SU2 Drag Sensitivities
 

Dear Developpers

I want to use SU2 to just compute the intial drag sensitivities for a full wing body configuration. Is it necessary to create an FFD box? Can I just run the adjoint solver with grid deformation to obtain the sensitivities?

Would be happy if some one can provide an initial ,cfg file for the above case. couldn't find one in the tutorials

Regards

kaus

In short: It is not necessary. Running the adjoint solver will give you the sensitivities in the sense "Objective function derivative with respect to grid node positions".

You can then use SU2_DOT_AD if you want to incorporate the sensitivities of a mesh deformation. Be careful that these sensitivities are then only accurate with respect to SU2's mesh deformation algorithm, of course.

A FFD box is "just" the build-in functionality in SU2 for real optimization steps, it can be done with every other technique/optimizer as well.

So no "special" .cfg files are necessary, run the solvers and use the output files (by default, they all contain the word _sens_ somewhere ;-))

Regards, Ole

Thanks I still have a question. I have a converged RANS (Direct) solution stored. Can I start my adjoint solver over this ?

what should be the options in the cfg file?

I have given
% ------------- DIRECT, ADJOINT, AND LINEARIZED PROBLEM DEFINITION ------------%
%
% Physical governing equations (EULER, NAVIER_STOKES,
% WAVE_EQUATION, HEAT_EQUATION, FEM_ELASTICITY,
% POISSON_EQUATION)
PHYSICAL_PROBLEM= NAVIER_STOKES
%
% Mathematical problem (DIRECT, CONTINUOUS_ADJOINT)
MATH_PROBLEM=CONTINUOUS_ADJOINT
%
% Restart solution (NO, YES)
RESTART_SOL= YES

and the restart file name. However the adjoint solver is failing after a few iterations.

do not know the reason.

regards

Kaus

Also is the adjoint grid different from the direct grid. One reason for this question is that the adjoint problem is backward in time and hence may require a different grid than the forward. What is surprising is that the forward problem converges without any problem wheras the adjoint problem diverges!

Yes, that makes sense.

However, I'm not familiar with the continous solver - so maybe someone else can help out.
What I said in previous post still accounts for both solvers.

If I were you, I would try with the discrete adjoint (Automatic Differentiation). Continuous adjoint could be tricky. Remember that RESTART_SOL= YES should be used if you already have a solution for the adjoint.

Best,
Francisco

Do we have a tutorial for the discrete adjoint solver? I could not find one

In this tutorial you find also information on how to switch to the discrete adjoint:

https://su2code.github.io/tutorials/...ined_NACA0012/