2
$\begingroup$

Let $X$ be a real Hilbert Space and $C \subseteq X$. Let $d_C$ be the infimal distance function to $C$ and $P_C(x) = C \cap S[x; d_C(x)]$ be the metric projection. We say $C$ is proximinal if $P_C(x) \neq \emptyset$ for all $x \in X$.

I'm wondering if the intersection of two proximinal subsets of $X$ must be proximinal? To avoid trivial counterexamples, I need their intersection to be non-empty.

The answer is most likely "no", or at least, something that is difficult to prove in the affirmative if true. Asplund proved that, given any non-convex Chebyshev subset $C$ (meaning $|P_C(x)| = 1$ for all $x$), there exists a closed half-space $H$ such that $X \cap H$ is not proximinal. Therefore, were proximinal sets closed under intersection, this would prove the Chebyshev conjecture: that all Chebyshev subsets are convex.

I was wondering if someone had an explicit example of two proximinal sets whose intersection is not proximinal?

$\endgroup$
3
  • $\begingroup$ @Jack I remember you, I think (this was you?). Thank you, I appreciate the support. Their behaviour is inconsistent with both their stated goals of collecting interesting mathematics and helping people. I think I first realised this with this question which I went to bat for in Meta. Check out the edit history! $\endgroup$ Aug 5, 2021 at 18:58
  • $\begingroup$ @Jack I'm still contributing, but it's getting harder. I'm coming around to the opinion that, due to the way MSE is being run, it is doing active harm to mathematics in general. I honestly believe that maths anxiety is pure maths' biggest issue, and that people coming away from maths education traumatised eventually leads to what we've seen for the past 20 years: dwindling support for pure mathematics. It's so important that we, as a community, engage in positive, encouraging outreach. MSE is pouring fuel on this dumpster fire. If I stay off meta, I still can do good, leading by example. $\endgroup$ Aug 5, 2021 at 19:05
  • 1
    $\begingroup$ Thank you for your messages. It is inspiring that you choose to continue your contribution to the main site. I found it difficult for me to make a peace with myself when seeing relentless daily close/deletion of good content in the site and a clique dominating the meta with nonsense while I was not able to do any change. But your words make me think that maybe we can still help others even without being able to stop those harmful activities. Staying off the meta (and the troubling users) and focusing on the good side may be the first step to make a peace. $\endgroup$
    – user14319
    Aug 5, 2021 at 19:44

1 Answer 1

1
$\begingroup$

This answer is strongly inspired by example 3.11 in the book by Bauschke and Combettes.

Let $H = \ell^2$ and consider a sequence $\{\alpha_n\} \in (1,\infty)$ with $\alpha_n \searrow 1$. Define \begin{align} C &= \{ \alpha_n \, e_n \mid n \in \mathbb N\} \cup \{0\}\\ D &= \{ x \in \ell^2 \mid \exists n \in \mathbb N : x_n \ge 1 \}. \end{align} The set $C$ is weakly closed, thus a proximinal set (e.g. Prop. 3.12 in Bauschke&Combettes).

Moreover, it is easy to check that $D$ is proximinal: For $y \in \ell^2$ one can construct a projection as follows: Take $j \in \operatorname{arg\,max}_j y_j$. If $y_j \ge 1$, then $y \in D$. Otherwise, obtain $x$ by replacing the entry $y_j$ in $y$ by $1$. Then, $x$ is a projection of $y$ onto $D$.

Finally, we have $C \cap D = C \setminus \{0\}$ and this set is not a proximinal set, see example 3.11 in Bauschke and Combettes, as $0$ has no projection onto $C \setminus \{0\}$.

$\endgroup$
2
  • $\begingroup$ Thank you for the answer! Just to clarify, do you mean $\operatorname{argmax}$ instead of $\operatorname{argmin}$? $\endgroup$ Oct 23, 2018 at 8:32
  • $\begingroup$ Yes, of course. $\endgroup$
    – gerw
    Oct 23, 2018 at 8:45

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.