7

It is well known that if $K$ is a compact operator on a Hilbert space, and $T_n$ is a sequence of operators converging strongly to $T$, then $K T_n$ converges in norm to $K T$.

Question : are there other operators $K$ that also satisfy this property, but are not compact ?

Thanks

  • 2
    On a separable Hilbert space, $K$ is necessarily compact: given an ONB $(x_n)n$, define $T_n x =\sum{\ell=1}^n x_\ell \langle x,x_\ell \rangle$. Then $T_n$ converges strongly to the identity. By assumption, $K T_n \to K$ in norm topology. But $T K_n$ has finite rank (why?). As a limit of finite rank operators in the norm topology, $K$ must be constant. I am not sure about non-separable Hilbert spaces, though. – PhoemueX Feb 04 '20 at 19:45
  • Oh...right thanks. – Marc_Adrien Feb 05 '20 at 20:09

1 Answers1

5

The statement made by the OP that:

It is well known that if $K$ is a compact operator on a Hilbert space, and $T_n$ is a sequence of operators converging strongly to $T$, then $K T_n$ converges in norm to $K T$.

is being questioned here. Nevertheless I have no qualms with the corresponding statement in which $K T_n$ and $KT$ are replaced by $T_nK$ and $TK$, so I will read the present question with that modification.

My goal it thus to give a negative answer, namely to prove that if an operator $K$ has this property, then it is necessarily compact.

$\newcommand{\one}{{\mathbb 1}}$ Lemma. Let $T$ be a bounded operator on a Hilbert space $H$ which is not compact. Then there exists an infinite dimensional subspace $K\subseteq H$, and a constant $c>0$, such that $$ \|T\xi \|\geq c\|\xi \|, \quad \forall \xi \in K. \tag 1 $$

Proof. Let us assume first that $T$ is positive semi-definite. For each $n\in \mathbb N$, let $\one_n$ be the characteristic function of the interval $[1/n, +\infty )$, and observe that $$ x\one_n(x) \ {\buildrel {n\to\infty}\over\longrightarrow}\ x, $$ uniformly on $[0,+\infty )$, so it follows that $$ T\one_n(T) \to T, \tag 2 $$ in norm. Notice that the $\one_n(T)$ form an increasing family of self-adjoint projections and we claim that they cannot all have finite rank. This is because otherwise $T\one_n(T)$ would also be finite rank, and then $T$ would be compact by (2).

Choosing any $n$ such that $\one_n(T)$ has infinite rank, let $K$ be the range of $\one_n(T)$, hence an infinite dimensional subspace. The proof will then be concluded once we show that $$ \|T\xi \|\geq \frac1n \|\xi \|, \quad \forall \xi \in K. \tag 3 $$ To prove this notice that $$ x\one_n(x) \geq \frac{\one_n(x)}n, \quad \forall x \in [0,+\infty ), $$ so we have that $$ nT\one_n(T) \geq \one_n(T). $$ For all $\xi $ in $K$, a.k.a. the range of $\one_n(T)$, we then have that $$ \|\xi \|^2 = \langle \xi , \xi \rangle = \langle \one_n(T)\xi , \xi \rangle \leq n\langle T\one_n(T)\xi ,\xi \rangle = n\langle T\xi ,\xi \rangle \leq n\|T\xi \|\|\xi \|, $$ from where (3) follows.

Returning to the general case, in which $T$ is only assumed to be a bounded operator, let $|T|=(T^*T)^{1/2}$, which is positive semi-definite. Observing that $$ \|T(\xi)\| = \|\, |T|(\xi) \, \|,\quad \forall \xi \in H, $$ we see that any pair $(K,c)$ that works for $|T|$, also works for $T$. QED

Theorem. Let $T$ be an operator on a Hilbert space $H$ such that, whenever $\{S_n\}_n$ is a sequence of bounded operators, strongly converging to zero, one has that $S_nT\to0$ in norm. Then $T$ is compact.

Proof. Assuming by contradiction that $T$ is not compact, use the Lemma to find an infinite dimensional subspace $K\subseteq H$, and a constant $c>0$, such that (1) holds.

Setting $L=T(K)$, is is easy to see that $T$ provides a bounded, linear, invertible map $$ \check T:K\to L, $$ with bounded inverse and, in particular, that $L$ is a closed infinite dimensional subspace of $H$.

One may then easily find a sequence of operators $\{S_n\}_n\subseteq B(L)$ which converges to zero strongly, but not in norm. Extending each $S_n$ to $H$ by setting it to be zero on $L^\perp$, it is clear that the extended operators $\hat S_n$ also satisfy $\hat S_n\to0$ strongly, but not in norm.

By the assumption on $T$ we have that $\|\hat S_nT\|\to 0$, so $$ \|S_n\| = \|S_n\check T\check T^{-1}\| \leq \|S_n\check T\| \|\check T^{-1}\| = $$$$ = \|\hat S_nT|_K\| \|\check T^{-1}\| \leq \|\hat S_nT\| \|\check T^{-1}\| \to 0 $$ contradicting the choice of $S_n$. QED

Ruy
  • 20,073
  • Very very nice! +1 – J. De Ro Dec 05 '20 at 19:56
  • Thanks! Would you like to try a shot at https://math.stackexchange.com/questions/3935048/does-left-multiplication-by-compact-operators-turn-strong-convergence-into-norm ? – Ruy Dec 05 '20 at 20:05
  • I see you have an answer deleted. Why is that? (Users with 10k+ rep can see deleted answers). – J. De Ro Dec 05 '20 at 20:08
  • I didn't know people could find out when there are deleted answers, and I'd be curious to learn how. Anyways, the reason is partly explained in PS(1) there. – Ruy Dec 05 '20 at 20:11
  • I edited my above comment, but maybe you missed it. But I spoiled myself by seeing your answer. I think it is better to edit the answer and place it in a spoiler box. A shift operator gives surprisingly often good counterexamples! – J. De Ro Dec 05 '20 at 20:13
  • Holy super users, Batman! OK, I did as you said. But I'm not going to undelete it yet since I don't want to dicourage answers. Anyway, weren't you surprised by it? @MathQED – Ruy Dec 05 '20 at 20:30
  • Disclaimer: I'm a functional analysis newbie. For me it is obvious that if $T_i \to T$ strongly, then $T_iK \to TK$ in norm for the simple reason that we can approximate $K$ by finite rank operators which reduces the proof to showing that $T_i F \to TF$ in norm when $T_i F \to TF$ strongly and $F$ is a finite rank operator. The latter is easily seen, for example by taking an orthonormal basis for the image of the finite rank operator. But frankly, I don't see why any reason you would expect that $KT_i \to KT$ strongly implies $K T_i \to KT$ in norm? – J. De Ro Dec 05 '20 at 20:42
  • You are totally correct. The statement that $KT_i\to KT$ is just a common misconception, as you can see in the statement of this question. Another indication that this is widespread is the comment by MaoWao, in which the questioning falls squarely outside the point by emphasizing the difference between nets and sequences. It surprised me because I am (or at least am supposed to be) a professional functional analyst and I seem to have been under the same misconception until recently! @MathQED – Ruy Dec 05 '20 at 20:51
  • This discussion reminds me of the highly interesting MO thread: https://mathoverflow.net/questions/23478/examples-of-common-false-beliefs-in-mathematics – J. De Ro Dec 05 '20 at 21:09