Development of a Method of Direct Numerical Modeling of Turbulent Flows Using a Supercomputer

Abstract

A method of direct numerical modeling of turbulent flows in the elements of nuclear power setups using a supercomputer is being developed at the Institute of Problems in the Safe Development of Nuclear Energy. The method is based on developed low numerical diffusion algorithms for which discrete approximations are constructed using finite-volume and completely separated grid methods. A regularized nonlinear monotonic operator separation scheme has been developed for solving the advection problem. Richardson’s iteration method with a preconditioner (i.e., a special matrix, premultiplication by which improves the convergence of the numerical method) in the form of the fast Fourier transform for the Laplace operator is used to solve the pressure equation. This approach to solving elliptic equations with variable coefficients gives multifold acceleration compared with the ordinary method of conjugate gradients. Quasi-direct numerical modeling is used to model the three-dimensional turbulent flows in single-phase flows. The method is implemented in the CONV-3D code, which is parallelized and efficient on the multiprocessor cluster computers Chebyshev and Lomonosov (Moscow State University).