7
$\begingroup$

Let $G$ be a group and let $A$ be an abelian group equipped with an action of $G$. Group extensions $$1 \longrightarrow A \longrightarrow \Gamma \longrightarrow G \longrightarrow 1$$ inducing the given action of $G$ on $A$ are classified by elements of $H^2(G;A)$.

We can understand this topologically as follows. Associated to our group extension is a map of classifying spaces $B \Gamma \rightarrow BG$ with homotopy fiber $BA$. If the action of $G$ on $A$ is trivial (so the extension is central), then you can choose your models for $B\Gamma$ and $BG$ and $BA$ such that $BA$ is a topological group and $B\Gamma \rightarrow BG$ is a principal $BA$-fibration. These are classified by maps $BG \rightarrow B (BA)$, and since $B (BA)$ is a $K(A,2)$ this classification amounts to an element of $H^2(BG;A) = H^2(G;A)$.

The above fails for non-central extensions, and $B\Gamma \rightarrow BG$ is not a principal $BA$-bundle.

Question: Is there a similar homotopical description of the element of $H^2(G;A)$ associated to a non-central extension? I've seen various descriptions of classifying spaces for general fibrations, but I don't understand their obstruction theory well enough to understand how the local coefficients would enter the picture.

$\endgroup$
7
  • $\begingroup$ Section 1.1.2 of arxiv.org/abs/1704.03021 is one such description. $\endgroup$ Commented Oct 30, 2023 at 19:51
  • $\begingroup$ Not a complete description but roughly how it should go: we can talk about principal $G$-bundles not only for $G$ a suitable kind of group but, over a base $X$, $G$ itself can be a suitable kind of group over $X$. When the action of $G$ on $A$ is nontrivial it defines a twisted version of $BA$ over $BG$ and then the claim is that $B \Gamma$ is the space of global sections of a twisted $BA$-bundle, classified by cohomology with local coefficients. $\endgroup$ Commented Oct 30, 2023 at 20:04
  • 4
    $\begingroup$ $H^2(G;A)$ is the group of pointed homotopy classes of maps $BG\rightarrow K(A,2)_{hAut(A)}$ that induce the given action $G\rightarrow Aut(A)$ on fundamental groups (here the h-subscript denotes homotopy orbits). Fibrations over $BG$ with fibre $BA\simeq K(A,1)$ and an identification of the fibre over a fixed basepoint are classified by pointed maps $BG\rightarrow BhAut(K(A,1))$ where $Aut(K(A,1))$ is the topological monoid of self-homotopy-equivalences of $K(A,1)$. $\endgroup$ Commented Oct 30, 2023 at 20:43
  • 4
    $\begingroup$ Now one can compute $BAut(K(A,1))\simeq K(A,2)_{hAut(A)}$ and under this equivalence those fibrations whose monodromy over the basepoint agrees with the given action correspond to those pointed maps $BG\rightarrow K(A,2)_{hAut(A)}$ that induce the given action on fundamental groups. $\endgroup$ Commented Oct 30, 2023 at 20:43
  • $\begingroup$ @archipelago : this seems like a complete answer :) $\endgroup$ Commented Oct 31, 2023 at 8:26

1 Answer 1

5
$\begingroup$

The short note Group Extensions and $H^3$ by P. J. Morandi works out the details of extensions by nonabelian groups. Any extension of $G$ by $A$ has an induced group homomorphism $\def\Out{{\sf Out}} ω\colon G→\Out(A)$. Conversely, given a group homomorphism $ω\colon G→\Out(A)$, we can construct a canonical cocycle with cohomology class $\def\H{{\sf H}}\def\Z{{\sf Z}}[c]∈\H^3(G,\Z(A))$, which vanishes if and only if $ω$ comes from a group extension.

If $[c]=0$, then trivializations of $c$ form a torsor over $\H^2(G,\Z(A))$ whose elements are in bijection with group extensions corresponding to $ω$.

This, in bundle-theoretic terms, a group homomorphism $ω\colon G→\Out(A)$ yields a bundle 2-gerbe over $\def\B{{\sf B}}\B G$ with structure group $\Z(A)$, whose trivializations are in bijection with group extensions corresponding to $ω$. Equivalently (using transgression), we can talk about a multiplicative gerbe over $G$ with structure group $\Z(A)$, whose (multiplicative) trivalizations are in bijection with group extensions corresponding to $ω$.

$\endgroup$
1
  • $\begingroup$ The stuff about extensions with non-abelian kernel is basically due to Eilenberg-Mac Lane and is sketched in Brown's book on group cohomology. Maybe I should've spelled this out more, but I am familiar with the classical group-theoretic treatment of extensions -- I've taught it many times, and I've used it in papers -- so what I'm really looking for is how to avoid all the explicit sections, cocycles, etc and do the classification at the level of classifying spaces and homotopy theory. You gesture at that in your 3rd paragraph, but it's unclear to me how to turn that into a rigorous proof. $\endgroup$ Commented Oct 31, 2023 at 13:28

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .