[Tommy Chen]

Hi Guys,

When running su2_cfd adjoint solver on one of my own case, I noticed that the sens_geo is sometimes a minus value, though it still does not converge, but I wonder how could this variable be a minus value, What is the meaning of this variable?

Thanks

[copeland]

Hi Tommy,

The Sens_Geo output corresponds to your specified objective function's sensitivity to local, normal geometric perturbations of the surface. This geometric sensitivity can be positive or negative, indicating that local outward perturbations of the surface increase or decrease your specified objective function accordingly.

Regarding the convergence of the adjoint solver -- there are many factors that affect the convergence of the linear system, including the CFL number, the numerical methods and the accuracy of the linear solver (in addition to the free-stream conditions, the quality of the mesh, and the geometry of interest!). As a first-step, try reducing the CFL number, increase the LINEAR_SOLVER_ITER and decrease the LINEAR_SOLVER_ERROR quantities in your configuration file.


Regards,
Sean

[Tommy Chen]

Originally Posted by copeland

Hi Tommy,

The Sens_Geo output corresponds to your specified objective function's sensitivity to local, normal geometric perturbations of the surface. This geometric sensitivity can be positive or negative, indicating that local outward perturbations of the surface increase or decrease your specified objective function accordingly.

Regarding the convergence of the adjoint solver -- there are many factors that affect the convergence of the linear system, including the CFL number, the numerical methods and the accuracy of the linear solver (in addition to the free-stream conditions, the quality of the mesh, and the geometry of interest!). As a first-step, try reducing the CFL number, increase the LINEAR_SOLVER_ITER and decrease the LINEAR_SOLVER_ERROR quantities in your configuration file.


Regards,
Sean

Thanks Sean,

In the convergence history file , there is only one single value of the Sens_Geo for every single iteration, so if this value corresponds to the specified objective function's sensitivity to local, normal geometric perturbations of the surface, then where is the location of the perturbation ?

Sens_Geo should be different on very single surface mesh node , but in the history file there is only one value on each iteration.

[copeland]

Hey Tommy,

You're absolutely right -- there is a single value of Sens_Geo in the history file and my explanation before was not quite correct.

If you look CAdjEulerSolver::Inviscid_Sensitivity (in solution_adjoint_mean.cpp), you can see the calculation of Sens_Geo, and it may clear things up. There is a CSensitivity which is calculated at every point on the boundary (THIS term is what I was describing my previous post), and Sens_Geo is the summation of all CSensitivities on the surface -- so Sens_Geo represents the total objective function sensitivity to shape perturbations.



Regards,
Sean

[Tommy Chen]

Originally Posted by copeland

Hey Tommy,

You're absolutely right -- there is a single value of Sens_Geo in the history file and my explanation before was not quite correct.

If you look CAdjEulerSolver::Inviscid_Sensitivity (in solution_adjoint_mean.cpp), you can see the calculation of Sens_Geo, and it may clear things up. There is a CSensitivity which is calculated at every point on the boundary (THIS term is what I was describing my previous post), and Sens_Geo is the summation of all CSensitivities on the surface -- so Sens_Geo represents the total objective function sensitivity to shape perturbations.



Regards,
Sean

Thanks Sean

Your explanation is quite clear.

I am thinking that it might be better if using root mean square of the Sens_Geo of each surface mesh node as measurement.

Anyway, thank you so much.