Let $(\lambda_{mn})_{(m,\:n)\in\mathbb N^2}\subseteq[0,\infty)$ with $$\lambda_{mn}^2\le\mu_m\mu_n\;\;\;\text{for all }(m,n)\in\mathbb N^2\tag1$$ for some $(\mu_n)_{n\in\mathbb N}\subseteq[0,\infty)$ such that $\sum_{n\in\mathbb N}\mu_n$ exists in $\mathbb R$. Let $(e_n)_{n\in\mathbb N}$ be an orthonormal basis of a separable $\mathbb R$-Hilbert space $H$ and $$e_m\otimes e_n:=\langle\;\cdot\;,e_m\rangle_He_n\;\;\;\text{for }(m,n)\in\mathbb N^2\;.$$ Let $\mathfrak L_1(H)$ denote the space of nuclear operators on $H$.

Can we show that $$A:=\sum_{(m,\:n)\in\mathbb N^2}\lambda_{mn}e_m\otimes e_n$$ is a bounded linear self-adjoint operator on $H$?

I've tried the following: Let $\operatorname{HS}(H)$ denote the space of Hilbert-Schmidt operators on $H$. Note that $(e_m\otimes e_n)_{(m,\:n)\in\mathbb N^2}$ is an orthonormal basis of $\operatorname{HS}(H)$ and hence $A$ exists in $\operatorname{HS}(H)$ if and only if $\sum_{(m,\:n)\in\mathbb N^2}\left\|\lambda_{mn}e_m\otimes e_n\right\|_{\operatorname{HS}(H)}^2$ exists in $\mathbb R$. Now, \begin{equation}\begin{split}\sum_{(m,\:n)\in\mathbb N^2}\left\|\lambda_{mn}e_m\otimes e_n\right\|_{\operatorname{HS}(H)}^2&=\sum_{(m,\:n)\in\mathbb N^2}\lambda_{mn}^2\underbrace{\left\|e_m\otimes e_n\right\|_{\operatorname{HS}(H)}^2}_{=\:1}\\&\le\sum_{(m,\:n)\in\mathbb N^2}\mu_m\mu_n\\&=\left(\sum_{m\in\mathbb N}\mu_m\right)\sum_{n\in\mathbb N}\mu_n\end{split}\tag2\end{equation} and hence $A$ exists in $\operatorname{HS}(H)$. However, I've no idea how I can show the self-adjointness.

  • $\begingroup$ Start with looking at $\langle A e_r, e_s\rangle$ and $\langle e_r, Ae_s\rangle$. $\endgroup$ – Daniel Fischer May 28 '17 at 19:37
  • $\begingroup$ @DanielFischer I've started with that. And, clearly, $e_m\otimes e_n$ itself is not self-adjoint. $\endgroup$ – 0xbadf00d May 28 '17 at 19:39
  • $\begingroup$ Can you find a necessary condition on the family $\{ \lambda_{mn}\}$ for self-adjointness? $\endgroup$ – Daniel Fischer May 28 '17 at 19:41
  • $\begingroup$ @DanielFischer I know you cannot see how this question is related, but do you know if the "square-root" (which is uniquely determined for any nonnegative and self-adjoint bounded linear operator) exists even when the operator in question isn't self-adjoint? (I didn't find a proof and I've noticed that some authors call an operator $A$ nonnegative if it is self-adjoint and $\langle Ax,x\rangle_H\ge0$ for all $x\in H$, while others (including me) call it nonnegative if only the second condition is satisfied) $\endgroup$ – 0xbadf00d May 28 '17 at 20:01
  • $\begingroup$ So you're asking whether for all non-negative $A$, there is a (non-negative, presumably) $B$ such that $A = B^2$? Off the top of my head, I don't know. (If we only were dealing with complex vector spaces.) $\endgroup$ – Daniel Fischer May 28 '17 at 20:07

You already know that $A$ is bounded, since it is Hilbert-Schmidt.

Knowing that $A$ is bounded, the following statements are equivalent:

  • $A$ is selfadjoint

  • $\lambda_{mn}=\lambda_{nm}$ for all $n,m$.

So, in general, your $A$ will not be selfadjoint.

The relevant computation is $$ \langle Ae_n,e_m\rangle =\lambda_{mn}, \ \ \ \ \langle A^*e_n,e_m\rangle=\lambda_{nm}. $$ together with the fact that it is enough to test the selfadjoint condition on elements of a fixed orthonormal basis.

| cite | improve this answer | |
  • $\begingroup$ In fact, "my" $\lambda_n$'s satisfy your condition. Could you provide a proof for the equivalence? $\endgroup$ – 0xbadf00d May 28 '17 at 19:42
  • $\begingroup$ Yes, I added a bit to the answer. $\endgroup$ – Martin Argerami May 28 '17 at 19:43
  • $\begingroup$ And let me note the following: If you're right and $A$ is self-adjoint, then your deleted answer to my other question was correct! The self-adjointness ensures the existence of a unique square-root and this allows us to show that a bounded, linear, nonnegative and self-adjoint operator is nuclear iff it has finite trace. $\endgroup$ – 0xbadf00d May 28 '17 at 19:43
  • $\begingroup$ I think there is a typo: It should be $\langle Ae_n,e_m\rangle =\lambda_{mn}$. $\endgroup$ – 0xbadf00d May 28 '17 at 19:51
  • $\begingroup$ I don't follow. In what sense would a self-adjoint operator have a "unique" root? $\endgroup$ – Martin Argerami May 29 '17 at 1:29

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.