I know from the spectral theorem for compact self-adjoint operators that $K$ can be written as $$K = \sum_{n} \lambda_n P_n$$ where $\lambda_n$ are eigenvalues of $K$ and $P_n$ are orthogonal projections. By the spectral theorem, the image of each $P_n$ is finite dimensional, but there can be a countably infinite many of them.

Given that, I'm not sure how to proceed. My instinct is that I need to show that $K$ cannot have infinitely many eigenvalues due to the condition that its image is closed.


Let $M$ be the kernel of $K$. Then $S:H/M \to K(H)$ defined by $S(x+M)=Kx$ is a well defined bounded operator which is also bijective. By Open Mapping Theorem its inverse is also continuous. Since $K$ is compact this implies that the closed unit ball of $K(x)$ is compact and hence $K(X)$ is finite dimensional.

Some details for the last part: suppose $(y_n)$ is a bounded sequence in $K(X)$. Let $x_n=S^{-1}y_n$. Then $\|x_n\| \leq \|S^{-1}\| \|y_n\|$ so $(x_n)$ is bounded. By the definition of the norm in $H/M$ we can find a sequence $(z_n)$ in $M$ such that $(x_n+z_n)$ is bounded. Since $K$ is compact $K(x_n+z_n)$ has a convergent subsequence. But $Kz_n=0$ so $(K(x_n))$ has a convergent subsequence. By definition of $x_n$ we have $y_n=S(x_n)=Kx_n$. Hence $(y_n)$ has a convergent subsequence. Thus every bounded sequence in $K(H)$ has a convergent subsequence. This implies that $K(H)$ is finite dimensional.

  • $\begingroup$ Where you wrote $H \vert M$, do you mean $S$ is defined on the quotient space $H/M$ or that $S$ is defined on $H$ when restricted to $M$? $\endgroup$ – kkc Aug 14 at 18:07
  • $\begingroup$ I meant the quotient space. $\endgroup$ – Kavi Rama Murthy Aug 14 at 23:12
  • $\begingroup$ could you please explain your answer with a bit more detail. I am having trouble understanding it $\endgroup$ – kkc Aug 14 at 23:34
  • 1
    $\begingroup$ @kkc I have added some details. $\endgroup$ – Kavi Rama Murthy Aug 14 at 23:44
  • $\begingroup$ Did you read my answer? Isn't $(y_n)$ any bounded sequence in $K(H)$ in my answer? $\endgroup$ – Kavi Rama Murthy Aug 15 at 23:11

The other answer is correct and helpful to know because it is quite general.

However, it is probably instructive to choose a more constructive path in your simple situation. For this, show that there exist $(a_n)\in \ell^2$ such that $(a_n \lambda_n^{-1})\not\in \ell^2$. Finally choose a unit vector $x_n$ in the image of each $P_n$ and conclude that $\sum_n x_n a_n $ is not in the image of your operator but in the closure of the image.


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.