Belos Version of the Day

A linear system to solve, and its associated information. More...
#include <BelosLinearProblem.hpp>
Public Member Functions  
Constructors/Destructor  
LinearProblem (void)  
Default constructor.  
LinearProblem (const Teuchos::RCP< const OP > &A, const Teuchos::RCP< MV > &X, const Teuchos::RCP< const MV > &B)  
Unpreconditioned linear system constructor.  
LinearProblem (const LinearProblem< ScalarType, MV, OP > &Problem)  
Copy constructor.  
virtual  ~LinearProblem (void) 
Destructor (declared virtual for memory safety of derived classes).  
Set methods  
void  setOperator (const Teuchos::RCP< const OP > &A) 
Set the operator A of the linear problem .  
void  setLHS (const Teuchos::RCP< MV > &X) 
Set lefthandside X of linear problem .  
void  setRHS (const Teuchos::RCP< const MV > &B) 
Set righthandside B of linear problem .  
void  setLeftPrec (const Teuchos::RCP< const OP > &LP) 
Set left preconditioner (LP ) of linear problem .  
void  setRightPrec (const Teuchos::RCP< const OP > &RP) 
Set right preconditioner (RP ) of linear problem .  
void  setCurrLS () 
Tell the linear problem that the solver is finished with the current linear system.  
void  setLSIndex (const std::vector< int > &index) 
Tell the linear problem which linear system(s) need to be solved next.  
void  setHermitian () 
Tell the linear problem that the (preconditioned) operator is Hermitian.  
void  setLabel (const std::string &label) 
Set the label prefix used by the timers in this object.  
Teuchos::RCP< MV >  updateSolution (const Teuchos::RCP< MV > &update=Teuchos::null, bool updateLP=false, ScalarType scale=Teuchos::ScalarTraits< ScalarType >::one()) 
Compute the new solution to the linear system using the given update vector.  
Teuchos::RCP< MV >  updateSolution (const Teuchos::RCP< MV > &update=Teuchos::null, ScalarType scale=Teuchos::ScalarTraits< ScalarType >::one()) const 
Compute the new solution to the linear system using the given update vector.  
Set / Reset method  
bool  setProblem (const Teuchos::RCP< MV > &newX=Teuchos::null, const Teuchos::RCP< const MV > &newB=Teuchos::null) 
Set up the linear problem manager.  
Accessor methods  
Teuchos::RCP< const OP >  getOperator () const 
A pointer to the (unpreconditioned) operator A.  
Teuchos::RCP< MV >  getLHS () const 
A pointer to the lefthand side X.  
Teuchos::RCP< const MV >  getRHS () const 
A pointer to the righthand side B.  
Teuchos::RCP< const MV >  getInitResVec () const 
A pointer to the initial unpreconditioned residual vector.  
Teuchos::RCP< const MV >  getInitPrecResVec () const 
A pointer to the preconditioned initial residual vector.  
Teuchos::RCP< MV >  getCurrLHSVec () 
Get a pointer to the current lefthand side (solution) of the linear system.  
Teuchos::RCP< const MV >  getCurrRHSVec () 
Get a pointer to the current righthand side of the linear system.  
Teuchos::RCP< const OP >  getLeftPrec () const 
Get a pointer to the left preconditioner.  
Teuchos::RCP< const OP >  getRightPrec () const 
Get a pointer to the right preconditioner.  
const std::vector< int >  getLSIndex () const 
(Zerobased) indices of the linear system(s) currently being solved.  
int  getLSNumber () const 
The number of linear systems that have been set.  
Teuchos::Array< Teuchos::RCP < Teuchos::Time > >  getTimers () const 
The timers for this object.  
State methods  
bool  isSolutionUpdated () const 
Has the current approximate solution been updated?  
bool  isProblemSet () const 
Whether the problem has been set.  
bool  isHermitian () const 
Whether the (preconditioned) operator is Hermitian.  
bool  isLeftPrec () const 
Whether the linear system is being preconditioned on the left.  
bool  isRightPrec () const 
Whether the linear system is being preconditioned on the right.  
Apply / Compute methods  
void  apply (const MV &x, MV &y) const 
Apply the composite operator of this linear problem to x , returning y .  
void  applyOp (const MV &x, MV &y) const 
Apply ONLY the operator to x , returning y .  
void  applyLeftPrec (const MV &x, MV &y) const 
Apply ONLY the left preconditioner to x , returning y .  
void  applyRightPrec (const MV &x, MV &y) const 
Apply ONLY the right preconditioner to x , returning y .  
void  computeCurrResVec (MV *R, const MV *X=0, const MV *B=0) const 
Compute a residual R for this operator given a solution X , and righthand side B .  
void  computeCurrPrecResVec (MV *R, const MV *X=0, const MV *B=0) const 
Compute a residual R for this operator given a solution X , and righthand side B . 
A linear system to solve, and its associated information.
This class encapsulates the general information needed for solving a linear system of equations using an iterative method.
ScalarType  The type of the entries in the matrix and vectors. 
MV  The (multi)vector type. 
OP  The operator type. (Operators are functions that take a multivector as input and compute a multivector as output.) 
BlockCG/BlockCGEpetraExFile.cpp, BlockCG/BlockPrecCGEpetraExFile.cpp, BlockCG/PseudoBlockCGEpetraExFile.cpp, BlockCG/PseudoBlockPrecCGEpetraExFile.cpp, BlockGmres/BlockFlexGmresEpetraExFile.cpp, BlockGmres/BlockGmresEpetraExFile.cpp, BlockGmres/BlockGmresPolyEpetraExFile.cpp, BlockGmres/BlockPrecGmresEpetraExFile.cpp, BlockGmres/PseudoBlockGmresEpetraExFile.cpp, BlockGmres/PseudoBlockPrecGmresEpetraExFile.cpp, GCRODR/GCRODREpetraExFile.cpp, GCRODR/PrecGCRODREpetraExFile.cpp, PCPG/PCPGEpetraExFile.cpp, and TFQMR/TFQMREpetraExFile.cpp.
Definition at line 82 of file BelosLinearProblem.hpp.
Belos::LinearProblem< ScalarType, MV, OP >::LinearProblem  (  void  ) 
Default constructor.
Creates an empty Belos::LinearProblem instance. The operator A, lefthandside X and righthandside B must be set using the setOperator()
, setLHS()
, and setRHS()
methods respectively.
Definition at line 561 of file BelosLinearProblem.hpp.
Belos::LinearProblem< ScalarType, MV, OP >::LinearProblem  (  const Teuchos::RCP< const OP > &  A, 
const Teuchos::RCP< MV > &  X,  
const Teuchos::RCP< const MV > &  B  
) 
Unpreconditioned linear system constructor.
Creates an unpreconditioned LinearProblem instance with the operator (A
), initial guess (X
), and right hand side (B
). Preconditioners can be set using the setLeftPrec()
and setRightPrec()
methods, and scaling can also be set using the setLeftScale()
and setRightScale()
methods.
Definition at line 576 of file BelosLinearProblem.hpp.
Belos::LinearProblem< ScalarType, MV, OP >::LinearProblem  (  const LinearProblem< ScalarType, MV, OP > &  Problem  ) 
Copy constructor.
Makes a copy of an existing LinearProblem instance.
Definition at line 597 of file BelosLinearProblem.hpp.
Belos::LinearProblem< ScalarType, MV, OP >::~LinearProblem  (  void  )  [virtual] 
Destructor (declared virtual for memory safety of derived classes).
Definition at line 623 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setOperator  (  const Teuchos::RCP< const OP > &  A  )  [inline] 
Set the operator A of the linear problem .
The operator is set by pointer; no copy of the operator is made.
Definition at line 123 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setLHS  (  const Teuchos::RCP< MV > &  X  )  [inline] 
Set lefthandside X of linear problem .
Setting the "lefthand side" sets the starting vector (also called "initial guess") of an iterative method. The multivector is set by pointer; no copy of the object is made. Belos' solvers will modify this multivector in place.
Definition at line 134 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setRHS  (  const Teuchos::RCP< const MV > &  B  )  [inline] 
Set righthandside B of linear problem .
The multivector is set by pointer; no copy of the object is made.
Definition at line 143 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setLeftPrec  (  const Teuchos::RCP< const OP > &  LP  )  [inline] 
Set left preconditioner (LP
) of linear problem .
The operator is set by pointer; no copy of the operator is made.
Definition at line 151 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setRightPrec  (  const Teuchos::RCP< const OP > &  RP  )  [inline] 
Set right preconditioner (RP
) of linear problem .
The operator is set by pointer; no copy of the operator is made.
Definition at line 156 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setCurrLS  (  ) 
Tell the linear problem that the solver is finished with the current linear system.
Definition at line 694 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setLSIndex  (  const std::vector< int > &  index  ) 
Tell the linear problem which linear system(s) need to be solved next.
Any calls to get the current RHS/LHS vectors after this method is called will return the new linear system(s) indicated by index
. The length of index
is assumed to be the blocksize. Entries of index
must be between 0 and the number of vectors in the RHS/LHS multivector. An entry of index
may also be 1, which means this column of the linear system is augmented using a random vector.
Definition at line 627 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setHermitian  (  )  [inline] 
Tell the linear problem that the (preconditioned) operator is Hermitian.
This knowledge may allow the operator to take advantage of the linear problem's symmetry. However, this method should not be called if the preconditioned operator is not Hermitian (or symmetric in real arithmetic).
We make no attempt to detect the symmetry of the operators, so we cannot check whether this method has been called incorrectly.
Definition at line 189 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::setLabel  (  const std::string &  label  )  [inline] 
Set the label prefix used by the timers in this object.
The default label prefix is "Belos". The timers are created during the first call to setProblem()
. Any calls to this method to change the label after that will not change the label used in the timer.
Definition at line 197 of file BelosLinearProblem.hpp.
Teuchos::RCP< MV > Belos::LinearProblem< ScalarType, MV, OP >::updateSolution  (  const Teuchos::RCP< MV > &  update = Teuchos::null , 
bool  updateLP = false , 

ScalarType  scale = Teuchos::ScalarTraits<ScalarType>::one() 

) 
Compute the new solution to the linear system using the given update vector.
Let be the update vector, the scale factor, and the current solution. If there is a right preconditioner , then we compute the new solution as . Otherwise, if there is no right preconditioner, we compute the new solution as .
This method always returns the new solution. If updateLP is false, it computes the new solution as a deep copy, without modifying the internally stored current solution. If updateLP is true, it computes the new solution in place, and returns a pointer to the internally stored solution.
update  [in/out] The solution update vector. If null, this method returns a pointer to the new solution. 
updateLP  [in] This is ignored if the update vector is null. Otherwise, if updateLP is true, the following things happen: (a) this LinearProblem's stored solution is updated in place, and (b) the next time GetCurrResVecs() is called, a new residual will be computed. If updateLP is false, then the new solution is computed and returned as a copy, without modifying this LinearProblem's stored current solution. 
scale  [in] The factor by which to multiply the solution update vector when computing the update. This is ignored if the update vector is null. 
Definition at line 730 of file BelosLinearProblem.hpp.
Teuchos::RCP<MV> Belos::LinearProblem< ScalarType, MV, OP >::updateSolution  (  const Teuchos::RCP< MV > &  update = Teuchos::null , 
ScalarType  scale = Teuchos::ScalarTraits<ScalarType>::one() 

)  const [inline] 
Compute the new solution to the linear system using the given update vector.
This method does the same thing as calling the threeargument version of updateSolution() with updateLP = false. It does not update the linear problem or change the linear problem's state in any way.
update  [in/out] The solution update vector. If null, this method returns a pointer to the new solution. 
scale  [in] The factor by which to multiply the solution update vector when computing the update. This is ignored if the update vector is null. 
Definition at line 257 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::setProblem  (  const Teuchos::RCP< MV > &  newX = Teuchos::null , 
const Teuchos::RCP< const MV > &  newB = Teuchos::null 

) 
Set up the linear problem manager.
Call this method if you want to solve the linear system with a different left or righthand side, or if you want to prepare the linear problem to solve the linear system that was already passed in. (In the latter case, call this method with the default arguments.) The internal flags will be set as if the linear system manager was just initialized, and the initial residual will be computed.
Many of Belos' solvers require that this method has been called on the linear problem, before they can solve it.
newX  [in/out] If you want to solve the linear system with a different lefthand side, pass it in here. Otherwise, set this to null (the default value). 
newB  [in] If you want to solve the linear system with a different righthand side, pass it in here. Otherwise, set this to null (the default value). 
Definition at line 803 of file BelosLinearProblem.hpp.
Teuchos::RCP<const OP> Belos::LinearProblem< ScalarType, MV, OP >::getOperator  (  )  const [inline] 
A pointer to the (unpreconditioned) operator A.
Definition at line 301 of file BelosLinearProblem.hpp.
Teuchos::RCP<MV> Belos::LinearProblem< ScalarType, MV, OP >::getLHS  (  )  const [inline] 
A pointer to the lefthand side X.
Definition at line 304 of file BelosLinearProblem.hpp.
Teuchos::RCP<const MV> Belos::LinearProblem< ScalarType, MV, OP >::getRHS  (  )  const [inline] 
A pointer to the righthand side B.
Definition at line 307 of file BelosLinearProblem.hpp.
Teuchos::RCP<const MV> Belos::LinearProblem< ScalarType, MV, OP >::getInitResVec  (  )  const [inline] 
A pointer to the initial unpreconditioned residual vector.
Definition at line 310 of file BelosLinearProblem.hpp.
Teuchos::RCP<const MV> Belos::LinearProblem< ScalarType, MV, OP >::getInitPrecResVec  (  )  const [inline] 
A pointer to the preconditioned initial residual vector.
Definition at line 316 of file BelosLinearProblem.hpp.
Teuchos::RCP< MV > Belos::LinearProblem< ScalarType, MV, OP >::getCurrLHSVec  (  ) 
Get a pointer to the current lefthand side (solution) of the linear system.
This method is called by the solver or any method that is interested in the current linear system being solved.
isSolutionUpdated()
).This method is not the same thing as getLHS()
. The getLHS()
method just returns a pointer to the original lefthand side vector. This method only returns a valid vector if the current subset of righthand side(s) to solve has been set (via the setLSIndex()
method).
Definition at line 872 of file BelosLinearProblem.hpp.
Teuchos::RCP< const MV > Belos::LinearProblem< ScalarType, MV, OP >::getCurrRHSVec  (  ) 
Get a pointer to the current righthand side of the linear system.
This method is called by the solver or any method that is interested in the current linear system being solved.
isSolutionUpdated()
).This method is not the same thing as getRHS()
. The getRHS()
method just returns a pointer to the original righthand side vector. This method only returns a valid vector if the current subset of righthand side(s) to solve has been set (via the setLSIndex()
method).
Definition at line 883 of file BelosLinearProblem.hpp.
Teuchos::RCP<const OP> Belos::LinearProblem< ScalarType, MV, OP >::getLeftPrec  (  )  const [inline] 
Get a pointer to the left preconditioner.
Definition at line 351 of file BelosLinearProblem.hpp.
Teuchos::RCP<const OP> Belos::LinearProblem< ScalarType, MV, OP >::getRightPrec  (  )  const [inline] 
Get a pointer to the right preconditioner.
Definition at line 354 of file BelosLinearProblem.hpp.
const std::vector<int> Belos::LinearProblem< ScalarType, MV, OP >::getLSIndex  (  )  const [inline] 
(Zerobased) indices of the linear system(s) currently being solved.
Since the block size is independent of the number of righthand sides for some solvers (GMRES, CG, etc.), it is important to know which linear systems are being solved. That may mean you need to update the information about the norms of your initial residual vector for weighting purposes. This information can help you avoid querying the solver for information that rarely changes.
setLSIndex()
has been called with a nonempty index vector argument, or if this linear problem was constructed via the copy constructor of a linear problem with a nonempty index vector. Definition at line 377 of file BelosLinearProblem.hpp.
int Belos::LinearProblem< ScalarType, MV, OP >::getLSNumber  (  )  const [inline] 
The number of linear systems that have been set.
This can be used by status test classes to determine if the solver manager has advanced and is solving another linear system. This is incremented by one every time that setLSIndex()
completes successfully.
Definition at line 385 of file BelosLinearProblem.hpp.
Teuchos::Array<Teuchos::RCP<Teuchos::Time> > Belos::LinearProblem< ScalarType, MV, OP >::getTimers  (  )  const [inline] 
The timers for this object.
The timers are ordered as follows:
Definition at line 393 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::isSolutionUpdated  (  )  const [inline] 
Has the current approximate solution been updated?
This only means that the current linear system for which the solver is solving (as obtained by getCurr{LHS, RHS}Vec()) has been updated by the solver. This will be true every iteration for solvers like CG, but not true for solvers like GMRES until the solver restarts.
Definition at line 410 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::isProblemSet  (  )  const [inline] 
Whether the problem has been set.
Definition at line 413 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::isHermitian  (  )  const [inline] 
Whether the (preconditioned) operator is Hermitian.
If preconditioner(s) are defined and this method returns true, then the entire preconditioned operator is Hermitian (or symmetric in real arithmetic).
Definition at line 420 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::isLeftPrec  (  )  const [inline] 
Whether the linear system is being preconditioned on the left.
Definition at line 423 of file BelosLinearProblem.hpp.
bool Belos::LinearProblem< ScalarType, MV, OP >::isRightPrec  (  )  const [inline] 
Whether the linear system is being preconditioned on the right.
Definition at line 426 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::apply  (  const MV &  x, 
MV &  y  
)  const 
Apply the composite operator of this linear problem to x
, returning y
.
This application is the composition of the left/right preconditioner and operator. Most Krylov methods will use this application method within their code.
Precondition:
getOperator().get()!=NULL
Definition at line 894 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::applyOp  (  const MV &  x, 
MV &  y  
)  const 
Apply ONLY the operator to x
, returning y
.
This method only applies the linear problem's operator, without any preconditioners that may have been defined. Flexible variants of Krylov methods will use this method. If no operator has been defined, this method just copies x into y.
Definition at line 980 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::applyLeftPrec  (  const MV &  x, 
MV &  y  
)  const 
Apply ONLY the left preconditioner to x
, returning y
.
This method only applies the left preconditioner. This may be required for flexible variants of Krylov methods. If no left preconditioner has been defined, this method just copies x into y.
Definition at line 994 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::applyRightPrec  (  const MV &  x, 
MV &  y  
)  const 
Apply ONLY the right preconditioner to x
, returning y
.
This method only applies the right preconditioner. This may be required for flexible variants of Krylov methods. If no right preconditioner has been defined, this method just copies x into y.
Definition at line 1008 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::computeCurrResVec  (  MV *  R, 
const MV *  X = 0 , 

const MV *  B = 0 

)  const 
Compute a residual R
for this operator given a solution X
, and righthand side B
.
This method will compute the residual for the current linear system if X
and B
are null pointers. The result will be returned into R. Otherwise R = OP(A)X  B
will be computed and returned.
Definition at line 1101 of file BelosLinearProblem.hpp.
void Belos::LinearProblem< ScalarType, MV, OP >::computeCurrPrecResVec  (  MV *  R, 
const MV *  X = 0 , 

const MV *  B = 0 

)  const 
Compute a residual R
for this operator given a solution X
, and righthand side B
.
This method will compute the residual for the current linear system if X
and B
are null pointers. The result will be returned into R. Otherwise R = OP(A)X  B
will be computed and returned.
Definition at line 1022 of file BelosLinearProblem.hpp.