Hi folks,

I am wondering if anybody knows of issues (in a finite volume formulation) with making dx=dy, but dz=something else. dx will be the same along the entire x axis, as will dy and dz along their respective axes, but dz will simply be a larger number, like 5*dx. I don't see any reason that this should cause issues, but want to know what issues I may need to anticipate.

I think the code should still be conservative as all finite volume codes are. But can this possibly cause other issues like numerical diffusion? Slow down convergence?

I can't find any references on this. The closest thing I see is discussion of how changing grid spacing ALONG an axis will lower the order of a method (i.e. - central difference becomes O(dx) not O(dx^2) because the errors do not cancel in a nonuniform grid). The only thing I can see is that it will probably take longer to converge (I am using the SIMPLER algorithm).

Any help or references would be appreciated. Thanks!

Best, Jon

Hi there,

you may to check for publication dealing with the impact of cell aspect ratio. From my experience, the greater the AR the slower the convergence.

Now, just something to think about: if your dx, dy and dz are all constant, then you may want to investigate Spectral methods instead of FV.

Sincerely,

Fred Felten