On Dec 11, 2012, at 3:15 PM, Hesameddin Ilatikhameneh <hilatikh at purdue.edu> wrote:
> Using -ksp_type preonly -pc_type lu didn't change the number of required > newton iterations. http://www.mcs.anl.gov/petsc/documentation/faq.html#newton > > I don't understand what you mean by "then backoff". > > Thanks > Hesam > ----- Original Message ----- > From: "Matthew Knepley" <petsc-maint at mcs.anl.gov> > To: "petsc-maint" <petsc-maint at mcs.anl.gov>, "Hesameddin Ilatikhameneh" > <hilatikh at purdue.edu> > Cc: "For users of the development version of PETSc" <petsc-dev at mcs.anl.gov> > Sent: Tuesday, December 11, 2012 4:09:46 PM > Subject: Re: [petsc-maint #143637] SNES takes 50 newton iterations > > On Tue, Dec 11, 2012 at 1:02 PM, Hesameddin Ilatikhameneh > <hilatikh at purdue.edu> wrote: >> Hello, >> >> I want to minimize a simple energy equation (E = E(x0,x1,...,xN)) using >> nonlinear solver SNES. This way SNES equation F(x) = dE/dxi = 0 and jacobian >> is J = dF/dx = d2E/dx2. I have attached a report explaining the problem and >> how I implemented it. >> >> The energy is related to 1D chain of balls connected with springs and the >> final answer is that all balls spaced equal to ideal distance. >> >> My problem is that SNES takes near 30-50 newton iterations to converge which >> seems to be too much for such a simple problem. >> I have also attached the code and appreciate your help on this. > > Start with -ksp_type preonly -pc_type lu, and then backoff. This gives > you the true Newton > convergence rate. > > Matt > >> Thanks >> Hesam >> > > > > -- > What most experimenters take for granted before they begin their > experiments is infinitely more interesting than any results to which > their experiments lead. > -- Norbert Wiener >
