Optimization speed and fair sets of functions

The Sharpened No Free Lunch theorem states that all optimization algorithms have the same performance on sets of functions that are closed under permutation, independently of the considered performance measure. However, not all performance measures are informative on how fast or how accurately an algorithm can solve a given problem. In this paper we focus on a particular performance measure, called optimization speed, that quantifies how fast a search algorithm is able to find an optimal solution, and we try to characterize the set of functions on which all possible search algorithms have the same optimization speed. We call fair these sets, and we prove some results about their structure, the number of such sets and the computational complexity of checking fairness.