This class implements the pure virtual class Epetra_Operator for the AztecOO iterative solve which only uses the Apply() method to compute the matrix vector product. More...

#include <trilinos_impl.h>

Inheritance diagram for TrilinosMatVec:

[legend]

Public Member Functions
	TrilinosMatVec (const Teuchos::RCP< Trilinos > &trilinos)

void	apply (const Tpetra_MultiVector &X, Tpetra_MultiVector &Y, Teuchos::ETransp mode=Teuchos::NO_TRANS, Scalar_d alpha=Teuchos::ScalarTraits< Scalar_d >::one(), Scalar_d beta=Teuchos::ScalarTraits< Scalar_d >::zero()) const override

Teuchos::RCP< const Tpetra_Map >	getDomainMap () const override

Teuchos::RCP< const Tpetra_Map >	getRangeMap () const override

Detailed Description

This class implements the pure virtual class Epetra_Operator for the AztecOO iterative solve which only uses the Apply() method to compute the matrix vector product.

Constructor & Destructor Documentation

◆ TrilinosMatVec()

TrilinosMatVec::TrilinosMatVec ( const Teuchos::RCP< Trilinos > & trilinos )

inline

Define matrix vector operation at each iteration of the linear solver adds on the coupled neuman boundary contribution to the matrix

Parameters

x	vector to be applied on the operator
y	result of sparse matrix vector multiplication

Member Function Documentation

◆ apply()

void TrilinosMatVec::apply	(	const Tpetra_MultiVector &	x,
		Tpetra_MultiVector &	y,
		Teuchos::ETransp	mode = `Teuchos::NO_TRANS`,
		Scalar_d	alpha = `Teuchos::ScalarTraits< Scalar_d >::one()`,
		Scalar_d	beta = `Teuchos::ScalarTraits<Scalar_d>::zero()`
	)		const

override

Define the matrix vector multiplication operation to do at each iteration of an iterative linear solver. This function is called by the AztecOO (K + v1*v1' + v2*v2' + ...) x = K*x + v1(v1'x) + v2(v2'*x), where [v1, v2, ...] = bdryVec_list

For coupled Neumann boundary terms (v1*v1', v2*v2'), we use efficient an vectorized operation rather than explicitly forming the rank 1 outer product matrix and adding it to the global stiffness matrix K.