search

ERGO-Code / HiGHS

Linear optimization software
MIT License
976 stars 180 forks source link

Solution objective worse than Primal bound #1570

Open jajhall opened 9 months ago

jajhall commented 9 months ago

Discussed in https://github.com/ERGO-Code/HiGHS/discussions/1568

Originally posted by **lpierezan** January 9, 2024 Hello everyone. I'm running the solver through the highspy (v 1.5.3) interface providing a MIP model in mps format. The solution found has an objective function worse than the primal bound. Is this a bug or am I missing some conceptual detail? Thanks! Detail: this only happens when I force mip_rel_gap to 0, which may not be recommended. ![image](https://github.com/ERGO-Code/HiGHS/assets/856658/d81cf726-a54a-45eb-abda-139279fd2536) I changed it to txt to be able to upload: [model_26.txt](https://github.com/ERGO-Code/HiGHS/files/13875712/model_26.txt)
jajhall commented 9 months ago

I've reproduced this with v1.5.3, and it also occurs with the most recent release (v1.6.0) but it doesn't occur with our "latest" branch. Although I've got some memory of this issue having been raised before, the fact that it solves OK with latest, doesn't mean it's due to a fix. It could be due to other changes in the MIP solver - there have been several bug fixes.

I'll see whether I've got time to go back to see why this failed in 1.6.0, and keep the issue open