2

I'm reading below result from a lecture note.

Corollary 0.18. Let $X$ be a normed space, $f: X \rightarrow(-\infty,+\infty]$ a convex function, $A \subset X$ an affine set, and $a: A \rightarrow \mathbb{R}$ a continuous affine function such that $a \leq f$ on $A$. Assume that $$ \operatorname{int}(\operatorname{dom}(f)) \cap A \neq \emptyset, $$ and either $f$ is continuous at some point of $\operatorname{dom}(f)$, or $f$ is l.s.c. and $X$ a Banach space. Then there exists a continuous affine extension $\hat{a}: X \rightarrow \mathbb{R}$ of a, such that $\hat{a} \leq f$.

Below I show that the continuity of $a$ is implied by that of $f$ and the hypothesis $a \le f$ on $A$. Could you have a check on my attempt?

Assume that $f$ is continuous at $x_0 \in \operatorname{dom}(f)$. Then $f$ is bounded on a neighborhood $U$ of $x_0$. Then $a$ is bounded from above on $U \cap A$ which is a neighborhood of $a$ in the subspace topology of $A$. This combining with the fact that $a$ is affine (a translation of a linear map) shows that $a$ is continuous at $x_0$ and thus on $A$.

Analyst
  • 6,351
  • It is not at all clear how you constructed an extension??? I would imagine that the Hahn Banach theorem would need to appear somewhere in the proof. – copper.hat Jun 20 '22 at 18:11
  • If $\operatorname{epi} f$ has a non empty interior, I think this is straightforward. – copper.hat Jun 20 '22 at 19:44
  • @copper.hat please see the proof (at the end of the pdf) by the author here. – Analyst Jun 20 '22 at 23:54
  • I got stuck at Observation 0.7 (iii) $\implies $(i). It is not true. Just because $a \in \operatorname{a-int} A$ and $x \in A$ does not mean that $(a,x) \subset \operatorname{a-int} A$ . I suspect that the $A$ needs to be convex, not just any subset. – copper.hat Jun 21 '22 at 04:27
  • @copper.hat You're right! – Analyst Jun 21 '22 at 04:39
  • 1
    It always takes me a while to work through stuff because I am a pendant. – copper.hat Jun 21 '22 at 04:40
  • @copper.hat I just remember that I asked this question about if Observation 0.7 lacks convexity assumption. – Analyst Jun 21 '22 at 04:43
  • I'm having second thoughts. The definition there is that $x \in \operatorname{a-int} E$ iff $x \in \operatorname{int}_L (E \cap L)$ for every line $L$ containing $x$. I am assuming that $\operatorname{int}_L (E \cap L) = \operatorname{ri} (E \cap L)$, the relative interior. I previously (mis)interpreted it as $x \in \operatorname{a-int} E$ iff for all $y$, there is some interval $I$ containing $0$ in its interior such that $x+ty \in E $ for all $t \in I$. I am not really sure I see the point of the definition in the PDF at the moment. – copper.hat Jun 21 '22 at 05:01
  • 1
    On a little more reflection it seems that $\operatorname{int}_L (E \cap L)$ is not $ \operatorname{ri} (E \cap L)$, but the interior of $E \cap L$ relative to $L$. So it coincides with my other interpretation. Pedantic emergency averted. – copper.hat Jun 21 '22 at 05:06
  • I do not follow the last theorem in the PDF you mentioned above. The author mention lsc. & $X$ Banach but I have no idea how that factors into the terse proof. – copper.hat Jun 22 '22 at 06:36
  • 1
    @copper.hat If $f:X \to \mathbb R$ is convex l.s.c. and $X$ a Banach space, then $f$ is continuous. – Analyst Jun 22 '22 at 07:20
  • Do you have a reference for that result by any chance, all my convex books are finite dimensional. – copper.hat Jun 22 '22 at 20:17
  • @copper.hat Please see here. – Analyst Jun 22 '22 at 23:22
  • Thanks! ${}{}{}$ – copper.hat Jun 23 '22 at 01:09

0 Answers0