10.03.2015 Views

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

344 CHAPTER 4. SEMIDEFINITE PROGRAMMING<br />

We find a good value of weight w ≈ 5 for randomized problems. Initial<br />

value of direction matrix W is not critical. With dimension n = 26 and<br />

number of equality constraints m = 10, convergence to a rank-ρ = 2 solution<br />

to semidefinite feasibility problem (776) is achieved for nearly all realizations<br />

in less than 10 iterations. (This finding is significant in so far as Barvinok’s<br />

Proposition 2.9.3.0.1 predicts existence of matrices having rank 4 or less in<br />

this intersection, but his upper bound can be no tighter than 3.) For small n ,<br />

stall detection is required; one numerical implementation is disclosed on<br />

Wıκımization.<br />

This diagonalization decomposition technique is extensible to other<br />

problem types, of course, including optimization problems having nontrivial<br />

objectives. Because of a greatly expanded feasible set, for example,<br />

find X ∈ S n<br />

subject to A svec X ≽ b<br />

X ≽ 0<br />

rankX ≤ ρ<br />

(785)<br />

this relaxation of (776) [195,III] is more easily solved by convex<br />

iteration rank-1. (The inequality in A remains in the constraints after<br />

diagonalization.) Because of the nonconvex nature of a rank constrained<br />

problem, more generally, there can be no proof of global convergence of convex<br />

iteration from an arbitrary initialization; although, local convergence is<br />

guaranteed by virtue of monotonically nonincreasing real objective sequences.<br />

[222,1.2] [37,1.1]

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

Saved successfully!

Ooh no, something went wrong!