Memetic Algorithms for molecular conformation and other optimization problems

With the name of`Memetic Algorithms' we recognize a class of metaheuristics which constitutes an emerging paradigm for optimization. For more than a decade now it has been applied with success to a large variety of combinatorial and nonlinear problems. In the last four years, and mostly in the last two, several techniques introduced in molecular optimization problems can be characterized in this way. We try to just introduce here the new techniques while we seek to emphasize the relation with previous and current work in Memetic Algorithms and Scatter Search. Memetic Algorithms (MAs) is a population-based approach to optimization 28]. It can be applied both to nonlinear and discrete (combinatorial) optimization problems. They are orders of magnitude faster than Genetic Algorithms in some problem domains. We call it a `metaheuristic' since it is a general purpose strategy that guides other basic heuristics or truncated exact methods. Other metaheuristics, like Simulated Annealing (SA) or Guided Local Search (GLS) (which guide some underlying \Hill-climbing" procedure), are applied to a single \optimizer" and thus they can not be categorized as`population-based'. Basic Tabu Search (TS) techniques also use a single optimizer, but more evolved metaheuristics like Scatter Search (SS), also use procedures based in \multiple-point" optimization. Genetic Algorithms (GAs), also try to evolve solutions in connguration space by using a \population" of individuals which represent alternative solutions of the problems. Good reviews stressing the similarities and diierences of these approaches can be found in 10] 11] and 28] (the latter, written in 1990-92, anticipates the relevance of MAs in protein landscapes). A main diierence between GAs and MAs is that the latter approach tries to use all possible knowledge available to solve the problem. Some SS procedures can be be classiied as MAs too, though none of the approaches properly includes the other. In essence, MAs constitute an exercise of humbleness and common sense; if several alternative algorithms for the problem already exist, it should be a wise idea to use them together. This is of particular interest for the practitioner , who has been relying on a well-known code which enables to give locally optimal solutions but requires some extra global optimization (adding stochasticity to the search). Thus MAs are sometimes called \Hybrid Genetic Algorithms", or \Lamarkian Evolutionary Algorithms" or even \knowledge-augmented GAs". Other denominations include \Genetic Local Search" or \Parallel Genetic Algorithms". They are easy to be recognized from the so-called \standard …