Memetic algorithms: the polynomial local search complexity theory perspective

In previous work we developed a syntax-only classification of evolutionary algorithms, in particular so-called memetic algorithms (MAs). When ‘syntactic sugar’ is added to our model, we are able to investigate the polynomial local search (PLS) complexity of memetic algorithms. In this paper we show the PLS-completeness of whole classes of problems that occur when memetic algorithms are applied to the travelling salesman problem using a range of mutation, crossover and local search operators. Our PLS-completeness results shed light on the worst case behaviour that can be expected of a memetic algorithm under these circumstances. Moreover, we point out in this paper that memetic algorithms for graph partitioning and maximum network flow (both with important practical applications) also give rise to PLS-complete problems.