Let $H$ be a separable infinite-dimensional real Hilbert space. We consider operators in $H.$

Nuclear norm of a nuclear operator is the sum of its singular values. A nuclear, positive and self-adjoint operator is called S-operator.

Does the following criterion hold true?

A sequence $A_n$ of S-operators converges in nuclear norm to S-operator $A$ if, and only if, $A_n$ weakly converges to $A$ and there exists an orthobasis $e_k$ such that the series $\Sigma_{k=1}^{\infty}(A_n e_k, e_k)$ converges uniformly in $n=1, 2\ldots$

I can prove the necessity only. But does the sufficiency hold true as well?

  • 1
    $\begingroup$ Does "weak convergence" mean "convergence with respect to the weak operator topology"? $\endgroup$ – Jochen Glueck Jun 10 '18 at 17:43
  • 1
    $\begingroup$ Yes, it means that $(A_n x, y)$ converges to $(A x, y),$ for all $x, y \in $H$. $\endgroup$ – Alexander Kukush Jun 10 '18 at 18:57

Looks correct. Split $H=H_n\oplus H^n$ where $H_n=span(e_1,\dots,e_n)$. Let $P_n$ and $P^n$ be the corresponding projections. The conditions are $P_nA_kP_n\to P_nAP_n$, $\operatorname{Tr}(P^nA_kP^n)\le\varepsilon_n$ with $\varepsilon_n\to 0$ as $n\to\infty$.

Now just observe that $A_k-A+\delta n^{-1} P_n+\delta(P_nA_kP_n+P_nAP_n)+(1+\delta^{-1})(P^nA_kP^n+P^nAP^n)$ is an $S$-operator for large $k$ and fixed $n,\delta$. However, its trace is $\le |\operatorname{Tr}P_n(A_k-A)P_n|+\delta(1+2\varepsilon_1)+(3+\delta^{-1})\varepsilon_n$, which can be made arbitrarily small by choosing $\delta, n, k$ in this order. Also, it is an $S$-perturbation of $A_k-A$ of small trace, so we are done.

| cite | improve this answer | |
  • $\begingroup$ Alas, in the proposed answer I do not see why this long operator is positive for large k and fixed n, $\delta.$ And I cannot understand the origin of the term $\delta n^{-1} P_n$. $\endgroup$ – Alexander Kukush Jun 10 '18 at 20:23
  • $\begingroup$ @AlexanderKukush $\delta(P_nA_kP_n+P_nAP_n)+\delta^{-1}(P^nA_kP^n+P^nAP^n)$ dominates the "off-diagonal" part of $A_k-A$; $\delta n^{-1}P_n$ is just to dominate $-P_n(A_k-A)P_n$. Is that much clear? $\endgroup$ – fedja Jun 10 '18 at 20:34
  • $\begingroup$ @fedija Now I understand your solution thoroughly. Brilliant! You helped me a lot. As a result I have a criterion for weak convergence of Gaussian measures in H: Measures $N(m_n, S_n)$ converge weakly to $N(m, S)$ if, and only if, $m_n$ converges strongly to $m$ and $S_n$ converges to $S$ in nuclear norm. $\endgroup$ – Alexander Kukush Jun 11 '18 at 7:07

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.