• 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.