OPTIMIZATION BY HIERARCHICAL MUTANT PRODUCTION

Inspired by the successful description of the first steps of molecular evolution by the quasispecies theory and the successful application o f quasispecies-like algorithms to optimization problems, we propose a hierarchically organized algorithm. This new algorithm is able to solv e a spin glass and a travelling salesman problem using only point muta tions. Furthermore, it performs better under comparable circumstances than the ordinary quasispecies algorithm. Depending on the structure o f the fitness landscape of the examined problem under consideration th e hierarchically organized algorithm proves to be much more suitable t han a simple quasispecies algorithm, especially in clustered landscape s. Tuning the error rates reveals the critical minimum copy fidelity n ecessary to guarantee optimization. We propose to incorporate hierarch ical concepts into optimization algorithms inspired by biological evol ution, such as genetic algorithms.