doc/details_integrate_functions.qbk - RealtimeRoboticsGroup/test - Gitiles

 [/============================================================================
   Boost.odeint

   Copyright 2011-2012 Karsten Ahnert
   Copyright 2011-2012 Mario Mulansky

   Use, modification and distribution is subject to the Boost Software License,
   Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
   http://www.boost.org/LICENSE_1_0.txt)
 =============================================================================/]

 [def _max_step_checker_   [classref boost::numeric::odeint::max_step_checker `max_step_checker`]]

 [section Integrate functions]

 Integrate functions perform the time evolution of a given ODE from some
 starting time ['t[sub 0]] to a given end time ['t[sub 1]] and starting at state ['x[sub 0]] by subsequent calls of a given stepper's `do_step` function.
 Additionally, the user can provide an __observer to analyze the state during time evolution, and
 a _max_step_checker_ to throw an exception if too many steps are taken between observer calls (i.e. too
 small step size).
 There are five different integrate functions which have different strategies on when to call the observer function during integration.
 All of the integrate functions except `integrate_n_steps` can be called with any stepper following one of the stepper concepts: __stepper , __error_stepper , __controlled_stepper , __dense_output_stepper.
 Depending on the abilities of the stepper, the integrate functions make use of step-size control or dense output.

 [heading Equidistant observer calls]

 If observer calls at equidistant time intervals /dt/ are needed, the
 `integrate_const` or `integrate_n_steps` function should be used.
 We start with explaining `integrate_const`:

 `integrate_const( stepper , system , x0 , t0 , t1 , dt )`

 `integrate_const( stepper , system , x0 , t0 , t1 , dt , observer )`

 `integrate_const( stepper , system , x0 , t0 , t1 , dt , observer , max_step_checker )`

 These integrate the ODE given by `system` with subsequent steps from `stepper`.
 Integration start at `t0` and `x0` and ends at some ['t' = t[sub 0] + n dt] with /n/ such that ['t[sub 1] - dt < t' <= t[sub 1]].
 `x0` is changed to the approximative solution ['x(t')] at the end of integration.
 If provided, the `observer` is invoked at times ['t[sub 0]], ['t[sub 0] + dt], ['t[sub 0] + 2dt], ... ,['t'].
 If provided, the `max_step_checker` counts the number of steps between observer calls and throws a
 `no_progress_error` this exceeds some limit (default: 500).
 `integrate_const` returns the number of steps performed during the integration.
 Note that if you are using a simple __stepper or __error_stepper and want to make exactly `n` steps you should prefer the `integrate_n_steps` function below.

 * If `stepper` is a __stepper or __error_stepper then `dt` is also the step size used for integration and the observer is called just after every step.
 * If `stepper` is a __controlled_stepper then `dt` is the initial step size.
 The actual step size will change due to error control during time evolution.
 However, if an observer is provided the step size will be adjusted such that the algorithm always calculates /x(t)/ at ['t = t[sub 0] + n dt] and calls the observer at that point.
 Note that the use of __controlled_stepper is reasonable here only if `dt` is considerably larger than typical step sizes used by the stepper.
 * If `stepper` is a __dense_output_stepper then `dt` is the initial step size.
 The actual step size will be adjusted during integration due to error control.
 If an observer is provided dense output is used to calculate /x(t)/ at ['t = t[sub 0] + n dt].

 [heading Integrate a given number of steps]

 This function is very similar to `integrate_const` above. The only difference
 is that it does not take the end time as parameter, but rather the number of
 steps. The integration is then performed until the time `t0+n*dt`.

 `integrate_n_steps( stepper , system , x0 , t0 , dt , n )`

 `integrate_n_steps( stepper , system , x0 , t0 , dt , n , observer )`

 `integrate_n_steps( stepper , system , x0 , t0 , dt , n , observer , max_step_checker )`

 Integrates the ODE given by `system` with subsequent steps from `stepper` starting at ['x[sub 0]] and ['t[sub 0]].
 If provided, `observer` is called after every step and at the beginning with
 `t0`, similar as above.
 Again, providing a `max_step_checker` will throw a `no_progress_error` if too many steps are performed
 between observer calls.
 The approximate result for ['x( t[sub 0] + n dt )] is stored in `x0`.
 This function returns the end time `t0 + n*dt`.


 [heading Observer calls at each step]

 If the observer should be called at each time step then the `integrate_adaptive` function should be used.
 Note that in the case of __controlled_stepper or __dense_output_stepper this leads to non-equidistant observer calls as the step size changes.

 `integrate_adaptive( stepper , system , x0 , t0 , t1 , dt )`

 `integrate_adaptive( stepper , system , x0 , t0 , t1 , dt , observer )`

 Integrates the ODE given by `system` with subsequent steps from `stepper`.
 Integration start at `t0` and `x0` and ends at ['t[sub 1]].
 `x0` is changed to the approximative solution ['x(t[sub 1])] at the end of integration.
 If provided, the `observer` is called after each step (and before the first step at `t0`).
 `integrate_adaptive` returns the number of steps performed during the integration.

 [note `integrate_adaptive` by design performs an observer call after each time step. Hence
 there is no need for a _max_step_checker_ as only exactly one step is ever performed between
 observer calls.
 ]

 * If `stepper` is a __stepper or __error_stepper then `dt` is the step size used for integration and `integrate_adaptive` behaves like `integrate_const` except that for the last step the step size is reduced to ensure we end exactly at `t1`.
 If provided, the observer is called at each step.
 * If `stepper` is a __controlled_stepper then `dt` is the initial step size.
 The actual step size is changed according to error control of the stepper.
 For the last step, the step size will be reduced to ensure we end exactly at `t1`.
 If provided, the observer is called after each time step (and before the first step at `t0`).
 * If stepper is a __dense_output_stepper then `dt` is the initial step size and `integrate_adaptive` behaves just like for __controlled_stepper above. No dense output is used.

 [heading Observer calls at given time points]

 If the observer should be called at some user given time points the `integrate_times` function should be used.
 The times for observer calls are provided as a sequence of time values.
 The sequence is either defined via two iterators pointing to begin and end of the sequence or in terms of a __boost_range object.

 `integrate_times( stepper , system , x0 , times_start , times_end , dt , observer )`

 `integrate_times( stepper , system , x0 , time_range , dt , observer )`

 Integrates the ODE given by `system` with subsequent steps from `stepper`.
 Integration starts at `*times_start` and ends exactly at `*(times_end-1)`.
 `x0` contains the approximate solution at the end point of integration.
 This function requires an observer which is invoked at the subsequent times `*times_start++` until `times_start == times_end`.
 If called with a __boost_range `time_range` the function behaves the same with `times_start = boost::begin( time_range )` and `times_end = boost::end( time_range )`.
 Additionally, a _max_step_checker_ can be provided, e.g.:

 `integrate_times( stepper , system , x0 , times_start , times_end , dt , observer , max_step_checker)`

 As above, this will throw a `no_progress_error` if too many steps are performed between observer calls.

 `integrate_times` returns the number of steps performed during the integration.

 * If `stepper` is a __stepper or __error_stepper `dt` is the step size used for integration.
 However, whenever a time point from the sequence is approached the step size `dt` will be reduced to obtain the state /x(t)/ exactly at the time point.
 * If `stepper` is a __controlled_stepper then `dt` is the initial step size.
 The actual step size is adjusted during integration according to error control.
 However, if a time point from the sequence is approached the step size is reduced to obtain the state /x(t)/ exactly at the time point.
 * If `stepper` is a __dense_output_stepper then `dt` is the initial step size.
 The actual step size is adjusted during integration according to error control.
 Dense output is used to obtain the states /x(t)/ at the time points from the sequence.

 [heading Convenience integrate function]

 Additionally to the sophisticated integrate function above odeint also provides a simple `integrate` routine which uses a dense output stepper based on `runge_kutta_dopri5` with standard error bounds ['10[super -6]] for the steps.

 `integrate( system , x0 , t0 , t1 , dt )`

 `integrate( system , x0 , t0 , t1 , dt , observer )`

 This function behaves exactly like `integrate_adaptive` above but no stepper has to be provided.
 It also returns the number of steps performed during the integration.

 [endsect]
	[/============================================================================
	Boost.odeint

	Copyright 2011-2012 Karsten Ahnert
	Copyright 2011-2012 Mario Mulansky

	Use, modification and distribution is subject to the Boost Software License,
	Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
	http://www.boost.org/LICENSE_1_0.txt)
	=============================================================================/]

	[def _max_step_checker_ [classref boost::numeric::odeint::max_step_checker `max_step_checker`]]

	[section Integrate functions]

	Integrate functions perform the time evolution of a given ODE from some
	starting time ['t[sub 0]] to a given end time ['t[sub 1]] and starting at state ['x[sub 0]] by subsequent calls of a given stepper's `do_step` function.
	Additionally, the user can provide an __observer to analyze the state during time evolution, and
	a _max_step_checker_ to throw an exception if too many steps are taken between observer calls (i.e. too
	small step size).
	There are five different integrate functions which have different strategies on when to call the observer function during integration.
	All of the integrate functions except `integrate_n_steps` can be called with any stepper following one of the stepper concepts: __stepper , __error_stepper , __controlled_stepper , __dense_output_stepper.
	Depending on the abilities of the stepper, the integrate functions make use of step-size control or dense output.

	[heading Equidistant observer calls]

	If observer calls at equidistant time intervals /dt/ are needed, the
	`integrate_const` or `integrate_n_steps` function should be used.
	We start with explaining `integrate_const`:

	`integrate_const( stepper , system , x0 , t0 , t1 , dt )`

	`integrate_const( stepper , system , x0 , t0 , t1 , dt , observer )`

	`integrate_const( stepper , system , x0 , t0 , t1 , dt , observer , max_step_checker )`

	These integrate the ODE given by `system` with subsequent steps from `stepper`.
	Integration start at `t0` and `x0` and ends at some ['t' = t[sub 0] + n dt] with /n/ such that ['t[sub 1] - dt < t' <= t[sub 1]].
	`x0` is changed to the approximative solution ['x(t')] at the end of integration.
	If provided, the `observer` is invoked at times ['t[sub 0]], ['t[sub 0] + dt], ['t[sub 0] + 2dt], ... ,['t'].
	If provided, the `max_step_checker` counts the number of steps between observer calls and throws a
	`no_progress_error` this exceeds some limit (default: 500).
	`integrate_const` returns the number of steps performed during the integration.
	Note that if you are using a simple __stepper or __error_stepper and want to make exactly `n` steps you should prefer the `integrate_n_steps` function below.

	* If `stepper` is a __stepper or __error_stepper then `dt` is also the step size used for integration and the observer is called just after every step.
	* If `stepper` is a __controlled_stepper then `dt` is the initial step size.
	The actual step size will change due to error control during time evolution.
	However, if an observer is provided the step size will be adjusted such that the algorithm always calculates /x(t)/ at ['t = t[sub 0] + n dt] and calls the observer at that point.
	Note that the use of __controlled_stepper is reasonable here only if `dt` is considerably larger than typical step sizes used by the stepper.
	* If `stepper` is a __dense_output_stepper then `dt` is the initial step size.
	The actual step size will be adjusted during integration due to error control.
	If an observer is provided dense output is used to calculate /x(t)/ at ['t = t[sub 0] + n dt].

	[heading Integrate a given number of steps]

	This function is very similar to `integrate_const` above. The only difference
	is that it does not take the end time as parameter, but rather the number of
	steps. The integration is then performed until the time `t0+n*dt`.

	`integrate_n_steps( stepper , system , x0 , t0 , dt , n )`

	`integrate_n_steps( stepper , system , x0 , t0 , dt , n , observer )`

	`integrate_n_steps( stepper , system , x0 , t0 , dt , n , observer , max_step_checker )`

	Integrates the ODE given by `system` with subsequent steps from `stepper` starting at ['x[sub 0]] and ['t[sub 0]].
	If provided, `observer` is called after every step and at the beginning with
	`t0`, similar as above.
	Again, providing a `max_step_checker` will throw a `no_progress_error` if too many steps are performed
	between observer calls.
	The approximate result for ['x( t[sub 0] + n dt )] is stored in `x0`.
	This function returns the end time `t0 + n*dt`.


	[heading Observer calls at each step]

	If the observer should be called at each time step then the `integrate_adaptive` function should be used.
	Note that in the case of __controlled_stepper or __dense_output_stepper this leads to non-equidistant observer calls as the step size changes.

	`integrate_adaptive( stepper , system , x0 , t0 , t1 , dt )`

	`integrate_adaptive( stepper , system , x0 , t0 , t1 , dt , observer )`

	Integrates the ODE given by `system` with subsequent steps from `stepper`.
	Integration start at `t0` and `x0` and ends at ['t[sub 1]].
	`x0` is changed to the approximative solution ['x(t[sub 1])] at the end of integration.
	If provided, the `observer` is called after each step (and before the first step at `t0`).
	`integrate_adaptive` returns the number of steps performed during the integration.

	[note `integrate_adaptive` by design performs an observer call after each time step. Hence
	there is no need for a _max_step_checker_ as only exactly one step is ever performed between
	observer calls.
	]

	* If `stepper` is a __stepper or __error_stepper then `dt` is the step size used for integration and `integrate_adaptive` behaves like `integrate_const` except that for the last step the step size is reduced to ensure we end exactly at `t1`.
	If provided, the observer is called at each step.
	* If `stepper` is a __controlled_stepper then `dt` is the initial step size.
	The actual step size is changed according to error control of the stepper.
	For the last step, the step size will be reduced to ensure we end exactly at `t1`.
	If provided, the observer is called after each time step (and before the first step at `t0`).
	* If stepper is a __dense_output_stepper then `dt` is the initial step size and `integrate_adaptive` behaves just like for __controlled_stepper above. No dense output is used.

	[heading Observer calls at given time points]

	If the observer should be called at some user given time points the `integrate_times` function should be used.
	The times for observer calls are provided as a sequence of time values.
	The sequence is either defined via two iterators pointing to begin and end of the sequence or in terms of a __boost_range object.

	`integrate_times( stepper , system , x0 , times_start , times_end , dt , observer )`

	`integrate_times( stepper , system , x0 , time_range , dt , observer )`

	Integrates the ODE given by `system` with subsequent steps from `stepper`.
	Integration starts at `times_start` and ends exactly at `(times_end-1)`.
	`x0` contains the approximate solution at the end point of integration.
	This function requires an observer which is invoked at the subsequent times `*times_start++` until `times_start == times_end`.
	If called with a __boost_range `time_range` the function behaves the same with `times_start = boost::begin( time_range )` and `times_end = boost::end( time_range )`.
	Additionally, a _max_step_checker_ can be provided, e.g.:

	`integrate_times( stepper , system , x0 , times_start , times_end , dt , observer , max_step_checker)`

	As above, this will throw a `no_progress_error` if too many steps are performed between observer calls.

	`integrate_times` returns the number of steps performed during the integration.

	* If `stepper` is a __stepper or __error_stepper `dt` is the step size used for integration.
	However, whenever a time point from the sequence is approached the step size `dt` will be reduced to obtain the state /x(t)/ exactly at the time point.
	* If `stepper` is a __controlled_stepper then `dt` is the initial step size.
	The actual step size is adjusted during integration according to error control.
	However, if a time point from the sequence is approached the step size is reduced to obtain the state /x(t)/ exactly at the time point.
	* If `stepper` is a __dense_output_stepper then `dt` is the initial step size.
	The actual step size is adjusted during integration according to error control.
	Dense output is used to obtain the states /x(t)/ at the time points from the sequence.

	[heading Convenience integrate function]

	Additionally to the sophisticated integrate function above odeint also provides a simple `integrate` routine which uses a dense output stepper based on `runge_kutta_dopri5` with standard error bounds ['10[super -6]] for the steps.

	`integrate( system , x0 , t0 , t1 , dt )`

	`integrate( system , x0 , t0 , t1 , dt , observer )`

	This function behaves exactly like `integrate_adaptive` above but no stepper has to be provided.
	It also returns the number of steps performed during the integration.

	[endsect]