13
$\begingroup$

Consider the following generalization of residual finiteness to topological groups.

A locally compact Hausdorff group $G$ is called residually compact if for every compact $K \subseteq G$ there is a normal cocompact lattice $\Lambda \subseteq G$ such that the projection $G \to G/\Lambda$ is injective on $K$.

Notice that this clashes with the normal definition of a residual property, but I don't know a better name.

Q1: Has this property been studied under another name?

Up to now we know the following. If $G$ contains a normal cocompact lattice which is residually finite, then the group is residually compact.

On the other hand a residually compact group embeds (algebraically) into a compact group, hence every finitely generated lattice in $G$ is residually finite by Malcev's theorem and the fact that the irreducible linear presentations of a compact group are finite dimensional. If $G$ is compactly generated, then every lattice is finitely generated. In the compactly generated case, residual finiteness is thus the same as containing a residually finite normal cocompact lattice.

Q2: Is there an example of a residually compact topological group that does not contain a residually finite normal lattice?

$\endgroup$
5
  • $\begingroup$ What is a "normal lattice"? Is it the same as a "lattice"? $\endgroup$
    – Misha
    Oct 27, 2015 at 2:21
  • $\begingroup$ @Misha With a "normal lattice" L in G I mean a normal subgroup that is a lattice (discrete subgroup such that G/L has finite measure). But I would also be interested in an example with "normal" being dropped. $\endgroup$ Oct 27, 2015 at 8:21
  • 2
    $\begingroup$ If you remove the normality assumption on the lattice (which quite relaxes the definition), it is called "residually systolic" in arxiv.org/abs/1403.5295, Definition 6.1. $\endgroup$
    – YCor
    Oct 27, 2015 at 8:29
  • $\begingroup$ A discrete cocompact normal subgroup $\Lambda$ is already a lot to ask for. For instance, if your group is compactly generated, then any such $\Lambda$ must have open centraliser, so if $G$ is connected then $\Lambda$ would be central, and if $G$ is totally disconnected you'd have a finite index subgroup of the form $\Lambda \times U$ where $U$ is a compact open subgroup. $\endgroup$
    – Colin Reid
    Oct 27, 2015 at 12:00
  • 1
    $\begingroup$ For an arbitrary compactly generated locally compact group $G$, it implies the existence of a compact normal subgroup $W$ such that the quotient $G/W$ is a isomorphic to an extension with kernel $\mathbf{R}^k$ and discrete quotient, whose action on $\mathbf{R}^k$ goes through $\mathrm{GL}_k(\mathbf{Z})$. Without the compact generation assumption, it still holds that if $W$ is the maximal compact normal subgroup in $G^\circ$, then $(G/W)^\circ=G^\circ/W$ is isomorphic to $\mathbf{R}^k$ for some $k$ and that the conjugation action preserves a lattice in this $\mathbf{R}^k$. $\endgroup$
    – YCor
    Oct 27, 2015 at 12:42

1 Answer 1

7
$\begingroup$

Here's an answer to the mathematical part of the question (namely Q2).

An example is the group of adeles. Start from $\mathbf{A'}$ defined as the inverse image of $\bigoplus_p\mathbf{Q}_p/\mathbf{Z}_p$ in $\prod\mathbf{Q}_p$, with $\prod\mathbf{Z}_p$ prescribed to be a compact open subgroup (the sum being over all primes); then $\mathbf{A}=\mathbf{R}\oplus\mathbf{A}'$. This is a locally compact ring with $\mathbf{Q}$ standing as a discrete cocompact subring through the diagonal embedding (let $Q$ denote its image).

Then $\mathbf{A}$ is "residually compact". Indeed, if we pick a compact subset, then there exists an integer $n\ge 0$, a finite set of primes $I$ with complement $J$ such that this compact subset is contained in the compact subset $$K_{I,n}=[-n,n]\times\bigoplus_{p\in I}p^{-n}\mathbf{Z}_p\oplus\prod_{p\in J}\mathbf{Z}_p.$$ Define $m=\prod_{p\in I}p^n$. The intersection $Q\cap K_{I,n}$ consists of those rationals of the form $m^{-1}k$ with $k\in\mathbf{Z}$ with real absolute value $|m^{-1}k|\le n$. So if we define $\phi$ as the topological group automorphism which the identity on $\mathbf{A}'$ and multiplies the real component by, say, $2mn$, then $\phi(Q)\cap K_{I,n}=\{0\}$. So $\mathbf{A}$ is "residually compact".

On the other hand, $\mathbf{A}$ has no residually finite lattice. Indeed, let $R$ be a lattice. The image of $R$ in $\bigoplus\mathbf{Q}_p/\mathbf{Z}_p$ has finite index, and since the latter group has no proper finite index subgroup, it is all of $\bigoplus\mathbf{Q}_p/\mathbf{Z}_p\simeq\mathbf{Q}/\mathbf{Z}$. The kernel of this projection is $R_1=R\cap (\mathbf{R}\oplus\prod\mathbf{Z}_p)$ and is a cocompact lattice in the open subgroup $\mathbf{R}\oplus\prod\mathbf{Z}_p$, which is torsion-free and compactly generated. In particular $R_1\cap\prod\mathbf{Z}_p$ is trivial and the projection of $R_1$ in $\mathbf{R}$ is injective and its image is a cocompact lattice in $\mathbf{R}$ (discreteness follows from compactness of $\prod\mathbf{Z}_p$). So $R_1$ is an infinite cyclic group. Now let us show that $R$ is divisible by any $p$ (this will imply that $R$ is not residually finite). Let $x\in R$. Then since $R/R_1$ is divisible, we can write $x=py+z$ with $z\in R_1$ and $y\in R$. Now let $M$ be the inverse image in $R$ of the subgroup of order $p$ of $R/R_1$. Then $M$ is torsion-free and contains $R_1$ with index $p$. Hence every element of $R_1$ has a $p$-root in $M$, hence in $R$. So $z=pz'$ with $z'\in R$, thus $x\in pR$. Thus $pR=R$ and $R$ is divisible, hence not residually finite (and actually, necessarily is isomorphic to $\mathbf{Q}$).

$\endgroup$
2
  • $\begingroup$ Thanks a lot for this extensive answer. It took me a while to digest this and I am still struggling to understand some bits. Is "in this open subgroup" referring to $R \oplus \prod Z_p$? $\endgroup$ Nov 30, 2015 at 9:43
  • $\begingroup$ OK, I edited to make this more precise. $\endgroup$
    – YCor
    Nov 30, 2015 at 12:36

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.