12.07.2015 Views

v2010.10.26 - Convex Optimization

v2010.10.26 - Convex Optimization

v2010.10.26 - Convex Optimization

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.4. RANK-CONSTRAINED SEMIDEFINITE PROGRAM 325which is a convex relaxation of the desired equality constraint[ ] [ ] I X I [I X ]X T =G X T (938)The rank constraint insures this equality holds, by Theorem A.4.0.1.3, thusrestricting solution to R n . Assuming full-rank solution (list) Xconvex equivalent problem statementrankZ = rankG = rankX (751)Problem statement (750) is nonconvex because of the rank constraint. Wedo not eliminate or ignore the rank constraint; rather, we find a convex wayto enforce it: for 0 < n < Nminimize 〈Z , W 〉G∈S N , X∈R n×Nsubject to d ij ≤ 〈G , (e i − e j )(e i − e j ) T 〉 ≤ d ij ∀(i,j)∈ I〈G , e i e T i 〉 = ‖ˇx i ‖ 2 , i = N − m + 1... N〈G , (e i e T j + e j e T i )/2〉 = ˇx T i ˇx j , i < j , ∀i,j∈{N − m + 1... N}X(:, N − m + 1:N) = [ ˇx N−m+1 · · · ˇx N ][ ] I XZ =X T≽ 0 (752)G<strong>Convex</strong> function trZ is a well-known heuristic whose sole purpose is torepresent convex envelope of rankZ . (7.2.2.1) In this convex optimizationproblem (752), a semidefinite program, we substitute a vector inner-productobjective function for trace;trZ = 〈Z , I 〉 ← 〈Z , W 〉 (753)a generalization of the trace heuristic for minimizing convex envelope of rank,where W ∈ S N+n+ is constant with respect to (752). Matrix W is normal toa hyperplane in S N+n minimized over a convex feasible set specified by theconstraints in (752). Matrix W is chosen so −W points in direction of rank-nfeasible solutions G . For properly chosen W , problem (752) becomes anequivalent to (750). Thus the purpose of vector inner-product objective (753)is to locate a rank-n feasible Gram matrix assumed existent on the boundaryof positive semidefinite cone S N + , as explained beginning in4.4.1; how tochoose direction vector W is explained there and in what follows:

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!