The kOmegaSSTLM transition model cannot calculate derivatives during field inversion

[6Mrzhang]

Dear professor He,
The version is dafoam 3.1.2, when I used the kOmegaSSTLM transition model for field inversion, the initial solution converged very well, but when using the runAdjoint task, there was a situation where the derivatives were all 0.

Then I modified the production term Pk (G) in the turbulence kinetic energy equation of the kOmegaSSTLM model

to

Then the full derivative appeared, and it was an intuitive full derivative. Below is a comparison of the derivatives of the two,

This is my runScript. py,

runScript_v2.txt
The only difference between the two cases is the production item, nothing else has changed, I don't know why, and using gammaInt instead of gammaIntEff is not a truly effective transition model. I want to know how to use gammaIntEff to calculate the correct full derivative

[friedenhe]

Is this channel flow fully laminar? If yes, the deriv wrt beta should be zero because it will not impact the flow.

[6Mrzhang]

its T3A transition plate, there must have been a transition here, my Cf curve is consistent with the curve provided by Openfoam official, so gammaInt ,so there is a derivative exist when changing gammaIntEff to gammaInt in the product source terms.

[6Mrzhang]

gammaInt does not appear in the KRes equation, and what truly affects KEqn is gammaIntEff

The expression for the full derivative is:
$$\dfrac{dF}{d\beta}=\dfrac{\partial F}{\partial \beta}-\psi^{T}\dfrac{\partial R}{\partial \beta}$$
in my case, the $$\psi^{T}\dfrac{\partial R}{\partial \beta}=0$$ ,
and $$\psi$$ can be get from
$$\dfrac{dR}{dw} \psi= \dfrac{dF}{dw}$$
I checked some derivatives through 'write Jacobians' in this equation, i get
$$\dfrac{dR}{dw}!=0$$ , $$(\dfrac{dR}{d\beta})^T\psi =0$$ and $$(\dfrac{dR}{d\beta})^T\psi=0$$
then i get $$\psi=0$$
i don't know why.

[friedenhe]

As I understand, gammaIntEff_ = max(gammaInt_(), gammaSep); , so they are related. I just looked at the SSTLM code and realized that something may be missing. Check my comment here:

dafoam/src/adjoint/DAModel/DATurbulenceModel/DAkOmegaSSTLM.C

Line 756 in 7c38ac1

// for SSTLM also update gammaIntEff_

Can you add gammaIntEff_ = max(gammaInt_(), gammaSep); to the above place and see if it fixes the problem? You may also need to add the codes to compute gammaSep. Basically, we need to add the code to update gammaIntEff in updateIntermediateVariables

[6Mrzhang]

Thanks Professor He. Now the kOmegaSSTLM model can work successfully in field inversion case. In addition, I believe this has reference value for other turbulence models as well. I'm happy to find the bug.

[6Mrzhang]

It's my code, I think adding these codes to the correctNut() function may be more concise and efficient, avoiding duplicate memory requests and calculations for grad (U), but it may increase the reading difficulty for those who are not familiar with this issue.

[friedenhe]

It is good to hear that it fixes the problem. No, we should put these lines in updateIntermediateVariables because this function is not called during the primal solution, and will be called only in the adjoint. Can you submit the above as a PR to DAFoam's repo?