Communication Dans Un Congrès Année : 2025

Enhancing the convergence of the Feasibility Pump through variable decomposition

Résumé

Feasibility Pump (FP) is a well-known heuristic for mixed integer linear programming that iteratively moves fractional solutions toward integer feasibility by minimizing their distance to target integer points while maintaining the original linear constraints. Due to its simplicity and effectiveness, FP is widely used in commercial solvers, with many variations improving its convergence and solution quality. This work introduces a novel acceleration technique relying on the intuition that not all binary variables are active in a feasible solution. A kernel search strategy identifies a subset of promising binary variables, whilethe remaining ones are grouped into ranked buckets based on their likelihood of appearing in a feasible solution. FP sub-problems prioritize the kernel and gradually incorporate other variables as needed. To our knowledge, this is the first decomposition-based approach within the FP framework. Computational experiments on MIPLIB 2017 benchmark and two additional problems in vehicle routing and scheduling show improved success rates and reduced running times. Notably, the proposed method surpasses FP’s performance and finds feasible solutions in cases where CPLEX fails within an hour.

Fichier non déposé

Dates et versions

hal-05130004 , version 1 (25-06-2025)

Identifiants

  • HAL Id : hal-05130004 , version 1

Citer

Lucas Assunção, Sebastián Urrutia, Andréa Cynthia Santos. Enhancing the convergence of the Feasibility Pump through variable decomposition. 34th European Conference on Operational Research (EURO 2025), Jun 2025, Leeds (Angleterre), United Kingdom. ⟨hal-05130004⟩
124 Consultations
0 Téléchargements

Partager

  • More