Let $X$ be a manifold, $G$ and $A$ finite abelian groups and $\epsilon \in H^2(G,A)$ a group cohomology class (for the moment I am assuming there is no action of $G$ on $A$). Given $\alpha \in H^1(X,G)$ there are two natural classes in $H^2(X,A)$ we can construct.

(i) First since $H^2(G,A)$ can be identified with the second cohomology group of $BG$ with $A$ coefficients, and $\alpha$ is associated with a homotopy equivalence class of maps $\widehat{\alpha} :X\rightarrow BG$, we can take the pull-back $\widehat{\alpha}^*\epsilon \in H^2(X,A)$.

(ii) Second, $\epsilon$ identifies uniquely a central extension $$ 1\rightarrow A\rightarrow \widetilde{G}\rightarrow G \rightarrow 1 $$ from which we get the Bockstein map $\beta : H^1(X,G)\rightarrow H^{2}(X,A)$. Then we can construct $\beta(\alpha)\in H^2(X,A)$.

  1. First question: is it true in general that $\widehat{\alpha}^*\epsilon=\beta(\alpha)$? If yes, how to prove it? If no in general, when it happen to be true?

  2. Then I am interested in if and how this story generalizes when $A$ is a non-trivial $G$ module speficied by a homomorphism $\rho :G \rightarrow \text{Aut}(A)$. In this case the twisted group cohomology $H_{\rho}^2(G,A)$ can be identified with the cohomology of $BG$ with local $A$ coefficients $H^2(BG,\widetilde{A})$, and the first construction similarly leads to a class $\widetilde{\alpha}^*\epsilon \in H^2(X,\widetilde{A})$, however I am not sure whether this cohomology group of $X$ with local coefficients makes sense... As for the second construction we again have an extension but this will be non-central. Nevertheless the first Bockstein map $\beta :H^1(X,G)\rightarrow H^2(X,A)$ can still be constructed in the same way and we can consider $\beta(\alpha)$. However it seems to me that this does not live in the cohomology with local $A$ coefficients, but I am really not sure since there might be subtleties in the construction of the Bockstein map. I would be really greatful to anybody could clarify this situation.

  • 1
    $\begingroup$ Even if $A$ is central, $\tilde G$ is usually not abelian. $\endgroup$ Jun 3, 2023 at 12:03
  • $\begingroup$ What is the simplest example? $\endgroup$ Jun 3, 2023 at 12:04
  • 3
    $\begingroup$ $1\to \mathbb{Z}/2 \to D_8 \to \mathbb{Z}/2\times\mathbb{Z}/2 \to 1$ ? $\endgroup$ Jun 3, 2023 at 12:06
  • $\begingroup$ Right, I'll edit the question. Do you agree that the first question doesn't change much since it is enough to have $\beta :H^1(X,G)\rightarrow H^2(X,A)$? $\endgroup$ Jun 3, 2023 at 12:25
  • 2
    $\begingroup$ In general, the Bockstein for a short exact sequence $0\to A \to B \to C \to 0$ of abelian groups is induced by composition with $K(C,n) \to K(A,n+1)$. When $B$ is non-abelian, this only makes sense when $n=1$. $\endgroup$ Jun 3, 2023 at 13:27

1 Answer 1


The general problem with this question is the definition of the Bockstein map. If $\widetilde G$ is non-abelian, then cochains with coefficients in $\widetilde G$ doesn't really make sense. The problem is the boundary map. With abelian coefficients, you get some linear combinations, but with non-abelian coefficients, in what order are you going to multiply?

Topologically, the Bockstein map for a short exact sequence $0 \to A \to B \to C \to 0$ of abelian groups comes from the fact that we have a fibre sequence $K(A,n) \to K(B,n) \to K(C,n) \to K(A,n+1)$, and composing with $K(C,n) \to K(A,n+1)$ gives the appropriate map. In your case, if $A \to \widetilde G \to G$ is a central extension, you do at least have a fibre sequence $$K(A,1) \to K(\widetilde G,1) \to K(G,1) \to K(A,2).$$ Then your Bockstein map $H^1(X,G) \to H^2(X,A)$ is defined, and the answer to your first question is yes.

But in your second question, if $A$ is only normal rather than central, you only get $K(A,1) \to K(\widetilde G,1) \to K(G,1)$. Yes, you get a map to cohomology with twisted coefficients, just by pulling back the class for $BG$ to $X$, but you have no Bockstein to compare it with.

  • $\begingroup$ In this case it seems to me that the pull-back of the extension class, taking values in cohomology with twisted coefficients is a sort of generalization of the Bockstein: it coincide with the Bockstein for trivial twisting but holds more generally. I am a bit surprised that this does not have an "algebraic" definition. $\endgroup$ Jun 5, 2023 at 8:43
  • $\begingroup$ Could you suggest a reference for the equivalence, in the central extension case, between the algebraic definition of the Bockstein and the geometric one composing with $K(G,1)\rightarrow K(A,2)$? $\endgroup$ Jun 5, 2023 at 8:45
  • $\begingroup$ Let me say it again. If $\widetilde G$ is not abelian then there is no algebraic definition of a Bockstein. $\endgroup$ Jun 5, 2023 at 8:53
  • $\begingroup$ Yes sorry, I meant a reference for the case in which the three groups are abelian. $\endgroup$ Jun 5, 2023 at 9:04
  • $\begingroup$ This is really just Yoneda's lemma. Every natural transformation between representable functors is representable. This is where the map $K(G,n) \to K(A,n+1)$ comes from. $\endgroup$ Jun 5, 2023 at 9:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.