Let $X$ be some reflexive Banach space. Suppose $x_n$ is some sequence in $X$ that weak converges to some $y \neq 0$. Is it the case that $$ \limsup \|x_n - y\| < \limsup \|x_n\| ?$$

Positive with Hilbert spaces

In the Hilbert space case this is true, as $$ \langle x_n - y, x_n -y \rangle = \| x_n\|^2 + \|y\|^2 - \langle x_n, y\rangle - \langle y, x_n\rangle $$ and by weak convergence the latter two terms both converges to $\|y\|^2$ and we get $$ \limsup \|x_n - y\|^2 = \limsup \|x_n\|^2 - \|y\|^2. $$

A negative example with a non-reflexive Banach space and weak* convergence

Let $X = L^\infty(\mathbb{R})$, and take $x_n = \mathbf{1}_{[-2,-1]} + \mathbf{1}_{[n,n+1]}$. Then $x_n$ weak* converges to $\mathbf{1}_{[-2,-1]}$ but $\|x_n - \mathbf{1}_{[-2,-1]}\|_{\infty} = 1 = \|x_n\|_\infty$.

  • $\begingroup$ [deleted previous comments which were based on guesswork that the answer below refutes] $\endgroup$ – Yemon Choi Dec 7 '20 at 18:09

The property you indicate is known as (strict) Opial’s Property (see https://en.m.wikipedia.org/wiki/Opial_property). It fails generally in reflexive spaces; in fact, it fails generally even for uniformly convex spaces where it is equivalent to Delta convergence (see https://en.m.wikipedia.org/wiki/Delta-convergence). Indeed, for $L^p[0,1]$, where $p\ne 2$, Opial’s Property fails. Using the duality pairing $\langle\cdot\,,\cdot\rangle\colon X\times X^*\to \mathbb{R}~or~\mathbb{C}$ and (the normalized) duality map $$J(x):=\{f\in X^*:\|f\|^2=f(x)=\|x\|^2\}\,,$$ Brailey Sims (https://www.mathshunter.edu.au/brailey/Research_papers/As%20Support%20Map%20Characterisation%20of%20the%20Opial%20Conditons.pdf) characterised spaces with the strict Opial Property as those spaces such that whenever $x_n$ converges weakly to $x\ne 0$, then $$\limsup_{\substack{n\to\infty\\j x_n\in J x_n}}\langle x\,,j x_n\rangle>0\,.$$ (The $\limsup$ may be equivalently replaced with $\liminf$). From this it is easy to deduce that the sequence spaces $\ell^p$, where $1\le p<\infty$ have Opial’s Property.


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.