search

ERGO-Code / HiGHS

Linear optimization software
MIT License
913 stars 170 forks source link

HiGHS Falsely Declares a MILP as Provably Optimal #937

Open jeffreydeankelly2 opened 2 years ago

jeffreydeankelly2 commented 2 years ago

Attached are the LP and MPS files for a MILP problem which has a provably optimal objective function of > -14.0 but HiGHS declares it as provably optimal at -18.2.

Even when the HiGHS relative and absolute gap tolerances are set to zero (0.0), the behavior still exists. It also occurs when the HiGHS presolve is turned "off" as well.

In addition, when an integer-feasible solution with an objective function > -18.2 is found by another MILP solver and all of its binary variables at one (1) are to fixed to one (1), then HiGHS correctly solves and finds the same objective function value.

CTAP-IMF.lp.txt CTAP-IMF.mps.txt

lgottwald commented 2 years ago

The lp and mps files seem to differ. The mps file produces positive objectives only and the .lp file produces objectives that negative as you report. Which version of HiGHS are you using?

jajhall commented 2 years ago

The only objective coefficient is for the last variable. It's +1 in the .lp file (with maximization), and -1 in the .mps file (without an OBJSENSE MAX section).