Generalized version of NonlinearSolver not limited to a single madness function. More...

#include <nonlinsol.h>

Collaboration diagram for madness::XNonlinearSolver< T, C, Alloc >:

[legend]

Public Member Functions
	XNonlinearSolver (const Alloc &alloc=Alloc(), bool print=false)

	XNonlinearSolver (const XNonlinearSolver &other)

void	clear_subspace ()

Tensor< C >	get_c () const

std::vector< T > &	get_plist ()

std::vector< T > &	get_rlist ()

std::vector< T > &	get_ulist ()

void	initialize (const T &u_initial)

void	set_maxsub (int maxsub)

T	update (const T &u, const T &r, const double rcondtol=1e-8, const double cabsmax=1000.0)
	Computes next trial solution vector.

T	update (const T &u_preliminary, const double rcondtol=1e-8, const double cabsmax=1000.0)
	Computes next trial solution vector.

Public Attributes
bool	do_print

Private Attributes
Alloc	alloc

Tensor< C >	c
	coefficients for linear combination

unsigned int	maxsub
	Maximum size of subspace dimension.

std::vector< T >	plist
	Subspace information.

Tensor< C >	Q

std::vector< T >	rlist

std::vector< T >	ulist

Detailed Description

template<class T, class C = double, class Alloc = default_allocator<T>>
class madness::XNonlinearSolver< T, C, Alloc >

Generalized version of NonlinearSolver not limited to a single madness function.

This solves the equation $r(u) = 0$ where u and r are both of type T and inner products between two items of type T produce a number of type C (defaulting to double). The type T must support storage in an STL vector, scaling by a constant of type C, inplace addition (+=), subtraction, allocation with value zero, and inner products computed with the interface inner(a,b). Have a look in examples/testsolver.cc for a simple but complete example, and in examples/h2dynamic.cc for a more complex example.

I've not yet tested with anything except C=double and I think that the KAIN routine will need extending for anything else.

Constructor & Destructor Documentation

◆ XNonlinearSolver() [1/2]

template<class T , class C = double, class Alloc = default_allocator<T>>

madness::XNonlinearSolver< T, C, Alloc >::XNonlinearSolver	(	const Alloc &	alloc = `Alloc()`,
		bool	print = `false`
	)

inline

◆ XNonlinearSolver() [2/2]

template<class T , class C = double, class Alloc = default_allocator<T>>

madness::XNonlinearSolver< T, C, Alloc >::XNonlinearSolver ( const XNonlinearSolver< T, C, Alloc > & other )

inline

Member Function Documentation

◆ clear_subspace()

template<class T , class C = double, class Alloc = default_allocator<T>>

void madness::XNonlinearSolver< T, C, Alloc >::clear_subspace ( )

inline

References madness::nonlinear_vector_solver(), madness::XNonlinearSolver< T, C, Alloc >::plist, madness::XNonlinearSolver< T, C, Alloc >::Q, madness::XNonlinearSolver< T, C, Alloc >::rlist, and madness::XNonlinearSolver< T, C, Alloc >::ulist.

Referenced by madness::XNonlinearSolver< T, C, Alloc >::initialize(), and iterate().

◆ get_c()

template<class T , class C = double, class Alloc = default_allocator<T>>

Tensor< C > madness::XNonlinearSolver< T, C, Alloc >::get_c ( ) const

inline

References madness::XNonlinearSolver< T, C, Alloc >::c.

◆ get_plist()

template<class T , class C = double, class Alloc = default_allocator<T>>

std::vector< T > & madness::XNonlinearSolver< T, C, Alloc >::get_plist ( )

inline

References madness::XNonlinearSolver< T, C, Alloc >::plist.

◆ get_rlist()

template<class T , class C = double, class Alloc = default_allocator<T>>

std::vector< T > & madness::XNonlinearSolver< T, C, Alloc >::get_rlist ( )

inline

References madness::XNonlinearSolver< T, C, Alloc >::rlist.

◆ get_ulist()

template<class T , class C = double, class Alloc = default_allocator<T>>

std::vector< T > & madness::XNonlinearSolver< T, C, Alloc >::get_ulist ( )

inline

References madness::XNonlinearSolver< T, C, Alloc >::ulist.

◆ initialize()

template<class T , class C = double, class Alloc = default_allocator<T>>

void madness::XNonlinearSolver< T, C, Alloc >::initialize ( const T & u_initial )

inline

References madness::XNonlinearSolver< T, C, Alloc >::alloc, madness::XNonlinearSolver< T, C, Alloc >::clear_subspace(), madness::nonlinear_vector_solver(), madness::XNonlinearSolver< T, C, Alloc >::plist, and madness::XNonlinearSolver< T, C, Alloc >::ulist.

Referenced by test_with_function().

◆ set_maxsub()

template<class T , class C = double, class Alloc = default_allocator<T>>

void madness::XNonlinearSolver< T, C, Alloc >::set_maxsub ( int maxsub )

inline

References madness::XNonlinearSolver< T, C, Alloc >::maxsub.

Referenced by iterate(), run(), and madness::PNO::t_solve().

◆ update() [1/2]

template<class T , class C = double, class Alloc = default_allocator<T>>

T madness::XNonlinearSolver< T, C, Alloc >::update	(	const T &	u,
		const T &	r,
		const double	rcondtol = `1e-8`,
		const double	cabsmax = `1000.0`
	)

inline

Computes next trial solution vector.

You are responsible for performing step restriction or line search (not necessary for linear problems).

Parameters

u	Current solution vector
r	Corresponding residual

Returns: Next trial solution vector

Parameters

[in]	rcondtol	rcond less than this will cause the subspace to be shrunk due to linear dependence
[in]	cabsmax	maximum element of c greater than this will cause the subspace to be shrunk due to li

References madness::XNonlinearSolver< T, C, Alloc >::alloc, madness::XNonlinearSolver< T, C, Alloc >::c, madness::check_linear_dependence(), madness::copy(), madness::XNonlinearSolver< T, C, Alloc >::do_print, madness::inner(), madness::KAIN(), madness::XNonlinearSolver< T, C, Alloc >::maxsub, madness::nonlinear_vector_solver(), madness::print(), madness::XNonlinearSolver< T, C, Alloc >::Q, madness::XNonlinearSolver< T, C, Alloc >::rlist, T(), u(), and madness::XNonlinearSolver< T, C, Alloc >::ulist.

Referenced by iterate(), madness::PNO::t_solve(), and test_with_function().

◆ update() [2/2]

template<class T , class C = double, class Alloc = default_allocator<T>>

T madness::XNonlinearSolver< T, C, Alloc >::update	(	const T &	u_preliminary,
		const double	rcondtol = `1e-8`,
		const double	cabsmax = `1000.0`
	)

inline

Computes next trial solution vector.

this version of update avoids using the residual by computing it on the fly note the distinction between u_preliminary (input) and u_kain (output) residual[i] = u_kain[i] - u_preliminary[i+1] ulist[i] = u_kain[i] plist[i] = u_preliminary[i]

Parameters