nonlinear solver api

[GoogleCodeExporter]

Instead of an evaluator instance, pass to nonlinear solvers a function
returning the residual and (optionally) a function computing the jacobian.

This has two main aims:
1. simplification: evaluator is a sfepy concept, not needed in an abstract
nonlinear solver setting - just the functions are needed.
2. quasi-newton (and other) methods do not need the jacobian at all.

Then we could interface some solvers from scipy.optimize.

Original issue reported on code.google.com by robert.c...@gmail.com on 28 Jan 2009 at 4:54

[GoogleCodeExporter]

The API is changed now, see [1]:

NonlinearSolver.__call__( self, state0, conf = None, fun = None, fun_grad = 
None,
                          lin_solver = None, status = None )

The newton solver (nls.py) works well. The Oseen solver has not been yet 
migrated,
but this is another issue.

[1] 
http://git.sympy.org/?p=sfepy.git;a=commit;h=a91176bbe15fcae4d4f3941bfcb0154f013
fbdbd

Original comment by robert.c...@gmail.com on 6 Mar 2009 at 9:27

• Changed state: Fixed

[GoogleCodeExporter]

Migrated to http://github.com/sfepy/sfepy/issues/81

Original comment by robert.c...@gmail.com on 30 Jan 2012 at 10:25

• Changed state: Migrated