gatesgh~asc.hpc.mil

The fast messy genetic algorithm (fmGA) belongs to a class of algorithms inspired by the principles of evolution, known appropriately as "evolutionary algo-rithms " (EAs). These techniques operate by applying biologically-inspired operators, such as recombination, mutation, and selection, to a population of individuals. EAs are frequently applied as optimum seeking tech-niques, by way of analogy to the principle of "survival of the fittest. " In contrast to many EAs, the fmGA con-sists of several evolutionary phases, each with distinct characteristics of local/global computation. These are explained in the paper. 1998 ACM 0-89791-969-6/98/0002 3B6 Previous scalability analyses of island-model EAs have been based on either fixed global population size or fixed subpopulation size. Recently developed popu-lation sizing theory enables scalability analysis based on fixed expected solution quality. Parallel computational experiments are performed to determine the effectiveness and efficiency of an MPI-based fmGA using each of these scaling techniques. The optimization problem for these experiments is the min-imization of the CHARMM energy model of the pen-tapeptide [Met]-Enkephalin.