6
$\begingroup$

Given a Hilbert space $H$, let $S_1(H)$ denote the space of trace-class operators on $H$, with the trace-class norm or Schatten 1-norm. That is $$ \Vert T \Vert_1 = \sum_{j\geq 1} |s_j| $$ where $(s_1,s_2,\dots)$ is the sequence of eigenvalues of the operator $|T|=(T^*T)^{1/2})$ written in any order.

Let $H_1$ and $H_2$ be Hilbert spaces of dimention $n$ and $m$ respectively. Let $H_1\otimes_2 H_2$ denote their Hilbert-space tensor product (hence, an inner product space of dimension $nm$). Consider the map $$\varphi:S_1(H_1)\hat{\otimes}S_1(H_2)\to S_1(H_1\otimes_2 H_2),$$ which is defined by $$\varphi( A \otimes B) (\xi_1\otimes\xi_2) = (A\xi_1 )\otimes (B\xi_2) \quad(\xi_1\in H_1, \xi_2\in H_2) $$

Is it the case that $$\Vert\varphi^{-1}\Vert=\min\lbrace{m‎, ‎n}\rbrace ?$$ Here, $\hat{\otimes}$ denotes the projective tensor product of Banach spaces.

$\endgroup$
9
  • 4
    $\begingroup$ And the question would be "How can you prove that the norm of $\phi^{-1}$ is the minimum of $m$ and $n$?"? This is not very clear from the way you phrased it. $\endgroup$ Jul 16, 2015 at 6:39
  • 4
    $\begingroup$ Please choose a title which describes your question in particular. $\endgroup$
    – Noah Stein
    Jul 16, 2015 at 12:06
  • 2
    $\begingroup$ When you say "projective tensor product norm " - is this the Banach space version or the operator space version? $\endgroup$
    – Yemon Choi
    Jul 16, 2015 at 14:32
  • 2
    $\begingroup$ And what is the norm on $H_1\hat{\otimes} H_2$? Hilbertian? or the projective tensor product of Banach spaces? I think the question in its current form needs to be clarified $\endgroup$
    – Yemon Choi
    Jul 16, 2015 at 14:36
  • 3
    $\begingroup$ Thanks for your precious time. Indeed,I am trying to find the proof of PROPOSITION 2.4 of johnson's paper(NON-AMENABILITY OF THE FOURIER ALGEBRA OF A COMPACT GROUP). (gen.lib.rus.ec/scimag/get.php?doi=10.1112%2Fjlms%2F50.2.361) @MikaeldelaSalle $\endgroup$
    – R.N
    Jul 16, 2015 at 20:44

1 Answer 1

4
$\begingroup$

Revised answer:

(Notation from p 61ff of the reference below). Note that for trace class norm we have $\mathcal{B}(H_1) = H_1 \hat\otimes H_1' = H_1\hat \otimes H_1$. For operator norm we have $\mathcal{B}(H_1) = H_1\hat{\hat\otimes} H_1'= H_1\hat{\hat\otimes} H_1$ where the tensor product is now the inductive one (also called $\epsilon$-tensor product). Using this and finite dimensionality we then can write
$\varphi:\mathcal{B}(H_1)\hat{\otimes}\mathcal{B}(H_2)\to\mathcal{B}(H_1\hat{\otimes}H_2)$ as the natural isomorphism $$ (H_1 \hat\otimes H_1)\, \hat\otimes\, (H_2 \hat\otimes H_2) \to (H_1 \hat\otimes H_2)\hat\otimes (H_1 \hat\otimes H_2)' = (H_1 \hat\otimes H_2)\hat\otimes (H_1 \hat{\hat\otimes} H_2) $$ which is $Id\,\hat\otimes\,j$ where $j:H_1 \hat\otimes H_2\to H_1 \hat{\hat\otimes} H_2$ is the embedding (here iso) of trace class norm into operator norm. The norm of $j$ is indeed $min\{\dim(H_1,\dim H_2)$, and the norm of $j^{-1}$ is its inverse, which can be seen by fixing bases and then writing any operator $A:H_1\to H_2$ as $U_2\circ D\circ U_1$ for isometries $U_i$ and a diagonal operator $D$.

Ah,

I see now that the question changed. Here is the version for the new question: $$ (H_1 \hat\otimes H_1)\, \hat\otimes\, (H_2 \hat\otimes H_2) = (H_1 \hat\otimes H_2)\, \hat\otimes\, (H_1 \hat\otimes H_2) \to (H_1 \otimes_2 H_2)\hat\otimes (H_1 \otimes_2 H_2)' = (H_1 \otimes_2 H_2)\hat\otimes (H_1 \otimes_2 H_2) $$ which is $i\hat\otimes i$ where $i: H_1 \hat\otimes H_2 \to H_1 \otimes_2 H_2$ is the natural iso. Its norm is the norm of the embedding $\ell^1\to \ell^2$ for the s-numbers.

  • Johann Cigler, Viktor Losert, Peter W. Michor: Banach modules and functors on categories of Banach spaces. Lecture Notes in Pure and Applied Mathematics 46, Marcel Dekker Inc., New York, Basel, (1979) (pdf)
$\endgroup$
3
  • 2
    $\begingroup$ I think your argument with reasonable tensor norms merely shows that $\varphi$ has norm $\leq 1$, not that $\varphi^{-1}$ has norm $\leq 1$. $\endgroup$
    – Yemon Choi
    Jul 16, 2015 at 14:37
  • $\begingroup$ You are right; my mistake. $\endgroup$ Jul 16, 2015 at 18:27
  • 3
    $\begingroup$ Just for sake of interest: it turns out that if one defines the projective tensor product in the category of operator spaces and completely bounded maps, then we have $T(H_1)\hat{\otimes}_{\rm op} T(H_2) = T(H_1\otimes_2 H_2)$, where $T$ denotes trace-class (i.e. Schatten-1) $\endgroup$
    – Yemon Choi
    Jul 16, 2015 at 22:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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