On Sun, Aug 28, 2022 at 12:44 PM Nicholas Arnold-Medabalimi < [email protected]> wrote:
> Hi > > I'm working on a finite volume implementation using DMPlex. I was > wondering what the best-considered way to add cell variables to the DM that > we don't want to be part of the ksp/snes solve. As far as I can tell, if I > add an extra section with those, they will still be part of the F and U > vectors in the solver callbacks (to be expected). But my impression is that > that will just inflate the system size even if they have an F=0? Or is the > solver able to notice variables that are not dependent for the solve. > > Alternatively, I have been just making a separate DM with those but that > feels a bit clunky especially if any mesh refinement occurs. > Mark is correct. We have been using a two DM strategy. You can attach this auxiliary data using https://petsc.org/main/docs/manualpages/DM/DMSetAuxiliaryVec/ The idea is that you define a layout for your extra information using a new DM, then create a Vec for it, and attach it to the DM. This can be pulled out by any callback function, and the DM can be pulled out of the vector. In the Plex pointwise functions for residual and Jacobian, auxiliary data is automatically interpolated to the quadrature points as well. Is this what you were looking for? Thanks, Matt > Any advice would be much appreciated. > > Thanks > Nicholas > > -- > Nicholas Arnold-Medabalimi > > Ph.D. Candidate > Computational Aeroscience Lab > University of Michigan > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
