Did we move forward with this at all ? I can help with getting this done, I
would really like to get adjoints going without penalties.
Thanks.
P.S: Even though it turns out that penalties are an easy (sometimes even
essential) way to get adjoint consistent formulations, they can contribute
hugely to QoI evaluation errors (especially boundary flux evaluations).
On Tue, Aug 28, 2012 at 6:22 PM, Roy Stogner <royst...@ices.utexas.edu>wrote:
>
> Vikram just cooked up a nice new libMesh example to demonstrate his
> multiphysics weighted adjoint residual error estimator, decided he'd
> try out the new DirichletBoundary code in the process... and ran into
> the natural problem when he discovered that the solver happily
> enforced the requested heterogenous Dirichlet constraints even in the
> adjoint problem where he wanted (and where penalty-method based codes
> would have automatically produced) homogeneous constraints instead.
>
> So this is going to take some changes. My best idea so far is to have
> the adjoint_solve "back up" and zero out the right hand side values on
> any heterogenous constraints, then do the solve, then restore the
> backups before returning. Any thoughts? I'm somewhat concerned about
> any possible weird interactions with arbitrary additions via
> System::user_constrain().
> ---
> Roy
--
Vikram Garg
http://users.ices.utexas.edu/~vikram/
http://www.runforindia.org/runners/vikramg
------------------------------------------------------------------------------
Got visibility?
Most devs has no idea what their production app looks like.
Find out how fast your code is with AppDynamics Lite.
http://ad.doubleclick.net/clk;262219671;13503038;y?
http://info.appdynamics.com/FreeJavaPerformanceDownload.html
_______________________________________________
Libmesh-devel mailing list
Libmesh-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-devel
