hypre/docs/usr_fei_solvers.tex
2007-01-24 16:03:31 +00:00

214 lines
8.5 KiB
TeX

\section{FEI Solvers}
\label{LSI_solvers}
%% \subsection{Solution Process in the FEI}
After the {\tt FEI} has been used to assemble the global linear system
(as described in Chapter \ref{ch-FEI}), a number of \hypre{} solvers
can be called to perform the solution.
This is straightforward, if \hypre's {\tt FEI} has been used.
If an external {\tt FEI} is employed, the user needs to link with \hypre's
implementation of the \code{LinearSystemCore} class, as described
in Section \ref{LSI_install}.
Solver parameters are specified as an array of strings, and a complete
list of the available options can be found in the FEI section of the reference
manual.
They are passed to the {\tt FEI} as in the following example:
\begin{display}
\begin{verbatim}
nParams = 5;
paramStrings = new char*[nParams];
for (i = 0; i < nParams; i++) }
paramStrings[i] = new char[100];
strcpy(paramStrings[0], "solver cg");
strcpy(paramStrings[1], "preconditioner diag");
strcpy(paramStrings[2], "maxiterations 100");
strcpy(paramStrings[3], "tolerance 1.0e-6");
strcpy(paramStrings[4], "outputLevel 1");
feiPtr -> parameters(nParams, paramStrings);
\end{verbatim}
\end{display}
To solve the linear system of equations, we call
\begin{display}
\begin{verbatim}
feiPtr -> solve(&status);
\end{verbatim}
\end{display}
where the returned value \code{status} indicates whether the solve was successful.
Finally, the solution can be retrieved by the following function call:
\begin{display}
\begin{verbatim}
feiPtr -> getBlockNodeSolution(elemBlkID, nNodes, nodeIDList,
solnOffsets, solnValues);
\end{verbatim}
\end{display}
where \code{nodeIDList} is a list of nodes in element block {\tt elemBlkID}, and
\code{solnOffsets[i]} is the index pointing to the first location
where the variables at node $i$ is returned in \code{solnValues}.
\subsection{Solvers Available Only through the FEI}
While most of the solvers from the previous sections are available through
the FEI interface, there are number of additional solvers and preconditioners
that are accessible only through the FEI.
These solvers are briefly described in this section (see also the reference manual).
%On the other side of the {\sf LinearSystemCore} interface is the linear
%solver package. Any linear solver package can be used as long as there
%is an implementation of the interface in the solver package itself. The
%{\sf LinearSystemCore} implementation in \hypre{} is encapsulated in the
%\hypre{} finite element conceptual interface called {\sf HYPRE\_LinSysCore}.
%The incoming element stiffness matrices and other information are
%converted into \hypre{}'s internal data structures.
%\subsection{Parallel Matrix and Vector Construction}
%
%The matrix class in \hypre{} accessible via the {\sf LinearSystemCore} interface
%is the parallel compressed sparse row ({\sf ParCSR}) matrix. The
%requirements about how the global matrix is partitioned among the
%processors are that each processor holds a contiguous block of rows and columns
%and the equation numbers in processors of lower rank are lower than those
%in processors of higher rank. The {\sf FEI} is responsible for ensuring
%that these two requirements are followed. The matrix can be loaded in
%parallel - a row or a block of rows at a time. The solution and right
%hand side vectors are constructed accordingly. The matrix rows corresponding
%to the shared nodes can be assigned to either processor, and is determined
%by the {\sf FEI} itself. Once the incoming matrix and vector data have
%been captured in the \hypre{} {\sf ParCSR} format, a whole of matrix and
%vector operators are available for use in the \hypre{} solvers.
%
\subsubsection{Sequential and Parallel Solvers}
\hypre{} currently has many iterative solvers. There is also internally a
version of the sequential {\sf SuperLU} direct solver (developed at U.C.
Berkeley) suitable to small problems (may be up to the size of $10000$).
In the following we list some of these internal solvers.
\begin{enumerate}
\item Additional Krylov solvers (FGMRES, TFQMR, symmetric QMR),
\item SuperLU direct solver (sequential),
\item SuperLU direct solver with iterative refinement (sequential),
\end{enumerate}
\subsubsection{Parallel Preconditioners}
The performance of the Krylov solvers can be improved by clever selection
of preconditioners. Besides those mentioned previously
in this chapter, the following preconditioners are
available via the {\sf LinearSystemCore} interface:
\begin{enumerate}
\item the modified version of MLI, which requires the finite element substructure matrices
to construct the prolongation operators,
\item parallel domain decomposition with inexact local solves ({\sf DDIlut}),
\item least-squares polynomial preconditioner,
\item $2 \times 2$ block preconditioner, and
\item $2 \times 2$ Uzawa preconditioner.
\end{enumerate}
Some of these preconditioners can be tuned by a number of internal parameters
modifiable by users. A description of these parameters is given in the reference manual.
\subsubsection{Matrix Reduction}
For some structural mechanics problems with multi-point constraints the
discretization matrix is indefinite (eigenvalues lie in both sides of
the imaginary axis). Indefinite matrices are much more difficult to solve
than definite matrices. Methods have been developed to reduce these
indefinite matrices to definite matrices. Two matrix reduction algorithms
have been implemented in \hypre{}, as presented in the following subsections.
\subsubsection{Schur Complement Reduction}
The incoming linear system of equations is assumed to be in the form :
\[ \left[
\begin{array}{cc}
D & B \\
B^T & 0
\end{array}
\right]
\left[
\begin{array}{c}
x_1 \\
x_2
\end{array}
\right]
=
\left[
\begin{array}{c}
b_1 \\
b_2
\end{array}
\right]
\]
where $D$ is a diagonal matrix. After Schur complement reduction is applied,
the resulting linear system becomes
$$
- B^T D^{-1} B x_2 = b_2 - B^T D^{-1} b_1.
$$
\subsubsection{Slide Surface Reduction}
With the presence of slide surfaces, the matrix is in the same form as in the
case of Schur complement reduction. Here $A$ represents the relationship
between the master, slave, and other degrees of freedom. The matrix block
$[B^T 0]$ corresponds to the constraint equations. The goal of reduction
is to eliminate the constraints. As proposed by Manteuffel, the trick is
to re-order the system into a $3 \times 3$ block matrix.
\[
\left[
\begin{array}{ccc}
A_{11} & A_{12} & N \\
A_{21} & A_{22} & D \\
N_{T} & D & 0 \\
\end{array}
\right]
=
\left[
\begin{array}{ccc}
A_{11} & \hat{A}_{12} \\
\hat{A}_{21} & \hat{A}_{22}.
\end{array}
\right]
\]
The reduced system has the form :
$$
(A_{11} - \hat{A}_{21} \hat{A}_{22}^{-1} \hat{A}_{12}) x_1 =
b_1 - \hat{A}_{21} \hat{A}_{22}^{-1} b_2,
$$
which is symmetric positive definite (SPD) if the original matrix is PD.
In addition, $\hat{A}_{22}^{-1}$ is easy to compute.
There are three slide surface reduction algorithms in \hypre{}.
The first follows the matrix formulation in this section. The second is
similar except that it replaces the eliminated slave equations with
identity rows so that the degree of freedom at each node is preserved.
This is essential for certain block algorithms such as the smoothed
aggregation multilevel preconditioners.
The third is similar to the second except that it is more general and
can be applied to problems with intersecting slide surfaces (sequential only
for intersecting slide surfaces).
\subsubsection{Other Features}
To improve the efficiency of the \hypre{} solvers, a few other features have been
incorporated. We list a few of these features below :
\begin{enumerate}
\item Preconditioner reuse - For multiple linear solves with matrices that are
slightly perturbed from each other, oftentimes the use of the same
preconditioners can save preconditioner setup times but suffer little
convergence rate degradation.
\item Projection methods - For multiple solves that use the same matrix,
previous solution vectors can sometimes be used to give a better initial
guess for subsequent solves. Two projection schemes have been implemented
in \hypre{} - A-conjugate projection (for SPD matrices) and minimal residual
projection (for both SPD and non-SPD matrices).
\item The sparsity pattern of the matrix is in general not destroyed after
it has been loaded to an \hypre{} matrix. But if the matrix is not to
be reused, an option is provided to clean up this pattern matrix to
conserve memory usage.
\end{enumerate}