Figure 10.2.2.1 Monte Carlo simulated annealing

From a random starting point, a random move is made on one of the dimensions of the search space; in this case, a residue/rotamer choice (a). The score (energy) of this new candidate E_new solution is compared to the previous step E_old. If the score is better, the move is always accepted (b). If it is worse, the move may be accepted if the Metropolis criterion is satisfied (c). This allows climbing out of local minima if the barrier is not large. This method is called simulated annealing because the simulation temperature T is high at the start of the run, and cools down (see simulation temperature section). This permits larger uphill moves early on; near the end of the run, only smaller moves are permitted, preventing escape from an optimal well. After mc_convergence iterations without improvement, the run is deemed to have converged. Multiple runs are required to prevent convergence to local minima (d).