[tomp1993]

Hi all,
I have a slightly strange issue going on with a supersonic adjoint design problem.
My Euler flow solution is converged down to 1e-10 (max.residual about 1e-8.5).
However, when i come to solve the adjoint problem, the solution is slowly diverging; with the solution file showing some unphysical noise which is seemingly propagated upstream from the trailing edge of the geometry - an area which does not, in reality, contribute to the drag of the upstream section which is being designed.

I have tried many fixes for this including:
-reducing adjoint CFL and allowing adaptive CFL,
-changing adjoint numerical scheme and associated 2nd and 4th order coefficients
-using MUSCL both on/off in adjoint problem
-increasing linear solver iterations to 500 and 1000
-using full adjoint grid adaptation with allowance for 25% increase in nodes and surface grid adaptation on.
-good quality fully-hexahedral grid created in Pointwise

here is an image of the surface sensitivity on the leading 20% of the wing - these are as expected and do reflect a true source of drag.
adjcdup.png
However, the trailing 90% of the wing shows unphysical sensitivity 'noise' as seen in the below image. not included in this image is the trailing edge fairing, which is left out of the drag integration so should theoretically have zero influence on either drag or sensitivity of the section of the wing shown here (due to the fact that the flow is supersonic and no pressure signals can be propagated upstream)
adjcduplarge.png

any ideas or suggestions for solving this issue would be very welcome!
many thanks
Tom

[tomp1993]

Hi all,
for anyone else that might encounter this issue; i have solved it.
I changed the downstream grid to fully unstructured instead of the hexa grid previously used. it seems that the adjoint converges very quickly and easily using standard settings in the config file now.
regards
Tom