Reference documentation for deal.II version Git d4ec8156ca 2019-11-18 14:03:00 -0700
\(\newcommand{\dealcoloneq}{\mathrel{\vcenter{:}}=}\)
Public Member Functions | Public Attributes | List of all members
RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData Class Reference

#include <deal.II/lac/relaxation_block.h>

Inheritance diagram for RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData:
[legend]

Public Member Functions

 AdditionalData (const double relaxation=1., const bool invert_diagonal=true, const bool same_diagonal=false, const typename PreconditionBlockBase< InverseNumberType >::Inversion inversion=PreconditionBlockBase< InverseNumberType >::gauss_jordan, const double threshold=0., VectorType *temp_ghost_vector=nullptr)
 
std::size_t memory_consumption () const
 
- Public Member Functions inherited from Subscriptor
 Subscriptor ()
 
 Subscriptor (const Subscriptor &)
 
 Subscriptor (Subscriptor &&) noexcept
 
virtual ~Subscriptor ()
 
Subscriptoroperator= (const Subscriptor &)
 
Subscriptoroperator= (Subscriptor &&) noexcept
 
void subscribe (std::atomic< bool > *const validity, const std::string &identifier="") const
 
void unsubscribe (std::atomic< bool > *const validity, const std::string &identifier="") const
 
unsigned int n_subscriptions () const
 
template<typename StreamType >
void list_subscribers (StreamType &stream) const
 
void list_subscribers () const
 
template<class Archive >
void serialize (Archive &ar, const unsigned int version)
 

Public Attributes

SparsityPattern block_list
 
double relaxation
 
bool invert_diagonal
 
bool same_diagonal
 
PreconditionBlockBase< InverseNumberType >::Inversion inversion
 
double threshold = 0.
 
unsigned int kernel_size = 0
 
std::vector< std::vector< unsigned int > > order
 
VectorType * temp_ghost_vector
 

Additional Inherited Members

- Static Public Member Functions inherited from Subscriptor
static ::ExceptionBaseExcInUse (int arg1, std::string arg2, std::string arg3)
 
static ::ExceptionBaseExcNoSubscriber (std::string arg1, std::string arg2)
 

Detailed Description

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
class RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData

Parameters for block relaxation methods. In addition to typical control parameters like relaxation, this object also contains the block structure in block_list and an optional ordering of the blocks in order.

Definition at line 82 of file relaxation_block.h.

Constructor & Destructor Documentation

◆ AdditionalData()

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::AdditionalData ( const double  relaxation = 1.,
const bool  invert_diagonal = true,
const bool  same_diagonal = false,
const typename PreconditionBlockBase< InverseNumberType >::Inversion  inversion = PreconditionBlockBase< InverseNumberType >::gauss_jordan,
const double  threshold = 0.,
VectorType *  temp_ghost_vector = nullptr 
)

Constructor.

Member Function Documentation

◆ memory_consumption()

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
std::size_t RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::memory_consumption ( ) const

Return the memory allocated in this object.

Member Data Documentation

◆ block_list

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
SparsityPattern RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::block_list

The mapping from indices to blocks. Each row of this pattern enumerates the indices constituting a diagonal block to be inverted.

Definition at line 101 of file relaxation_block.h.

◆ relaxation

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
double RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::relaxation

Relaxation parameter.

Definition at line 106 of file relaxation_block.h.

◆ invert_diagonal

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
bool RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::invert_diagonal

Invert diagonal during initialization. Alternatively, diagonal blocks are inverted on the fly, whenever they are used. While inverting blocks in advance requires more memory, it usually saves a lot of computation. See same_diagonal on how you can avoid memory overhead.

Definition at line 114 of file relaxation_block.h.

◆ same_diagonal

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
bool RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::same_diagonal

Assume all diagonal blocks are equal to save memory. If this flag is true, then only the first diagonal block of the matrix is inverted and stored. It is then used for all other blocks.

Note
Avoid setting this true if your blocks are not equal, in particular if their sizes differ.

Definition at line 124 of file relaxation_block.h.

◆ inversion

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
PreconditionBlockBase<InverseNumberType>::Inversion RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::inversion

Choose the inversion method for the blocks.

Definition at line 129 of file relaxation_block.h.

◆ threshold

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
double RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::threshold = 0.

If inversion is SVD, we can compute the Penrose-Moore inverse of the blocks. In order to do so, we can specify here the threshold below which a singular value will be considered zero and thus not inverted. Setting this parameter to a value greater than zero takes precedence over threshold, i.e. kernel_size must be zero if you want to use threshold. This parameter is used in the call to LAPACKFullMatrix::compute_inverse_svd().

Definition at line 140 of file relaxation_block.h.

◆ kernel_size

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
unsigned int RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::kernel_size = 0

If inversion is SVD, we can compute the Penrose-Moore inverse of the blocks. In order to do so, we can specify here the size of the kernel that will not be inverted but considered zero. Setting this parameter to a value greater than zero takes precedence over threshold, i.e. kernel_size must be zero if you want to use threshold. This parameter is used in the call to LAPACKFullMatrix::compute_inverse_svd().

Definition at line 151 of file relaxation_block.h.

◆ order

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
std::vector<std::vector<unsigned int> > RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::order

The order in which blocks should be traversed. This vector can initiate several modes of execution:

  1. If the length of the vector is zero, then the relaxation method will be executed from first to last block.

  2. If the length is one, then the inner vector must have the same size as the number of blocks. The relaxation method is applied in the order given in this vector.

  3. If the outer vector has length greater one, then the relaxation method is applied several times, each time in the order given by the inner vector of the corresponding index. This mode can for instance be used for ADI methods and similar direction sweeps.

Definition at line 173 of file relaxation_block.h.

◆ temp_ghost_vector

template<typename MatrixType, typename InverseNumberType = typename MatrixType::value_type, typename VectorType = Vector<double>>
VectorType* RelaxationBlock< MatrixType, InverseNumberType, VectorType >::AdditionalData::temp_ghost_vector
mutable

Temporary ghost vector that is used in the relaxation method when performing parallel MPI computations. The user is required to have this point to an initialized vector that contains all indices that appear in the block_list sa ghost values. Typically, this the set of locally active level DoFs. Unused when VectorType is a serial vector type like Vector<double>.

Definition at line 183 of file relaxation_block.h.


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