maspack.solvers

Class KKTSolver

java.lang.Object

maspack.solvers.KKTSolver

public class KKTSolver
extends java.lang.Object

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	KKTSolver.Status Described whether or not a solution was found.

Field Summary

Fields

Modifier and Type	Field and Description
static boolean	computeResidualMG
static boolean	myDebug
static java.lang.String	myQPTestCaseFile
static boolean	useBlockSolves

Constructor Summary

Constructors

Constructor and Description
KKTSolver()
KKTSolver(SparseSolverId solverType)

Method Summary

Modifier and Type	Method and Description
void	analyze(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, int typeM) Performs symbolic analysis on the equality portion of the KKT system defined by matrices M and GT.
void	analyze(VectorNd Mdiag, int sizeM, SparseBlockMatrix GT, VectorNd Rg) Performs symbolic analysis on the equality portion of the KKT system defined by a diagonal M matrix and G.
void	analyzeMG(java.lang.Object M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, int typeM) Performs symbolic analysis on the equality portion of the KKT system defined by M and G.
void	dispose()
void	factor(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg) Does a numeric factorization of a KKT system containg only equality constraints.
void	factor(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, SparseBlockMatrix NT, VectorNd Rn) Does a numeric factorization of a KKT system containing equality and inequality constraints.
void	factor(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, SparseBlockMatrix NT, VectorNd Rn, SparseBlockMatrix DT, VectorNd Rd) Does a numeric factorization of a KKT system containg equality, inequality, and friction constraints.
void	factor(VectorNd Mdiag, int sizeM, SparseBlockMatrix GT, VectorNd Rg) Does a numeric factorization of a KKT system containg only equality constraints.
void	factor(VectorNd Mdiag, int sizeM, SparseBlockMatrix GT, VectorNd Rg, SparseBlockMatrix NT, VectorNd Rn) Does a numeric factorization of a KKT system containg equality and inequality constraints.
KKTSolver.Status	factorAndSolve(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, VectorNd vel, VectorNd lam, VectorNd bm, VectorNd bg, int tolExp) Does a numeric factorization of a KKT system containg only equality constraints.
void	finalize()
VectorNi	getLcpState()
MatrixNd	getLinearMatrix()
DirectSolver	getMatrixSolver()
int	getNumNonZerosInFactors()
void	initialize()
boolean	isFactored()
boolean	lastSolveWasIterative()
void	printLinearProblem(java.io.PrintWriter pw, VectorNd bf, VectorNd bg, java.lang.String fmtStr, boolean omitLowerRightDiagonal)
void	printStructure(java.io.PrintStream ps)
void	printValues(java.io.PrintStream ps, java.lang.String fmtStr)
void	printValues(java.io.PrintStream ps, java.lang.String fmtStr, double[] vals, int[] rowOffs, int[] colIdxs, int size, int numVals)
double	residual(SparseBlockMatrix M, int sizeM, SparseBlockMatrix GT, VectorNd Rg, SparseBlockMatrix NT, VectorNd Rn, VectorNd vel, VectorNd lam, VectorNd the, VectorNd bm, VectorNd bg, VectorNd bn)
KKTSolver.Status	solve(VectorNd vel, VectorNd lam, VectorNd bm, VectorNd bg) Solves the equality parts of a factored system.
KKTSolver.Status	solve(VectorNd vel, VectorNd lam, VectorNd the, VectorNd bm, VectorNd bg, VectorNd bn) Solves the equality and inequality parts of a factored system.
KKTSolver.Status	solve(VectorNd vel, VectorNd lam, VectorNd the, VectorNd phi, VectorNd bm, VectorNd bg, VectorNd bn, VectorNd bd, VectorNd flim) Solves the equality, inequality, and frictional parts of a factored system.
KKTSolver.Status	solve(VectorNd vel, VectorNd lam, VectorNd the, VectorNd phi, VectorNd bm, VectorNd bg, VectorNd bn, VectorNd bd, VectorNd flim, VectorNi state, VectorNi contactIdxs) Solves the equality, inequality, and frictional parts of a factored system.
KKTSolver.Status	solve(VectorNd vel, VectorNd lam, VectorNd the, VectorNd bm, VectorNd bg, VectorNd bn, VectorNi state) Solves the equality and inequality parts of a factored system.
void	solveMG(double[] Xbuf, double[] Bbuf, int nrhs)
void	solveMG(VectorNd x, VectorNd b) Solve the system
void	solveMG(VectorNd xm, VectorNd xg, VectorNd bm, VectorNd bg) Solve the system

Methods inherited from class java.lang.Object

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

Field Detail

computeResidualMG

public static boolean computeResidualMG

useBlockSolves

public static boolean useBlockSolves

myQPTestCaseFile

public static java.lang.String myQPTestCaseFile

myDebug

public static boolean myDebug

Constructor Detail

KKTSolver

public KKTSolver(SparseSolverId solverType)

KKTSolver

public KKTSolver()

Method Detail

analyzeMG

public void analyzeMG(java.lang.Object M,
                      int sizeM,
                      SparseBlockMatrix GT,
                      VectorNd Rg,
                      int typeM)

Performs symbolic analysis on the equality portion of the KKT system defined by M and G. This step must be called before any calls are made to factor().

analyze

public void analyze(VectorNd Mdiag,
                    int sizeM,
                    SparseBlockMatrix GT,
                    VectorNd Rg)

Performs symbolic analysis on the equality portion of the KKT system defined by a diagonal M matrix and G. This step must be called before any calls are made to factor().

Parameters:

Mdiag - vector defining the diagonal entries of M

sizeM - size of M

GT - Sparse matrix defining the transpose of G

Rg - if non-null, supplies the diagonal regularization matrix R

analyze

public void analyze(SparseBlockMatrix M,
                    int sizeM,
                    SparseBlockMatrix GT,
                    VectorNd Rg,
                    int typeM)

Performs symbolic analysis on the equality portion of the KKT system defined by matrices M and GT. This step must be called before any calls are made to factor().

Parameters:

M - Sparse matrix defining M

sizeM - size of M

GT - Sparse matrix defining the transpose of G

Rg - if non-null, supplies the diagonal regularization matrix R

typeM - describes the type of M, to be used in determining how the resulting KKT system should be factored. Should be either Matrix.INDEFINITE, Matrix.SYMMETRIC, or Matrix.SPD.

factor

public void factor(SparseBlockMatrix M,
                   int sizeM,
                   SparseBlockMatrix GT,
                   VectorNd Rg)

Does a numeric factorization of a KKT system containg only equality constraints. analyze() must have been previously called with M and G matrices having the same symbolic structure as the ones supplied to this method.

factor

public void factor(SparseBlockMatrix M,
                   int sizeM,
                   SparseBlockMatrix GT,
                   VectorNd Rg,
                   SparseBlockMatrix NT,
                   VectorNd Rn)

Does a numeric factorization of a KKT system containing equality and inequality constraints. analyze() must have been previously called with M and G matrices having the same symbolic structure as the ones supplied to this method.

factor

public void factor(SparseBlockMatrix M,
                   int sizeM,
                   SparseBlockMatrix GT,
                   VectorNd Rg,
                   SparseBlockMatrix NT,
                   VectorNd Rn,
                   SparseBlockMatrix DT,
                   VectorNd Rd)

Does a numeric factorization of a KKT system containg equality, inequality, and friction constraints. analyze() must have been previously called with M and G matrices having the same symbolic structure as the ones supplied to this method.

factor

public void factor(VectorNd Mdiag,
                   int sizeM,
                   SparseBlockMatrix GT,
                   VectorNd Rg)

Does a numeric factorization of a KKT system containg only equality constraints. analyze() must have been previously called with M and GT matrices having the same symbolic structure as the ones supplied to this method.

factor

public void factor(VectorNd Mdiag,
                   int sizeM,
                   SparseBlockMatrix GT,
                   VectorNd Rg,
                   SparseBlockMatrix NT,
                   VectorNd Rn)

Does a numeric factorization of a KKT system containg equality and inequality constraints. analyze() must have been previously called with M and G matrices having the same symbolic structure as the ones supplied to this method.

isFactored

public boolean isFactored()

solve

public KKTSolver.Status solve(VectorNd vel,
                              VectorNd lam,
                              VectorNd the,
                              VectorNd phi,
                              VectorNd bm,
                              VectorNd bg,
                              VectorNd bn,
                              VectorNd bd,
                              VectorNd flim)

Solves the equality, inequality, and frictional parts of a factored system.

solve

public KKTSolver.Status solve(VectorNd vel,
                              VectorNd lam,
                              VectorNd the,
                              VectorNd phi,
                              VectorNd bm,
                              VectorNd bg,
                              VectorNd bn,
                              VectorNd bd,
                              VectorNd flim,
                              VectorNi state,
                              VectorNi contactIdxs)

Solves the equality, inequality, and frictional parts of a factored system.

solve

public KKTSolver.Status solve(VectorNd vel,
                              VectorNd lam,
                              VectorNd the,
                              VectorNd bm,
                              VectorNd bg,
                              VectorNd bn)

Solves the equality and inequality parts of a factored system.

solve

public KKTSolver.Status solve(VectorNd vel,
                              VectorNd lam,
                              VectorNd the,
                              VectorNd bm,
                              VectorNd bg,
                              VectorNd bn,
                              VectorNi state)

Solves the equality and inequality parts of a factored system.

residual

public double residual(SparseBlockMatrix M,
                       int sizeM,
                       SparseBlockMatrix GT,
                       VectorNd Rg,
                       SparseBlockMatrix NT,
                       VectorNd Rn,
                       VectorNd vel,
                       VectorNd lam,
                       VectorNd the,
                       VectorNd bm,
                       VectorNd bg,
                       VectorNd bn)

solve

public KKTSolver.Status solve(VectorNd vel,
                              VectorNd lam,
                              VectorNd bm,
                              VectorNd bg)

Solves the equality parts of a factored system.

factorAndSolve

public KKTSolver.Status factorAndSolve(SparseBlockMatrix M,
                                       int sizeM,
                                       SparseBlockMatrix GT,
                                       VectorNd Rg,
                                       VectorNd vel,
                                       VectorNd lam,
                                       VectorNd bm,
                                       VectorNd bg,
                                       int tolExp)

Does a numeric factorization of a KKT system containg only equality constraints. analyze() must have been previously called with M and G matrices having the same symbolic structure as the ones supplied to this method.

getLcpState

public VectorNi getLcpState()

printStructure

public void printStructure(java.io.PrintStream ps)

printValues

public void printValues(java.io.PrintStream ps,
                        java.lang.String fmtStr)

getLinearMatrix

public MatrixNd getLinearMatrix()

printLinearProblem

public void printLinearProblem(java.io.PrintWriter pw,
                               VectorNd bf,
                               VectorNd bg,
                               java.lang.String fmtStr,
                               boolean omitLowerRightDiagonal)
                        throws java.io.IOException

Throws:

java.io.IOException

printValues

public void printValues(java.io.PrintStream ps,
                        java.lang.String fmtStr,
                        double[] vals,
                        int[] rowOffs,
                        int[] colIdxs,
                        int size,
                        int numVals)

solveMG

public void solveMG(VectorNd x,
                    VectorNd b)

Solve the system

 [ M -G ] [ x ]   [ b ]
 [      ] [   ] = [   ]
 [ G Rg ] [lam]   [ a ]
 

using the factorization of

 [ M  G ]
 [      ] 
 [ G -Rg ]
 

This requires negating the value of lam from the original solve

solveMG

public void solveMG(double[] Xbuf,
                    double[] Bbuf,
                    int nrhs)

solveMG

public void solveMG(VectorNd xm,
                    VectorNd xg,
                    VectorNd bm,
                    VectorNd bg)

Solve the system

 [ M -G ] [ x ]   [ b ]
 [      ] [   ] = [   ]
 [ G Rg ] [lam]   [ a ]
 

using the factorization of

 [ M  G ]
 [      ] 
 [ G -Rg ]
 

This requires negating the value of lam from the original solve

getNumNonZerosInFactors

public int getNumNonZerosInFactors()

lastSolveWasIterative

public boolean lastSolveWasIterative()

dispose

public void dispose()

finalize

public void finalize()

Overrides:

finalize in class java.lang.Object

getMatrixSolver

public DirectSolver getMatrixSolver()

initialize

public void initialize()