belfem::fem::dofmgr::SolverData Class Reference

#include <cl_FEM_DofMgr_SolverData.hpp>

Public Member Functions
	SolverData (DofManager *aParent, DofData *aDofData, BlockData *aBlockData, SideSetData *aSideSetData)
	~SolverData ()
void	extract_graph_from_mesh (Vector< id_t > &aGraphData)
void	allocate_matrices (const Vector< id_t > &aGraphData, Graph &aFreeDofs, Graph &aFixedDofs)
	Allocates sparse matrices for the finite element system.
void	compute_memory ()
void	create_assembly_tables ()
void	reset ()
void	reset_convection ()
void	reset_matrices (const bool aFullForce=false)
void	reset_rhs_vector ()
void	reset_rhs_matrix ()
void	assemble_jacobian (Element *aElement, const Matrix< real > &aJacobian)
void	assemble_full_matrices (Element *aElement, const Matrix< real > &aMass, const Matrix< real > &aStiffness)
void	assemble_newton (Element *aElement, const Matrix< real > &adJdx)
void	assemble_rhs (Element *aElement, const Vector< real > &aRHS)
void	assemble_volume_loads (Element *aElement, const Vector< real > &aRHS)
void	asseble_surface_loads (Element *aElement, const Vector< real > &aRHS)
void	assemble_rhs (Element *aElement, const Matrix< real > &aRHS)
void	collect_matrices (const bool aFullForce=false)
void	collect_rhs_vector ()
void	collect_vector (Vector< real > &aVector)
void	collect_rhs_matrix ()
void	update_field_values ()
void	set_solver (const SolverParameters &aParams)
	set the solver type for this field
Solver *	solver ()
	expose the solver
void	solve ()
	solve the system.
void	compute_residual ()
	Phase 1 of the two-phase iterative solve: prepare the load-adjusted RHS b, capture ||b||, and overwrite mRhsVector with the residual r = A x - b of the COMMITTED state under the current assembly — without solving.
void	solve_from_residual ()
	Phase 2: consume the residual left by compute_residual(), run the linear solve and the algorithm's update ( relaxed / Anderson Picard, or the Newton step with its post-update recompute ).
void	invalidate_residual ()
	invalidate the two-phase handshake; called by a fresh assembly
real	residual (const uint aIteration)
	return the residual
real	absolute_residual () const
	absolute residual norm of the last residual() call ( rank-uniform, broadcast alongside the relative value )
real	pre_update_residual () const
	pre-update residual of the entry state under the current assembly ( Newton iterates only, QUIET_NAN otherwise; rank-uniform after residual() ).
real	fixed_point_residual () const
	fixed-point residual ||G(x)-x|| / ||x|| of the last Anderson update ( rank-uniform after residual() )
void	set_anderson_depth (const uint aDepth)
	activate ( depth > 0 ) or deactivate ( depth = 0 ) Anderson mixing of the Picard branch; allocates the history registers
void	anderson_commit ()
	move the staged ( x, r ) pair into the history; called by the controller once the trial iterate is accepted.
void	anderson_discard ()
	drop the staged pair ( trial was rejected )
void	anderson_clear ()
	invalidate the history ( new attempt, algorithm switch, restore, ... ); also drops any staged pair
SpMatrix *	system_matrix ()
SpMatrix *	enforcement ()
SpMatrix *	dirichlet ()
SpMatrix *	imposition ()
SpMatrix *	matrix (const MatrixType aType)
Cell< Vector< index_t > > &	tables (const MatrixType aType)
Vector< real > &	rhs_vector ()
real	rhs_norm ()
Vector< real > &	volume_loads ()
Vector< real > &	surface_loads ()
index_t	number_of_free_dofs () const
index_t	number_of_fixed_dofs () const
index_t	my_number_of_free_dofs () const
index_t	my_number_of_fixed_dofs () const
void	save_system (const string &aPath)
void	load_system (const string &aPath)
void	remember_initialization_values (const bool aSaveRHS)
void	use_reset_values (const bool aFlag)
void	use_jedi_force (const bool aFlag)
void	use_full_force (const bool aFlag)
void	populate_graph (const Vector< id_t > &aData, const MatrixType aMatrixType, Graph &aFreeDofs, Graph &aFixedDofs, bool aLinkToSelf=true)
	Builds vertex connectivity for sparse matrix sparsity patterns.

Constructor & Destructor Documentation

SolverData()

belfem::fem::dofmgr::SolverData::SolverData	(	DofManager *	aParent,
		DofData *	aDofData,
		BlockData *	aBlockData,
		SideSetData *	aSideSetData )

~SolverData()

belfem::fem::dofmgr::SolverData::~SolverData

(

)

Member Function Documentation

absolute_residual()

real belfem::fem::dofmgr::SolverData::absolute_residual ( ) const

inline

absolute residual norm of the last residual() call ( rank-uniform, broadcast alongside the relative value )

allocate_matrices()

void belfem::fem::dofmgr::SolverData::allocate_matrices	(	const Vector< id_t > &	aGraphData,
		Graph &	aFreeDofs,
		Graph &	aFixedDofs )

Allocates sparse matrices for the finite element system.

Purpose:

• Creates sparse matrix structures with optimized sparsity patterns
• Supports both JEDI Force matrices (J, D, E, I) and Full matrices (M, K)
• Allocates RHS vectors/matrices for solution

Matrix Types: JEDI Force:

• Jacobian (J): n×n matrix (free-free interactions)
• Dirichlet (D): n×m matrix (free-fixed interactions, negative sign in assembly)
• Enforcement (E): m×n matrix (fixed-free, computes reaction forces) [if mUseJediForce]
• Imposition (I): m×m matrix (fixed-fixed self-coupling) [if mUseJediForce]

Full Force (if mUseFullForce):

• FullMass (M): (n+m)×(n+m) matrix (all DOFs, eigenvalue analysis)
• FullStiffness (K): (n+m)×(n+m) matrix (all DOFs, eigenvalue analysis)

Workflow:

1. If mUseFullForce: allocate FullMass / FullStiffness in the combined index space
2. Allocate the Dirichlet matrix (always)
3. If mUseJediForce: allocate Enforcement and Imposition
4. Allocate the System matrix from the System graph (must be last), then the Jacobian as its child (shared pointers and indices)
5. Allocate RHS vector or matrix

Index Space Handling for Full Matrices:

• Free DOFs normally use indices 0..n-1 (SEPARATE mode)
• Fixed DOFs normally use indices 0..m-1 (SEPARATE mode)
• Full matrices require combined indexing: free=0..n-1, fixed=n..n+m-1
• Temporarily shift fixed DOF indices, merge graphs, create matrices, restore

Graph Requirements:

• aFreeDofs: Contains free DOF vertices with reordered indices 0..n-1
• aFixedDofs: Contains fixed DOF vertices with reordered indices 0..m-1
• For matrix rows: Graph passed to SpMatrix must match row DOF type • Jacobian (J) → aFreeDofs (free rows × free cols) • Dirichlet (D) → aFreeDofs (free rows × fixed cols) • Enforcement (E) → aFixedDofs (fixed rows × free cols) • Imposition (I) → aFixedDofs (fixed rows × fixed cols)

Preconditions:

• DOFs must be reordered (via reorder_dofs())
• aGraphData contains complete DOF-to-DOF connectivity
• Solver must be initialized

Postconditions:

• All required sparse matrices allocated with correct sparsity patterns
• RHS vectors/matrices allocated

anderson_clear()

void belfem::fem::dofmgr::SolverData::anderson_clear

(

)

invalidate the history ( new attempt, algorithm switch, restore, ... ); also drops any staged pair

anderson_commit()

void belfem::fem::dofmgr::SolverData::anderson_commit

(

)

move the staged ( x, r ) pair into the history; called by the controller once the trial iterate is accepted.

No-op if mixing is off or nothing is staged.

anderson_discard()

void belfem::fem::dofmgr::SolverData::anderson_discard

(

)

drop the staged pair ( trial was rejected )

asseble_surface_loads()

void belfem::fem::dofmgr::SolverData::asseble_surface_loads	(	Element *	aElement,
		const Vector< real > &	aRHS )

assemble_full_matrices()

void belfem::fem::dofmgr::SolverData::assemble_full_matrices	(	Element *	aElement,
		const Matrix< real > &	aMass,
		const Matrix< real > &	aStiffness )

assemble_jacobian()

void belfem::fem::dofmgr::SolverData::assemble_jacobian	(	Element *	aElement,
		const Matrix< real > &	aJacobian )

assemble_newton()

void belfem::fem::dofmgr::SolverData::assemble_newton	(	Element *	aElement,
		const Matrix< real > &	adJdx )

assemble_rhs() [1/2]

void belfem::fem::dofmgr::SolverData::assemble_rhs	(	Element *	aElement,
		const Matrix< real > &	aRHS )

assemble_rhs() [2/2]

void belfem::fem::dofmgr::SolverData::assemble_rhs	(	Element *	aElement,
		const Vector< real > &	aRHS )

assemble_volume_loads()

void belfem::fem::dofmgr::SolverData::assemble_volume_loads	(	Element *	aElement,
		const Vector< real > &	aRHS )

collect_matrices()

void belfem::fem::dofmgr::SolverData::collect_matrices

(

const bool

aFullForce = false

)

collect_rhs_matrix()

void belfem::fem::dofmgr::SolverData::collect_rhs_matrix

(

)

collect_rhs_vector()

void belfem::fem::dofmgr::SolverData::collect_rhs_vector

(

)

collect_vector()

void belfem::fem::dofmgr::SolverData::collect_vector

(

Vector< real > &

aVector

)

compute_memory()

void belfem::fem::dofmgr::SolverData::compute_memory

(

)

compute_residual()

void belfem::fem::dofmgr::SolverData::compute_residual

(

)

Phase 1 of the two-phase iterative solve: prepare the load-adjusted RHS b, capture ||b||, and overwrite mRhsVector with the residual r = A x - b of the COMMITTED state under the current assembly — without solving.

residual() then reports the head certificate. One-shot per assembly ( load terms would double-add ); requires a fresh compute_jacobian_and_rhs.

create_assembly_tables()

void belfem::fem::dofmgr::SolverData::create_assembly_tables

(

)

dirichlet()

SpMatrix * belfem::fem::dofmgr::SolverData::dirichlet ( )

inline

enforcement()

SpMatrix * belfem::fem::dofmgr::SolverData::enforcement ( )

inline

extract_graph_from_mesh()

void belfem::fem::dofmgr::SolverData::extract_graph_from_mesh

(

Vector< id_t > &

aGraphData

)

fixed_point_residual()

real belfem::fem::dofmgr::SolverData::fixed_point_residual ( ) const

inline

fixed-point residual ||G(x)-x|| / ||x|| of the last Anderson update ( rank-uniform after residual() )

imposition()

SpMatrix * belfem::fem::dofmgr::SolverData::imposition ( )

inline

invalidate_residual()

void belfem::fem::dofmgr::SolverData::invalidate_residual ( )

inline

invalidate the two-phase handshake; called by a fresh assembly

load_system()

void belfem::fem::dofmgr::SolverData::load_system

(

const string &

aPath

)

matrix()

SpMatrix * belfem::fem::dofmgr::SolverData::matrix ( const MatrixType aType )

inline

my_number_of_fixed_dofs()

index_t belfem::fem::dofmgr::SolverData::my_number_of_fixed_dofs ( ) const

inline

my_number_of_free_dofs()

index_t belfem::fem::dofmgr::SolverData::my_number_of_free_dofs ( ) const

inline

number_of_fixed_dofs()

index_t belfem::fem::dofmgr::SolverData::number_of_fixed_dofs ( ) const

inline

number_of_free_dofs()

index_t belfem::fem::dofmgr::SolverData::number_of_free_dofs ( ) const

inline

populate_graph()

void belfem::fem::dofmgr::SolverData::populate_graph	(	const Vector< id_t > &	aData,
		const MatrixType	aMatrixType,
		Graph &	aFreeDofs,
		Graph &	aFixedDofs,
		bool	aLinkToSelf = true )

Builds vertex connectivity for sparse matrix sparsity patterns.

Purpose:

• Creates graph connectivity by linking DOF vertices based on element topology
• Filters connections by DOF type (free/fixed) to match specific matrix structure
• Supports both JEDI Force matrices (J, D, E, I) and Full matrices (M, K)
• Optionally excludes self-connections for graph reordering algorithms

Matrix Type Mapping (rows × columns):

• Jacobian (J): free × free (standard stiffness, used for solving)
• Dirichlet (D): free × fixed (BC enforcement, negative sign in assembly)
• Enforcement (E): fixed × free (computes reaction forces at constraints)
• Imposition (I): fixed × fixed (self-coupling of prescribed DOFs)
• FullMass (M): all × all (eigenvalue analysis, combined index space)
• FullStiffness (K):all × all (eigenvalue analysis, combined index space)

Algorithm:

1. Determine row/column DOF types from matrix type
2. For each DOF in connectivity data: a. Check if DOF should be a row vertex (matches tRowIsFixed) b. Loop over all connected DOFs from elements c. Filter connections by column type (matches tColIsFixed) d. Mark connections in bitset using my_index + offset e. [Optional] Remove self-connection if aLinkToSelf=false f. Convert bitset to index list g. Insert vertex pointers into DOF's connectivity list

Index Space Handling:

• Free DOFs: my_index = 0..n-1, offset = 0
• Fixed DOFs: my_index = 0..m-1, offset = tMyNumberOfFreeDofs
• Bitset size: (n+m) to accommodate both types with offset
• For full matrices: uses combined space, fixed DOFs already offset

Bitset Mechanism:

• Efficiently tracks unique connections (prevents duplicates)
• Position i in bitset represents DOF with my_index i (or i-offset for fixed)
• bitset.where() extracts set positions into index list

Graph Vertex Container:

• Each Dof inherits from graph::Vertex
• Vertex has container of connected vertices (adjacency list)
• SpMatrix constructor reads this connectivity to build sparsity pattern

Parameters:

Parameters

aData	DOF-to-DOF connectivity from extract_graph_from_mesh() Format: [num_dofs, dof_id, num_connected, dof_id, ...]
aMatrixType	Which matrix to populate (J, D, E, I, FullMass, FullStiffness)
aFreeDofs	Graph vertices for free DOFs (indices 0..n-1)
aFixedDofs	Graph vertices for fixed DOFs (indices 0..m-1)
aLinkToSelf	Include diagonal connections (default true) true: Include self-connections (for matrix allocation) false: Exclude self-connections (for METIS/SCOTCH reordering)

Preconditions:

• aData contains complete DOF connectivity from extract_graph_from_mesh()
• aFreeDofs and aFixedDofs contain reordered DOF vertices
• Each DOF's my_index is set to its position in respective graph (0..n-1 or 0..m-1)

Postconditions:

• Each row DOF has vertex container populated with column DOF vertices
• Connectivity matches requested matrix type structure
• Ready for SpMatrix construction

Usage Example: // For Jacobian (free × free): populate_graph(graphData, Jacobian, freeDofs, fixedDofs, true); // Result: Each free DOF points to all connected free DOFs

// For Dirichlet (free × fixed): populate_graph(graphData, Dirichlet, freeDofs, fixedDofs, false); // Result: Each free DOF points to all connected fixed DOFs, no self-links

pre_update_residual()

real belfem::fem::dofmgr::SolverData::pre_update_residual ( ) const

inline

pre-update residual of the entry state under the current assembly ( Newton iterates only, QUIET_NAN otherwise; rank-uniform after residual() ).

The honest backtracking reference — see the member comment

remember_initialization_values()

void belfem::fem::dofmgr::SolverData::remember_initialization_values

(

const bool

aSaveRHS

)

reset()

void belfem::fem::dofmgr::SolverData::reset

(

)

reset_convection()

void belfem::fem::dofmgr::SolverData::reset_convection

(

)

reset_matrices()

void belfem::fem::dofmgr::SolverData::reset_matrices

(

const bool

aFullForce = false

)

reset_rhs_matrix()

void belfem::fem::dofmgr::SolverData::reset_rhs_matrix

(

)

reset_rhs_vector()

void belfem::fem::dofmgr::SolverData::reset_rhs_vector

(

)

residual()

real belfem::fem::dofmgr::SolverData::residual

(

const uint

aIteration

)

return the residual

rhs_norm()

real belfem::fem::dofmgr::SolverData::rhs_norm ( )

inline

rhs_vector()

Vector< real > & belfem::fem::dofmgr::SolverData::rhs_vector ( )

inline

save_system()

void belfem::fem::dofmgr::SolverData::save_system

(

const string &

aPath

)

set_anderson_depth()

void belfem::fem::dofmgr::SolverData::set_anderson_depth

(

const uint

aDepth

)

activate ( depth > 0 ) or deactivate ( depth = 0 ) Anderson mixing of the Picard branch; allocates the history registers

set_solver()

void belfem::fem::dofmgr::SolverData::set_solver

(

const SolverParameters &

aParams

)

set the solver type for this field

solve()

void belfem::fem::dofmgr::SolverData::solve

(

)

solve the system.

For the iterative vector-RHS path this is the fused composition compute_residual() + solve_from_residual(); Direct mode and matrix-RHS callers keep the single-call contract unchanged.

solve_from_residual()

void belfem::fem::dofmgr::SolverData::solve_from_residual

(

)

Phase 2: consume the residual left by compute_residual(), run the linear solve and the algorithm's update ( relaxed / Anderson Picard, or the Newton step with its post-update recompute ).

Must be entered on EVERY rank — the collective barrier pairing lives here.

solver()

Solver * belfem::fem::dofmgr::SolverData::solver ( )

inline

expose the solver

surface_loads()

Vector< real > & belfem::fem::dofmgr::SolverData::surface_loads ( )

inline

system_matrix()

SpMatrix * belfem::fem::dofmgr::SolverData::system_matrix ( )

inline

tables()

Cell< Vector< index_t > > & belfem::fem::dofmgr::SolverData::tables ( const MatrixType aType )

inline

update_field_values()

void belfem::fem::dofmgr::SolverData::update_field_values

(

)

use_full_force()

void belfem::fem::dofmgr::SolverData::use_full_force

(

const bool

aFlag

)

use_jedi_force()

void belfem::fem::dofmgr::SolverData::use_jedi_force

(

const bool

aFlag

)

use_reset_values()

void belfem::fem::dofmgr::SolverData::use_reset_values ( const bool aFlag )

inline

volume_loads()

Vector< real > & belfem::fem::dofmgr::SolverData::volume_loads ( )

inline

---

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

• src/fem/kernel/cl_FEM_DofMgr_SolverData.hpp
• src/fem/kernel/cl_FEM_DofMgr_SolverData.cpp