Discrete vs Continuous Adjoint solution variable: Scaling/normalization issue?
4 Attachment(s)
Hello all,
Why are the Discrete adjoint solution (psiRho, for ex) values order of magnitude different from the Continuous adjoint solution values? Referring to the QuickStart tutorial, I could match exactly with the continuous adjoint solution. However, when run with MATH_PROBLEM=DISCRETE_ADJOINT option with the same config file (inv_NACA0012.cfg) the adjoint solution variables is order of magnitude different than the continuous adjoint. See attached figures. The figures also include Continuous vs Discrete surface sensitivity plot. I actually verified the Discrete adjoint sensitivity with the Direct Derivatives, besides verifying dCd/dM from FD with Disc Adj. Is there some internal normalization for Disc Adjoint solution that is saved in adjoint restart file? Any pointers would help. |
This is something I am also experiencing. Did you find the reason for this in the end?
Z |
I guess you used two different configuration files for continuous and discrete runs.
You might look at the REF_DIMENSIONALIZATION in the configuration file and make sure that the non-dimensionalization is the same between the two adjoint runs. |
All times are GMT -4. The time now is 03:12. |