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

This thesis revisits the Parallel Tempering (PT) algorithm and presents a novel CPU-based parallel implementation designed explicitly for Operations Research (OR) problems. This implementation utilizes a dataflow-driven parallel programming model to enhance performance and scalability. The study introduces a general-purpose, publicly available API that facilitates customizable components and efficient parallel execution, enabling the application of PT to complex combinatorial problems represented as permutations. The algorithm was validated through three challenging case studies: the uniform job sequencing and tool switching problem (SSP), the identical parallel machines with tooling constraints (IPMTC), and the resource-constrained parallel machine scheduling (RCPMS).  In each case, PT achieved competitive or superior results compared to state-of-the-art methods, with improvements of up to 42\% in solution quality and reductions in execution times of up to 93\%.  This research led to three publications in international journals, including one in the high-impact ACM Computing Surveys, as well as three national conference papers.