|
[Sponsors] |
September 3, 2013, 22:09 |
Adjoint sensitivities at Mach 1.0 and above
|
#1 |
New Member
Join Date: May 2012
Posts: 6
Rep Power: 14 |
Hi,
I'm running the continuous adjoint solver for the naca0012 airfoil from the provided test case in the SU2 package. I am manually comparing the adjoint sensitivities produced against finite differences. I have noticed that for Mach 1 and above that the adjoint sensitivities do not match with finite differences. Is there something that I need to change in the test case configuration other than just the Mach number? Do I need to define different boundary conditions in the farfield, such as supersonic inlet and supersonic outlet conditions? Thank you in advance, KW |
|
September 12, 2013, 05:07 |
|
#2 |
Super Moderator
Thomas D. Economon
Join Date: Jan 2013
Location: Stanford, CA
Posts: 271
Rep Power: 14 |
Hi KW,
As you know, obtaining good agreement between continuous adjoint and finite difference gradients can often be tricky business. The results can be very mesh dependent, very sensitive to the level of numerical dissipation used in the primal and adjoint problems, and the finite difference step size must also be selected carefully. In order to obtain good agreement, you might consider running an array of finite difference step sizes for the problem (the finite_differences.py script assumes 1e-4 by default, but can be adjusted with the '-s' option). You may also want to try a different/finer mesh which is better suited for the particular problem (note that the shock(s) will be in a different location) or to adjust the levels of artificial dissipation if you are using the JST scheme, for instance. The current far-field boundary condition implementation can handle the supersonic conditions. Hope some of these suggestions help. Thanks for using SU2, and all the best, Tom |
|
September 12, 2013, 22:27 |
|
#3 |
New Member
Join Date: May 2012
Posts: 6
Rep Power: 14 |
Thank you Tom.
I will give those suggestions a try. |
|
Thread Tools | Search this Thread |
Display Modes | |
|
|