I have a lp problem that gives me numerical instability on the standard
simplex method in glpk. I've been able to solve it using another solver
that I compiled with long doubles, but I would prefer to use glpk if I can
make it work for this.
I tried following the recommendations in the manual for using glp_exact,
but the computation is still taking too long and I'm not sure I'm doing this correctly.
In the section for the glp_exact function, the manual reads:
"Computations in exact arithmetic are very time-consuming, so solving LP
problem with the routine glp_exact from the very beginning is not a good
idea. It is much better at first to find an optimal basis with the routine
glp_simplex and only then to call glp_exact, in which case only a few
simplex iterations need to be performed in exact arithmetic."
And here is what I ran (my problem is in the CPLEX lp format):
glpsol --lp problem.lp -w inexact-solution.txt
glpsol --exact --ini inexact-solution.txt --lp problem.lp
Am I doing this properly? If I am, then I would try to do this using long
doubles. Would the developer(s) welcome a contribution of support for long
doubles to the source code?
Many thanks and regards,