solver_abtb - mixed systems solver (rheolef-7.2)
Contents
Copyright
Copyright (C) 2000-2018 Pierre Saramito <Pierre.Saramito@imag.fr> GPLv3+: GNU GPL version 3 or
later <http://gnu.org/licenses/gpl.html>. This is free software: you are free to change and
redistribute it. There is NO WARRANTY, to the extent permitted by law.
rheolef Version 7.2 solver_abtb(4rheolef)
Description
This class provides both direct and iterative algorithms for solving mixed problem:
[ A B^T ] [ u ] [ Mf ]
[ ] [ ] = [ ]
[ B -C ] [ p ] [ Mg ]
where A is symmetric positive definite and C is symmetric positive. By default, iterative algorithms are
considered for tridimensional problems and direct methods otherwise. A solver_option(4) argument can
change this default behavior. Mixed linear problems appears for instance with the discretization of
Stokes and elasticity problems. The C matrix can be zero and then the corresponding argument can be
omitted when invoking the constructor. Non-zero C matrix appears for of Stokes problems with stabilized
P1-P1 element, or for nearly incompressible elasticity problems.
Recall that, for 1D and 2D problems, the direct method is default, since it is more efficient: see e.g.
usersguide. The solver_option(4) allows one to change this default behavior.
Direct Algorithm
When the kernel of B^T is reduced to zero, the global system is non-singular and the solver(4) method
could be applied. Otherwise, when the kernel of B^T is not reduced to zero, then the pressure p is
defined up to a constant and the system is singular. This is a major difficulty for any direct method.
Thus, the system is first completed by the imposition of an additional constraint on the pressure term:
it should have a zero average value. By this way, the system becomes non-singular and the solver(4)
method could be applied.
Example
See the usersguide for practical examples for the nearly incompressible elasticity, the Stokes and the
Navier-Stokes problems.
Implementation
This documentation has been generated from file linalg/lib/solver_abtb.h
Iterative Algorithm
The cg(5) preconditionned conjugate gradient algorithm or the gmres(5) one is used, depending on the
symmetry of the matrix. The mp matrix is used as preconditionner: it can be customized by the
set_preconditionner member function. The linear sub-systems related to the A matrix are also solved by an
inner solver. This inner solver is automatically defined by default: it can be customized by the
set_inner_solver member function and e.g. uses it own inner iterative algorithm and preconditionner.
Name
solver_abtb - mixed systems solver (rheolef-7.2)
Synopsis
solver_abtb stokes (a,b,mp);
solver_abtb elasticity (a,b,c,mp);
