Choosing Solver

[Kun-Dar Li]

Hi Daniel,
Do you have more clues or informations for choosing Solver? Why LU, not PCG, 
or others? I cann't find more detailed information about Solvers that were 
included in FiPy.
May I get a totally different result by using different solver? Thank you so 
much for your help.

Kundar


----- Original Message ----- 
From: "Daniel Wheeler" <daniel.wheel...@gmail.com>
To: "Multiple recipients of list" <fipy@nist.gov>
Sent: Wednesday, July 08, 2009 11:15 AM
Subject: Re: Error messages



Hi Kun-Dar,

Reply below.

2009/7/7 Kun-Dar Li <kun...@umich.edu>:
Hello all,

I ran FiPy under RedHat 5 to solve 4th-order PDE with anisotropic 
diffusion
coefficient and periodic boundary in 2D model. During the first few steps,
it shows:

-----------------------------------------------------------------------------------------------------------------------------------------------------------
/usr/lib/python2.4/site-packages/numpy/core/ma.py:609: UserWarning:
Cannot automatically convert masked array to numeric because data
is masked in one or more locations.
warnings.warn("Cannot automatically convert masked array to "\
/usr/lib/python2.4/site-packages/FiPy-2.0.2-py2.4.egg/fipy/solvers/pysparse/pysparseSolver.py:51:

Just a warning, don't worry about it.

DeprecationWarning: 'iterations' should be used instead of
'steps'

Again, just a warning about an argument name change, don't worry about it.

Solver.__init__(self, *args, **kwargs)
regularapatite.py:115: DeprecationWarning: 'Viewer' should be used
instead of 'make'
viewer = fipy.viewers.make(vars = var, limits = {'datamin': 0.,
'datamax': 1.})

Again, just a warning. To get rid of these just update your script
based on the message so, "make" can be changed to "Viewer", but not
really important.

regularapatite.py:136: MaximumIterationWarning: Iterations: 15001.
Relative error: 5.96005

The solver is not converging. This is bad. First off, try using the LU
solver on a small problem to make sure things are working. Once you
are sure thing are working you can figure out which iterative solver
to use for bigger systems.

eqch.solve(var, solver = solver, dt = dt)
step 0 dt 0.5
regularapatite.py:136: StagnatedSolverWarning: The solver stagnated.
Iterations: 1. Relative error: nan
eqch.solve(var, solver = solver, dt = dt)
step 1 dt 0.5
step 2 dt 0.5
step 3 dt 0.5

---------------------------------------------------------------------------------------------------------------------
I guessed there are four errors. But, I'm not sure how to fix it. Can
someone explain me what's the meanings about these messages? Or, I can not
solve the anisotropic diffusion coefficient under periodic boundary?

You should be able to. This is something we have never tried, so I
cannot be sure. There are certainly a lot of complications with
boundary conditions and anisotropy, but shouldn't be for periodic
boundaries.

Thank
you so much for any help.





--
Daniel Wheeler