Revisiting the parallel tempering algorithm: High-performance computing and applications in operations research

Abstract

This study explores the parallel tempering method, an approach that has shown promising results in simulation and is suited for modern multiprocessor platforms. Though relatively unexplored in operations research, the algorithm has significant potential. The study evaluates a newly developed single-node CPU-based parallel implementation of parallel tempering in three case studies involving challenging operations research problems. The evaluation considers both the quality of solutions and response time. The study also proposes an API containing the implementation of multi-core parallel tempering to encourage its use and facilitate future implementations. Evaluation results confirm the effectiveness of parallel tempering, demonstrating significant performance against state-of-the-art methods.