[petsc-dev] KSP rtol

[Barry Smith]

  BTW: petsc-dev has the new option SNESSetPicard() which can be used to to make SNES a Picard solver and it always uses the Jed-correct defect correction approach so there is no reason to over solve the linear system. See its manual page.

   Barry

On Dec 14, 2011, at 9:29 PM, Jed Brown wrote:

> On Wed, Dec 14, 2011 at 13:50, Mark F. Adams <mark.adams at columbia.edu> wrote:
> What I am doing is classic inexact Newton (or it looks like these folks are doing Picard).
> 
> It appears that they are doing Picard in one-shot mode instead of in defect correction mode. If you write it as a defect correction, then the initial guess will be zero and the standard tolerance is the correct one. I maintain that defect correction is better in every way for solvers, and it naturally gives you a nonlinear residual so that you can use a reliable nonlinear solver tolerance.