Residuals
 

hi Friends

I am trying to use mass residual as a criterion for grid sensitivity analysis. Unfortunately when I double the number of cells of my simulation I get higher mass residual, I expected to get lower mass residual. Is that means that the simulation with smaller number of cells are more accuarte than that of higher number of cells? Is there any one who can expalin this?

regards

mohaba

Re: Residuals
 

Hi,
This is very interesting. Sometimes people think exactly as you do, but usually you don't know why this happen. It could be several things. When you double the number of cells in your grid, in certain regions you might see some situation that you were not able to see in a coarse grid, such as recirculations, stagnation zones, etc. Differencing scheme is another thing. When you use a 1st order scheme in a coarse grid, you might expect a very low level of the residuals. However when you use a fine grid (5x more cells than the coarse one) you can expect false diffusion due the discretization scheme. Well, we can discuss this for weeks. Let's see what the other colleagues of this forum say.
Regards, cfd guy

Re: Residuals
 

Hi Mohaba,

Comparing mass residual is really not a good way of testing grid sensitivity since it is a criteria for convergence. You should really compare a global quantity like a total force, drag, pressure drop or something of that sort.

Robin

Re: Residuals
 

Hi Mohaba,

you're confusing two distinct concepts: error (which should decrease as the mesh is refined) with residuals (how well the discretized equations are being solved). Residuals tell you if you've obtained a converged solution on a particular mesh, but tell you nothing at all about the solution error on that mesh. For that information you need to do a mesh refinement study tracking some relevant quantity... Robin has given some possibilities.

Phil

Re: Residuals
 

Hi Phil/Robin,

How do you do the mesh refinement study to check for the error!? I get mixed up too, because i would like to see that my discretized equations are solved well before i can rely on the refined messh results.

Please guide me here: I do get small mass residuals (more convergence) with a coarse mesh and less convergence with a refined mesh. How do I check for error?

regards oduor

Re: Residuals
 

Hello, i think that your original mesh was not good enough. The first mesh should be dense enough to capture any time dependent effects like flow separation, shock. Having a denser grid allows you to see these effects. I don't think that false diffusion is a problem here as you have a fine grid, but if it is the case ..it will disappear with increased mesh density. So, if i were you i would drop the course mesh case and use the finer grid as my coarse mesh, then start refining from there. You should be careful as your refinement should be solution adaptive oherwise you might mess up the simulation. Hope it helps!