Article View: sci.math.research
Article #87Re: Permutation polytopes
From: Greg Kuperberg
Date: Wed, 07 Oct 1992 23:39
Date: Wed, 07 Oct 1992 23:39
39 lines
2645 bytes
2645 bytes
As I understand it Michael Cohen's question is this: Let A be the anti-symmetric n x n matrix with all 1's above the diagonal and all -1's below. Let U be the convex hull of the orbit of A under the action of conjugation by permutation matrices. What are the codimension one faces of U? I can't answer this question, but I have evidence that the answer could be very complicated. First, here are some properties of U: U actually lives in a vector subspace of A because the k'th row of a matrix in U must have the same sum as the k'th column. However, since the whole matrix adds to zero this is really only n-1 constraints, so we've identified a subspace V of dimension (n-1)(n-2)/2 that contains U. But in this subspace U has content, because the linear span of U must be a representation of the symmetric group, and V is an irreducible respresentation. Note also that U is a centrally symmetric polytope, because conjugation by the antidiagonal negates the matrix A. There is a theorem due to Milman (and perhaps Bourgain?) that the number of faces times the number of edges of a centrally symmetric polytope in k dimensions is bounded below by some C^k. Perhaps C=2 is a viable value, but the proof only derives that C is 1 plus a small number not worth worrying about. For us, however, k=n^2, and C^n^2 grows much faster than n!. There will asymptotically be a huge number of orbits of faces under the action of the permutation group. Therefore there is no a priori symmetry reason that the faces are nice. However, I do know of one example of a polytope with many more faces than symmetries which is nevertheless nice. It's the convex hull of permutation matrices which are symmetric and have zeroes on the diagonal. Do you want to do linear programming? For the polytope I have just mentioned, there is a way to do linear programming in guaranteed polynomial time where you never consider all faces at once, only those you encounter. It's possible that there is also such an algorithm for your polytope. Greg
Technical Details
Message-ID:
<9210072339.AA20805@zaphod.uchicago.edu>Path:
arc1-ca.pugleaf.net!sparky!uunet!think.com!sdd.hp.com!ux1.cso.uiuc.edu!news.cso.uiuc.edu!usenetReferences:
<97468@bu.edu>Show Full Headers
Complete Article Headers
Newsgroups: sci.math.research Path: arc1-ca.pugleaf.net!sparky!uunet!think.com!sdd.hp.com!ux1.cso.uiuc.edu!news.cso.uiuc.edu!usenet From: greg@math.uchicago.edu (Greg Kuperberg) Subject: Re: Permutation polytopes References: <97468@bu.edu> Message-ID: <9210072339.AA20805@zaphod.uchicago.edu> Sender: Daniel Grayson <dan@math.uiuc.edu> X-Submissions-To: sci-math-research@uiuc.edu Organization: University of Illinois at Urbana X-Administrivia-To: sci-math-research-request@uiuc.edu Approved: Daniel Grayson <dan@math.uiuc.edu> Date: Wed, 7 Oct 1992 23:39:36 GMT Lines: 40