> I have tested some more with gurobi, both raising and lowering the > primal feasibility tolerance (range: 1e-2,1e-9) but it finds the > optimal solution of 2.445336666667e+05 regardless. > What is the default primal feasibility tolerance of glpk? I can't seem > to find any info regarding the primal feasibility tolerance in the > manual. > And is there no parameter to change it?
The relative primal feasibility tolerance is specified by tol_bnd, which is a member of the struct smcp passed to the glp_simplex api routine (for more details please see the glpk reference manual included in the distribution). The default value is 1e-7. This parameter is available only on api level and cannot be changed on using glpsol. BTW, the glpk exact simplex solver (--exact) also reports primal infeasibility, so I see only two reasons why the results may be wrong: i) your instance is almost infeasible and is highly sensitive to round-off errors in input data; ii) something wrong in mps file. To check the latter reason you can convert your mps file to cplex lp format (glpsol foo.mps --wlp foo.lp --check) and try to solve foo.lp with gurobi rather than foo.mps. _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
