---

Ma27SolverBase Class Reference
[LinearSolvers]

#include <Ma27Solver.h>

Inheritance diagram for Ma27SolverBase:

List of all members.

Public Methods
	Ma27SolverBase (int n, int nnz)
double	thresholdPivoting ()
double	treatAsZero ()
virtual void	copyMatrixElements (double fact[], int lfact)=0
virtual void	getIndices (int irowM[], int jcolM[])=0
virtual void	diagonalChanged (int idiag, int extent)
virtual void	matrixChanged ()
virtual void	basicSolve (double drhs[], int n)
virtual void	solve (OoqpVector &rhs)=0
virtual	~Ma27SolverBase ()
Protected Methods
int	ierror ()
virtual void	firstCall ()
Protected Attributes
double	precision
int *	irowM
int *	jcolM
double *	fact
int	n
int	nnz
int	la
int *	ikeep
int *	iw
int	liw
int *	iw1
int *	iw2
int	nsteps
int	maxfrt
double *	w
double	ipessimism
double	rpessimism

---

Detailed Description

implements the linear solver class using the HSL MA27 solver

---

Constructor & Destructor Documentation

Ma27SolverBase::Ma27SolverBase (  int    n, int    nnz )

base class constructor. Allocates values for kTreatAsZero, kThresholdPivoting, kThresholdPivotingMax, kThresholdPivotingFactor, kPrecision, ipessimism,rpessimism

virtual Ma27SolverBase::~Ma27SolverBase (    )  [virtual]

destructor

---

Member Function Documentation

virtual void Ma27SolverBase::basicSolve (  double    drhs[], int    n )  [virtual]

version of the main solve routine that takes argument as an array of doubles Parameters: drhs  on input contains the right-hand side; on output contains the solution n  dimension of the system

virtual void Ma27SolverBase::copyMatrixElements (  double    fact[], int    lfact )  [pure virtual]

copy elements from matrix into the fact data structure, in preparation for factorization (or refactorization). Implemented in Ma27Solver.

virtual void Ma27SolverBase::diagonalChanged (  int    idiag, int    extent )  [virtual]

called if the diagonal elements of the matrix have changed. Triggers a refactorization of the matrix, if necessary. Parameters: idiag  index of the first diagonal element that changed extent  the number of diagonal element that changed. Implements DoubleLinearSolver.

virtual void Ma27SolverBase::firstCall (    )  [protected, virtual]

called the very first time a matrix is factored. Allocates space for the factorization and performs ordering

virtual void Ma27SolverBase::getIndices (  int    irowM[], int    jcolM[] )  [pure virtual]

change format for row/column index matrices, in preparation for call to MA27 FORTRAN routines Parameters: irowM  array of nnz elements indicating row index (in range 1..n) of the corresponding matrix element jcolM  array of nnz elements indicating col index (in range 1..n) of the corresponding matrix element Implemented in Ma27Solver.

int Ma27SolverBase::ierror (    )  [inline, protected]

ierror, used by ma27 to send context-specific supplimental info to the user

virtual void Ma27SolverBase::matrixChanged (    )  [virtual]

called if some elements of the matrix have changed. Triggers a refactorization of the matrix, if necessary. Implements DoubleLinearSolver.

virtual void Ma27SolverBase::solve (  OoqpVector &    rhs )  [pure virtual]

version of the main solve routine that takes argument as an OoqpVector Parameters: drhs  on input contains the right-hand side; on output contains the solution n  dimension of the system Implements DoubleLinearSolver. Implemented in Ma27Solver.

double Ma27SolverBase::thresholdPivoting (    )  [inline]

the Threshold Pivoting parameter, stored as U in the ma27dd common block. Takes values in the range [0,1]. Larger values enforce greater stability in the factorization as they insist on larger pivots. Smaller values preserve sparsity at the cost of using smaller pivots.

double Ma27SolverBase::treatAsZero (    )  [inline]

the "Treat As Zero" parameter, stored as pivtol in the common block ma27td. The factorization will not accept a pivot whose absolute value is less than this parameter as a 1x1 pivot or as the off-diagonal in a 2x2 pivot.

---

Member Data Documentation

double* Ma27SolverBase::fact [protected]

nonzero element of the factors

int* Ma27SolverBase::ikeep [protected]

pivot sequence and temporary storage information

double Ma27SolverBase::ipessimism [protected]

amounts by which to increase allocated factorization space when inadequate space is detected. ipessimism is for array "iw", rpessimism is for the array "fact".

int* Ma27SolverBase::irowM [protected]

index array for the factorization

int * Ma27SolverBase::iw [protected]

pivot sequence and temporary storage information

int * Ma27SolverBase::iw1 [protected]

pivot sequence and temporary storage information

int * Ma27SolverBase::iw2 [protected]

pivot sequence and temporary storage information

int * Ma27SolverBase::jcolM [protected]

index array for the factorization

int Ma27SolverBase::la [protected]

length of the array containing factors; may be increased during the numerical factorization if the estimate obtained during the symbolic factorization proves to be inadequate.

int Ma27SolverBase::liw [protected]

pivot sequence and temporary storage information

int Ma27SolverBase::maxfrt [protected]

pivot sequence and temporary storage information

int Ma27SolverBase::n [protected]

dimension of the whole matrix

int Ma27SolverBase::nnz [protected]

number of nonzeros in the matrix

int Ma27SolverBase::nsteps [protected]

pivot sequence and temporary storage information

double Ma27SolverBase::precision [protected]

precision we demand from the linear system solver. If it isn't attained on the first solve, we use iterative refinement and possibly refactorization with a higher value of kThresholdPivoting.

double Ma27SolverBase::rpessimism [protected]

amounts by which to increase allocated factorization space when inadequate space is detected. ipessimism is for array "iw", rpessimism is for the array "fact".

double* Ma27SolverBase::w [protected]

temporary storage for the factorization

---

The documentation for this class was generated from the following file:

• Ma27Solver.h

---