7
$\begingroup$

This is the isoperimetric type question. We know that in $\mathbb{R}^d$, balls are the sets that minimize the isoperimetric ratio $\frac{S^{d}}{V^{d-1}}$, where $S$ is the surface area and $V$ is the volumn.

Now consider the polytopes with $f$ faces. Lindelof's theorem says, among all proper convex polytopes in $\mathbb{R}^d$ with given exterior normals of the facets, it is precisely the polytopes circumscribed to a ball that have minimum isoperimetric quotient. This theorem can be found in http://link.springer.com/book/10.1007%2F978-3-540-71133-9, page 308, Theorem 18.4.

However, on Page 309, the author made a Corollary 18.2 that among all proper convex polytopes in $\mathbb{R}^d$ with a given number of facets, there are polytopes with minimum isoperimetric quotient and these polytopes are circumscribed to a ball.

Now my question is, I think the two claims above are different. To prove the Corollary 18.2, one has to prove the existence of the polytopes minimizing the isoperimetric constant among all polytopes circumscribed to a unit ball. I searched a lot of references, but I didn't find any proofs of such an existence. Is this an obvious result?

$\endgroup$
2
  • 5
    $\begingroup$ It follows since the space of configurations of $n$ unit vectors is compact. $\endgroup$ Jul 3, 2015 at 20:44
  • $\begingroup$ @Anton Petrunin, I'm confused because the $f$ unit exterior normal vectors are given in Lindelof's Theorem, but in the Corollary, only the number of faces are given. Maybe I didn't understand your comment. What's the meaning of "the space of configurations of n unit vectors "? $\endgroup$
    – student
    Jul 3, 2015 at 21:25

1 Answer 1

3
$\begingroup$

This is an explanation of Anton's comment. For each set of $f$ unit vectors, one finds the polytope minimizing the isoperimetric quotient. The set of configurations of $f$ (not necessarily distinct) unit vectors is compact (it is homeomorphic to $\mathbb{S}^{(d-1) \times f}$), and the isoperimetric quotient is continuous thereon (this requires an argument), therefore it achieves its minimum for some set of $f$ unit vectors. These vectors might not be distinct a priori, but it is easy to see that counting a face with multiplicity (which is what "not distinct" means) does not help you. So, the faces are distinct, and by Lindelof the minimizer is circumscribed.

$\endgroup$
4
  • $\begingroup$ Could you explain "counting a face with multiplicity does not help you"? I'm sorry I didn't understand... If $\frac{S^d}{V^{d-1}}=G(v_1, \dots \, , v_f)$ for some continuous function $G$, then what does it mean if the minimum is attained at a point of which two components are the same? $\endgroup$
    – student
    Jul 4, 2015 at 0:51
  • $\begingroup$ @student The volume is the same as when you would count the face one, but the area is bigger. $\endgroup$
    – Igor Rivin
    Jul 4, 2015 at 1:08
  • $\begingroup$ Sorry I still didn't get it. If $\min_{(v_1, \dots \, ,v_f) \in \mathbb{S}^{(n-1) \times f}}\frac{S(v_1, \dots \, ,v_f)^d}{V(v_1, \dots \, ,v_f)^{d-1}}$ is attained at $(v_1, v_1, v_2, \dots \, ,v_{f-1})$, then this means the minimizer is polytope with $f-1$ faces with outer normals $v_1, \dots \, ,v_{f-1}$, but if then how to conclude this is not the minimizer? My main confusion is, I don't know the correspondence of the $(v_1, v_1, \dots \, ,v_{f-1})$ to the geometry of the convex body. It seems that you interpreted it in a different way I explained. $\endgroup$
    – student
    Jul 4, 2015 at 1:26
  • $\begingroup$ I interpreted it as a polytope with $f-1$ faces with outer normals $v_1, \dots \, ,v_{f-1}$, but you suggested it should mean counting the multiplicity of the face with $v_1$ vector. Maybe in order to understand the argument, I need to know the expression of the isoperimetric ratio function? $\endgroup$
    – student
    Jul 4, 2015 at 1:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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