3
$\begingroup$

Suppose $X$ is a Banach space with the following property: For any $x\in X$ there exists a two dimensional subspace $E$ isometric with $l_2^2$ such that $x\in E$. Does this property characterize a (separable) Hilbert space?

What about the stronger property: For any $x$ and any $n\in\mathbb{N}$ there exists a $n$-dimensional subspace $E$ isometric with $l_2^n$ such that $x\in E$?

$\endgroup$
1
  • 2
    $\begingroup$ C'mon, take a Cartesian product $\ \ell_2^b\times\mathbb R^2\ $ of $\ \ell_2^n\ $ and $\ \mathbb R^2\ $ with the max distance. You can make it into a Banach space in many different way but such that $\ \ell_2^n\times\{x\}\ $ will be isometric to $\ \ell_2^n\times\{y\};\ $ and $\ \{x\}\times\mathbb R^2\ $ will be isometric to $\ \mathbb R^2\ $ with the max distance. However, such a Banach space would contain isometrically $\ \mathbb R^2\ $ with the max distance hence it cannot be Hilbert (there would be two different points that have more than one metric center). $\endgroup$
    – Wlod AA
    Commented Oct 25, 2022 at 21:15

3 Answers 3

6
$\begingroup$

For any Banach space $X$ you can consider $X\oplus l^2$, with norm $||(x,y)||:=(||x||^2+||y||^2)^\frac{1}{2}$. Then for each $x\in X$, span$(x)\oplus l^2$ is isometric to $l^2$, so $X\oplus l^2$ is covered by isometric copies of $l^2$

$\endgroup$
5
  • $\begingroup$ Hmm, the Answer-format is easier to edit than the Comment-format hence you finished about 30s before me. :) $\endgroup$
    – Wlod AA
    Commented Oct 25, 2022 at 21:19
  • $\begingroup$ Yeah we ended almost at the same time haha. Tbh I was going to post it as a comment but I thought it would make more sense as an answer. $\endgroup$
    – Saúl RM
    Commented Oct 25, 2022 at 21:19
  • 1
    $\begingroup$ Sure, have fun rather than to suffer with the awful MO-comment editor -- not that the MO-editor is much better to be fair, both are HORRIBLE. $\endgroup$
    – Wlod AA
    Commented Oct 25, 2022 at 21:23
  • $\begingroup$ Yes, I overcomplicated myself. This is too trivial for MO. Sorry about that. Maybe an admin should delete it. $\endgroup$
    – Markus
    Commented Oct 25, 2022 at 21:36
  • 1
    $\begingroup$ @Markus, you could still upvote Saul :) (since you didn't care, I took care ot it :) ). $\endgroup$
    – Wlod AA
    Commented Oct 25, 2022 at 21:48
3
$\begingroup$

Let $\ B\ $ bee an arbitrary Banach space. If each of its 3-dimensional linear space is isomorphic to $\ \ell_2^3\ $ than the inner product can be introduced there properly, and $\ B\ $ is Hilbert then.

I seem to remember that this works even for $\ \ell_2^2\ $ in place of $\ \ell_2^3.$

Otherwise, $\ B\ $ is not Hilbert, far from this (as the examples above show).

$\endgroup$
0
$\begingroup$

No. That property does not characterize Hilbert spaces. Let $H$ be an infinite dimensional separable Hilbert space. Let $K$ be the space of all bounded sequences $(x_n)_{n=0}^{\infty}$ where $x_n\in H$ for each $n$. Then $K$ is a normed vector space with norm $\|(x_n)_n\|=\sup\{\|x_n\|:n\geq 0\}$, and it is not too hard to see that $K$ is a Banach space.

Suppose now that $(x_n)_{n\in\omega}\in K$. Then let $\alpha_n=\|x_n\|$ whenever $n\geq 0$. Let $(x_{n,m})_{n\geq 0,m\geq 0}$ be a collection such $x_{n,0}=x_n$ for each $n$, for each $n$, the set $(x_{n,m})_{m\geq 0}$ is orthogonal, and where $\|x_{n,m}\|=\alpha_n$ for each $n.$

Let $\mathbf{x}_m=(x_{n,m})_{n\geq 0}$. Let $W$ be the subspace of $K$ spanned by $(\mathbf{x}_m)_{m\geq 0}$. Then whenever $\|\beta_1\mathbf{x}_1+\dots+\beta_r\mathbf{x}_r\|=|\beta_1|^2\cdot\|\mathbf{x}_1\|+\dots+|\beta_r\|^2\cdot\|\mathbf{x}_r\|^2$. In other words, the induced norm on $W$ is induced by an inner product.

On the other hand, $K$ is very far from being an inner product space since $K$ has a copy of $\ell^\infty$.

$\endgroup$
2
  • 1
    $\begingroup$ In fact, I am interested in the separable case. Can you think of a separable counterexample? $\endgroup$
    – Markus
    Commented Oct 25, 2022 at 20:28
  • $\begingroup$ If $1\leq p\leq\infty$, then you can take the norm $\|(x_n)_n\|_p=\|(\|x_n\|)_n\|$ and define $K_p$ to be the subset of $K$ consisting of all sequences where $\|(x_n)_n\|_p$ converges. This should be a non-Hilbert Banach space (which does not satisfy the parallelogram law) as well and it should be separable whenever $1<p<\infty$. $\endgroup$ Commented Oct 25, 2022 at 21:13

You must log in to answer this question.

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