[Haolin Zhi]

Hello, every SU2 developer:

I'm reading the python codes of SU2-7.3.1 to figure out the whole framework of shape optimization. Here, I find a small irrational point in the definition of evalution of constraint gradients. (in SU2_PY/SU2/eval/design.py)

For example, we may add a weight coefficient to each equality/inequality constraint in the .cfg file. So in the python framework, the weight will be stored in <config['OPT_CONSTRAINT']['EQUALITY']['SCALE']>, right? When we compute the constraint as well as its gradients using SU2_CFD solver, the python code will get the constraint and gradients from the history files, and then multiple them with the weight (or SCALE), and then return the results to scipy_tools.py for optimization.

Look at line 429 in the function "con_ceq" in file "design.py", the resulted constraint is scaled by the SCALE parameter, which is no problem. Then look at line 477 in the gradient function "con_dceq", the resulted gradients are not scaled by the SCALE parameter, that I think it is irrational. The same problem also happens in the function "con_dcieq", the gradients of inequality constraints.

I think that the two gradients function for constraints should be scaled with the SCALE parameter, because at line 386 in the objective gradient function "obj_df", the computed objective gradients are indeed scaled.

I really appreciate any SU2 developer could check and critize my doubt. Maybe I'm wrong, because I am just a novice of SU2 code.

ZHI

[Haolin Zhi]

Sorry, everyone. I have seen the statement in .cfg file that this push factor does not affect the gradients of constraints.

So I mistook it.