Take a nice space $X$. Let us call a local system on $X$ a functor from the fundamental groupoid of $X$ to groups, so that $G$ is a local system on $X$ if for each $x \in X$ there is a group $G_x$ and for each homotopy class of paths $h$ from $x$ to $y$ there is an isomorphism $h_* : G_x \to G_y$ (plus additional conditions coming from the functor axioms).

A local system morphism $\alpha : G \to H$ is a natural transformation between such functors, that is for each $x$ we have a group morphism $\alpha_x : G_x \to H_x$ and these are compatible with the isomorphisms $h_*$ so that $h\alpha_x = \alpha_yh$. A local system morphism has well-defined kernel and image, which are both local systems. A local system extension is a short exact sequence of local systems $$1 \to A \to G \to H \to 1.$$

Question: assume now $A$ is abelian and moreover it is a trivial local system, that is the constant functor with value $A$ at every $x \in X$. What is the necessary and sufficient condition for such an extension to split, that is when does that projection $G \to H$ have a right inverse as a local system morphism?

Note that for this extension to split at a single point $x$ it is necessary and sufficient that the class of the extension at $x$ vanish in $H^2(H_x; A)$, but it is not enough because $\pi_1(X,x)$ may act nontrivially on $G_x$ and $H_x$, which yields a further obstruction living in $H^1(\pi_1(X,x);\text{Hom}(H_x,A))$. As soon as this obstruction vanishes, the extension splits in the sense that I'm interested in. However, splitting the problem into two separate obstructions seems a bit unnatural to me, and I would like to gain better understanding of this, maybe as some higher-level obstruction in sheaf cohomology or something similar.


There is a version of group cohomology for groups with groups actions, equivariant group cohomology, which encodes an obstruction to finding a splitting, and subsumes both the obstructions I mentioned, and probably is equivalent to them.

In this paper the authors give a thorough description of equivariant group cohomology, and in particular establish a bijection between equivariant extensions and second equivariant group cohomology.

More precisely, given groups $\Gamma$ and $G$ where $\Gamma$ acts on $G$ by automorphisms, there is a notion of $\Gamma$-equivariant $G$-modules $A$, and for such a module the authors define the equivariant group cohomology $$H^*_\Gamma(G;A).$$ Extensions of the form $$0 \to A \to H \to G \to 0,$$ which are $\Gamma$-equivariant, are classified by $H^2_\Gamma(G;A)$.

A short exact sequence of local systems over a nice connected space $X$ is nothing else than an equivariant group extension where the background group is $\pi_1X$. I haven't checked this in detail but it seems that $H^2_{\pi_1X}(G;\mathbb Z_2) = 0$ if and only if $H^2(G;\mathbb Z_2) = 0$ and $H^1(\pi_1X;\text{Hom}(G,\mathbb Z_2)) = 0$, thus answering my question satisfactorily.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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