About smagorinsky_sfs_term

[bronise777]

In the pc_smagorinsky_sfs_term function, sigma is solved in different directions.
const amrex::Real sigmadx = Cs2 * alphaij[0] - third * CI * alpha * static_castamrex::Real(dir == 0);
const amrex::Real sigmady = Cs2 * alphaij[1] - third * CI * alpha * static_castamrex::Real(dir == 1);
const amrex::Real sigmadz = Cs2 * alphaij[2] - third * CI * alpha * static_castamrex::Real(dir == 2);
But in the pc_dynamic_smagorinsky_sfs_term function, sigma is a little different, why subtract CI*alpha/ 3 only in the x direction?
const amrex::Real sigmadx = coeff(iv, comp_Cs2) * alphaij(iv, 0) - coeff(iv, comp_CI) * alpha(iv) / 3.0;
const amrex::Real sigmady = coeff(iv, comp_Cs2) * alphaij(iv, 1);
const amrex::Real sigmadz = coeff(iv, comp_Cs2) * alphaij(iv, 2);

[marchdf]

I think you have indeed identified another bug. Thank you for taking the time to go through this code so carefully and raising issues when you encounter them. It is much appreciated. As you can tell, the dynamic Smagorinsky LES has not received much attention since it was implemented more than 5 years ago. I hope this is the last of them. PR created for this issue in #849

[bronise777]

Have you calculated with the corrected model? In the case of my Taylor-Green vortex, the corrected calculation appears to be less good than the pre-modified one. The following figures are the results before and after correction respectively, using the case of PeleLMeX script drawing pictures
before
after

[marchdf]

(Just a note, you are using PeleC, right? your labels and messages indicate PeleLMeX but this is the PeleC repo)

Something does look strange. I pored over the code again and I do think you did indeed find a bug (the one that is fixed). It is possible that, in fixing the bug, we made an other more subtle bug more obvious. I remember now that my main issue when implementing all this was that I couldn't figure out a good unit test for this. I guess we are now facing this issue. Since you've been pouring over the equations much more recently than I have, maybe you could help out with this? In an ideal world, we could set up a known initial condition and compute the expected sfs term manually/some python script and compare to what we get out of PeleC.