Drake

An abstract class for an integrator for ODEs and DAEs as represented by a Drake System. More...
#include <drake/systems/analysis/integrator_base.h>
Public Types  
enum  StepResult { kReachedPublishTime = 1, kReachedZeroCrossing = 2, kReachedUpdateTime = 3, kTimeHasAdvanced = 4, kReachedBoundaryTime = 5, kReachedStepLimit = 6 } 
Status returned by StepOnceAtMost(). More...  
Public Member Functions  
IntegratorBase (const System< T > &system, Context< T > *context=nullptr)  
Maintains references to the system being integrated and the context used to specify the initial conditions for that system (if any). More...  
virtual  ~IntegratorBase ()=default 
Destructor. More...  
virtual bool  supports_error_estimation () const =0 
Indicates whether an integrator supports error estimation. More...  
void  set_fixed_step_mode (bool flag) 
Sets an integrator with error control to fixed step mode. More...  
bool  get_fixed_step_mode () const 
Gets whether an integrator is running in fixed step mode. More...  
void  set_target_accuracy (double accuracy) 
Request that the integrator attempt to achieve a particular accuracy for the continuous portions of the simulation. More...  
double  get_target_accuracy () const 
Gets the target accuracy. More...  
double  get_accuracy_in_use () const 
Gets the accuracy in use by the integrator. More...  
void  set_maximum_step_size (const T &max_step_size) 
Sets the maximum step size that may be taken by this integrator. More...  
const T &  get_maximum_step_size () const 
Gets the maximum step size that may be taken by this integrator. More...  
void  Reset () 
Resets the integrator to initial values, i.e., default construction values. More...  
void  Initialize () 
An integrator must be initialized before being used. More...  
void  request_initial_step_size_target (const T &step_size) 
Request that the first attempted integration step have a particular size. More...  
const T &  get_initial_step_size_target () const 
Gets the target size of the first integration step. More...  
StepResult  IntegrateAtMost (const T &publish_dt, const T &update_dt, const T &boundary_dt) 
Integrates the system forward in time by a single step with step size subject to integration error tolerances (assuming that the integrator supports error estimation). More...  
double  get_stretch_factor () const 
Gets the stretch factor (> 1), which is multiplied by the maximum (typically userdesignated) integration step size to obtain the amount that the integrator is able to stretch the maximum time step toward hitting an upcoming publish or update event in IntegrateAtMost(). More...  
void  IntegrateWithMultipleSteps (const T &dt) 
Stepping function for integrators operating outside of Simulator that advances the continuous state exactly by dt . More...  
void  IntegrateWithSingleFixedStep (const T &dt) 
Stepping function for integrators operating outside of Simulator that advances the continuous state exactly by dt and using a single fixed step. More...  
const T &  get_ideal_next_step_size () const 
Return the step size the integrator would like to take next, based primarily on the integrator's accuracy prediction. More...  
const Context< T > &  get_context () const 
Returns a const reference to the internallymaintained Context holding the most recent state in the trajectory. More...  
Context< T > *  get_mutable_context () 
Returns a mutable pointer to the internallymaintained Context holding the most recent state in the trajectory. More...  
void  reset_context (Context< T > *context) 
Replace the pointer to the internallymaintained Context with a different one. More...  
const System< T > &  get_system () const 
Gets a constant reference to the system that is being integrated (and was provided to the constructor of the integrator). More...  
bool  is_initialized () const 
Indicates whether the integrator has been initialized. More...  
virtual int  get_error_estimate_order () const =0 
Derived classes must override this function to return the order of the integrator's error estimate. More...  
const T &  get_previous_integration_step_size () const 
Gets the size of the last (previous) integration step. More...  
const ContinuousState< T > *  get_error_estimate () const 
Gets the error estimate (used only for integrators that support error estimation). More...  
Does not allow copy, move, or assignment  
IntegratorBase (const IntegratorBase &)=delete  
IntegratorBase &  operator= (const IntegratorBase &)=delete 
IntegratorBase (IntegratorBase &&)=delete  
IntegratorBase &  operator= (IntegratorBase &&)=delete 
Methods for minimum integration step size selection and behavior  
Variable step integrators reduce their step sizes as needed to achieve requirements such as specified accuracy or step convergence. However, it is not possible to take an arbitrarily small step. Normally integrators choose an appropriate minimum step and throw an exception if the requirements can't be achieved without going below that. Methods in this section allow you to influence two aspects of this procedure:
By default, integrators allow a very small minimum step which can result in long run times. Setting a larger minimum can be helpful as a diagnostic to figure out what aspect of your simulation is requiring small steps. You can set the minimum to what should be a "reasonable" minimum based on what you know about the physical system. You will then get an std::runtime_error exception thrown at any point in time where your model behaves unexpectedly (due to, e.g., a discontinuity in the derivative evaluation function). If you disable the exception (via DetailsBecause time is maintained to finite precision, there is an absolute minimum step size You may request a larger minimum step size Under some circumstances the integrator may legitimately take a step of size  
void  set_requested_minimum_step_size (const T &min_step_size) 
Sets the requested minimum step size h_min that may be taken by this integrator. More...  
const T &  get_requested_minimum_step_size () const 
Gets the requested minimum step size h_min for this integrator. More...  
void  set_throw_on_minimum_step_size_violation (bool throws) 
Sets whether the integrator should throw a std::runtime_error exception when the integrator's step size selection algorithm determines that it must take a step smaller than the minimum step size (for, e.g., purposes of error control). More...  
bool  get_throw_on_minimum_step_size_violation () const 
Reports the current setting of the throw_on_minimum_step_size_violation flag. More...  
T  get_working_minimum_step_size () const 
Gets the current value of the working minimum step size h_work(t) for this integrator, which may vary with the current time t as stored in the integrator's context. More...  
Integrator statistics methods.  
These methods allow the caller to manipulate and query integrator statistics. Generally speaking, the larger the integration step taken, the faster a simulation will run. These methods allow querying (and resetting) the integrator statistics as one means of determining how to make a simulation run faster.  
void  ResetStatistics () 
Forget accumulated statistics. More...  
int64_t  get_num_substep_failures () const 
Gets the number of failed substeps (implying one or more step size reductions was required to permit solving the necessary nonlinear system of equations). More...  
int64_t  get_num_step_shrinkages_from_substep_failures () const 
Gets the number of step size shrinkages due to substep failures (e.g., integrator convergence failures) since the last call to ResetStatistics() or Initialize(). More...  
int64_t  get_num_step_shrinkages_from_error_control () const 
Gets the number of step size shrinkages due to failure to meet targeted error tolerances, since the last call to ResetStatistics or Initialize(). More...  
int64_t  get_num_derivative_evaluations () const 
Returns the number of ODE function evaluations (calls to CalcTimeDerivatives()) since the last call to ResetStatistics() or Initialize(). More...  
const T &  get_actual_initial_step_size_taken () const 
The actual size of the successful first step. More...  
const T &  get_smallest_adapted_step_size_taken () const 
The size of the smallest step taken as the result of a controlled integration step adjustment since the last Initialize() or ResetStatistics() call. More...  
const T &  get_largest_step_size_taken () const 
The size of the largest step taken since the last Initialize() or ResetStatistics() call. More...  
int64_t  get_num_steps_taken () const 
The number of integration steps taken since the last Initialize() or ResetStatistics() call. More...  
Methods for dense output computation  
In general, dense output computations entail both CPU load and memory footprint increases during numerical integration. For some applications, the performance penalty may be prohibitive. As such, these computations are only carried out by explicit user request. The API to start and stop a dense integration process (i.e. a numerical integration process that also computes dense output) is consistent with this design choice. Once dense integration is started, and until it is stopped, all subsequent integration steps taken will update the allocated dense output.  
void  StartDenseIntegration () 
Starts dense integration, allocating a new dense output for this integrator to use. More...  
const DenseOutput< T > *  get_dense_output () const 
Returns a const pointer to the integrator's current DenseOutput instance, holding a representation of the continuous state trajectory since the last StartDenseIntegration() call. More...  
std::unique_ptr< DenseOutput< T > >  StopDenseIntegration () 
Stops dense integration, yielding ownership of the current dense output to the caller. More...  
Methods for weighting state variable errors  
This group of methods describes how errors for state variables with heterogeneous units are weighted in the context of errorcontrolled integration. This is an advanced topic and most users can simply specify desired accuracy and accept the default state variable weights. A collection of state variables is generally defined in heterogeneous units (e.g. length, angles, velocities, energy). Some of the state variables cannot even be expressed in meaningful units, like quaternions. Certain integrators provide an estimate of the absolute error made in each state variable during an integration step. These errors must be properly weighted to obtain an "accuracy" with respect to each particular variable. These pervariable accuracy determinations can be compared against the user's requirements and used to select an appropriate size for the next step [Sherman 2011]. The weights are normally determined automatically using the system's characteristic dimensions, so most users can stop reading now! Custom weighting is primarily useful for performance improvement; an optimal weighting would allow an errorcontrolled integrator to provide the desired level of accuracy across all state variables without wasting computation achieving superfluous accuracy for some of those variables. Users interested in more precise control over state variable weighting may use the methods in this group to access and modify weighting factors for individual state variables. Changes to these weights can only be made prior to integrator initialization or as a result of an event being triggered and then followed by reinitialization. Relative versus absolute accuracyState variable integration error, as estimated by an integrator, is an absolute quantity with the same units as the variable. At each time step we therefore need to determine an absolute error that would be deemed "good enough", i.e. satisfies the user's accuracy requirement. If a variable is maintained to a relative accuracy then that "good enough" value is defined to be the required accuracy How to choose weightsThe weight Some subtleties for secondorder dynamic systemsSystems governed by 2ndorder differential equations are typically split into second order (configuration) variables q, and rate (velocity) variables v, where the time derivatives qdot of q are linearly related to v by the kinematic differential equation Note that generalized quasicoordinates To summarize, separate weights can be provided for each of
Weights on the generalized velocity variables How the weights are usedThe errors in the The weighting matrices Eq = Wq Ev: Ev(i,i) = { min(Wv(i,i), 1/vᵢ) if vᵢ is relative { Wv(i,i) if vᵢ is absolute Ez: Ez(i,i) = { min(Wz(i,i), 1/zᵢ) if zᵢ is relative { Wz(i,i) if zᵢ is absolute ( Now given an error estimate vector Determining weights for qThe kinematic differential equations v = N⁺ qdot Inverse kinematic differential equation dꝗ/dt = N⁺ dq/dt Use synonyms for v and qdot dꝗ = N⁺ dq Change time derivatives to differentials Wꝗ dꝗ = Wꝗ N⁺ dq Premultiply both sides by Wꝗ N Wꝗ dꝗ = N Wꝗ N⁺ dq Premultiply both sides by N N Wꝗ dꝗ = Wq dq Define Wq := N Wꝗ N⁺ N Wꝗ v = Wq qdot Back to time derivatives. The last two equations show that Finally, note that a diagonal entry of one of the weighting matrices can be set to zero to disable error estimation for that state variable (i.e., auxiliary variable or configuration/velocity variable pair), but that setting an entry to a negative value will cause an exception to be thrown when the integrator is initialized.
 
const Eigen::VectorXd &  get_generalized_state_weight_vector () const 
Gets the weighting vector (equivalent to a diagonal matrix) applied to weighting both generalized coordinate and velocity state variable errors, as described in the group documentation. More...  
Eigen::VectorBlock< Eigen::VectorXd >  get_mutable_generalized_state_weight_vector () 
Gets a mutable weighting vector (equivalent to a diagonal matrix) applied to weighting both generalized coordinate and velocity state variable errors, as described in the group documentation. More...  
const Eigen::VectorXd &  get_misc_state_weight_vector () const 
Gets the weighting vector (equivalent to a diagonal matrix) for weighting errors in miscellaneous continuous state variables z . More...  
Eigen::VectorBlock< Eigen::VectorXd >  get_mutable_misc_state_weight_vector () 
Gets a mutable weighting vector (equivalent to a diagonal matrix) for weighting errors in miscellaneous continuous state variables z . More...  
Protected Member Functions  
virtual void  DoResetStatistics () 
Resets any statistics particular to a specific integrator. More...  
void  CalcTimeDerivatives (const Context< T > &context, ContinuousState< T > *dxdt) 
Evaluates the derivative function (and updates call statistics). More...  
template<typename U >  
void  CalcTimeDerivatives (const System< U > &system, const Context< U > &context, ContinuousState< U > *dxdt) 
Evaluates the derivative function (and updates call statistics). More...  
void  set_accuracy_in_use (double accuracy) 
Sets the working ("in use") accuracy for this integrator. More...  
void  InitializeAccuracy (double default_accuracy, double loosest_accuracy, double max_step_fraction) 
Generic code for validating (and resetting, if need be) the integrator working accuracy for error controlled integrators. More...  
bool  StepOnceErrorControlledAtMost (const T &dt_max) 
Default code for advancing the continuous state of the system by a single step of dt_max (or smaller, depending on error control). More...  
T  CalcStateChangeNorm (const ContinuousState< T > &dx_state) const 
Computes the infinity norm of a change in continuous state. More...  
std::pair< bool, T >  CalcAdjustedStepSize (const T &err, const T &attempted_step_size, bool *at_minimum_step_size) const 
Calculates adjusted integrator step sizes toward keeping state variables within error bounds on the next integration step. More...  
virtual void  DoInitialize () 
Derived classes can override this method to perform special initialization. More...  
virtual void  DoReset () 
Derived classes can override this method to perform routines when Reset() is called. More...  
virtual std::unique_ptr< StepwiseDenseOutput< T > >  DoStartDenseIntegration () 
Derived classes can override this method to provide a continuous extension of their own when StartDenseIntegration() is called. More...  
StepwiseDenseOutput< T > *  get_mutable_dense_output () 
Returns a mutable pointer to the internallymaintained StepwiseDenseOutput instance, holding a representation of the continuous state trajectory since the last time StartDenseIntegration() was called. More...  
virtual bool  DoStep (const T &dt)=0 
Derived classes must implement this method to (1) integrate the continuous portion of this system forward by a single step of size dt and (2) set the error estimate (via get_mutable_error_estimate()). More...  
virtual bool  DoDenseStep (const T &dt) 
Derived classes may implement this method to (1) integrate the continuous portion of this system forward by a single step of size dt , (2) set the error estimate (via get_mutable_error_estimate()) and (3) update their own dense output implementation (via get_mutable_dense_output()). More...  
ContinuousState< T > *  get_mutable_error_estimate () 
Gets an error estimate of the state variables recorded by the last call to StepOnceFixedSize(). More...  
void  set_actual_initial_step_size_taken (const T &dt) 
void  set_smallest_adapted_step_size_taken (const T &dt) 
Sets the size of the smalleststeptaken statistic as the result of a controlled integration step adjustment. More...  
void  set_largest_step_size_taken (const T &dt) 
void  set_ideal_next_step_size (const T &dt) 
An abstract class for an integrator for ODEs and DAEs as represented by a Drake System.
Integrators solve initial value problems of the form:
ẋ(t) = f(t, x(t)) with f : ℝ × ℝⁿ → ℝⁿ
(i.e., f()
is an ordinary differential equation) given initial conditions (t₀, x₀). Thus, integrators advance the continuous state of a dynamical system forward in time.
Apart from solving initial value problems, for which the integrator is a key component of a simulator, integrators can also be used to solve boundary value problems (via numerical methods like the Multiple Shooting Method) and trajectory optimization problems (via numerical methods like direct transcription). This class and its derivatives were developed primarily toward the former application (through IntegrateAtMost() and the Simulator class). However, the IntegratorBase architecture was developed to support these ancillary applications as well using the IntegrateWithMultipleSteps() and IntegrateWithSingleFixedStep() methods; the latter permits the caller to advance time using fixed steps in applications where variable stepping would be deleterious (e.g., direct transcription).
For applications that require a more dense sampling of the system continuous state than what would be available through either fixed or errorcontrolled step integration (for a given accuracy), dense output support is available (through StartDenseIntegration() and StopDenseIntegration() methods). The accuracy and performance of these outputs may vary with each integration scheme implementation. Unless specified otherwise, an HermitianDenseOutput is provided by default.
A natural question for a user to ask of an integrator is: Which scheme (method) should be applied to a particular problem? The answer is whichever one most quickly computes the solution to the desired accuracy! Selecting an integration scheme for a particular problem is presently an artform. As examples of some selection criteria: multistep methods (none of which are currently implemented in Drake) generally work poorly when events (that require state reinitializations) are common, symplectic methods generally work well at maintaining stability for large integration steps, and stiff integrators are often best for computationally stiff systems. If ignorant as to the characteristics of a particular problem, it is often best to start with an explicit, RungeKutta type method. Statistics collected by the integrator can help diagnose performance issues and possibly point to the use of a different integration scheme.
Some systems are known to exhibit "computational stiffness", by which it is meant that (excessively) small integration steps are necessary for purposes of stability: in other words, steps must be taken smaller than that required to achieve a desired accuracy over a particular interval. Thus, the nature of computationally stiff problems is that the solution to the ODE is smooth in the interval of stiffness (in contrast, some problems possess such high frequency dynamics that very small steps are simply necessary to capture the solution accurately). Implicit integrators are the goto approach for solving computationally stiff problems, but careful consideration is warranted. Implicit integrators typically require much more computation than nonimplicit (explicit) integrators, stiffness might be an issue on only a very small time interval, and some problems might be only "moderately stiff". Put another way, applying an implicit integrator to a potentially stiff problem might not yield faster computation. The first chapter of [Hairer, 1996] illustrates the issues broached in this paragraph using various examples.
Established methods for integrating ordinary differential equations invariably make provisions for estimating the "local error" (i.e., the error over a small time interval) of a solution. Although the relationship between local error and global error (i.e., the accumulated error over multiple time steps) can be tenuous, such error estimates can allow integrators to work adaptively, subdividing time intervals as necessary (if, e.g., the system is particularly dynamic or stationary in an interval). Even for applications that do not recommend such adaptive integration like direct transcription methods for trajectory optimization error estimation allows the user to assess the accuracy of the solution.
IntegratorBase provides numerous settings and flags that can leverage problemspecific information to speed integration and/or improve integration accuracy. As an example, set_maximum_step_size() allows the user to prevent overly large integration steps (that integration error control alone might be insufficient to detect). As noted previously, IntegratorBase also collects a plethora of statistics that can be used to diagnose poor integration performance. For example, a large number of shrinkages due to error control could indicate that a system is computationally stiff.
T  The vector element type, which must be a valid Eigen scalar. 
enum StepResult 
Status returned by StepOnceAtMost().
When a step is successful, it will return an indication of what caused it to stop where it did. When unsuccessful it will throw an exception so you won't see any return value. When return of control is due ONLY to reaching a publish time, (status is kReachedPublishTime) the context may return an interpolated value at an earlier time.

delete 

delete 

inlineexplicit 
Maintains references to the system being integrated and the context used to specify the initial conditions for that system (if any).
system  A reference to the system to be integrated; the integrator will maintain a reference to the system in perpetuity, so the integrator must not outlive the system. 
context  A pointer to a writeable context (nullptr is ok, but a nonnull pointer must be set before Initialize() is called). The integrator will advance the system state using the pointer to this context. The pointer to the context will be maintained internally. The integrator must not outlive the context. 

virtualdefault 
Destructor.

protected 
Calculates adjusted integrator step sizes toward keeping state variables within error bounds on the next integration step.
Note that it is not guaranteed that the (possibly) reduced step size will keep state variables within error bounds; however, the process of (1) taking a trial integration step, (2) calculating the error, and (3) adjusting the step size can be repeated until convergence.
err  The norm of the integrator error that was computed using attempted_step_size .  
attempted_step_size  The step size that was attempted.  
[in,out]  at_minimum_step_size  If true on entry, the error control mechanism is not allowed to shrink the step because the integrator is stepping at the minimum step size (note that this condition will only occur if get_throw_on_minimum_step_size_violation() == false  an exception would be thrown otherwise). If true on entry and false on exit, the error control mechanism has managed to increase the step size above the working minimum; if true on entry and true on exit, error control would like to shrink the step size but cannot. If false on entry and true on exit, error control shrank the step to the working minimum step size. 
true
if the integration step was to be considered successful and false
otherwise. The value of the T type will be set to the recommended next step size.

protected 
Computes the infinity norm of a change in continuous state.
We use the infinity norm to capture the idea that, by providing accuracy requirements, the user can indirectly specify error tolerances that act to limit the largest error in any state vector component.

inlineprotected 
Evaluates the derivative function (and updates call statistics).
Subclasses should call this function rather than calling system.CalcTimeDerivatives() directly.

inlineprotected 
Evaluates the derivative function (and updates call statistics).
Subclasses should call this function rather than calling system.CalcTimeDerivatives() directly. This version of this function exists to allow integrators to count AutoDiff'd systems in derivative function evaluations.
Derived classes may implement this method to (1) integrate the continuous portion of this system forward by a single step of size dt
, (2) set the error estimate (via get_mutable_error_estimate()) and (3) update their own dense output implementation (via get_mutable_dense_output()).
This method is called during the default Step() method.
dt  The integration step to take. 
true
if successful, false
if either the integrator was unable to take a single step of size dt
or to advance its dense output an equal step.

inlineprotectedvirtual 
Derived classes can override this method to perform special initialization.
This method is called during the Initialize() method. This default method does nothing.

inlineprotectedvirtual 
Derived classes can override this method to perform routines when Reset() is called.
This default method does nothing.

inlineprotectedvirtual 
Resets any statistics particular to a specific integrator.
The default implementation of this function does nothing. If your integrator collects its own statistics, you should reimplement this method and reset them there.

inlineprotectedvirtual 
Derived classes can override this method to provide a continuous extension of their own when StartDenseIntegration() is called.
Derived classes must implement this method to (1) integrate the continuous portion of this system forward by a single step of size dt
and (2) set the error estimate (via get_mutable_error_estimate()).
This method is called during the default Step() method.
dt  The integration step to take. 
true
if successful, false
if the integrator was unable to take a single step of size dt
(due to, e.g., an integrator convergence failure). t
, the time and state will be advanced to t+dt
if the method returns true
; otherwise, the time and state should be reset to those at t
. true
for some, albeit possibly very small, positive value of dt
. The derived integrator's stepping algorithm can make this guarantee, for example, by switching to an algorithm not subject to convergence failures (e.g., explicit Euler) for very small step sizes.

inline 
Gets the accuracy in use by the integrator.
This number may differ from the target accuracy if, for example, the user has requested an accuracy not attainable or not recommended for the particular integrator.

inline 
The actual size of the successful first step.
Returns a const reference to the internallymaintained Context holding the most recent state in the trajectory.
This is suitable for publishing or extracting information about this trajectory step.

inline 
Returns a const pointer to the integrator's current DenseOutput instance, holding a representation of the continuous state trajectory since the last StartDenseIntegration() call.
This is suitable to query the integrator's current dense output, if any (may be nullptr).

inline 
Gets the error estimate (used only for integrators that support error estimation).
If the integrator does not support error estimation, nullptr is returned.

pure virtual 
Derived classes must override this function to return the order of the integrator's error estimate.
If the integrator does not provide an error estimate, the derived class implementation should return 0.
Implemented in ImplicitEulerIntegrator< T >, SemiExplicitEulerIntegrator< T >, RungeKutta3Integrator< T >, ExplicitEulerIntegrator< T >, and RungeKutta2Integrator< T >.

inline 
Gets whether an integrator is running in fixed step mode.
If the integrator does not support error estimation, this function will always return true
. If the integrator runs in fixed step mode, it will always take the maximum step size directed (which may be that determined by get_maximum_step_size() or may be smaller, as directed by, e.g., Simulator for event handling purposes).

inline 
Gets the weighting vector (equivalent to a diagonal matrix) applied to weighting both generalized coordinate and velocity state variable errors, as described in the group documentation.
Only used for integrators that support error estimation.

inline 
Return the step size the integrator would like to take next, based primarily on the integrator's accuracy prediction.
This value will not be computed for integrators that do not support error estimation and NaN will be returned.

inline 
Gets the target size of the first integration step.
You can find out what step size was actually used for the first integration step with get_actual_initial_step_size_taken()
.

inline 
The size of the largest step taken since the last Initialize() or ResetStatistics() call.

inline 
Gets the maximum step size that may be taken by this integrator.
This is a soft maximum: the integrator may stretch it by as much as 1% to hit a discrete event.

inline 
Gets the weighting vector (equivalent to a diagonal matrix) for weighting errors in miscellaneous continuous state variables z
.
Only used for integrators that support error estimation.
Returns a mutable pointer to the internallymaintained Context holding the most recent state in the trajectory.

inlineprotected 
Returns a mutable pointer to the internallymaintained StepwiseDenseOutput instance, holding a representation of the continuous state trajectory since the last time StartDenseIntegration() was called.
This is useful for derived classes to update the integrator's current dense output, if any (may be nullptr).

inlineprotected 
Gets an error estimate of the state variables recorded by the last call to StepOnceFixedSize().
If the integrator does not support error estimation, this function will return nullptr.

inline 
Gets a mutable weighting vector (equivalent to a diagonal matrix) applied to weighting both generalized coordinate and velocity state variable errors, as described in the group documentation.
Only used for integrators that support error estimation. Returns a VectorBlock to make the values mutable without permitting changing the size of the vector. Requires reinitializing the integrator after calling this method; if Initialize() is not called afterward, an exception will be thrown when attempting to call StepOnceAtMost(). If the caller sets one of the entries to a negative value, an exception will be thrown when the integrator is initialized.

inline 
Gets a mutable weighting vector (equivalent to a diagonal matrix) for weighting errors in miscellaneous continuous state variables z
.
Only used for integrators that support error estimation. Returns a VectorBlock to make the values mutable without permitting changing the size of the vector. Requires reinitializing the integrator after calling this method. If Initialize() is not called afterward, an exception will be thrown when attempting to call StepOnceAtMost(). If the caller sets one of the entries to a negative value, an exception will be thrown when the integrator is initialized.

inline 
Returns the number of ODE function evaluations (calls to CalcTimeDerivatives()) since the last call to ResetStatistics() or Initialize().
This count includes all such calls including (1) those necessary to compute Jacobian matrices; (2) those used in rejected integrated steps (for, e.g., purposes of error control); (3) those used strictly for integrator error estimation; and (4) calls that exhibit little cost (due to results being cached).

inline 
Gets the number of step size shrinkages due to failure to meet targeted error tolerances, since the last call to ResetStatistics or Initialize().

inline 
Gets the number of step size shrinkages due to substep failures (e.g., integrator convergence failures) since the last call to ResetStatistics() or Initialize().

inline 
The number of integration steps taken since the last Initialize() or ResetStatistics() call.

inline 
Gets the number of failed substeps (implying one or more step size reductions was required to permit solving the necessary nonlinear system of equations).

inline 
Gets the size of the last (previous) integration step.
If no integration steps have been taken, value will be NaN.

inline 
Gets the requested minimum step size h_min
for this integrator.

inline 
The size of the smallest step taken as the result of a controlled integration step adjustment since the last Initialize() or ResetStatistics() call.
This value will be NaN for integrators without error estimation.

inline 
Gets the stretch factor (> 1), which is multiplied by the maximum (typically userdesignated) integration step size to obtain the amount that the integrator is able to stretch the maximum time step toward hitting an upcoming publish or update event in IntegrateAtMost().
Gets a constant reference to the system that is being integrated (and was provided to the constructor of the integrator).

inline 
Gets the target accuracy.

inline 
Reports the current setting of the throw_on_minimum_step_size_violation flag.

inline 
Gets the current value of the working minimum step size h_work(t)
for this integrator, which may vary with the current time t as stored in the integrator's context.
See this section for more detail.

inline 
An integrator must be initialized before being used.
The pointer to the context must be set before Initialize() is called (or an std::logic_error will be thrown). If Initialize() is not called, an exception will be thrown when attempting to call StepOnceAtMost(). To reinitialize the integrator, Reset() should be called followed by Initialize().
std::logic_error  If the context has not been set or a userset parameter has been set illogically (i.e., one of the weighting matrix coefficients is set to a negative value this check is only performed for integrators that support error estimation; the maximum step size is smaller than the minimum step size; the requested initial step size is outside of the interval [minimum step size, maximum step size]). 

inlineprotected 
Generic code for validating (and resetting, if need be) the integrator working accuracy for error controlled integrators.
This method is intended to be called from an integrator's DoInitialize() method.
default_accuracy  a reasonable default accuracy setting for this integrator. 
loosest_accuracy  the loosest accuracy that this integrator should support. 
max_step_fraction  a fraction of the maximum step size to use when setting the integrator accuracy and the user has not specified accuracy directly. 
std::logic_error  if neither the initial step size target nor the maximum step size has been set. 
IntegratorBase< T >::StepResult IntegrateAtMost  (  const T &  publish_dt, 
const T &  update_dt,  
const T &  boundary_dt  
) 
Integrates the system forward in time by a single step with step size subject to integration error tolerances (assuming that the integrator supports error estimation).
The integrator must already have been initialized or an exception will be thrown. The context will be integrated forward by an amount that will never exceed the minimum of publish_dt
, update_dt
, and 1.01 * get_maximum_step_size()
.
publish_dt  The step size, >= 0.0 (exception will be thrown if this is not the case) at which the next publish will occur. 
update_dt  The step size, >= 0.0 (exception will be thrown if this is not the case) at which the next update will occur. 
boundary_dt  The step size, >= 0.0 (exception will be thrown if this is not the case) marking the end of the userdesignated simulated interval. 
std::logic_error  If the integrator has not been initialized or one of publish_dt, update_dt, or boundary_dt is negative. 
Simulator::StepTo()
. In other circumstances, users will typically call IntegratorBase::IntegrateWithMultipleSteps()
.This method at a glance:

inline 
Stepping function for integrators operating outside of Simulator that advances the continuous state exactly by dt
.
This method is designed for integrator users that do not wish to consider publishing or discontinuous, midinterval updates. This method will step the integrator multiple times, as necessary, to attain requested error tolerances and to ensure the integrator converges.
Simulator::StepTo()
in place of this function (which was created for offsimulation purposes), generally. dt  The nonnegative integration step to take. 
std::logic_error  If the integrator has not been initialized or dt is negative. 
dt
in a single step.This method at a glance:
dt

inline 
Stepping function for integrators operating outside of Simulator that advances the continuous state exactly by dt
and using a single fixed step.
This method is designed for integrator users that do not wish to consider publishing or discontinuous, midinterval updates. One such example application is that of direct transcription for trajectory optimization, for which the integration process should be consistent: it should execute the same sequence of arithmetic operations for all values of the nonlinear programming variables. In keeping with the naming semantics of this function, error controlled integration is not supported (though error estimates will be computed for integrators that support that feature), which is a minimal requirement for "consistency".
Simulator::StepTo()
in place of this function (which was created for offsimulation purposes), generally. dt  The nonnegative integration step to take. 
std::logic_error  If the integrator has not been initialized or dt is negative or if the integrator is not operating in fixed step mode. 
std::runtime_error  If the integrator was unable to take a step of the requested size. 
dt
.This method at a glance:

inline 
Indicates whether the integrator has been initialized.

delete 

delete 

inline 
Request that the first attempted integration step have a particular size.
If no request is made, the integrator will estimate a suitable size for the initial step attempt. If the integrator does not support error control, this method will throw a std::logic_error (call supports_error_estimation() to verify before calling this method). For variablestep integration, the initial target will be treated as a maximum step size subject to accuracy requirements and event occurrences. You can find out what size actually worked with get_actual_initial_step_size_taken()
.
std::logic_error  If the integrator does not support error estimation. 

inline 
Resets the integrator to initial values, i.e., default construction values.
Replace the pointer to the internallymaintained Context with a different one.
This is useful for supplying a new set of initial conditions or wiping out the current context (by passing in a null pointer). You should invoke Initialize() after replacing the Context unless the context is null.
context  The pointer to the new context or nullptr to wipe out the current context without replacing it with another. 

inline 
Forget accumulated statistics.
These are reset to the values they have post construction or immediately after Initialize()
.

inlineprotected 
Sets the working ("in use") accuracy for this integrator.
The working accuracy may not be equivalent to the target accuracy when the latter is too loose or tight for an integrator's capabilities.

inlineprotected 

inline 
Sets an integrator with error control to fixed step mode.
If the integrator runs in fixed step mode, it will always take the maximum step size directed (which may be that determined by get_maximum_step_size(), or may be smaller, as directed by, e.g., Simulator for event handling purposes).
std::logic_error  if integrator does not support error estimation and flag is set to false . 

inlineprotected 

inlineprotected 

inline 
Sets the maximum step size that may be taken by this integrator.
The integrator may stretch the maximum step size by as much as 1% to reach a discrete event. For fixed step integrators, all steps will be taken at the maximum step size unless an event would be missed.

inline 
Sets the requested minimum step size h_min
that may be taken by this integrator.
No step smaller than this will be taken except under circumstances as described above. This setting will be ignored if it is smaller than the absolute minimum h_floor
also described above. Default value is zero.
min_step_size  a nonnegative value. Setting this value to zero will cause the integrator to use a reasonable value instead (see get_working_minimum_step_size()). 

inlineprotected 
Sets the size of the smalleststeptaken statistic as the result of a controlled integration step adjustment.

inline 
Request that the integrator attempt to achieve a particular accuracy for the continuous portions of the simulation.
Otherwise a default accuracy is chosen for you. This may be ignored for fixedstep integration since accuracy control requires variable step sizes. You should call supports_error_estimation() to ensure that the integrator supports this capability before calling this function; if the integrator does not support it, this method will throw an exception.
Integrators vary in the range of accuracy (loosest to tightest) that they can support. If you request accuracy outside the supported range for the chosen integrator it will be quietly adjusted to be in range. You can find out the accuracy setting actually being used using get_accuracy_in_use()
.
The precise meaning of accuracy is a complicated discussion, but translates roughly to the number of significant digits you want in the results. By convention it is supplied as 10^digits
, meaning that an accuracy of 1e3 provides about three significant digits. For more information, see [Sherman 2011].
std::logic_error  if integrator does not support error estimation. 

inline 
Sets whether the integrator should throw a std::runtime_error exception when the integrator's step size selection algorithm determines that it must take a step smaller than the minimum step size (for, e.g., purposes of error control).
Default is true
. If false
, the integrator will advance time and state using the minimum specified step size in such situations. See this section for more detail.

inline 
Starts dense integration, allocating a new dense output for this integrator to use.
std::logic_error  if any of the preconditions is not met. 
Default code for advancing the continuous state of the system by a single step of dt_max
(or smaller, depending on error control).
This particular function is designed to be called directly by an error estimating integrator's DoStep() method to effect errorcontrolled integration. The integrator can effect error controlled integration without calling this method, if the implementer so chooses, but this default method is expected to function well in most circumstances.
[in]  dt_max  The maximum step size to be taken. The integrator may take a smaller step than specified to satisfy accuracy requirements, to resolve integrator convergence problems, or to respect the integrator's maximum step size. 
std::logic_error  if integrator does not support error estimation. 
true
if the full step of size dt_max
is taken and false
otherwise (i.e., a smaller step than dt_max
was taken).

inline 
Stops dense integration, yielding ownership of the current dense output to the caller.
std::logic_error  if any of the preconditions is not met. 

pure virtual 
Indicates whether an integrator supports error estimation.
Without error estimation, target accuracy will be unused.
Implemented in ImplicitEulerIntegrator< T >, SemiExplicitEulerIntegrator< T >, RungeKutta3Integrator< T >, ExplicitEulerIntegrator< T >, and RungeKutta2Integrator< T >.