BifurcationKit.jl

This Julia package aims at performing automatic bifurcation analysis of possibly large dimensional equations F(u, λ)=0 where λ∈ℝ by taking advantage of iterative methods, dense / sparse formulation and specific hardwares (e.g. GPU).

It incorporates continuation algorithms (PALC, deflated continuation, ...) based on a Newton-Krylov method to correct the predictor step and a Matrix-Free/Dense/Sparse eigensolver is used to compute stability and bifurcation points.

Despite initial focus on large scale problems, the package can easily handle low dimensional problems.

By leveraging on the above method, the package can also seek for periodic orbits of Cauchy problems by casting them into an equation $F(u,p)=0$ of high dimension. It is by now, one of the only softwares which provides shooting methods AND methods based on finite differences or collocation to compute periodic orbits.

The current package focuses on large scale nonlinear problems and multiple hardwares. Hence, the goal is to use Matrix Free methods on GPU (see PDE example and Periodic orbit example) or on a cluster to solve non linear PDE, nonlocal problems, compute sub-manifolds...

One design choice is that we try not to require u to be a subtype of an AbstractArray as this would forbid the use of spectral methods like the one from ApproxFun.jl. For now, our implementation does not allow this for Hopf continuation and computation of periodic orbits.

Installation

This package requires Julia >= v1.3.0 because of the use of methods added to abstract types (see #31916).

To install it, please run

] add BifurcationKit

To install the bleeding edge version, please run

] add BifurcationKit#master

Citing this work

If you use this package for your work, we ask that you cite the following paper!! Open source development strongly depends on this. It is referenced on HAL-Inria as follows:

@misc{veltz:hal-02902346,
  TITLE = {{BifurcationKit.jl}},
  AUTHOR = {Veltz, Romain},
  URL = {https://hal.archives-ouvertes.fr/hal-02902346},
  INSTITUTION = {{Inria Sophia-Antipolis}},
  YEAR = {2020},
  MONTH = Jul,
  KEYWORDS = {pseudo-arclength-continuation ; periodic-orbits ; floquet ; gpu ; bifurcation-diagram ; deflation ; newton-krylov},
  PDF = {https://hal.archives-ouvertes.fr/hal-02902346/file/354c9fb0d148262405609eed2cb7927818706f1f.tar.gz},
  HAL_ID = {hal-02902346},
  HAL_VERSION = {v1},
}

Other softwares

There are many good softwares already available.

  • For continuation in small dimension, most softwares are listed on DSWeb. One can mention the widely used AUTO-07p, or also, XPPAUT, MATCONT, PyDSTool and COCO. All these are very reliable and some address high codimensional bifurcations.

  • For large scale problems, there is the versatile and feature full pde2path but also Trilinos, CL_MATCONTL, COCO, GetFEM and the python libraries pyNCT and pacopy.

  • For deflated continuation, there is defcont (by the inventor of the algo. P. E. Farrell) and this code by N. M. Evstigneev.

In Julia, we also have Bifurcations.jl.

A word on performance

The examples which follow have not all been written with the goal of performance but rather simplicity (except maybe 2d Ginzburg-Landau equation (finite differences, codim 2, Hopf aBS) and 1d Langmuir–Blodgett transfer model (advanced)). One could surely turn them into more efficient codes. The intricacies of PDEs make the writing of efficient code highly problem dependent and one should take advantage of every particularity of the problem under study.

For example, in one of the simplest tutorials, Temperature model (Simplest example), one could use BandedMatrices.jl for the jacobian and an inplace modification when the jacobian is called ; using a composite type would be favored. Porting them to GPU could be another option.

Main features

  • Newton-Krylov solver with generic linear / eigen preconditioned solver. Idem for the arc-length continuation.
  • Newton-Krylov solver with nonlinear deflation and preconditioner. It can be used for branch switching for example.
  • Continuation written as an iterator
  • Monitoring user functions along curves computed by continuation, see events
  • Continuation methods: PALC, Deflated continuation
  • Bifurcation points are located using a bisection algorithm
  • detection of Branch, Fold, Hopf bifurcation point of stationary solutions and computation of their normal form.
  • Automatic branch switching at branch points (whatever the dimension of the kernel)
  • Automatic branch switching at simple Hopf points to periodic orbits
  • Automatic bifurcation diagram computation of equilibria
  • Fold / Hopf continuation based on Minimally Augmented formulation, with Matrix Free / Sparse Jacobian.
  • detection all codim 2 bifurcations of equilibria and computation of the normal forms of Bogdanov-Takens, Bautin and Cusp
  • Branching from Bogdanov-Takens points to Fold / Hopf curve
  • Periodic orbit computation and continuation using Shooting, Finite Differences or Orthogonal Collocation.
  • detection of Branch, Fold, Neimark-Sacker, Period Doubling bifurcation point of periodic orbits.
  • Continuation of Fold of periodic orbits

Custom state means, we can use something else than AbstractArray, for example your own struct.

Note that you can combine most of the solvers, like use Deflation for Periodic orbit computation or Fold of periodic orbits family.

FeaturesMatrix FreeCustom stateTutorialGPU
(Deflated) Krylov-NewtonYesYesAllY
Continuation (Natural, Secant, Tangent, Polynomial)YesYesAllY
Deflated ContinuationYesYesLinkY
Branching / Fold / Hopf point detectionYesYesAll / All / LinkY
Fold Point continuationYesYesLink, LinkY
Hopf continuationYesAbstractArrayLink
Branch switching at Branch / Hopf pointsYesAbstractArrayLinkY
Automatic bifurcation diagram computation of equilibriaYesAbstractArrayLink
Periodic Orbit (Trapezoid) Newton / continuationYesAbstractVectorLink, LinkY
Periodic Orbit (Collocation) Newton / continuationYesAbstractVectorLink
Periodic Orbit with Parallel Poincaré / Standard Shooting Newton / continuationYesAbstractArrayLink
Fold, Neimark-Sacker, Period doubling detectionYesAbstractVectorLink
Continuation of Fold of periodic orbitsYesAbstractVectorLinkY
Bogdanov-Takens / Bautin / Cusp / Zero-Hopf / Hopf-Hopf point detectionYesYesY
Branching from Bogdanov-Takens points to Fold / Hopf curveNoAbstractVector

Requested methods for Custom State

Needless to say, if you use regular arrays, you don't need to worry about what follows.

We make the same requirements as KrylovKit.jl. Hence, we refer to its docs for more information. We additionally require the following methods to be available:

  • Base.length(x): it is used in the constraint equation of the pseudo arclength continuation method (see continuation for more details). If length is not available for your "vector", define it length(x) = 1 and adjust tuning the parameter theta in ContinuationPar.
  • Base.copyto!(dest, in) this is required to reduce the allocations by avoiding too many copies
  • Base.eltype must be extended to your vector type. It is mainly used for branching.

Citations

The papers citing this work are collected on google scholar.