Anasazi::LOBPCGSolMgr< ScalarType, MV, OP > Class Template ReferenceThe Anasazi::LOBPCGSolMgr provides a powerful and fully-featured solver manager over the LOBPCG eigensolver.
[Anasazi Eigensolver Framework]
Inheritance diagram for Anasazi::LOBPCGSolMgr< ScalarType, MV, OP >:
List of all members.
template<class ScalarType, class MV, class OP>
The Anasazi::LOBPCGSolMgr provides a powerful and fully-featured solver manager over the LOBPCG eigensolver.
class Anasazi::LOBPCGSolMgr< ScalarType, MV, OP >
This solver manager exists to provide a flexible manager over the Anasazi::LOBPCG eigensolver intended for general use. Features provided by this solver manager include:
- locking of converged eigenpairs
- global convergence on only the significant eigenpairs (instead of any eigenpairs with low residual)
- recovery from Anasazi::LOBPCGRitzFailure when full orthogonalization is disabled
These options are all driven by a list of parameters handed to the solver manager at construction. For more information, see Anasazi::LOBPCGSolMgr::LOBPCGSolMgr().
- Chris Baker, Ulrich Hetmaniuk, Rich Lehoucq, Heidi Thornquist
Definition at line 84 of file AnasaziLOBPCGSolMgr.hpp.
Constructor & Destructor Documentation
Basic constructor for LOBPCGSolMgr.
This constructor accepts the Eigenproblem to be solved in addition to a parameter list of options for the solver manager. These options include the following:
"Which" - a
string specifying the desired eigenvalues: SM, LM, SR or LR. Default: "SR"
"Block Size" - a
int specifying the block size to be used by the underlying LOBPCG solver. Default: problem->getNEV()
"Full Ortho" - a
bool specifying whether the underlying solver should employ the full orthogonalization scheme. Default: true
"Recover" - a
bool specifying whether the solver manager should attempt to recover in the case of a LOBPCGRitzFailure when full orthogonalization is disabled. Default: true
"Maximum Iterations" - a
int specifying the maximum number of iterations the underlying solver is allowed to perform. Default: 100
"Verbosity" - a sum of MsgType specifying the verbosity. Default: Anasazi::Errors
"Convergence Tolerance" - a
MagnitudeType specifying the level that residual norms must reach to decide convergence. Default: machine precision.
"Relative Convergence Tolerance" - a
bool specifying whether residuals norms should be scaled by their eigenvalues for the purposing of deciding convergence. Default: true
"Use Locking" - a
bool specifying whether the algorithm should employ locking of converged eigenpairs. Default: false
"Max Locked" - a
int specifying the maximum number of eigenpairs to be locked. Default: problem->getNEV()
"Locking Quorum" - a
int specifying the number of eigenpairs that must meet the locking criteria before locking actually occurs. Default: 1
"Locking Tolerance" - a
MagnitudeType specifying the level that residual norms must reach to decide locking. Default: 0.1*convergence tolerance
"Relative Locking Tolerance" - a
bool specifying whether residuals norms should be scaled by their eigenvalues for the purposing of deciding locking. Default: true
"Init" - a LOBPCGState<ScalarType,MV> struct used to initialize the LOBPCG eigensolver.
Definition at line 186 of file AnasaziLOBPCGSolMgr.hpp.
Member Function Documentation
This method performs possibly repeated calls to the underlying eigensolver's iterate() routine until the problem has been solved (as decided by the solver manager) or the solver manager decides to quit.
This method calls LOBPCG::iterate(), which will return either because a specially constructed status test evaluates to Passed or an exception is thrown.
A return from LOBPCG::iterate() signifies one of the following scenarios:
- the maximum number of iterations has been exceeded. In this scenario, the solver manager will place
all converged eigenpairs into the eigenproblem and return Unconverged.
- the locking conditions have been met. In this scenario, some of the current eigenpairs will be removed
from the eigensolver and placed into auxiliary storage. The eigensolver will be restarted with the remaining
eigenpairs and some random information to replace the removed eigenpairs.
- global convergence has been met. In this case, the most significant NEV eigenpairs in the solver and locked storage
have met the convergence criterion. (Here, NEV refers to the number of eigenpairs requested by the Eigenproblem.)
In this scenario, the solver manager will return Converged.
- an LOBPCGRitzFailure exception has been thrown. If full orthogonalization is enabled and recovery from this exception
is requested, the solver manager will attempt to recover from this exception by gathering the current eigenvectors,
preconditioned residual, and search directions from the eigensolver, orthogonormalizing the basis composed of these
three, projecting the eigenproblem, and restarting the eigensolver with the solution of the project eigenproblem. Any
additional failure that occurs during this recovery effort will result in the eigensolver returning Unconverged.
- ReturnType specifying:
- Converged: the eigenproblem was solved to the specification required by the solver manager.
- Unconverged: the eigenproblem was not solved to the specification desired by the solver manager
Implements Anasazi::SolverManager< ScalarType, MV, OP >.
Definition at line 280 of file AnasaziLOBPCGSolMgr.hpp.
The documentation for this class was generated from the following file:
Generated on Thu Sep 18 12:31:39 2008 for Anasazi by 220.127.116.11