CHANGELOG.md

v3.8.0

Even more amazing level of integration with ModelingToolkit.jl!

• Now it is possible to partially set specific state variables by providing a dictionary or vector of pairs that map indices to variables in set_state! or reinit!.
• Partial setting of states via providing a dictionary also works in reinit!.
• Partial setting of states also works when initializing a ParallelDynamicalSystem by providing a vector of dictionaries or vector of pairs.
• It was already possible to provide a dictionary of parameter indices to values in set_parameters!, now it was made more clear in the docstring.
• bugfix where reinit! or set_state! of continuous time out of place parallel dynamical systems was not working correctly.
• Dynamical systems are now broadacastable, e.g., you can do observe_state.(ds, (a, b, c)).
• Fixed a bug where set_state!(ds, val, index) only worked for in-place systems.

v3.7.0

Symbolic indices used in observe_state can now also be used in trajectory.

It is possible to specify time to observe a state at now.

v3.6.0

New convenience functions state_name, parameter_name for obtaining a name::String that matches the indexed state/parameter.

v3.5.0

Dynamical systems that have been constructed from DEProblems that themselves have been constructed from ModelingToolkit.jl keep a reference to the symbolic model and all symbolic variables. Accessing a DynamicalSystem using symbolic variables is possible via the functions observe_state, set_state!, current_parameter and set_parameter!. The referenced MTK model corresponding to the dynamical system can be obtained with model = referrenced_sciml_model(ds::DynamicalSystem).

See also the online overarching tutorial for an example.

At the moment this is only possible for CoupledODEs and its derivative systems, as these are the only systems that can be made by a DEProblem, however it should be easy to allow e.g., DeterministicIteratedMap to be created by a DiscreteProblem.

Also:

• Crucial bugfix for successful_step that was not working properly for discrete time systems.

v3.4.0

Better error handling for diffeq and CoupledODEs: passing a keyword was possible but should not have been.

v3.3.0

Sorry, I forgot what goes here!

v3.2.0

The keyword Dt can now be used instead of Δt in trajectory if access to unicode isn't available.

v3.1.0

New error checking functionality added via successful_step. Returns true if the last step! call to ds was successful, false otherwise. For continuous time systems this uses DifferentialEquations.jl error checking,for discrete time it checks if any variable is Inf or NaN. Defaults to true for other types.

v3.0

Complete rewrite of the package. The DynamicalSystems.jl v3 changelog summarizes the highlights. Here we will list all changes to this specific package.

Majorly Breaking

• The Discrete/ContinuousDynamicalSystem constructors no longer accept a Jacobian. Use the dedicated TangentDynamicalSystem for something that represents the tangent space and can be given to downstream functions such as lyapunovspectrum. As a result, none of the predefined systems come with a hand coded Jacobian. The function is still available for manual use nevertheless.
• The keyword diffeq does not exist anymore and is not given to any downstream functions such as lyapunovspectrum. The only struct that cares about DifferentialEquations.jl arguments is CoupledODEs so it is the only one that accepts diffeq keyword.
• trajectory now returns the actual trajectory and the time vector: X, t = trajectory(ds, ...).
• There is no longer a special type for 1D discrete time dynamical systems. You will have to use an 1-element SVector for them.
• The Systems submodule has been completely removed. Now a package PredefinedDynamicalSystems.jl exists. It is untested and it is not recommended to use in any other setting other a demonstrational example. The DynamicalSystems module exports Systems as an alias to PredefinedDynamicalSystems, so that should not be breaking if you were using DynamicalSystems.
• The default ODE solver is now Tsit5 from OrdinaryDiffEq.jl. SimpleDiffEq.jl is no longer used. It is almost certain that the evolution of all chaotic systems will yield ever-so-slightly different trajectories due to the different default integrator.

Enhancements

• DynamicalSystem now defines a proper, well-thought-out interface that is implemented from all its concrete subtypes. See its docstring for details.
• All DynamicalSystem implementations are now "integrators": mutable objects that are stepped with step!. They need to be deepcopied for parallelizing via threads.
• ParallelDynamicalSystem now works for any kind of dynamical system.
• trajectory works for any conceivable thing that extends the DynamicalSystem interface
• New struct ArbitrarySteppable for linking DynamicalSystems.jl with other libraries like Agents.jl.

Deprecations

• DiscreteDynamicalSystem -> DeterministicIteratedMap
• ContinuousDynamicalSystem -> CoupledODEs
• parallel_integrator -> ParallelDynamicalSystem
• tangent_integrator -> TangentDynamicalSystem
• stroboscopicmap -> StroboscopicMap
• poincaremap -> PoincareMap

v2.9

• All code related to the poincare map integrator have been moved here from ChaosTools.jl.

v2.8

• New functions get_parameters and get_parameter that simplify interacting with the parameters of a dynamical system (and drastically simplify source code of set_parameter!).

v2.7.2

• Integer Δt is now enforced in trajectory calculation of discrete systems. Not doing so led to silently incorrect behavior that the user wouldn't expect.
• Fixed set_parameter! not working for projected_integrator.

v2.7

• New function get_states that returns an iterator over states contained in a parallel_integrator.

v2.6

• Added new GeneralizedDynamicalSystem abstract type that is an umbrella term.

v2.5

• Added new current_time(integ) function.
• Added trajectory for projected_integrator and stroboscopicmap.
• Source code of trajectory has been expended and generalized. It should now work out of the box for any object that defines the integrator API, as defined in the DynamicalSystems.jl documentation.
• Added new boolean function isdiscretetime that works for all systems/integrators.

v2.4

• Added new type of integrator: stroboscopicmap.
• Added new type of integrator: projected_integrator.

v2.3

• Added several new famous systems.

v2.2

• get_state(parallel_integrator, k) now returns a view in case the integrator is in-place.

v2.1.1

• Fix set_state! not working for in-place parallel integrator

v2.1.0

• Now set_parameter! can also work with parameter containers that are composite types.
• Added some new pre-defined systems used in synchronization studies, like the Kuramoto model or coupled Roessler oscillators or nonlinearly coupled logistic maps.
• The integrator SimpleTsit5 is also exported. This integrator is non-adaptive, while the default integrator of the library SimpleATsit5 is adaptive.

v2.0.0

• The keyword dt of trajectory has been renamed to Δt. This keyword had conflicts with the options of DifferentialEquations.jl. No warning can be thrown for this change, and users still using dt will have it silently propagated as keyword to the diffeq solvers.

v1.8.11

• More performant version of set_deviations!(integ, Q) when Q is the result of a QR-decomposition.

v1.8.3

• Added grebogi_map dynamical system.

v1.8

• trajectory now supports keyword save_idxs.

v1.7

• New system: Thomas cyclically symmetric

v1.6.1

• Arbitrary keywords can be propagated into trajectory for discrete systems (which are unused, this is simply for unifying syntax with continuous systems)

v1.6.0

• Bugfix of ODEProblem(ContinuousDynamicalSystem) where the parameters of the system where not passed properly.
• Move to Julia 1.5.
• Allow ODEProblem(cds) to take a different state optionally

v1.5.4

• Critical bugfix/mentioning of how 1-D systems work.

v1.5

• Added function that provides initial conditions for henon-heiles system at given energy, Systems.henonheiles_ics(E, n)
• More famous systems and a lot more Jacobians to existing systems (#93)

v1.4

• added Pomaeu-Manneville map into the famous systems.

v1.3

• The specialized integrators (tangent & parallel) now implement an internal norm that only evaluates norm of the main state, instead of using the other parallel states or deviation vectors. (#86)
• Bunch of bugfixes and performance improvements (see git history)

v1.2

In version 1.2 we have moved to SciMLBase 5.0+. In addition to that we have changed the default integrator to SimpleATsit5 from module SimpleDiffEq. This is not a breaking change and you can use any of the previous integrators. It must be stated though that numeric results you obtained using the default integrator will now be slightly different.

In addition, be sure that you have version SimpleDiffEq 0.3.0 or greater (which is what DynamicalSystemsBase 1.2.3 guarantees).

v1.1

• All functionality related to neighborhood, reconstruct and StateSpaceSet has moved to a new package: DelayEmbeddings. It is reexported by DynamicalSystemsBase, which is now (as the name suggests) the basis package for defining DynamicalSystem.

v1.0

• First major release (long term stability).

v0.12.0

New Features

• orthonormal is 100x more performant.
• Super duper cool new printing for DynamicalSystem.
• Added tests etc. and now we are sure we can support Sparse matrix jacobians.
• Testing of specialized integrators with DiffEqCallbacks.
• Parameters are printed in the DynamicalSystem.

0.11.0

Breaking

• Dropped support of Julia versions < 0.7

• Reconstruction does not exist anymore. Instead, reconstruct is used in it's place. The function now also always returns a StateSpaceSet.

• In the reconstruct function, D now stands for the number of temporal neighbors which is one less than the dimensionality of the reconstructed space.

  • This change was done because it is more intuitive and more scalable when considering multiple timeseries or spatio temporal timeseries.

• Re-worked the internals of DynamicalSystem (although the API remained the same): Now DynamicalSystem only stores what is absolutely necessary and creates directly integrators when need be. A "problem" is not stored anymore. This also lead to re-working of how keyword arguments are handled. I am very happy to say that these changes reduced tremendously the source code!

New Features

• Exported a 3-argument reinit!(integ, u0::AbstractVector, Q0::AbstractMatrix) that takes Q0 as the third argument and reinits safely any integrator.

• reconstruct creates internally a subtype of AbstractEmbedding. These objects can be used as functors to create the i-th reconstructed vector on demand. This also improved the source code clarity significantly.

• tangent_ and parallel_integrator can now accept callbacks for continuous systems. In general you could pass to the constructors any keyword acceptable by init of DiffEq.

v0.10

some of the following changes are breaking

• state function no longer exists and has been merged into get_state.
• Created specialized tangent integrator for discrete systems, which is about 20% faster. This is a "breaking" change.
• Created functions get_state, get_deviations, set_state! and set_deviations! that always correctly return either the system state or the deviation vectors (in a form of a matrix) from any integrator!!!
• Dynamical Systems with matrices as states are no longer available.
• Added note in reconstruction about how many temporal neighbors there are.

v0.9

• Theiler window is now part of the neighborhood function
• Methods that estimate parameters for Reconstruction have moved back to ChaosTools.

v0.8

New features

• Multi-dimensional Reconstruction
• Multi-time Reconstruction
• Multi-dimensional, multi-time Reconstruction
• Methods for estimating Reconstruction parameters are now here.
• Reconstruction from StateSpaceSet

Breaking

• Corrected name helies -> heiles
• Reconstructions now have field delay which is the delay time/times and are instead parameterized on the type of the delay.

v0.6

Massively Breaking

• Complete overhaul of all of DynamicalSystems. More details are in the official documentation (because I would had to write 10 pages of changelog otherwise). The constructors also changed signature.
• All DynamicalSystems are now immutable and contain a problem, a jacobian and a jacobian matrix.
• Advantage of Multiple dispatch is taken to distinguish between in-place, out of place, continuous, discrete, heaven, hell, whatever.
• All internals use integrator and step!.
• variational_integrator renamed to tangent_integrator.

Non-breaking

• Massive performance boost of up to 8x in system evolution of continuous systems.
• increased the interaction between DiffEq and DynamicalSystems.
• A dedicated integrator for discrete systems that is super fast is now employed by default. It follows the same high level DiffEq interface but the implementation is totally internal.
• Out-of-place continuous systems are allowed. All famous systems are actually out of place now.
• parallel_integrator added.

v0.5.0

Massively Breaking

• All system type definitions have changed. See the new documentation strings or the new docs! We are now having the syntax eom!(du, u, p, t) and the parameters are passed directly into the function!!!

Non-breaking

• Improved the algorithm that converts a StateSpaceSet to a Matrix. It is now not only faster, but also more clear!

v0.4.1

• (Breaking?) Moved estimate_delay to ChaosTools.jl
• Added gissinger system.
• Added columns function for columns of dataset.

v0.4.0

Breaking

• By default now all ContinuousDynamicalSystems are solved using Vern9 as a solver and tolerances of 1e-9, both abstol and reltol.

Non-breaking

• Added energy conservation option to the Henon Heiles system
• All ContinuousDS evolution now internally passes thrgouh the get_sol function, which improves the clarity and stability of the ecosystem greatly!!!
• Improved stability in propagating solve keywords.
• get_sol now returns solution and time vector for generality purposes.
• get_sol is now also exported.
• Internal ODEProblem constructor correctly merges different callbacks.

v0.3.3

Non-breaking

• Bugfix of eltype of Reconstruction.
• Added circlemap to Systems.
• Bugfix on StateSpaceSet that incorrect methods were being called due to missing <:.
• Method Base.getindex(d::AbstractStateSpaceSet{D,T}, ::Colon, j<:AbstractVector{Int}) now exists!
• Added function evolve!.
• Clearly state that we do not support matrices in the equations of motion.
• Bugfixes regarding ODEProblem.

v0.3.2

Non-breaking

• Orders of magnitude speed-up in conversions between Matrix and StateSpaceSet, because now methods use transpose internally and only reinterpret.

v0.3.1

• Added jacobian function
• Removed EomVector nonsense.
• Now trajectory correctly gives equi-spaced points when the ODEProblem has "special" callbacks.

v0.3.0

BREAKING

• The type ContinuousDS has been completely overhauled.
  • Now the type contains an ODEProblem instead of state and eom!.
  • Now the form of equations of motion is expected as (t, u, du).
  • We are able to handle non-autonomous systems with "simple" time dependence.
  • This also allows to use callbacks natively tied to the system! This means that you can create the simple form of Hybrid systems!

v0.2.1

Non-breaking

• Added function minmaxima that calculates maxima and minima at the same time for datasets.

v0.2.0

Non-breaking

• Minor docstring improvements and IO improvements.
• Datasets can now be accessed with ranges as well.
• Added the Duffing oscillator in the predefined systems.
• Changed 🐛 where get_solver would remove the entry.
• Added the possibility to provide a state in all evolving functions.
• Added Shinriki oscillator (chaotic 3D flow with period doubling).

v0.1.0

Initial release, see: https://juliadynamics.github.io/DynamicalSystems.jl/v0.8.0/ for the features up to this point.