0
$\begingroup$

Is the set of convex polygons dense in the set of convex domains in $\mathbb{R}^2$, for the Banach-Mazur distance?

Any insight for a negative or positive answer is very much welcome!

$\endgroup$
7
  • $\begingroup$ The BM distance is usually defined as a distance between normed spaces. What is the distance between convex subsets of $\mathbb R^2$ ? $\endgroup$ Commented Dec 25, 2023 at 9:27
  • 1
    $\begingroup$ But you can also be more rigorous as: K,L convex bodies, then $d_{BM}(K,L)=inf\{c\in \mathbb{R}_{>0}|K\subset L\subset cK\}$. $\endgroup$
    – kvicente
    Commented Dec 25, 2023 at 9:34
  • 1
    $\begingroup$ Yeah i think, you refer to the original definition, however, is equivalent to this last one I gave you (which I personally like more). See for example Corollary 2.5 in universiteitleiden.nl/binaries/content/assets/science/mi/… $\endgroup$
    – kvicente
    Commented Dec 25, 2023 at 9:47
  • 1
    $\begingroup$ The paper you give requires $L=T(K)$ for some invertible linear map $T$, and that $T$ need not be identity, so the definition you give is wrong. (Another way to see its failure is to consider the case that $K$ and $L$ do not contain each other) $\endgroup$
    – Zerox
    Commented Dec 25, 2023 at 12:02
  • 1
    $\begingroup$ And also, I think you should replace "convex domain" by "closed (or open) convex neighbourhood of the origin". $\endgroup$
    – Zerox
    Commented Dec 25, 2023 at 12:04

1 Answer 1

1
$\begingroup$

Let $U\subset \mathbb R^2$ be a bounded open convex neighborhood of $0$. For every $\epsilon>0$ we have $\displaystyle \overline U=\bigcap_{H\supset U\atop H \text{ closed half-plane}} H \subset (1+\epsilon)U$, therefore $\displaystyle \bigcap_{H\supset U\atop H \text{ closed half-plane}} H \setminus (1+\epsilon)U=\emptyset.$ Hence, by compactness, the intersection is already empty on a finite sub-family of these half-planes $H_1,\dots, H_m$, so $\displaystyle U\subset \bigcap_{1\le j\le m} H_j \subset (1+\epsilon)U$, which exhibits a convex domain bounded by a polygon close to $U$.

[edit] I think that given $\epsilon_0>0$ one can produce another positive $\epsilon<\epsilon_0$ and polygon whose boundary is inscribed to $(1+\epsilon)\partial U$ and circumscribed to $\partial U$, by the construction sketched here below:

Fix a point $x_0\in\partial U$. Form a sequence (depending on $\epsilon$, $y_k=y_k(\epsilon)$, for $k\ge0$, of points on $(1+\epsilon)\partial U $, starting from $y_0:=(1+\epsilon)x_0$, and defining inductively $y_{k+1}\in (1+\epsilon)\partial U$ as the point on the line from $y_k$ which is tangent to $\partial U$ and such that $(y_{k+1}-y_k)\cdot y_k>0$ (this is to fix a sense of rotation). The key point is that for each $k$ the point $y_k$ depends continuously from $\epsilon$. Call $m_\epsilon$ the first index $m$ such that the corresponding polygonal paths with vertices $y_0(\epsilon),\dots y_m(\epsilon)$ cross itself. Since $m_\epsilon\to\infty$ as $\epsilon\to0$, by continuity there is $\epsilon <\epsilon_0$ such that $y_{m_{\epsilon_0}}(\epsilon)= y_0(\epsilon)$, which means that the polygon with vertices $y_k(\epsilon)$ is circumscribed to $\partial U$ and inscribed to $(1+\epsilon)\partial U$.

$\endgroup$
7
  • $\begingroup$ I think your answer is right. However I have a questions before accepting it. Can we show that this polygon is inscribed in the boundary of $(1+\varepsilon)U$ and circumscribed in the boundary of $U$ at the same time? $\endgroup$
    – kvicente
    Commented Dec 25, 2023 at 12:45
  • $\begingroup$ Actually not by the above argument; it just produces a convex polygon between $U$ and $(1+\epsilon)U$, whose boundary could even be disjoint from both $\partial U$ and $(1+\epsilon)U$. This argument can be repeated in any dimension though. $\endgroup$ Commented Dec 25, 2023 at 14:00
  • $\begingroup$ However I think one can produce a polygon as you say with a possibly smaller $\epsilon$: I’ve added a sketch of the construction. $\endgroup$ Commented Dec 25, 2023 at 14:42
  • $\begingroup$ This is good @pietromajer. However I don't see clearly why the $\varepsilon<\varepsilon_0$ is, as a matter of fact, a positive number. I mean, why the closing of the produced polygon happens strictly before $\varepsilon=0$? Can you hopefully elaborate a little bit more in this detail. Thanks very much for taking the time with this btw. $\endgroup$
    – kvicente
    Commented Dec 26, 2023 at 7:17
  • $\begingroup$ The reason is that this would happen exactly in case $U$ itself is a polygon. If it is not a polygon, more and more edges are needed for a complete turn by a polygonal between $\partial U$ and $(1+\epsilon)\partial U$, as $\epsilon\to0$. $\endgroup$ Commented Dec 26, 2023 at 7:44

You must log in to answer this question.

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