Claim: Let $A$ be a bounded linear operator on a Banachspace $\mathfrak{X}$. Denote $\sigma(A)$ as the spectrum of A. Let $\lambda$ be a point in the boundary of the $\sigma(A)$. Then there exist a sequence $\{x_n\}_{n\geq 0}\subset \mathfrak{X}$, $\|x_n\| = 1$ such that

$$(\lambda\mathbb{1} - A)x_n\rightarrow 0.$$

How can I prove this?

  • $\begingroup$ Hint: $\sigma(A)$ is a compact subset of $\mathbb C$, so $\partial \sigma(A)\subset \sigma(A)$. $\endgroup$ – Math1000 Apr 15 '16 at 14:38
  • $\begingroup$ @Math1000 sorry I still don't see it. $\endgroup$ – john Apr 15 '16 at 15:50
  • $\begingroup$ @Math1000 This means that $\lambda \in \sigma(A)$ and therefore $(\lambda \mathbb{1} - A)$ is not invertible, so therefore the kernel is non trivial? But wouldn't this mean there is even an $x \in \mathfrak{X}$ such that $(\lambda \mathbb{1} - A) x = 0$ $\endgroup$ – john Apr 15 '16 at 15:57
  • $\begingroup$ @Math1000 Ok I realize that this argument is only valid if $\lambda$ is in the point spectrum. If $\lambda$ is in the continues spectrum then I see that the sequence exists, now all that is left is the residual spectrum right? $\endgroup$ – john Apr 15 '16 at 16:08
  • 1
    $\begingroup$ If $\lambda\in\partial\sigma(A)$ then there exists a sequence $\lambda_n\subset\rho(A)$ with $\lambda_n\to\lambda$. Then what can we say about $\|(\lambda_n I-A)^{-1}\|$? $\endgroup$ – Math1000 Apr 15 '16 at 16:08

Because $\lambda\in\partial\sigma(A)$, then there exists $\{ \lambda_n \} \subset\rho(A)$ that converges to $\lambda$. Suppose for the moment that the following holds for all $x$: $$ l(x)=\sup_{n}\|(\lambda_n I-A)^{-1}x\| < \infty. $$ Then, by the uniform boundedness principle, $$ M=\sup_{n} \|(\lambda_n I -A)^{-1} \| < \infty. $$ Hence, for $|\lambda-\lambda_n|M < 1$, it must hold that $\lambda I-A$ is invertible because $(\lambda_n I-A)$ is invertible and \begin{align} (\lambda I-A) & = (\lambda-\lambda_n)I+(\lambda_n I-A)\\ & = \{(\lambda-\lambda_n)(\lambda_nI-A)^{-1}+I\}(\lambda_n I-A) \end{align} This contradiction proves that the original assumption was false. Hence, there exists $x$ such that $\lim_{k} \|(\lambda_{n_k} I-A)^{-1}x\|=\infty$. Let $$ y_k = \|(\lambda_{n_k} I-A)^{-1}x\|^{-1}(\lambda_{n_k}I-A)^{-1}x. $$ Then $\{ y_k \}_{k=1}^{\infty}$ is a sequence of unit vectors for which $$ (\lambda I-A)y_k = (\lambda-\lambda_{n_k})y_k+\|(\lambda_{n_k}I-A)^{-1}x\|^{-1}x $$ Hence, $\lim_{k}(\lambda I-A)y_k = 0$.


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.