# Convergence on non-uniform grid

 Register Blogs Members List Search Today's Posts Mark Forums Read

 January 29, 2005, 10:55 Convergence on non-uniform grid #1 dragon Guest   Posts: n/a hello I have a code for 3d laminar steady flow simulation, the code uses a collocated grid,and convection terms uses cds scheme,MIM used for pressure,I used the code to solve the cavity flow,I have the following problems: 1. when solving uniform grid steady flow ,the code works very well,but when I use non-uniform grid,even in a relative coarse grid ,the pressure always diverges,I wanna know why?The grid ratio is 1.2 near the wall. 2.I changed the code for unsteady simulation, 2nd runge-kutta for time marching,uniform grid this time for 20*20*20 grid the code works well,but when I refine the grid to 80*80*80 the pressure fails to converge. The converge criteria is the sum of all abs(residules) less than 1e-3.WHy? besides what is the best criteria for converge ? all simulation done on re=1000,based on lid velocity I would appreciate any comment,thanks dragon

 January 29, 2005, 13:32 Re: Convergence on non-uniform grid #2 agg Guest   Posts: n/a 1) Your code runs OK on a uniform grid. This means that you solver is ok. 2) Were you ever able to run any simulation successfully on an non-uniform grid? If not, goto STEP 3. 3) When you use non-uniform grid, one generally uses tranformation techniques (jacobians). If that is the case may be there is an error somewhere when you compute the derivatives. Also check your pressure poisson equation (if you are solving one). For a non-uniform grid there will be extra terms in the poisson equation. 4) Also use an RMS criterion for checking convergence.

 January 31, 2005, 02:23 Re: Convergence on non-uniform grid #3 Chandra Shekhar Guest   Posts: n/a On non-uniform grid, error in fluxes at opposite faces of a grid doesnt cancel out....and so, it is impossible to have any FD Scheme which is conservative as well as can maintain more than 2nd order accuracy... In your case, to get yr solution converged, try to use highr orger upwinding scheme (3rd order may be sufficient). good luck!