Manual implementation pressure-strain model

[testal]

Hello,

When a Reynolds stress turbulence model is selected, Fluent allows to specify which scale variable (epsilon or omega) and which formulation for the pressure-strain source term (LRR or SSG) the user wants to use. 

The LRR model is made of four terms which depend on the coefficients C_1 and C_2. The first affects the slow term and the second affects the remaining 3 terms (rapid terms). See Theory Guide, equation 4.247. No wall reflection is needed when BSL-omega is the scale-trasporting equation (like in my case). 

I've been trying to reimplement the rapid terms manually through UDFs:

1. the slow term remains untouched.
2. the coefficient C_2 is changed so that the term with gamma_0 goes to 0.
3. the remaining terms (with alpha_0 and beta_0) are canceled out by a UDF containing six DEFINE_SOURCE functions, one for each of the components of the turbulent stress tensor. This UDF works as a sink.
4. a second UDF, containing six DEFINE_SOURCE functions, reinjects the rapid terms. 

The idea is that simulations with this model built through UDFs should give the same result as the simulation using the standard built-in LRR term. I found this to be somehow true after a few thousands of iterations. After that, if the simulation keeps running, the solution begins to deteriorate and the residuals steadily rise. The domain is 2D, axisymmetric swirl. The deterioration seems to begin in the axis region and then propagates to the rest of the domain.

My question is whether Fluent imposes limiters and/or realizability checks that are not reported in the guides.

Thank you.

[Muhammad Hassan]

That's an interesting observation. Since your UDF implementation matches the built-in LRR model for the first few thousand iterations before diverging, it does make me wonder if Fluent is applying some internal limiters, realizability checks, or implicit source-term treatment that isn't documented.

Another possibility is that the built-in model handles source-term linearization differently than a UDF, which could explain why the solution gradually becomes unstable, especially near the axis.

It would be helpful if someone from the Fluent team could clarify whether any additional stability or realizability corrections are applied internally to the Reynolds Stress Model.