A Breeder Algorithm for Stellarator Optimization
Abstract
An optimization algorithm that combines the global parameter space search properties of a genetic algorithm (GA) with the local parameter search properties of a LevenbergMarquardt (LM) algorithm is described. Optimization algorithms used in the design of stellarator configurations are often classified as either global (such as GA and differential evolution algorithm) or local (such as LM). While nonlinear leastsquares methods such as LM are effective at minimizing a costfunction based on desirable plasma properties such as quasisymmetry and ballooning stability, whether or not this is a local or global minimum is unknown. The advantage of evolutionary algorithms such as GA is that they search a wider range of parameter space and are not susceptible to getting stuck in a local minimum of the cost function. Their disadvantage is that in some cases the evolutionary algorithms are ineffective at finding a minimum state. Here, we describe the initial development of the Breeder Algorithm (BA). BA consists of a genetic algorithm outer loop with an inner loop in which each generation is refined using a LM step. Initial results for a quasipoloidal stellarator optimization will be presented, along with a comparison to existing optimization algorithms.
 Publication:

APS Division of Plasma Physics Meeting Abstracts
 Pub Date:
 October 2003
 Bibcode:
 2003APS..DPPRP1025W