BALL::ConjugateGradientMinimizer Class Reference
[Energy Minimizer]

A minimizer for geometry optimization based on different conjugate gradient (CG) algorithms. More...

#include <conjugateGradient.h>

Inheritance diagram for BALL::ConjugateGradientMinimizer:

[legend]

List of all members.

Classes
struct	Default
	Defaults for all options. More...
struct	Option
	Options names. More...
Public Types
Enums
enum	UpdateMethod { POLAK_RIBIERE = 1, FLETCHER_REEVES = 2, SHANNO = 3 }
	The different conjugate gradient methods implemented. More...
Public Member Functions
Constructors and Destructors
	ConjugateGradientMinimizer ()
	Default constructor.
	ConjugateGradientMinimizer (ForceField &force_field)
	Constructor expecting a valid force field.
	ConjugateGradientMinimizer (ForceField &force_field, SnapShotManager *ssm)
	Constructor expecting a valid force field and a snapshot manager.
	ConjugateGradientMinimizer (ForceField &force_field, SnapShotManager *ssm, const Options &options)
	Constructor expecting a valid force field, a snapshot manager and options.
	ConjugateGradientMinimizer (ForceField &force_field, const Options &options)
	Constructor expecting a valid force field and options.
	ConjugateGradientMinimizer (const ConjugateGradientMinimizer &rhs)
	Copy constructor.
virtual	~ConjugateGradientMinimizer ()
	Destructor.
Assignments
const ConjugateGradientMinimizer &	operator= (const ConjugateGradientMinimizer &rhs)
	Assignment operator.
Setup methods. They do all necessary preparations.
virtual bool	specificSetup ()
	Specific setup.
Accessors
void	setUpdateMethod (UpdateMethod method)
	Set explicitly the criterion used for updateDirection.
UpdateMethod	getUpdateMethod () const
	Returns the current method for updateDirection.
virtual double	findStep ()
	Calculate the next step.
virtual void	updateDirection ()
	Update the search direction.
virtual bool	minimize (Size iterations=0, bool resume=false)
	Minimize the energy of the system.
Protected Attributes
Protected Attributes
LineSearch	line_search_
	The line search.
Gradient	unscaled_direction_
	Unscaled last search direction.
Size	number_of_atoms_
	Number of movable atoms.
Size	updt_method_
	Update method used for the CG.
bool	first_iter_
	Is this the (initial) first iteration?
double	old_gtg_
	where is the gradient of the last iteration
vector< Vector3 >	a_i_
	Variable needed for the Shanno direction calculation.
vector< Vector3 >	b_i_
	Variable needed for the Shanno direction calculation.
vector< Vector3 >	p_t_
	Variable needed for the Shanno direction calculation.
vector< Vector3 >	y_t_
	Variable needed for the Shanno direction calculation.
vector< Vector3 >	p_i_
	Variable needed for the Shanno direction calculation.
vector< Vector3 >	y_i_
	Variable needed for the Shanno direction calculation.
double	D_1_
	Variable needed for the Shanno direction calculation.
double	D_4_
	Variable needed for the Shanno direction calculation.
Size	restart_frequency_
	Frequency for restarts.
Size	last_restart_iter_
	We count the iterations since the last restart.

---

Detailed Description

A minimizer for geometry optimization based on different conjugate gradient (CG) algorithms.

See

[1] Fletcher, R. and Reeves, C.M. "Function Minimization by Conjugate Gradients" Comp. J., 7, 149-154, 1964

[2] David Shanno "Conjugate Gradient Methods With Inexact Searches" Mathematics of Operations Research, Vol. 3, No. 3, August 1978, pp. 244-256

[3] S. Watowich et. al "A Stable Rapidly Converging Conjugate Gradient Method for Energy Minimization" Journal of Computational Chemistry, Vol. 9, No. 6, pp. 650-661 (1988)

[4] M.J.D. Powell "Convergence properties of algorithms for nonlinear optimization" Report No. DAMPT 1985/NA1, University of Cambridge, 1985

[5] D.F. Shanno "Globally convergent conjugate gradient algorithms" Mathematical Programming 33 (1985), pp. 61-67

---

Member Enumeration Documentation

enum BALL::ConjugateGradientMinimizer::UpdateMethod

The different conjugate gradient methods implemented.

See also:

updateDirection for details on the implementation and references

Enumerator:

POLAK_RIBIERE	Polak-Ribiere method.
FLETCHER_REEVES	Fletcher-Reeves method.
SHANNO	Shanno.

---

Member Function Documentation

virtual double BALL::ConjugateGradientMinimizer::findStep

(

)

[virtual]

Calculate the next step.

First, this method updates the model. Second, it performs a line search along the calculated direction afterwards.

Returns:

double if the line search found an acceptable solution, otherwise -1.

See also:

EnergyMinimizer::findStep

Reimplemented from BALL::EnergyMinimizer.

virtual bool BALL::ConjugateGradientMinimizer::minimize	(	Size	iterations = 0,
		bool	resume = false
	)			[virtual]

Minimize the energy of the system.

This method executes at most iterations minimization steps. If the number of iterations is not given, the number specified in the options is taken.

Parameters:

	iterations	the maximum number of iterations
	resume	true to resume a previous run

See also:

EnergyMinimizer::minimize

Reimplemented from BALL::EnergyMinimizer.

void BALL::ConjugateGradientMinimizer::setUpdateMethod

(

UpdateMethod

method

)

Set explicitly the criterion used for updateDirection.

If the method is not recognized, this function fails.

virtual void BALL::ConjugateGradientMinimizer::updateDirection

(

)

[virtual]

Update the search direction.

This method updates the search direction. It uses the different conjugate gradient caculations dependend on the options.

Reimplemented from BALL::EnergyMinimizer.

---

Member Data Documentation

Size BALL::ConjugateGradientMinimizer::last_restart_iter_ [protected]

We count the iterations since the last restart.

There are other necessities for a restart so the iteration counter may not coincide (by modulo) with the restart frequency.