Verifying adjoint results - Complex step or Finite Difference

kmandar · February 4, 2016, 09:16

Dear SU2 developers,

Is there a provision to verify the SU2 adjoint results?

Is the Complex Step method implemented in any part of SU2? If not, are there plans for doing so?

I am currently performing Finite Difference to check some of the results I am generating. This involves running SU2_DEF to create a new perturbed mesh and then running SU2_CFD to perform analysis of the perturbed geometry.

How do the SU2 developers do this? Is there a python script already written? (I am curious because do not want to reinvent the wheel.)

Thanks for your help,
Mandar

talbring · February 5, 2016, 05:57

Hi kmandar,

you can use Finite differences or forward mode of AD to verify the gradient.

For FD you can simply use the finite_differences.py script.

For Forward mode (or called direct differentation mode) you have to first of all compile the correct binary. Follow the information here: https://github.com/su2code/SU2/wiki/AD-Build
Instead of or in addition to --enable-autodiff add the flag --enable-direct-diff.
After compilation and setting the correct paths, you can use the script direct_differentiation.py.

kmandar · February 5, 2016, 13:37

Hi Tim,

Thanks for your reply. I will try both these options.

Mandar

February 4, 2016, 09:16	Verifying adjoint results - Complex step or Finite Difference	#1
kmandar Member Mandar Kulkarni Join Date: Nov 2013 Location: Virginia Tech, Blacksburg, VA Posts: 52 Rep Power: 13	Dear SU2 developers, Is there a provision to verify the SU2 adjoint results? Is the Complex Step method implemented in any part of SU2? If not, are there plans for doing so? I am currently performing Finite Difference to check some of the results I am generating. This involves running SU2_DEF to create a new perturbed mesh and then running SU2_CFD to perform analysis of the perturbed geometry. How do the SU2 developers do this? Is there a python script already written? (I am curious because do not want to reinvent the wheel.) Thanks for your help, Mandar

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
solving finite journal bearing problem by finite difference	Atit Koonsrisuk	Main CFD Forum	6	March 29, 2015 09:09
How to write k and epsilon before the abnormal end	xiuying	OpenFOAM Running, Solving & CFD	8	August 27, 2013 16:33
dynamic Mesh is faster than MRF????	sharonyue	OpenFOAM Running, Solving & CFD	14	August 26, 2013 08:47
Finite Volume Vs Finite Difference	Mukkarum	Main CFD Forum	3	February 8, 2003 12:16
Difference in results using different Mesh?	Julie	FLUENT	7	July 2, 2002 11:28

February 5, 2016, 05:57		#2
talbring Super Moderator Tim Albring Join Date: Sep 2015 Posts: 195 Rep Power: 11	Hi kmandar, you can use Finite differences or forward mode of AD to verify the gradient. For FD you can simply use the finite_differences.py script. For Forward mode (or called direct differentation mode) you have to first of all compile the correct binary. Follow the information here: https://github.com/su2code/SU2/wiki/AD-Build Instead of or in addition to --enable-autodiff add the flag --enable-direct-diff. After compilation and setting the correct paths, you can use the script direct_differentiation.py.

February 5, 2016, 13:37		#3
kmandar Member Mandar Kulkarni Join Date: Nov 2013 Location: Virginia Tech, Blacksburg, VA Posts: 52 Rep Power: 13	Hi Tim, Thanks for your reply. I will try both these options. Mandar