Simulated Annealing

When to use classical SA

Use solver="sa" when your problem has fewer than ~500 variables and lowest-latency results matter more than solution quality improvement from multiple runs. Classical SA starts in milliseconds versus ~1 second GPU initialization overhead.

It's also the right choice for reproducibility-critical workflows: with a fixed seed, classical SA produces identical results across every run, making it easier to debug QUBO formulations without introducing stochastic noise.

Usage

import nerox

client = nerox.Client()

job = client.optimize.qubo(
    Q=Q_matrix,
    solver="sa",           # classical simulated annealing
    n_sweeps=100000,       # more sweeps compensate for single chain
    seed=0,                # fully reproducible
)

result = job.wait()
print(result.solution, result.objective)

vs. GPU Annealing