4

Given a compact operator, it is well-known that the Spectrum consists only of eigenvalues and possibly 0. Now I'm thinking about the inverse implication with additional conditions.

So, given a normal operator T such that

$\sigma(T) \subseteq \sigma_p(T) \cup \{ 0 \} $

and such that $\sigma_p(T)$ has no non-zero accumulation points and no non-zero eigenvalue $\lambda$ of infinite multiplicity (i.e. such that $\dim \ker (\lambda - T) = \infty$).

Then, I have a source that states that $T$ is compact (used in the proof of 6.14 in Invariant Subspaces by Rosenthal and Radjavi).

Actually, now that I've written it down I think it's also possible that I only need the latter two conditions. Sorry if the solution is obvious and I'm not seeing the wood for the rees and thanks in advance.

Myshkin
  • 36,898
  • 28
  • 168
  • 346
Haluter
  • 115
  • My thought is that the "obvious approach" would be to try to approximate $T$ with (normal) operators of finite rank. Could we use the Gelfand transform here? – Ben Grossmann Jun 15 '16 at 20:48
  • Have you learned any type of functional calculus for normal operators? Without that, it's going to be tough to prove what you want. – Disintegrating By Parts Jun 17 '16 at 01:18
  • @TrialAndError: Yes, I know the continous functional calculus. I've heard of the Borel one, too, but I think it won't help here since it only works fors self-adjoint elements? I'll think about the problem again with focus on the continous functional calculus. – Haluter Jun 17 '16 at 07:55
  • @Omnomnomnom: Unfortunately I'm not able to see how to get finite-rank operators in this setting. Lack of creativity or experience on my part, probably... – Haluter Jun 17 '16 at 07:56
  • 1
    Use the continuous functional calculus, and consider functions that are zero on a neighborhood of $0$. – Ben Grossmann Jun 17 '16 at 12:38
  • Perhaps it is sufficient to show that $T^*T$ is compact? I don't remember if that's true – Ben Grossmann Jun 17 '16 at 13:48

1 Answers1

3

The continuous Functional Calculus will give you want you want, under your assumptions. If $\lambda \ne 0$ is a point of the spectrum, you can construct a continuous function $F_{\epsilon}$ that is $1$ at $\lambda$ and is $0$ on the remaining part of the spectrum. Then $P_{\lambda}=F_{\epsilon}(T)$ is selfadjoint and $P_{\lambda}^2=P_{\lambda}$. Because $(z-\lambda)F_{\epsilon}(z)$ is identically $0$ on the spectrum, then $TP_{\lambda}=\lambda P_{\lambda}$. You end up with $P_{\lambda}$ being the orthogonal projection onto $\mathcal{N}(T-\lambda I)$, a space which you have assumed to be finite-dimensional.

Let $\epsilon > 0$. The functional calculus can be used to construct a finite rank operator $T_{\epsilon}$ such that $\|T-T_{\epsilon}\| < \epsilon$ as follows. First, there exists $r \in (0,\epsilon)$ such that $\sigma(T)\cap\{ z : |z|=r \}=\emptyset$; this follows from your assumption that there is no finite non-zero point of accumulation of the spectrum. There are at most a finite number of points in $\sigma(T)\cap\{\lambda : |\lambda| > r \}$. Therefore, there is a continuous function $F_{\epsilon}$ on $\mathbb{C}$ such that $$ |F_{\epsilon}(z)| \le 1 \mbox{ for all $z\in\mathbb{C}$, and } F_{\epsilon}(z)= \left\{\begin{array}{ll} 1, & |z| \le r \\ 0, & |z| > r, z\in \sigma(T). \end{array} \right. $$ Let $G_{\epsilon}(z)=1-F_{\epsilon}(z)$. Then $I=F_{\epsilon}(T)+G_{\epsilon}(T)$. Both $F_{\epsilon}(T)$ and $G_{\epsilon}(T)$ are projections because they're either $0$ or $1$ at every point of the spectrum, which makes them idempotent and selfadjoint. You can further decompose $G_{\epsilon}(T)$ into functions that are $0$ on every one of the finite number of points in $\sigma(T)\cap\{\lambda : |\lambda| > r \}$, except for one point where the function is $1$; that gives $$ G_{\epsilon}(T) = \sum_{\lambda\in\sigma(T),\;|\lambda| > r}P_{\lambda} $$ where $P_{\lambda}$ is the orthogonal projection onto $\mathcal{N}(T-\lambda I)$. Therefore, $$ T = TF_{\epsilon}(T)+\sum_{\lambda\in\sigma(T),\;|\lambda| > r}\lambda P_{\lambda}, $$ and $$ \|TF_{\epsilon}(T)\|=\sup_{z\in\sigma(T)}|zF_{\epsilon}(z)| \le r < \epsilon. $$ Hence, $T$ is a norm limit of finite-rank operators $\sum_{\lambda\in\sigma(T),\;|\lambda| > r}\lambda P_{\lambda}$.

Disintegrating By Parts
  • 91,908
  • 6
  • 76
  • 168