Global sensitivity analysis on numerical solver parameters of Particle-In-Cell models in particle accelerator systems

Every computer model depends on numerical input parameters that are chosen according to mostly conservative but rigorous or empirical estimates. These could for example be the step size time integrators, a seed pseudo-random number generators, threshold of grid points discretize computational domain. In case is enhanced with new algorithms and modelling techniques influence quantities interest, running as well accuracy often initially unknown. Usually trial-and-error basis neglecting cost versus aspects. As consequence per simulation might unnecessarily high which wastes computing resources. Hence, it essential identify most critical analyse systematically their effect result in order minimize time-to-solution without losing significantly accuracy. Relevant identified by global sensitivity studies where Sobol’ indices common measures. sensitivities obtained surrogate models based polynomial chaos expansion. this paper, we first introduce general methods uncertainty quantification. We then demonstrate use solver reduce costs discuss further improvements analysis. The evaluated neighbouring bunch simulations existing PSI Injector II Ring proposed DAEδALUS cyclotron rf electron gun Argonne Wakefield Accelerator.