1
$\begingroup$

Let $(E, |\cdot|)$ be an infinite-dimensional Banach space. Assume that

  • $T:E\to E$ is a compact (bounded linear) operator, and
  • $(\lambda_n)$ is a sequence of distinct eigenvalues of $T$.

Let $E_n$ be the corresponding eigenspace of $\lambda_n$. Then $(E_n)$ is a sequence of finite-dimensional subspaces of $E$ such that $E_n \cap E_m = \{0\}$ for all $n \neq m$.

Is there a bounded sequence $(e_n)$ such that $e_n \in E_n$ for all $n$ and that $(e_n)$ does not have any convergent subsequence?

Thank you so much for your elaboration!

$\endgroup$
4
  • 2
    $\begingroup$ Your question is a bit unclear. Which of the following did you mean to ask: (i) "Is it true that for any such $E$ and $T$ there is a bounded sequence $(e_n)$ such that $e_n \in E_n$ for all $n$ and that $(e_n)$ does not have any convergent subsequence?" and (i) "Is it true that for some such $E$ and $T$ there is a bounded sequence $(e_n)$ such that $e_n \in E_n$ for all $n$ and that $(e_n)$ does not have any convergent subsequence?" $\endgroup$ Commented May 26, 2023 at 16:22
  • $\begingroup$ @IosifPinelis I meant (i)... $\endgroup$
    – Analyst
    Commented May 26, 2023 at 16:24
  • 1
    $\begingroup$ If you mean (i) then you will need to impose some more conditions, to ensure that T has an infinite number of distinct eigenvalues... $\endgroup$
    – Yemon Choi
    Commented May 26, 2023 at 17:16
  • $\begingroup$ @YemonChoi The existence of such sequence $(\lambda_n)$ is actually part of the assumption. $\endgroup$
    – Analyst
    Commented May 26, 2023 at 17:28

2 Answers 2

6
$\begingroup$

The answer is negative. Take a biorthogonal sequence $(x_n,x_n^*)$ in $E$ such that $(x_n)$ converges to some unit norm vector $x_0$. Set

$T=\sum_n 2^{-n} \|x_n\|^{-1} x_n^* \otimes x_n$.

Such a sequence exists in every infinite dimensional space. For example, in $\ell_2$ let $x_n = e_0 + n^{-1} e_n$, where $(e_n)_{n=1}^\infty$ is orthonormal, and $x_n^*$ is $\langle \cdot, n e_n \rangle$.

$\endgroup$
0
$\begingroup$

Partial answer: yes for normal operators on Hilbert spaces. We can write any normal compact operator as $\sum_{n\geq 1} \lambda_n f_n g_n^*$ with $(f_n)$, $(g_n)$ orthonormal families and $\lambda_n \to 0$. A $\lambda\neq 0$ is an eigenvalue if and only if $N_\lambda = \{ n : \lambda_n = \lambda\}$ is non-empty, in which case the eigenspace is $\mathrm{Vect}(f_n : n\in N_\lambda) = \mathrm{Vect}(g_n : n\in N_\lambda)$. This means that eigenspaces associated to different eigenvalues are necessarily orthogonal. Given an infinite sequence of distinct eigenvalues, one can then easily find a sequence of eigenvectors from which no subsequence converges.

$\endgroup$
2
  • $\begingroup$ That would be the case only for normal compact operators, right ? $\endgroup$ Commented May 27, 2023 at 16:40
  • $\begingroup$ You are correct; wikipedia lacks the hypothesis... I have updated my answer. It does weaken it even more than before. $\endgroup$ Commented May 27, 2023 at 16:52

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .