Ipopt::Ma86SolverInterface Class Reference

Base class for interfaces to symmetric indefinite linear solvers for sparse matrices. More...

#include <IpMa86SolverInterface.hpp>

Inheritance diagram for Ipopt::Ma86SolverInterface:

Public Member Functions
	Ma86SolverInterface ()
	~Ma86SolverInterface ()
bool	InitializeImpl (const OptionsList &options, const std::string &prefix)
	overloaded from AlgorithmStrategyObject More...
Methods for requesting solution of the linear system.
ESymSolverStatus	InitializeStructure (Index dim, Index nonzeros, const Index *ia, const Index *ja)
	Method for initializing internal stuctures. More...
double *	GetValuesArrayPtr ()
	Method returning an internal array into which the nonzero elements (in the same order as ja) will be stored by the calling routine before a call to MultiSolve with a new_matrix=true (or after a return of MultiSolve with SYMSOLV_CALL_AGAIN). More...
ESymSolverStatus	MultiSolve (bool new_matrix, const Index *ia, const Index *ja, Index nrhs, double *rhs_vals, bool check_NegEVals, Index numberOfNegEVals)
	Solve operation for multiple right hand sides. More...
Index	NumberOfNegEVals () const
	Number of negative eigenvalues detected during last factorization. More...
bool	IncreaseQuality ()
	Request to increase quality of solution for next solve. More...
bool	ProvidesInertia () const
	Query whether inertia is computed by linear solver. More...
EMatrixFormat	MatrixFormat () const
	Query of requested matrix type that the linear solver understands. More...
Methods related to the detection of linearly dependent
rows in a matrix
bool	ProvidesDegeneracyDetection () const
	Query whether the indices of linearly dependent rows/columns can be determined by this linear solver. More...
ESymSolverStatus	DetermineDependentRows (const Index *ia, const Index *ja, std::list< Index > &c_deps)
	This method determines the list of row indices of the linearly dependent rows. More...
Public Member Functions inherited from Ipopt::SparseSymLinearSolverInterface
	SparseSymLinearSolverInterface ()
virtual	~SparseSymLinearSolverInterface ()
Public Member Functions inherited from Ipopt::AlgorithmStrategyObject
bool	Initialize (const Journalist &jnlst, IpoptNLP &ip_nlp, IpoptData &ip_data, IpoptCalculatedQuantities &ip_cq, const OptionsList &options, const std::string &prefix)
	This method is called every time the algorithm starts again - it is used to reset any internal state. More...
bool	ReducedInitialize (const Journalist &jnlst, const OptionsList &options, const std::string &prefix)
	Reduced version of the Initialize method, which does not require special Ipopt information. More...
	AlgorithmStrategyObject ()
	Default Constructor. More...
virtual	~AlgorithmStrategyObject ()
	Default Destructor. More...
Public Member Functions inherited from Ipopt::ReferencedObject
	ReferencedObject ()
virtual	~ReferencedObject ()
Index	ReferenceCount () const
void	AddRef (const Referencer *referencer) const
void	ReleaseRef (const Referencer *referencer) const

Static Public Member Functions
static void	RegisterOptions (SmartPtr< RegisteredOptions > roptions)

Private Types
enum	order_opts { ORDER_AUTO, ORDER_AMD, ORDER_METIS }

Private Attributes
int	ndim_
double *	val_
int	numneg_
Index *	order_
void *	keep_
bool	pivtol_changed_
struct ma86_control	control_
double	umax_
int	ordering_

Additional Inherited Members
Public Types inherited from Ipopt::SparseSymLinearSolverInterface
enum	EMatrixFormat {   Triplet_Format, CSR_Format_0_Offset, CSR_Format_1_Offset, CSR_Full_Format_0_Offset,   CSR_Full_Format_1_Offset }
	Enum to specify sparse matrix format. More...
Protected Member Functions inherited from Ipopt::AlgorithmStrategyObject
const Journalist &	Jnlst () const
IpoptNLP &	IpNLP () const
IpoptData &	IpData () const
IpoptCalculatedQuantities &	IpCq () const
bool	HaveIpData () const

Detailed Description

Base class for interfaces to symmetric indefinite linear solvers for sparse matrices.

This defines the general interface to linear solvers for sparse symmetric indefinite matrices. The matrices can be provided either in "triplet format" (like for Harwell's MA27 solver), or in compressed sparse row (CSR) format for the lower triangular part of the symmetric matrix.

The solver should be able to compute the interia of the matrix, or more specifically, the number of negative eigenvalues in the factorized matrix.

This interface is used by the calling objective in the following way:

1. The InitializeImpl method is called at the very beginning (for every optimization run), which allows the linear solver object to retrieve options given in the OptionsList (such as pivot tolerances etc). At this point, some internal data can also be initialized.
2. The calling class calls MatrixFormat to find out which matrix representation the linear solver requires. The possible options are Triplet_Format, as well as CSR_Format_0_Offset and CSR_Format_1_Offset. The difference between the last two is that for CSR_Format_0_Offset the couning of the element position in the ia and ja arrays starts are 0 (C-style numbering), whereas for the other one it starts at 1 (Fortran-style numbering).
3. After this, the InitializeStructure method is called (once). Here, the structure of the matrix is provided. If the linear solver requires a symbolic preprocessing phase that can be done without knowledge of the matrix element values, it can be done here.
4. The calling class will request an array for storing the actual values for a matrix using the GetValuesArrayPtr method. This array must be at least as large as the number of nonzeros in the matrix (as given to this class by the InitializeStructure method call). After a call of this method, the calling class will fill this array with the actual values of the matrix.
5. Every time lateron, when actual solves of a linear system is requested, the calling class will call the MultiSolve to request the solve, possibly for mulitple right-hand sides. The flag new_matrix then indicates if the values of the matrix have changed and if a factorization is required, or if an old factorization can be used to do the solve.

Note that the GetValuesArrayPtr method will be called before every call of MultiSolve with new_matrix=true, or before a renewed call of MultiSolve if the most previous return value was SYMSOLV_CALL_AGAIN.

1. The calling class might request with NumberOfNegEVals the number of the negative eigenvalues for the original matrix that were detected during the most recently performed factorization.
2. The calling class might ask the linear solver to increase the quality of the solution. For example, if the linear solver uses a pivot tolerance, a larger value should be used for the next solve (which might require a refactorization).
3. Finally, when the destructor is called, the internal storage, also in the linear solver, should be released.

Note, if the matrix is given in triplet format, entries might be listed multiple times, in which case the corresponsing elements have to be added.

A note for warm starts: If the option "warm_start_same_structure" is specified with "yes", the algorithm assumes that a problem with the same sparsity structure is solved for a repeated time. In that case, the linear solver might reuse information from the previous optimization. See Ma27TSolverInterface for an example.

Definition at line 104 of file IpMa86SolverInterface.hpp.

Member Enumeration Documentation

enum Ipopt::Ma86SolverInterface::order_opts

private

Enumerator
ORDER_AUTO
ORDER_AMD
ORDER_METIS

Definition at line 107 of file IpMa86SolverInterface.hpp.

Constructor & Destructor Documentation

Ipopt::Ma86SolverInterface::Ma86SolverInterface ( )

inline

Definition at line 127 of file IpMa86SolverInterface.hpp.

Ipopt::Ma86SolverInterface::~Ma86SolverInterface

(

)

Member Function Documentation

static void Ipopt::Ma86SolverInterface::RegisterOptions ( SmartPtr< RegisteredOptions >  roptions )

static

bool Ipopt::Ma86SolverInterface::InitializeImpl ( const OptionsList &  options, const std::string &  prefix  )

virtual

overloaded from AlgorithmStrategyObject

Implements Ipopt::SparseSymLinearSolverInterface.

ESymSolverStatus Ipopt::Ma86SolverInterface::InitializeStructure ( Index  dim, Index  nonzeros, const Index *  ia, const Index *  ja  )

virtual

Method for initializing internal stuctures.

Here, ndim gives the number of rows and columns of the matrix, nonzeros give the number of nonzero elements, and ia and ja give the positions of the nonzero elements, given in the matrix format determined by MatrixFormat.

Implements Ipopt::SparseSymLinearSolverInterface.

double* Ipopt::Ma86SolverInterface::GetValuesArrayPtr ( )

inlinevirtual

Method returning an internal array into which the nonzero elements (in the same order as ja) will be stored by the calling routine before a call to MultiSolve with a new_matrix=true (or after a return of MultiSolve with SYMSOLV_CALL_AGAIN).

The returned array must have space for at least nonzero elements.

Implements Ipopt::SparseSymLinearSolverInterface.

Definition at line 155 of file IpMa86SolverInterface.hpp.

ESymSolverStatus Ipopt::Ma86SolverInterface::MultiSolve ( bool  new_matrix, const Index *  ia, const Index *  ja, Index  nrhs, double *  rhs_vals, bool  check_NegEVals, Index  numberOfNegEVals  )

virtual

Solve operation for multiple right hand sides.

Solves the linear system A * x = b with multiple right hand sides, where A is the symmtric indefinite matrix. Here, ia and ja give the positions of the values (in the required matrix data format). The actual values of the matrix will have been given to this object by copying them into the array provided by GetValuesArrayPtr. ia and ja are identical to the ones given to InitializeStructure. The flag new_matrix is set to true, if the values of the matrix has changed, and a refactorzation is required.

The return code is SYMSOLV_SUCCESS if the factorization and solves were successful, SYMSOLV_SINGULAR if the linear system is singular, and SYMSOLV_WRONG_INERTIA if check_NegEVals is true and the number of negative eigenvalues in the matrix does not match numberOfNegEVals. If SYMSOLV_CALL_AGAIN is returned, then the calling function will request the pointer for the array for storing a again (with GetValuesPtr), write the values of the nonzero elements into it, and call this MultiSolve method again with the same right-hand sides. (This can be done, for example, if the linear solver realized it does not have sufficient memory and needs to redo the factorization; e.g., for MA27.)

The number of right-hand sides is given by nrhs, the values of the right-hand sides are given in rhs_vals (one full right-hand side stored immediately after the other), and solutions are to be returned in the same array.

check_NegEVals will not be chosen true, if ProvidesInertia() returns false.

Implements Ipopt::SparseSymLinearSolverInterface.

Index Ipopt::Ma86SolverInterface::NumberOfNegEVals ( ) const

inlinevirtual

Number of negative eigenvalues detected during last factorization.

Returns the number of negative eigenvalues of the most recent factorized matrix. This must not be called if the linear solver does not compute this quantities (see ProvidesInertia).

Implements Ipopt::SparseSymLinearSolverInterface.

Definition at line 206 of file IpMa86SolverInterface.hpp.

bool Ipopt::Ma86SolverInterface::IncreaseQuality ( )

virtual

Request to increase quality of solution for next solve.

The calling class asks linear solver to increase quality of solution for the next solve (e.g. increase pivot tolerance). Returns false, if this is not possible (e.g. maximal pivot tolerance already used.)

Implements Ipopt::SparseSymLinearSolverInterface.

bool Ipopt::Ma86SolverInterface::ProvidesInertia ( ) const

inlinevirtual

Query whether inertia is computed by linear solver.

Returns true, if linear solver provides inertia.

Implements Ipopt::SparseSymLinearSolverInterface.

Definition at line 225 of file IpMa86SolverInterface.hpp.

EMatrixFormat Ipopt::Ma86SolverInterface::MatrixFormat ( ) const

inlinevirtual

Query of requested matrix type that the linear solver understands.

Implements Ipopt::SparseSymLinearSolverInterface.

Definition at line 233 of file IpMa86SolverInterface.hpp.

bool Ipopt::Ma86SolverInterface::ProvidesDegeneracyDetection ( ) const

inlinevirtual

Query whether the indices of linearly dependent rows/columns can be determined by this linear solver.

Reimplemented from Ipopt::SparseSymLinearSolverInterface.

Definition at line 244 of file IpMa86SolverInterface.hpp.

ESymSolverStatus Ipopt::Ma86SolverInterface::DetermineDependentRows ( const Index *  ia, const Index *  ja, std::list< Index > &  c_deps  )

inlinevirtual

This method determines the list of row indices of the linearly dependent rows.

Reimplemented from Ipopt::SparseSymLinearSolverInterface.

Definition at line 250 of file IpMa86SolverInterface.hpp.

Member Data Documentation

int Ipopt::Ma86SolverInterface::ndim_

private

Definition at line 113 of file IpMa86SolverInterface.hpp.

double* Ipopt::Ma86SolverInterface::val_

private

Definition at line 114 of file IpMa86SolverInterface.hpp.

int Ipopt::Ma86SolverInterface::numneg_

private

Definition at line 115 of file IpMa86SolverInterface.hpp.

Index* Ipopt::Ma86SolverInterface::order_

private

Definition at line 116 of file IpMa86SolverInterface.hpp.

void* Ipopt::Ma86SolverInterface::keep_

private

Definition at line 117 of file IpMa86SolverInterface.hpp.

bool Ipopt::Ma86SolverInterface::pivtol_changed_

private

Definition at line 118 of file IpMa86SolverInterface.hpp.

struct ma86_control Ipopt::Ma86SolverInterface::control_

private

Definition at line 121 of file IpMa86SolverInterface.hpp.

double Ipopt::Ma86SolverInterface::umax_

private

Definition at line 122 of file IpMa86SolverInterface.hpp.

int Ipopt::Ma86SolverInterface::ordering_

private

Definition at line 123 of file IpMa86SolverInterface.hpp.

---

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

• src/Algorithm/LinearSolvers/IpMa86SolverInterface.hpp