Modifying Cell Aspect Ratio

[JhonCordova]

Hi, Everyone

I noticed that the cell aspect ratio is always set to 1 (ΔX=ΔY=ΔZ​). Is there any way to adjust this so that ΔX≠ΔY≠ΔZ​? If not, how hard would it be to implement such a feature?

Thanks in advance!

[baperry2]

Right now, ΔX=ΔY=ΔZ​ is a firm requirement in PeleC.

It would definitely be useful to relax this for many problems. That's an idea we've thrown around but I'm not aware of anyone having gone through thoroughly to figure out what would actually be required. There are definitely some places in the code that are already set up to allow differing grid spacing for each direction, but there probably are other places where the shortcut of treating them as equal has been used. I'm not sure if allowing differing grid spacing would merely be an issue of going through and ensuring proper bookkeeping is done everywhere for ΔX≠ΔY≠ΔZ​ (probably not trivial to do, but feasible), or if there would be more major roadblocks anywhere (e.g. using capability from AMReX that would not support this).

@marchdf do you remember why the limitation of ΔX=ΔY=ΔZ​ was originally enforced? Do you know of any major roadblocks that would prevent this from being relaxed?

[marchdf]

I think it was a thing to do with EB. I wouldn't be totally surprised if non-EB mostly just worked without that. There's probably just a couple of places where that assumption is explicitly made.

[baperry2]

Yeah, it having something to do with EB rings a bell for me as well.

@JhonCordova, I'd recommend starting with a non-EB case, remove the error for ΔX≠ΔY≠ΔZ​, and then try a case with non-unity aspect ratio and see what (if anything) breaks. You can turn off different types of physics using these flags to see what works and what doesn't:

pelec.do_hydro = 1
pelec.do_react = 1
pelec.diffuse_temp=1
pelec.diffuse_enth=1
pelec.diffuse_spec=1
pelec.diffuse_vel=1

[drummerdoc]

It would be a bug if it required equal cell sizes in all direction without EB. I fully expect that it should work. We should modify the assertion to enforce this only if EB is being actually used.....my $.02

[JhonCordova]

Thanks. It is working

[drummerdoc]

Excellent. Is there any test changes we should push to formalize this? Thanks for checking back @JhonCordova

[JhonCordova]

The simulation is running for the channel flow case, but I still need to include some modifications, such as a variable external force to maintain the Reynolds number. As soon as I obtain results that can be compared with DNS data, I will let you know.

[baperry2]

Still a bit worried that there's possibly something buried where we don't account for different dx in different directions. That would likely cause a silent issue where the simulation runs to completion but potentially gives a bad result. So the validation comparisons will be important.

[yjh124]

Hello, I wonder if the problem has been solved? How was it solved? I have encountered the same problem. Thank you very much.

[baperry2]

The isotropic grid size restriction has been lifted for non-EB simulations (#964) but remains in place for EB simulations despite some progress in #967. Further development in that direction is not planned but contributions would be welcome.

[yjh124]

Thanks very much for the informative reply and for sharing the PR links. I'll note the current status and look into it further.