SU2 Discrete Adjoint Error!!!!

vishu343 · September 12, 2019, 18:13

Hello All,

I am trying to verify sensitivity of Discrete Adjoint against Finite differences for tutorial case ONERAM6 (SU2 version 6.2.0).

I have plotted dCd/dX (Change of drag coefficient against change in one coordinate of FFD box like delx = 0.001).

In this case, what I see is that at trailing edge of the wing, values differ significantly compared to Finite difference and in other words, Discrete Adjoint seems to be matching with Continuous Adjoint and there is no clear benefit of using Discrete Adjoint in this case.

Would anybody shed some light on such issue or anybody might have experienced something like this!

Thanks a lot!

pcg · September 15, 2019, 06:22

Are you using the sharp corner filter option?

vishu343 · September 15, 2019, 07:50

Hi, Thanks for your reply.

No, I am not aware of it (sharp corner filter option). What is that actually! Does it affect discrete adjoint sensitivity.

Thanks.

pcg · September 16, 2019, 05:48

It is a limiter for the sensitivities in areas where geometry changes very rapidly, it avoids numerical issues but it could explain why you see a mismatch close to the trailing edge.
The name of the option is SENS_REMOVE_SHARP, by default it is off so if you don't see it in your cfg file it is not being used.

Have you checked the FD value you get is (more or less) independent of the perturbation you use?

cfdjetman · October 15, 2019, 18:55

I've posted my question here

the gradients calculated through discrete adjoint are near zero

Could you someone help with this issue.

September 12, 2019, 18:13	SU2 Discrete Adjoint Error!!!!	#1
vishu343 New Member Vishal Chauhan Join Date: Jan 2012 Posts: 14 Rep Power: 14	Hello All, I am trying to verify sensitivity of Discrete Adjoint against Finite differences for tutorial case ONERAM6 (SU2 version 6.2.0). I have plotted dCd/dX (Change of drag coefficient against change in one coordinate of FFD box like delx = 0.001). In this case, what I see is that at trailing edge of the wing, values differ significantly compared to Finite difference and in other words, Discrete Adjoint seems to be matching with Continuous Adjoint and there is no clear benefit of using Discrete Adjoint in this case. Would anybody shed some light on such issue or anybody might have experienced something like this! Thanks a lot!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[blockMesh] blockMesh with double grading.	spwater	OpenFOAM Meshing & Mesh Conversion	92	January 12, 2019 09:00
[OpenFOAM] Native ParaView Reader Bugs	tj22	ParaView	270	January 4, 2016 11:39
[swak4Foam] GroovyBC the dynamic cousin of funkySetFields that lives on the suburb of the mesh	gschaider	OpenFOAM Community Contributions	300	October 29, 2014 18:00
OpenFOAM without MPI	kokizzu	OpenFOAM Installation	4	May 26, 2014 09:17
Problem with compile the setParabolicInlet	ivanyao	OpenFOAM Running, Solving & CFD	6	September 5, 2008 20:50

September 15, 2019, 06:22		#2
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 465 Rep Power: 13	Are you using the sharp corner filter option?

September 15, 2019, 07:50		#3
vishu343 New Member Vishal Chauhan Join Date: Jan 2012 Posts: 14 Rep Power: 14	Hi, Thanks for your reply. No, I am not aware of it (sharp corner filter option). What is that actually! Does it affect discrete adjoint sensitivity. Thanks.

September 16, 2019, 05:48		#4
pcg Senior Member Pedro Gomes Join Date: Dec 2017 Posts: 465 Rep Power: 13	It is a limiter for the sensitivities in areas where geometry changes very rapidly, it avoids numerical issues but it could explain why you see a mismatch close to the trailing edge. The name of the option is SENS_REMOVE_SHARP, by default it is off so if you don't see it in your cfg file it is not being used. Have you checked the FD value you get is (more or less) independent of the perturbation you use?

October 15, 2019, 18:55		#5
cfdjetman New Member cfdjetman Join Date: Mar 2019 Posts: 25 Rep Power: 7	I've posted my question here the gradients calculated through discrete adjoint are near zero Could you someone help with this issue.