Machine based optimization using genetic algorithms in a storage ring

The genetic algorithm (GA) has been a popular technique in optimizing the design of particle accelerators. As a population based algorithm, GA requires a large number of evaluations of the objective functions, which can be time consuming. One can benefit from parallel computing with significantly reduced computing time when fulfilling the function evaluation by a numerical machine model in simulation codes. Indeed, this is the most common approach in GA applications. In this paper, instead of applying GA in the conventional numerical calculations as described above, we present a successful experimental demonstration of implementing GA in real machine based optimization. We conduct the minimization of the average vertical beam size of the SPEAR3 storage ring using GA. Beam loss rate is chosen as the sole objective function because it is inversely proportional to the vertical beam size and can be measured instantaneously in SPEAR3. The decision variables are the strengths of SPEAR3 skew quadrupoles, by varying which we can change both the betatron coupling and the vertical dispersion while searching for the minimum beam size. The results in this paper can shed light on new applications of GAs in the particle accelerator community, for example, optimizing the luminosity of a high energy collider or the injection efficiency of a diffraction limited storage ring in real time.