3
$\begingroup$

Let $A$ and $B$ be two self-adjoint, positive definite Compact operators on a Hilbert space $\mathcal{H}$. Further, let $A$ be trace class. Define $C_n \equiv AB(\frac{I}{n} + BAB)^{-1}$. Does $\frac{tr(C_nC_n^{T})}{n}$ converge to $0$ as $n \to \infty$. The answer is affirmative when $A$ and $B$ commute as can be found in section 7.1 of the following paper- https://arxiv.org/pdf/1103.2692.pdf

$\endgroup$
4
  • $\begingroup$ Maybe you just mean "positive"? A positive definite operator is not compact unless H is finite dimensional. $\endgroup$ May 31, 2016 at 19:10
  • $\begingroup$ That is incorrect. For example, any operator with eigenvalues $\rho_i >0$ with $\rho_i \to 0$ as $i \to \infty$ is compact and positive definite. $\endgroup$
    – Madhuresh
    May 31, 2016 at 19:42
  • $\begingroup$ I suggest that you give the definition of "positive definite" operator that you are adopting (which is clearly not this one: encyclopediaofmath.org/index.php/Positive-definite_operator ) $\endgroup$ May 31, 2016 at 20:55
  • $\begingroup$ @PietroMajer I must admit that my definition of (semi-)positive definite is the one Madhuresh has given $\endgroup$
    – Yemon Choi
    Jun 1, 2016 at 1:30

1 Answer 1

4
$\begingroup$

Yes, this is true. Denote $Q_n=\sqrt{n}\sqrt{A}B$. We have to prove that $c_n:=tr(\sqrt{A}Q_n(I+Q_n^*Q_n)^{-2}Q_n^*\sqrt{A})\to 0$. We have $c_n=tr(AQ_n(I+Q_n^*Q_n)^{-2}Q_n^*)$. We need a

Lemma. For any compact operator $Q$ we have $Q(I+Q^*Q)^{-2}Q^*\leqslant \frac14I$. Moreover, for any vector $x$ we have $\|Q_n(I+Q_n^*Q_n)^{-2}Q_n^* x\|\to 0$.

Proof. We may choose orthonormal bases $(v_i)$ and $(u_i)$ and positive numbers $\lambda_i$ so that $Qx=\sum \lambda_i (x,u_i)v_i$, $Q^*x=\sum \lambda_i (x,v_i)u_i$, then $(v_i)$ is eigenbasis for $Q(I+Q^*Q)^{-2}Q^*$ with eigenvalues $\lambda_i^2/(1+\lambda_i^2)^2\leqslant 1/4$. For $Q=Q_n$ bases $v_i,u_i$ are fixed, $\lambda_i=\sqrt{n}\mu_i$ for some fixed positive numbers $(\mu_i)$, thus each specific eigenvalue $n\mu_i^2/(1+n\mu_i^2)^2$ tends to 0. Representing $x$ as a sum of small vector and a finite linear combination of $v_i$ we get the second claim.

Now let $e_i$ be an eigenbasis for $A$, $Ae_i=t_ie_i$, $\sum t_i<\infty$. We have $$c_n=\sum t_i (Q_n(I+Q_n^*Q_n)^{-2}Q_n^*e_i,e_i),$$ where $i$-th term always does not exceed $t_i$ and each specific term tends to 0. So we may simply for given $\varepsilon>0$ choose $N$ so that $\sum_{i>N} t_i<\varepsilon$, estimate the tail of (1) for $i>N$ as $\varepsilon/4$ and say that first $N$ summands are less than $\varepsilon/4$ provided that $n$ is large enough. Totally $|c_n|<\varepsilon/2$ for large $n$, as desired.

$\endgroup$
2
  • 3
    $\begingroup$ Though not needed in the answer to original problem, I believe that the Lemma holds even for non-compact operator $Q$. More specifically, let $R=(I+Q^{*}Q)^{-1}Q^{*}$. Then by the functional calculus for positive operators, $RR^{*} \leq \frac{1}{4}I$, which implies that $Q(I+QQ^{*})^{-2}Q^{*}=R^{*}R\leq\frac{1}{4}I$ as well. Furthermore, if $Q_n = \sqrt{n}Q$ for all $n$ and $R_n = (I+Q_n^{*}Q_n)^{-1}Q_n^{*}$ then for any vector $x$, we also have $\|R_nx\|\to 0$. $\endgroup$
    – T. Le
    Jun 1, 2016 at 0:16
  • $\begingroup$ Could you elaborate on your comment please? Perhaps an answer rather than a comment. I am new to Operator theory. It shall be very helpful. $\endgroup$
    – Madhuresh
    Jun 6, 2016 at 4:32

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.