search

mdolab / adflow

ADflow is a finite volume RANS solver tailored for gradient-based aerodynamic design optimization.
Other
220 stars 98 forks source link

Time spectral stability derivatives are broken #295

Open sseraj opened 1 year ago

sseraj commented 1 year ago

Description

As mentioned in the code, the time spectral stability derivatives are broken. I am creating this issue for greater visibility.

hntter commented 1 year ago

Thanks for remingding the stability derivatives bugs. I notice the time spectral test file (test_time_spectral_naca64A010.py) used 'useexternaldynamicmesh' to define grid motion. I also tested only using 'qMode' of 'alphaMode' to run time spectral analysis but it seems has no effect for the results, which shows the same values for different timeintervals. Is using 'useexternaldynamicmesh' to define grid motion is the only way for now to run TS analysis? For another question, I can fit the data cm versus φ to get cmzq, but using cmzq as a constraint, is it also needed to evaluate it sensitive?

Besh wishes

sseraj commented 1 year ago

Is using 'useexternaldynamicmesh' to define grid motion is the only way for now to run TS analysis?

Yes, this is the only way that currently works.

For another question, I can fit the data cm versus φ to get cmzq, but using cmzq as a constraint, is it also needed to evaluate it sensitive?

Yes, to use cmzq as a constraint, you will also need the gradient. The stability code is differentiated with Tapenade, but it is also broken because the analysis is broken. The time spectral dynamic mesh code is not differentiated (tracked in #198).