> On May 20, 2016, at 10:11 AM, Juha Jaykka <[email protected]> wrote: > >> not. This usually indicates a bad guess for Newton. I would now advocate > > Oh, did I forget to mention, that other initial conditions DO converge. The > funny thing is the initial condition which does NOT converge is actually > CLOSER to the solution than ones which DO. Had it been the other way around, > I > probably would not have thought twice of it.
Yeah, "closeness" for convergence of Newton's method can be rather complicated. Send the code. Python though it may be. Barry > >> 1) Grid sequencing: This is easy if you use a DMDA. You just use >> -snes_grid_sequence >> and its automatic. Since you report that smaller grids converge, this >> is usually enough. > > Unfortunately, no. > >> 2) NPC: If the above fails, I would try preconditioning Newton with GS. >> You can do this >> just using multiplicative composition, or left preconditioning. > > I'm not sure what you mean here. Do you mean I should run normal newton line > search snes with SNESSetNPC(snes_of_type_gs)? > > If so, does the inner snes_of_type_gs and the outer snes_of_type_newtonls get > all the same routines? I guess so, but just checking. > >> If 1 works, then its likely that FAS would work even better. > > I wish it was so easy: I did try all the snes types before posting the first > post. Also all KSP types. Except those that need things I don't have, of > course. > > Cheers, > Juha
