[dune-pdelab] Dune reported error in newton.hh
Markus Blatt
markus at dr-blatt.de
Fri Jan 6 20:54:14 CET 2012
Hey Jared,
On Fri, Jan 06, 2012 at 04:31:31PM +0100, Jared Okiro wrote:
> I have used the example DG code (Symmetric Weighted Interior Penalty
> Galerkin(SWIPG)) in pdelab/localoperator/convectiondiffusiondg.hh to
> solve a scalar nonlinear reaction-diffusion equation and it worked
> well.
> I have extended the code to solve a two component nonlinear
> reaction-diffusion system. However, at run time I get the error
> below: This error occurs when I evaluate the jacobian manually and
> also when I use numerical differentiation to determine the jacobian.
>
> /Dune reported error: NewtonLinearSolverError [linearSolve:/homes/okiro/DUNE/dune-svn/dune-pdelab/dune/pdelab/backend/../newton/newton.hh:174]:
> NewtonSolver::linearSolve(): Linear solver did not converge in 1
> iterations./
>
I would guess that the linear solver has real problems solving your
equation. As it tried only 1 step, I assume that it somehow broke
down.
> When I print the matrix of the system with the coarse mesh and few
> iterations, I see that the matrix assembled is for the initial
> condition for each of the children and this does not change with
> even slight more iterations. The matrix looks like this:
>
> const residual [blocks=8,dimension=32]
> row 0 9.0e-01 9.0e-01 9.0e-01 9.0e-01
> row 4 9.0e-01 9.0e-01 9.0e-01 9.0e-01
> row 8 9.0e-01 9.0e-01 9.0e-01 9.0e-01
> row 12 9.0e-01 9.0e-01 9.0e-01 9.0e-01
> row 16 9.5e-01 9.5e-01 9.5e-01 9.5e-01
> row 20 9.5e-01 9.5e-01 9.5e-01 9.5e-01
> row 24 9.5e-01 9.5e-01 9.5e-01 9.5e-01
> row 28 9.5e-01 9.5e-01 9.5e-01 9.5e-01
>
How did you print the matrix? It looks like it contains zero rows
(only each 4th row has nonzeros). Additionally the dimension 32x4
seems weired.
Markus
--
Do you need more support with DUNE or HPC in general?
Dr. Markus Blatt - HPC-Simulation-Software & Services http://www.dr-blatt.de
Rappoltsweilerstr. 5, 68229 Mannheim, Germany
Tel.: +49 (0) 160 97590858 Fax: +49 (0)322 1108991658
More information about the dune-pdelab
mailing list