[elio_xh]

Is there any way to extract the geometric sensitivities (design velocities) matrix for a Hicks-Henne 2D airfoil adjoint optimization?

I see in history output of a continuous adjoint run the "Sens_Geo" column but my understanding is this value is the summation of all sensitivities on the surface, but what I'm looking for is the full matrix that measures the influence of the design variables on the position of each grid point on the surface mesh.

I've tried editing the CAdjEulerSolver::Inviscid_Sensitivity module (in solution_adjoint_mean.cpp), to print all the CSensitivity values calculated at every point on the boundary, but my understanding is even this already sums each design variable contribution. Is there any way to extract that matrix?

[hlk]

Originally Posted by elio_xh

Is there any way to extract the geometric sensitivities (design velocities) matrix for a Hicks-Henne 2D airfoil adjoint optimization?

I see in history output of a continuous adjoint run the "Sens_Geo" column but my understanding is this value is the summation of all sensitivities on the surface, but what I'm looking for is the full matrix that measures the influence of the design variables on the position of each grid point on the surface mesh.

I've tried editing the CAdjEulerSolver::Inviscid_Sensitivity module (in solution_adjoint_mean.cpp), to print all the CSensitivity values calculated at every point on the boundary, but my understanding is even this already sums each design variable contribution. Is there any way to extract that matrix?

Thanks for your question.
The output file surface_adjoint.dat contains the surface sensitivity on your plotted marker, as well as the values of the adjoint variables.
The output file of_grad_{} contains the sensitivity of the objective with respect to the design variables defined in the config file.
The routines in SU2_DOT project the surface sensitivites onto particular design variables.

[kmandar]

Hi elio_xh,

One way I think of getting discrete design velocities from SU2 would be using finite difference as follows:

1. Use the Hicks_Henne design variable definition to perturb the mesh with a small step, say DV_VALUE= 0.01. You can output the deformed mesh using SU2_DEF suite.

2. Use finite difference to get the design velocity at each grid point using the two meshes and the value of the perturbation.

Of course, the result will depend on the step size, so you may have to do a step size study to get the correct step size.

I believe there would be a way to get these values without using finite difference, but I am not aware of it.

Another way would be to get the design velocities from software you use to discretize the domain using this geometry.

Hope this helps,
Mandar