10.03.2015 Views

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

v2009.01.01 - 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.

4.6. CARDINALITY AND RANK CONSTRAINT EXAMPLES 311<br />

and<br />

minimize<br />

W ∈ S 2n+1 〈G ⋆ , W 〉<br />

subject to 0 ≼ W ≼ I<br />

trW = 2n<br />

(708)<br />

which has an optimal solution W that is known in closed form (p.599). These<br />

two problems are iterated until convergence and a rank-1 G matrix is found.<br />

A good initial value for direction matrix W is 0. Optimal Q ⋆ equals [x ⋆ y ⋆ ].<br />

Numerically, this Procrustes problem is easy to solve; a solution seems<br />

always to be found in one or few iterations. This problem formulation is<br />

extensible, of course, to orthogonal (square) matrices Q . <br />

4.6.0.0.3 Example. Combinatorial Procrustes problem.<br />

In case A,B∈ R n , when vector A = ΞB is known to be a permutation of<br />

vector B , solution to orthogonal Procrustes problem<br />

minimize ‖A − XB‖ F<br />

X∈R n×n<br />

(1593)<br />

subject to X T = X −1<br />

is not necessarily a permutation matrix Ξ even though an optimal objective<br />

value of 0 is found by the known analytical solution (C.3). The simplest<br />

method of solution finds permutation matrix X ⋆ = Ξ simply by sorting<br />

vector B with respect to A .<br />

Instead of sorting, we design two different convex problems each of whose<br />

optimal solution is a permutation matrix: one design is based on rank<br />

constraint, the other on cardinality. Because permutation matrices are sparse<br />

by definition, we depart from a traditional Procrustes problem by instead<br />

demanding a vector 1-norm which is known to produce solutions more sparse<br />

than Frobenius norm.<br />

There are two principal facts exploited by the first convex iteration design<br />

(4.4.1) we propose. Permutation matrices Ξ constitute:<br />

1) the set of all nonnegative orthogonal matrices,<br />

2) all points extreme to the polyhedron (91) of doubly stochastic matrices.<br />

That means:

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

Saved successfully!

Ooh no, something went wrong!