Consider the Hilbert Schmidt operator $K: L^2(\Omega) \rightarrow L^2(\Omega)$, $\Omega \subset \subset \mathbb R^N$, with $k \in L^2(\Omega \times \Omega)$ and $f \in L^2(\Omega)$,

$$(Kf)(x) := \int_\Omega k(x,y)f(y)\, dy.$$

I want to show that the Hilbert Schmidt operator $K$ is a compact operator. Therefore I'm using this characterization.

Let $X$, $Y$ be normed linear spaces and $X$ reflexive. A continuous linear operator $T: X \rightarrow Y$ that maps weakly convergent sequences onto strongly convergent sequences is compact.

(We already know that $K$ is well-defined as is proven here.)

My question here is, isn't it obvious that $K$ is compact?

  • We know that $K$ is linear and bounded, hence continuous.
  • Every continuous map takes weakly convergent sequences to weakly convergent sequences.
  • The norm itself is also continuous.
  • Weak convergence together with convergence of the norms implies convergence.

Thus $K$ is compact. Am I missing something here? Or better: What am I missing here?


I'm also adding the proof from the textbook for completeness:

Proof. Let $(f_n)_{n \in \mathbb N} \subset L^2(\Omega)$ a weakly convergent sequence, then $(f_n)_{n \in \mathbb N}$ is bounded. That is, $\exists C > 0 $ such that $||f_n||_{L^2(\Omega)} \leq C$, $\forall n \in \mathbb N$. By Fubini's theorem we have for almost every $x\in \Omega$ that $$ || k(x,\cdot) ||_{L^2(\Omega)} = \int_\Omega |k(x,y)|^2 \, dy < \infty .$$

Thus for almost every $x \in \Omega$ we have

$\begin{align} \lim_{n \rightarrow \infty} (Kf_n)(x) & = \int_\Omega k(x,y)f_n(y) \, dy = \lim_{n \rightarrow \infty} \langle k(x,\cdot), f_n \rangle_{L^2(\Omega)} \\ & = \langle k(x,\cdot), f \rangle_{L^2(\Omega)} = \int_\Omega k(x,y)f(y) \, dy = (Kf)(x) \end{align}$

By Cauchy-Schwarz's inequality we have $$ (Kf_n)(x) \leq ||f_n||_{L^2(\Omega)} \int_\Omega |k(x,y)|^2 \, dy \leq C \, \int_\Omega |k(x,y)|^2 \, dy $$

Hence by Lebesgue's dominant convergence theorem we have convergence of the norms $$ \lim_{n \rightarrow \infty} \int_\Omega |(Kf_n)(x)| \, dx = \lim_{n \rightarrow \infty} \int_\Omega |(Kf)(x)| \, dx ,$$ that is $|| Kf_n ||_{L^2(\Omega)} \rightarrow || Kf ||_{L^2(\Omega)}\, \, (n\rightarrow \infty)$. Since weak convergence together with (strong or normal) convergence of the norms implies (strong) convergence, $K$ is compact.

  • $\begingroup$ How would your argument not apply to a general bounded operator? Not all bounded operators are compact. $\endgroup$ – DisintegratingByParts Jan 22 '17 at 18:15
  • 1
    $\begingroup$ agreed. but where is the error in the argument? $\endgroup$ – cesare borgia Jan 22 '17 at 19:57
  • $\begingroup$ it would be good to add a citation for the textbook where you found the proof you wrote $\endgroup$ – qbert Jun 26 '18 at 22:28

The norm is continuous as a map $\|\cdot\|: (X,\|\cdot\|_X)\to \mathbb R$ but not when defined on $X$ with its weak topology. This is where your general argumentation fails. The proof from your textbook is fine, however one can in general show that every Hilbert-Schmidt operator is already compact: One can represent the finite rank operators in a Hilbert space as a tensor product. We have several natural norms on this space, whose completions lead to several classes of operators (nuclear operators, Hilbert-Schmidt operators and compact operators) and those norms dominate each other in such a way that we have the inclusions nuclear operator is Hilbert-Schmidt operator is compact operator.

  • 1
    $\begingroup$ what exactly do you mean by "... but not when defined on $X$ with its weak topology"? the weak topology is not metrizable in general, hence exists no correponding norm or am I mistaken? $\endgroup$ – cesare borgia Jan 23 '17 at 23:56
  • 2
    $\begingroup$ In general not, but the weak topology is, as the name suggests, a topology. Consider $l^2(\mathbb N)$. Then $\delta_n$ converges weakly to zero, but $\|\delta_n\|=1$, hence the norm is not continuous wrt the weak topology. $\endgroup$ – Sebastian Bechtel Jan 24 '17 at 7:03

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.