On the behaviour of the (1,λ)-es for a simple constrained problem
Dirk V. Arnold · 2011
We study the behaviour of the (1,λ)-ES for a linear problem with a single linear constraint. The algorithm produces offspring until λ feasible candidate solutions have been generated and selects the best of those as the next generation's parent. Integral expressions that describe the strategy's one-generation behaviour are developed and used in a simple zeroth order model for the steady state of the strategy. Implications for the performance of cumulative step size adaptation are discussed and a comparison with the (1+1)-ES is drawn.