As we know, a good CFD code should be grid independent, does that mean as we increase the grid points (no matter how much), the program should converge? In other words, if a program converges with 2000 time step, if we go by 5000, and it starts diverging after say iteration 3000, is there anything wrong with the code?????

What I meant by going from 2000 to 5000 iteration was after increasing the grid numbers; i.e.:

(iteration = 2000 grid# = 100,000. > converge) and (iteration = 5000 grid# = 100,000. > converge)

AND ALSO:

(iteration = 2000 grid# = 300,000. > converge)

BUT:

(iteration = 5000 grid# = 300,000. > DIVERGE!!!!!!)

I don't know what you're trying to compute, but a coarse grid will tend to be dissipative and have a stabilizing effect on the computed flow field. Perhaps this is part of what you're seeing.

"Grid independent" means that as you refine the grid, the solution should stop changing. The change in solution from a coarse grid to successively finer grids should decrease as you go from one grid to the next.

I agree with "cfd dude" that grid independence should refer to the solution and not the code.

With regard to instability under grid refinement, for incompressible flow there are finite element pairs that do not satisfy the inf-sup condition and the condition number of the matrix deteriorates under refinement, making it increasingly difficult to find a solution. Refinement doesn't always result in improvement of the solution.

CFd dude - Thanks.

My solution doesn't change by itertion # 2000, but as it goes to higher iterations, difference bewteen solutions from fine and coarse grids will increase so at iteration 5000, there is a huge difference.

Also, for the case of fine grids, after iteration 2000, the program starts diverging, which this does not happen when I use a coarse mesh.

What do you think?

Again it's hard to say without knowing more about your case, your numerical scheme, and what type of flow it is. But two points to think about:

1) With finer meshes, your time step will be smaller for the same CFL number. So are you reaching the same solution at the same physical time on both grids by iteration 2000? And at 5000?

2) It's possible that the coarse mesh may be damping an instability in your solution due to numerical dissipation (since the unstable feature is not resolved), so you only see the divergence on the finer mesh.

It is possible that your are capturing some new phenomena with the fine grid and that cause the problems.

However, it is more likely that your finest grid is "too fine" somewhere in the domain so that the precision of your solver is insufficient. The numerical inaccuracies cause the solution to diverge. Been there, done that. More than once too

-- Jarmo