[dune-pdelab] AMG preconditioner

[Amir Peiraviminaei]

Hi everyone,
I'm using
Dune::PDELab::ISTLBackend_BCGS_AMG_SSOR<IGO> LS
as linear solver within newton method to slove a system of PDEs discretized
with DG method.
the problem is when it build a hierarchy of at least 2 levels, it does not
converge.
this is the message I get:

When AMG preconditioner sets for 2 levels:

TIME STEP [implicit Euler]      1 time (from):   0.0000e+00 dt:
3.1450e+01 time (to):   3.1450e+01
STAGE 1 time (to):   3.1450e+01.
  Initial defect:   1.2878e+01
Using a direct coarse solver (UMFPack)
Building hierarchy of 2 levels (including coarse solver) took 1.351 seconds.
=== Dune::BiCGSTABSolver
=== rate=0.9541, T=72.91, TIT=0.7291, IT=100
Catched Error, Dune reported error: NewtonLinearSolverError
[linearSolve:../dune-pdelab/dune/pdelab/newton/newton.hh:244]:
NewtonSolver::linearSolve(): Linear solver did not converge in 100
iterations



When AMG preconditioner sets for 1 level:

TIME STEP [implicit Euler]      1 time (from):   0.0000e+00 dt:
3.1450e+01 time (to):   3.1450e+01
STAGE 1 time (to):   3.1450e+01.
  Initial defect:   1.7495e+01
Using a direct coarse solver (UMFPack)
Building hierarchy of 1 levels (including coarse solver) took 12.65 seconds.
=== Dune::BiCGSTABSolver
=== rate=7.55e-29, T=0.6516, TIT=inf, IT=0
  Newton iteration  1.  New defect:   4.8970e-03.  Reduction (this):
2.7991e-04.  Reduction (total):   2.7991e-04

Is it possible that this implementation of Linear solver(LS) does not work
with DG method?
How to use linear solvers from dune-solvers module for newton method from
dune-pdelab?

Thank you.
Amir
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune-pdelab/attachments/20201123/1abb3f3e/attachment.htm>