In examples/cahnHilliard/mesh2D.py, we call eq.solve(..., solver=LinearLUSolver()). I don't know why the default Trilinos solver (probably GMRES) doesn't show the same evolution, but it's possibly the nonlinearities in the 2nd-order DiffusionTerm mean we should be sweeping and GMRES is more sensitive to it than LU for some reason.
On Jan 27, 2014, at 12:33 PM, Jane Hung <[email protected]> wrote: > Another issue: I tried this Cahn Hilliard example > http://pastebin.com/9UZJ2h24 with trilinos instead of pysparse, and the > results seem to differ. The phase separation doesn't seem to occur with the > trilinos solver. > > > On Fri, Jan 24, 2014 at 7:23 PM, Jane Hung <[email protected]> wrote: > Could you give more details about how to extend the overlaps along the top > and bottom of the mesh? Should it be something like this > mesh=Grid2D(nx=500,ny=500,dx=0.25, dy=0.25, overlap=2) ? How could you > specify the top and bottom need to be joined? > > > On Fri, Jan 24, 2014 at 11:54 AM, Daniel Wheeler <[email protected]> > wrote: > On Fri, Jan 24, 2014 at 10:44 AM, Jane Hung <[email protected]> wrote: > > Actually I got it working in Ubuntu, but the parallel implementation doesn't > > give expected results, as in different from running in serial. The > > parallelization setup seems fine from the command line test on the site. Do > > periodic grids work when running in parallel? > > As far as I can tell, they do not work in parallel. This simple test > doesn't do the right thing in parallel. > > import fipy as fp > > m = fp.PeriodicGrid2D(nx=10, ny=10, dx=1., dy=1.) > v = fp.CellVariable(mesh=m, value=0.) > v.setValue(1., where=(m.x < 1) & (m.y < 1)) > (fp.TransientTerm() == fp.DiffusionTerm()).solve(v, dt=0.1) > > from fipy.tools.debug import PRINT > PRINT(v.value, stall=False) > > It is clear that the grid is not communicating in the top to bottom > direction in parallel. > > > If not, how could periodic > > boundary conditions be implemented? > > Ideally, Gmsh would do this for us, but I don't believe that it has > any notion of periodicity. To make it work for the "sliced > partitioned" grids in FiPy is probably not that difficult, but I don't > have time to do this right now. It is just a question of extending the > overlaps to join along the top and bottom of the mesh. The left-right > overlaps already work being on the same processor since the > partitioning is only along the y-axis. > > Hope that helps. > > -- > Daniel Wheeler > _______________________________________________ > fipy mailing list > [email protected] > http://www.ctcms.nist.gov/fipy > [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ] > > > > -- > Jane Hung > Graduate Student | MIT Department of Chemical Engineering > Hatton Lab 66-325 | Doyle Lab E18-509 > [email protected] | 415.952.6325 > > > > -- > Jane Hung > Graduate Student | MIT Department of Chemical Engineering > Hatton Lab 66-325 | Doyle Lab E18-509 > [email protected] | 415.952.6325 > _______________________________________________ > fipy mailing list > [email protected] > http://www.ctcms.nist.gov/fipy > [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ] _______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
