Continuous Adjoint computation with two separate walls

CarlosLozano · November 2, 2022, 08:05

Hi,
I am reproducing the nozzle test case but with inviscid flow and the complete nozzle geometry (i.e., adding also a lower wall). I have two separate markers for the upper and lower walls (UWALL and DWALL, resp), and when I compute the drag adjoint solution it only obeys the adj bc
(psi2,psi3).(nx,ny) = (cos aoa,sin aoa).(nx,ny) on the upper wall. On the lower wall it obeys (psi2,psi3).(nx,ny) =0. I am using v5 (yes, I know....). Any clues?

In the .su2 file I have
.
.
.
NMARK= 4
MARKER_TAG= DWALL
MARKER_ELEMS= 943
3 0 1
3 1 2
.
.
.
3 941 942
3 942 943
MARKER_TAG= OUTFLOW
MARKER_ELEMS= 200
3 943 1887
3 1887 2831
.
.
.
3 187855 188799
3 188799 189743
MARKER_TAG= UWALL
MARKER_ELEMS= 943
3 189743 189742
3 189742 189741
.
.
.
3 188802 188801
3 188801 188800
MARKER_TAG= INFLOW
MARKER_ELEMS= 200
3 188800 187856
3 187856 186912
.
.
.

In the .cfg file I have

% ---------------- BOUNDARY CONDITION DEFINITION -------------------%
%
% Marker of the Euler boundary (0 = no marker)
MARKER_EULER= ( UWALL, DWALL )
%
% Marker of the far field (0 = no marker)
MARKER_FAR= ( INFLOW,OUTFLOW )
%
% Marker of the surface which is going to be plotted or designed
MARKER_PLOTTING= ( UWALL, DWALL )
%
% Marker of the surface where the functional (Cd, Cl, etc.) will be evaluated
MARKER_MONITORING= ( UWALL, DWALL )
%

(Yes, I am using far-field type inlt/outlet since in v5 inflow and outflow for adjoint are not defined). Shouldn't make much of a difference for the problem at hand).

Thanks in advance!

bigfootedrockmidget · November 2, 2022, 08:48

Quote:

Originally Posted by CarlosLozano

I am using v5

...... why?

CarlosLozano · November 2, 2022, 08:54

Short answer: the cluster i work with has a very old SO version and v6 and higher do not compile properly. I am not the admin, so there's little i can do. Besides, v5 works fine for me and i am not sure that v6 and higher have brought much improvement for the continuous adjoint solver.

bigfootedrockmidget · November 2, 2022, 09:23

It will be impossible to help you with this. I think it is better if you try to recompile the latest su2, and we can then support you with the compilation difficulties. When the problem with continuous adjoints that you are now reporting still exists, we can then dive into the latest code and try to fix it.

Also note that system admins are there to make sure that the cluster is working properly for its users.

CarlosLozano · November 2, 2022, 09:29

Quote:

Originally Posted by bigfootedrockmidget

It will be impossible to help you with this. I think it is better if you try to recompile the latest su2, and we can then support you with the compilation difficulties. When the problem with continuous adjoints that you are now reporting still exists, we can then dive into the latest code and try to fix it.

Also note that system admins are there to make sure that the cluster is working properly for its users.

Ok, thanks

November 2, 2022, 08:05	Continuous Adjoint computation with two separate walls	#1
CarlosLozano New Member Join Date: Feb 2018 Posts: 27 Rep Power: 8	Hi, I am reproducing the nozzle test case but with inviscid flow and the complete nozzle geometry (i.e., adding also a lower wall). I have two separate markers for the upper and lower walls (UWALL and DWALL, resp), and when I compute the drag adjoint solution it only obeys the adj bc (psi2,psi3).(nx,ny) = (cos aoa,sin aoa).(nx,ny) on the upper wall. On the lower wall it obeys (psi2,psi3).(nx,ny) =0. I am using v5 (yes, I know....). Any clues? In the .su2 file I have . . . NMARK= 4 MARKER_TAG= DWALL MARKER_ELEMS= 943 3 0 1 3 1 2 . . . 3 941 942 3 942 943 MARKER_TAG= OUTFLOW MARKER_ELEMS= 200 3 943 1887 3 1887 2831 . . . 3 187855 188799 3 188799 189743 MARKER_TAG= UWALL MARKER_ELEMS= 943 3 189743 189742 3 189742 189741 . . . 3 188802 188801 3 188801 188800 MARKER_TAG= INFLOW MARKER_ELEMS= 200 3 188800 187856 3 187856 186912 . . . In the .cfg file I have % ---------------- BOUNDARY CONDITION DEFINITION -------------------% % % Marker of the Euler boundary (0 = no marker) MARKER_EULER= ( UWALL, DWALL ) % % Marker of the far field (0 = no marker) MARKER_FAR= ( INFLOW,OUTFLOW ) % % Marker of the surface which is going to be plotted or designed MARKER_PLOTTING= ( UWALL, DWALL ) % % Marker of the surface where the functional (Cd, Cl, etc.) will be evaluated MARKER_MONITORING= ( UWALL, DWALL ) % (Yes, I am using far-field type inlt/outlet since in v5 inflow and outflow for adjoint are not defined). Shouldn't make much of a difference for the problem at hand). Thanks in advance!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Continuous Adjoint - Linear solver diverged	pedropcb13	SU2	0	August 6, 2021 18:34
Discrete Adjoint Euler Computation	doan.nak	SU2	2	June 6, 2014 05:01
Inaccurate gradient results from continuous Adjoint method	Yminjo	SU2	1	January 11, 2014 23:08
adjoint method- uniformity	Envolly	OpenFOAM Programming & Development	8	July 9, 2013 02:29
Question about Continuous Adjoint Method	fumiya	Main CFD Forum	6	November 1, 2011 11:09

November 2, 2022, 08:54		#3
CarlosLozano New Member Join Date: Feb 2018 Posts: 27 Rep Power: 8	Short answer: the cluster i work with has a very old SO version and v6 and higher do not compile properly. I am not the admin, so there's little i can do. Besides, v5 works fine for me and i am not sure that v6 and higher have brought much improvement for the continuous adjoint solver.

November 2, 2022, 09:23		#4
bigfootedrockmidget Senior Member bigfoot Join Date: Dec 2011 Location: Netherlands Posts: 676 Rep Power: 21	It will be impossible to help you with this. I think it is better if you try to recompile the latest su2, and we can then support you with the compilation difficulties. When the problem with continuous adjoints that you are now reporting still exists, we can then dive into the latest code and try to fix it. Also note that system admins are there to make sure that the cluster is working properly for its users.