A fast but sometimes unreliable optimization method. When searching for the minimum/maximum value of a function a random step is taken; if the value improves it replaces the current value, then another random step is taken. This method is fast and relatively easy to program but does not allow backtracking and therefore can become trapped on local minima/maxima in the search space.

A heuristic variation uses an evaluation function to examine and select the best successor from the current position. This produces a faster ascent through the problem space.