TrustRegionSubProblem Class Reference

Detailed Description

The trust region subproblem introduced in

‍Yuan, Ya-xiang. Trust region algorithms for nonlinear equations. Hong Kong Baptist University, Department of Mathematics, 1994.

#include <TrustRegionSubProblem.h>

Inheritance diagram for TrustRegionSubProblem:

Public Member Functions
	TrustRegionSubProblem (const OptionSet &options)
virtual void	reinit (const NonlinearSystem &system, const Scalar &delta)
BatchTensor	preconditioned_direction (const Scalar &s) const
Public Member Functions inherited from NonlinearSystem
	NonlinearSystem (const OptionSet &options)
virtual void	init_scaling (const bool verbose=false)
	Compute algebraic Jacobian-based automatic scaling following https://cs.stanford.edu/people/paulliu/files/cs517-project.pdf.
BatchTensor	scale_residual (const BatchTensor &r) const
	Apply scaling to the residual.
BatchTensor	scale_Jacobian (const BatchTensor &J) const
	Apply scaling to the Jacobian.
BatchTensor	scale_direction (const BatchTensor &p) const
	Remove scaling from the search direction, i.e. \( J^{-1} r \).
virtual void	set_solution (const BatchTensor &x)
	Set the solution vector.
virtual BatchTensor	solution () const
	Get the solution vector.
BatchTensor	residual (const BatchTensor &x)
	Convenient shortcut to set the current solution, assemble and return the system residual.
void	residual ()
	Convenient shortcut to assemble and return the system residual.
BatchTensor	Jacobian (const BatchTensor &x)
	Convenient shortcut to set the current solution, assemble and return the system Jacobian.
void	Jacobian ()
	Convenient shortcut to assemble and return the system Jacobian.
std::tuple< BatchTensor, BatchTensor >	residual_and_Jacobian (const BatchTensor &x)
	Convenient shortcut to set the current solution, assemble and return the system residual and Jacobian.
void	residual_and_Jacobian ()
	Convenient shortcut to assemble and return the system residual and Jacobian.
const BatchTensor &	residual_view () const
const BatchTensor &	Jacobian_view () const
BatchTensor	residual_norm () const
	The residual norm.

Protected Member Functions
virtual void	assemble (bool residual, bool Jacobian) override
	Compute the residual and Jacobian.
BatchTensor	preconditioned_solve (const Scalar &s, const BatchTensor &v) const

Protected Attributes
TorchShape	_batch_sizes
torch::TensorOptions	_options
Protected Attributes inherited from NonlinearSystem
TorchSize	_ndof
	Number of degrees of freedom.
BatchTensor	_solution
	View for the solution of this nonlinear system.
BatchTensor	_residual
	View for the residual of this nonlinear system.
BatchTensor	_Jacobian
	View for the Jacobian of this nonlinear system.
BatchTensor	_scaled_residual
BatchTensor	_scaled_Jacobian
const bool	_autoscale
	If true, do automatic scaling.
const Real	_autoscale_tol
	Tolerance for convergence check of the iterative automatic scaling algorithm.
const unsigned int	_autoscale_miter
	Maximum number of iterations allowed for the iterative automatic scaling algorithm.
bool	_scaling_matrices_initialized
	Flag to indicate whether scaling matrices have been computed.
BatchTensor	_row_scaling
	Row scaling "matrix" – since it's a batched diagonal matrix, we are only storing its diagonals.
BatchTensor	_col_scaling
	Column scaling "matrix" – since it's a batched diagonal matrix, we are only storing its diagonals.

Additional Inherited Members
Static Public Member Functions inherited from NonlinearSystem
static OptionSet	expected_options ()
static void	disable_automatic_scaling (OptionSet &options)
static void	enable_automatic_scaling (OptionSet &options)

Constructor & Destructor Documentation

TrustRegionSubProblem()

TrustRegionSubProblem

(

const OptionSet &

options

)

Member Function Documentation

assemble()

void assemble ( bool residual, bool Jacobian )

overrideprotectedvirtual

Compute the residual and Jacobian.

Parameters

residual	Whether residual is requested
Jacobian	Whether Jacobian is requested

Implements NonlinearSystem.

preconditioned_direction()

BatchTensor preconditioned_direction

(

const Scalar &

s

)

const

preconditioned_solve()

BatchTensor preconditioned_solve ( const Scalar & s, const BatchTensor & v ) const

protected

reinit()

void reinit ( const NonlinearSystem & system, const Scalar & delta )

virtual

Member Data Documentation

_batch_sizes

TorchShape _batch_sizes

protected

_options

torch::TensorOptions _options

protected