On Mon, 26 Oct 2015, Miguel Angel Salazar de Troya wrote: > Can libmesh handle computing the adjoint solution for several QoI? It looks > like it can do so by looping over the QoISet, but then when it comes to > calculate the adjoint solution in ImplicitSystem::adjoint_solve() we don't > loop over different diff_qoi in FEMSystem::assemble_qoi_derivative (). In > fact we cannot store several diff_qoi.
Nope. A single DifferentiableQoI gets asked for all QoI (or for a QoISet subset of them) at once. If you want better factored code than that (which is certainly reasonable!) then it's up to you to implement a CompoundQoI subclass that combines other DifferentiableQoI members. I'd have put such a thing in libMesh long ago, except that Paul Bauman beat me to it in https://github.com/grinsfem/grins/blob/master/src/qoi/include/grins/composite_qoi.h with a class which has left me in limbo: the design there is IMHO not quite good enough to move straight into libMesh but not nearly bad enough that I've felt any impulse to rewrite it. My main complaint with Paul's CompositeQoI is just that the members aren't themselves DifferentiableQoI subclasses; they subclass a separate and somewhat redundant GRINS::QoIBase. But that probably wasn't his fault; IIRC he started on GRINS before I even had DifferentiableQoI properly refactored out, and he wasn't yet confident enough with libMesh code to call me out rather than come up with workarounds when he spotted something poorly factored. --- Roy ------------------------------------------------------------------------------ _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
