1
$\begingroup$

Let $A\subset \ell^2$ consist of all $x\in \ell^2$ with $|x|_2=1$ which does not belong to any $\ell^p$ for all $0<p<2$.

Note that $A$ is non-empty with a Baire category argument.

I am interested in (somewhat) distribution and configuration of $A$ and its complement in the unit sphere $S$ of $\ell^2$.

Note that $S\setminus A$ is path connected.

Is $S\setminus A$ a contractible space?

What can be said about connected components of $A$? In particular can one find at least one non constant curve?

Note: I was thinking to a possible density of $A$ in the sphere, or to a possible measure of $A$ as a subset of sphere. Note that density of a subset $A$ at a point $p$ of a metric space with measure $\mu$ is the limit $\frac{\mu(A\cap B_\epsilon)}{\mu (B_{\epsilon})}$ as $\epsilon$ goes to 0. ($B_\epsilon$ is the ball around $p$).

But I realized that the measure theory on infinite Banach spaces is a technical matter:

What is known about the Gaussian measure of the unit ball in a Hilbert Space?

What is the 'right' definition of zero measure subsets of Banach spaces?

Invariant probability on a unit ball of a Banach space

https://math.stackexchange.com/questions/75932/measure-on-hilbert-space

$\endgroup$
15
  • $\begingroup$ $A$ is dense, also by Baire category. (Or, really, it’s just the unit sphere minus a proper subspace. Of course it’s dense.) $\endgroup$
    – David Gao
    Commented Aug 17 at 10:17
  • $\begingroup$ @DavidGao Obviously both $A$ and $S\setminus A$ are dense. I do not get what do you meant? $\endgroup$ Commented Aug 17 at 10:21
  • $\begingroup$ I thought your note meant you’re thinking about asking whether $A$ is dense? I suppose I was mistaken. $\endgroup$
    – David Gao
    Commented Aug 17 at 10:25
  • 1
    $\begingroup$ For your second question, read Chapter 6 in the book of Benyamini and Lindenstrauss for concepts of null sets and their uses. $\endgroup$ Commented Aug 17 at 15:57
  • 1
    $\begingroup$ The book is in the list on Wikipedia: Geometric nonlinear functional analysis (with Yoav Benyamini). Colloquium publications, 48. American Mathematical Society, 2000. $\endgroup$ Commented Aug 19 at 13:37

2 Answers 2

4
$\begingroup$

Umm, $A$ is path-connected and it is fairly simple -- just change the coordinates one by one. Say we want to go from $u\in A$ to $v\in A$. On $[0, \frac{1}{2}]$ change linearly $u_1$ to $v_1$, then on $[\frac{1}{2}, \frac{2}{3}]$ change linearly $u_2$ to $v_2$ and so on. The only slight issue is that we are leaving the sphere, but just consider $v(t) = u(t)/||u(t)||$, then $v(0) = u, v(1) = v$, and $v(t)\notin \ell^p$ for all $0 < p < 2$ and all $0\le t < 1$ since $v(t)$ is equal to $u$ up to finitely many coordinates and a multiplication by a constant, and $u\notin \ell^p$ for all $0 < p < 2$, and for $t = 1$ we have $v(t) = v\notin \ell^p, 0 < p < 2$ as well.

The only thing that is a tiny bit not obvious is continuity of $v(t)$ at $t = 1$, but this follows since for any vector $w\in \ell^2$ and any $\epsilon > 0$ there exists $N(w, \epsilon)$ such that $\sum_{n > N} |w_n|^2 < \epsilon$.

$\endgroup$
3
  • 1
    $\begingroup$ Ah, this is far simpler. I was thinking we had to rotate to points with disjoint support before uniformly moving from one to the other. Forgot we could just alter coordinates one-by-one. I suppose this spared me of the effort needed to write down my approach then. +1 $\endgroup$
    – David Gao
    Commented Aug 17 at 13:17
  • $\begingroup$ Thank you for your intersting argument.So this question gives an example of an infinite dimensional sphere with two complemented dense set each of them path connected..I doubt this could be hold in finite dimensional case $\endgroup$ Commented Aug 17 at 16:36
  • $\begingroup$ I wonder if A is contractible too? $\endgroup$ Commented Aug 17 at 16:44
2
$\begingroup$

For the first question, yes, $S \setminus A$ is contractible. Rewrite the standard basis so that it is indexed by $\mathbb{Z}$. Let $U$ be the bilateral shift. Then $U$ is an isomorphism on $\ell^p$ for any $0 < p \leq 2$. Recall, from this paper, that $\varphi(x) = \frac{1}{2}(1 - \|x\|)e_0 + U(x)$ is a homeomorphism from the unit ball of $\ell^2$ onto itself without fixed point. From the geometric description of $\varphi$, contained in the proof of Theorem 1 of the cited paper, and using the facts that $e_0 \in \ell^p$ and $U$ is an isomorphism of $\ell^p$ for all $0 < p < 2$, it is not hard to see that $\varphi$ restricts to a homeomorphism from the unit ball of $\bigcup_{0 < p < 2} \ell^p$ (under the $\ell^2$ norm) to itself. Then $\psi$, which sends $x$ to the point at which the ray from $\varphi(x)$ to $x$ meets the unit sphere, is a retraction from the unit ball of $\bigcup_{0 < p < 2} \ell^p$ onto the unit sphere of $\bigcup_{0 < p < 2} \ell^p$ (i.e., $S \setminus A$). Thus, $S \setminus A$ is contractible, as $\Phi(x, t) = \psi(tx)$ is a homotopy between a constant map and the identity map on $S \setminus A$.


On second thought, I decided to still write down my approach to proving $A$ is path-connected despite Aleksei's wonderful (and simpler) answer, to ensure there is one complete answer.

For any $x \in A$, denote $x_e$ to be the part of $x$ supported on even indices and $x_o$ to be the part of $x$ supported on odd indices. Let $x, y \in A$. We first show that there is a path from $x$ to $x'$ in $A$, and from $y$ to $y'$ in $A$, s.t. $x'$ and $y'$ have disjoint supports.

Since $x \notin \bigcup_{0 < p < 2} \ell^p$, we must have at least one of $x_e$ and $x_o$ is not in $\bigcup_{0 < p < 2} \ell^p$ either. The same holds for $y$. If $x_e \notin \bigcup_{0 < p < 2} \ell^p$ and $y_o \notin \bigcup_{0 < p < 2} \ell^p$, then simply let $x' = \frac{x_e}{\|x_e\|}$. Now define,

$$f(t)_n = \begin{cases} x_n &, \text{ if }n\text{ is even}\\ (1 - t)x_n &, \text{ if }n\text{ is odd} \end{cases}$$

$$g(t) = \frac{f(t)}{\|f(t)\|}$$

It is then easy to verify that $g$ is a path from $x$ to $x'$ in $A$. Similarly, there is a path from $y$ to $y' = \frac{y_o}{\|y_o\|}$ in $A$. Clearly, $x'$ and $y'$ have disjoint supports. Similar arguments also apply when $x_o \notin \bigcup_{0 < p < 2} \ell^p$ and $y_e \notin \bigcup_{0 < p < 2} \ell^p$.

Now, consider when $x_o \notin \bigcup_{0 < p < 2} \ell^p$ and $y_o \notin \bigcup_{0 < p < 2} \ell^p$. The point here is we can first rotate $x$ to $x''$ s.t. $(x'')_e \notin \bigcup_{0 < p < 2} \ell^p$, then apply the result of the previous paragraph. This can be easily achieved by,

$$f(t)_n = \begin{cases} \cos(\frac{\pi}{2}t)x_n + \sin(\frac{\pi}{2}t)x_{n+1} &, \text{ if }n\text{ is odd}\\ -\sin(\frac{\pi}{2}t)x_{n-1} + \cos(\frac{\pi}{2}t)x_n &, \text{ if }n\text{ is even} \end{cases}$$

Then set $x'' = f(1)$. Since $(x'')_e$ is simply $x_o$ shifted one index forward, we have $(x'')_e \notin \bigcup_{0 < p < 2} \ell^p$. $f(t)$ is simply rotation in each pair of adjacent coordinates, so it preserves the $\ell^2$ norm. Any rotation matrix is also both bounded below and above under $\ell^p$ norm for any $0 < p < 2$, so as $x \notin \bigcup_{0 < p < 2} \ell^p$, $f(t) \notin \bigcup_{0 < p < 2} \ell^p$ either for any $t$. Thus, $f(t)$ is indeed a path in $A$. Now apply the result in the previous paragraph to reach the desired conclusion. The case where $x_e \notin \bigcup_{0 < p < 2} \ell^p$ and $y_e \notin \bigcup_{0 < p < 2} \ell^p$ can be proved similarly.

Now, it suffices to show that for $x, y \in A$ with disjoint supports, they can be connected by a path. Such a path is simply given by,

$$f(t) = \frac{tx + (1 - t)y}{\|tx + (1 - t)y\|}$$

$\endgroup$
12
  • $\begingroup$ I highly suspect $A$ is path-connected. It should be possible to “rotate” two elements of $A$ in $A$ so that their supports become disjoint, then just use a line segment (properly normalized) to connect the two. But I couldn’t figure out how to formalize this. $\endgroup$
    – David Gao
    Commented Aug 17 at 11:30
  • $\begingroup$ I am leaving for a few hours I will come back here, thanks again $\endgroup$ Commented Aug 17 at 11:34
  • $\begingroup$ To be honnest I doubt even A is connected but I have no idea to proof. Some how I am curious : which one is bigger A or its complement??!! $\endgroup$ Commented Aug 17 at 11:35
  • $\begingroup$ befor I read the details, in the linked paper the set $|x|\leq 1$ is called sphere but the usual terminology is ball or disk I think. any way in my question I mean the sphere the point of unit norm not less than 1. But I guess your argument still work, yes? $\endgroup$ Commented Aug 17 at 12:30
  • 1
    $\begingroup$ @AliTaghavi I believe so. The path constructed in Aleksei’s answer is rather uniform in the starting point, so you can just combine those paths into a contraction of $A$ onto a single point. There are some technical details to check, but I’m reasonably sure it should work. $\endgroup$
    – David Gao
    Commented Aug 17 at 17:25

You must log in to answer this question.

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