Re: [Help-glpk] moving interior point solution to simplex solver

[Michael Hennebry]

On Thu, 15 Jul 2004, Neil Burch wrote:

>   I am trying to solve a number of ill conditioned LPs.  lpx_interior
> is definitely faster than lpx_simplex, but often terminates early with
> a numerical instability error.  However, the values at termination, if
> I just go ahead and get the row primal values, are often very close to
> the solution lpx_simplex returns.  Is there any fast way to take those
> values and use the simplex method to clean it up (ie other than just
> starting from scratch with lpx_simplex?)

You might try using dual values and complementary slackness.
A nonzero dual value indicates that the corresponding primal
constraint is tight.
That should vastly reduce the dimensionality of the simplex problem.
Absent degeneracy and numerical error, the simplex problem will
reduce to a set of linear equations.

My recollection is that the interior point solver does not produce
dual values, so you will need to solve the dual problem directly.

-- 
Mike   address@hidden
"Nothing says it like words if you know how to use them."
                    --  the Professional Organization of English Majors