Let $G$ be a semisimple Lie group with finite center, $K$ a maximal compact subgroup, and $d=\dim(G/K)$. Let $\Gamma$ be a non-cocompact discrete subgroup of $G$. [Edit: assume that $\Gamma$ is virtually torsion-free (which is automatic if $G$ is linear and $\Gamma$ is finitely generated).]

Is it true that the virtual cohomological dimension (vcd) of $\Gamma$ is $<d$?

The virtual cohomological dimension in this case is the cohomological dimension of some/every torsion-free finite-index subgroup. I guess it's also the rational cohomological dimension.


  • if $\Gamma$ is cocompact the vcd equals $d$;
  • in general, the vcd is $\le d$;
  • if $\Gamma$ is a non-cocompact lattice, then the vcd is $<d$ (at least in the arithmetic case, where it's related to the $\mathbf{Q}$-rank, cf work of Borel-Serre, and also in the rank-1 case; I think the general case follows).

I'd also be interested by variants of this question, where vcd is replaced by the asymptotic dimension, or by Roe's coarse cohomological dimension.

  • $\begingroup$ In case $\Gamma$ is not virtually torsion-free, its vcd is infinite so one has to assume this, as emphasized by Misha in his answer. (When $G$ is non-linear, in some cases it can happen that $\Gamma$ is finitely generated but not virtually torsion-free.) $\endgroup$ – YCor Apr 11 '19 at 8:34

First of all, since you are not assuming finite generation, you should at least assume that $\Gamma$ is virtually torsion-free. (Otherwise, you need to work rationally and reprove Whitehead's lemma, in the setting of orbifolds/orbicomplexes; see below.)

With this extra assumption, here is a proof:

WLOG, $\Gamma$ is torsion-free; let $X=G/K$ be the associated symmetric space so $M=\Gamma\backslash X$ is an open $d$-dimensional manifold. Whitehead proved in Lemma 2.1 in

"The immersion of an open 3-manifold in euclidean space", Proc. London Math. Soc 11 1961, 81-90.

that every open triangulated $d$-dimensional manifold $M$ deformation retracts to its $d-1$-dimensional skeleton $Y\subset M$.

See also Andy Putman's write-up of the proof here.

Thus, in your case, there exists a $d-1$-dimensional $Y=K(\Gamma,1)$. Hence, $cd(\Gamma)\le d-1$.

Edit: If $X$ is an irreducible symmetric space of rank $\ge 2$, I do not know of any examples of discrete subgroups $\Gamma< G$ (which are not non-uniform lattices) of vcd equal to $d-1$. (Most likely, such subgroups simply do not exist.) In contrast, such examples abound in the rank 1 case (more precisely, for real and complex hyperbolic spaces).

| cite | improve this answer | |
  • $\begingroup$ Excellent, thanks! I'm happy with the case when $\Gamma$ is finitely generated, so I accept the answer. $\endgroup$ – YCor Apr 10 '19 at 20:13
  • 1
    $\begingroup$ In addition, it works for an arbitrary virtually connected Lie group $G$. $\endgroup$ – YCor Apr 10 '19 at 20:52

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.