This is a homework question and I find myself ending up with the "opposite" of what I am supposed to prove. Below is the question and my attempt:

Suppose that $\lambda \in \mathbb{R}$. Show that if $\lambda \in \sigma_p(T)$ and $\lambda \notin \sigma_p(T^*)$, then $\lambda \in \sigma_r(T^*)$. I assume $T$ is a bounded linear operator on $X$ and maps to $X$, where $X$ is Banach. $\sigma_p(T)$ is the point spectrum of $T$, and $\sigma_r(T)$ is the residual spectrum.

I thought $\lambda \in \sigma_p(T)$ implies $\lambda I - T$ is not injective and $\exists v \neq 0$ such that $\lambda v = Tv$, and $\lambda \notin \sigma_p(T^*)$ implies $\lambda I - T^*$ is injective.

Since $\sigma(T) = \sigma(T^*)$ (reference: Spectrum for a bounded linear operator and its adjoint on a Banach space are same., or Conway, "A Course in Functional Analysis", pg.208 Prop 6.1), then $\lambda$ must be in either $\sigma_c(T^*)$ or $\sigma_r(T^*)$, where $\sigma_c(T^*)$ is the continuous spectrum. However, as $Ker((\lambda I - T)^*) = Ker(\lambda I - T^*) = \{0\}$, and $Ran(\lambda I - T)^{\perp} = Ker(\lambda I - T^*)$, so $Ran(\lambda I - T)$ must be dense in $X$, so $\lambda \in \sigma_c(T^*)$ by definition. However, I'm supposed to show $\lambda \in \sigma_r(T^*)$.

Can anyone point the mistake I made?


Well I don't know how you obtained $\lambda\in\sigma_c(T^*)$, as you showed was that $\operatorname{Ran}(\lambda I-T)$ is dense, not that $\operatorname{Ran}(\lambda I-T^*)$ is dense. In any case, the calculation you want is $$^\perp\operatorname{Ran}(\lambda I-T^*)=\ker(\lambda I-T)\neq\{0\}.$$ Thus $\lambda I-T$ does not have dense range. But by hypothesis, $\lambda\notin\sigma_p(T^*)$, and thus $\lambda\in\sigma_r(T)$.

  • $\begingroup$ I thought that $\lambda I - T^*$ is injective implies $Ker(\lambda I - T^*) = \{0\}$, and $Ran(\lambda I - T)^{\perp} = Ker(\lambda I - T^*)$, so $Ran(\lambda I - T)$ must be dense. And the continuous spectrum is defined as $\sigma_c(A) := \{\lambda \in \mathbb{C}: \lambda I - A \text{ is injective, } Ran(\lambda I - A) \text{ is dense, and } A \text{ is not surjective}\}$ $\endgroup$ – Yang Apr 23 '18 at 22:15
  • $\begingroup$ Yes, but you showed $\operatorname{Ran}(\lambda I-T)$ is dense, not that $\operatorname{Ran}(\lambda I-T^*)$ is dense. $\endgroup$ – Aweygan Apr 23 '18 at 22:18
  • $\begingroup$ Oh this is a stupid mistake. Thanks a lot! $\endgroup$ – Yang Apr 23 '18 at 22:20
  • $\begingroup$ You're welcome. Glad to help! $\endgroup$ – Aweygan Apr 23 '18 at 22:20
  • $\begingroup$ Something is wrong here. Didn't you have shown that $\lambda-T'$ (not $\lambda-T$) does not have dense range and hence $\lambda\in\sigma_r(T')$ (not $\lambda\in\sigma_r(T)$, which would contradict $\lambda\in\sigma_p(T)$ by the way)? $\endgroup$ – 0xbadf00d Feb 2 at 5:52

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.