> I think I have stumbled on a glpk bug (version 4.28). > On a particular problem, I get an assertion failure:
> Assertion failed: spx->p != 0 > Error detected in file glpspx02.c at line 602 > Before getting the assertion failure, I get a numerical instability > warning > spx_simplex: warning: numerical instability (primal simplex, phase I) Thank you for the bug report. The error appears, because the glpk lp solver is not sufficiently robust and the constraint matrix has large constraint coefficients that makes it ill-conditioned leading to excessive round-off errors in computing basic solution components. > BTW, I get this message even when calling glp_simplex with GLP_DUALP. The dual simplex is used only if the current basic solution is dual feasible. > When I reverse the rows, I get PROBLEM HAS UNBOUNDED SOLUTION > This happens when I call glp_simplex from another language. > The funny thing is that when I write the problem into a file (cplex > format), > glpsol does not complain and finds an optimal value. Try to scale the instance (lpx_scale_prob) and then to construct an initial basis (lpx_adv_basis or lpx_cpx_basis) before a call to glp_simplex (of course, if you do not intend to do reoptimization). Sometimes this helps. _______________________________________________ Bug-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-glpk
