Detailed Description

Multifrontal solver for nonlinear factor graphs.

This class extends MultifrontalSolver to solve nonlinear problems. The linearization is provided externally via load(), or via eliminateInPlace() which combines loading and elimination.

Inheritance diagram for gtsam::NonlinearMultifrontalSolver:

Public Member Functions
	NonlinearMultifrontalSolver (const NonlinearFactorGraph &graph, const Values &values, const Ordering &ordering, MultifrontalSolver::Parameters params={}, DampingParams dampingParams=DampingParams())
	Construct the solver from a nonlinear factor graph and linearization point.
void	load (const GaussianFactorGraph &graph)
	Load new numerical values from the factor graph.
void	eliminateInPlace (double lambda=0.0)
	Eliminate with optional LM-style damping.
void	eliminateInPlace (const GaussianFactorGraph &graph, double lambda=0.0)
	Load and eliminate the graph in a single traversal with optional damping.
Public Member Functions inherited from gtsam::MultifrontalSolver
	MultifrontalSolver (const GaussianFactorGraph &graph, const Ordering &ordering, const Parameters &params=Parameters{})
	Construct the solver from a factor graph and an ordering.
	MultifrontalSolver (const GaussianFactorGraph &graph, const Ordering &ordering, size_t firstPhaseSize, const Parameters &params=Parameters{})
	Construct a solver configured for partial multifrontal elimination.
	MultifrontalSolver (PrecomputedData data, const Ordering &ordering, const Parameters &params=Parameters{})
	Construct the solver from precomputed symbolic data.
	MultifrontalSolver (PrecomputedData data, const Ordering &ordering, size_t firstPhaseSize, const Parameters &params=Parameters{})
	Construct a partial solver from precomputed symbolic data.
void	load (const GaussianFactorGraph &graph)
	Load new numerical values from the factor graph.
void	eliminateInPlace ()
	Eliminate the graph using Cholesky factorization.
void	eliminateInPlace (const GaussianFactorGraph &graph)
	Load and eliminate the graph in a single traversal.
void	eliminatePartialInPlace ()
	Eliminate the configured ordering prefix and assemble the retained clique factors without factorizing the retained variables.
void	eliminatePartialInPlace (const GaussianFactorGraph &graph)
	Load a graph and partially eliminate it in one bottom-up traversal.
GaussianFactorGraph	remainingFactorGraph () const
	Materialize one Hessian factor per retained clique after partial elimination.
GaussianBayesTree	computeBayesTree () const
	Compute a Bayes tree from the in-place Cholesky factorization.
const VectorValues &	updateSolution ()
	Solve for the update vector.
const VectorValues &	updateSolution (const VectorValues &retainedSolution)
	Seed retained variables and back-substitute through eliminated cliques.
double	deltaError (double oldError=nullptr, double newError=nullptr) const
	Return the linearized delta error from the last updateSolution() call.
const std::vector< CliquePtr > &	roots () const
	Accessor for the roots of the elimination tree.
size_t	cliqueCount () const
	Get the total number of cliques in the solver.
void	print (const std::string &s="", const KeyFormatter &keyFormatter=DefaultKeyFormatter) const
	Print the solver state.
Public Member Functions inherited from gtsam::ForestTraversal< MultifrontalSolver, MultifrontalClique >
	ForestTraversal (size_t numThreads=std::thread::hardware_concurrency())
	Construct a helper with a fixed thread budget (used by TBB when enabled).
void	runTopDown (Fn fn, int parallelThreshold=10)
	Scheduler-based top-down traversal.
void	runBottomUp (Fn fn, int parallelThreshold=10, size_t leafAggregationProblemSize=0)
	Scheduler-based bottom-up traversal.

Static Public Member Functions
static MultifrontalSolver::PrecomputedData	Precompute (const NonlinearFactorGraph &graph, const Values &values, const Ordering &ordering)
	Precompute symbolic structure from a nonlinear factor graph and values.
Static Public Member Functions inherited from gtsam::MultifrontalSolver
static PrecomputedData	Precompute (const GaussianFactorGraph &graph, const Ordering &ordering)
	Precompute symbolic structure and sizing data from a factor graph.

Public Types
using	DampingParams = LMDampingParams
Public Types inherited from gtsam::MultifrontalSolver
using	Parameters = MultifrontalParameters
	Tuning parameters for traversal and reporting.
using	CliquePtr = std::shared_ptr<MultifrontalClique>
	Shared pointer to a MultifrontalClique.
using	Node = MultifrontalClique
	Node type for tree traversal utilities.

Additional Inherited Members
Protected Attributes inherited from gtsam::MultifrontalSolver
std::vector< CliquePtr >	roots_
	Roots of the elimination tree.
std::vector< CliquePtr >	cliques_
	All cliques in the solver.
std::map< Key, size_t >	dims_
	Map from variable key to dimension.
VectorValues	solution_
	Cached solution vector.
std::unordered_set< Key >	fixedKeys_
	Keys fixed by constrained factors.
bool	loaded_ = false
	Whether load() has been called.
bool	eliminated_ = false
	Whether eliminateInPlace() ran.
bool	partiallyEliminated_ = false
	Whether partial elimination ran.
Ordering	ordering_
	Complete variable ordering.
size_t	firstPhaseSize_ = 0
	Prefix eliminated in partial mode.
Parameters	params_
	Tunable solver parameters.
double	lastOldError_ = 0.0
	Cached old linearized error.
double	lastNewError_ = 0.0
	Cached new linearized error.
bool	hasDeltaError_ = false
	Whether updateSolution computed it.

Constructor & Destructor Documentation

◆ NonlinearMultifrontalSolver()

gtsam::NonlinearMultifrontalSolver::NonlinearMultifrontalSolver	(	const NonlinearFactorGraph &	graph,
		const Values &	values,
		const Ordering &	ordering,
		MultifrontalSolver::Parameters	params = {},
		DampingParams	dampingParams = DampingParams() )

Construct the solver from a nonlinear factor graph and linearization point.

This computes the symbolic structure (including fixed keys) from the nonlinear graph and uses the values to determine variable dimensions. Call load() with a linearized graph before eliminating.

Parameters

graph	The nonlinear factor graph to build structure from.
values	The linearization point used to determine variable dims.
ordering	The variable ordering to use.
params	Tunable parameters for traversal and reporting.

Member Function Documentation

◆ eliminateInPlace() [1/2]

void gtsam::NonlinearMultifrontalSolver::eliminateInPlace	(	const GaussianFactorGraph &	graph,
		double	lambda = 0.0 )

Load and eliminate the graph in a single traversal with optional damping.

This calls fillAb() and factorization per clique in post-order.

Parameters

graph	The linearized factor graph (structure must match).
lambda	Optional damping value; non-positive disables damping.

◆ eliminateInPlace() [2/2]

void gtsam::NonlinearMultifrontalSolver::eliminateInPlace ( double lambda = 0.0 )

Eliminate with optional LM-style damping.

When lambda is provided, adds damping on frontal blocks before factorization.

◆ load()

void gtsam::NonlinearMultifrontalSolver::load ( const GaussianFactorGraph & graph )

Load new numerical values from the factor graph.

This overrides the base load() to optionally cache diag(J^T J) for exact diagonal damping when enabled.

◆ Precompute()

MultifrontalSolver::PrecomputedData gtsam::NonlinearMultifrontalSolver::Precompute	(	const NonlinearFactorGraph &	graph,
		const Values &	values,
		const Ordering &	ordering )

static

Precompute symbolic structure from a nonlinear factor graph and values.

This avoids linearization; call load() with a linearized graph before eliminating.

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

/tmp/gtsam-4.3.0-doxygen.rsXPUS/source/gtsam/nonlinear/NonlinearMultifrontalSolver.h
/tmp/gtsam-4.3.0-doxygen.rsXPUS/source/gtsam/nonlinear/NonlinearMultifrontalSolver.cpp

Detailed Description

Public Member Functions

Static Public Member Functions

Public Types