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 me first recall the Stone-von Neumann theorem that if two one-parameter groups of unitary operators $U_t$ and $V_s$ over a Hilbert space satisfy $U_tV_s=e^{ist}V_sU_t$ for every $s,t\in{\mathbb R}$ (Weyl relations), then their generators $P$ and $Q$ satisfy the canonical commutation relation $[Q,P]\psi=i\psi$ for all $\psi$ in the common dense domain. See also this related question.

I am interested in discrete one-parameter groups $U^p$ and $V^q$ where $p,q\in{\mathbb Z}$, and $U$ and $V$ are unitary matrices. There are simple examples of pairs $U,V\in{\mathbb U}_n$ that $\omega$-commute, which means that $UV=\omega VU$. In this case, $\omega$ is some root of unity, say $\omega\ne1$, and we have the Weyl relation $U^jV^k=\omega^{jk}V^kU^j$.

Is there any such pair with the property that $\|I_n-U\|<1$ and $\|I_n-V\|<1$, where $\|\cdot\|$ is the operator norm?

My gess is No : if such a pair existed then we could define the logarithms of $U$ and $V$ by the converging series $\log(1-x)=-x-\frac{x}2-\cdots$. It seems to me that we should obtain two matrices $X$ and $Y$ satisfying $[X,Y]=\alpha I_n$ where $e^\alpha=\omega$. But a finite dimensional commutator has zero trace, thus $\alpha=0$ and $\omega=1$.

Edit. Just to let you know that $\omega$-commuting matrices are not abstract non-sense, here is a nice relation when $\omega^p=1$ : if $(A,B)$ $\omega$-commute, then $(A+B)^p=A^p+B^p$. I could be due to H. S. A. Potter (does anyone knows if H is for Harry?).

I replaced the spectral radius by the operator norm because it is equal for normal matrices, such as $I_n-U$.

re-Edit. I realize that there is a trivial answer to my question: $U$ is unitary equivalent to $\omega U$, thus its spectrum is a union of regular $m$-agons over the unit circle, where $m$ is the order or the root of unity $\omega$. Then there must be an eigenvalue with non-positive real part, which implies $\|I_n-U\|\ge\sqrt2$.

share|cite|improve this question
I'm a bit confused about finite dimensions. If you want $U$ and $V$ in finite dim, then taking the determinant (which is the elder brother of the trace) of the relation $UV = \omega VU$ gives $\omega = 1$??? What have I got wrong here? – Stefan Waldmann Oct 19 '11 at 15:48
Yes you got wrong. What you obtain is $\omega^n=1$, because $\det(aM)=a^n\det M$. – Denis Serre Oct 19 '11 at 16:03
Oh bulls* I'm sorry. It seems to be already late... Just ignore what I wrote :) – Stefan Waldmann Oct 19 '11 at 16:55
Since the relation you mention is quite magical it could be well due to H. Potter... – François Brunault Oct 20 '11 at 8:47
It seems $H$ stands for Harold :… – François Brunault Oct 20 '11 at 8:52
up vote 4 down vote accepted

I did not try to follow your argument, but here is another proof of your claim. In fact it proves a slightly stronger result, i.e. that if $\rho(1-U)$ and $\rho(1-V)$ are strictly smaller that $\sqrt 2$, $U,V$ cannot $\omega$-commute if $\omega \neq 1$.

First note that for a couple of unitaries $(U,V)$ that $\omega$-commute, all the couples $(V,U^*)$, $(V^*,U)$ and $(U^*,V^*)$ also $\omega$-commute. In particular, by linearity the trace of $(U+U^*)(V+V^*)$ is zero if $\omega \neq 0$.

But the condition $\rho(1-U)<\sqrt 2$ exactly means that $U+U^*>0$ (as usual $>0$ means that the matrix is positive definite). Therefore if $\rho(1-U)<\sqrt 2$ and $\rho(1-V)<\sqrt 2$, the trace of $(U+U^*)(V+V^*)$ is positive.

This proves the claim.

share|cite|improve this answer
I have a doubt. I find that two of these pairs are $\omega$-commuting and the two others $\bar\omega$-commute. We cannot say anything about the pair $(U+U^*,V+V^*)$. – Denis Serre Oct 19 '11 at 17:40
Maybe you are right (but I do not think so, I tried to be precise in the way I order the couples), but anyway my point is that $Tr(U^{\varepsilon_1} V^{\varepsilon_2})= 0$ whenever $\varepsilon_1,\varepsilon_2 \in \{1,*\}$, and this holds because the couple $(U^{\varepsilon_1},V^{\varepsilon_2})$ something-commute, with something different from $1$. By linearity, this implies that $Tr( (U+U^*)(V+V^*) )=0$. – Mikael de la Salle Oct 19 '11 at 21:18
To expand my comment: $(U,V)$ $\omega$-commute if and only if $(V,U)$ $\bar \omega$-commute. – Mikael de la Salle Oct 19 '11 at 22:08

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.