Re: [Help-glpk] [Fwd: Glpsol Changing Tolerance]

[Andrew Makhorin]

> I'm having a problem with using glpsol to solve an MIP problem using the
> default tolerances. Using the MIP variant of my problem glpsol find an
> integral optimal solution; however when I take that solution and solve
> the problem again with the integer variables converted to parameters the
> solution is vastly different. I have had this issue before with
> commercial solvers and have solved it by tightening the integrality
> tolerance to 0. Unfortunately I cannot determine a way to do this with
> glpsol. Any assistance would be appreciated.
> 

You need to change the control parameter iocp.tol_int. You cannot do it
from the command-line, however, you may patch the glpsol source code.
Take file glpk/src/glpapi21.c, find the call to glp_inopt, and add a
line above it as follows:

csa->iocp.tol_int = 1e-20;
glp_intopt(csa->prob, &csa->iocp);

Note that tol_int should be positive, i.e. zero is not allowed.