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 x be an element in a C*-algebra A, is it true that if x approximately commute with every element in A, then x is near the centre of A? More precisely, I want to know whether the following is true: Let x be an element in a C*-algebra A with norm 1. Then for any $\epsilon>0$, there exist a $\delta>0$ such that the following holds: $\forall y\in A,||y||=1,||xy-yx||<\delta\Rightarrow dist(x, Centre(A))<\epsilon$. This is true if A is the matrix algebra, but I was wondering whether it can be generalized to any C*-algebra

share|cite|improve this question
Are you allowing $\delta$ to depend on both $\epsilon$ and $x$? Or do you want $\delta$ to depend only on $\epsilon$ and $\Vert x \Vert$? – Yemon Choi Jan 10 '11 at 0:47
up vote 3 down vote accepted

The present question seems very close to the following one, which has been well-studied in the literature:

Let $A$ be a C*-algebra. Does there exist a constant $K>0$ such that, for each $x\in A$, the distance from $x$ to $Z(A)$ is bounded above by $$ K \sup\{ \Vert xy-yx \Vert \;:\; y\in A, \Vert y\Vert \leq 1\} ?$$

A useful discussion of what is known can be found in Section 1 of

[MR2274022] R. J. Archbold, D. W. B. Somerset, Measuring noncommutativity in C*-algebras. JFA 242 (2007) no. 1, 247--271

According to this paper, the answer turns out to be "yes" if $A$ is a von Neumann algebra, or one of several natural classes of C*-algebra; but the answer is "no" in general, by an example in

[MR0215111] R. V. Kadison, E. C. Lance, J. R. Ringrose, Derivations and automorphisms of operator algebras. II. JFA 1 1967 204–221.

I haven't been able to read the Kadison-Lance-Ringrose paper, but a similar example seems to be discussed in Example 3.3 of

[MR0482236] R. J. Archbold, On the norm of an inner derivation of a C*-algebra. Math. Proc. Cambridge Philos. Soc. 84 (1978), no. 2, 273–291.

share|cite|improve this answer
@Yemon: The original question is a bit weaker than the boxed statement here, and is effectively asking whether, for each $\epsilon > 0$, there is a $K(\epsilon)$ such that the distance of any point from the center is bounded by $\epsilon\Vert x\Vert + K(\epsilon)\sup_{\Vert y\Vert\le 1}\Vert xy-yx\Vert$. Your boxed statement is asking whether $K(\epsilon)$ is bounded above as $\epsilon\to0$, which is rather stronger (but maybe you can reduce it to the same thing). – George Lowther Jan 10 '11 at 2:32
...and replacing $\Vert x\Vert$ by $\inf\Vert x-z\Vert$ for $z$ in the center does reduce it to the same thing, for all $\epsilon < 1$. So, they are in fact the same statement. – George Lowther Jan 10 '11 at 2:35
This is exactly what I was looking for. Thank you guys! – Qingyun Jan 10 '11 at 17:29
I'd encourage people to vote up George Lowther's answer, since unlike me he used thought rather than MathSciNet and actually gave an explicit, informative counter-example – Yemon Choi Jan 10 '11 at 17:38
It seems that there is quite a bit of research about on this problem. E.g., if the algebra has trivial center then $K\in\{\frac12,1,\frac12+\frac{1}{\sqrt{3}}\}\cup[\frac32,\infty]$. And lots more surprising results besides that. Eg, – George Lowther Jan 10 '11 at 23:05

In general, the answer to this has to be "no". To construct a counterexample, let $S$ be a dense subset of the unit interval (0,1], and let $\mathcal{H}=\ell^2(S)$ be the Hilbert space with orthonormal basis $(e_s)_{s\in S}$. For each positive integer $n$, let $\mathcal{H}_n$ be the subspace of $\mathcal{H}$ generated by linear combinations of $e_s$ over $1/(n+1) < s\le 1/n$. We can define a $C^*$ sub-algebra $\mathcal{A}$ of the bounded linear operators $B(\mathcal{H})$ as follows: $A\in B(\mathcal{H})$ is in $\mathcal{A}$ if and only if the following are satisfied

  1. $A$ maps $\mathcal{H}_n$ into $\mathcal{H}_n$.
  2. There exists a sequence of complex numbers $a_n$ such that, for each fixed $n$, $$\Vert Ae_s-a_ne_s\Vert+\Vert A^*e_s-\bar a_ne_s\Vert\to0$$ as $s\to 1/n$.

It is easily seen that this is a sub-$C^*$-algebra. I'm thinking of $\mathcal{A}$ intuitively as being those elements of the direct product (sum?) $\prod_nB(\mathcal{H}_n)$ "joined continuously at the edges", and roughly as continuous functions $S\to\mathbb{C}$ but allowing some non-commutativity within the intervals $(1/(n+1),1/n]$.

If operator $A$ is in its center then it must restrict to the center of each $B(\mathcal{H}_n)$, so must be a constant on each of these. That is, $Ae_s=\lambda_se_s$ for some $\lambda_s\in\mathbb{C}$, which is independent of $s$ over $1/(n+1) < s\le 1/n$. Also, $s\mapsto\lambda_s$ must be continuous at $s=1/n$, from which we see that $s\mapsto\lambda_s$ is constant. So, $\mathcal{A}$ has trivial center.

Fixing a positive integer $m$, define an operator $A\in\mathcal{A}$ by $Ae_s=\min(ms,1)e_s$. Then, $\Vert A\Vert = 1$ and ${\rm dist}(A,{\rm Center}(\mathcal{A}))=1/2$. On the other hand, the restriction of $A$ to each $\mathcal{H}_n$ is within a distance $1/(2(m+1))$ from its center. This gives $\Vert AB - BA\Vert \le \Vert B\Vert / (m+1)$ for all $B\in\mathcal{A}$.

Choosing $m$ as large as we like, this contradicts your claim.

share|cite|improve this answer

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.