Deformation gerbes of flat modules beyond finite rank
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).
For a flat module \(M\) over \(R=B/I\), the sign comparison of the lesson on the gerbe of local deformations holds as soon as the first two Ext sheaves of \(M\) into \(N=I\otimes_RM\) vanish. This lesson proves this and two related facts:
- the cohomology \(H^n(T,H)\) of the band \(H=\mathcal Hom_R(M,N)\) agrees with \(\operatorname{Ext}^n_R(M,N)\) for \(n\le2\), and the class of the gerbe is \(-\omega(B,M)\);
- in infinite rank, the band \(H\) need not be \(\mathcal End(M)\otimes I\).
Prerequisites.
- The gerbe of local deformations of a locally free module: Conventions with Lemma 0.1, Lemma 1.1, Lemma 1.4, Definition 1.5, Section 2, and the proof of Theorem 3.1.
- Deformations of modules and of their morphisms: Definition 2.2.
- From the AI-integrated Stacks project: [Tag 05TG] and morphisms out of objects in degrees \(\le n\) [Tag 06XS].
Conventions
The notation of the lesson on the gerbe of local deformations is used, with \(N\xrightarrow{j}J^\bullet\) an injective resolution and \(\mathcal K^q=\mathcal Hom_R(M,J^q)\). Put \[ \mathcal Ext^q_R(M,N)=H^q(\mathcal K^\bullet). \] This does not depend on \(J^\bullet\): two resolutions are homotopy equivalent over \(1_N\), and \(\mathcal Hom_R(M,-)\) preserves homotopies.
1. The comparison and the sign
1.0. Let \(M\) be flat. By Lemma 0.1 there, the \(\mathcal K^q\) are injective, and \(H^0(\mathcal K^\bullet)=H\). Let \(H\xrightarrow{j_L}L^\bullet\) be an injective resolution. By [Tag 05TG], there is a map \(\rho\colon L^\bullet\to\mathcal K^\bullet\), unique up to homotopy, with \(\rho j_L\) the inclusion \(a_H\colon H\to\mathcal K^0\); equality holds because \(\mathcal K^{-1}=0\). The comparison map is \[ \begin{aligned}&c_n =H^n\Gamma(\ast,\rho) \colon H^n(T,H)\\&\quad \to\operatorname{Ext}^n_R(M,N),\end{aligned} \] where \(H^n\Gamma(\ast,\mathcal K^\bullet)=\operatorname{Ext}^n_R(M,N)\) by [Tag 05TG].
Theorem 1.1. Let \(M\) be a flat \(R\)-module with \(\mathcal Ext^1_R(M,N)=\mathcal Ext^2_R(M,N)=0\).
- The map \(c_n\) is an isomorphism for \(n\le2\).
- \(\mathcal G\) is locally nonempty and locally connected, banded by \(H\), and presentations exist.
- Every presentation has \(\check\delta c=0\), and \(c_2([\mathcal G])=-\omega(B,M)\).
Proof. (1) The cone \(C_\rho\) of \(\rho\) is a bounded below complex of injectives. On cohomology sheaves, \(\rho\) is an isomorphism in degree \(0\), and \(H^q(L)=0\) for \(q>0\) while \(H^1(\mathcal K)=H^2(\mathcal K)=0\). So \(H^q(C_\rho)=0\) for \(q\le2\). Then \[ H^n\Gamma(\ast,C_\rho)=\operatorname{Hom}_{D(R)}(R,C_\rho[n])=0 \] for \(n\le2\) [Tags 05TG, 06XS]. The long exact sequence of \(\Gamma(\ast,L^\bullet)\to\Gamma(\ast,\mathcal K^\bullet)\to\Gamma(\ast,C_\rho)\) then shows that \(c_n\) is an isomorphism for \(n\le2\).
(2) It suffices to replace Proposition 2.4 (4) there, which concerns local existence; the rest of the proof of Theorem 3.1 (1) applies verbatim.
Local lifts. Let \(b\colon M|_U\to A_0|_U\). Then \(\bar d^0b\) is a cocycle of \(\mathcal K^1\). Since \(H^1(\mathcal K)=0\), locally \(\bar d^0b=d^0a\) with \(a\colon M\to J^0\). Then \(\bar d^0(\bar a-b)=0\), so \(\bar a=b\), \(\bar d^0\) being injective. This applies to \(b=k^{-1}(t-s)\).
Local sections. The map \(\theta\) is a cocycle of \(\mathcal K^2\). Since \(H^2(\mathcal K)=0\), locally \(\theta=d^1w\) with \(w\colon M\to J^1\). Then \(d^1(g-w\pi)=0\), so \(g-w\pi\) lands in \(\ker d^1=\bar d^0(A_0)\); write \(g-w\pi=\bar d^0r\) with \(r\colon E\to A_0\). From \(rk=1\), the restriction of \(\pi\) to \(\ker r\) is an isomorphism onto \(M\), and its inverse is a section \(s\) of \(\pi\).
(3) The proof of Theorem 3.1 (3) there used finite rank only through (2). It gives \(jc=\check\delta a\), hence \(\check\delta c=0\), and \(D(a-w)=a_Hc-\theta|_{K_0}\) in \(\operatorname{Tot}C^\bullet(K,\mathcal K^\bullet)\).
By Lemma 1.4 there, applied to \(\mathcal K^\bullet\) and to \(L^\bullet\), and by its naturality along \(\rho\), the class \([j_Lc]\) maps under \(\operatorname{Tot}(\rho)\) to \([a_Hc]=[\theta|_{K_0}]\). So \(c_2[\mathcal G]=[\theta]\), which is \(-\omega(B,M)\) by Lemma 2.3 there. \(\square\)
Corollary 1.2. If \(\mathcal Hom_R(M,-)\) is exact and \(M\) is flat, then Theorem 1.1 applies. This holds, for instance, for every free module over a field in the topos of sets, of any rank.
Proof. If \(\mathcal Hom_R(M,-)\) is exact, the complex \(\mathcal K^\bullet\) is exact in degrees \(>0\). In the topos of sets, a free module is projective, so \(\operatorname{Hom}(M,-)\) is exact. \(\square\)
2. The tensor description of the band
Proposition 2.1. Let \(T\) be the topos of sets and \(k=\mathbf Q\). Put \(M=\bigoplus_{n\ge1}ke_n\) and \(I=\bigoplus_{n\ge1}kt_n\), and let \(B=k\oplus I\) with \((a,x)(b,y)=(ab,ay+bx)\), so that \(R=k\). Then the map \[ \operatorname{End}_k(M)\otimes_kI\to\operatorname{Hom}_k(M,I\otimes_kM), \] \(A\otimes v\mapsto(m\mapsto v\otimes Am)\), is not surjective.
Proof. The ring \(B\) is commutative, and \(I\) is an ideal with \(I^2=0\). Let \(h(e_n)=t_n\otimes e_1\), and let \(e_1^*\) be the first coordinate.
An element \(\sum_{j\le r}A_j\otimes v_j\) maps to a homomorphism whose composite with \(1\otimes e_1^*\) takes values in the span of \(v_1,\dots,v_r\). The composite of \(h\) with \(1\otimes e_1^*\) sends \(e_n\) to \(t_n\), and the \(t_n\) span an infinite-dimensional space. So \(h\) is not in the image. \(\square\)
By Corollary 1.2, the band in this example is \(H=\operatorname{Hom}_k(M,I\otimes M)\), not \(\operatorname{End}(M)\otimes I\).
3. Exercises
Exercise 3.1. Show that a finite locally free \(M\) satisfies the hypotheses of Theorem 1.1, and that \(c_2=\iota\) in that case.
4. Complete solutions
3.1. The functor \(\mathcal Hom_R(M,-)\) is exact, so \(\mathcal K^\bullet\) is an injective resolution of \(H\). One may take \(L=\mathcal K\) and \(\rho=1\), and then \(c_2=\iota\).
Sources
The question of the comparison is due to Illusie, in his work on the cotangent complex, and gerbes are due to Giraud. The main argument follows a solution produced with ChatGPT (OpenAI) in October 2026, completed here. The homological algebra is that of the Stacks project.