LBM

[AliKarakus]

DG / Lattice-Boltzmann solver with basic functionality.

[codecov]

Codecov Report

Merging #71 (5f734db) into master (d5a931a) will not change coverage.
The diff coverage is n/a.

❗ Current head 5f734db differs from pull request most recent head 40b5863. Consider uploading reports for the commit 40b5863 to get more accurate results

@@           Coverage Diff           @@
##           master      #71   +/-   ##
=======================================
  Coverage   70.22%   70.22%           
=======================================
  Files          12       12           
  Lines         131      131           
=======================================
  Hits           92       92           
  Misses         39       39           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d5a931a...40b5863. Read the comment docs.

[tcew]

LGTM.

Can you elaborate a bit on the difference to bns.

[AliKarakus]

LGTM.

Can you elaborate a bit on the difference to bns.

Both solvers govern the same physics and expecting to achieve similar results for low Mach flows. I think there are couple points of using DG- LBM, ​
--> it is very straight-forward to extend to different physics i.e. heat transfer, natural convection, multi-phase (I am working on connecting with Level-Set method for sharp interfaces) and multi-relaxation.
​--> comparing with BNS especially in boundary layers.
--> possible publications with extending to deforming meshes, pml, multi-phase, etc.

[noelchalmers]

This looks really good to me. A couple questions/comments:

• I think it's a good idea to make the default final time t=10 in the setup.rcs so that the runs produce several frames.
• Same comment for the testing runs. Otherwise only a small number of timesteps are actually performed. (Changing this will require updated 'golden' norms).
• I suspect there is a subtle bug in the Quad2D code path. Running the default setup to t=10 doesn't pass the 'eyeball' norm. The Gaussian doesn't seem to move at all.

I see a several places where early PML and/or Multirate support is enabled/commented. @AliKarakus is the plan to add these features in the future? They're not hurting anything so no worries either way.

I also did some quick profiling on large problem setups. Probably unsurprisingly, several kernels are register heavy, the volume and phaseField kernels in particular are often at <30% peak DRAM BW. Not a problem, just an FYI.

I second @tcew 's comment. I'm not particularly familiar with LBM, so it would be cool to see the full formulation written somewhere!

Thanks @AliKarakus !

[noelchalmers]

@AliKarakus @tcew These test runs are going to be bitten by the latest OCCA changes complaining about @barrier("local"). Let ignore these test fails for the moment.

Should we make occa a git submodule to avoid this kind of thing in the future?

[tcew]

Should we make occa a git submodule to avoid this kind of thing in the future?

Yes. I think this is a very good and timely idea.

I anticipate quite a few changes in occa and we do not want to be continually held up.

[AliKarakus]

@noelchalmers Thanks for the feedback, I double checked all the tests on my local machine. Everything works smoothly.

--> I see a several places where early PML and/or Multirate support is enabled/commented. @AliKarakus is the plan to add these features in the future? They're not hurting anything so no worries either way.

I am going to work on a PML formulation, but multi-rate time-stepping creates some issues because of the time-step size dependent properties in LBM. Need to check details a little more. Currently my target is to combine this basic DG-LBM formulation with Level-Set for bubble dynamics and contact line problems.

--->I also did some quick profiling on large problem setups. Probably unsurprisingly, several kernels are register heavy, the volume and phaseField kernels in particular are often at <30% peak DRAM BW. Not a problem, just an FYI.

Yes, I think this problem is arising from the high number of fields i.e. 9 for 2D and 15 or 19 for 3D. At the moment, I don't know how to overcome this performance issue quickly as BNS outperforms LBM even in 2D test cases.

--->I'm not particularly familiar with LBM, so it would be cool to see the full formulation written somewhere!
It will be available soon (hopefully) with DG-LBM/LS paper.

Thanks a lot @noelchalmers and @tcew .

[noelchalmers]

@AliKarakus Can you merge the latest master branch? That should fix our OCCA issues and re-run the CI checks.

[tcew]

@AliKarakus @noelchalmers - the Lbs tests seem to take quite a while. In particular the tet 3d case seems to take 20 minutes unless I misread the test report.

[noelchalmers]

LGTM. The whole suite of tests take around 24 minutes.

[tcew]

LGTM. The whole suite of tests take around 24 minutes.

@noelchalmers would you like to do the honors and merge this at your convenience ?

[tcew]

Great work @AliKarakus and thanks to @noelchalmers for overseeing the merge !

[AliKarakus]

Thanks @tcew and @noelchalmers. I am deleting the lbm branch now.