User talk:Wotao.yin

From Wikimization

Revision as of 03:43, 5 November 2010 by Dattorro (Talk | contribs)
Jump to: navigation, search

I regard the following as a difficult problem, having spent considerable time with it.

LaTeX: \begin{array}{cl}\mbox{minimize}_X&c^{\rm T}\mbox{vec}\,X\\ \mbox{subject to}&A\,\mbox{vec}\,X=b\\ &X^{\rm T\!}X=I\\ &X\geq_{}\mathbf{0}\end{array}

Vector LaTeX: c\, is left unspecified beause I may want to vary it later in a convex iteration. For your purposes, it may arbitrarily be set to LaTeX: \mathbf{0} or LaTeX: \mathbf{1}.

Rectangular submatrix LaTeX: \,X\!\in\mathbb{R}^{1024\times256}\, comes from a square permutation matrix LaTeX: \,\Xi\!\in\mathbb{R}^{1024\times1024}\, having three out of every consecutive four columns discarded.

Matrix LaTeX: A\in\mathbb{R}^{10565\times262144} is sparse having only 979,444 nonzeros. It contains integers from the set LaTeX: \{-1,0,1,2\}.

Vector LaTeX: b\, is quite sparse having only a single nonzero entry.

A good presolver can eliminate about 60,000 columns of LaTeX: A\, because one of the constraints (fifth row from the bottom) has all nonnegative entries. This means that about 60,000 entries in permutation submatrix LaTeX: X\, can be set to zero before solution begins.

--Dattorro 03:31, 5 November 2010 (PDT)

Retrieved from "http://www.convexoptimization.com/wikimization/index.php/User_talk:Wotao.yin"
Personal tools