The current point hops over a bumpy 1-D function. Worse moves are accepted with probability exp(-Δ/T); T cools over time. Best-so-far is tracked separately.
| Quantity | Cost | Note |
|---|---|---|
| Per iteration | O(1) | incremental Δ + one exp |
| Whole run | O(I · c) | I iterations, c per-move cost |
| Memory | O(state) | hold current + best only |
| Global-opt guarantee | log cooling | Tₕ = c/log(k+2), impractical |