News.md

BifurcationKit.jl, Changelog

All notable changes to this project will be documented in this file (hopefully). No performance improvements will be notified but mainly the addition of new methods, the modifications of internal structs, etc.

[Unreleased]

Breaking changes

• get_adjoint_basis is replaced by the methods _get_kernel_basis_1d_from_eigensolver / _get_kernel_basis_1d_from_bls (resp. the Nd helpers), which now compute the right and left (adjoint) kernel vectors together
• __compute_bordered_vectors_fold / __compute_bordered_vectors_hopf are now also used to build the kernel basis of the normal forms (start_with_eigen = Val(false))

Added

• add ∫_gauss to integrate arrays sampled at the Gauss points, i.e. in the row layout of the collocation operator, consistently with ∫
• add the keyword start_with_eigen = Val(true) to get_normal_form1d, get_normal_formNd and the codim 2 normal forms (hopf, cusp, bogdanov_takens, bautin, zero_hopf, hopf_hopf): with start_with_eigen = Val(false), the (right/left) kernel basis is built with a bordered linear system (keywords bls, bls_adjoint, bls_block) instead of the eigensolver, which is more robust for large-scale problems. The 1d basis is computed with the bordered-vector solvers __compute_bordered_vectors_fold / __compute_bordered_vectors_hopf and the Nd basis with __compute_nd_basis_from_bls. bautin_normal_form also accepts bls/bls_adjoint like hopf_normal_form
• compute the parameter derivative dFdp in the Moore-Penrose algorithm (newton_moore_penrose) with the (user provided) R01/R01! instead of first-order finite differences, and accumulate the number of linear iterations (itlineartot)

Fixed

• evaluate all the integrands in neimark_sacker_normal_form_iooss at the Gauss points and build the RHS of the homological equations h20/h11 in the row layout of the collocation operator. This removes the former Icoll mass-matrix pre-scaling and the ad-hoc h20 ./= 2Ntst / h11 ./= 2Ntst renormalizations.
• evaluate all the integrands in period_doubling_normal_form_iooss at the Gauss points and build the RHS of the homological equations h₂/h₀₁ in the row layout of the collocation operator, normalizing ψ₁★ with ∫_gauss so that <ψ₁★, F(u₀)> = 1/2. This removes the former Icoll mass-matrix pre-scaling and fixes the scaling of the normal form coefficients.

[0.8.3]

Breaking changes

• detect_codim2_parameters is renamed modify_contparams_for_codim2
• the method _compute_bordered_vectors of the Fold/Hopf/PD/NS MA formulations is refactored into the methods __compute_bordered_vectors_fold / __compute_bordered_vectors_hopf

Added

• add the bordered-vector kernels __compute_bordered_vectors_fold / __compute_bordered_vectors_hopf which compute the right/left null vectors of the Fold / Hopf jacobians by a bordered linear system (factoring the Fold/Hopf initial eigenvector computation). These methods are also used to build the kernel basis of the normal forms with start_with_eigen = Val(false)
• allow plotting during NS (Neimark-Sacker) continuation
• Plots backend: plot several codim 1 branches in one figure
• improve codim 2 continuation of periodic orbits (PD/NS Jacobian borders, events on Fold of PO, simplified get_normal_form for codim 2 points of PO)

Fixed

• correct update! for NSMAProblem / PDMAProblem
• robust mesh adaptation for Collocation (improved predictor)
• correct the Hopf predictor at Hopf-Hopf points

[0.8.2]

• filter infinite eigenvalues in generalized eigenvalue computations (DefaultEig)

[0.8.1]

Breaking changes

• 🚦🚦🚦 update!(prob, x) becomes restore_problem!(prob, x, pars)
• getlinsolver/getbls become get_bordered_linsolver
• OneParamCont, TwoParamCont, TwoParamPeriodicOrbitCont become AbstractOneParamCont, AbstractTwoParamCont, AbstractTwoParamPeriodicOrbitCont
• _getsolution becomes saved_solution
• tangent field in MoorePenrose becomes predictor
• remove autodiff keyword argument for computing normal forms
• FlowDE construction is now keyword-based
• merge BT nfsupp into nf

Added

• add derivatives of the flow w.r.t. the parameter and higher-order differentials R01, R11, R20, R30 to Flow/FlowDE, computed with ForwardDiff by default and usable in the Poincaré return map and the normal forms
• use monodromy_matrix! in Shooting and Poincaré Shooting
• rework the Poincaré return map around _evolve_flow_prm; add R01, R11, R20, R30 methods for PoincaréMap and collocation
• rework the Jet traits: R01/R02/R11 are now computed by dispatch on AutoDiff()/FiniteDifferences() and generalized to any AbstractBifurcationProblem
• save/restore of the Poincaré section for PoincareShooting (POSavedSolutionAndState_PSH, BVPSavedSolutionAndState_PSH), fixing issue #334
• add Accumulator to DeflationOperator
• add eigenvalue solvers EigenWave and GEigenWave for waves
• add __sort_spectrum and correct the spectrum sorting of DefaultEig
• add mechanism to switch the plot backend: set_plot_backend!, get_plot_backend with BK_Plots(), BK_Makie(), BK_NoPlot()
• Makie: add plot_stability_segments
• Plots: add :dots style for the unstable part of the branch
• add defaults to ContState fields and EmptyContState
• refactor the generalized eigenvalue computation into GeneralizedEigenSolver.jl and the tangents into Tangents.jl
• use an out-of-place formulation for waves
• add ∫_gauss to integrate arrays sampled at the Gauss points, i.e. in the row layout of the collocation operator, consistently with ∫
• add the keyword start_with_eigen = Val(true) to get_normal_form1d, get_normal_formNd and the codim 2 normal forms (hopf, cusp, bogdanov_takens, bautin, zero_hopf, hopf_hopf): with start_with_eigen = Val(false), the (right/left) kernel basis is built with a bordered linear system (keywords bls, bls_adjoint, bls_block) instead of the eigensolver, which is more robust for large-scale problems. bautin_normal_form also accepts bls/bls_adjoint like hopf_normal_form
• compute the parameter derivative dFdp in the Moore-Penrose algorithm (newton_moore_penrose) with the (user provided) R01/R01! instead of first-order finite differences, and accumulate the number of linear iterations (itlineartot)

Fixed

• evaluate all the integrands of the NS Iooss normal form (neimark_sacker_normal_form_iooss) at the Gauss points and build the RHS of the homological equations h20/h11 in the row layout of the collocation operator. This removes the former Icoll mass-matrix pre-scaling and the ad-hoc h20 ./= 2Ntst / h11 ./= 2Ntst renormalizations.

[0.8.0]

• add a new BVP interface: BVPModel, PeriodicOrbitModel, DiscretizedBVP, discretize, generate_solution, the discretizers Shooting, Trapeze, Collocation and the problem BVPBifProblem
• add mesh adaptation for BVP problems
• add user-specified time interval for BVP problems
• add BoundaryValueProblemCont for the continuation of BVP problems
• add wrapper for deflated continuation of BVP problems
• add TimeMesh structure to support non-uniform meshes for Trapeze
• SolPeriodicOrbit becomes BVPSolution, POSolution becomes POInterpolation, POSolutionAndState becomes POSavedSolutionAndState
• rename Trap into Trapeze in the BVP interface
• jacobian = :auto becomes jacobian = AutoDiffDense() in BVP Shooting
• add normal_form alias for get_normal_form

[0.7.3]

• Remove AbstractPeriodicOrbitDiscretization, make subtypes inherit directly from AbstractBoundaryValueDiscretization
• AbstractPODifferentialDiscretization → AbstractDifferentialDiscretization
• AbstractPOFiniteDifferencesDiscretization → AbstractFiniteDifferencesDiscretization
• AbstractPOShootingDiscretization → AbstractShootingDiscretization

[0.7.2]

• Change names *ShootingProblem for *Shooting, PeriodicOrbitOCollProblem for Collocation and PeriodicOrbitTrapProblem for Trapeze

[0.6.1]

• emove fields params and lens from PeriodicOrbitFunctionalTrap, PeriodicOrbitFunctionalSh, PeriodicOrbitFunctionalColl, WrapTW
• remove param field from *MAProblem structs (FoldMAProblem, HopfMAProblem, PDMAProblem, NSMAProblem, BTMAProblem) and from WrapTW; getparams now delegates to getparams(get_formulation(prob))
• add re_make methods for AbstractMABifurcationProblem, AbstractWaveProblem, AbstractWrapperPeriodicOrbitProblem, AbstractPODifferentialDiscretization, AbstractPOShootingDiscretization, TWModel, AbstractMinimallyAugmentedFormulation
• add getlens method for TWModel

[0.6.0]

• add AbstractBoundaryValueDiscretization, AbstractPeriodicOrbitDiscretization, AbstractPODifferentialDiscretization, AbstractPOFiniteDifferencesDiscretization
• AbstractPoincareShootingProblem becomes AbstractPoincareShootingDiscretization
• AbstractShootingProblem becomes AbstractPOShootingDiscretization
• BTProblemMinimallyAugmented becomes BTMinimallyAugmentedFormulation
• TWProblem becomes TWModel
• remove modify_po_finalise, modify_po_record, modify_tw_record and Finalizer
• change Fold/Hopf/PeriodDoubling/NeimarkSackerProblemMinimallyAugmented into Fold/Hopf/PeriodDoubling/NeimarkSackerMinimallyAugmentedFormulation
• AbstractProblemMinimallyAugmented becomes AbstractMinimallyAugmentedFormulation
• make AbstractMABifurcationProblem{T, Tjac} dependent on 2 parameters
• add method finalise_solution(iter::ContIterable, state::AbstractContinuationState, contRes)
• remove update_minaug_hopf, update_minaug_fold, update_min_aug_ns, update_min_aug_pd
• WrapPOColl becomes PeriodicOrbitFunctionalColl
• WrapPOSh becomes PeriodicOrbitFunctionalSh
• WrapPOTrap becomes PeriodicOrbitFunctionalTrap
• correct_bifurcation becomes _correct_event_labels
• add PeriodicOrbit{Tdisc}, TravellingWave{Tdisc} to bridge discretization and problem
• add MASolution, MASolutionFreq to hold codim1 MA solutions (for mesh adaptation in codim2)
• add RecordForFold, RecordForHopf, RecordForPeriodicOrbits, RecordForNS, RecordForPD, RecordForTW to replace closure-based callbacks
• add accessor methods get_discretization, get_formulation, get_solution, getparams
• add update! for PDMAProblem, NSMAProblem, HopfMAProblem and periodic orbit wrappers
• TWModel inherits from AbstractTravelingWaveDiscretization instead of AbstractBifurcationProblem
• change signature of continuation for periodic orbits: probPO::AbstractPeriodicOrbitProblem becomes disc::AbstractPeriodicOrbitDiscretization
• DefaultLS uses VI.Zero() / VI.One() instead of literal 0 / 1
• closure removal for record_from_solution in all continuation types (codim1 and codim2, periodic orbits, waves)
• add mesh adaptation in codim2 continuation
• change initialization of delta in BifurcationProblem to use getdelta
• simplify is_event_crossed and refactor events
• remove dead code in _continuation(gh::Bautin)
• various bug fixes: correction of jacobian selection in PD continuation, fix plotting in NS continuation, correct PD formulation, correct linear bordered solver solve_bls_block, correct type stability of Hopf predictor, correct type assert in Bautin normal form, do not update MA problem when in bisection

[0.5.8]

• add AbstractBifurcationPointCodim2, NdBranchPoint for bifurcation points with dim(Ker) > 1
• add method minus(x::POSolutionAndState, y::POSolutionAndState) for detect_loop with collocation and mesh adaptation
• add branching to curve of periodic orbits from curve of Hopf points
• set PALC(tangent = Bordered()) as default in autoswitch constructor
• remove type constraint Tlens <: AllOpticTypes in BifurcationPoints
• improve type stability of get_normal_formNd, biorthogonalise, etc.
• refactor BranchSwitching.jl
• simplify multicontinuation to return a Branch instead of Vector{Branch}
• bump compat for RecursiveArrayTools to 4

[0.5.6] (future)

• reorganise tests:
  • each test belongs to a test category (like in previous versions)
  • the test directory arborescence reflects this organisation
  • the runtests.jl file is now generic and runs accepts arguments which define which tests will be run
  • the .github/workflows/ci.yml can now run a subset of tests (using the previous runtests.jl modifications) bases on the labels of PR/commit/..
  • examples of valid labels: Run test(s): wave, Run test(s): wave | newton

[0.5.5]

• most Standard shooting code works with VI.MinimalVec

[0.5.4]

• modify dispatch for get_time_slice and get_time_slices

[0.5.3]

• make BorderedArray comply with VectorInterface.jl (VI)
• most of BK code complies with VI
• bug correction in mesh adaptation for collocation
• massive improvement in computation of reduced equations. Improve type stability as well.

[0.5.2]

• export ODEBifProblem
• add new jacobian types
• jad becomes jacobian_adjoint

[0.5.1]

• _eig_floquet_col is now _eig_floquet_coll
• FloquetCollGEV is now called FloquetGEV
• extract_period is now called _extract_period
• the struct COPCACHE has been changed a lot.

[0.5.0]

• add ShiftInvert, new general composite eigen solver
• add option use_adapted_mesh to generate_ci_problem(pb::PeriodicOrbitOCollProblem
• add option optimal_period to generate_ci_problem(shooting::ShootingProblem
• 🚦🚦🚦 change jacobian parameter in PeriodicOrbitTrapProblem from Symbol to custom type. See ?PeriodicOrbitTrapProblem for more information
• correct bug in Natural which improves AutoSwitch
• remove (some) closures from codim1 continuation
• add structures/normal form for CuspPO, R1, R2, etc

[0.4.16]

• make WrapPOColl a AbstractWrapperFDProblem
• 🚦🚦🚦 change jacobian_ma argument from Symbol to type. See docs for codim 2 continuation. For example, continuation(br ,1; jacobian_ma = :minaug) becomes continuation(br ,1; jacobian_ma = MinAug())
• 🚦🚦🚦 add update! function to BifurcationProblem. This allows to adapt the problem during continuation

[0.4.15]

• solve compile time issue for continuation of Fold of periodic orbits
• move test functions out of continuation function for MinAug
• emove the saving of AD generated BifFunction. Rely on dispatch instead.

[0.4.4]

• change type parameters in AbstractCodim2EigenSolver
• add Krylov.jl as possible linear solver

[0.4.3]

• add jacobian option MinAugMatrixBased to Fold/Hopf continuation
• remove reference to RecursiveVec

[0.4.2]

• change bordered linear solvers' interface
• record_from_solution has been changed to the following definition

[0.4.0]

• Setfield.jl is not anymore the main component of BifurcationKit to support parameter axis. It has been changed in favour of Accessors.jl

[0.3.7]

• remove Requires.jl and use extensions. This requires julia>=1.9

[0.3.5]

• add field save_solution to BifurcationProblem. Allows to save problem state along with the current solution. Useful for periodic orbits for example where we can save the phase condition and the mesh when the latter is adapted.

[0.3.4]

• add function _keep_opts_cont to filter continuation options
• order 2 prediction for periodic orbit from Hopf bifurcation
• add jacobian(Π::PoincaréMap{ <: WrapPOSh }, x, pars)
• hasstability becomes _hasstability
• getvectortype becomes _getvectortype
• add condensation of parameters, great speedup for collocation of periodic orbits
• add in_bisection field in struct ContState
• allow to do codim 2 continuation of PO with collocation and mesh adaptation
• update_section_every_step becomes a UInt
• add fields in PeriodDoublingProblemMinimallyAugmented and NeimarkSackerProblemMinimallyAugmented for holding Resonance test values
• add specific finalizer for Fold of PO when using Collocation or shooting
• add struct FinalisePO
• update_minaug_every_step = 1 by default for Hopf / Fold continuation
• update_minaug_every_step = 1 is default for for PD / NS continuation

[0.2.8] - 2023-05-18

• add getDelta to the interface of AbstractFlow
• remove finDiffEps from ContinuationPar.

[0.2.8] - 2023-04-23

• use jvp function name in Flow interface
• add radius to section of Poincare Shooting
• add _mesh field to reconstruct POColl problem (adapted mesh) from previous solution
• add new jacobian parametrization using structs instead of Symbol
• remove θ from ContinuationPar

[0.2.8] - 2023-04-17

• add delta keyword to BifurcationProblem constructor

MISSSINF

[0.2.0] - 2022-07-23

• new interface based on the problem BifurcationProblem

[0.1.8] - 2021-12-12

• switch from DiffEqBase to SciMLBase
• change function name closesttozero to rightmost

[0.1.8] - 2021-11-27

• ⛳️ add a new interface for Flows
• add custom distance for DeflationOperator
• add possibility to use forward diff (AD) with deflation operator

[0.1.8] - 2021-11-20

• the method for periodic orbits getM becomes getMeshSize

[0.1.7] - 2021-11-6

• add abstract types AbstractDirectLinearSolver and AbstractIterativeLinearSolver
• the function getTrajectory becomes getPeriodicOrbit
• add struct SolPeriodicOrbit to allow for unified plotting interface with all methods for computing periodic orbits
• ⛳️ the keyword argument linearPO is renamed into jacobianPO
• add newton / continuation methods for TWProblem
• add GEigArpack generalized eigensolver

[0.1.5] - 2021-10-23

• remove documentation from package, it is now located in BifurcationKitDocs.jl

[0.1.5] - 2021-10-16

• change function name problemForBS into reMake for aBS of periodic orbits
• add function generateSolution to generate guess for computing orbits from a function solution t -> orbit(t)
• ⛳️ add orthogonal collocation method for periodic orbits
• add additional method for computing Floquet multipliers based on generalized eigenvalue problem

[0.1.5] - 2021-09-26

• add new problem for symmetries TWProblem

[0.1.5] - 2021-09-25

• add example for wave computation

[0.1.5] - 2021-09-25

• refactoring, extractTimeSlices becomes getTimeSlices

[0.1.5] - 2021-09-05

• add a simple callback to limit residuals in Newton iterations cbMaxNorm
• ⛳️ add branch switching for branches of PO at BP / PD
• auto generate more tutorials

[0.1.5] - 2021-07-18

• rename get3Jet into getJet
• remove BlockArrays.setblock! occurrences which are deprecated

[0.1.5] - 2021-07-10

• add perturbGuess option to multicontinuation
• change option printSolution to recordFromSolution in continuation and similar functions

[0.1.5] - 2021-06-26

• add new function getFirstPointsOnBranch to allow fine grained control of aBS
• add full automatic differentiation for Deflated Problems

[0.1.5] - 2021-06-20

• ⛳️ add computing full transcritical/pitchfork branch (not half) in bifurcationDiagram

[0.1.4] - 2021-06-06

• move toward automatic generation of docs with figures
• add applytoX, applytoY option to plotting

[0.1.4] - 2021-05-30

• add function get3Jet to compute Taylor expansion
• getLensParam becomes getLensSymbol
• add detection of codim 2 singularities

[0.0.1] - 2021-05-16

• rename HopfBifPoint -> Hopf
• rename GenericBifPoint into SpecialPoint and bifpoint to specialpoint
• add applytoY keyword to plot recipe

[0.0.1] - 2021-05-9

• remove p->nothing as default argument in continuationHopf
• add bordered linear solver option in newtonHopf

[0.0.1] - 2021-05-2

• remove type piracy for iterate
• put the computation of eigenvalues in the iterator
• correct mistake in bracketing interval in locateBifurcation!
• remove GMRESIterativeSolvers! from linearsolvers

[0.0.1] - 2021-04-3

• correct bug in the interval locating the bifurcation point (in bisection method)

[0.0.1] - 2021-01-24

• ⛳️ add bothside kwargs to continuation to compute a branch on both sides of initial guess
• update the Minimally augmented problem during the continuation. This is helpful otherwise the codim 2 continuation fails.
• [WIP] detection of Bogdanov-Takens and Fold-Hopf bifurcations
• remove field foldpoint from ContResult

[0.0.1] - 2020-11-29

• improve bordered solvers for POTrap based on the cyclic matrix

[0.0.1] - 2020-11-7

• ⛳️ update phase condition during continuation for shooting problems and Trapezoid method

[0.0.1] - 2020-11-7

• remove fields n_unstable, n_imag and stability from ContResult and put it in the field branch.

[0.0.1] - 2020-10-25

• the keyword argument Jt for the jacobian transpose is written Jᵗ

[0.0.1] - 2020-9-18

• new way to use the argument printSolution in continuation. You can return (Named) tuple now.

[0.0.1] - 2020-9-17

• add new type GenericBifPoint to record bifurcation points and also an interval which contains the bifurcation point
• add kwargs to arguments finaliseSolution
• add kwargs to callback from newton. In particular, newton passes fromNewton=true, newtonPALC passes fromNewton = false
• save intervals for the location of bifurcation points in the correct way, (min, max)

[0.0.1] - 2020-9-16

• better estimation of d2f/dpdx in normal form computation
• change of name HyperplaneSections -> SectionPS for Poincare Shooting

[0.0.1] - 2020-9-12

• clamp values in [pMin, pMax] for continuation
• put arrow at the end of the branch (plotting)

[0.0.1] - 2020-9-6

• add eta parameter in ContinuationPar
• change name PALCStateVariables into ContState and PALCIterable into ContIterable
• ⛳️ add Deflated Continuation

[0.0.1] - 2020-8-21

• ⛳️ add Multiple predictor (this is needed to implement the pmcont algorithm from pde2path (Matlab)

[0.0.1] - 2020-7-26

• ⛳️ add Polynomial predictor

[0.0.1] - 2020-7-19

• ⛳️ add Branch switching for non-simple branch points

[0.0.1] - 2020-7-9

The package is registered.

[0.0.1] - 2020-6-20

Deprecated

• Rename option ContinuationPar: saveSolEveryNsteps --> saveSolEveryStep
• Rename option ContinuationPar: saveEigEveryNsteps --> saveEigEveryStep
• Rename option ContinuationPar: plotEveryNsteps --> plotEveryStep

[0.0.1] - 2020-6-10

• change the name of the package into BifurcationKit.jl

Deprecated

• The options computeEigenvalue in ContinuationPar has been removed. It is now controlled with detectBifurcation.

[0.0.1] - 2020-5-2

Added

• ⛳️ automatic branch switching from simple Hopf points
• ⛳️ automatic normal form computation for any kernel dimension

[0.0.1] - 2020-4-27

Added

• ⛳️ automatic branch switching from simple branch points (equilibrium)
• ⛳️ automatic normal form computation