File kernel/src/simulationTools/OSNSMatrixProjectOnConstraints.hpp#

Go to the source code of this file

Specific storage for matrices used in OneStepNSProblem with a projection scheme

class OSNSMatrixProjectOnConstraints : public OSNSMatrix
#include <OSNSMatrixProjectOnConstraints.hpp>

Interface to some specific storage types for matrices used in OneStepNSProblem.

This class is used to define an interface for various storage used for matrices in OneStepNSProblem

. Its aim is to fill the Numerics structure NumericsMatrix, required in many XXX_problem structures of Numerics as input argument for drivers.

The idea is to remove all matrix storage management problems from OSNS classes (

LCP

…) and to leave it into this class.

Two main functions:

  • fill(indexSet, interactionBlocks): fill the matrix using a list of “active” Interaction, in indexSet, and a MapOfMapOfInteractionMatrices, interactionBlocks, which determines which Interaction are connected or not (ie have common DynamicalSystem).

  • convert(): fill the NumericsMatrix structure (indeed only pointers links to the components of the present class)

Note that OSNSMatrix are square.

For example, if in a LCP, constraints of interest are indexSet={inter2,inter3,inter8,inter12}, whith common DynamicalSystem between 2 and 3, 2 and 8 and 8 and 12.

interactionBlocks contains matrices for all (interi,interj) which have common DS, for (interi,interj) in I0, the set of all Interaction.

(for details on how interactionBlocks is computed see OneStepNSProblem.h).

We denote interactionBlocks[interi][interj] = mij

Then, a call to fill(indexSet, interactionBlock) results in a matrix which looks like:

M=\left\lbrace\begin{array}{cccc} m22 & m23 & m28 & 0 \\ m32 & m33 & 0 & 0 \\ 0 & 0 & m88 & m812 \\ 0 & 0 & m128& m1212 \end{array}\right.

Note: at the time the available storage types are:

  • full matrix in a SiconosMatrix (_storageType = 0). In this case, for each call to fill(), the SiconosMatrix M is resized according to the sizes of the Interaction present in indexSet and then all the required interactionBlocks mij are COPIED into M.

  • Sparse Block Storage (_storageType = 1): corresponds to SparseBlockStructuredMatrix structure of Numerics. Only non-null interactionBlocks are saved in the matrix M and there is no copy of sub-interactionBlocks, only links thanks to pointers.

Public Functions

OSNSMatrixProjectOnConstraints(unsigned int n, unsigned int m, NM_types stor)

Constructor with dimRow and DimColumn of the matrix.

Parameters:
  • n – row size of the rectangle matrix

  • m – column size of the rectangle matrix

  • stor – storage type (NM_DENSE, NM_SPARSE_BLOCK)

unsigned int computeSizeForProjection(SP::Interaction inter)

compute the size of the vector to project for a given Interaction.

Parameters:

inter – the corresponding interaction

Returns:

unsigned int

virtual ~OSNSMatrixProjectOnConstraints()

destructor

virtual void fillM(InteractionsGraph &indexSet, bool update = true)

fill the current class using an index set and a map of interactionBlocks

Parameters:
  • indexSet – the index set of the active constraints

  • update – if true update the size and position

Protected Functions

ACCEPT_SERIALIZATION(OSNSMatrixProjectOnConstraints)#
inline OSNSMatrixProjectOnConstraints()#
virtual unsigned updateSizeAndPositions(InteractionsGraph &indexSet)#

For each Interaction in the graph, compute its absolute position.

Parameters:

indexSet – the index set ot the concerned interactios.

Returns:

the dimension of the problem (or size of the matrix), computed as the sum of the nslaw of all the Interaction in indexSet

unsigned updateSizeAndPositions(InteractionsGraph &indexSet)

For each Interaction in the graph, compute its absolute position.

Parameters:

indexSet – the index set ot the concerned interactios.

Returns:

the dimension of the problem (or size of the matrix), computed as the sum of the nslaw of all the Interaction in indexSet

unsigned updateSizeAndPositions(DynamicalSystemsGraph &DSG)#

For each DynamicalSystem in the graph, compute its absolute position.

Parameters:

DSG – the index set of the dynamical systems

Returns:

the dimension of the problem (or size of the matrix), computed as the sum of the nslaw of all the Interaction in indexSet