7
$\begingroup$

Consider a closed, bounded and convex set $C \subset \mathbb{R}^{2}$ and denote its boundary with $\partial C$. It is very well-known that the Minkowski sum of two convex sets is convex again. What about the Minkowski sum of its boundary? Is the Minkowski sum $\partial C + \partial C$ again a convex set and how can one prove that? Does this property hold in other dimensions?

$\endgroup$
4
  • $\begingroup$ @IgorBelegradek No, $\partial C +\partial C = 2C$ in that case. I believe this equality is true in general (equivalently, any point in a convex body is the midpoint of boundary points) and should be provable by a topological argument. $\endgroup$ May 7, 2020 at 19:37
  • $\begingroup$ Take $C$ to be the unit disk and $D$ to be the disk of radius, say, $1/100$. The sum of the convex sets is the disk of radius $11/10$ but the sum of the boundaries is the annulus with inner radius $9/10$ and outer radius $11/10$. $\endgroup$
    – Deane Yang
    May 7, 2020 at 20:18
  • 1
    $\begingroup$ @DeaneYang The question is about the sum of $\partial C$ with itself. $\endgroup$
    – M. Winter
    May 7, 2020 at 20:20
  • $\begingroup$ @M.Winter, oy. Thanks. $\endgroup$
    – Deane Yang
    May 7, 2020 at 20:24

2 Answers 2

11
$\begingroup$

Yes, $\partial C + \partial C$ is convex since it equals $2C$. Equivalently, every point in $z \in C$ is a midpoint of two boundary points. This is obvious if $z \in \partial C$. Otherwise, let $f :S^{n-1} \to \mathbf{R}$ be the continuous function which sends $u$ to the length of the segment going from $z$ to $\partial C$ in direction $u$. Since $n > 1$, this function takes equal values at a pair of antipodal points (a very simple corollary to Borsuk-Ulam, if you want), which gives the desired property.

$\endgroup$
3
  • 3
    $\begingroup$ If you consider segments going from $z$ to $\partial C$ in directions $u$ and $-u$, and you compare their lengths, then I think all you need here is the intermediate value theorem to find that they will be equal in length at some point (by continuously moving $u$ to $-u$). $\endgroup$
    – M. Winter
    May 7, 2020 at 20:19
  • $\begingroup$ @GuillaumeAubrun Although it seems intuitively correct, can you give an argument why f is continuous? $\endgroup$
    – Hermann
    May 8, 2020 at 0:02
  • 1
    $\begingroup$ This follows e.g. from the fact that the gauge function of a convex body is continuous. The gauge function $g_K$ of a convex body $K$ is defined on $\mathbf{R}^n$ by $g_K(x) = \inf \{ t \geq 0 \, : \, x \in tK \}$ ; it is convex hence continuous. When $z=0$ (which you can assume), our $f$ is the restriction of $1/g_C$ to $S^{n-1}$. $\endgroup$ May 8, 2020 at 7:12
-1
$\begingroup$

Convex hull of Minkowski sum is the Minkowski sum of convex hulls. The proof is Theorem 1.1.2 in

Schneider, Rolf, Convex bodies: the Brunn-Minkowski theory, Encyclopedia of Mathematics and Its Applications. 44. Cambridge: Cambridge University Press. xiii, 490 p. (1993). ZBL0798.52001.

...But the answer to the question is: NO. See the picture in https://www.geometrie.tuwien.ac.at/peternell/ms_paper_v1.pdf (Peternell, Minkowski sum of boundary surfaces...) Can't find a citation.

$\endgroup$
6
  • 1
    $\begingroup$ I don't see how this answers the question. Also, the desired property is false in dimension $1$, so this restriction should enter at some point. $\endgroup$ May 7, 2020 at 20:02
  • $\begingroup$ @GuillaumeAubrun Fair point. $\endgroup$
    – Igor Rivin
    May 7, 2020 at 20:06
  • $\begingroup$ @GuillaumeAubrun but see the edit. $\endgroup$
    – Igor Rivin
    May 7, 2020 at 20:10
  • $\begingroup$ @GuillaumeAubrun And thanks for the downvote! $\endgroup$
    – Igor Rivin
    May 7, 2020 at 21:09
  • 1
    $\begingroup$ I did note downvote your answer and I'm sorry that you believe this! I think the picture in the paper you mention addresses a different question, namely whether $\partial C + \partial D$ is convex. $\endgroup$ May 7, 2020 at 21:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.