simulated-annealing

Simulated annealing optimizes continuous multimodal objective functions using a physics-inspired approach with Metropolis acceptance. You provide an objective function and variable bounds, and it performs a stochastic search to find near-optimal solutions for black-box optimization tasks.

Simulated annealing escapes local optima by using a Metropolis acceptance criterion that probabilistically accepts worse solutions during the search. It supports Gaussian, Uniform, and Cauchy neighborhood generation along with multiple cooling schedules to navigate complex landscapes.

Yes, this approach is suitable for parameter tuning, engineering design, and other black-box optimization tasks. It handles continuous, non-convex functions where gradients are unreliable, using adaptive cooling schedules and optional constraint handling.

The best way to set up global optimization is to instantiate the optimizer with your objective function and variable bounds, then call the optimize method. This executes the stochastic search across the defined solution space and returns the best solution found.

Yes, the simulated annealing approach supports optional constraint handling alongside variable bounds. You define the search space boundaries before optimization, and the algorithm respects these limits while generating neighbors during the stochastic search.

Simulated annealing is designed for continuous multimodal objective functions and relies on stochastic search rather than gradients. Its effectiveness depends on selecting appropriate cooling schedules and neighbor generation methods, making it less suitable for discrete or purely combinatorial problems.