0

$C$ is a non-empty convex set in $\mathbb{R}^n$. If $\forall y \in \mathbb{R}^n,$ there exists some $\epsilon>0$ such that $z+\epsilon y\in C$, then $z\in \rm{interior}(C)$

I cannot prove this conclusion. How to prove this theorem?

copper.hat
  • 172,524

1 Answers1

5

I assume you consider the standard metric topology in $\mathbb{R}^n$, induced by the Euclidean norm. Let $y=e_i$, $i=1,2,\dots n$, $e_i=(0,0,0,\dots 1,\dots 0)$ where the $1$ is in the $i$-th position. For each one of these you have some $\epsilon_i>0$ satisfying the given condition. Take $\epsilon=\min \epsilon_i$. By convexity, the set $z+\textrm{conv}\, \{0,\pm \epsilon e_{i_1}, \pm \epsilon e_{i_2},\dots \pm \epsilon e_{i_n}\}$ with $i_k\neq i_j$ for $k\neq j$ is contained in $C$. But the ball centered at $z$ with radius $\epsilon/\sqrt{n}$ is contained in the above set (think of the two or three dimensional cases). Therefore that ball is contained in $C$ and $z$ is interior.

GReyes
  • 16,446
  • 11
  • 16
  • Haha thanks for the proof. Your certification process is very detailed. This question is a corollary in the book "Convex Analysis", which is not explained in detail. I've been stuck with this problem all day and didn't expect this way of proof. Your proof is very ingenious and I am amazed. By the way, should I remove the 0 in the conv set and just use $z$ + conv{±1,±2,⋯±} ?Thank you very much! – Yilin Cheng May 03 '22 at 02:12
  • @YilinCheng Yes. You do not need to include $0$ (it does not hurt though..) – GReyes May 03 '22 at 03:02
  • @YilinCheng There are many books with the same title. Are you talking about Boyd-Vandenberghe? – GReyes May 03 '22 at 03:03
  • I am reading the Convex Analysis of RT Rockafellar, It's a bit difficult for me to understand this book. ๐•ᴗ•๐. @GReyes – Yilin Cheng May 03 '22 at 03:14
  • By the way, how do I use @? Can only use "@ symbol and copy paste his id after"? – Yilin Cheng May 03 '22 at 03:20
  • That is a classic. Very good book. I believe you just type @ and when you start typing the user name, it recognizes it and you just click. That is what I do at least.. – GReyes May 03 '22 at 03:23
  • @YilinCheng There is an easier to digest book by Bertsekas – GReyes May 03 '22 at 03:24
  • Ok, thanks a lot for your suggestion, I will search and read Bertsekas related books. Regarding @, I found that I typed the @ symbol along with your id, and nothing popped up. But I was able to pop up under the comments under other questions. This might be a Bug. . . Or there is something wrong with my web page. .So anyway, thank you very much – Yilin Cheng May 03 '22 at 03:32
  • The result is proved in Rockafellar's book. The essential result (indeed, perhaps the essence of convexity in $\mathbb{R}^n$ to some extent) is Theorem 6.1. – copper.hat May 03 '22 at 03:34
  • @GReyes I just refreshed the page a few times, now I can @ and the id pops up automatically..... – Yilin Cheng May 03 '22 at 03:36
  • @copper.hat Yes, but this Proposition is not proven in the book. I don't know how to use the necessary and sufficient conditions for relative interior points of Theorem 6.1 to prove this proposition – Yilin Cheng May 03 '22 at 03:40
  • @YilinCheng You are correct, excuse me. However, he proves Theorem 6.4 using Theorem 6.1, and Corollary 6.4.1 shows that $z \in \operatorname{ri} C$, and by choosing appropriate $y$s you can show that the affine hull of $C$ is $\mathbb{R}^n$. – copper.hat May 03 '22 at 03:46
  • @copper.hat OK, I haven't read the latter part yet. I'll keep reading the following theorems to see if they work for the proof. Thank you! – Yilin Cheng May 03 '22 at 04:43
  • @YilinCheng I believe that the same proof works for the more general result $z\in ri, C$. All you need is as many independent directions to move out of $z$ as the dimension of the affine hull. – GReyes May 03 '22 at 16:55
  • @GReyes Yes, HaHa. For the necessary and sufficient conditions of relative interior points, there are detailed proofs in Rockafellar's Convex Analysis book. It is based on some theorems earlier in the book. Therefore, the proof method of the relative interior point is easier to think of. – Yilin Cheng May 05 '22 at 02:27