2
$\begingroup$

Let $E$ be a Banach space, let $T:E\to E$ have norm $1$ and let $\nu\in E^*\setminus\{0\}$ be such that $T^*\nu=\nu$. Under which conditions there is $e\in E$ such that $Te=e$ and $\langle e,\nu\rangle\ne 0$?

This is true for reflexive spaces. Indeed, WLOG $\|\nu\|=1$. Consider $D=\nu^{-1}(1)\cap \overline{B}_{E}$, which is weakly compact and non-empty. Then $T_{D}$ is weak-to-weak continuous self-map of $D$. Therefore, there is a fixed point due to Tychonoff fixed point theorem. Thus, I am looking for conditions significantly weaker than reflexivity.

REMARK: In fact, the argument works if there is a linear topology on $E$ such that $T$ and $\nu$ are continuous with respect to that topology and $\overline{B}_{E}$ is compact in that topology.

$\endgroup$
3
  • 1
    $\begingroup$ The existence of such an $e$ for each $\nu$ is equivalent to $T$ being mean ergodic. $\endgroup$ Jan 5, 2020 at 10:55
  • $\begingroup$ @JochenGlueck do you mean for every $\nu$ that is a fixed point of $T^*$? If so, could you please elaborate? $\endgroup$
    – erz
    Jan 5, 2020 at 18:24
  • $\begingroup$ Yes, I mean for every $\nu$ that is a fixed point of $T^*$. I added an answer with details and a reference. $\endgroup$ Jan 6, 2020 at 15:14

1 Answer 1

4
$\begingroup$

Part 1 of the answer. In terms of $T$, the property you are looking for is characterized by the mean ergodic theorem:

Theorem. Let $E$ be a Banach space and let $T$ be a bounded linear operator on $E$ that is power-bounded in the sense that $\sup_{n \in \mathbb{N}_0}\|T^n\| < \infty$. Then the following assertions are equivalent:

(i) $T$ is mean ergodic, i.e. the sequence of Cesàro means $(\frac{1}{n}\sum_{k=0}^{n-1} T^k)_{n \in \mathbb{N}}$ converges strongly.

(ii) The fixed space of the dual operator $T^*$ is separated by the fixed space of the operator $T$, i.e. for every $0 \not= \nu \in \ker(1 - T^*)$ there exists $0 \not= e \in \ker(1-T)$ such that $\langle \nu, e \rangle \not= 0$.

(iii) The space $E$ is the topologically direct sum of the fixed space $\ker(1-T)$ and the closure of the range of $1-T$.

This is a classical theorem in operator theory. You can find it, for instance, in Theorem 8.20 of [T. Eisner, B. Farkas, M. Haase, R. Nagel: Operator Theoretic Aspects of Ergodic Theory (2015)] (actually, the theorem there contains several more equivalent assertions and is not only stated for power-bounded operators but for the slightly larger class of Cesàro bounded operators).

The result for reflexive spaces that is stated in the question is a special case of the above theorem since every power-bounded operator on a reflexive Banach space is mean ergodic [op. cit., Theorem 8.22].

Part 2 of the answer. The question on which spaces every power-bounded (or contractive) operator is mean-ergodic is a very subtle one.

  • In the paper [V. P. Fonf, M. Lin, P. Wojtaszczyk: Ergodic Characterizations of Reflexivity of Banach Spaces (Journal of Functional Analysis, 2001)] it is shown that a Banach space with a Schauder basis is necessarily reflexive if every power-bounded operator on it is mean ergodic.

  • In the paper [V. P. Fonf, M. Lin, P. Wojtaszczyk: A non-reflexive Banach space with all contractions mean ergodic (Israel Journal of Mathematics, 2010)] it is shown that there exist non-reflexive Banach spaces an which every operator $T$ of norm $\|T\| \le 1$ is mean ergodic.

$\endgroup$
5
  • 1
    $\begingroup$ We may add that the limit in (i) is the linear projector onto $\ker(I-T)$ associated with the decomposition in (iii) $\endgroup$ Jan 6, 2020 at 18:04
  • 1
    $\begingroup$ Thank you! A small correction: in (ii) the latter is separated by the former. $\endgroup$
    – erz
    Jan 7, 2020 at 3:48
  • $\begingroup$ This is however, not exactly what I was looking for; I was more interested in conditions on $E$ and $\nu$... $\endgroup$
    – erz
    Jan 7, 2020 at 3:50
  • $\begingroup$ @erz: Thank you for the correction; I corrected (ii) accordingly. Concerning your second comment, I added a second part to the answer: for most Banach spaces, one cannot weaken the reflexivity assumption if one wants all power-bounded operators to be mean ergodic. $\endgroup$ Jan 7, 2020 at 9:55
  • $\begingroup$ The thing is, I don't need this property to hold for every $\nu$, I need it to hold for a given $\nu$, and so it may depend on $\nu$. This should be a geometric condition that takes into account the way the kernel of $\nu$ dissects the unit ball of $E$. Thank you nonetheless! $\endgroup$
    – erz
    Jan 7, 2020 at 16:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.