MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $G$ be a countable group and $\lambda \colon G \to U(\ell^2 G)$ its left-regular representation. Suppose that there exists a constant $C>0$ such that for all $T \in B(\ell^2 G)$

$$\inf \lbrace\|T-S\| \mid S \in \lambda(G)' \rbrace \leq C \cdot \sup\lbrace \|\lambda(g)T- T \lambda(g) \| \mid g \in G\rbrace.$$

(Here $\lambda(G)'$ denotes the commutant of $G$ in $B(\ell^2 G)$.)

Question: Is $G$ amenable?

It is fairly easy to see that amenability of $G$ implies the existence of such a constant. Indeed, one may take $S$ to be some fixed point for the conjugation action on $\overline{\rm conv}\lbrace \lambda(g)T\lambda(g)^* \mid g \in G\rbrace$. I am asking for the converse of this statement.

EDIT: Since the derivation problem came up in Kate's comment, I want to clarify to what version of it my question is related. The inequality above holds for some $C$ if and only if the first bounded cohomology of $G$ with coefficients in $B(\ell^2 G)$ (with the conjugation action induced by $\lambda$) is reduced. This is a straightforward application of the open mapping theorem. Now, two things are unclear:

Question: Can $H^1_b(G,B(\ell^2 G))$ be reduced without being zero?


Question: Can $H^1_b(G,B(\ell^2 G))$ be zero without $G$ being amenable?

share|cite|improve this question
Can you (directly) show that such C doesn't exist for, say, free group? – Łukasz Grabowski Jan 10 '11 at 23:04
Lukasz, this follows from the work of Pytlik-Szwarc. The idea is roughly as follows. If $G$ admits such a constant $C$ as above, then the constant for any subgroup is at most $C$. Now, for $\mathbb F_n \subset \mathbb F_2$ you consider the operator $T$ on $\ell^2 \mathbb F_n$ which moves a vertex towards the root in a standard Cayley graph of $\mathbb F_n$. One can show directly that the size of $C(\mathbb F_n)$ which is enforced by $T$ tends to infinity. The details need some clever work, but that (I think) is at least the idea. – Andreas Thom Jan 11 '11 at 6:41
You must have thought about it already, but for the sake of completeness: Maybe using Whyte's metric solution to von Neumann problem and imitating the T you described could help? – Łukasz Grabowski Jan 11 '11 at 10:59
The problem is that Kevin Whyte's construction is not equivariant at all. Random forrests have been used by Nicolas Monod and Inessa Epstein to extend the result to non-amenable groups without free subgroups. – Andreas Thom Jan 11 '11 at 14:02
@Yemon: Yes, exactly. – Theo Buehler Jan 12 '11 at 22:09

Let $M=C^*_\lambda(G)''$ be group von Neumann algebra of $G$. The the condition above implies:

$d(T, M')\leq C ||ad(T)|_{M}||$ for every $T\in B(l^2 G)$. The last inequality is equivalent to saying that every derivation of $M$ into $B(l^2 G)$ is inner.

Edit: the above inequality is satisfied automatically (was clarified to me by Stuart White).

It is known that if $M\subset B(H)$ has a cyclic vector, then every bounded derivation from $M$ into $B(H)$ is inner [E. Christensen, Extensions of derivations II, Math. Scand., 1982]. Thus [Christensen, op cit, Cor 5.4] we have

$d(T,M')\leq 3/2\|(\mathrm{ad}| T)_{M}\|$

for every $T\in B(l^2 G)$.

share|cite|improve this answer
I do not understand how you conclude the inequality in your answer. Isn't it is known that all derivations from $M$ into $B(H)$ are inner? What is your conclusion? – Andreas Thom Jan 10 '11 at 20:16
I should have added this as a comment, not as an answer.. Your inequality is stronger than the inequality in the answer and might actually imply amenability. the derivation problem seems to be still unknown. also $||ad(T)|_M||$ is norm of operator $ad(T)$ restricted to $M$, which is not less than $sup ||\lambda(g)T-T\lambda(g)||$. – Kate Juschenko Jan 10 '11 at 20:40
also and references there are related – Kate Juschenko Jan 10 '11 at 21:21

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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