Thanks for the answer: we have discovered an error in our formulation. The huge coefficients were caused by an error in our code which produces the ilp formulation due to erroneous use of unsigned values.
Best regards Marco Lattuada On Wed, 13 Jun 2007 14:02:55 +0400 Andrew Makhorin <[EMAIL PROTECTED]> wrote: >> we have found an ilp formulation (file glpk.mps) on >>which >> glpsol with default options finds a solution >> (the produced output is in default.txt) and on which >> advanced MIP solver doesn't find (output: intopt.txt). >> Is it a bug of glpk or are we doing something wrong? > > Thank you for your report. > > This is not a bug. The difference appears due to huge >coefficients > in some constraints, for example: > > R0001162: + C0000011 + 4294967295 C0000140 + C0000516 - >C0000527 > <= 4294967295 > > Most probably such huge coefficients introduce excessive >round-off > errors during the preprocessing phase. Note that the >successful > solution of your instance with the standard b&b was a >happy chance. > > The "big M" method is not a good modeling technique, >because it makes > the problem too hard to be solved in floating-point >arithmetic. > > > Andrew Makhorin > _______________________________________________ Bug-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-glpk
