Estudo de parâmetros do método Multigrid geométrico para equações 2D em CFD e volumes finitos

Abstract

Resumo: A influencia de alguns parametros do metodo multigrid geometrico sobre o tempo de CPU para tres diferentes modelos matematicos bidimensionais do escopo da CFD (Computational Fluid Dynamics) e investigada. Os modelos matematicos sao: a equacao de Laplace, a equacao de Adveccao-Difusao e as Equacoes de Burgers. Os parametros em estudo sao: numero de iteracoes internas do solver (ƒË); numero de malhas (L); numero de incognitas (N); solvers e operadores de prolongacao. O multigrid e empregado com esquema FAS (Full Approximation Scheme) e tecnica FMG (Full Multigrid) com ciclo V e razao de engrossamento r = 2. As equacoes diferenciais sao discretizadas pelo Metodo dos Volumes Finitos (MVF) em geometrias simples e malhas bidimensionais uniformes por direcao, com aproximacoes de 2a ordem CDS e correcao adiada. As condicoes de contorno, do tipo Dirichlet, sao aplicadas mediante a tecnica de volumes ficticios. Os sistemas de equacoes algebricas sao resolvidos com o emprego do solver Gauss-Seidel Lexicografico (GS-Lex) e, no caso do problema de Burgers, tambem com o emprego do Gauss-Seidel red-black (GS-RB). Verificou-se principalmente que: o esquema FAS-FMG e cerca de duas vezes mais rapido do que o FAS padrao; que o numero de equacoes ou complexidade do problema nao interfere na eficiencia do multigrid; que o operador de prolongacao bilinear e o mais eficiente para interpolar as solucoes entre os niveis do FMG.

This work investigates the influence of some parameters from the Multigrid Geometric method over CPU processing time for three different mathematical bidimensional methods that make up the Computational Fluid Dynamics scope. These mathematical models are: Laplace equation, Advection-Diffusion equation and Burgers' equations. In order to achieve the main target, which consists on optimize the employed algorithms to solve the problems above, the computational time minimization is sought through parameters modifications at the algorithms. The considered parameters are: number of solver's internal iteration (v); number of grids (L); number of incognites (N); solvers and prolongation operators. The multigrid is employed besides FAS (Full Approximation Scheme) and FMG (Full Multigrid) technique, with V cycle and coarsening ratio r = 2. The differential equations discretization is made by the Finite Volume Method (MVF) over simple geometries and direction uniform bidimensional grids, with second order CDS and delayed correction. The Dirichlet type boundary conditions are applied through fictitious volume technique. The system of algebraic equations are solved by the Gauss-Seidel Lexicographic (GS-Lex) solver and, at the Burgers problem, the Gauss-Seidel red-black (GS-RB) is also employed. The main results that should be emphasized are: the FAS-FMG scheme is about twice faster than the standard FAS; the multigrid efficiency ate not affected by the number of equations or complexity of the problem; the bilinear prolongation operator is the most efficient to interpolate the solution among the FMG levels.