Population partitioning in genetic algorithms

When an optimisation landscape is highly multimodal, even a genetic algorithm can become stuck in local minima. Increasing the population size is one way to ensure that the search space is more comprehensively sampled. But it can be a more efficient use of population members to allow a number of subpopulations to evolve separately and then interbreed with one another. One example of such a multimodal problem is scattering from a conductive object, which is important in applications such as radar cross-section optimisation