Spectral element method and high order time splitting method for navies stokes equation

陈雪江,秦国良,徐忠
DOI: https://doi.org/10.3969/j.issn.1007-4708.2002.03.006
2002-01-01
Abstract:The calculational formula of the spectral element method that combines the generality of the finite element method with the accuracy of spectral techniques is induced. The spectral element method is a high order weighted residual technique. The computational domain is broken into a series of elements, and the variables such as the velocity and the pressure are represented as a high order interpolation polynomial through Chebyshev collocation points, and the element matrix is formed by finite element method. Then the system matrix is constructed by element stiffness matrix summation, and the answer is obtained by solving the linear equation set. A formulation for splitting methods is also developed that results in high order time accurate schemes for the solution of incompressible Navier Stokes equations. The normal time splitting method is improved by applying third order explicit Adams Bashforth method and fourth order Runge Kutta method to the nonlinear connective terms, and third order implicit Adams Molten method to the linear terms. And the third order pressure boundary conditions for Possion equation for the pressure is employed as well. Using be high order time splitting method and high order pressure boundary conditions the precision of the time discretization is advanced, and we can calculate the flow by longer time step to improve the efficiency of calculation. The advance of the precision for the time discretization is also useful for high Re fluid calculation. Then the high order time splitting method to split the incompressible Navier Stokes Equation, combining with the spectral element method, is utilized to simulate one typical sample in computational fluid dynamics, lid driven flow in closed square cavity. In the case of the same calculational accuracy and the same grid, the paper increases the Reynolds umber from Re=100 to Re=600, and the time of convergence is shortened by using longer time step. The results show agreement with the accepted benchmark solutions.
What problem does this paper attempt to address?