7
$\begingroup$

Let $G$ be a countable amenable group with a (left) Følner sequence $(F_n)$. Let $\Gamma$ be a subset of $G$ that has density $1$ with respect to $(F_n)$, in the sense that $$ \lim_{n \to \infty} \frac{\lvert\Gamma \cap F_n\rvert}{\lvert F_n\rvert} \ = \ 1. $$ Now define $$ \Gamma' := \{ (s,t) \in G^2 : st^{-1} \in \Gamma \}. $$
Is it true that $\Gamma'$ has density $1$ in $G^2$ with respect to $(F_n \times F_n)$?

A starting point is the calculation $$ \lvert\Gamma' \cap F_n^2\rvert = \sum_{\gamma \in \Gamma} \#\{(s,t) \in F_n^2 : st^{-1} = \gamma \} \ = \ \sum_{\gamma \in \Gamma} \lvert F_n \cap \gamma^{-1} F_n\rvert. $$

In the special case $G = \mathbb{Z}$, $F_n = [-n,n]$, the result is true, but my argument uses the geometry of $\mathbb{Z}$ quite strongly. The argument is: in this case, $\lvert F_n \cap \gamma^{-1}F_n\rvert = \max(2n+1 - \lvert\gamma\rvert, 0)$, so if $n$ is chosen large enough so that $\lvert\Gamma \cap F_n\rvert > (1-\epsilon)|F_n|$, then $$ \sum_{\gamma \in \Gamma} \lvert F_n \cap \gamma^{-1} F_n\rvert \ \geq \ \sum_{\epsilon \cdot n < \lvert\gamma\rvert \leq 2n} (2n+1-\lvert\gamma\rvert) \ \geq \ (2n+1)^2 \cdot (1 - O(\epsilon)) \ = \ \lvert F_n^2\rvert(1-O(\epsilon)) $$ as desired.

Is this result true for all countable amenable groups?

$\endgroup$
2
  • $\begingroup$ $\#\{(s, t) \in F_n^2 : s t^{-1} = \gamma\}$ and $\lvert F_n \cap \gamma^{-1}F_n\rvert$ both mean cardinality, right? $\endgroup$
    – LSpice
    Nov 8, 2022 at 3:09
  • 1
    $\begingroup$ @LSpice Yes, that's right $\endgroup$
    – Adam
    Nov 9, 2022 at 3:14

1 Answer 1

2
$\begingroup$

The answer to your question as stated is "no", but a variant of it is true (see the proposition below).

Proof that the answer is "no": Let $(F_n)$ be the Følner sequence in $\mathbb{Z}$ given by $F_n = [2^n, 2^n + n]$, and let $\Gamma = \bigcup_{n \in \mathbb{N}}{F_n}$ so that $\Gamma$ has full density along $(F_n)$. Then $$|\Gamma' \cap (F_n \times F_n)| = \sum_{t \in F_n}{|\Gamma \cap (F_n - t)|}.$$ But for $t \in F_n$, $F_n - t \subset F_n - F_n = [-n, n]$, so $$\frac{|\Gamma' \cap (F_n \times F_n)|}{|F_n|^2} \le \frac{|\Gamma \cap [-n,n]|}{n},$$ and it's easy to check that $$\lim_{n \to \infty}{\frac{|\Gamma \cap [-n,n]|}{n}} = 0.$$//

The key to this counterexample is that $F_n - F_n$ behaves very differently from $F_n$ (in particular, they are disjoint sets). For the special example you considered (with $F_n = [-n,n]$), the difference set $F_n - F_n = [-2n, 2n]$ has substantial overlap with $F_n$, and that's what you were able to utilize.

For general Følner sequences, the basic idea behind your argument can still be adapted to prove the following weaker result:

Proposition: Let $G$ be a countable amenable group. If $\Gamma \subseteq G$ has full density along a Følner sequence $(F_n)$, then there exists a Følner sequence $(\Phi_n)$ in $G^2$ such that $\Gamma'$ has full density along $(\Phi_n)$.

Proof sketch: For each $n \in \mathbb{N}$, choose $N_n$ so that $F_{N_n}$ is almost-invariant with respect to shifts coming from $F_n$, say $$\frac{|F_{N_n} \cap (F_{N_n}t^{-1})|}{|F_{N_n}|} > 1 - \frac{1}{n}$$ for $t \in F_n$. By assumption, $|\Gamma \cap F_{N_n}| > (1 - o(1)) |F_{N_n}|$, so $$|\Gamma' \cap (F_{N_n} \times F_n)| = \sum_{t \in F_n}{|\Gamma \cap F_{N_n}t^{-1}|} \ge |F_n| \left( 1 - o(1) - \frac{1}{n} \right) |F_{N_n}| = (1 - o(1)) |F_{N_n} \times F_n|.$$//

The Proposition can also be proved another (in my opinion, easier) way using the fact that a set has upper Banach density equal to 1 (i.e., full density along some Følner sequence) if and only if the set is thick (contains a translate of every finite set).

Proof sketch for thick sets: Assume $\Gamma$ is thick in $G$. We want to show that $\Gamma'$ is thick in $G^2$. Let $F \subseteq G^2$ be a finite set. Then let $\tilde{F} = \{st^{-1} : (s, t) \in F\}$. This is a finite set, so there exists $x \in G$ such that $x\tilde{F} \subset \Gamma$, since $\Gamma$ is a thick subset of $G$. A simple calculation shows $(x,1)F \subset \Gamma'$.

$\endgroup$

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.