On 3/30/15 9:32 PM, Andrew Spott wrote:
I’m sorry, I didn’t want to imply you were taking too much time. I was just interested in trying to solve the problem myself, especially if it was a simpler fix.
Nothing of that sort. I just expected to figure it out sooner, especially when you were kind enough to produce a very simple case and an external Crank–Nicolson. We welcome any help. However, as far as I can tell the issue may not be in the way the theta.c is implemented.
Emil
I don’t have enough of an understanding of the internals of PETSc though, if it depends on the compiler flags. Good luck! -Andrew — Andrew On Mon, Mar 30, 2015 at 7:23 PM, Emil Constantinescu <emcon...@mcs.anl.gov <mailto:emcon...@mcs.anl.gov>> wrote: Hi Andrew, Sorry for the delay; things are going slower with me over the spring break but we are working on it. The problem that you see may not be related to the integrator, which runs well (we get expected convergence rates and good results) for other problems, in particular, for problems with real variables (and no c++11). So the high level th->endpoint implementation seems fine. For some reason the linear solver is misbehaving when using cn. If you use option -snes_mf the problem disappears. Please do not use -ts_theta_adapt until it is fixed. In conclusion I don't think that looking at theta.c will be productive. We are currently looking at what happens at the lower levels. Thanks, Emil On 3/30/15 4:25 PM, Andrew Spott wrote: > Emil: If I want to fix this myself, do you have any idea where I should > start looking? I assume in the th->endpoint section of TSStep_Theta in > http://www.mcs.anl.gov/petsc/petsc-current/src/ts/impls/implicit/theta/theta.c.html? > > Thanks, > > -Andrew > > >
