Inverse Solution for Parameter Estimation of Computer Simulation by an Empirical Bayesian Code Tuning Method

This article presents an empirical Bayesian code tuning method based on a Gaussian process model for estimating adjustable theory parameters in a complex computer simulation code by using both computer simulation data and real experimental data. Some parameters of the metamodel are estimated from the data by the maximum likelihood method, and those estimates are then used to obtain the maximum a posterior estimate of theory parameters. Four transport parameters of the theoretical nuclear fusion model are estimated by applying this method to computational nuclear fusion devices (tokamak). The approximated standard errors of estimates are obtained by using the Fisher information matrix. The posterior probability of a parameter is computed to test a hypothesis about the parameter.