Hello Vachan, Quick note. I'm only suggesting looping yourself because I am not totally familiar with MeshWorker. Looping myself exposed the mechanics a bit more for me, which I wanted.
I was pointing to step-33 for the loop, not the operators. I don't remember seeing a deal.II example implementing deal.II in operator form. So, I suggested looping though the cells to compute the operators that are not common to all matrices. This is usually the mass matrix if your cells are not linear (aka cartesian mesh). The other operators can be made the same for all the cells, except for some cofactor Jacobian matrix to transfer derivatives and normals from reference space to physical space. If you only ever plan on using a cartesian mesh, then all your local matrices will be the same. There is no point in assembling a global matrix that will operate on your solution. If I were you, I'd take a look at chapter 3 & 6 of Hesthaven's book where he builds the same local operators that will be used on each cell. Therefore, you wouldn't have an array of differentiation matrices. Just one small differentiation matrix that applies to all cells. Feel free to look at how I do the [mass matrix]( https://github.com/dougshidong/PHiLiP/blob/master/src/dg/dg.cpp#L752) Step-33 is CG by the way, therefore, two cells at the same level (no hanging nodes) don't need to be added to the residual. Therefore, they don't show that case. Feel free to look at my [loops]( https://github.com/dougshidong/PHiLiP/blob/master/src/dg/dg.cpp#L246), where I loop over all the cells, and all the faces. For internal faces, only one cell will be responsible to evaluate the flux and add it to both sides. If someone wants to chime in with a better way to loop over all the faces just once, I would love to know about it too. You can also keep on eye out on the code I linked. It currently does not pre-compute the operators, but it's only a matter of moving the loops out to pre-processing. Praveen, I did inspire my loops out of your dflo ;). Note that if you use a distributed grid, the user_index() will not be consistent across MPI boundaries. Therefore, you might end up with doubly-counted faces. Here is a [commit]( https://github.com/dougshidong/PHiLiP/commit/d0d7e6bd663b0cef032c7f953d2b2dddce4950a7#diff-d2da12f45bddf4e965acafc118ff8366) to your if with an if-statement that works for refine/distributed grids. On Wednesday, September 18, 2019 at 12:33:43 AM UTC-4, vachan potluri wrote: > > Doug and Praveen, > > Thanks for your answers. I had a look at step-33. As far as I understand, > although the looping through cells is not through MeshWorker, the assembly > is still global! So, for a non-cartesian mesh, I think you are suggesting > using such a loop over all cells to calculate local matrices. Will these > cell-local matrices stored as an array of matrices in the conservation law > class? > > I now have one more question. In assemble_face_term function of step-33, > the normal numerical flux is calculated. This function is called for every > cell-neighbor pair from the assemble_system function. This means, if I am > not wrong, at every interior face quadrature point, the numerical flux is > calculated twice. This might be very costly. Is there a way to avoid this? > One can probably force only one cell to calculate numerical flux at a face > based on the face normal's orientation. But then how can we communicate the > calculated flux to the other cell. Or even better, is it possible to loop > over faces? We could then add contributions to both the cells sharing this > face without double computation. > > Thanks again! > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/6f1ae5d3-a150-457a-8a6d-fa393910b15e%40googlegroups.com.
