On Thu, 10 Mar 2016, Adriano Côrtes wrote:
*1. Why only the timestep 0 printed the line bellow?*
Nonlinear step: |du|/|u| = 0.99946, |du| = 194.712
That's only printed for intermediate nonlinear steps, IIRC. In later
time steps the problem is linear enough to converge in one
quasi-Newton step.
*2. Why the number of linear iterations is so different from the libmesh
page and my run?
Probably those were generated with different PETSc solver options. We
pull $LIBMESH_OPTIONS from your environment, and some of the libMesh
developers habitually keep it set to more robust or to faster values
(for their problems) than the PETSc defaults.
Also the finite elements types? *
Libmesh page:
Finite Element Types="LAGRANGE", "JACOBI_20_00" "LAGRANGE", "JACOBI_20_00"
Infinite Element Mapping="CARTESIAN" "CARTESIAN"
Approximation Orders="SECOND", "THIRD" "FIRST", "THIRD"
This was generated with infinite element support enabled; looks like
your output was generated with it disabled.
*3. Where in Libmesh the PETSc class SNES is used?
In PetscNonlinearSolver for NonlinearImplicitSystem or in
PetscDiffSolver for FEMSystem.
Or better asking does FEMSystem communicates in the solver level
with SNES, or is only in the KSP level?*
Either. With PetscDiffSolver it uses SNES, with NewtonSolver it just
uses KSP.
---
Roy
------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785111&iu=/4140
_______________________________________________
Libmesh-users mailing list
Libmesh-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-users
