This is a question on how to relate two different distances in the matrix setting. Everywhere below, $M_n$ denotes the square matrices $n\times n$ whose entries are in $\mathbb C$. We consider the operator norm $\|\cdot\|$ on $M_n$.

By a subalgebra we mean a $C^*$-subalgebra of $M_n$. If $A\subseteq M_n$ is a subalgebra and $x\in M_n$, define the distance from $x$ to $A$ as the real number $d(x,A)=\inf_{a \in A}\|x-a\|$.

The question is the following: suppose that $A$ is a unital $C^*$-subalgebra of $M_n$ and that $x\in M_n$ is a contraction with $d(x,A)\geq\frac{1}{4}$. Does there exist $u\in A'$ (the commutant of $A$) such that $\|[x,uxu^*]\|\geq\frac{1}{16}$ (or, for what matter, $\frac{1}{64}$, the only important thing is that this number doesn't depend on the choice of $n$, $A$, or $x$)?

Beware, I am not asking whether I can find a $u$ with $\|[u,x]\|\geq\frac{1}{16}$. This is clearly possible since $y=\int_{\mathcal U(A')}uxu^*d\mu(u)$, when I am integrating over the Haar measure on the unitary group of $A'$, is the conditional expectation onto $A''=A$. Since $y\in A$, we have $\|x-y\|\geq\frac{1}{4}$, and thefore there must be a unitary $u\in \mathcal U(A')$ such that $\|x-uxu^*\|\geq\frac{1}{16}$, or in other words, such that $\|[x,u]\|\geq\frac{1}{16}$.

Thanks and best,


Take $A=\mathbb C\oplus\mathbb C\subset M_2(\mathbb C)$, and $x=\begin{bmatrix} 0&1/4\\0&0\end{bmatrix}.$

Then $A'=A$. If $u\in A'=A$, then $u=\begin{bmatrix} \lambda&0\\0&\mu\end{bmatrix}$, with $\lambda,\mu\in\mathbb T$. Then $$ uxu^*=\begin{bmatrix} 0&\lambda\overline\mu/4\\ 0&0\end{bmatrix}, $$ and $xuxu^*=uxu^*x=0$. In particular, $[x,uxu^*]=0$.

  • $\begingroup$ Right, make sense, thanks! To follow up though, what if $x$ is positive, or even just self-adjoint? $\endgroup$ – Alessandro Vignati Oct 16 '18 at 3:40
  • $\begingroup$ If you take the same $A$ as above and $x=\begin{bmatrix} 1&1/4\\1/4&1\end{bmatrix}$, then $x\geq0$ and $xuxu^*=uxu^*x$ for all unitary $u\in A$. $\endgroup$ – Martin Argerami Oct 16 '18 at 4:31
  • $\begingroup$ No..In particular in the top left corner of $uxu^*x$ you'll get $1+\frac{\lambda\overline\mu}{4}$ but in the top left corner you'll get it's adjoint, and unless $\lambda=\mu$ they're different. $\endgroup$ – Alessandro Vignati Oct 16 '18 at 12:19
  • $\begingroup$ Yes, you are right. $\endgroup$ – Martin Argerami Oct 16 '18 at 14:08

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.