Monitors whether some variable has converged to a definite value. More...

#include <Convergence_monitor.hpp>

Public Member Functions
	Convergence_monitor (double iteration_fraction)

void	add_sample (Int iteration, double value)
	Adds a sample to the value and noise monitors.

Int	n_sample () const
	Forwards to `History_stats::n_sample`

Int	last_iter () const
	Forwards to `History_stats::last_iter`

double	iteration_fraction () const
	Forwards to `History_stats::iteration_fraction`

double	last_value () const
	Forwards to `History_stats::last_value` for the value `History_stats`

double	smoothed () const
	Forwards to `History_stats::smoothed` for the value `History_stats`

double	trend () const
	Forwards to `History_stats::trend` for the value `History_stats`

double	curvature () const
	Forwards to `History_stats::curvature` for the value `History_stats`

double	noise () const
	Forwards to `History_stats::smoothed` for the noise `History_stats`

double	noise_trend () const
	Forwards to `History_stats::trend` for the noise `History_stats`

double	noise_curvature () const
	Forwards to `History_stats::curvature` for the noise `History_stats`

bool	converged (math::Tolerance trend_tol, math::Tolerance noise_tol) const
	Checks whether the variable has converged to within the specified tolerances.

void	reset ()
	Clears the sample history.

Detailed Description

Monitors whether some variable has converged to a definite value.

This is essentially a package of two History_stats objects, one of which monitors the history of the variable itself, and the other of which monitors the history of the deviation of the variable from the curve fit, which is referred to as the noise. The variable is considered to be converged if there is no significant drift over time (the trend() and curvature() are sufficiently small) and either of the following is true:

The noise() is sufficiently small, indicating that the variable has converged to a single value.
The noise() has stagnated (its trend() and curvature() are sufficiently small), indicating that the variable is displaying an oscillatory pattern that is not changing over time, and running more iterations will not provide any new information.

Constructor & Destructor Documentation

◆ Convergence_monitor()

hexed::Convergence_monitor::Convergence_monitor ( double iteration_fraction )

Parameters

iteration_fraction The iteration_fraction passed to the History_stats

Member Function Documentation

◆ add_sample()

void hexed::Convergence_monitor::add_sample	(	Int	iteration,
		double	value )

Adds a sample to the value and noise monitors.

See also: History_stats::add_sample

◆ converged()

bool hexed::Convergence_monitor::converged	(	math::Tolerance	trend_tol,
		math::Tolerance	noise_tol ) const

Checks whether the variable has converged to within the specified tolerances.

trend_tol applies to the variation of the curve fit over the last iteration_fraction iterations, where the absolute value of the linear and quadratic components are taken separately so that they can't cancel out. trend_tol.rel is relative to smoothed() + noise() so that it can still be applicable to quantities that converge to zero if they do so in an oscillatory manner. noise_tol.rel is relative to smoothed(), and trend_tol is used to assess whether the noise has stagnated, with trend_tol.rel being relative to noise() in this case.

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