Development of closures based on machine learning for high-collisionality plasmas with multiple ion species

Fluid closures for multi-ion-species plasmas are required to describe plasma phenomena such as edge pedestal transport in tokamaks and multi-component ion flows in the Earth ionosphere. Although quantitative closures can be obtained by solving the general moment equations, analytic formulas of converged closure coefficients become impractical as the number of moments increases. Additional complexities come from combinations of mass, temperature, number density, and charge ratios and the Hall parameter as the number of ion species increases. This work aims to find fitting functions that represent the analytic closure coefficients and can be used conveniently for practical fluid problems. In order to establish the closure functions, we use machine learning for multivariate polynomial regression. Machine learning is based on the training data set constructed by the analytical formulas. The multiple ion parameters constitute the input data set and the analytic closure coefficients constitute the well-labeled output set. The fitted closure functions will be presented for two ion species and an effective-parameter method will be introduced for multiple ion species.