Arrow obstructions of diagrams and the lifted multiplication

Written by Claude Opus 5.5 (Anthropic) on 23 October 2026. Self-checked by the writing AI. Public domain (CC0).

Suppose that every scheme of a diagram has been deformed. The diagram itself deforms when the maps between the deformed schemes can be chosen, and when the chosen maps satisfy the relations of the diagram. For a diagram indexed by a free category there are no relations, and the obstruction to deforming the diagram is then built from the obstructions to lifting the individual maps. This lesson proves this in the language of the cotangent complex: the obstruction to deforming a diagram indexed by a free category is minus the image, under a connecting map, of the family of obstructions to lifting its generating arrows. The connecting map comes from a two-term description of the Ext groups of modules over such a diagram, and the comparison of the two obstructions is a diagram chase. In the same way the obstruction to lifting a morphism of such diagrams is minus the image of the defects of compatibility of lifts chosen at each level, a statement that rests on the extension class of a trivial extension twisted along the arrows.

Applied to the strict truncations of the nerves of the lesson Truncated nerves and the level filtration, this identifies three graded pieces of deformation obstructions. For a flat group scheme \(G\), the component of the obstruction to deforming \(G\) in the second step of its filtration is the class of minus the obstruction to lifting the multiplication to a deformation of \(G\) as a scheme, in a quotient that removes exactly the dependence on that deformation. For a homomorphism \(u\), the component of the obstruction to extending \(u\) in the second step is the class of minus the defect of multiplicativity of a lift of \(u\) as a morphism of schemes. For an action, the component in the first step is the class of minus the obstruction to extending the action to a deformation of the scheme acted on.

Prerequisites. Truncated nerves and the level filtration (Conventions, 1.1, Theorem 3.1, Corollaries 3.2 and 3.3, Sections 4 and 5); Morphisms of fibred topoi and modules on total topoi (Proposition 4.2); Deformations of diagrams of schemes (Proposition 1.4, Construction 4.2, Theorem 4.3); Equivariant extensions and deformations (Theorem 3.3, Lemma 5.1); Deformations of flat group schemes and of homomorphisms (Conventions, Lemma 1.1, Theorem 3.1); The adjoint co-Lie complex of a group scheme (Lemma 3.1); Deforming a morphism together with its target (Conventions). From the programme lesson The cotangent complex: Theorems 10.3 and 10.4, the class (10.6) and its description by relation maps, the graph boundary (10.7) and its identification with the connecting map, and the exact sequence (10.9). From the AI-integrated Stacks project: Hom complexes into bounded below complexes of injectives [Stacks, Tag 05TG]; cones and their distinguished triangles [Tag 014I]; morphisms of triangles [Tags 014A, 0145]; right adjoints of exact functors preserve injectives [Tag 015Z]; the horseshoe lemma [Tag 013T].

Conventions

For complexes \(A\), \(B\) of modules, \(\operatorname{Hom}^n(A,B)=\prod_p\operatorname{Hom}(A^p,B^{p+n})\) with the differential \(d(f)=d_B\circ f-(-1)^nf\circ d_A\); for a bounded above \(A\) and a bounded below complex \(J\) of injectives, \(H^n\operatorname{Hom}^\bullet(A,J)=\operatorname{Hom}_D(A,J[n])\) [Tag 05TG]. The cone of a map of complexes \(g\colon A\to B\) is \(C(g)^n=B^n\oplus A^{n+1}\) with \(d(b,a)=(db+g(a),-da)\); the triangle \(A\xrightarrow{g}B\to C(g)\xrightarrow{p}A[1]\), with the inclusion and the projection, is distinguished [Tag 014I]. The complex \(\operatorname{Hom}^\bullet(A[1],B)\) is \(\operatorname{Hom}^\bullet(A,B)[-1]\), with the same terms and differentials. For an exact sequence \(0\to P\to Q\to R\to0\) of complexes, the connecting map sends the class of a cocycle \(r\in R^n\) to the class of \(dq\in P^{n+1}\), for any \(q\in Q^n\) mapping to \(r\).

Extensions of rings of a topos and their classes \(\kappa\in\operatorname{Ext}^1(L_{\mathcal B/\mathcal Z},M)\) are those of Theorem 10.3 of The cotangent complex, defined by (10.6) there; \(\kappa\) is additive for the Baer sum and natural in the coefficients and under inverse images of morphisms of topoi. For a morphism of ringed topoi \(\varphi\), the cotangent complexes and their inverse images are as in the Conventions of the lesson on deforming a morphism together with its target.

1. Modules over a diagram indexed by a free category

1.1. Setting. In Sections 1–5, \(C\) is a finite category freely generated by a set \(A\) of arrows: every arrow of \(C\) is, in a unique way, a composite \(a_k\circ\cdots\circ a_1\) of generators, the identities being the composites with \(k=0\); so \(C\) has no endomorphisms other than identities. For a generator \(a\) write \(a\colon s(a)\to t(a)\). The categories \(\Delta'_{\leq1}\) (objects \([0]\), \([1]\), generators \(\delta^0,\delta^1\)) and \(\Delta'_{[1,2]}\), the full subcategory of \(\Delta'\) on \([1]\) and \([2]\) (generators \(\delta^0,\delta^1,\delta^2\colon[1]\to[2]\)), are of this kind. \(X\) is a topos fibred over \(C^{\mathrm{op}}\), with the morphisms \(X(\alpha)\colon X_{c'}\to X_c\) for \(\alpha\colon c\to c'\), and \(\mathcal O\) a ring of \(\operatorname{Top}(X)\) for which every \(X(\alpha)\) is a flat morphism of ringed topoi. An \(\mathcal O\)-module \(K\) has levels \(K_c\) and structure maps \(K_\alpha\colon X(\alpha)^*K_c\to K_{c'}\), determined by those of the generators (Proposition 4.2 of the lesson on morphisms of fibred topoi). As in 1.1 of the lesson on truncated nerves, the restriction to the level \(c\) has an exact left adjoint \(c_!\), \((c_!F)_w=\bigoplus_{q\colon c\to w}X(q)^*F\), and carries injective modules to injective modules.

Proposition 1.2 (the resolution of a module). For every \(\mathcal O\)-module \(K\) the sequence \[ \begin{aligned}0&\to\bigoplus_{a\in A}t(a)_!\bigl(X(a)^*K_{s(a)}\bigr)\xrightarrow{\ \iota\ }\bigoplus_{c}c_!K_c\\&\quad\xrightarrow{\ \varepsilon\ }K\to0\end{aligned} \] is exact. Here \(\varepsilon\) is the sum of the counits, and on the summand of \(a\), \(\iota\) is adjoint to the map \(X(a)^*K_{s(a)}\to\bigl(\bigoplus_cc_!K_c\bigr)_{t(a)}\) sending \(x\) to \(x\), placed in the summand of the path \(a\) of \(\bigl(s(a)_!K_{s(a)}\bigr)_{t(a)}\), minus \(K_a(x)\), placed in the summand of the identity of \(\bigl(t(a)_!K_{t(a)}\bigr)_{t(a)}\). The sequence is functorial in \(K\).

Proof. Fix a level \(w\). The level \(w\) of the middle term is \(\bigoplus X(q)^*K_c\) over the arrows \(q\colon c\to w\), that of the first term is \(\bigoplus X(pa)^*K_{s(a)}\) over the pairs of a generator \(a\) and an arrow \(p\colon t(a)\to w\), with \(X(p)^*X(a)^*\cong X(pa)^*\). On these summands \(\varepsilon\) is \(K_q\), and \(\iota\) sends \(x\) in the summand of \((a,p)\) to \(x\) in the summand of \(pa\) minus \(X(p)^*(K_a)(x)\) in the summand of \(p\); so \(\varepsilon\iota=0\), because \(K_{pa}=K_p\circ X(p)^*(K_a)\). Every arrow \(q\neq\mathrm{id}\) is uniquely \(q=q'a\) with \(a\) a generator, its first factor. Define \(\sigma\colon K_w\to\bigoplus X(q)^*K_c\) as the inclusion of the summand of \(\mathrm{id}_w\), and \(h\) from the middle term to the first by \(h=0\) on the summand of \(\mathrm{id}_w\) and, by induction on the length of \(q=q'a\), \[ \begin{aligned}h(x\text{ at }q)&=x\text{ at }(a,q')\\&\quad+h\bigl(X(q')^*(K_a)(x)\text{ at }q'\bigr).\end{aligned} \] Then \(\varepsilon\sigma=\mathrm{id}\). By induction on the length, \(\iota h(x\text{ at }q)=x\text{ at }q-\sigma\varepsilon(x\text{ at }q)\): for \(q=q'a\), \(\iota(x\text{ at }(a,q'))=x\text{ at }q-K_a(x)\text{ at }q'\), and the induction hypothesis for \(q'\) gives the term \(K_a(x)\text{ at }q'-\sigma(K_{q'}K_a(x))\). Finally \(h\iota=\mathrm{id}\), because the first factor of \(pa\) is \(a\). So at every level the complex is contractible as a complex of abelian sheaves, hence exact. The maps are natural in \(K\). \(\square\)

For complexes \(K\) and \(J\) of \(\mathcal O\)-modules put \[ \begin{gathered}\operatorname{Lev}(K,J)=\prod_c\operatorname{Hom}^\bullet(K_c,J_c),\\ \operatorname{Arr}(K,J)\\ =\prod_{a\in A}\operatorname{Hom}^\bullet\bigl(X(a)^*K_{s(a)},J_{t(a)}\bigr),\end{gathered} \] and let \(\Phi\colon\operatorname{Lev}(K,J)\to\operatorname{Arr}(K,J)\) be the map of complexes \[ \Phi(f)_a=J_a\circ X(a)^*(f_{s(a)})-f_{t(a)}\circ K_a. \]

Corollary 1.3 (the cocone description). Let \(K\) be a bounded above complex of \(\mathcal O\)-modules and \(J\) a bounded below complex of injective \(\mathcal O\)-modules. The sequence \[ \begin{aligned}0&\to\operatorname{Hom}^\bullet(K,J)\to\operatorname{Lev}(K,J)\\&\quad\xrightarrow{\ \Phi\ }\operatorname{Arr}(K,J)\to0,\end{aligned} \] whose first map is the restriction to the levels, is exact, and it is natural in \(K\) and \(J\). Its connecting maps \[ \partial_C\colon H^n\operatorname{Arr}(K,J)\to H^{n+1}\operatorname{Hom}^\bullet(K,J) \] give a long exact sequence \[ \begin{aligned}&\cdots\to\operatorname{Ext}^n(K,J)\to\textstyle\bigoplus_c\operatorname{Ext}^n(K_c,J_c)\\&\quad\to\textstyle\bigoplus_a\operatorname{Ext}^n(X(a)^*K_{s(a)},J_{t(a)})\\&\quad\xrightarrow{\ \partial_C\ }\operatorname{Ext}^{n+1}(K,J)\to\cdots.\end{aligned} \] If the terms of \(K\) are flat, \(X(a)^*K_{s(a)}\) represents \(LX(a)^*K_{s(a)}\).

Proof. Apply \(\operatorname{Hom}(-,J^q)\) to the sequence of Proposition 1.2 for every term of \(K\); since \(J^q\) is injective, the result is exact. By adjunction \(\operatorname{Hom}(c_!F,J^q)=\operatorname{Hom}(F,J^q_c)\), and the map induced by \(\iota\) on the summand of \(a\) is \(f\mapsto J^q_a\circ X(a)^*(f_{s(a)})-f_{t(a)}\circ K_a\): the adjoint of \(f_{s(a)}\) is \(J^q_a\circ X(a)^*(f_{s(a)})\) on the summand of the path \(a\). Taking Hom complexes gives the sequence. The levels \(J_c\) are bounded below complexes of injective \(\mathcal O_c\)-modules, so the cohomology of \(\operatorname{Lev}\) and \(\operatorname{Arr}\) consists of the Ext groups shown [Tag 05TG]. \(\square\)

So the Ext groups over \(C\) are the cohomology of the cocone of \(\Phi\): a class is a family of classes at the levels together with homotopies between the two composites along every generator.

2. Two diagram chases

In this section, the rows \(0\to P_1\to P_2\to P_3\to0\), \(0\to Q_1\to Q_2\to Q_3\to0\), \(0\to R_1\to R_2\to R_3\to0\) and the columns \(0\to P_j\to Q_j\to R_j\to0\) of a commutative diagram of complexes of abelian groups are exact. Write \(\rho_1\colon P_1\to P_2\), \(\rho_2\colon P_2\to P_3\) and likewise for \(Q\) and \(R\) for the maps along the rows, \(\gamma\) for the maps along the columns, \(\partial^{\mathrm r}\) for the connecting maps of the rows and \(\partial^{\mathrm c}\) for those of the columns.

Lemma 2.1. Let \(x\in H^n(R_2)\), with image \(y\in H^n(R_3)\), and assume \(H^n(P_3)=0\). Then \(\partial^{\mathrm c}(y)\in H^{n+1}(P_3)\) is the image of \(\partial^{\mathrm c}(x)\in H^{n+1}(P_2)\) under \(\rho_2\); suppose it vanishes. Then \(y\) has a unique preimage \(\tilde y\in H^n(Q_3)\), \(\partial^{\mathrm c}(x)\) has a unique preimage \(c\in H^{n+1}(P_1)\), and \[ \partial^{\mathrm r}(\tilde y)=\gamma(c)\quad\text{in }H^{n+1}(Q_1). \]

Proof. The first statement is the naturality of connecting maps. The preimages exist by exactness, and are unique: the ambiguity of \(\tilde y\) is the image of \(H^n(P_3)\), that of \(c\) the image of \(\partial^{\mathrm r}\colon H^n(P_3)\to H^{n+1}(P_1)\). Let \(\hat x\in R_2^n\) be a cocycle representing \(x\) and \(\bar x\in Q_2^n\) a preimage. Then \(d\bar x=\gamma(\hat c)\) for a cocycle \(\hat c\in P_2^{n+1}\) representing \(\partial^{\mathrm c}(x)\). As the class of \(\rho_2(\hat c)\) vanishes, \(\rho_2(\hat c)=d\hat e\) for some \(\hat e\in P_3^n\); write \(\hat e=\rho_2(\hat f)\) with \(\hat f\in P_2^n\). Then \(\hat y=\rho_2(\bar x)-\gamma(\hat e)\in Q_3^n\) is a cocycle, as \(d\rho_2(\bar x)=\gamma(\rho_2(\hat c))=\gamma(d\hat e)\), and it maps to \(\rho_2(\hat x)\), so it represents \(\tilde y\). The element \(\hat c'=\hat c-d\hat f\) satisfies \(\rho_2(\hat c')=0\), so it is \(\rho_1\) of a cocycle of \(P_1^{n+1}\), also written \(\hat c'\), whose image in \(H^{n+1}(P_2)\) is that of \(\hat c\); so it represents \(c\). Finally \(\bar x-\gamma(\hat f)\in Q_2^n\) maps to \(\hat y\), and its differential is \(\gamma(\hat c)-\gamma(d\hat f)=\gamma(\rho_1(\hat c'))=\rho_1(\gamma(\hat c'))\); so \(\partial^{\mathrm r}(\tilde y)\) is the class of \(\gamma(\hat c')\), which is \(\gamma(c)\). \(\square\)

Lemma 2.2. Let \(z\in H^n(R_1)\), and suppose that \(\rho_1(z)\in H^n(R_2)\) is the image of some \(\tilde x\in H^n(Q_2)\), and that \(H^{n-1}(R_3)=0\). Then \(\rho_2(\tilde x)\in H^n(Q_3)\) maps to zero in \(H^n(R_3)\), so \(\rho_2(\tilde x)=\gamma(w)\) for a unique \(w\in H^n(P_3)\), and \[ \partial^{\mathrm c}(z)=-\partial^{\mathrm r}(w)\quad\text{in }H^{n+1}(P_1). \] The class \(\partial^{\mathrm r}(w)\) does not depend on the choice of \(\tilde x\).

Proof. The image of \(z\) in \(H^n(R_3)\) is \(\rho_2\rho_1(z)=0\); so \(w\) exists, and it is unique because the kernel of \(\gamma\colon H^n(P_3)\to H^n(Q_3)\) is the image of \(H^{n-1}(R_3)=0\). Let \(\hat z\in R_1^n\) be a cocycle representing \(z\) and \(q\in Q_1^n\) a preimage, so that \(\gamma(q)=\hat z\) and \(\partial^{\mathrm c}(z)\) is the class of the cocycle \(c_1\in P_1^{n+1}\) with \(\gamma(c_1)=dq\). Let \(\hat x\in Q_2^n\) be a cocycle representing \(\tilde x\); changing it by a coboundary, we may assume \(\gamma(\hat x)=\rho_1(\hat z)\). Then \(\rho_1(q)-\hat x=\gamma(p)\) for some \(p\in P_2^n\). Applying \(\rho_2\), and using \(\rho_2\rho_1=0\), gives \(\rho_2(\hat x)=-\gamma(\rho_2(p))\), so \(w\) is the class of the cocycle \(-\rho_2(p)\). To compute \(\partial^{\mathrm r}(w)\), lift \(-\rho_2(p)\) to \(-p\in P_2^n\): then \(\partial^{\mathrm r}(w)\) is the class of the element of \(P_1^{n+1}\) whose image under \(\rho_1\) is \(-dp\). Now \(\gamma(dp)=\) \(\rho_1(dq)-d\hat x=\) \(\rho_1(\gamma(c_1))=\) \(\gamma(\rho_1(c_1))\), so \(dp=\rho_1(c_1)\), \(\gamma\) being injective. Hence \(\partial^{\mathrm r}(w)=-[c_1]=-\partial^{\mathrm c}(z)\). Another choice of \(\tilde x\) changes it by an element of the image of \(H^n(P_2)\), which changes \(w\) by an element of the image of \(\rho_2\), killed by \(\partial^{\mathrm r}\). \(\square\)

3. Trivial extensions twisted along the arrows

3.0. Setting. In 1.1, let \(\mathcal Z\to\mathcal O\) be a map of rings of \(\operatorname{Top}(X)\), \(M\) an \(\mathcal O\)-module and \(L=L_{\mathcal O/\mathcal Z}\), the cotangent complex of the ring map, a bounded above complex of flat \(\mathcal O\)-modules with the levels \(L_{\mathcal O_c/\mathcal Z_c}\). An extension of \(\mathcal O\) by \(M\) over \(\mathcal Z\) is an extension of \(\mathcal Z\)-algebras of the ring \(\mathcal O\) of the topos \(\operatorname{Top}(X)\) by \(M\), as in Theorem 10.3 of The cotangent complex. Its levels \(\mathcal E_c\) are extensions of \(\mathcal O_c\) by \(M_c\) over \(\mathcal Z_c\), and its structure maps \(\mathcal E_a\colon X(a)^{-1}\mathcal E_{s(a)}\to\mathcal E_{t(a)}\) are maps of \(X(a)^{-1}\mathcal Z_{s(a)}\)-algebras lifting the structure map \(\mathcal O_a\) of \(\mathcal O\) and inducing on the kernels the map \(X(a)^{-1}M_{s(a)}\to M_{t(a)}\) of \(M\). Conversely, such levels and such ring maps along the generators define an extension, \(C\) being free. For a generator \(a\), \(\operatorname{Ext}^0\bigl(X(a)^*L_{s(a)},M_{t(a)}\bigr)\) is the group of derivations \(X(a)^{-1}\mathcal O_{s(a)}\to M_{t(a)}\) over \(X(a)^{-1}\mathcal Z_{s(a)}\), since \(H^0(L_{s(a)})=\Omega_{\mathcal O_{s(a)}/\mathcal Z_{s(a)}}\).

Definition 3.1. For a family \(D=(D_a)_{a\in A}\) of such derivations, the twisted trivial extension \(T_D\) has the levels \(\mathcal O_c\oplus M_c\), the trivial extensions, and along the generator \(a\) the ring maps \[ \begin{gathered}X(a)^{-1}(\mathcal O_s\oplus M_s)\to\mathcal O_t\oplus M_t,\\ (b,m)\longmapsto\bigl(\mathcal O_a(b),M_a(m)+D_a(b)\bigr),\end{gathered} \] with \(s=s(a)\), \(t=t(a)\); they are ring maps because \(D_a\) is a derivation and \(M_t^2=0\), and maps over \(\mathcal Z\) because \(D_a\) vanishes on \(X(a)^{-1}\mathcal Z_s\).

Proposition 3.2. In \(\operatorname{Ext}^1(L,M)\), \[ \kappa(T_D)=-\partial_C(D), \] where \(D\) is read as a class in \(H^0\operatorname{Arr}(L,J)=\) \(\prod_a\operatorname{Ext}^0\bigl(X(a)^*L_{s(a)},M_{t(a)}\bigr)\) for an injective resolution \(J\) of \(M\), and \(\partial_C\) is the connecting map of Corollary 1.3.

Proof. A graph boundary. Let \(\Omega=\Omega_{\mathcal O/\mathcal Z}\), with the levels \(\Omega_{\mathcal O_c/\mathcal Z_c}\) and the universal derivation \(d\colon\mathcal O\to\Omega\). For a generator \(a\) let \(h_a\colon X(a)^*\Omega_s\to M_t\) be the \(\mathcal O_t\)-linear map with \(h_a(1\otimes db)=D_a(b)\). Let \(I'\) be the module with the levels \(M_c\oplus\Omega_c\) and the structure maps \[ (m,\omega)\longmapsto\bigl(M_a(m)+h_a(\omega),\Omega_a(\omega)\bigr) \] along the generators. The inclusion of \(M\) and the projection to \(\Omega\) are maps of modules, and \(0\to M\to I'\to\Omega\to0\) is exact and split at every level. The graph boundary of \(d\) for this sequence, the pullback of the trivial extension \(\mathcal O\oplus I'\to\mathcal O\oplus\Omega\) along \(b\mapsto(b,db)\) ((10.7) of The cotangent complex), has the elements \((b,m,db)\), so its levels are identified with \(\mathcal O_c\oplus M_c\) by \((b,m)\); along \(a\), \((b,m,db)\) goes to \(\bigl(\mathcal O_a(b),M_a(m)+D_a(b),d(\mathcal O_a(b))\bigr)\). So the graph boundary is \(T_D\). By the identification of the graph boundary with the connecting map (proof of Theorem 10.3 there), \(\kappa(T_D)=\delta(d)\), where \(\delta\colon\operatorname{Ext}^0(L,\Omega)\to\operatorname{Ext}^1(L,M)\) is the connecting map of the sequence and \(d\) is read in \(\operatorname{Ext}^0(L,\Omega)=\operatorname{Hom}(\Omega,\Omega)\) as the identity. (That identification is made there for a sequence whose middle term is injective; for ours, choose an injective \(I''\) containing \(M\) and a map \(I'\to I''\) extending the inclusion, with the induced map \(\Omega\to I''/M\). The graph boundaries for the two sequences are isomorphic extensions of \(\mathcal O\) by \(M\), the map between them being the identity on \(M\) and on \(\mathcal O\), and the connecting maps are compatible.)

The chase. By the horseshoe lemma [Tag 013T], choose injective resolutions \(J^M\), \(J^{I'}\), \(J^\Omega\) of \(M\), \(I'\), \(\Omega\) forming an exact sequence of complexes \(0\to J^M\to J^{I'}\to J^\Omega\to0\), split in each degree. Applying the sequences of Corollary 1.3 for \(K=L\) gives a commutative diagram whose rows are those sequences for \(J^M\), \(J^{I'}\), \(J^\Omega\) and whose columns, \(\operatorname{Hom}^\bullet(L,-)\), \(\operatorname{Lev}(L,-)\), \(\operatorname{Arr}(L,-)\) applied to the split sequence, are exact. Let \(z=d\in H^0\operatorname{Hom}^\bullet(L,J^\Omega)\); then \(\partial^{\mathrm c}(z)=\delta(d)\). Its image in \(H^0\operatorname{Lev}(L,J^\Omega)=\prod_c\operatorname{Hom}(\Omega_c,\Omega_c)\) is the family of identities, which lifts to the family \(\tilde x\) of the inclusions \(\Omega_c\to M_c\oplus\Omega_c\), the sequence being split at every level. For the generator \(a\), \(\Phi(\tilde x)_a\) is the map \(X(a)^*\Omega_s\to I'_t\), \(\omega\mapsto\) \(\bigl(h_a(\omega),\Omega_a(\omega)\bigr)-\bigl(0,\Omega_a(\omega)\bigr)=\) \(\bigl(h_a(\omega),0\bigr)\), using that the structure map of \(L\) along \(a\) induces \(\Omega_a\) on \(H^0\). So \(\rho_2(\tilde x)=\gamma(w)\) with \(w=(h_a)_a=D\). The group \(H^{-1}\operatorname{Arr}(L,J^\Omega)\) vanishes, \(L\) lying in degrees \(\leq0\). Lemma 2.2 gives \(\delta(d)=-\partial_C(D)\). \(\square\)

Corollary 3.3 (two extensions with identified levels). Let \(\mathcal E\) and \(\mathcal E'\) be extensions of \(\mathcal O\) by \(M\) over \(\mathcal Z\), and \(\theta_c\colon\mathcal E_c\to\mathcal E'_c\) isomorphisms of extensions at every level. For a generator \(a\colon s\to t\), the maps \(\theta_t\circ\mathcal E_a\) and \(\mathcal E'_a\circ X(a)^{-1}(\theta_s)\) from \(X(a)^{-1}\mathcal E_s\) to \(\mathcal E'_t\) lift the same map and induce the same map on the kernels, so their difference is \(D_a\circ(X(a)^{-1}\mathcal E_s\to X(a)^{-1}\mathcal O_s)\) for a derivation \(D_a\). Then \[ \kappa(\mathcal E)-\kappa(\mathcal E')=-\partial_C(D). \]

Proof. Transport \(\mathcal E'\) along the \(\theta_c\): it becomes the extension \(\mathcal E''\) with the levels \(\mathcal E_c\) and the structure maps \(\theta_t^{-1}\mathcal E'_aX(a)^{-1}(\theta_s)=\mathcal E_a-D_a\), isomorphic to \(\mathcal E'\). The Baer sum \(\mathcal E+T_{-D}\), the quotient of \(\mathcal E\times_{\mathcal O}T_{-D}\) by the elements \((m,(0,-m))\), has the levels \(\mathcal E_c\), through \((e,(\bar e,m))\mapsto e+m\), and the structure maps \(e+m\mapsto\mathcal E_a(e)+M_a(m)-D_a(\bar e)\); so it is \(\mathcal E''\). As \(\kappa\) is additive, \(\kappa(\mathcal E')=\kappa(\mathcal E)+\kappa(T_{-D})=\kappa(\mathcal E)+\partial_C(D)\) by Proposition 3.2. \(\square\)

4. Deformations of a diagram indexed by a free category

4.0. Setting. Let \(f\colon X\to Y\) be a morphism of diagrams of schemes of type \(u\colon C^{\mathrm{op}}\to J\), with \(C\) as in 1.1 and every \(X(\alpha)\) flat, and let \(Y'\) be an extension of \(Y\) with a termwise quasi-coherent ideal \(\mathcal J\) (Construction 4.2 of the lesson on deformations of diagrams, with its notation): on \(\operatorname{Top}(X_{\mathrm{Zar}})\), \(\mathcal A'=f^{-1}\mathcal O_{Y'}\), \(\mathcal A=f^{-1}\mathcal O_Y\), \(\mathcal B=\mathcal O_X\), \(M=f^*\mathcal J\), the canonical map \(c\colon f^{-1}\mathcal J\to M\), the transitivity triangle \(D\xrightarrow{g}L'\to L\xrightarrow{w}D[1]\) of \(\mathcal A'\to\mathcal A\to\mathcal B\), with \(L'=L_{\mathcal B/\mathcal A'}\), \(L=L_{\mathcal B/\mathcal A}=L_{X/Y}\) and \(D=L_{\mathcal A/\mathcal A'}\otimes^{\mathbf L}_{\mathcal A}\mathcal B\), and the obstruction \(\omega(f,Y')=\tilde c[1]\circ w\). By the proof of Theorem 4.3 there, a deformation of \(f\) over \(Y'\) is an extension of \(\mathcal A'\)-algebras of \(\mathcal B\) by \(M\) with the structure map \(c\), and its class \(\kappa\in\operatorname{Ext}^1(L',M)\) restricts along \(g\) to \(c\in\operatorname{Hom}_{\mathcal A}(f^{-1}\mathcal J,M)=\operatorname{Ext}^1(D,M)\) ((10.9) of The cotangent complex).

For a generator \(a\colon s\to t\) of \(C\), the morphism \(X(a)\colon X_t\to X_s\) lies over \(Y(u(a))\colon Y_{u(t)}\to Y_{u(s)}\). If \(X'_s\) and \(X'_t\) are deformations of \(f_s\) and \(f_t\) over \(Y'_{u(s)}\) and \(Y'_{u(t)}\), with the rings \(\mathcal E_s\) and \(\mathcal E_t\), the obstruction to lifting \(X(a)\) is the class \[ o_a\in\operatorname{Ext}^1\bigl(X(a)^*L_{X_s/Y_{u(s)}},M_t\bigr) \] defined as in the proof of Theorem 3.1 of the lesson on deformations of flat group schemes: Lemma 5.1 of the lesson on equivariant extensions, for the rings \(X(a)^{-1}\mathcal A'_s\to X(a)^{-1}\mathcal B_s\to\mathcal B_t\), the extensions \(X(a)^{-1}\mathcal E_s\) and \(\mathcal E_t\) and the structure map \(c_a\colon X(a)^{-1}M_s\to M_t\) of \(M\), gives the class \(\kappa(\mathcal E_t)\circ\rho_a-c_a\circ\kappa(X(a)^{-1}\mathcal E_s)\) in \(\operatorname{Ext}^1(X(a)^*L'_s,M_t)\), and \(o_a\) is the unique class whose composite with \(X(a)^*\) of the functoriality map \(L'_s\to L_s\) is that class. It vanishes exactly when \(X(a)\) lifts to a morphism \(X'_t\to X'_s\) compatible with the thickenings and lying over \(Y'(u(a))\).

Theorem 4.1. In 4.0, suppose that every level \(f_c\) has a deformation \(X'_c\) over \(Y'_{u(c)}\), and let \(o_a\) be the obstructions to lifting the generators for these deformations. Then \[ \begin{aligned}&\omega(f,Y')=-\partial_C\bigl((o_a)_{a\in A}\bigr)\\&\quad\text{in }\operatorname{Ext}^2(L_{X/Y},M).\end{aligned} \]

Proof. Models. The complexes \(L'\) and \(D\) are bounded above complexes of flat \(\mathcal B\)-modules, and \(g\) is a map of complexes, the functoriality map of the polynomial constructions. Let \(C(g)\) be its cone; the triangle \(D\to L'\to C(g)\xrightarrow{p}D[1]\) is distinguished, so there is an isomorphism \(\varphi\colon L\to C(g)\) with \(p\circ\varphi=w\), compatible with the maps from \(L'\) [Tags 0145, 014A]. Let \(J\) be an injective resolution of \(M\).

The diagram. The exact sequence \(0\to L'\to C(g)\to D[1]\to0\), split in each degree, and the sequences of Corollary 1.3 for \(K=D[1]\), \(C(g)\), \(L'\) form a commutative diagram with exact rows \(P\), \(Q\), \(R\) (for \(D[1]\), \(C(g)\), \(L'\)) and exact columns \(\operatorname{Hom}^\bullet(-,J)\), \(\operatorname{Lev}(-,J)\), \(\operatorname{Arr}(-,J)\). Let \(x=\) \((\kappa(\mathcal E_c))_c\in\) \(H^1(R_2)=\) \(\prod_c\operatorname{Ext}^1(L'_c,M_c)\).

The two images of \(x\). Its image in \(H^1(R_3)\) is \(\Phi(x)_a=c_a\circ X(a)^*\kappa(\mathcal E_s)-\kappa(\mathcal E_t)\circ L'_a\), with \(L'_a=\rho_a\) the functoriality map; this is minus the class defining \(o_a\). So the preimage \(\tilde y\) of Lemma 2.1 is \((-o_a)_a\), read through \(\varphi\). The connecting map of the column \(\operatorname{Lev}\) is restriction along \(g\): for a cocycle \(\psi\) of \(\operatorname{Hom}^1(L'_c,J_c)\), the element \((l,\delta)\mapsto\psi(l)\) of \(\operatorname{Hom}^1(C(g)_c,J_c)\) lifts it, and its differential is \((l,\delta)\mapsto\psi(g(\delta))\), by the formula for the differential of the cone. So \(\partial^{\mathrm c}(x)=(\kappa(\mathcal E_c)\circ g_c)_c=(c_c)_c\), the canonical maps at the levels. This family is the restriction to the levels of the canonical map \(c\in\operatorname{Ext}^1(D,M)\), so its image under \(\Phi\) vanishes; and \(H^1(P_3)=\prod_a\operatorname{Ext}^0(X(a)^*D_s,M_t)=0\), as \(D\) lies in degrees \(\leq-1\).

Conclusion. By Lemma 2.1, \(\partial^{\mathrm r}(\tilde y)=\gamma(c)\), where \(c\in H^2(P_1)=\operatorname{Ext}^1(D,M)\) is the unique class restricting to the \(c_c\), that is, the canonical map \(c\). The map \(\gamma\colon\) \(H^2\operatorname{Hom}^\bullet(D[1],J)\to\) \(H^2\operatorname{Hom}^\bullet(C(g),J)\) is composition with \(p\), so through \(\varphi\), \(\gamma(c)=\tilde c[1]\circ p\circ\varphi=\tilde c[1]\circ w=\omega(f,Y')\). And \(\partial^{\mathrm r}(\tilde y)=\partial_C(-(o_a)_a)\). \(\square\)

So once the levels are deformed, the diagram deforms if and only if the family of the obstructions to lifting the generators lies in the image of \(\Phi\): changing the level deformations by classes \(x_c\in\operatorname{Ext}^1(L_c,M_c)\) changes the family \((o_a)\) by \(-\Phi(x)\), and the generators then lift.

5. Lifting a morphism of diagrams

5.0. Setting. Let \(X\) and \(Y\) be diagrams of schemes of type \(C^{\mathrm{op}}\), with \(C\) as in 1.1 and flat maps, over a scheme \(S\), flat over \(S\); let \(X'\), \(Y'\) be deformations of \(X\to S\) and \(Y\to S\) over a first-order thickening \(S\subseteq S'\) with the quasi-coherent ideal \(I\), and \(v\colon X\to Y\) a morphism of diagrams of type \(\mathrm{id}_C\) over \(S\). An extension of \(v\) is a morphism of diagrams \(v'\colon X'\to Y'\) over \(S'\) compatible with the thickenings. As in the proof of Theorem 3.1 of the lesson on deformations of flat group schemes, with \(\mathcal Z=f_X^{-1}\mathcal O_{S'}\), \(\mathcal A=v^{-1}\mathcal O_Y\), \(\mathcal B=\mathcal O_X\), \(\mathcal E=v^{-1}\mathcal O_{Y'}\), \(\mathcal E'=\mathcal O_{X'}\) and \(M=f_X^*I\), the extensions of \(v\) are the lifts of \(v^\sharp\colon\mathcal A\to\mathcal B\) in the sense of Lemma 5.1 of the lesson on equivariant extensions, and there is a class \[ o(v)\in\operatorname{Ext}^1(Lv^*L_{Y/S},M), \] the unique class whose composite with \(Lv^*\) of the functoriality map \(L_{Y/S'}\to L_{Y/S}\) is \(\kappa(\mathcal E')\circ\rho-c\circ\kappa(\mathcal E)\); it vanishes exactly when \(v\) has an extension.

Theorem 5.1. In 5.0, suppose that every level \(v_c\) extends to a morphism \(v'_c\colon X'_c\to Y'_c\) over \(S'\) compatible with the thickenings, and choose such \(v'_c\). For a generator \(a\colon s\to t\), the two morphisms \(v'_s\circ X'(a)\) and \(Y'(a)\circ v'_t\colon X'_t\to Y'_s\) extend the same morphism; let \(D_a\) be the derivation by which the ring map of the first exceeds that of the second, a class in \(\operatorname{Ext}^0\bigl(X(a)^*Lv_s^*L_{Y_s/S},M_t\bigr)\) (Lemma 5.1 (3) of the lesson on equivariant extensions). Then \[ \begin{aligned}&o(v)=-\partial_C\bigl((D_a)_{a\in A}\bigr)\\&\quad\text{in }\operatorname{Ext}^1(Lv^*L_{Y/S},M).\end{aligned} \]

Proof. The ring \(\mathcal A\) of \(\operatorname{Top}(X_{\mathrm{Zar}})\) satisfies 1.1, its structure maps being \(v^{-1}\) of those of \(\mathcal O_Y\), which are flat. By Lemma 5.1 (1), (2) of the lesson on equivariant extensions, the lifts of \(v^\sharp\) are the isomorphisms between the extensions \(c_*\mathcal E\) and \(v^*\mathcal E'=\mathcal E'\times_{\mathcal B}\mathcal A\) of \(\mathcal A\) by \(M\) over \(\mathcal Z\), and \(\kappa(\mathcal E')\circ\rho-c\circ\kappa(\mathcal E)=\kappa(v^*\mathcal E')-\kappa(c_*\mathcal E)\). The same holds at every level, so the \(v'_c\) give isomorphisms \(\theta_c\colon(c_*\mathcal E)_c\to(v^*\mathcal E')_c\), \(\theta_c[e+m]=(v'^\sharp_c(e)+m,\bar e)\), with \(v'^\sharp_c\colon v_c^{-1}\mathcal O_{Y'_c}\to\mathcal O_{X'_c}\) the ring map of \(v'_c\). The structure map of \(c_*\mathcal E\) along \(a\) comes from \(v_t^{-1}(Y'(a)^\sharp)\), that of \(v^*\mathcal E'\) from \(X'(a)^\sharp\). So \(\theta_t\circ(c_*\mathcal E)_a-(v^*\mathcal E')_a\circ X(a)^{-1}(\theta_s)\) is, on the first component, the ring map of \(Y'(a)\circ v'_t\) minus that of \(v'_s\circ X'(a)\), that is, \(-D_a\). Corollary 3.3, for the ring \(\mathcal A\) over \(\mathcal Z\) and the \(\mathcal A\)-module \(M\), gives \(\kappa(c_*\mathcal E)-\kappa(v^*\mathcal E')=-\partial_C(-D)\), so \(\kappa(v^*\mathcal E')-\kappa(c_*\mathcal E)=-\partial_C(D)\) in \(\operatorname{Ext}^1_{\mathcal A}(L_{\mathcal A/\mathcal Z},M)=\operatorname{Ext}^1(Lv^*L_{Y/S'},M)\), the Ext groups over \(\mathcal A\) of the levels and arrows being identified with those over \(\mathcal B\) by the adjunction of restriction of scalars and \(-\otimes^{\mathbf L}_{\mathcal A}\mathcal B\), as in the proof of Theorem 3.1 of the lesson on deformations of flat group schemes. Here the classes \(D_a\) are read with \(L_{\mathcal A/\mathcal Z}\); composition with the functoriality map \(w\colon Lv^*L_{Y/S'}\to Lv^*L_{Y/S}\) commutes with \(\partial_C\), by the naturality of Corollary 1.3 in \(K\), and it is injective on \(\operatorname{Ext}^1\) and bijective on \(\operatorname{Ext}^0\) (proof of Theorem 3.1 of the lesson on deformations of flat group schemes). So the identity descends to \(Lv^*L_{Y/S}\). \(\square\)

6. Group schemes, homomorphisms and actions

In this section \(S\to T\) and the \(T\)-extension \(S\subseteq S'\) by a quasi-coherent \(\mathcal O_S\)-module \(I\) are as in 4.0 of the lesson on truncated nerves, whose notation we use: \(N=N(G)\), the strict truncations \(\tau_iN\), the complex \(K\) and the isomorphisms \(\theta_i\) of its Theorem 3.1, the filtration \(F^\bullet\) and the groups \(Z^k\), \(B^k\) of its Section 5. Write \(\Gamma=\Delta'_{[1,2]}\), the free category of 1.1 with the generators \(\delta^0,\delta^1,\delta^2\colon[1]\to[2]\), \(\partial_\Gamma\) for the connecting map \(\partial_C\) of Corollary 1.3 for \(C=\Gamma\), and \(\rho\) for the restriction of a diagram of type \((\Delta'_{\leq2})^{\mathrm{op}}\) to \(\Gamma^{\mathrm{op}}\).

Lemma 6.1. Let \(Z\) be a diagram of flat \(S\)-schemes of type \((\Delta'_{\leq2})^{\mathrm{op}}\) with flat maps and \(Z_0=S\), and \(M\) a module of \(\operatorname{Top}(Z_{\mathrm{Zar}})\). The restriction map \(R\operatorname{Hom}(L_{Z/S},M)\to\) \(R\operatorname{Hom}(L_{\rho Z/S},\rho M)\) is an isomorphism, and restriction carries the obstruction \(\omega(Z)\) to \(\omega(\rho Z)\) and deformations of \(Z\) to deformations of \(\rho Z\). The same holds with \(L_{Z/S}\) replaced by \(Lv^*L_{Y/S}\) for a morphism \(v\colon Z\to Y\) of type \(\mathrm{id}\) to another such diagram, and with the obstruction to extending \(v\).

Proof. No arrow of \(\Delta'\) goes from \([1]\) or \([2]\) to \([0]\), so the extension by zero at the level \(0\) is an exact left adjoint of \(\rho\) on modules: a map from a module that vanishes at the level \(0\) is a pair of maps at the levels \(1\), \(2\), compatible with the structure maps among them. Hence \(\rho\) is exact and preserves injectives [Tag 015Z], and \(R\operatorname{Hom}(e(F),M)=R\operatorname{Hom}(F,\rho M)\) for the extension by zero \(e(F)\) of a complex \(F\). The counit \(e(\rho L_{Z/S})\to L_{Z/S}\) is the identity at the levels \(1\), \(2\) and \(0\to L_{S/S}\) at the level \(0\), a quasi-isomorphism, \(L_{S/S}\) being acyclic. So \(R\operatorname{Hom}(L_{Z/S},M)\cong R\operatorname{Hom}(\rho L_{Z/S},\rho M)\), and this composite of the counit and the adjunction is the restriction; moreover \(\rho L_{Z/S}=L_{\rho Z/S}\), the cotangent complex being formed levelwise. The same argument applies to \(Lv^*L_{Y/S}\), whose level \(0\) is \(L_{S/S}\). The statements on obstructions and deformations are proved as Proposition 4.1 of the lesson on truncated nerves and Proposition 4.3 (3) there: their proofs use only that restriction along a functor of index categories is the inverse image of a morphism of ringed topoi, exact and compatible with the polynomial constructions. \(\square\)

Theorem 6.2 (lifting the multiplication). Let \(G\) be a group scheme over \(S\), flat and locally of finite presentation, and \((G'_1,\iota)\) a deformation of \(G\) as a scheme (Proposition 4.2 (1) of the lesson on truncated nerves). Write \(\pi_2\colon G\times_SG\to S\), and let \[ o_\mu(G'_1)\in\operatorname{Ext}^1_{G^2}(d_1^*L_{G/S},\pi_2^*I) \] be the obstruction of 4.0 to lifting \(d_1\colon G\times_SG\to G\), \((g_1,g_2)\mapsto g_2g_1\), to an \(S'\)-morphism \(G'_1\times_{S'}G'_1\to G'_1\) compatible with \(\iota\). Let \(\partial_\mu\colon\) \(\operatorname{Ext}^1_{G^2}(d_1^*L_{G/S},\pi_2^*I)\to\) \(\operatorname{Ext}^2(L_{\tau_2N/S},f^*I)\) be the map \(v\mapsto\partial_\Gamma(0,v,0)\), the components being those of \(\delta^0,\delta^1,\delta^2\), read through Lemma 6.1.

  1. \(\omega(\tau_2N)=-\partial_\mu\bigl(o_\mu(G'_1)\bigr)\).
  2. For \(x\in\operatorname{Ext}^1_G(L_{G/S},\pi^*I)\), let \(x^{(2)}\in\operatorname{Ext}^1_{G^2}(L_{G^2/S},\pi_2^*I)\) be the class with \(x^{(2)}\circ\lambda_0=d_0^*x\) and \(x^{(2)}\circ\lambda_2=d_2^*x\), where \(\lambda_i\colon d_i^*L_{G/S}\to L_{G^2/S}\) are the functoriality maps; it exists and is unique by Lemma 3.1 of the lesson on the adjoint co-Lie complex, \(d_2\) and \(d_0\) being the two projections. Put \(\delta(x)=x^{(2)}\circ\lambda_1-d_1^*x\), an additive map. If \(G''_1\) is the deformation of \(G\) as a scheme whose class exceeds that of \(G'_1\) by \(x\) (Theorem 4.3 (2) of the lesson on deformations of diagrams), then \(o_\mu(G''_1)=o_\mu(G'_1)+\delta(x)\). The kernel of \(\partial_\mu\) is the image of \(\delta\).
  3. For \(v\in\operatorname{Ext}^1_{G^2}(d_1^*L_{G/S},\pi_2^*I)\), let \(\chi(v)\in E_1^{2,1}/B^2\) be the class of any \(y\in E_1^{2,1}=H^3(F^2K/F^3K)\) whose image in \(H^3(F^1K/F^3K)\) is \(\theta_2(\partial_\mu(v))\). Then \(\chi\) is a well defined surjective homomorphism with kernel \(\operatorname{Im}\delta\). The obstruction \(\omega(G,i)\) lies in \(F^2\), and its component in \(F^2/F^3=Z^2/B^2\) is \(-\chi\bigl(o_\mu(G'_1)\bigr)\).

So \(\chi\) identifies the cokernel of \(\delta\) with \(E_1^{2,1}/B^2\), where \(B^2=d_1(E_1^{1,1})\) (Section 5 of the lesson on truncated nerves), and the component of the obstruction in \(F^2/F^3\) is the image of minus the obstruction to lifting the multiplication to a deformation of \(G\) as a scheme. The indeterminacy \(\operatorname{Im}\delta\) is exactly the change of that obstruction with the choice of the deformation.

Proof. (1) By Lemma 6.1, \(\omega(\tau_2N)\) corresponds to the obstruction \(\omega(\rho\tau_2N)\) of the diagram \(\rho\tau_2N\), formed by the three faces \(d_0,d_1,d_2\colon G\times_SG\to G\), over \(S\). The faces are flat: \(d_0\) and \(d_2\) are projections, and \(d_1\) is a projection composed with the automorphism \((g_1,g_2)\mapsto(g_1,g_2g_1)\). Apply Theorem 4.1 to \(\rho\tau_2N\to S\), with the deformations \(G'_1\) at the level \([1]\) and \(G'_1\times_{S'}G'_1\) at the level \([2]\), a deformation of \(G\times_SG\to S\), \(G'_1\) being flat over \(S'\). The faces \(d_0\) and \(d_2\) lift to the projections, so their obstructions vanish (Lemma 5.1 (3) of the lesson on equivariant extensions), and the obstruction for \(d_1\) is \(o_\mu(G'_1)\). So \(\omega(\rho\tau_2N)=-\partial_\Gamma(0,o_\mu(G'_1),0)\).

(2) In 4.0 for \(\rho\tau_2N\to S\), let \(\pi_c\colon L'_c\to L_c\) be the second maps of the transitivity triangles at the levels, \(L_1=L_{G/S}\) and \(L_2=L_{G^2/S}\). By Theorem 4.3 of the lesson on deformations of diagrams and its proof, the class of the level \(1\) of \(G''_1\) is that of \(G'_1\) plus \(x\circ\pi_1\), and the classes of \(G''_1\times_{S'}G''_1\) and \(G'_1\times_{S'}G'_1\) differ by \(x_2\circ\pi_2\) for a class \(x_2\in\operatorname{Ext}^1_{G^2}(L_{G^2/S},\pi_2^*I)\). In the proof of Theorem 4.1, the class defining \(o_a\) is \(-\Phi(\kappa(\mathcal E_c))_a\), for \(\Phi\) formed with \(L'\). Since \(\pi_t\circ L'_a=L_a\circ X(a)^*(\pi_s)\), by the naturality of the functoriality maps, the change of the levels changes that class by \(-\Phi(x,x_2)_a\circ X(a)^*(\pi_s)\), for \(\Phi\) formed with \(L\); so the obstructions satisfy \(o''_a=o'_a-\Phi(x,x_2)_a\), the composition with \(X(a)^*(\pi_s)\) being injective on \(\operatorname{Ext}^1\) (4.0). The structure maps of \(f^*I\) are the identities, so \(\Phi(x,x_2)_{\delta^i}=d_i^*x-x_2\circ\lambda_i\). For \(i=0,2\) both obstructions vanish; hence \(x_2\circ\lambda_i=d_i^*x\) and \(x_2=x^{(2)}\). For \(i=1\) this gives \(o_\mu(G''_1)=\) \(o_\mu(G'_1)-d_1^*x+x^{(2)}\circ\lambda_1=\) \(o_\mu(G'_1)+\delta(x)\). By the exactness of Corollary 1.3, \(\partial_\Gamma(0,v,0)=0\) if and only if \((0,v,0)=\Phi(x_1,x_2)\) for classes \(x_1\in\operatorname{Ext}^1_G(L_{G/S},\pi^*I)\) and \(x_2\in\operatorname{Ext}^1_{G^2}(L_{G^2/S},\pi_2^*I)\). The components of \(\delta^0\) and \(\delta^2\) of this equation say \(x_2=x_1^{(2)}\), and that of \(\delta^1\) then says \(v=d_1^*x_1-x_1^{(2)}\circ\lambda_1=\delta(-x_1)\). So \(\ker\partial_\mu=\operatorname{Im}\delta\).

(3) By Corollary 1.3, \(\partial_\Gamma\) followed by the restriction to the levels vanishes; so \(\partial_\mu(v)\) restricts to zero on \(\tau_1N\), whose Ext groups are those of the level \(1\) (Corollary 3.2 (3) of the lesson on truncated nerves). By Theorem 3.1 there, \(\theta_2(\partial_\mu(v))\) lies in the kernel of \(H^3(F^1K/F^3K)\to H^3(F^1K/F^2K)\), which is the image of \(H^3(F^2K/F^3K)\); the elements \(y\) are unique modulo the kernel of \(H^3(F^2K/F^3K)\to H^3(F^1K/F^3K)\), which is \(B^2\). So \(\chi\) is a well defined homomorphism. Surjectivity: the image of \(E_1^{2,1}\) in \(H^3(F^1K/F^3K)\) corresponds, through \(\theta_2\) and Lemma 6.1, to the classes \(\xi\in\operatorname{Ext}^2(L_{\rho\tau_2N/S},f^*I)\) restricting to zero at the level \(1\). Such a \(\xi\) also restricts to zero at the level \(2\): by Corollary 1.3, its restrictions \(\xi_1\), \(\xi_2\) satisfy \(\Phi(\xi_1,\xi_2)=0\), that is, \(\xi_2\circ\lambda_i=d_i^*\xi_1=0\) for \(i=0,2\), and \((\lambda_0,\lambda_2)\) is an isomorphism. So \(\xi=\partial_\Gamma(v_0,v_1,v_2)\) by Corollary 1.3. Let \(x_2\) be the class with \(x_2\circ\lambda_0=v_0\) and \(x_2\circ\lambda_2=v_2\); then \((v_0,v_1,v_2)+\Phi(0,x_2)=\) \((0,v_1-x_2\circ\lambda_1,0)\), and \(\xi=\partial_\mu(v_1-x_2\circ\lambda_1)\), as \(\partial_\Gamma\circ\Phi=0\). Next, \(\chi(v)=0\) if and only if \(\theta_2(\partial_\mu(v))=0\), that is, \(v\in\ker\partial_\mu=\operatorname{Im}\delta\). Finally, \(\omega(G,i)\in F^2\) by Theorem 5.2 (2) of the lesson on truncated nerves, \(G\) having the deformation \(G'_1\) as a scheme, and by (1) and Theorem 5.2 (3) there, its component is \(\chi(-o_\mu(G'_1))=-\chi(o_\mu(G'_1))\). \(\square\)

Theorem 6.3 (the multiplicativity defect). Let \((F',\iota_F)\) and \((G',\iota_G)\) be deformations of the group schemes \(F\) and \(G\), \(u\colon F\to G\) a homomorphism, and \(\omega=\omega(u,F',G')\), filtered as in Proposition 5.3 of the lesson on truncated nerves. Suppose that \(u\) extends to an \(S'\)-morphism of schemes \(u'_1\colon F'\to G'\) with \(u'_1\iota_F=\iota_Gu\). Write \(\mu_{F'}(a,b)=ba\) and \(\mu_{G'}(a,b)=ba\) for the faces \(d_1\) of the nerves of the deformations, and \(\pi_2\colon F\times_SF\to S\). Let \[ D_\mu(u'_1)\in\operatorname{Ext}^0\bigl(d_1^*Lu^*L_{G/S},\pi_2^*I\bigr) \] be the derivation by which the ring map of \(u'_1\circ\mu_{F'}\) exceeds that of \(\mu_{G'}\circ(u'_1\times u'_1)\), two morphisms \(F'\times_{S'}F'\to G'\) extending \(u\circ d_1=d_1\circ(u\times u)\), and let \(\partial_u\) be the map \(v\mapsto\partial_\Gamma(0,v,0)\) for the complex \(LN(u)^*L_{N(G)/S}\) restricted to \(\rho\tau_2N(F)\), read through Lemma 6.1.

  1. The restriction of \(\omega\) to \(\tau_2N(F)\) is \(-\partial_u\bigl(D_\mu(u'_1)\bigr)\).
  2. For \(e\in\operatorname{Ext}^0(Lu^*L_{G/S},\pi_F^*I)\), let \(u'_1+e\) be the morphism whose ring map exceeds that of \(u'_1\) by \(e\) (Lemma 5.1 (3) of the lesson on equivariant extensions). Then \(D_\mu(u'_1+e)=D_\mu(u'_1)-\delta_u(e)\), where \(\delta_u(e)=e^{(2)}\circ\lambda_1-d_1^*e\) and \(e^{(2)}\in\operatorname{Ext}^0\bigl(L(u\times u)^*L_{G^2/S},\pi_2^*I\bigr)\) is the class with \(e^{(2)}\circ\lambda_i=d_i^*e\) for \(i=0,2\), the \(\lambda_i\) standing for \(L(u\times u)^*\) of the functoriality maps of Theorem 6.2 (2) for \(G\). The kernel of \(\partial_u\) is \(\operatorname{Im}\delta_u\).
  3. The obstruction \(\omega\) lies in \(F^2\), and its component in \(F^2=Z^2/B^2\subseteq E_1^{2,0}/B^2\) is \(-\chi_u\bigl(D_\mu(u'_1)\bigr)\), where \(\chi_u\) is defined from \(\partial_u\) as \(\chi\) in Theorem 6.2 (3); \(\chi_u\) is surjective with kernel \(\operatorname{Im}\delta_u\), and \(B^2=d_1(E_1^{1,0})\).

Proof. (1) By Proposition 4.3 (3) of the lesson on truncated nerves and Lemma 6.1, the restriction of \(\omega\) is the obstruction \(o(v)\) of 5.0 for the morphism \(v=\rho\tau_2N(u)\) of diagrams of type \(\Gamma^{\mathrm{op}}\) and the deformations \(\rho\tau_2N(F')\), \(\rho\tau_2N(G')\). Extend the levels by \(u'_1\) and \(u'_1\times_{S'}u'_1\). For the generators \(\delta^0\) and \(\delta^2\) the two morphisms of Theorem 5.1 agree, the faces being projections; for \(\delta^1\) they are \(u'_1\circ\mu_{F'}\) and \(\mu_{G'}\circ(u'_1\times u'_1)\). Theorem 5.1 gives \(o(v)=-\partial_\Gamma(0,D_\mu(u'_1),0)\).

(2) Changing \(u'_1\) by \(e\) changes the ring map of \(u'_1\circ\mu_{F'}\) by the pullback \(d_1^*e\) of \(e\), and that of \(u'_1\times u'_1\) by \(e\otimes u'^\sharp_1+u'^\sharp_1\otimes e\), whose class is \(e^{(2)}\), the product of two elements of the ideal being zero; so it changes the ring map of \(\mu_{G'}\circ(u'_1\times u'_1)\) by \(e^{(2)}\circ\lambda_1\). Hence \(D_\mu\) changes by \(d_1^*e-e^{(2)}\circ\lambda_1=-\delta_u(e)\). The kernel of \(\partial_u\) is computed as in Theorem 6.2 (2), with \(\operatorname{Ext}^0\) in place of \(\operatorname{Ext}^1\) and Lemma 3.1 of the lesson on the adjoint co-Lie complex pulled back along \(u\times u\).

(3) By Proposition 5.3 of the lesson on truncated nerves, \(\omega\in F^2\), as \(u\) lifts to the morphism of schemes \(u'_1\). The rest is proved as Theorem 6.2 (3), with Proposition 5.3 there, Lemma 5.1 there for \(m=2\), and the remark after it for \(B^2\). \(\square\)

Theorem 6.4 (extending an action). Let \(G\) be a group scheme over \(S\), flat and locally of finite presentation, \(f\colon X\to Y\), \(g\colon Y\to Z\) and \((Y',j,g')\) as in Proposition 4.4 of the lesson on truncated nerves, \(a\colon G\times_SX\to X\) and \(a_{Y'}\) the actions, \(\varphi=N(G,f)\), \(\omega_1\) the obstruction of that proposition for \(\tau_1\varphi\), and \(\omega(f,j)\) filtered as in Proposition 5.4 there. Let \(M=\tau_1\varphi^*\underline J\), with the levels \(M_0=f^*J\) and \(M_1\) and the structure maps \(M_{\delta^0}\colon a^*M_0\to M_1\) and \(M_{\delta^1}\colon\mathrm{pr}_X^*M_0\to M_1\), isomorphisms. Let \((X',\iota,f')\) be a deformation of \(f\) over \(Y'\), and \[ o_a(X')\in\operatorname{Ext}^1\bigl(a^*L_{X/Y},M_1\bigr) \] the obstruction of 4.0 to lifting \(a\) to a morphism \(G\times_SX'\to X'\) compatible with \(\iota\) and lying over \(a_{Y'}\). Let \(\partial_a\) be the map \(v\mapsto\partial_{\Delta'_{\leq1}}(v,0)\), the components being those of \(\delta^0\) and \(\delta^1\), for the morphism \(\tau_1\varphi\).

  1. \(\omega_1=-\partial_a\bigl(o_a(X')\bigr)\).
  2. For \(x\in\operatorname{Ext}^1(L_{X/Y},f^*J)\), let \(x^{(1)}\in\operatorname{Ext}^1(L_{G\times X/G\times Y},M_1)\) be the class with \(x^{(1)}\circ\lambda_1=M_{\delta^1}\circ\mathrm{pr}_X^*x\), where \(\lambda_1\colon\mathrm{pr}_X^*L_{X/Y}\to L_{G\times X/G\times Y}\) is the functoriality map, an isomorphism by flat base change (Proposition 1.4 of the lesson on deformations of diagrams), and put \(\delta_a(x)=x^{(1)}\circ\lambda_0-M_{\delta^0}\circ a^*x\), with the functoriality map \(\lambda_0\colon a^*L_{X/Y}\to L_{G\times X/G\times Y}\). If the class of the deformation \(X''\) of \(f\) exceeds that of \(X'\) by \(x\), then \(o_a(X'')=o_a(X')+\delta_a(x)\), and \(\ker\partial_a=\operatorname{Im}\delta_a\).
  3. The obstruction \(\omega(f,j)\) lies in \(F^1\), and its component in \(F^1/F^2=Z^1/B^1\subseteq E_1^{1,1}/B^1\) is \(-\chi_a\bigl(o_a(X')\bigr)\), where \(\chi_a\) is defined from \(\partial_a\) as \(\chi\) in Theorem 6.2 (3), through Proposition 4.4 (4) of the lesson on truncated nerves in place of \(\theta_2\); \(\chi_a\) is surjective with kernel \(\operatorname{Im}\delta_a\), and \(B^1=d_1(E_1^{0,1})\).

Proof. (1) Apply Theorem 4.1 to \(\tau_1\varphi\colon\tau_1N(G,X)\to\tau_1N(G,Y)\), of type \(\mathrm{id}\), over \(\tau_1N(G,Y')\), with the deformations \(X'\) at the level \([0]\) and \(G\times_SX'\) at the level \([1]\), a deformation of \(\varphi_1\) over \(G\times_SY'\) by base change along the flat morphism \(G\to S\). The maps of \(\tau_1N(G,X)\) are flat, \(a\) being a projection composed with the automorphism \((g,x)\mapsto(g,gx)\). The face \(d_1=\mathrm{pr}_X\) lifts to the projection, so its obstruction vanishes; that of \(d_0=a\) is \(o_a(X')\).

(2) As in Theorem 6.2 (2): the class of the level \(1\) changes by \(x_1\circ\pi_1\) for a class \(x_1\), the obstructions change by \(-\Phi(x,x_1)\), and \(\Phi(x,x_1)_{\delta^i}=M_{\delta^i}\circ d_i^*x-x_1\circ\lambda_i\). The obstruction of \(\delta^1\) vanishes for both deformations, so \(x_1=x^{(1)}\), and the component of \(\delta^0\) changes by \(-(M_{\delta^0}\circ a^*x-x^{(1)}\circ\lambda_0)=\delta_a(x)\). The kernel of \(\partial_a\) is computed in the same way: \((v,0)=\Phi(x_0,x_1)\) forces \(x_1=x_0^{(1)}\) and \(v=\delta_a(-x_0)\).

(3) By Proposition 5.4 of the lesson on truncated nerves, \(\omega(f,j)\in F^1\), as \(f\) has the deformation \(X'\). The rest is proved as Theorem 6.2 (3), with Proposition 5.4 there and the remark after its Lemma 5.1; a class vanishing at the level \(0\) vanishes at the level \(1\), \(\lambda_1\) being an isomorphism. \(\square\)

7. Exercises

Exercise 7.1. For \(C=\Delta'_{\leq1}\) and a module \(K\), write out the sequence of Proposition 1.2 at the levels \([0]\) and \([1]\), and the maps \(\sigma\) and \(h\) of its proof.

Exercise 7.2. Let \(C\) have two objects \(s\), \(t\) and two generators \(a,b\colon s\to t\); let \(X\) be the diagram with \(X_s=X_t=\mathbf A^1_k\) over a field \(k\), with the Zariski topoi, and both maps the identity, \(\mathcal O=\mathcal O_X\), \(\mathcal Z=k\), \(M\) the module with \(M_s=0\) and \(M_t=\mathcal O_{\mathbf A^1}\), and \(D=(D_a,D_b)=(\partial/\partial x,0)\). Show directly that \(T_D\) is not isomorphic to the trivial extension, and compare with Proposition 3.2 and Corollary 1.3.

Exercise 7.3. Let \(k\) be a field of characteristic \(p>0\), \(S=T=\operatorname{Spec}k\), \(S'=\operatorname{Spec}k[\epsilon]\) with \(\epsilon^2=0\), \(I=k\epsilon\), and \(G=\alpha_p=\operatorname{Spec}k[x]/(x^p)\) with the addition. For a polynomial \(c\in k[x]\) of degree \(<p\) let \(G'_c=\operatorname{Spec}k[\epsilon][x]/(x^p-\epsilon c(x))\).

  1. Show that \(G'_c\) is a deformation of \(G\) as a scheme, and that the addition lifts to an \(S'\)-morphism \(G'_c\times_{S'}G'_c\to G'_c\) compatible with \(\iota\) if and only if \(c(x+y)=c(x)+c(y)\) in \(k[x,y]/(x^p,y^p)\), that is, if and only if \(c=\lambda x\) for some \(\lambda\in k\).
  2. Show that \(\operatorname{Ext}^1_G(L_{G/S},\pi^*I)\cong k[x]/(x^p)\), and that the class of \(G'_c\) exceeds that of \(G'_0\) by the element \(c\). Explain the result of (1) with Theorem 6.2 (2). Why does this not contradict the existence of a deformation of \(G\) as a group scheme?

8. Complete solutions

7.1. The arrows of \(\Delta'_{\leq1}\) into \([0]\) are the identity, into \([1]\) the identity and \(\delta^0,\delta^1\). The first term is \(\bigoplus_{j}[1]_!(X(\delta^j)^*K_0)\), with the levels \(0\) and \(X(\delta^0)^*K_0\oplus X(\delta^1)^*K_0\); the middle term \([0]_!K_0\oplus[1]_!K_1\) has the levels \(K_0\) and \(X(\delta^0)^*K_0\oplus X(\delta^1)^*K_0\oplus K_1\). At the level \([0]\) the sequence is \(0\to0\to K_0\xrightarrow{\mathrm{id}}K_0\to0\). At the level \([1]\), \(\iota(y_0,y_1)=(y_0,y_1,-K_{\delta^0}y_0-K_{\delta^1}y_1)\) and \(\varepsilon(y_0,y_1,z)=K_{\delta^0}y_0+K_{\delta^1}y_1+z\). The map \(\sigma\) is \(z\mapsto(0,0,z)\), and \(h(y_0,y_1,z)=(y_0,y_1)\), the first factors of \(\delta^0\) and \(\delta^1\) being themselves; indeed \(\iota h+\sigma\varepsilon=\mathrm{id}\) and \(h\iota=\mathrm{id}\).

7.2. The levels of \(T_D\) are \(\mathcal O\oplus0=\mathcal O\) at \(s\) and \(\mathcal O\oplus\mathcal O\) at \(t\), and the two ring maps \(\mathcal O\to\mathcal O\oplus\mathcal O\) are \(r\mapsto(r,r')\) along \(a\) and \(r\mapsto(r,0)\) along \(b\), where \(r'\) is the derivative. An isomorphism \(\theta\) from \(T_D\) to the trivial extension is an isomorphism of extensions at every level, compatible with the structure maps: at \(s\) it is the identity, at \(t\) it is \((r,m)\mapsto(r,m+E(r))\) for a derivation \(E\) of \(\mathcal O\) over \(k\). Compatibility along \(b\) gives \((r,E(r))=(r,0)\), so \(E=0\); compatibility along \(a\) then gives \((r,r')=(r,0)\) for all \(r\), which fails for \(r=x\). In Corollary 1.3, \(L\) is quasi-isomorphic at both levels to \(\Omega=\mathcal O\,dx\) in degree \(0\), so \(H^0\operatorname{Lev}=\operatorname{Hom}(\Omega,M_t)=\Gamma(\mathbf A^1,\mathcal O)\) at \(t\) and \(0\) at \(s\), \(H^0\operatorname{Arr}=\Gamma(\mathbf A^1,\mathcal O)^2\), and \(\Phi(f_t)=(-f_t,-f_t)\), the maps \(M_a\), \(M_b\) being zero; \(H^1\operatorname{Lev}=0\), the levels being affine and \(\Omega\) free. So \(\partial_C\) identifies the cokernel of the diagonal map \(\Gamma(\mathbf A^1,\mathcal O)\to\Gamma(\mathbf A^1,\mathcal O)^2\) with \(\operatorname{Ext}^1(L,M)\). Reading \(D\) as \((1,0)\), through \(dx\), Proposition 3.2 gives \(\kappa(T_D)=-\partial_C(1,0)=\partial_C(-1,0)\), and \((-1,0)\) is not on the diagonal: a nonzero class, as found directly.

7.3. (1) The ring \(k[\epsilon][x]/(x^p-\epsilon c(x))\) is free over \(k[\epsilon]\) with the basis \(1,x,\ldots,x^{p-1}\), the polynomial \(x^p-\epsilon c(x)\) being monic of degree \(p\), and it reduces to \(k[x]/(x^p)\) modulo \(\epsilon\); so \(G'_c\) is flat over \(S'\) with the fibre \(G\), and it is a deformation of \(G\) as a scheme (Lemma 1.1 of the lesson on deformations of flat group schemes). Let \(R'\) be the ring of \(G'_c\times_{S'}G'_c\), the quotient of \(k[\epsilon][x,y]\) by \(x^p-\epsilon c(x)\) and \(y^p-\epsilon c(y)\), free over \(k[\epsilon]\) with the basis \(x^iy^j\), \(i,j<p\). A lift of the addition is a \(k[\epsilon]\)-algebra map sending \(x\) to \(x+y+\epsilon h\) for some \(h\in k[x,y]\); it is well defined if and only if the image of \(x^p-\epsilon c(x)\) vanishes in \(R'\). In characteristic \(p\), with \(\epsilon^2=0\), \((x+y+\epsilon h)^p=\) \((x+y)^p=\) \(x^p+y^p=\) \(\epsilon(c(x)+c(y))\), and \(\epsilon c(x+y+\epsilon h)=\epsilon c(x+y)\). So the condition is \(\epsilon\bigl(c(x)+c(y)-c(x+y)\bigr)=0\) in \(R'\), that is, \(c(x+y)=c(x)+c(y)\) in \(k[x,y]/(x^p,y^p)\), as \(\epsilon r=0\) in \(R'\) exactly when \(r\in(x^p,y^p)\). Write \(c=\sum_{n<p}c_nx^n\). Then \(c(x+y)-c(x)-c(y)=\) \(-c_0+\sum_{2\leq n<p}c_n\bigl((x+y)^n-x^n-y^n\bigr)\), and for \(2\leq n<p\) the polynomial \((x+y)^n-x^n-y^n\) has the coefficient \(n\neq0\) at \(x^{n-1}y\); these monomials are distinct and of degrees \(<p\) in \(x\) and in \(y\). So the condition holds exactly when \(c=c_1x\). For \(c=\lambda x\) the map \(x\mapsto x+y\) is a lift, and \(G'_{\lambda x}\), the kernel of \(x\mapsto x^p-\epsilon\lambda x\) on the additive group over \(k[\epsilon]\), is a deformation of \(G\) as a group scheme.

(2) \(G\) is the closed subscheme \(x^p=0\) of \(\mathbf A^1_k\), so the truncation \(\tau_{\geq-1}L_{G/S}\) is represented by the complex \((x^p)/(x^{2p})\to\Omega_{k[x]/k}\otimes k[x]/(x^p)\), \(x^p\mapsto px^{p-1}dx=0\), of free \(k[x]/(x^p)\)-modules of rank one in degrees \(-1\) and \(0\), and \(\operatorname{Ext}^1\) only depends on this truncation. As \(G\) is affine and \(\pi^*I=(k[x]/(x^p))\epsilon\) is quasi-coherent, \(\operatorname{Ext}^1_G(L_{G/S},\pi^*I)=\) \(\operatorname{Hom}\bigl((x^p)/(x^{2p}),\pi^*I\bigr)\), and sending \(x^p\) to \(\epsilon r\) identifies it with \(k[x]/(x^p)\ni r\). The class (10.6) of an extension is given by its relation map: the image of each relation under a lift of the presentation (proof of Theorem 10.3 of The cotangent complex). Present \(k[x]/(x^p)\) over \(k[\epsilon]\) by \(k[\epsilon][x]\), with the relations \(\epsilon\) and \(x^p\); the lift \(x\mapsto x\) into \(G'_c\) sends \(\epsilon\) to \(\epsilon\) and \(x^p\) to \(\epsilon c(x)\). The functoriality map to \(L_{G/S}\), formed with the presentation \(k[x]\), sends the relations \(\epsilon\) and \(x^p\) to \(0\) and \(x^p\). So the classes of \(G'_c\) and \(G'_0\) differ by the composite of the class \(x^p\mapsto\epsilon c(x)\), that is \(c\), with the functoriality map, and the class of \(G'_c\) exceeds that of \(G'_0\) by \(c\). In the same way \(\operatorname{Ext}^1_{G^2}(d_1^*L_{G/S},\pi_2^*I)\) is identified with \(k[x,y]/(x^p,y^p)\) through the image of the pulled back relation \(d_1^*(x^p)\), and \(\operatorname{Ext}^1_{G^2}(L_{G^2/S},\pi_2^*I)\) with pairs, through the images of the relations \(x^p\) and \(y^p\) of \(k[x,y]\). Then \(c^{(2)}\) is \((c(x),c(y))\); the functoriality map \(\lambda_1\) of the addition, formed with \(x\mapsto x+y\), sends \(d_1^*(x^p)\) to \((x+y)^p=x^p+y^p\); and \(d_1^*c\) sends \(d_1^*(x^p)\) to \(\epsilon c(x+y)\). So \(\delta(c)=c(x)+c(y)-c(x+y)\). As the addition lifts to \(G'_0\), \(o_\mu(G'_0)=0\), and Theorem 6.2 (2) gives \(o_\mu(G'_c)=\delta(c)\), which vanishes exactly in the case found in (1). The obstruction \(o_\mu(G'_c)\) depends on the choice of \(G'_c\): by Theorem 6.2 (3), only its class modulo \(\operatorname{Im}\delta\) determines the component of \(\omega(\alpha_p,i)\), and here that class is zero, \(o_\mu(G'_c)\) lying in \(\operatorname{Im}\delta\). This agrees with \(\omega(\alpha_p,i)=0\): the additive group scheme \(\alpha_p\) over \(k[\epsilon]\), the case \(c=0\), is a deformation of \(G\) as a group scheme.

Sources

The description of the graded pieces of the deformation obstructions of group schemes, homomorphisms and actions by the obstruction to lifting the multiplication, the defect of multiplicativity and the obstruction to extending the action is due to Illusie, in his work on the cotangent complex. For smooth groups these obstructions are those of the description of deformations of group schemes by Hochschild cohomology in the third exposé of the seminar on group schemes of Demazure and Grothendieck. The description of the Ext groups of modules over a diagram indexed by a free category by a cocone of level terms and arrow terms is the analogue of the standard resolution of a representation of a quiver. The homological algebra used here is that of the Stacks project.