HiGHS reports infeasible for LP models that Gurobi solve successfully

[siddharth-krishna]

HiGHS v1.10.0 reports that the following two LP models are infeasible, but Gurobi v12.0.0 is able to solve them successfully: https://storage.googleapis.com/solver-benchmarks/TIMES-GEO_E4SMA_Base_scenariocplex.lp https://storage.googleapis.com/solver-benchmarks/TIMES-GEO_E4SMA_NetZero_scenariocplex.lp

Output log:

LP   TIMES-GEO-global-base-31-20ts has 6027253 rows; 6449851 cols; 36624474 nonzeros
Coefficient ranges:
  Matrix [1e-07, 3e+05]
  Cost   [1e+00, 1e+00]
  Bound  [1e-07, 9e+08]
  RHS    [1e-08, 5e+06]
Presolving model
Problem status detected on presolve: Infeasible
Model name          : TIMES-GEO-global-base-31-20ts
Model status        : Infeasible
Objective value     :  0.0000000000e+00
HiGHS run time      :          3.88
Writing the solution to /solver-benchmark/runner/solutions/TIMES-GEO-global-base-31-20ts-highs-1.10.0.sol

Any idea what the issue could be? Thank you.

[jajhall]

I don't get spurious infeasibility for TIMES-GEO_E4SMA_Base_scenariocplex.lp, but I guess that's because the logging you give is for TIMES-GEO-global-base-31-20ts, not TIMES-GEO_E4SMA_Base_scenariocplex.lp or TIMES-GEO_E4SMA_NetZero_scenariocplex.lp

[jajhall]

Now trying to get https://storage.googleapis.com/solver-benchmarks/TIMES-GEO-global-base-31-20ts.lp

[siddharth-krishna]

Thanks for looking into it. Does Highs v 1.10.0 solve TIMES-GEO_E4SMA_Base_scenariocplex.lp successfully for you?

Our benchmark runner downloads the LP/MPS files from cloud storage and renames it to a common <name>-<size> format before running: https://github.com/open-energy-transition/solver-benchmark/blob/8ae4d3ab208c6f69d26abdcee9a46e34139a1968/results/metadata.yaml#L2121-L2155

So TIMES-GEO-global-base-31-20ts is the same as https://storage.googleapis.com/solver-benchmarks/TIMES-GEO_E4SMA_Base_scenariocplex.lp

[jajhall]

Thanks for looking into it. Does Highs v 1.10.0 solve TIMES-GEO_E4SMA_Base_scenariocplex.lp successfully for you?

I've not tried the interior point solver, having decided to see what the PDLP solver makes of it on a GPU

[jajhall]

Our benchmark runner downloads the LP/MPS files from cloud storage and renames it to a common <name>-<size> format before running:

Goodness, how opaque!