Race — eXchange — Improve
Status: research prototype · preprint in preparation · release TBD
ReXi is a parallel solver for mixed-integer programming, built on SCIP and SoPlex, focused on quickly finding strong primal solutions.
ReXi is a private research project by Gioni Mexi and Daniel Rehfeldt.
ReXi includes customized and optimized versions of SCIP and SoPlex. We intend to contribute these changes back to the SCIP project for inclusion in a future release. ReXi gratefully builds on the work of the current and previous SCIP developers at Zuse Institute Berlin and beyond.
MIPfeas benchmark
ReXi currently leads among the non-commercial solvers on the public MIPfeas benchmark by Michael Bussieck / Hans Mittelmann.
On the MIPLIB 2017 feasibility benchmark (233 instances, 600 s timelimit), ReXi reaches a geometric mean primal integral of 0.0165. The aggregated commercial-solver reference VMCS reaches 0.0105; the next-best-performing non-commercial solver reaches 0.0383.
Geometric mean primal integral · lower is better
ReXi found feasible solutions for 223 / 233 instances and proved optimality for 110.
ReXi and the other CPU solvers ran on Intel Core Ultra 9 285K desktop machines; the NVIDIA cuOpt entries ran on B200 and GB10 GPU systems.
Citation: a preprint is in preparation. If you'd like to get more information on ReXi or if you have questions in the meantime, please get in touch.