MIPLIB 2017 Infeas
A benchmark of primal- and dual-infeasible linear programs

MIPLIB 2017 Infeas

334 linear programs with no optimal solution, each derived from a MIPLIB 2017 instance with its integrality dropped. 229 have no feasible point; 105 are unbounded. They run up to 20,682,487 variables and 19,912,111 constraints.

Scatter of every published instance, variables against constraints on logarithmic axes, coloured by class, with the mark area growing with the nonzero count. Scatter of every published instance, variables against constraints on logarithmic axes, coloured by class, with the mark area growing with the nonzero count.
Every published instance, on logarithmic axes. The mark area grows with the nonzero count. Infeasible instances keep the aspect ratio of the source model; unbounded ones are far wider than they are tall, because relaxing removes constraints and leaves the variables in place.

Primal-infeasible

229 instances

A branch-and-bound run splits on a variable and explores one side of the split. This set collects the other side, at nodes where that side has no feasible point.

Browse them all →

Dual-infeasible

105 instances

Of those, 103 come from removing an inclusion-minimal set of bounds and constraints from a feasible, bounded root relaxation; the other 2 were already unbounded.

Browse them all →
Every instance

Two files

<name>.mps.gz

The linear program, gzipped MPS, continuous variables only. Nothing to reconstruct before handing it to a solver.

<name>.meta.json

How that program was derived from its source instance: the recorded branch for a primal-infeasible file; for a dual-infeasible file, either the bounds and constraints removed or that the source relaxation was already unbounded.

Citing

How to cite

@misc{miplib2017_infeas,
  author       = {Haihao Lu and Bo Tang},
  title        = {MIPLIB 2017 Infeas: Primal- and Dual-Infeasible Linear Programs from MIPLIB 2017},
  howpublished = {\url{https://mit-lu-lab.github.io/MIPLIB2017-Infeas/}},
  year         = {2026}
}

Every instance derives from MIPLIB 2017. Please also cite Gleixner et al., “MIPLIB 2017: Data-Driven Compilation of the 6th Mixed-Integer Programming Library”.

Contact

This collection is maintained by Haihao Lu and Bo Tang. If a published file does not have the status it claims, or a source instance carries terms other than the default, please open an issue and the affected file will be corrected or withdrawn.