Adjoint Improvements

[talbring]

This pull request introduces major structural changes to the discrete adjoint solver. I moved some of the routines to the driver level so that multi-zone problems work with the discrete adjoint. The basic implementation works for Fluid-Fluid zones. Other drivers follow soon (e.g. turbomachinery).

Along with that there also some other minor changes:

• The low memory tecplot output (which was not working properly anyway) is now removed.
• The issue with the unsteady cont. adjoint referenced in Unsteady optimization in V5.0 #377 is solved.
• Restart for the discrete adjoint works properly (as long as the same number of procs is used)
• I added options for frozen eddy viscosity and frozen limiter. This can be enabled with the options FROZEN_VISC_DISC/FROZEN_LIMITER_DISC. The default is NO for both.

[ghost]

Hi Tim, looks like to me a great and useful implementation. I have just one comment for the future... would it be possible to unify the continuous and discrete adjoint options in the config file? In the past I have successfully combined upwind schemes for the primal problems with centered schemes for the dual problem (using the discrete adjoint!). I know it is not consistent but... it works! and I'm not the only one using this strategy. On the other hand, I'm looking forward to try the new restart, I don't think previous versions are working correctly.

Thanks again,
Francisco

[talbring]

I also thought about unifying the options. However, for example for the frozen viscosity option we have different default values (NO for disc. adj. and YES for cont. adj.). That's why I did it that way for now. But I'll think of some solutions for that problem for the future.

Best,
Tim

[ghost]

Thanks Tim, From the practical view of point, it is important to decide which numerical spatial integration do you want to use for the AD stuff… even if it is not consistent with the direct solver. Excellent job! Thanks, Francisco

…

On Jul 12, 2017, at 11:57 PM, Tim Albring ***@***.***> wrote: I also thought about unifying the options. However, for example for the frozen viscosity option we have different default values (NO for disc. adj. and YES for cont. adj.). That's why I did it that way for now. But I'll think of some solutions for that problem for the future. Best, Tim — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#407 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AccuRtpsNdEcP-XjJy86bhhu1LXesbZ-ks5sNb_kgaJpZM4OS2wu>.