slow/non-convergence of qp() on convex problems - is it to be expected?

[Joshua Redstone]

Followup: I've poked around more and it looks like this is a bug in qp().I've
sent an email describing the problem in detail and included a '.dat' file to
reproduce it.
The email sent to bug-octave has subject line: qp() in octave-3.0.0 gets
stuck on wrong answer to convex problem
Josh

On Tue, Jun 3, 2008 at 8:52 AM, Joshua Redstone <redstone at gmail.com> wrote:

> Hi all,I'm having difficulty getting qp() to find the global minimum of
> the particular class of convex problems I'm trying to solve.
> The success of qp() in finding the global solution seems, even when
> executing for thousands of iterations, seems to be very sensitive to the
> initial value x0 specified.
> I'm wondering if this is to be expected, and what I can do to make my
> problem more amenable to qp()'s algorithm.
>
> Details:
> I modified qp() to take an extra parameter, maxiter, specifying the maximum
> number of iterations.
> For a particular problem instance, I find that even with maxiter set at
> 3000, I have to call qp() many times with different
> values of x0 before it will return info=0 - which means 'The problem is
> feasible and convex.  Global solution found.'
> The other times, qp() returns info=3 - which means 'Maximum number of
> iterations reached.'
> Furthermore, when qp() returns info=0, the objective function value is
> 1e-28, while, when it returns info=3, the
> objective function value is more like 1e-2.  Is this to be expected?  Is
> there a way I could modify either my problem or choose another
> convex solver besides qp() that might have better luck?
> The class of convex problems I'm trying to solve is of dimension 100, and
> there are two equality constraints and six inequality constraints.
> Any suggestions?
> I can attach an octave file with an instance of this behavior if it would
> help.
> Thanks,
> Josh
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://www.cae.wisc.edu/pipermail/help-octave/attachments/20080605/ecf282fc/attachment.html