Local Search

Local search repeatedly changes part of a candidate solution—such as swapping TSP edges or recoloring graph vertices—to find an improvement within a chosen neighborhood. Researchers study how neighborhood design, problem structure, and landscape analysis affect the quality of local optima, while combining local search with methods such as differential evolution, Bayesian optimization, and learned or diffusion-based solvers. Evaluations span combinatorial benchmarks, numerical optimization, clustering, and routing, with attention to both solution quality and runtime. A central tension is how to reach stronger solutions or certify local optimality without making each search step too expensive or getting trapped in poor local optima.