalks are the maps \(c\) of the stalk pairs \((A_p,I_p)\).
- For envelopes relative to a divided power base, the analogous map is surjective, and it is an isomorphism after tensoring with \(\mathbb Q\). It need not be injective, even for an ideal generated by a nonzerodivisor and a base whose divided power filtration vanishes from some index on. No single inverted integer repairs this.
- The envelope of \((\mathbb F_p[x],(x))\) is not flat over \(\mathbb F_p[x]\).
Whether \(c\) itself is an isomorphism for every weakly regular ideal is not decided in this lesson.
Prerequisites. Weakly regular ideals (Section 2 with 2.3, Proposition 2.2, Definition 2.4, Theorem 3.3, Proposition 4.2); Koszul complexes and the shift formulas (Conventions: the divided power algebra \(\Gamma\) by generators and relations). From the AI-integrated Stacks project: existence of divided power envelopes [Stacks, Tag 07H7].
Conventions
\(T\) is a topos and rings are commutative.
A PD pair \((C,K,\gamma)\) consists of a ring \(C\) of \(T\), an ideal \(K\), and maps of sheaves of sets \(\gamma_n\colon K\to C\), \(n\geq0\). They satisfy \(\gamma_0=1\), \(\gamma_1(x)=x\), \(\gamma_n(K)\subseteq K\) for \(n\geq1\), and the identities \[ \begin{aligned} \gamma_n(x+y)&=\textstyle\sum_{i=0}^n\gamma_i(x)\gamma_{n-i}(y),\\ \gamma_n(ax)&=a^n\gamma_n(x),\\ \gamma_m(x)\gamma_n(x)&=\tbinom{m+n}{m}\gamma_{m+n}(x),\\ \gamma_m(\gamma_n(x))&=e_{m,n}\,\gamma_{mn}(x), \end{aligned} \] the last one for \(m,n\geq1\). Here \(e_{m,n}=(mn)!/(m!(n!)^m)\) is an integer. Iterating the product identity gives \[ \gamma_i(x)^m=\tfrac{(mi)!}{(i!)^m}\gamma_{mi}(x) \] and \(x^n=n!\gamma_n(x)\).
A PD morphism is a ring map that carries the ideal into the ideal and commutes with all the \(\gamma_n\). A pair \((A,I)\) is a ring with an ideal. A morphism of pairs is a ring map carrying the ideal into the ideal.
For a \(B\)-module \(M\), \(\Gamma_B(M)=\bigoplus_n\Gamma^n_B(M)\) is the divided power algebra of the Conventions of the lesson on Koszul complexes. It is the sheaf associated with the construction on sections, with generators \(\gamma_k(\xi)\) and the relations listed there. Let \(G=\bigoplus_nG_n\) be a commutative graded \(B\)-algebra, and let \(g_n\colon M\to G_n\) be maps of sheaves with \(g_0=1\) and \(g_1\) linear, satisfying the relations. On sections the \(g_n\) define compatible ring maps \(\Gamma_{B(U)}(M(U))\to G(U)\), hence a homomorphism \(\Gamma_B(M)\to G\). It is unique, since the \(\gamma_k(\xi)\) generate.
For a sheaf \(X\) of abelian groups, \(X_{\mathbb Q}=X\otimes_{\mathbb Z}\mathbb Q\). Its local sections are locally of the form \(x/s\) with \(s\) a nonzero integer. The functor \(X\mapsto X_{\mathbb Q}\) is exact.
1. Envelopes and their filtration
Definition 1.1. An envelope of a pair \((A,I)\) is a PD pair \((D,J,\gamma)\) with a morphism of pairs \(u\colon(A,I)\to(D,J)\) such that composition with \(u\) is a bijection \[ \begin{aligned} &\operatorname{Hom}_{\mathrm{PD}}\bigl((D,J,\gamma),(C,K,\delta)\bigr)\\ &\quad\to\operatorname{Hom}_{\mathrm{pair}}\bigl((A,I),(C,K)\bigr) \end{aligned} \] for every PD pair \((C,K,\delta)\). Two envelopes are isomorphic by a unique PD isomorphism under \(A\). For rings in the usual sense, envelopes exist [Tag 07H7].
Put \(F^0D=D\). For \(n\geq1\), let \(F^nD\) be the ideal generated by the monomials of weight \(\geq n\). These are the products \(\prod_\nu\gamma_{i_\nu}(y_\nu)\) with \(y_\nu\in J\), \(i_\nu\geq1\) and \(\sum_\nu i_\nu\geq n\). Then \(F^1D=J\). A product of monomials of weights \(\geq a\) and \(\geq b\) is a monomial of weight \(\geq a+b\), so \(F^aD\cdot F^bD\subseteq F^{a+b}D\). Hence \[ \operatorname{gr}D=\textstyle\bigoplus_nF^nD/F^{n+1}D \] is a graded ring.
Lemma 1.2. For \(m,r\geq1\), \(\gamma_m(F^rD)\subseteq F^{mr}D\).
Proof. Let \(z=c\prod_{\nu=1}^k\gamma_{i_\nu}(y_\nu)\) with \(c\in D\) and \(\sum i_\nu\geq r\geq1\), so \(k\geq1\). The scalar identity, with the scalar \(c\prod_{\nu\geq2}\gamma_{i_\nu}(y_\nu)\) and \(\gamma_{i_1}(y_1)\in J\), gives \[ \begin{aligned} \gamma_m(z)&=c^m\textstyle\prod_{\nu\geq2}\gamma_{i_\nu}(y_\nu)^m\\ &\quad\cdot\gamma_m\bigl(\gamma_{i_1}(y_1)\bigr). \end{aligned} \] Here \(\gamma_{i_\nu}(y_\nu)^m\) is an integer multiple of \(\gamma_{mi_\nu}(y_\nu)\), and \(\gamma_m(\gamma_{i_1}(y_1))=e_{m,i_1}\gamma_{mi_1}(y_1)\). So \(\gamma_m(z)\) is an element of \(D\) times a monomial of weight \(m\sum i_\nu\geq mr\).
Every section of \(F^rD\) is locally a finite sum \(z=\sum_tz_t\) of such elements. The addition identity writes \(\gamma_m(z)\) as a sum of products \(\prod_t\gamma_{m_t}(z_t)\) with \(\sum_tm_t=m\). The factors with \(m_t=0\) equal \(1\), and the others lie in \(F^{m_tr}D\). So each product lies in \(F^{mr}D\). \(\square\)
Lemma 1.3. Let \((D,J,\gamma)\) be an envelope of \((A,I)\).
- \(D\) is generated as a ring by \(u(A)\) and the \(\gamma_n(u(x))\), \(x\in I\), \(n\geq1\). The ideal \(J\) is generated by the \(\gamma_n(u(x))\).
- \(u\) induces ring isomorphisms \(B\to D/J\) and \(A/I^2\to D/F^2D\). The second restricts to an isomorphism of \(B\)-modules \[ \alpha\colon I/I^2\to J/F^2D,\quad\bar x\mapsto\overline{u(x)}. \]
Proof. (1) Let \(C\subseteq D\) be the subring generated by the stated sections, and \(K\subseteq C\) the ideal of \(C\) generated by the \(\gamma_n(u(x))\). For \(c\in C\), \(m,n\geq1\), \[ \begin{aligned}&\gamma_m(c\gamma_n(u(x)))\\&\quad =c^me_{m,n}\gamma_{mn}(u(x)) \in K.\end{aligned} \] By the addition identity, \(\gamma_m\) of a finite sum of such elements is a sum of products each of which has a factor in \(K\); so \(\gamma_m(K)\subseteq K\) for \(m\geq1\). Thus \((C,K,\gamma)\) is a PD pair. Moreover \(u\) factors through it, since \(u(x)=\gamma_1(u(x))\). The universal property gives a PD morphism \(r\colon D\to C\) under \(A\). The inclusion composed with \(r\) is a PD endomorphism of \(D\) under \(A\), hence the identity. So \(C=D\), and \(J=r(J)\subseteq K\subseteq J\).
(2) The map \(B\to D/J\). The map \(A\to D/J\) kills \(I\), so it induces a ring map \(B\to D/J\). The zero ideal of \(B\), with \(\gamma_0=1\) and \(\gamma_n=0\) for \(n\geq1\), is a PD pair. So \(A\to B\) induces a PD morphism \(D\to B\) under \(A\), which kills \(J\). The composite \(B\to D/J\to B\) is the identity. The ring \(D/J\) is generated by the image of \(u(A)\) by (1), since \(\gamma_n(u(x))\in J\) for \(n\geq1\). So the other composite is the identity as well.
The zero structure on \(I/I^2\). On the ideal \(I/I^2\) of \(A/I^2\) put \(\gamma_0=1\), \(\gamma_1=\mathrm{id}\), and \(\gamma_n=0\) for \(n\geq2\). Products of two sections of \(I/I^2\) vanish, and the identities hold:
- In the addition identity for \(n\geq2\), the only surviving terms are the products \(\gamma_1(x)\gamma_1(y)=0\).
- The scalar identity is clear.
- In the product identity with \(m,n\geq1\), both sides vanish: the left side is a product of two sections of the ideal, and the right side contains \(\gamma_{m+n}=0\).
- In the iteration identity, for \(n\geq2\) both sides vanish. For \(n=1\) it reads \(\gamma_m(x)=\gamma_m(x)\), since \(e_{m,1}=1\).
The map \(A/I^2\to D/F^2D\). The universal property gives a PD morphism \(\rho\colon D\to A/I^2\) under \(A\). It kills every monomial of weight \(\geq2\), so it kills \(F^2D\), and it carries \(J\) into \(I/I^2\). Since \(u(I^2)\subseteq F^1D\cdot F^1D\subseteq F^2D\), \(u\) induces a ring map \(A/I^2\to D/F^2D\). The composite through \(\rho\) is the identity of \(A/I^2\). The ring \(D/F^2D\) is the image of \(u(A)\), because the \(\gamma_n(u(x))\) with \(n\geq2\) lie in \(F^2D\). So the other composite is the identity as well.
The isomorphism \(\alpha\). The first map carries \(I/I^2\) into \(J/F^2D\), and its inverse, induced by \(\rho\), carries \(J/F^2D\) into \(I/I^2\). So the restriction \(\alpha\) is bijective. It is \(A\)-linear between modules killed by \(I\), hence \(B\)-linear. \(\square\)
2. The comparison map
By Lemma 1.3, \(\operatorname{gr}^0D=D/J\cong B\). So \(\operatorname{gr}D\) is a commutative graded \(B\)-algebra.
Proposition 2.1. Let \(M=I/I^2\). There is a unique homomorphism of graded \(B\)-algebras \[ c\colon\Gamma_B(M)\to\operatorname{gr}D \] with \(c(\gamma_n(\bar x))=\gamma_n(u(x))\bmod F^{n+1}D\) for local sections \(x\) of \(I\). It is the identity of \(B\) in weight \(0\) and \(\alpha\) in weight \(1\), so it is an isomorphism in these weights.
Proof. For \(n\geq1\), \(\gamma_n(u(x))\) is a monomial of weight \(n\), so it lies in \(F^nD\).
Independence of the lift. Let \(h\in I^2\). Then \(u(h)\in F^2D\), and \[ \begin{aligned} &\gamma_n(u(x+h))-\gamma_n(u(x))\\ &\quad=\textstyle\sum_{i=1}^n\gamma_{n-i}(u(x))\gamma_i(u(h)). \end{aligned} \] The term of index \(i\) lies in \(F^{n-i}F^{2i}\subseteq F^{n+i}D\subseteq F^{n+1}D\), by Lemma 1.2. So the map of sheaves \[ \begin{gathered}I\to\operatorname{gr}^nD,\\ x\mapsto\gamma_n(u(x))\bmod F^{n+1}D,\end{gathered} \] agrees on \(x\) and \(x+h\). It therefore factors through the quotient \(I\to M\), the coequalizer of the projection and the addition \(I\times I^2\rightrightarrows I\). Call the factorization \(g_n\), and put \(g_0=1\).
The relations. The product of \(\operatorname{gr}D\) is induced by that of \(D\), and \(B\) acts through \(A\).
- \(g_1=\alpha\) is linear.
- \(g_k(b\xi)=b^kg_k(\xi)\): lift \(b\) to \(a\in A\) and use \(\gamma_k(u(ax))=u(a)^k\gamma_k(u(x))\).
- The addition and product relations hold in \(D\) for the \(\gamma_k(u(x))\), \(\gamma_k(u(y))\), and \(u(x+y)=u(x)+u(y)\).
So the \(g_n\) define \(c\) (Conventions), and \(c\) is unique. In weight \(0\) it is the identity of \(B=\operatorname{gr}^0D\), and in weight \(1\) it is \(\alpha\). Lemma 1.3 (2) gives the last assertion. \(\square\)
Proposition 2.2. For every ideal \(I\), \(c\) is surjective.
Proof. A standard monomial is a product \(a\prod_{t=1}^k\gamma_{s_t}(u(x_t))\) with \(a\in u(A)\), \(x_t\in I\), \(s_t\geq1\) and \(k\geq0\). Its weight is \(\sum s_t\). By Lemma 1.3 (1), every section of \(D\) is locally a finite sum of standard monomials. Every section of \(J\) is locally a sum of products of sections of \(D\) with elements \(\gamma_n(u(x))\), hence a sum of standard monomials of weight \(\geq1\). Products of standard monomials are standard monomials, with the weights added.
Divided powers of standard monomials. Let \(z\) be a standard monomial of weight \(w\geq1\), so \(k\geq1\). The computation of Lemma 1.2 shows that \(\gamma_i(z)\) is an integer multiple of \[ a^i\textstyle\prod_t\gamma_{is_t}(u(x_t)), \] a standard monomial of weight \(iw\). By the addition identity, \(\gamma_i\) of a sum of standard monomials of weight \(\geq1\) is a sum of products of such multiples. So it is a sum of standard monomials of weight \(\geq i\).
The ideal \(F^nD\). Consider a monomial \(\prod_\nu\gamma_{i_\nu}(y_\nu)\) with \(\sum i_\nu\geq n\). Writing each \(y_\nu\) locally as a sum of standard monomials of weight \(\geq1\) shows that the monomial is locally a sum of standard monomials of weight \(\geq n\). Multiplying by a section of \(D\) keeps this property. So \(F^nD\) consists locally of sums of standard monomials of weight \(\geq n\).
Hence \(\operatorname{gr}^nD\) is generated as an abelian sheaf by the classes of standard monomials of weight exactly \(n\). The class of \(a\prod_t\gamma_{s_t}(u(x_t))\) with \(\sum s_t=n\) is the product in \(\operatorname{gr}D\) of the class \(\bar a\in B\) and the classes of the \(\gamma_{s_t}(u(x_t))\). So it is \[ c\bigl(\bar a\textstyle\prod_t\gamma_{s_t}(\bar x_t)\bigr).\qquad\square \]
3. Rational isomorphism
Lemma 3.1. Let \((C,K,\gamma)\) be a PD pair of \(T\).
- The formula \(\gamma_n(y/s)=\gamma_n(y)/s^n\) defines a PD structure on the ideal \(K_{\mathbb Q}\) of \(C_{\mathbb Q}\). The map \(C\to C_{\mathbb Q}\) is a PD morphism.
- If \((D,J,\gamma)\) is an envelope of \((A,I)\), then \(D_{\mathbb Q}\) is an envelope of \((A_{\mathbb Q},I_{\mathbb Q})\), and \(F^n(D_{\mathbb Q})=(F^nD)_{\mathbb Q}\).
- Every ideal \(K\) of a \(\mathbb Q\)-algebra \(C\) has exactly one PD structure, \(\gamma_n(z)=z^n/n!\). The pair \(A_{\mathbb Q}\) with \(I_{\mathbb Q}\) and these divided powers, with \(u=\mathrm{id}\), is an envelope of \((A_{\mathbb Q},I_{\mathbb Q})\). Its filtration is the \(I_{\mathbb Q}\)-adic one.
Proof. (1) Let \(y/s=z/t\), so that \(v(ty-sz)=0\) for some nonzero integer \(v\). The scalar identity gives \[ \begin{aligned}&(vt)^n\gamma_n(y) =\gamma_n(vty)\\&\quad =\gamma_n(vsz) =(vs)^n\gamma_n(z),\end{aligned} \] so \(\gamma_n(y)/s^n=\gamma_n(z)/t^n\). This defines maps on sections, compatible with restriction, hence maps of sheaves. Writing local sections with a common denominator, the identities follow from those in \(C\); for instance \(\gamma_m(\gamma_n(y)/s^n)=\gamma_m(\gamma_n(y))/s^{mn}\).
(2) Let \((C,K,\delta)\) be a PD pair and \(\psi\colon(A_{\mathbb Q},I_{\mathbb Q})\to(C,K)\) a morphism of pairs. Then \(C\) is a \(\mathbb Q\)-algebra. The PD morphism \(\varphi\colon D\to C\) induced by \(\psi|_A\) extends uniquely to a ring map \(\varphi_{\mathbb Q}\colon D_{\mathbb Q}\to C\). It is a PD morphism, since \[ \begin{aligned}&\delta_n(\varphi(y)/s) =\delta_n(\varphi(y))/s^n\\&\quad =\varphi(\gamma_n(y))/s^n.\end{aligned} \] It agrees with \(\psi\) on \(A_{\mathbb Q}\), the two agreeing on \(A\). Conversely, a PD morphism \(D_{\mathbb Q}\to C\) under \(A_{\mathbb Q}\) restricts along the PD morphism \(D\to D_{\mathbb Q}\) to the PD morphism \(\varphi\). So it equals \(\varphi_{\mathbb Q}\). The monomials of \(D_{\mathbb Q}\) are those of \(D\) divided by integers, which gives the equality of filtrations.
(3) The identity \(n!\gamma_n(z)=z^n\) gives uniqueness. The maps \(z\mapsto z^n/n!\) satisfy the identities, because these are identities of polynomials with rational coefficients: \[ \begin{aligned} \tfrac{(x+y)^n}{n!}&=\textstyle\sum_i\frac{x^i}{i!}\frac{y^{n-i}}{(n-i)!},\\ \tfrac1{m!}\bigl(\tfrac{z^n}{n!}\bigr)^m&=e_{m,n}\tfrac{z^{mn}}{(mn)!}. \end{aligned} \] A morphism of pairs from \((A_{\mathbb Q},I_{\mathbb Q})\) into a PD pair \((C,K,\delta)\) goes into a \(\mathbb Q\)-algebra, where \(\delta_n(z)=z^n/n!\). So it is a PD morphism, and the identity of \(A_{\mathbb Q}\) has the universal property. Finally, a monomial \(\prod y_\nu^{i_\nu}/i_\nu!\) of weight \(\geq n\) lies in \(I_{\mathbb Q}^n\), and a product of \(n\) sections of \(I_{\mathbb Q}\) is a monomial of weight \(n\). \(\square\)
Lemma 3.2. Let \(I\) be a weakly regular ideal of a ring \(A\) of \(T\), \(B=A/I\) and \(M=I/I^2\). Then \(I_{\mathbb Q}\) is a weakly regular ideal of \(A_{\mathbb Q}\). Moreover \(A_{\mathbb Q}/I_{\mathbb Q}=B_{\mathbb Q}\), and \(I_{\mathbb Q}/I_{\mathbb Q}^2=M_{\mathbb Q}\).
Proof. Quotient and conormal module. The functor \(X\mapsto X_{\mathbb Q}\) is exact and carries \(I^2\) to \(I_{\mathbb Q}^2\), both being images of \(I\otimes I\). This gives the two identifications.
Flatness. For a \(B_{\mathbb Q}\)-module \(N\), regarded as a \(B\)-module, \[ \begin{aligned}&M_{\mathbb Q}\otimes_{B_{\mathbb Q}}N =M\otimes_BB_{\mathbb Q}\otimes_{B_{\mathbb Q}}N\\&\quad =M\otimes_BN.\end{aligned} \] A monomorphism of \(B_{\mathbb Q}\)-modules is a monomorphism of \(B\)-modules. So \(M_{\mathbb Q}\) is flat over \(B_{\mathbb Q}\).
The Tor algebras. Let \(P\to B\) be the resolution of Section 2 of the lesson on weakly regular ideals. Then \(P_{\mathbb Q}\to B_{\mathbb Q}\) is a quasi-isomorphism of simplicial \(A_{\mathbb Q}\)-algebras, because \(-\otimes\mathbb Q\) is exact. Its terms \(P_n\otimes_AA_{\mathbb Q}\) are flat over \(A_{\mathbb Q}\), by the identity above with \(A\) in place of \(B\). Moreover \[ P_{\mathbb Q}\otimes_{A_{\mathbb Q}}B_{\mathbb Q}=(P\otimes_AB)_{\mathbb Q}. \] By Proposition 2.2 (2) of that lesson, \(\pi_\bullet\bigl((P\otimes_AB)_{\mathbb Q}\bigr)=\operatorname{Tor}^A_\bullet(B,B)_{\mathbb Q}\) is therefore \(\operatorname{Tor}^{A_{\mathbb Q}}_\bullet(B_{\mathbb Q},B_{\mathbb Q})\) as a graded ring. The homomorphism of Proposition 2.2 (3) there, for the square from \(A\to B\) to \(A_{\mathbb Q}\to B_{\mathbb Q}\), is the map of Tor sheaves induced by the square. Computed with the map \(P\to P_{\mathbb Q}\), it is the localization map. So it extends to an isomorphism of graded rings \[ \operatorname{Tor}^A_\bullet(B,B)_{\mathbb Q}\to\operatorname{Tor}^{A_{\mathbb Q}}_\bullet(B_{\mathbb Q},B_{\mathbb Q}). \]
The maps \(\theta\). By the naturality of \(\partial\) and \(\theta\) (2.3 and Definition 2.4 there), \(\theta_{A_{\mathbb Q},I_{\mathbb Q}}\circ\Lambda(M\to M_{\mathbb Q})\) is \(\theta_{A,I}\) followed by this map. Both \(\theta_{A_{\mathbb Q},I_{\mathbb Q}}\) and the localization of \(\theta_{A,I}\) are \(B_{\mathbb Q}\)-linear algebra homomorphisms out of \(\Lambda_{B_{\mathbb Q}}(M_{\mathbb Q})=\Lambda_B(M)_{\mathbb Q}\), and they agree on \(M\). So they are equal, and \(\theta_{A_{\mathbb Q},I_{\mathbb Q}}\) is an isomorphism. \(\square\)
Theorem 3.3. Let \(I\) be a weakly regular ideal of a ring \(A\) of \(T\) with an envelope \((D,J,\gamma)\), and let \(M=I/I^2\). Then \[ c\otimes\mathbb Q\colon\Gamma_B(M)_{\mathbb Q}\to(\operatorname{gr}D)_{\mathbb Q} \] is an isomorphism.
Proof. The target. By exactness and Lemma 3.1 (2), \((\operatorname{gr}D)_{\mathbb Q}=\operatorname{gr}(D_{\mathbb Q})\). By Lemma 3.1 (2), (3) and the uniqueness of envelopes, there is a PD isomorphism \(D_{\mathbb Q}\to A_{\mathbb Q}\) under \(A_{\mathbb Q}\). It carries \(\gamma_n(u(x))\) to \(x^n/n!\) and the filtration to the \(I_{\mathbb Q}\)-adic one. So \((\operatorname{gr}D)_{\mathbb Q}\) is identified with \(\operatorname{gr}_{I_{\mathbb Q}}(A_{\mathbb Q})\). By Lemma 3.2 and Theorem 3.3 (1) of the lesson on weakly regular ideals, the multiplication map \[ \mu\colon S_{B_{\mathbb Q}}(M_{\mathbb Q})\to\operatorname{gr}_{I_{\mathbb Q}}(A_{\mathbb Q}) \] is an isomorphism.
The source. Let \(\sigma\colon S_{B_{\mathbb Q}}(M_{\mathbb Q})\to\Gamma_B(M)_{\mathbb Q}\) be the algebra map that is the identity on \(M_{\mathbb Q}=\Gamma^1_B(M)_{\mathbb Q}\). The maps \(\xi\mapsto\xi^n/n!\), from \(M\) to \(S^n_{B_{\mathbb Q}}(M_{\mathbb Q})\), satisfy the relations of \(\Gamma\), as in Lemma 3.1 (3). So they define a homomorphism \(\Gamma_B(M)\to S_{B_{\mathbb Q}}(M_{\mathbb Q})\), which extends to \(\tau\colon\Gamma_B(M)_{\mathbb Q}\to S_{B_{\mathbb Q}}(M_{\mathbb Q})\). Then \(\tau\sigma\) is the identity on the generators \(M_{\mathbb Q}\). Since \(\xi^n=n!\gamma_n(\xi)\) in \(\Gamma\), \(\sigma\tau(\gamma_n(\xi))=\gamma_n(\xi)\). So \(\sigma\) is an isomorphism of graded algebras.
The comparison. Under these identifications \((c\otimes\mathbb Q)\circ\sigma\) sends \(\xi_1\cdots\xi_n\) to the class of \(x_1\cdots x_n\), where \(\xi_i=\bar x_i\). So it is \(\mu\), and \(c\otimes\mathbb Q=\mu\circ\sigma^{-1}\) is an isomorphism. \(\square\)
4. Topoi with enough points
Proposition 4.1. Let \(p\) be a point of \(T\), with stalk functor \(p^*\) and right adjoint \(p_*\). Let \((D,J,\gamma)\) be an envelope of \((A,I)\) in \(T\). Then \((D_p,J_p,\gamma_p)\) is an envelope of \((A_p,I_p)\), \((F^nD)_p=F^n(D_p)\), and \(c_p\) is the map \(c\) of \((A_p,I_p)\).
Proof. Envelopes. The functor \(p^*\) preserves finite limits and colimits, and \(p_*\) preserves limits. Rings, ideals and the PD identities are described by finite products, monomorphisms and equalities of maps. So both functors carry pairs to pairs and PD pairs to PD pairs. The adjunction, which respects finite products, gives bijections natural in \((C,K,\delta)\): \[ \begin{aligned} &\operatorname{Hom}_{\mathrm{PD}}(D_p,C)\cong\operatorname{Hom}_{\mathrm{PD}}(D,p_*C)\\ &\quad\cong\operatorname{Hom}_{\mathrm{pair}}((A,I),p_*(C,K))\\ &\quad\cong\operatorname{Hom}_{\mathrm{pair}}((A_p,I_p),(C,K)). \end{aligned} \] The composite bijection is composition with \(u_p\).
Filtration. \(F^nD\) is the image of the map \(\mathbb Z^{(D\times W_n)}\to D\), \((d,w)\mapsto dw\). Here \(W_n\) is the image of \(\coprod J^k\to D\), the coproduct over the sequences \((i_1,\dots,i_k)\) with \(\sum i_\nu\geq n\). The functor \(p^*\) commutes with images, coproducts, finite products and free abelian groups. So \((F^nD)_p=F^n(D_p)\), and \((\operatorname{gr}D)_p=\operatorname{gr}(D_p)\).
The divided power algebra. Let \(G\) be a commutative graded \(B_p\)-algebra. Then \(\bigoplus_np_*G_n\) is a commutative graded \(B\)-algebra through the unit \(B\to p_*B_p\). By adjunction, families of maps \(M_p\to G_n\) satisfying the relations correspond to families \(M\to p_*G_n\) satisfying them, that is, to graded homomorphisms \(\Gamma_B(M)\to\bigoplus_np_*G_n\). These correspond to families of maps \(\Gamma^n_B(M)\to p_*G_n\) compatible with products, and hence to graded homomorphisms \(p^*\Gamma_B(M)\to G\). So \(p^*\Gamma_B(M)\) and \(\Gamma_{B_p}(M_p)\) represent the same functor. They are isomorphic by the map carrying \(\gamma_n(\xi)\) to \(\gamma_n(\xi_p)\).
Finally, \(c_p\) and the map \(c\) of \((A_p,I_p)\) agree on the generators \(\gamma_n(\bar x)\), so they are equal. \(\square\)
Theorem 4.2. Let \(T\) have enough points, and let \(I\) be an ideal of a ring \(A\) of \(T\) with an envelope \((D,J,\gamma)\).
- For every point \(p\), the stalk of \(\theta_{A,I}\) is \(\theta_{A_p,I_p}\), and \((I/I^2)_p=I_p/I_p^2\). If \(I\) is weakly regular, then \(\theta_{A_p,I_p}\) is an isomorphism, and \(I_p\) is weakly regular whenever \((I/I^2)_p\) is flat over \(B_p\).
- \(c\) is an isomorphism if and only if the map \(c\) of every pair \((A_p,I_p)\) is.
So, if the map \(c\) is an isomorphism for every weakly regular ideal of a ring in the usual sense, then the map \(c\) of \((A,I)\) is an isomorphism for every weakly regular \(I\) whose conormal module has flat stalks.
Proof. (1) The resolution \(P\to B\) of Section 2 of the lesson on weakly regular ideals is built from free algebras on sheaves of sets and their counits. The free algebra \(A[X]=A\otimes\mathbb Z^{(\coprod_nX^n/\Sigma_n)}\) is built from finite products and colimits, which \(p^*\) preserves. So \(p^*P\to B_p\) is the resolution of \(B_p\) over \(A_p\), and \(p^*(P\otimes_AB)=p^*P\otimes_{A_p}B_p\). The functor \(p^*\) is exact, so it commutes with \(\pi_\bullet\). It also commutes with the shuffle product and with the connecting maps of \(0\to P\otimes I\to P\to P\otimes B\to0\). So the stalks of \(\operatorname{Tor}^A_\bullet(B,B)\), with its product, and of \(\partial\) are those of \((A_p,I_p)\).
The map \((\theta_{A,I})_p\) and \(\theta_{A_p,I_p}\) are algebra homomorphisms out of \(\Lambda_{B_p}(M_p)=(\Lambda_BM)_p\) that agree in degree one, so they are equal. Exactness gives \((I/I^2)_p=I_p/I_p^2\). The last assertion is Definition 2.4 there.
(2) A map of sheaves on a topos with enough points is an isomorphism exactly when all its stalks are. Apply Proposition 4.1. The final sentence follows from (1) and (2). \(\square\)
The rational statement needs no points: Theorem 3.3 holds in every topos.
5. Relative envelopes
5.0. Setting. Let \((\Lambda,H,\delta)\) be a PD pair of rings, and \(\lambda\colon\Lambda\to A\) a ring map with \(\lambda(H)\subseteq I\). A relative envelope is a PD pair \((D_r,J_r,\gamma)\) with a morphism of pairs \(v\colon(A,I)\to(D_r,J_r)\) such that \(v\lambda\colon(\Lambda,H,\delta)\to(D_r,J_r,\gamma)\) is a PD morphism. It is required to be universal: for every PD pair \((C,K,\epsilon)\) and every morphism of pairs \(w\colon(A,I)\to(C,K)\) with \(w\lambda\) a PD morphism, there is a unique PD morphism \(D_r\to C\) under \(A\). Relative envelopes exist [Tag 07H7].
By the absolute universal property there is a unique PD morphism \(q\colon D\to D_r\) with \(qu=v\). Put \[ c_r=\operatorname{gr}(q)\circ c\colon\Gamma_B(M)\to\operatorname{gr}D_r. \]
Proposition 5.1.
- \(q\) is surjective, \(q(F^nD)=F^nD_r\) for every \(n\), and \(c_r\) is surjective.
- If \(I\) is weakly regular, \(c_r\otimes\mathbb Q\) is an isomorphism.
Proof. (1) Surjectivity of \(q\). The image \(q(D)\) is a subring of \(D_r\), and \(q(J)\) is an ideal of it. It is stable under the \(\gamma_n\), since \(\gamma_n(q(y))=q(\gamma_n(y))\). Moreover \(v(I)=q(u(I))\subseteq q(J)\). The map \(v\lambda\) takes values in \(v(A)\subseteq q(D)\) and carries \(H\) into \(q(J)\). It is a PD morphism into \((q(D),q(J))\), since that pair carries the restricted divided powers. The relative universal property gives a PD morphism \(r\colon D_r\to q(D)\) under \(A\). The inclusion composed with \(r\) is a PD endomorphism of \(D_r\) under \(A\), hence the identity. So \(q(D)=D_r\), and \(J_r=r(J_r)\subseteq q(J)\subseteq J_r\).
The filtrations. A monomial \(\prod\gamma_{i_\nu}(y_\nu)\) of \(D_r\) has \(y_\nu=q(z_\nu)\) with \(z_\nu\in J\), so it is \(q\) of a monomial of the same weight. Since \(q\) is surjective, \(q(F^nD)\) is an ideal of \(D_r\), and \(q(F^nD)=F^nD_r\). So \(\operatorname{gr}(q)\) is surjective, and \(c_r\) is surjective by Proposition 2.2.
(2) A map back. In \(\Lambda\), \(n!\,\delta_n(h)=h^n\). So \(n!\,\lambda(\delta_n(h))=\lambda(h)^n\) in \(A\), and the composite \(\Lambda\to A\to A_{\mathbb Q}\) is a PD morphism into \(A_{\mathbb Q}\) with the divided powers \(z^n/n!\) on \(I_{\mathbb Q}\). By Lemma 3.1 (2), (3), \(A_{\mathbb Q}\cong D_{\mathbb Q}\) under \(A\). So the relative universal property gives a PD morphism \(r\colon D_r\to D_{\mathbb Q}\) under \(A\).
Bijectivity. By the absolute uniqueness, \(rq\colon D\to D_{\mathbb Q}\) is the localization map. Localizing gives \(r_{\mathbb Q}q_{\mathbb Q}=\mathrm{id}\), so \(q_{\mathbb Q}\) is injective. By (1) and exactness, \(q_{\mathbb Q}\) is surjective and maps \((F^nD)_{\mathbb Q}\) onto \((F^nD_r)_{\mathbb Q}\). So \(\operatorname{gr}(q)\otimes\mathbb Q\) is an isomorphism, and Theorem 3.3 gives the assertion. \(\square\)
Proposition 5.2. Let \(\Lambda=A=\mathbb Z_{(2)}\), \(\lambda=\mathrm{id}\), \(H=I=2A\) and \(\delta_n(z)=z^n/n!\). Then \(I\) is weakly regular, \(M\cong\mathbb F_2\), and \(c_r\) is zero in every positive weight. In particular \(c_r\) is not injective.
Proof. The PD structure. By Legendre's formula \(v_2(n!)=n-s_2(n)\leq n-1\), where \(s_2(n)\) is the sum of the binary digits of \(n\). So \(v_2(2^na^n/n!)\geq1\) for \(n\geq1\), and \(\delta_n(I)\subseteq I\). The identities hold because they hold in \(\mathbb Q\).
Weak regularity. The element \(2\) is a nonzerodivisor, so its Koszul complex \(A\xrightarrow{2}A\) has no homology in positive degrees. By Proposition 4.2 of the lesson on weakly regular ideals, \(I\) is weakly regular and \(M\) is free on \(\bar2\) over \(B=\mathbb F_2\).
The relative envelope. The pair \((A,I,\delta)\) with \(v=\mathrm{id}\) is a relative envelope. Indeed, for a PD pair \((C,K,\epsilon)\) and a morphism of pairs \(w\colon(A,I)\to(C,K)\) with \(w\lambda=w\) a PD morphism, \(w\) itself is the unique PD morphism under \(A\).
The filtration. For \(m=2^k\), \(v_2(m!)=m-1\), so \(\delta_m(2)\) is \(2\) times a unit. For \(n\geq1\), choose \(2^k\geq n\): then \(I\subseteq I^{[n]}\subseteq I\). So \(\operatorname{gr}^nD_r=0\) for \(n\geq1\), while \(\Gamma^1_B(M)=M\neq0\). \(\square\)
Proposition 5.3. Let \(p\) be a prime, \(\Lambda=\mathbb F_p[\epsilon,\eta]/(\epsilon,\eta)^2\) and \(H=(\epsilon,\eta)\). For \(u,v\in\mathbb F_p\) put \(\delta_0=1\), \(\delta_1(z)=z\), \(\delta_p(u\epsilon+v\eta)=u\eta\), and \(\delta_n=0\) for \(n\geq2\), \(n\neq p\).
- \(\delta\) is a PD structure with \(H^{[p+1]}=0\).
- Let \(R=\mathbb F_p[a,b]/(a,b)^2\), \(A=R[x]\), \(I=(x)\), and \(\lambda(\epsilon)=ax\), \(\lambda(\eta)=bx\). Then \(I\) is weakly regular, with \(B=R\) and \(M=R\bar x\) free. The element \(b\bar x\neq0\) lies in the kernel of \(c_r\), so \(c_r\) is not injective.
Proof. (1) The ideal \(H\) consists of the elements \(z=u\epsilon+v\eta\), and \(H\cdot H=0\). Since \(u^p=u\), \(\delta_p(z)=u^p\eta\). We check the identities.
- Addition. The mixed terms are products of two elements of \(H\), hence zero. The map \(\delta_p\) is additive because \((u+u')^p=u^p+u'^p\).
- Scalars. Write \(w=c+w'\) with \(c\in\mathbb F_p\) and \(w'\in H\). Then \(wz=cz\), and \(w^p=c^p\) because \(w'^2=0\). So \(\delta_p(wz)=c^pu^p\eta=w^p\delta_p(z)\). For the other indices both sides are \(0\), or the identity is clear.
- Products, \(m,n\geq1\). The left side lies in \(H\cdot H=0\). The right side vanishes unless \(m+n=p\), and then \(p\) divides \(\binom pm\).
- Iteration. For \(m=1\) or \(n=1\) the identity is clear. For \(m,n\geq2\), \(mn\neq p\), so the right side is \(0\). On the left, \(\delta_n(z)\in\mathbb F_p\eta\), and \(\delta_m\) of an element of \(\mathbb F_p\eta\) is zero, its \(\epsilon\)-coefficient being \(0\).
A monomial with two factors vanishes, and \(\delta_n=0\) for \(n>p\). So \(H^{[p+1]}=0\).
(2) The base map. The map \(\lambda\) is a ring map, because the products of \(ax\) and \(bx\) vanish, and \(\lambda(H)\subseteq I\).
Weak regularity. Comparing coefficients shows that \(x\) is a nonzerodivisor of \(R[x]\). By Proposition 4.2 of the lesson on weakly regular ideals, \(I\) is weakly regular and \(M\) is free on \(\bar x\). Since \(b\neq0\) in \(R\), \(b\bar x\neq0\).
The kernel. Put \(t=v(x)\). In \(D_r\), using \(a^p=0\), \[ \gamma_p(v\lambda(\epsilon))=\gamma_p(v(a)t)=v(a)^p\gamma_p(t)=0. \] The compatibility gives \(\gamma_p(v\lambda(\epsilon))=v\lambda(\delta_p(\epsilon))=v\lambda(\eta)=v(b)t\). So \(v(b)t=0\), and \[ c_r(b\bar x)=v(b)t\bmod F^2D_r=0.\qquad\square \]
Corollary 5.4. For no nonzero integer \(N\) is \(c_r\otimes\mathbb Z[1/N]\) an isomorphism for all relative envelopes of weakly regular ideals. This holds even when the divided power filtration of the base is required to vanish from some index on.
Proof. Choose a prime \(p\) not dividing \(N\). In Proposition 5.3 every ring is an \(\mathbb F_p\)-algebra, in which \(N\) is a unit. So \(c_r\otimes\mathbb Z[1/N]=c_r\), which is not injective. \(\square\)
6. Flatness over the ambient ring
Proposition 6.1. Let \(A=\mathbb F_p[x]\) and \(I=(x)\). Then \(I\) is weakly regular with free conormal module, and every envelope \(D\) of \((A,I)\) fails to be flat over \(A\).
Proof. Weak regularity. The element \(x\) is a nonzerodivisor. Proposition 4.2 of the lesson on weakly regular ideals applies.
A test PD ring. Let \(E=\mathbb F_p[y]/(y^p)\), with \(\epsilon_n(z)=z^n/n!\) for \(n<p\) and \(\epsilon_n=0\) for \(n\geq p\) on \(K=(y)\). Then \(K^p=0\), and \(\epsilon_n(K)\subseteq K\) for \(n\geq1\). We check the identities.
- All indices \(<p\). When every index involved, including \(m+n\) or \(mn\), is \(<p\), the identities are identities of rational polynomials with denominators invertible in \(\mathbb F_p\).
- Addition, \(n\geq p\). Both sides vanish. Each term on the right is either a product of total degree \(n\geq p\) in \(K\), or contains a factor \(\epsilon_k\) with \(k\geq p\).
- Products, \(m+n\geq p\). The right side is \(0\). On the left, if \(m,n<p\), the product is \(z^{m+n}\) times a unit, which is \(0\); otherwise a factor vanishes.
- Iteration, \(mn\geq p\). The right side is \(0\). If \(n\geq p\), then \(\epsilon_n(z)=0\). If \(m\geq p\), then \(\epsilon_m=0\). If \(m,n<p\), the left side is \(z^{mn}\) times a unit, which is \(0\).
Non-flatness. The morphism of pairs \(x\mapsto y\) induces a PD morphism \(D\to E\). It sends \(\gamma_{p-1}(u(x))\) to \(y^{p-1}/(p-1)!\neq0\), so \(\gamma_{p-1}(u(x))\neq0\). Since \(D\) is an \(\mathbb F_p\)-algebra, \[ u(x)\gamma_{p-1}(u(x))=p\,\gamma_p(u(x))=0. \] Multiplication by \(x\) is injective on \(A\), but after \(-\otimes_AD\) it becomes multiplication by \(u(x)\) on \(D\), which is not injective. \(\square\)
7. Exercises
Exercise 7.1. Let \(p\) be an odd prime, \(\Lambda=A=\mathbb Z_{(p)}\), \(\lambda=\mathrm{id}\), \(H=I=pA\) and \(\delta_n(z)=z^n/n!\). Show that \(c_r\) is an isomorphism in weights \(<p-1\) and zero in weight \(p-1\).
Exercise 7.2. Let \(A\) be a \(\mathbb Q\)-algebra of \(T\) and \(I\) a weakly regular ideal. Show that \(c\) is an isomorphism and that \(\operatorname{gr}D\cong\operatorname{gr}_I(A)\).
8. Complete solutions
7.1. The PD pair. By Legendre's formula \(v_p(k!)=\) \((k-s_p(k))/(p-1)\leq\) \((k-1)/(p-1)\), where \(s_p(k)\geq1\) is the sum of the base-\(p\) digits of \(k\geq1\). Put \(w(k)=k-v_p(k!)\). Then \(w(k)\geq1\), so \((A,I,\delta)\) is a PD pair. As in Proposition 5.2, it is its own relative envelope.
The filtration. A monomial \(\prod\delta_{i_\nu}(y_\nu)\) with \(y_\nu\in I\) has valuation \(\geq\sum w(i_\nu)\). This is \(\geq w(\sum i_\nu)\), because a multinomial coefficient is an integer, so \(v_p((\sum i_\nu)!)\geq\sum v_p(i_\nu!)\). Moreover \(\delta_k(p)\) is a unit times \(p^{w(k)}\). Since \(A\) is a discrete valuation ring, \[ I^{[n]}=p^{e(n)}A,\qquad e(n)=\min_{k\geq n}w(k). \]
The values of \(e\). For \(k<p\), \(w(k)=k\). For \(k\geq p\), \[ w(k)\geq k-\tfrac{k-1}{p-1}=\tfrac{k(p-2)+1}{p-1}, \] which increases with \(k\) and equals \(p-1\) at \(k=p\). Moreover \(w(p)=p-1\). Hence \(e(n)=n\) for \(n\leq p-1\), and \(e(p)=p-1\).
The comparison. The module \(\Gamma^n_{\mathbb F_p}(M)\) is spanned by \(\gamma_n(\bar p)\): a product \(\prod\gamma_{n_i}(a_i\bar p)\) equals \(\prod a_i^{n_i}\) times an integer multiple of \(\gamma_n(\bar p)\). For \(n<p\), \(\gamma_n(\bar p)\neq0\): the PD ring \(E\) of Proposition 6.1 gives a map \(\Gamma_{\mathbb F_p}(M)\to E\), \(\bar p\mapsto y\), sending \(\gamma_n(\bar p)\) to \(y^n/n!\neq0\).
For \(n<p-1\), \(\operatorname{gr}^n=p^nA/p^{n+1}A\cong\mathbb F_p\). The surjection \(c_r\) of Proposition 5.1 (1) between one-dimensional spaces is an isomorphism. In weight \(p-1\), \(\operatorname{gr}^{p-1}=p^{p-1}A/p^{p-1}A=0\), while \(\gamma_{p-1}(\bar p)\neq0\).
7.2. Here \(X_{\mathbb Q}=X\) for every sheaf of \(A\)-modules. So \(c\otimes\mathbb Q=c\), which is an isomorphism by Theorem 3.3. By Lemma 3.1 (3), \(A\) with \(z^n/n!\) is an envelope of \((A,I)\) with the \(I\)-adic filtration. So \(\operatorname{gr}D\cong\operatorname{gr}_I(A)\), as in the proof of Theorem 3.3.
Sources
The question whether \(\Gamma_B(I/I^2)\to\operatorname{gr}D\) is an isomorphism for weakly regular ideals, and its affirmative answer for Koszul-regular ideals following Berthelot's treatment of regular sequences, are due to Illusie, in his work on the cotangent complex. Divided power envelopes and their filtration are Berthelot's, in Cohomologie cristalline des schémas de caractéristique \(p>0\); their existence is taken from the Stacks project, cited above. Weakly regular ideals are Quillen's quasi-regular ideals, in his MIT notes on the homology of commutative rings (1968), available from Haynes Miller's page. Z. Mao, Revisiting derived crystalline cohomology, arXiv:2107.02921, studies divided power envelopes of quasiregular pairs in the animated setting. The main argument follows a solution produced with ChatGPT (OpenAI) in October 2026 and completed here.