Group: sci.op-research
From: hmobahi@gmail.com
Date: Friday, February 15, 2008 3:44 PM
Subject: Plz recommend a fast and robust QP solver

Hello,

I am trying to solve a *convex* quadratic problem iin Matlab
environment. There is only a single equality constraints and all
variables are bounded.
I have tried a few solvers, but none was satisfactory.

1. cvx : very slow (probably because it is too general)

2. quadprog : stops with exit flag set to "Maximum Number of
Iterations Reached". I even set this parameter as large as 100000 and
still get the same error.

3.BPMPD_MEX2.21.1 : Doesn't seem to work correctly. I tried
[x,y,s,w,how]=bp([],ones(1,n),1,ones(1,n),0,1:n,zeros(1,n),
1:n,ones(1,n)) which means given n variables bounded between 0 and
one, minmize their sum of squares, when the sum of vars is equal to
one. It sets how to 'infeasible primal', which is incorrect (the
solution must be x(i)=1/n)

Any recommendation for a good *free* solver?

Thanks

H.M.

Safety Articles | Usenet Groups | Usenet News | Bluegrass