Inaccurate gradient results from continuous Adjoint method
I'm studying global optimization methods.
Currently, I consider airfoil design as the application.
However, I have found inaccurate gradient results when I use the continuous Adjoint method which is enclosed in the SU2-Suite.
Here is the information of the airfoil design.
Mach number = 0.75
Design variables= 5 Hicks-Henne bump functions on the upper surface of the airfoil
In order to measure the accuracy of the Adjoint method,
I considered 2 bump functions, and compared gradient values
from the Adjoint method and finite difference method.
All computation conditions(grid, # of CPUs, etc) were same except the method.
The above is the two plots.
White and red plots were each calculated using the finite difference and the Adjoint method.
As you can see, the overall trends and the magnitudes of the plots are different each other.
Since step size for the finite difference method was selected by some tests,
I think the finite difference method is accurate.
What makes the difference? Is this a limitation of the Adjoint method?
If you have a while, please take a look at
|All times are GMT -4. The time now is 17:49.|