• C. Lingwood, G. Burt, K. Gunn
  Cockcroft Institute, Lancaster University, Lancaster
• J.D.A. Smith
  Tech-X, Boulder, Colorado

The design of RF cavities is a multivariate multi-objective problem. Manual optimisation is poorly suited to this class of investigation, and the use of numerical methods results in a non-differentiable problem. Thus the only reliable optimisation algorithms employ heuristic methods. Using an evolutionary algorithm guided by Pareto ranking methods, a crab cavity design can be optimised for transverse voltage (V_T) while maintaining acceptable surface fields and the correct operating frequency. Evolutionary algorithms are an example of a parallel meta-heuristic search technique inspired by natural evolution. They allow complex, epistatic (non-linear) and multimodal (multiple optima and/or sub-optima) optimization problems to be efficiently explored. Using the concept of domination the solutions can be ordered into Pareto fronts. The first of which contains a set of cavity designs for which no one objective (e.g. the transverse voltage) can be improved without decrementing other objectives.