MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

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 $H$ and $K$ be Hilbert spaces, and let $u$ be a partial isometry in $\mathcal{B}(H \otimes K)$ between projections $p_0 = u^\ast u$ and $p_1 = u u^\ast$ such that $p_0, p_1 \leq 1 \otimes (1-q)$ for some projection $q \in \mathcal B(K)$ equivalent to $1 \in \mathcal B (K)$. Does $p_1 u (a \otimes 1)p_0 = p_1 (a \otimes 1)u p_0$ for all $a \otimes 1 \in \mathcal{B}(H) \otimes 1$ imply that $u$ can be extended to a unitary operator in the commutant $(\mathcal B(H)\otimes 1)' = 1 \otimes \mathcal B(K)$?

I've verified the implication for several simple examples, but I'm having trouble proving it in full generality.

Update: I've reworded the question for clarity thanks to Yemon's comments. By a unitary operator extending $u$, I mean a unitary operator that agrees with $u$ on $p_0 (\mathcal H \otimes \mathcal K)$.

Update: Here's one example. Let $K$ be infinite dimensional with basis $e_1,e_2,\ldots$, and $H$ be $n$-dimensional with bases $\xi_1, \ldots, \xi_n$ and $\zeta_1, \ldots, \zeta_n$. Let $\xi = \frac{1}{\sqrt n}\sum_i \xi_i \otimes e_i$ and $\zeta = \frac{1}{\sqrt n}\sum_I \zeta_i \otimes e_i$. Define $u\colon\eta \mapsto \langle \eta, \xi \rangle \zeta$. Note that the assumptions in the first sentence are satisfied with $q$ the projection on the span of $\{e_{n+1}, e_{n+2}, \ldots \}$. The condition that $p_1 u (a \otimes 1)p_0 = p_1 (a \otimes 1)u p_0$ for all $a \otimes 1 \in \mathcal{B}(H) \otimes 1$ is here equivalent to $\langle u (a \otimes 1) \xi, \zeta\rangle = \langle (a \otimes 1) u \xi , \zeta \rangle $ for all $a \in \mathcal B(H)$, which is in turn equivalent to $\langle(a \otimes 1)\xi, \xi \rangle = \langle (a \otimes 1) \zeta, \zeta \rangle$ for all $a \in \mathcal B(H)$. The latter equation holds because both sides are equal to the trace of $a$. Thus, all conditions are satisfied.

To see that the conclusion holds, let $K_n$ be the span of $e_1,\ldots,e_n$, and let $c_{ij}$ be the unitary $n \times n$ matrix such that $\zeta_i = \sum_j c_{ij} \xi_j$. Then $$u \xi = \zeta = \frac{1}{\sqrt n}\sum_i \zeta_i \otimes e_i = \frac{1}{\sqrt n} \sum_{ij} c_{ij}\xi_j \otimes e_i = \frac{1}{\sqrt n}\sum_j \xi_j \otimes we_j = (1 \otimes w) \xi,$$ where $w\colon K_n \rightarrow K_n$ is the unitary defined by $we_j = \sum_i c_{ij}e_i$. We extend $w$ to all of $K$ by setting $\tilde{w} = w \oplus 1$. Then $1 \otimes \tilde w$ is a unitary operator in $1 \otimes B(K)$ that extends $u$.

I realized that the other examples I was thinking about are trivial.

share|cite|improve this question
I think I must be misunderstanding the terminology here. Your partial isometry $u$ is an operator on $H\otimes K$; and you want to extend it to a unitary on some large space? (I guess I'm confused by your reference to the commutant of $B(H)\otimes 1$, since the commutant of that algebra inside $B(H\otimes K)$ is surely going to be $1\otimes B(K)$...) – Yemon Choi Mar 30 '10 at 7:22
By the way, it would be nice if you updated your questions rather than editing them, since I have no way to edit comments unless I delete and rewrite. Also, it would be nice if you acknowledged that the update comes from the comments; you still haven't addressed the thing I was confused about... – Yemon Choi Mar 30 '10 at 7:32
While I'm still here: what simple examples did you try? – Yemon Choi Mar 30 '10 at 7:32
(Placeholder comment here, to say that it's getting late here in QC and so I'll have to wait till tomorrow to come back and think more.) – Yemon Choi Mar 30 '10 at 7:34
Steven, to be consistent with the notation of the question, let $w$ be the arbitrary isometry in $\mathcal B (K)$, and let $u = 1 \otimes w$. Then, $p_0 = 1 \otimes w^\ast w \not \leq 1 \otimes (1-q)$ for any projection $q$ equivalent to $1 \in B(K)$. – Andre Mar 30 '10 at 12:45
up vote 3 down vote accepted

If I get it right this time, the following Zorn argument proves the implication.

Denote by $q_0=\bigvee_{v\in \mathcal{U}(H)} (v\otimes 1)p_0(v^\ast\otimes 1)$ and similarly $q_1=\bigvee_{v\in \mathcal{U}(H)} (v\otimes 1)p_1(v^\ast\otimes 1)$. Apply Zorn's lemma to the set $I=\left\{w\in \mathcal{B}(H\otimes K)\left\vert \begin{array}{l}\,\,w\text{ is a partial isometry, }\\w^\ast w\leq q_0,\\ww^\ast\leq q_1,\\w\text{ extends u, and }\\\!\!\!ww^\ast(v\otimes 1)w=w(v\otimes 1)ww^\ast\text{ for all }v\in\mathcal{U}(H)\end{array}\right\}\right.$

ordered by extension: $w_1\leq w_2$ iff $w_2 w_1^\ast=w_1 w_1^\ast$ (in other words: $w_2$ has larger initial support and agrees with $w_1$ on the initial support of $w_1$)

A maximal element $w\in I$ must have $w^\ast w=q_0$, for otherwise there would have been a unitary $v\in\mathcal{B}(H)$ such that $f=(v^\ast\otimes 1)w^\ast w(v\otimes 1)\wedge (q_0-w^\ast w)\not=0$. Then $(v^\ast\otimes 1)w(v\otimes 1)f + w$ is a partial isometry in $I$ that extends $w$. Since $1-q_0$ and $1-q_1$ are equivalent projections in $1\otimes \mathcal{B}(K)$, we can extend $w$ to a unitary.

share|cite|improve this answer
Steven, is there a typo in that last paragraph? – Andre Mar 31 '10 at 14:37
There were indeed several typos in the last paragraph – Steven Deprez Mar 31 '10 at 16:24
Thank you very much for your solution. – Andre Mar 31 '10 at 16:50

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.