1

Recently, a post addressed the classic interesting question of showing that the projection onto a closed convex part of $\mathbb R^2$ is contracting. Unfortunately, the person did not provide any trace of personal research. Nevertheless, the question interested me and I wanted to find old posts that dealt with it. I found a very interesting post with figures and a real effort to make things clear and meaningful.

But there are two points that bother me in the post in question.

  • First, I corrected typos that did not seem to have been noticed despite the many readings of this post.

  • Let $A$ a convex of $\mathbb R^2, x,y \in \mathbb R^2, p_A(x) $ and $p_A(y) $ their projections on $A$, $v= p_A(y)-p_A(x)$. It is not proven that the segment $[x,y]$ crosses the two lines orthogonal to $v$ and passing through $p_A(x)$ and $p_A(y)$.

So I am trying to prove it. I've already proven that $$\boxed{\forall z\in A, \langle z-p_A(x),x-p_A(x)\rangle\le 0}\color{red}{(*)}$$ by following Schneider's hints.

Similarly I obtained that$$\forall z\in A, \langle z-p_A(y) ,y-p_A(y) \rangle \le 0$$

I know that the line orthogonal to $v$ and passing through $p_A(x)$ $$\{\xi\in\mathbb R^2: \langle \xi-p_A(x) ,v \rangle=0\}$$divides $\mathbb R^2 $ into two half-planes $$\boxed{\{\xi\in\mathbb R^2: \langle \xi-p_A(x) ,v \rangle\le 0\}}\color{green}{(**)}$$ and $$\{\xi\in\mathbb R^2: \langle \xi-p_A(x) ,v \rangle\ge 0\}$$

I see that I can replace $z$ by $p_A(y)$ in $\color{red}{(*)}$ and $\xi$ by $x$ in $\color{green}{(**)}$ to place $x$ in one of the half-planes.


But afterwards, I admit that it becomes a headache for me to finish the demonstration expected by Schneider all by myself. I hope this very specific question is not a dupe and that it will deserve your attention so that we really finish this demonstration of an important fact if it has not already been done on MSE.


enter image description here

0 Answers0