# Relations of diagrams and the associativity of an action

*Written by Claude Opus 5.5 (Anthropic) on 23 October 2026. Self-checked by the writing AI. Public domain (CC0).*

When the generating arrows of a diagram of schemes have been lifted to deformations of its schemes, the lifted diagram may still fail to satisfy the relations among the arrows: two composites that agree in the diagram may differ after lifting, by a derivation. For diagrams indexed by categories of height two, in which every relation identifies two composites of two generators, this lesson proves that the obstruction to deforming the diagram is the image of the family of these relation defects under a double connecting map. The proof puts the deformations of the levels in a normal form, as graph boundaries inside one trivial extension by an injective module, in which lifts of arrows become derivations and relation defects become differences of composed derivations.

The double connecting map is then compared with the strict cosimplicial description of the Ext groups of cartesian modules on strict 2-truncated simplicial topoi. For a group scheme acting on a scheme, this identifies the component of the obstruction to deforming an equivariant morphism in the last step of its level filtration: it is minus the class of the associativity defect of an extended action.

**Prerequisites.** [*Arrow obstructions of diagrams and the lifted multiplication*](arrow-obstructions-of-diagrams-and-the-lifted-multiplication.md) (Conventions, 1.1, Proposition 1.2, Corollary 1.3, Lemmas 2.1 and 2.2, 4.0, Theorem 4.1); [*Truncated nerves and the level filtration*](truncated-nerves-and-the-level-filtration.md) (1.1, Lemma 1.2, Proposition 1.3, Proposition 4.4, Section 5); [*Deformations of diagrams of schemes*](deformations-of-diagrams-of-schemes.md) (Construction 4.2, Theorem 4.3); [*Equivariant extensions and deformations*](equivariant-extensions-and-deformations.md) (Lemma 5.1); [*Morphisms of fibred topoi and modules on total topoi*](morphisms-of-fibred-topoi-and-modules-on-total-topoi.md) (Proposition 4.2). From the programme lesson [*The cotangent complex*](course:AG-HP/the-cotangent-complex#extensions-of-sheaves-of-algebras): Theorem 10.3, the class (10.6), 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](https://kokunoyumeto.github.io/stacks-zh-hans-cn/en/derived.html#derived-lemma-morphisms-into-injective-complex)]; the horseshoe lemma [[Tag 013T](https://kokunoyumeto.github.io/stacks-zh-hans-cn/en/homology.html#homology-lemma-horseshoe)]; the snake lemma [[Tag 010H](https://kokunoyumeto.github.io/stacks-zh-hans-cn/en/homology.html#homology-lemma-snake)].

## Conventions

Hom complexes, cones, connecting maps and the classes \(\kappa\) of extensions of rings are as in the Conventions of the lesson on arrow obstructions. For a commutative diagram of complexes whose rows and columns are short exact sequences, the notation \(\rho_1,\rho_2,\gamma,\partial^{\mathrm r},\partial^{\mathrm c}\) of Section 2 of that lesson is used, and its Lemmas 2.1 and 2.2 are applied as stated there.

## 1. Modules over a diagram of height two

**1.1. Setting.** In Sections 1 and 2, \(C\) is a finite category with a function \(h\) from its objects to \(\{0,1,2\}\), the height, such that:

1. every arrow \(c\to c'\) other than an identity has \(h(c')>h(c)\); the arrows with \(h(c')=h(c)+1\) are called *generators*, and \(A\) is their set;
2. every arrow \(\alpha\colon c\to c'\) with \(h(c')=h(c)+2\) has exactly two factorizations \(\alpha=b_1a_1=b_2a_2\) into generators, with \((a_1,b_1)\neq(a_2,b_2)\).

Write \(R\) for the set of these arrows \(\alpha\), called *relations*; for \(r\in R\) the two factorizations are numbered once and for all, and \(s(r)\), \(t(r)\) are the source and target. The categories \(\Delta'_{\leq2}\) and \(\Delta'_{[1,3]}\), the full subcategory of \(\Delta'\) on \([1]\), \([2]\), \([3]\), are of this kind, with the height \(k\) or \(k-1\) on \([k]\): an injective map \([k]\to[k+2]\) misses two elements, and its two factorizations correspond to the order in which they are added. Every arrow of \(C\) is an identity, a generator or a relation. As in 1.1 of the lesson on arrow obstructions, \(X\) is a topos fibred over \(C^{\mathrm{op}}\) with flat morphisms \(X(\alpha)\) of ringed topoi for a ring \(\mathcal O\) of \(\operatorname{Top}(X)\); a module \(K\) has the levels \(K_c\) and the structure maps \(K_\alpha\), with \(K_{b a}=K_b\circ X(b)^*(K_a)\); and \(c_!\) is the exact left adjoint of the restriction to the level \(c\), \((c_!F)_w=\bigoplus_{q\colon c\to w}X(q)^*F\).

**Proposition 1.2 (the resolution of a module).** For every \(\mathcal O\)-module \(K\) the sequence
\[
\begin{aligned}&0 \to\bigoplus_{r\in R}t(r)_!\bigl(X(r)^*K_{s(r)}\bigr)\\ &\quad \xrightarrow{\ \iota_2\ }\bigoplus_{a\in A}t(a)_!\bigl(X(a)^*K_{s(a)}\bigr)\\ &\quad \quad \xrightarrow{\ \iota_1\ }\bigoplus_{c}c_!K_c \xrightarrow{\ \varepsilon\ }K \to0\end{aligned}
\]
is exact. Here \(\varepsilon\) and \(\iota_1\) are as in Proposition 1.2 of the lesson on arrow obstructions, and on the summand of \(r\), with the factorizations \(b_1a_1=b_2a_2\), \(\iota_2\) is adjoint to the map sending \(x\) to
\[
\begin{aligned}&\bigl(x\text{ at }(a_1,b_1)\\&\quad+X(b_1)^*(K_{a_1})(x)\\&\qquad\text{at }(b_1,\mathrm{id})\bigr)\\&\quad-\bigl(x\text{ at }(a_2,b_2)\\&\quad+X(b_2)^*(K_{a_2})(x)\\&\qquad\text{at }(b_2,\mathrm{id})\bigr),\end{aligned}
\]
where \(y\text{ at }(a,p)\) denotes \(y\) in the summand of the arrow \(p\colon t(a)\to t(r)\) of \(\bigl(t(a)_!X(a)^*K_{s(a)}\bigr)_{t(r)}\). The sequence is functorial in \(K\).

**Proof.** Write \(y\text{ at }q\) for \(y\) in the summand of \(q\) of \(\bigl(\bigoplus_cc_!K_c\bigr)_w\). On the summand of \((a,p)\), \(\iota_1\) sends \(y\) to \(y\text{ at }pa-X(p)^*(K_a)(y)\text{ at }p\). So \(\iota_1\iota_2(x)\) is \(x\text{ at }b_1a_1-K_{b_1}X(b_1)^*(K_{a_1})(x)\text{ at }\mathrm{id}\) minus the same expression for the second factorization, which vanishes because \(b_1a_1=b_2a_2\) and \(K_{b_1a_1}=K_{b_2a_2}\); and \(\varepsilon\iota_1=0\) as in the lesson on arrow obstructions. Fix a level \(w\). If \(h(w)\leq1\), no relation has the target \(w\), and every arrow \(q\neq\mathrm{id}\) into \(w\) is a generator; the proof of Proposition 1.2 of the lesson on arrow obstructions applies without change. Let \(h(w)=2\), and choose for every relation \(\alpha\) with the target \(w\) its first factorization \(b_1a_1\) as the preferred one. Define \(\sigma\colon K_w\to\bigl(\bigoplus_cc_!K_c\bigr)_w\) as the inclusion of the summand of \(\mathrm{id}_w\); \(h_1\) on the middle term at the level \(w\) by \(h_1=0\) on the summand of \(\mathrm{id}_w\), \(h_1(y\text{ at }b)=y\text{ at }(b,\mathrm{id})\) for a generator \(b\) with the target \(w\), and
\[
\begin{aligned}&h_1(x\text{ at }\alpha) =x\text{ at }(a_1,b_1)\\&\quad +X(b_1)^*(K_{a_1})(x)\text{ at }(b_1,\mathrm{id})\end{aligned}
\]
for a relation \(\alpha\); and \(h_2\) on the term of the generators at the level \(w\) by \(h_2(x\text{ at }(a_2,b_2))=-x\text{ at }\alpha\) on the second factorization of a relation \(\alpha\), and \(h_2=0\) on the summands \((a_1,b_1)\) and \((b,\mathrm{id})\). Then \(\varepsilon\sigma=\mathrm{id}\), and a direct computation on each summand gives \(\iota_1h_1+\sigma\varepsilon=\mathrm{id}\), \(\iota_2h_2+h_1\iota_1=\mathrm{id}\) and \(h_2\iota_2=\mathrm{id}\). For instance, on the summand of \(x\text{ at }(a_2,b_2)\), \(h_1\iota_1\) gives \(x\text{ at }(a_1,b_1)+\) \(X(b_1)^*(K_{a_1})(x)\text{ at }(b_1,\mathrm{id})-\) \(X(b_2)^*(K_{a_2})(x)\text{ at }(b_2,\mathrm{id})\), and \(\iota_2h_2=-\iota_2(x\text{ at }\alpha)\) adds the difference between \(x\text{ at }(a_2,b_2)\) and that expression; on \(x\text{ at }\alpha\), \(\iota_1h_1(x\text{ at }\alpha)=x\text{ at }\alpha-K_\alpha(x)\text{ at }\mathrm{id}\). So at every level the complex is contractible as a complex of abelian sheaves, hence exact. \(\square\)

For complexes \(K\) and \(J\) of \(\mathcal O\)-modules, let \(\operatorname{Lev}(K,J)\), \(\operatorname{Arr}(K,J)\) and \(\Phi\) be as in the lesson on arrow obstructions, put
\[
\begin{aligned}&\operatorname{Rel}(K,J)\\&\quad =\prod_{r\in R}\operatorname{Hom}^\bullet\bigl(X(r)^*K_{s(r)},J_{t(r)}\bigr),\end{aligned}
\]
and let \(\Psi\colon\operatorname{Arr}(K,J)\to\operatorname{Rel}(K,J)\) be the map of complexes
\[
\begin{aligned}\Psi(g)_r&=\bigl(J_{b_1}\circ X(b_1)^*(g_{a_1})\\&\quad+g_{b_1}\circ X(b_1)^*(K_{a_1})\bigr)\\&\quad-\bigl(J_{b_2}\circ X(b_2)^*(g_{a_2})\\&\quad+g_{b_2}\circ X(b_2)^*(K_{a_2})\bigr).\end{aligned}
\]
The expression in the first bracket is the *composite of \(g\) along the path* \((a_1,b_1)\): for \(g=\Phi(f)\) it is \(\Phi(f)\) computed for the arrow \(b_1a_1\), by the composition rules of \(J\) and \(K\).

**Corollary 1.3 (the relation sequence).** Let \(K\) be a bounded above complex of \(\mathcal O\)-modules and \(J\) a bounded below complex of injective \(\mathcal O\)-modules. The sequence
\[
\begin{aligned}&0 \to\operatorname{Hom}^\bullet(K,J) \to\operatorname{Lev}(K,J)\\&\quad \xrightarrow{\ \Phi\ }\operatorname{Arr}(K,J) \xrightarrow{\ \Psi\ }\operatorname{Rel}(K,J)\\&\quad \to0\end{aligned}
\]
is exact, and natural in \(K\) and \(J\). Let \(Z(K,J)=\operatorname{Im}\Phi=\ker\Psi\), and write \(\partial_1\) and \(\partial_2\) for the connecting maps of the short exact sequences
\[
\begin{gathered}(S_1) \colon\ 0 \to\operatorname{Hom}^\bullet(K,J)\\ \to\operatorname{Lev}(K,J) \to Z(K,J) \to0,\\ (S_2) \colon\ 0 \to Z(K,J) \to\operatorname{Arr}(K,J)\\ \to\operatorname{Rel}(K,J) \to0.\end{gathered}
\]
The *double connecting map* is \(\partial_R=\) \(\partial_1\circ\partial_2\colon\) \(H^n\operatorname{Rel}(K,J)\to\) \(H^{n+2}\operatorname{Hom}^\bullet(K,J)=\) \(\operatorname{Ext}^{n+2}(K,J)\). For \(J\) an injective resolution of \(M\), \(H^n\operatorname{Rel}(K,J)=\) \(\prod_r\operatorname{Ext}^n(X(r)^*K_{s(r)},M_{t(r)})\).

**Proof.** Apply \(\operatorname{Hom}(-,J^q)\) to the sequence of Proposition 1.2 for every term of \(K\), as in the proof of Corollary 1.3 of the lesson on arrow obstructions. The map induced by \(\iota_2\) on the summand of \(r\) is \(\Psi(\cdot)_r\): the map adjoint to \(g_{a_1}\) is \(J_{b_1}\circ X(b_1)^*(g_{a_1})\) on the summand of the path \(b_1\), and that adjoint to \(g_{b_1}\) is \(g_{b_1}\) on the summand of the identity. The cohomology of \(\operatorname{Rel}\) is computed by [Tag 05TG], the levels of \(J\) being injective. \(\square\)

## 2. The obstruction for a diagram of height two

**2.0. Setting.** Let \(f\colon X\to Y\), \(Y'\) and the notation \(\mathcal A'\to\mathcal A\to\mathcal B\), \(M=f^*\mathcal J\), \(c\), \(L'=L_{\mathcal B/\mathcal A'}\), \(L=L_{\mathcal B/\mathcal A}\), the transitivity triangle \(D\xrightarrow{g}L'\to L\xrightarrow{w}D[1]\) and \(\omega(f,Y')=\tilde c[1]\circ w\) be as in 4.0 of the lesson on arrow obstructions, with \(C\) as in 1.1 and every \(X(\alpha)\) flat. Suppose that every level \(f_c\) has a deformation \(X'_c\) over \(Y'_{u(c)}\), with the ring \(\mathcal E_c\), an extension of \(\mathcal A'_c\)-algebras of \(\mathcal B_c\) by \(M_c\), and that every generator \(a\colon s\to t\) lifts to a morphism \(X'_a\colon X'_t\to X'_s\) compatible with the thickenings and lying over \(Y'(u(a))\), with the ring map \(\varphi_a\colon X(a)^{-1}\mathcal E_s\to\mathcal E_t\). For a relation \(r\colon s\to t\) with the factorizations \(b_1a_1=b_2a_2\), the morphisms \(X'_{a_1}\circ X'_{b_1}\) and \(X'_{a_2}\circ X'_{b_2}\) from \(X'_t\) to \(X'_s\) lift the same morphism \(X(r)\); let \(D_r\) be the derivation by which the ring map of the first exceeds that of the second, a class in \(\operatorname{Ext}^0\bigl(X(r)^*L_s,M_t\bigr)\) (Lemma 5.1 (3) of the lesson on equivariant extensions; derivations over \(\mathcal A'\) and over \(\mathcal A\) into \(M\) agree, \(D\) lying in degrees \(\leq-1\)). Call \((D_r)_{r\in R}\) the family of *relation defects*.

**Lemma 2.1 (graph boundaries).** In 2.0, let \(I\) be an injective \(\mathcal B\)-module of \(\operatorname{Top}(X)\) containing \(M\), \(Q=I/M\) and \(q\colon I\to Q\) the projection.

1. For every level \(c\) there is a derivation \(\Delta_c\colon\mathcal B_c\to Q_c\) vanishing on \(\mathcal A'_c\) and an isomorphism of extensions of \(\mathcal A'_c\)-algebras from \(\mathcal E_c\) onto the graph boundary
\[
\begin{aligned}G(\Delta_c)=\{&(b,i)\in\mathcal B_c\oplus I_c:\\&q(i)=\Delta_c(b)\},\end{aligned}
\]
a subring of the trivial extension \(\mathcal B_c\oplus I_c\). Then \(\kappa(\mathcal E_c)=\delta(\Delta_c)\), where \(\delta\colon\operatorname{Ext}^0(L'_c,Q_c)\to\operatorname{Ext}^1(L'_c,M_c)\) is the connecting map of \(0\to M_c\to I_c\to Q_c\to0\) and \(\Delta_c\) is read in \(\operatorname{Hom}(\Omega_{\mathcal B_c/\mathcal A'_c},Q_c)=\operatorname{Ext}^0(L'_c,Q_c)\).
2. Fix such isomorphisms. For a generator \(a\colon s\to t\), the ring maps \(X(a)^{-1}G(\Delta_s)\to G(\Delta_t)\) lifting \(\mathcal B_a\) and inducing \(M_a\) on the kernels are the maps \((b,i)\mapsto\bigl(\mathcal B_a(b),I_a(i)+h(b)\bigr)\) for the derivations \(h\colon X(a)^{-1}\mathcal B_s\to I_t\) over \(\mathcal B_a\), vanishing on \(X(a)^{-1}\mathcal A'_s\), with \(q\circ h=\) \(\Delta_t\circ\mathcal B_a-Q_a\circ X(a)^{-1}(\Delta_s)\).
3. If the lifts along \(a\colon s\to t\) and \(b\colon t\to t'\) correspond to \(h_a\) and \(h_b\), their composite corresponds to \(I_b\circ X(b)^{-1}(h_a)+h_b\circ X(b)^{-1}(\mathcal B_a)\).
4. If \(h_a\) corresponds to \(\varphi_a\) for every generator \(a\), then \(D_r=\Psi(h)_r\), where \(h=(h_a)_a\) is read in \(\operatorname{Arr}\) of the complex \(\Omega_{\mathcal B/\mathcal A'}\) in degree \(0\) with the coefficients \(I\).

**Proof.** (1) The push-out \(\mathcal E_c\oplus_{M_c}I_c\) is an extension of \(\mathcal A'_c\)-algebras of \(\mathcal B_c\) by \(I_c\), whose class lies in \(\operatorname{Ext}^1(L'_c,I_c)=0\), \(I_c\) being an injective \(\mathcal B_c\)-module (restriction to a level has the exact left adjoint \(c_!\)). By Theorem 10.3 of *The cotangent complex* it is isomorphic to the trivial extension \(\mathcal B_c\oplus I_c\). Under such an isomorphism \(\mathcal E_c\) is a sub-\(\mathcal A'_c\)-algebra of \(\mathcal B_c\oplus I_c\) containing \(M_c\) and mapping onto \(\mathcal B_c\), so its elements over \(b\) form a coset \((b,i_b+M_c)\); \(\Delta_c(b)=q(i_b)\) is a derivation, \(\mathcal E_c\) being closed under products, and it vanishes on \(\mathcal A'_c\), which maps to \((a,0)\). So \(\mathcal E_c=G(\Delta_c)\). This is the graph boundary (10.7) of *The cotangent complex* for the sequence \(0\to M_c\to I_c\to Q_c\to0\), whose middle term is injective, and its class is \(\delta(\Delta_c)\) by the identification of the graph boundary with the connecting map in the proof of Theorem 10.3 there.

(2) A lift \(\varphi\colon X(a)^{-1}G(\Delta_s)\to G(\Delta_t)\subseteq\mathcal B_t\oplus I_t\) and the map \(I_a\colon X(a)^{-1}I_s\to I_t\) agree on \(X(a)^{-1}M_s\), so they define a ring map on the push-out \(X(a)^{-1}\bigl(G(\Delta_s)\oplus_{M_s}I_s\bigr)=\) \(X(a)^{-1}(\mathcal B_s\oplus I_s)\), the identification being \(((b,i),i')\mapsto(b,i+i')\). A ring map \(X(a)^{-1}(\mathcal B_s\oplus I_s)\to\mathcal B_t\oplus I_t\) over \(X(a)^{-1}\mathcal A'_s\) lifting \(\mathcal B_a\) and equal to \(I_a\) on \(I\) is \((b,i)\mapsto(\mathcal B_a(b),I_a(i)+h(b))\) for a derivation \(h\) as stated, and it maps \(G(\Delta_s)\) into \(G(\Delta_t)\) exactly when \(Q_a(q(i))+q(h(b))=\Delta_t(\mathcal B_a(b))\) whenever \(q(i)=\Delta_s(b)\), which is the stated condition. Conversely such a map restricts to a lift.

(3) \(\bigl(\mathcal B_b\mathcal B_a(b),I_b(I_a(i)+h_a(b))+h_b(\mathcal B_a(b))\bigr)\) is the composite.

(4) By (3), the ring maps of \(X'_{a_k}\circ X'_{b_k}\), the composites \(\varphi_{b_k}\circ X(b_k)^{-1}(\varphi_{a_k})\), correspond to the composites of \(h\) along the paths \((a_k,b_k)\), \(k=1,2\). Their difference is \(\Psi(h)_r\), with values in \(M_t\), the two composites having the same image under \(q\). \(\square\)

**Theorem 2.2.** In 2.0,
\[
\begin{aligned}&\omega(f,Y')\\&\quad =\partial_R\bigl((D_r)_{r\in R}\bigr)\quad\text{in }\operatorname{Ext}^2(L_{X/Y},M),\end{aligned}
\]
the family of relation defects being read in \(H^0\operatorname{Rel}(L,J)=\) \(\prod_r\operatorname{Ext}^0(X(r)^*L_{s(r)},M_{t(r)})\) for an injective resolution \(J\) of \(M\).

**Proof.** *The relation defects.* Choose \(I\), \(\Delta_c\) and \(h\) as in Lemma 2.1, and, by the horseshoe lemma [Tag 013T], injective resolutions \(J^M\), \(J^I\), \(J^Q\) of \(M\), \(I\), \(Q\) forming a sequence \(0\to J^M\to J^I\to J^Q\to0\) split in each degree; put \(J=J^M\). Consider the commutative diagram whose rows are the sequences \((S_2)\) for \(K=L'\) and the coefficients \(J^M\), \(J^I\), \(J^Q\), and whose columns are \(Z(L',-)\), \(\operatorname{Arr}(L',-)\), \(\operatorname{Rel}(L',-)\) applied to the split sequence. The columns \(\operatorname{Arr}\) and \(\operatorname{Rel}\) are exact, and so is the column \(Z\), by the snake lemma [Tag 010H], \(\Psi\) being surjective. Let \(z\in H^0Z(L',J^Q)\) be the image of \(\Delta=(\Delta_c)_c\in H^0\operatorname{Lev}(L',J^Q)\). Its image in \(H^0\operatorname{Arr}(L',J^Q)\) is \(\Phi(\Delta)_a=Q_a\circ X(a)^*(\Delta_s)-\Delta_t\circ\Omega_a\), which is the image of \(-h\in H^0\operatorname{Arr}(L',J^I)\) by Lemma 2.1 (2). The group \(H^{-1}\operatorname{Rel}(L',J^Q)\) vanishes, \(L'\) lying in degrees \(\leq0\). Lemma 2.2 of the lesson on arrow obstructions gives \(\partial^{\mathrm c}(z)=-\partial^{\mathrm r}(\varpi)\), where \(\varpi\in H^0\operatorname{Rel}(L',J^M)\) is the class with image \(\Psi(-h)=-(D_r)_r\), by Lemma 2.1 (4); so \(\varpi=-(D_r)_r\) and \(\partial^{\mathrm c}(z)=\partial_2\bigl((D_r)_r\bigr)\). The connecting maps of the columns commute with \(\Phi\colon\operatorname{Lev}\to Z\), so \(\partial^{\mathrm c}(z)\) is the image of \(\delta(\Delta)=(\kappa(\mathcal E_c))_c\) (Lemma 2.1 (1)). Hence the image \(\bar x\) of \(x=(\kappa(\mathcal E_c))_c\in H^1\operatorname{Lev}(L',J)\) in \(H^1Z(L',J)\) is \(\partial_2\bigl((D_r)_r\bigr)\), the \(D_r\) read with \(L'\).

*The chase.* As in the proof of Theorem 4.1 of the lesson on arrow obstructions, let \(\varphi\colon L\to C(g)\) be an isomorphism with \(p\circ\varphi=w\), and consider the commutative diagram whose rows \(P\), \(Q\), \(R\) are the sequences \((S_1)\) for \(K=D[1]\), \(C(g)\), \(L'\) and whose columns are \(\operatorname{Hom}^\bullet(-,J)\), \(\operatorname{Lev}(-,J)\), \(Z(-,J)\) applied to the sequence \(0\to L'\to C(g)\to D[1]\to0\), split in each degree; the column \(Z\) is exact by the snake lemma. Here \(H^1(P_3)=H^1Z(D[1],J)=0\): by \((S_2)\) it lies between \(H^0\operatorname{Rel}(D[1],J)\) and \(H^1\operatorname{Arr}(D[1],J)\), products of groups \(\operatorname{Ext}^{-1}\) and \(\operatorname{Ext}^0\) out of complexes in degrees \(\leq-1\), which vanish. The element \(x\in H^1(R_2)\) has the image \(\bar x\in H^1(R_3)\), and \(\partial^{\mathrm c}(x)=(c_c)_c\), as in that proof; it is the restriction to the levels of \(\tilde c\in H^2(P_1)=\operatorname{Ext}^1(D,M)\), so its image in \(H^2(P_3)\) vanishes, \(\operatorname{Hom}^\bullet\to\operatorname{Lev}\to Z\) being zero. By Lemma 2.1 of the lesson on arrow obstructions, \(\bar x\) has a unique preimage \(\tilde y\in H^1(Q_3)\), and \(\partial^{\mathrm r}(\tilde y)=\gamma(\tilde c)=\tilde c[1]\circ p\circ\varphi=\omega(f,Y')\), through \(\varphi\). The class \(\partial_2((D_r)_r)\in H^1Z(L,J)\), with the \(D_r\) read with \(L\), is carried to \(\bar x\) by the map induced by \(L'\to L\), by the naturality of \(\partial_2\) in \(K\); by the uniqueness of \(\tilde y\), it is \(\tilde y\) read through \(\varphi\). So \(\omega(f,Y')=\partial_1\partial_2\bigl((D_r)_r\bigr)\). \(\square\)

So once the levels are deformed and the generators lifted, the diagram deforms if the relation defects lie in the image of \(\Psi\) on \(H^0\); changing the lifts of the generators by derivations \(e_a\) changes the defects by \(\Psi(e)\).

## 3. Comparison with the cosimplicial description

**3.0. Setting.** In this section \(C=\Delta'_{\leq2}\), with the height \(k\) on \([k]\), and \(X\) and \(\mathcal O\) are as in 1.1 of the lesson on truncated nerves for \(n=2\). The relations are the three injective maps \(v\colon[0]\to[2]\), indexed by \(v(0)\in\{0,1,2\}\). With \(\delta^i\colon[0]\to[1]\) and \(\delta^j\colon[1]\to[2]\), the six composites \(\delta^j\delta^i\) give each \(v\) twice: \(v(0)=2\) as \((i,j)=(0,0),(0,1)\); \(v(0)=1\) as \((1,0),(0,2)\); \(v(0)=0\) as \((1,1),(1,2)\). The first factorization of \(v\) is the one with the smaller \(j\); for it \((-1)^{i+j}=(-1)^{v(0)}\), and for the second \((-1)^{i+j}=-(-1)^{v(0)}\).

**Lemma 3.1.** Let \(E\) be a cartesian \(\mathcal O\)-module. The maps of modules \(F_0\), \(F_1\), \(F_2\) from the terms of the cartesian resolution of Lemma 1.2 of the lesson on truncated nerves to those of the resolution of Proposition 1.2, adjoint to
\[
\begin{gathered}F_0\colon E_0\to E_0\text{ at }\mathrm{id}_{[0]},\\ F_1 \colon y\\ \mapsto\textstyle\sum_{i=0,1}(-1)^iE_{\delta^i}^{-1}(y)\text{ at }(\delta^i,\mathrm{id}),\\ F_2 \colon w\\ \mapsto\textstyle\sum_v(-1)^{v(0)}E_v^{-1}(w)\text{ at }v,\end{gathered}
\]
form a map of complexes over the identity of \(E\).

**Proof.** \(F_0\) commutes with the augmentations. For \(y\in E_1\), \(\iota_1F_1(y)=\) \(\sum_i(-1)^i\bigl(E_{\delta^i}^{-1}(y)\text{ at }\delta^i-y\text{ at }\mathrm{id}\bigr)\), which is \(F_0\) of the image \(\sum_i(-1)^iE_{\delta^i}^{-1}(y)\text{ at }\delta^i\) of \(y\) in the cartesian resolution, the terms at \(\mathrm{id}\) cancelling. For \(w\in E_2\), the image of \(w\) in the cartesian resolution is \(\sum_j(-1)^jE_{\delta^j}^{-1}(w)\text{ at }\delta^j\), and \(F_1\) of it is
\[
\textstyle\sum_{i,j}(-1)^{i+j}E_{\delta^j\delta^i}^{-1}(w)\text{ at }(\delta^i,\delta^j),
\]
because \(E_{\delta^j\delta^i}=E_{\delta^j}\circ X(\delta^j)^*(E_{\delta^i})\). On the other hand \(\iota_2F_2(w)\) has, on the summands \((\delta^i,\delta^j)\), the terms \((-1)^{v(0)}E_v^{-1}(w)\) for the first factorization of \(v\) and \(-(-1)^{v(0)}E_v^{-1}(w)\) for the second, which are the same by 3.0. Its terms on the summands \((\delta^j,\mathrm{id})\) are \(\sum(-1)^{i+j}X(\delta^j)^*(E_{\delta^i})(E_{\delta^j\delta^i}^{-1}(w))=\) \(\sum(-1)^{i+j}E_{\delta^j}^{-1}(w)\) over the factorizations \(\delta^j\delta^i\) through \(\delta^j\); for each \(j\) these are \(i=0\) and \(i=1\), so they cancel. A map of modules out of \(k_!E_k\) is determined by its restriction to \(E_k\), so \(\iota_2F_2=F_1\partial\) and \(\iota_1F_1=F_0\partial\). \(\square\)

**Lemma 3.2.** Let \(0\to H\to C^0\xrightarrow{d}C^1\xrightarrow{d}C^2\to0\) be an exact sequence of complexes, \(T\) the total complex of the double complex with the columns \(C^0\), \(C^1\), \(C^2\), and \(H\to T\) the inclusion through \(C^0\), a quasi-isomorphism. Let \(\partial\partial\colon H^n(C^2)\to H^{n+2}(H)\) be the composite of the connecting maps through the image of \(C^0\to C^1\). For a cocycle \(u\) of \(C^2\), placed in \(T\), the class of \(u\) in \(H^{n+2}(T)\) is the image of \(-\partial\partial[u]\).

**Proof.** Write \(d_v\) for the differentials of the complexes \(C^k\). Choose \(\nu\in(C^1)^n\) with \(d\nu=u\) and \(\mu\in(C^0)^{n+1}\) with \(d\mu=d_v\nu\); then \(d_v\mu\) is a cocycle of \(H\) representing \(\partial\partial[u]\). With the total differential \(d+(-1)^kd_v\) on the column \(k\), \(\mu+\nu\) has the boundary \(d_v\mu+d\mu-d_v\nu+u=d_v\mu+u\); with \(d_v+(-1)^nd\) in the vertical degree \(n\), \(\mu+(-1)^n\nu\) has the boundary \(d_v\mu+u\) as well. So \([d_v\mu]=-[u]\) in \(H^{n+2}(T)\) for either sign rule. \(\square\)

**Proposition 3.3.** Let \(E\) be a bounded above complex of cartesian \(\mathcal O\)-modules and \(J\) a bounded below complex of injective \(\mathcal O\)-modules, and \(u=\) \((u_v)_v\in\) \(H^n\operatorname{Rel}(E,J)=\) \(\prod_v\operatorname{Ext}^n(X(v)^*E_0,J_2)\). Under the quasi-isomorphism \(\operatorname{Hom}^\bullet(E,J)\to\operatorname{Tot}\mathcal H(E,J)\) of Proposition 1.3 of the lesson on truncated nerves, \(\partial_R(u)\) maps to the class of the cocycle
\[
-\textstyle\sum_v(-1)^{v(0)}u_v\circ E_v^{-1}
\]
of the column \(\mathcal H^2(E,J)=\operatorname{Hom}^\bullet(E_2,J_2)\).

**Proof.** Applying \(\operatorname{Hom}(-,J)\) to the map of resolutions of Lemma 3.1, termwise in \(E\), gives a map from the sequence of Corollary 1.3 to the exact sequence \(0\to\) \(\operatorname{Hom}^\bullet(E,J)\to\) \(\mathcal H^0\to\) \(\mathcal H^1\to\) \(\mathcal H^2\to\) \(0\) of the proof of Proposition 1.3 of the lesson on truncated nerves, the identity on \(\operatorname{Hom}^\bullet(E,J)\); on \(\operatorname{Rel}\) it is \(u\mapsto\sum_v(-1)^{v(0)}u_v\circ E_v^{-1}\), the adjoint of \(F_2\). Connecting maps are natural, so \(\partial_R(u)\) is the double connecting map of the cosimplicial sequence applied to that element, and Lemma 3.2 gives the sign. \(\square\)

## 4. The associativity of an action

**Theorem 4.1.** 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, \(\varphi=N(G,f)\), and \(\omega(f,j)\) filtered as in Proposition 5.4 there, so that \(F^3=0\). Let \(E\) be the cartesian complex on the fppf topos of \(\tau_2N(G,X)\) and \(J\) the coefficients of the proof of Proposition 4.4 (4) there, through which \(\operatorname{Ext}^k(L_{\tau_2\varphi},\tau_2\varphi^*\underline J)\) is computed by \(\operatorname{Tot}\mathcal H(E,J)\), and \(E_1^{2,0}=H^0\operatorname{Hom}^\bullet(E_2,J_2)\). Let \((X',a')\) be a pair as in Proposition 4.4 (2) there, and \(v_2\colon G\times_SG\times_SX\to X\), \((g_1,g_2;x)\mapsto g_2g_1x\). The morphisms \(a'\circ(\mathrm{id}_G\times a')\) and \(a'\circ(\mu\times\mathrm{id}_{X'})\), \((g_1,g_2;x)\mapsto a'(g_2,a'(g_1,x))\) and \(a'(g_2g_1,x)\), lift \(v_2\); let \(A(a')\in\operatorname{Ext}^0(v_2^*L_{X/Y},M_2)\) be the derivation by which the ring map of the first exceeds that of the second, the *associativity defect*. Then \(\omega(f,j)\in F^2\), and its component in \(F^2=Z^2/B^2\subseteq E_1^{2,0}/B^2\) is the class of
\[
-A(a')\circ E_{v_2}^{-1},
\]
\(A(a')\) being read in \(\operatorname{Ext}^0(v_2^*E_0,J_2)\) through the identifications of Proposition 4.4 (4). In particular the class of \(A(a')\circ E_{v_2}^{-1}\) modulo \(B^2\) does not depend on the choice of \((X',a')\), and it vanishes exactly when some pair has an associative \(a'\), that is, when \(f\) has an equivariant deformation.

**Proof.** By Proposition 5.4 of the lesson on truncated nerves, \(\omega(f,j)\in F^2\), the pair \((X',a')\) existing, and its component in \(F^2\) is the class of any element of \(E_1^{2,0}\) whose image in \(\operatorname{Ext}^2(L_{\tau_2\varphi},\tau_2\varphi^*\underline J)\) is the obstruction \(\omega_2\) of \(\tau_2\varphi\); \(\omega(f,j)=0\) exactly when \(f\) has an equivariant deformation (Proposition 4.4 (3) there). Apply Theorem 2.2 to \(\tau_2\varphi\colon\tau_2N(G,X)\to\tau_2N(G,Y)\) over \(\tau_2N(G,Y')\), with \(C=\Delta'_{\leq2}\), the level deformations \(X'\), \(G\times_SX'\), \(G\times_SG\times_SX'\), and the lifts of the generators \(d'_1=\mathrm{pr}_{X'}\), \(d'_0=a'\) from the level \(1\), and \(d'_2(g_1,g_2;x)=(g_1;x)\), \(d'_1(g_1,g_2;x)=(g_2g_1;x)\), \(d'_0(g_1,g_2;x)=(g_2;a'(g_1,x))\) from the level \(2\); they lie over the faces of \(\tau_2N(G,Y')\), \(a'\) lying over \(a_{Y'}\), and the maps of \(\tau_2N(G,X)\) are flat. For \(v=\delta^j\delta^i\) the morphism \(X'_{\delta^i}\circ X'_{\delta^j}\) is \(d'_i\circ d'_j\). For \(v(0)=0\), \(d'_1d'_1\) and \(d'_1d'_2\) are both the projection to \(X'\); for \(v(0)=1\), the first factorization gives \(d'_1d'_0\) and the second \(d'_0d'_2\), both \((g_1,g_2;x)\mapsto a'(g_1,x)\); for \(v(0)=2\), the first factorization gives \(d'_0d'_0=a'(g_2,a'(g_1,x))\) and the second \(d'_0d'_1=a'(g_2g_1,x)\). So the relation defects are \(0\), \(0\) and \(A(a')\), and \(\omega_2=\partial_R(0,0,A(a'))\). The identifications of Proposition 4.4 (4) of the lesson on truncated nerves are induced by an exact inverse image of fibred topoi commuting with the functors \(c_!\) and \(X(\alpha)^*\), and by levelwise isomorphisms of complexes of modules; they carry the sequences of Corollary 1.3 to each other, by naturality in \(K\) and \(J\). By Proposition 3.3, the image of \(\omega_2\) in \(H^2\operatorname{Tot}\mathcal H(E,J)\) is the class of \(-(-1)^2A(a')\circ E_{v_2}^{-1}\) in the column \(2\), which is the image of that element of \(E_1^{2,0}\). \(\square\)

So the last graded piece of the equivariant obstruction is computed by any extension of the action to a deformation of \(f\): it is minus the class of the associativity defect of that extension, transported to the level \(2\) of the nerve by its cartesian structure. Together with Theorem 6.4 of the lesson on arrow obstructions, which computes the graded piece in \(F^1/F^2\) from the obstruction to extending the action, and Proposition 5.4 of the lesson on truncated nerves, which computes the piece in \(F^0/F^1\) as the obstruction to deforming \(f\), every graded piece of \(\omega(f,j)\) is described by a deformation problem of the truncated nerves.

## 5. Exercises

**Exercise 5.1.** In 3.0, list the six composites \(\delta^j\delta^i\colon[0]\to[2]\) with their values at \(0\), and check that \(\sum_{i}(-1)^{i+j}\), taken over the composites through a fixed \(\delta^j\), vanishes, and that the two factorizations of every \(v\) carry opposite signs \((-1)^{i+j}\).

**Exercise 5.2.** Let \(k\) be a ring, \(S=Y=Z=\operatorname{Spec}k\), \(S'=Y'=\operatorname{Spec}k[\epsilon]\) with \(\epsilon^2=0\) and \(I=J=k\epsilon\), and \(G=\{e,\sigma\}\) the constant group of order \(2\) acting on \(X=\operatorname{Spec}B\) through an involution \(\sigma\) of the \(k\)-algebra \(B\). Take \(X'=\operatorname{Spec}B[\epsilon]\) and the lifts \(a'_e=\mathrm{id}\), \(a'_\sigma=\sigma'\) with \(\sigma'^\sharp=\sigma+\epsilon\theta\).

1. Show that \(\sigma'^\sharp\) is a ring map exactly when \(\theta\colon B\to B\) satisfies \(\theta(xy)=\sigma(x)\theta(y)+\theta(x)\sigma(y)\), that is, \(\theta=\sigma\circ\eta\) for a \(k\)-derivation \(\eta\) of \(B\).
2. Show that the associativity defect \(A(a')\) vanishes on the components \((g_1,g_2)=(e,e),(e,\sigma),(\sigma,e)\) of \(G\times G\times X'\), and that on the component \((\sigma,\sigma)\) it is the derivation \(\eta+\sigma\eta\sigma\).
3. Show that replacing \(\eta\) by \(\eta+\zeta\), for a derivation \(\zeta\), changes the defect by \(\zeta+\sigma\zeta\sigma\), and deduce that, when \(2\) is invertible in \(k\), \(\eta\) can be chosen with \(A(a')=0\), so that \(X\) has an equivariant deformation over \(k[\epsilon]\) with the given \(X'\). Compare with Theorem 4.1.

## 6. Complete solutions

**5.1.** \(\delta^i\) sends \(0\) to \(1-i\), and \(\delta^j\) sends \(0,1\) to \(0,1\) with \(j\) skipped. The composites: \((0,0)\mapsto2\), \((0,1)\mapsto2\), \((0,2)\mapsto1\), \((1,0)\mapsto1\), \((1,1)\mapsto0\), \((1,2)\mapsto0\). For fixed \(j\) the pairs \((0,j)\), \((1,j)\) carry the signs \((-1)^j\) and \(-(-1)^j\), whose sum vanishes; the two factorizations of each value, \((0,0),(0,1)\), \((1,0),(0,2)\) and \((1,1),(1,2)\), have \(i+j\) of different parities.

**5.2.** (1) \(\sigma'^\sharp(xy)=\sigma(x)\sigma(y)+\epsilon\theta(xy)\) and \(\sigma'^\sharp(x)\sigma'^\sharp(y)=\) \(\sigma(x)\sigma(y)+\epsilon(\sigma(x)\theta(y)+\theta(x)\sigma(y))\), \(\epsilon^2=0\); and \(\sigma^{-1}\theta=\sigma\theta\) is then a derivation. (2) \(G\times G\times X'\) is the disjoint union of four copies of \(X'\), and \(A(a')\) on the component \((g_1,g_2)\) compares \(a'_{g_2}\circ a'_{g_1}\) with \(a'_{g_2g_1}\). With \(a'_e=\mathrm{id}\) the first three components agree. On \((\sigma,\sigma)\), \(g_2g_1=e\), and the ring map of \(\sigma'\circ\sigma'\) is \((\sigma+\epsilon\sigma\eta)^2=\) \(\sigma^2+\epsilon(\sigma\sigma\eta+\sigma\eta\sigma)=\) \(\mathrm{id}+\epsilon(\eta+\sigma\eta\sigma)\), \(\sigma\) being an involution; so it exceeds that of \(\mathrm{id}\) by \(\eta+\sigma\eta\sigma\), a derivation of \(B\). (3) The defect is additive in \(\eta\). If \(2\) is invertible, put \(\zeta=-\tfrac12(\eta+\sigma\eta\sigma)\); then \(\sigma\zeta\sigma=\zeta\), and the defect becomes \((\eta+\sigma\eta\sigma)+2\zeta=0\). The pair is then associative, and by Proposition 4.4 (3) of the lesson on truncated nerves it is an equivariant deformation. In Theorem 4.1 the component of \(\omega(f,j)\) in \(F^2\) is the class of minus the defect; here it vanishes, and \(\omega(f,j)=0\), \(f\) having an equivariant deformation. The defect itself depends on the lift \(\sigma'\), through the term \(\zeta+\sigma\zeta\sigma\); by Theorem 4.1 its class modulo \(B^2\) does not.

## Sources

The description of the graded pieces of the obstruction to deforming an equivariant morphism, in particular of the last one by the associativity defect of an extended action, is due to Illusie, in his work on the cotangent complex. The resolution of a module over a category with relations of length two is the analogue of the standard resolution of a representation of a quiver with relations. The identification of the class of a graph boundary with a connecting map is that of the programme lesson on the cotangent complex cited above. The homological algebra used here is that of the Stacks project.
