SA explores the TSP energy landscape by accepting worse solutions with probability e−ΔE/T. As temperature cools, the algorithm converges to near-optimal tours — escaping local minima via thermal fluctuations.