On Tue, Oct 15, 2013 at 6:20 PM, Jed Brown <[email protected]> wrote:
> Christophe Ortiz <[email protected]> writes: > > > Yes ! Actually, that was the problem ! Now I set ARKIMEX1BEE after > > TSSetFromOptions() and it is much faster ! ARKIMEXA2 seems even faster. > > I still recommend configuring this stuff from the command line, and I > would put TSSetFromOptions last so that you can override anything at > run-time. > Ok. > > > Thanks Jed ! > > > > Now I see that the convergence speed of ARKIMEX depends on where I put > > TSSetFromOptions(). > > > > If I do: > > TSCreate(); > > TSSetFromOptions(); > > TSSetProblemType(); > > TSSetType(); > > TSARKIMEXSetType(); > > > > then ARKIMEXA2 is very fast. > > > > Instead, if I put TSSetFromOptions() somewhere else: > > TSCreate(); > > > > TSSetProblemType(); > > TSSetType(); > > *TSSetFromOptions();* > > TSARKIMEXSetType(); > > I would use this order: > > TSCreate(); > TSSetProblemType(); // optional > TSSetType(); > TSARKIMEXSetType(); > TSSetFromOptions(); > > > ARKIMEXA2 slows down significantly. I guess that TSSetFromOptions > overrides > > something...Any idea ? > > Compare the output from -ts_view in both cases. > I found where is the problem. I had set TSARKIMEXSetFullyImplicit(ts,PETSC_TRUE); I found it produces wrong solutions, either with ARKIMEX3, 1BEE or A2. Solution becomes negative. Without this option, all ARKIMEX types work very well and very fast. Christophe
