Examples of ptolemy.domains.continuous.kernel.ContinuousDirector

• ptolemy.domains.continuous.kernel.ContinuousDirector
  The continuous time domain is a timed domain that supports continuous-time signals, discrete-event signals, and mixtures of the two. There is a global notion of time that all the actors are aware of. The semantics of this domain is given in: Edward A. Lee and Haiyang Zheng, "Operational Semantics of Hybrid Systems," Invited paper in Proceedings of Hybrid Systems: Computation and Control (HSCC) LNCS 3414, Zurich, Switzerland, March 9-11, 2005.

  A signal is a set of "events," each of which has a tag and value. The set of values includes a special element, called "absent", denoting the absence of a (normal) value. This director uses superdense time, where every event has a tag that is a member of the set RxN. R is a connected subset of the real numbers (giving "time", and approximated by instances of the Time class), and N is the natural numbers (giving an "index"). At a time t, a signal may have multiple values in sequence with tags (t, 0), (t, 1)... Its "initial value" is the value at tag (t, 0). It typically settles to a "final value" after a finite number of indices. If it fails to settle to a final value, the signal is said to have a "stuttering Zeno" condition, and time will not progress.

  In our semantics, all signals are piecewise continuous. This means that the initial value, as a function of time, is continuous on the left, the final value, as a function of time, is continuous on the right, and the signal has exactly one value (meaning the initial value and the final value are the same) at all times except those on a discrete subset D.

  A purely continuous signal has exactly one value at all times, meaning that the final value equals the initial value at all times. A purely discrete signal has initial value "absent" and final value "absent" at all times, and at a discrete subset of the times, it may have non-absent values. The only signal that is both purely continuous and purely discrete is the one that is absent at all tags.

  A signal may be mostly continuous, but has multiple values at a discrete subset of times. These multiple values semantically represent discontinuities in a continuous signal that is not purely continuous.

  The set of times where signals have more than one distinct value are a discrete subset D of the time set. These times are called "breakpoints" and are treated specially in the execution. Between these times, an ordinary differential equation (ODE) solver governs the execution. Initial values are always given by the ODE solver.

  The parameters of this director are:

  • startTime: The start time of the execution. This parameter has no effect if this director is not within a top-level model.
  • stopTime: The stop time of the execution. When the current time reaches this value, postfire() will return false. This will occur whether or not this director is at the top level.
  • initStepSize: The suggested integration step size. If the ODE solver is a fixed step size solver, then this parameter gives the step size taken. Otherwise, at the start of execution, this provides the first guess for the integration step size. In later iterations, the integrators provide the suggested step size. This is a double with default value 0.1
  • maxStepSize: The maximum step size. This can be used to prevent the solver from too few samples of signals. That is, for certain models, it might be possible to get accurate results for very large step sizes, but plots of the signals may be misleading (even if they are accurate) because they represent the signal with only a few samples. The default value is 1.0.
  • maxIterations: The maximum number of iterations that an ODE solver can use to resolve the states of integrators. Implicit solvers, for example, iterate until they converge, and this parameter bounds the number of iterations. An example of an implicit solver is the BackwardsEuler solver. The default value is 20, and the type is int. FIXME: Currently, this package implements no implicit solvers.
  • ODESolver: The class name of the ODE solver used for integration. This is a string that defaults to "ExplicitRK23Solver", a solver that tends to deliver smooth renditions of signals, at the expense of computing more points than the "ExplicitRK45Solver". Solvers are all required to be in package "ptolemy.domains.continuous.kernel.solver". If there is another ContinuousDirector above this one in the hierarchy, then the value of this parameter is ignored and the solver given by the first ContinuousDirector above will be used.
  • errorTolerance: This is the local truncation error tolerance, used for controlling the integration accuracy in variable step size ODE solvers, and also for determining whether unpredictable breakpoints have been accurately identified. Any actor that implements ContinuousStepSizeControl may use this error tolerance to determine whether the current step is accurate. For example, if the local truncation error in some integrator is greater than this tolerance, then the integration step is considered to have failed, and should be restarted with a reduced step size. The default value is 1e-4.

  This director maintains a breakpoint table to record all predictable breakpoints that are greater than or equal to the current time. The breakpoints are sorted in their chronological order. Breakpoints at the same time are considered to be identical, and the breakpoint table does not contain duplicate time points. A breakpoint can be inserted into the table by calling the fireAt() method. The fireAt method may be requested by the director, which inserts the stop time of the execution. The fireAt method may also be requested by actors and the requested firing time will be inserted into the breakpoint table.

  This director is designed to work with any other director that implements the actor semantics. As long as the other director does not commit state changes except in postfire, this director can be used within it, and it can be used within the model controlled by this director. Of course, the enclosing model must advance time, or model time will not progress beyond zero.

  This director is based on the CTDirector by Jie Liu and Haiyang Zheng, but it has a much simpler scheduler. FIXME: the design of clone method should be examined and reimplemented. All Continuous files need this. @author Haiyang Zheng and Edward A. Lee @version $Id: ContinuousDirector.java,v 1.50 2007/12/07 06:27:42 cxh Exp $ @since Ptolemy II 6.0 @Pt.ProposedRating Yellow (hyzheng) @Pt.AcceptedRating Red (hyzheng)

Related Classes of ptolemy.domains.continuous.kernel.ContinuousDirector