Spalart-Allmaras 2D Axisymmetric Source Term Question

[mskim0401]

Hi.
I'm trying to run 2D axisymmetric case using RANS SA and SST model.
I ran my case using SU2 version 5 that has no viscous axisymmetric source term.
Now I use latest version where the issues are resolved.

In the issue #1106 , @jtneedels and @WallyMaier ran verification cases using SA model below:

But in my case, SA model in latest version show as like version 5 results (no axisymeetric source term).
When I see the source code, however, it seems that SST model axisymmetric source term exists, but SA model is not.
SU2_CFD/src/numerics/turbulent/turb_sources.cpp

Can't I run axisymmetric case using SA model?
I'm confused this. I really appreciate it if someone help me.

Thanks.

[WallyMaier]

Hi, you should be able to run the Axisymmetric and SA. What type of case are you running/can you provide your cfg file?

I believe the SST model requires more source terms (seen in turb_sources) to fully capture the relaxation effects. @FlorianDm introduced the SST source terms, so he may be a better resource on this case.

[WallyMaier]

@FlorianDm, @ALL. Please disregard my previous posts, I was mistaken.

[FlorianDm]

@bigfooted This is indeed confusing so I checked it a bunch of times when I did it and now again and I really think it is correct. The signs are the same as those of the normal turbulent sources when you compare the production terms:

SU2/SU2_CFD/src/numerics/turbulent/turb_sources.cpp

Lines 884 to 885 in 1f92147

	Residual[0] += pk*Volume;
	Residual[1] += pw*Volume;

SU2/SU2_CFD/include/numerics/turbulent/turb_sources.hpp

Lines 365 to 366 in 1f92147

	Residual[0] += yinv*Volume*(pk_axi-cdk_axi);
	Residual[1] += yinv*Volume*(pw_axi-cdw_axi);

For the (averaged) flow equations the sources block is added (LHS):

SU2/SU2_CFD/src/solvers/CEulerSolver.cpp

Line 2406 in 1f92147

LinSysRes.AddBlock(iPoint, residual);

For the turbulence equations the sources block is subtracted (RHS):

SU2/SU2_CFD/src/solvers/CTurbSSTSolver.cpp

Line 347 in 1f92147

LinSysRes.SubtractBlock(iPoint, residual);

Or I am still missing something...

[bigfooted]

OK, now I see where the difference comes from. But then the question is why we add source terms for the flow and subtract source terms for turbulence. This is really confusing...

[bigfooted]

Ok, so I am now convinced that the source terms for SST are correct. I think we can add the Jacobian as well. We take the gradient of the residual wrt rho.k and rho.w and then J[0][0]= (-v + P/rhok).Vol/r and J[1][1]=-v.Vol/r (ignoring derivatives of the gradients).

And then for SA, we need to add at least the convection-diffusion terms, and some other terms perhaps depending on the production terms.

[FlorianDm]

I had actually added the Jacobian in the PR but ended up removing it after testing a lot and not finding any (positive) effect on convergence. There is a little discussion with Pedro in the PR about conserving diagonal dominance but in the end it did not make a difference. You can find the removed Jacobian terms here dd1b1b3