Why results are worse with a finer mesh
 

Hi all,

I'm solving a natural convection with user defined viscosity in square cavity in Fluent. Fluent solves the problem with no issues on a 40x40 orthogonal mesh (mapped+edge sizing) and the results are valid, however when I make the mesh 80x80 with the same settings; firstly, after almost a week of run time (for comparison 40x40 took 5hrs to converge) the residuals doesn't seem to converge to the same residual that the 40x40 model was converged to. Secondly, I stopped the iterations but the results are not valid (e.g. should be 1e-3 but it's 1e-4).
can you please help me with the possibilities that are causing this issue. I expect that when I increase the grids not only the solution should converge but also it has to have more accurate result and the 80x80 is not a very small mesh size in the scale of my simulations.
I'm using SIMPLE algorithm and Green-Gauss node based for gradient terms and second order for other terms. an image of my residuals are attached.

thanks,
Amir

Okay you need to be looking at more than residuals. Residuals are not a measure of accuracy or validity. The result is the actual solution. You should not have any broad expectations of residuals vs grid size.

Thank you very much for your response!
ok I got your point so I shouldn't expect the same behavior of residuals for different meshes and also they are not a measure of validating my results. now the question is why the solution (comparing Nusselt number and max velocity with a benchmark) is not valid on finer mesh while every other factor is the same?