jvx.numeric

Class PnConjugateGradientMatrix

java.lang.Object

jvx.numeric.PnConjugateGradientMatrix

Direct Known Subclasses:

PnBiconjugateGradient

public class PnConjugateGradientMatrix
extends java.lang.Object

Solve symmetric sparse linear equation systems using preconditioned conjugate gradient method.

Author:

Jonas Hurrelmann

Version:

09.12.2010, 1.01 revised (fk) Removed unused field m_eps to avoid confusion with m_tol. 02.12.2005, 1.00 created (jh)

Constructor Summary

Constructors

Constructor and Description
PnConjugateGradientMatrix() Constructor.

Method Summary

Modifier and Type	Method and Description
boolean	converged() Returns whether the residual norm is within the tolerance.
double	getActualEstimatedError() Get estimated error of last solve call.
int	getActualNumOfTakenIterations() Get number of iterations used in last solve call.
double	getConvergenceTest() There are four different methods measure the convergence of the solution of the linear equation used in solve.
int	getDimension() Get dimension.
int	getMaxNumIterations() Get maximum number of allowed iterations.
PnPreconditioner	getPreconditioner() Set the preconditioner to use for the conjugate gradient.
double	getTolerance() Get desired convergence tolerance of solve.
void	setConvergenceTest(int aTest) There are four different methods used to measure the convergence of the solution of the linear equation in solve, the default is 1.
void	setDimension(int dim) Set dimension of surface matrix is based on.
void	setMaxNumIterations(int itmax) Set maximum number of allowed iterations.
void	setPreconditioner(PnPreconditioner prec) Set the preconditioner to use for the conjugate gradient.
void	setTolerance(double tol) Set desired convergence tolerance of solve.
double	solve(PnSparseMatrix smat, PdVector x, PdVector b) Performs a conjugate gradient on a sparse matrix to solve the linear system Ax = b.
void	updatePreconditioner(PnSparseMatrix smat) When the sparse matrix to solve has changed update the preconditioner with this method.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

PnConjugateGradientMatrix

public PnConjugateGradientMatrix()

Constructor.

Method Detail

converged

public boolean converged()

Returns whether the residual norm is within the tolerance.

Returns:

true if convergence happened within tolerance.

setPreconditioner

public void setPreconditioner(PnPreconditioner prec)

Set the preconditioner to use for the conjugate gradient.

Parameters:

prec -

getPreconditioner

public PnPreconditioner getPreconditioner()

Set the preconditioner to use for the conjugate gradient.

getActualEstimatedError

public double getActualEstimatedError()

Get estimated error of last solve call.

getActualNumOfTakenIterations

public int getActualNumOfTakenIterations()

Get number of iterations used in last solve call.

setMaxNumIterations

public void setMaxNumIterations(int itmax)

Set maximum number of allowed iterations. Default value is 10000.

getMaxNumIterations

public int getMaxNumIterations()

Get maximum number of allowed iterations.

setTolerance

public void setTolerance(double tol)

Set desired convergence tolerance of solve.

getTolerance

public double getTolerance()

Get desired convergence tolerance of solve.

setDimension

public void setDimension(int dim)

Set dimension of surface matrix is based on. Coordinate vectors must have length equal dim*(dim of sparseMatrix).

getDimension

public int getDimension()

Get dimension.

setConvergenceTest

public void setConvergenceTest(int aTest)

There are four different methods used to measure the convergence of the solution of the linear equation in solve, the default is 1.

 
 1. |A*x-b|/|b|<tol<br>
 2. |Diag(A)^-1*x-b|/|Diag(A)^-1*b|<tol<br>
 With Diag(A) denoting the diagonal part of A.<br>
 3. This routine uses its own estimate of error in x, and
 requires its magnitude, divided by the magnitude of x, to
 be less than tol.<br>
 4. Same as 3., except that the largest (in absolute value)
 component of the error and the largest component of x are
 used instead of the vector magnitude.
 

getConvergenceTest

public double getConvergenceTest()

There are four different methods measure the convergence of the solution of the linear equation used in solve. Get the index of the method actually used.

See Also:

setConvergenceTest(int)

updatePreconditioner

public void updatePreconditioner(PnSparseMatrix smat)

When the sparse matrix to solve has changed update the preconditioner with this method.

solve

public double solve(PnSparseMatrix smat,
                    PdVector x,
                    PdVector b)

Performs a conjugate gradient on a sparse matrix to solve the linear system Ax = b. The matrix has to be square.

Returns:

size of residual

Version:

02.05.2008, 1.10 revised (fk) speedup. 02.12.2005, 1.00 created (jh)