From: Kirk, Benjamin (JSC-EG311) <benjamin.kirk-1@na...> - 2012-07-27 15:03:41
We've got an interest here in loosely coupling two codes that communicate
through boundary data. It will also be possible to tightly couple them, as
they are both based on libMesh, but because of disparate time scales I think
we actually want the ability to run them loosely coupled.
We're planning to use something like inverse distance interpolation from a
point cloud, where each set of physics will populate its own point cloud
that the other interpolates from.
I'm thinking there would be a pretty wide audience for this, so I'm going to
add it into libMesh, but I wanted to start some discussion on the list first
to see if anyone has particular design goals they'd like incorporated.
From: Roy Stogner <roystgnr@ic...> - 2012-07-27 15:28:36
On Fri, 27 Jul 2012, Kirk, Benjamin (JSC-EG311) wrote:
> We're planning to use something like inverse distance interpolation from a
> point cloud, where each set of physics will populate its own point cloud
> that the other interpolates from.
What's the advantage over just doing an L2/H1/whatevers-appropriate
projection? I know, explicit vs implicit, but global projections are
about as well conditioned and well behaved as implicit gets; the solve
rarely takes long compared to the subsequent real PDE solve.