Stochastic tunneling

Stochastic tunneling ( STUN) is a method of global optimization, in which is sampled at minimizing function with the Monte Carlo method.

Principle

Optimization algorithms based on the Monte Carlo method, scan the investigated function by jumping them at random from the current solution to another solution, the function values ​​differ by. The probability of such an attempt jump is usually chosen as the defined Metropolis criterion, wherein the parameter is selected appropriately.

The principle of STUN is to circumvent the slow dynamics of unfavorable shaped energy functions that are encountered for example in spin glasses, in that such barriers are tunneled. This goal is achieved by the Monte Carlo sampling of the transformed function that is not subject to this slow dynamics. The transformation is in the "standard form " defined by, the previous lowest function value found is. This transformation preserves the loci of the minima. The effect of such a transformation is shown in the figure.

An adaptive version of the method allows itself to estimate the parameters of the method "online" and thereby improve the efficiency of the algorithm.

Other approaches

• Simulated annealing
• Parallel tempering
• Evolutionary algorithm
• Differential evolution