Dear Ronghong Fan, Yes, I implemented several versions. The question I asked above got resolved as explained above. If you have a specific question feel free to ask, I will help if I can. Keep in mind that Multiscale FEMs are not generic, only work in certain (scaling) scenarios and do not work so well if you do not have scale separation (at least not in their standard version).
Let me know about questions. Cheers, Konrad On Wednesday, November 14, 2018 at 3:50:27 AM UTC+1, Ronghong Fan wrote: > > Dear Konrad, > Do you have some progress about Multiscale Finite Element Method with > dealii? > I am very interested in mfem too and want to use dealii to implement. > > > On Thursday, August 30, 2018 at 7:27:21 PM UTC+8, Konrad wrote: >> >> On Wednesday, August 29, 2018 at 5:25:29 PM UTC+2, Wolfgang Bangerth >> wrote: >>> >>> >>> I think that's actually an important point you are making here, maybe >>> inadvertently: You have as many *meshes* as there are coarse cells. >>> >>> Now, just refining the global mesh a number of more times gives you >>> another global mesh, but it does not give you a mesh for each coarse >>> cell. >>> >>> In my deal.II classes, I've had students solve problems like yours and >>> they typically just build one new triangulation for each coarse mesh >>> cell, then attach a DoFHandler to it, and solve problems on them. That >>> is likely not very efficient if you want to solve real problems, but >>> it's good enough for small testcases and it allows for easy >>> parallelization using threads. >>> >>> On the other hand, if you were to build a global DoFHandler on the >>> refined mesh, you'd have to define what to do with those DoFs that lie >>> outside the patch (that results from a single coarse mesh) you are >>> currently considering. You'd likely want to set these to zero, but >>> they're still there, and the resulting patch problem will be large. It's >>> also much more difficult to do things in parallel because now you have >>> one global (fine) mesh that you want to use for multiple things at the >>> same time. >>> >> >> This is exactly the way I decided to do it. As you say, it is probably >> not the most efficient way but easy to thread (and maybe even easy to bring >> on different nodes on a cluster). Thanks again for your comments. >> >> Cheers, >> Konrad >> > -- 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. For more options, visit https://groups.google.com/d/optout.
