While testing some changes I've been finding different objective values for map10 (https://miplib.zib.de/instance_details_map10.html) Given this is a MIPLIB 2017 benchmark instance we can't blame the instance
On c4012f564 (some recent latest commit) a valid optimal solution test.sol.txt gets separated by HighsTableauSeparator when computing a lifted binary cover cut. I can't seem to find any numerical error that would cause this from the cut logic. The tableau row (aggregation) before separation is violated by some tiny amount 1e-17, but that shouldn't necessarily cause the bug..... I'm putting it here to track the issue and because I've failed to fix it today.
Warning: It's a bit annoying to recreate because the size of the problem and wait time.
While testing some changes I've been finding different objective values for
map10(https://miplib.zib.de/instance_details_map10.html) Given this is a MIPLIB 2017 benchmark instance we can't blame the instanceOn
c4012f564(some recentlatestcommit) a valid optimal solution test.sol.txt gets separated byHighsTableauSeparatorwhen computing a lifted binary cover cut. I can't seem to find any numerical error that would cause this from the cut logic. The tableau row (aggregation) before separation is violated by some tiny amount1e-17, but that shouldn't necessarily cause the bug..... I'm putting it here to track the issue and because I've failed to fix it today.Warning: It's a bit annoying to recreate because the size of the problem and wait time.