Bifurcation analysis of Bogdanov-Takens bifurcations in delay differential equations_2

2025-04-27 0 0 9.12MB 128 页 10玖币
侵权投诉
Bifurcation analysis of Bogdanov-Takens bifurcations in delay
differential equations
M.M. BosschaertYu.A. Kuznetsov
October 7, 2022
Abstract: In this paper, we will perform the parameter-dependent center manifold reduction
near the generic and transcritical codimension two Bogdanov–Takens bifurcation in classical delay
differential equations (DDEs). Using a generalization of the Lindstedt-Poincaré method to approx-
imate the homoclinic solution allows us to initialize the continuation of the homoclinic bifurcation
curves emanating from these points. The normal form transformation is derived in the functional
analytic perturbation framework for dual semigroups (sun-star calculus) using a normalization
technique based on the Fredholm alternative. The obtained expressions give explicit formulas,
which have been implemented in the freely available bifurcation software package DDE-BifTool.
The effectiveness is demonstrated on various models.
Keywords: generic Bogdanov–Takens bifurcation, transcritical Bogdanov–Takens bifurcation, ho-
moclinic solutions, delay differential equations, sun-star calculus, strongly continuous semi-groups,
Center Manifold Theorem, DDE-BifTool
2020 MSC: 37G05, 37G10, 65P30, 34K16, 34K18, 34K19
1 Introduction
The Bogdanov–Takens bifurcation caused by the presence of an equilibrium with a double zero eigen-
value is a well-studied singularity in dynamical systems. It implies existence of saddle homoclinic
orbits nearby, which is a global phenomenon. In particular, the codimension two Bogdanov–Takens
bifurcation in finite-dimensional ordinary differential equations (ODEs) has been studied theoretically
and applied in numerous reserach publications. The same is true for the infinite-dimensional dynamical
systems generated by delay differential equations (DDEs). In the simplest case, often encountered in
applications, such DDEs have the form
˙x(t) = f(x(t), x(tτ1), . . . , x(tτm), α), t 0,(1)
where x(t)Rn, α Rp, while 0< τ1< τ2<··· < τmare constant delays, and f:Rn×(m+1) ×Rp
Rnis a smooth mapping. These are known as discrete DDEs.
Up to date, the standard available parameter-dependent center manifold theorem for DDEs in [13]
assumed that the equilibrium persists for all nearby parameter values. This was a serious limitation,
since in generic unfoldings of the codimension two Bogdanov–Takens singularity it is not the case.
Submitted to the editors DATE.
Department of Mathematics, Hasselt University, Diepenbeek Campus, Agoralaan Gebouw D, 3590 Diepenbeek,
Belgium (maikel.bosschaert@uhasselt.be).
Department of Mathematics, Utrecht University, Budapestlaan 6, 3508 TA Utrecht, The Netherlands and De-
partment of Applied Mathematics, University of Twente, Zilverling Building, 7500AE Enschede, The Netherlands
(I.A.Kouznetsov@uu.nl).
1
arXiv:2210.02560v1 [math.DS] 5 Oct 2022
However, recently, in [2], this obstruction has been removed and the existence of finite-dimensional
smooth parameter-dependent local center manifolds has been rigorously established in the functional
analytic perturbation framework for dual semigroups (sun-star calculus) developed in [5,6,7,8]. Once
the existence of these invariant manifolds is proved, the normalization technique for local bifurcations
of ODEs developed in [27] can be lifted rather easily to the infinite dimensional setting of DDEs. The
advantages of this normalization technique are that the center manifold reduction and the calculation of
the normal form coefficients are performed simultaneously by solving the so-called homological equation.
This method gives explicit expressions for the coefficients rather than a procedure as developed in
[17,18]. The explicit expressions make them particularly suitable for both symbolic and numerical
evaluation.
Indeed, utilizing the normalization method, the authors in [2] obtained asymptotics to initialize the
continuation of codimension one bifurcation curves of nonhyperbolic equilibria and cycles emanating
from the codimension two generalized Hopf, fold-Hopf, Hopf-Hopf and transcritical-Hopf bifurcation
points in DDEs of the form (1). These asymptotics have been implemented into the fully GNU Octave
compatible MATLAB package DDE-BifTool [16,35].
Another recent developent is the rigorous derivation of higher-order asymptotics for the codimen-
sion one homoclinic bifurcation curve emanating from the generic codimension two Bogdanov–Takens
bifurcation point in in two-parameter ODEs [3]. Thus, by combining the results of the parameter-
dependent center manifolds in DDEs from [2] and the homoclinic asymptotics in ODEs from [3], we
are in the position to perform the parameter-dependent center manifold reduction and normalization
for the generic and transcritical codimension two Bogdanov–Takens bifurcations. This will allow us
to initialize the continuation of codimension one bifurcation curves of nonhyperbolic equilibria and
homoclinic solutions emanating from these codimension two points. Hopefully, our results and their
software implementation will make the numerical analysis of Bogdanov–Takens bifurcations in DDEs
from applications rather routine.
This paper is organized as follows. We begin in Section 2 with a short summary from [2] on
parameter-dependent center manifolds for classical DDEs and we state various results needed for the
normalization technique. In Section 3 we describe the general technique that we use to derive the trans-
formation from the orbital normal form on the parameter-dependent center manifold in the infinite-
dimensional setting of DDEs. In Section 4 the method is then applied to the generic and transcritical
codimension two Bogdanov–Takens bifurcations. We provide explicit transformations necessary for the
predictors of codimension one bifurcation curves. We do this in a form suitable for classical DDEs,
covering cases that are more general than (1). It is here where we see the true benefit of allowing for
orbital normal forms on the center manifold. Indeed, we do not need to derive homoclinic asymptotics
for the transcritical codimension two Bogdanov–Takens bifurcation separately. Instead, we only need
to derive the center manifold transformation for the transcritical codimension two Bogdanov–Takens
bifurcation. Then, using the blow-up transformations (82), we obtain the same perturbed Hamiltonian
system (up to order three) as in the generic Bogdanov–Takens bifurcation.
We employ our implementation in DDE-BifTool to illustrate the accuracy of the codimension one
bifurcation curve predictors through various example models, displaying the generic and transcritical
codimension two Bogdanov–Takens bifurcations in Section 5. An in-depth treatment of the examples,
including the MATLAB and Julia source code to reproduce the obtained results, as well as a more in-depth
analysis of the examples, are provided in the Supplement.
2 Parameter-dependent center manifolds for DDEs
Here we summarize those results from [2] on parameter-dependent center manifold for classical DDEs,
which are required for the normalization technique in Section 4. For a general introduction on pertur-
bation theory for dual semigroups (also known as sun-star calculus) we refer to [13].
2
Consider the classical parameter-dependent DDE
˙x(t) = F(xt, α), t 0,(DDE)
where F:X×RpRnis Ck-smooth for some k1with F(0,0) = 0 and X:=C([h, 0],Rn). Here
for each t0, the history function xt: [h, 0] Rndefined by
xt(θ):=x(t+θ),θ[h, 0].
It is convenient to split the right hand-side into its linear and nonlinear parts and write
F(φ, α) = hζ, φi+D2F(0,0)α+G(xt, α).(2)
Here ζ: [0, h]Rn×nis a matrix-valued function of bounded variation, normalized by the requirement
that ζ(0) = 0 and is right-continuous on the open interval (0, h), and G:XRnis a Ck-smooth
nonlinear operator with G(0,0) = G1(0,0) = G2(0,0) = 0. The pairing is defined by
hζ, φi:=Zh
0
(θ)φ(θ),(3)
where the integral is of the Riemann–Stieltjes type.
Let Tbe the C0-semigroup on Xcorresponding to the linearization of (DDE) at 0Xfor the
critical parameter value α= 0. Suppose that the generator
D(A) = {φ|˙
φX, ˙
φ(0) = hζ, φi}, =˙
φ,
of Thas 1n0<purely imaginary eigenvalues with corresponding n0-dimensional real center
eigenspace X0. Then by [2, Corollary 20] there exists a Ck-smooth map C:U×UpXdefined in a
neighborhood of the origin in X0×Rpand such that for every sufficiently small αRpthe manifold
Wc
loc(α):=C(U, α)is locally positively invariant for the semiflow generated by (DDE) at parameter
value α.
Since Xis not reflexive, i.e. does not isomorphic to its dual space X?, the adjoint semigroup T?is
only weak?continuous on X?and A?generates T?only in the weak?sense. The maximal subspace of
strong continuity
X:={x?X?:t7→ T?(t)x?is norm-continuous on R+}
is invariant under T?, and we have the representation
X=Rn? ×L1([0, h],Rn?).(4)
The duality pairing between φ= (c, g)Xand φXis
hφ, φi=(0) + Zh
0
g(θ)φ(θ). (5)
At this stage, we again have a C0-semigroup Twith generator Aon a Banach space Xso we
can iterate the above construction once more. On the dual space
X?=Rn×L([h, 0],Rn),
we obtain the adjoint semigroup T?with the weak?generator
D(A?) = {(α, φ)X?|φLip(α)}, A?(α, φ)=(hζ, φi,˙
φ).(6)
3
The duality pairing between φ?= (a, ψ)X?and φ= (c, g)Xis
hφ?, φi=ca +Zh
0
g(θ)ψ(θ). (7)
By restriction to the maximal subspace of strong continuity X =D(A?), we end up with the
C-semigroup T. Its generator A is the part of A?in X. The canonical injection j:XX?
is given by
j(φ)=(φ(0), φ),(8)
mapping Xonto X. Therefore, Xis sun-reflexive with respect to the shift semigroup T.
We are now in the position to state the second part of [2, Corollary 20]. That is, if the history xt
associated with a solution of (DDE) exists on some nondegenerate interval Iand xt∈ Wc
loc(α)for all
tI, then u:IXdefined by u(t):=xtis differentiable and satisfies
j˙u(t) = A?ju(t)+(D2F(0,0)α)r?+G(u(t), α)r?,tI.
Here, for i= 1, . . . , n, we denote r?
i:= (ei,0), where eiis the ith standard basis vector of Rnand
wr?:=
n
X
i=1
wir?
i,w= (w1, . . . , wn)Rn.
2.1 Spectral computations for classical DDEs in case of multiple eigenvalues
It is well known that for classical DDEs all spectral information about the generator Ais contained in
a holomorphic characteristic matrix function ∆ : CCn×ndefined by
∆(z):=zI ˆ
ζ(z)with ˆ
ζ(z):=Zh
0
ezθ (θ),(9)
where ζis the real kernel from (3), see [13, Sections IV.4 and IV.5]. In particular, the eigenvalues of A
are the roots of the characteristic equation
det ∆(z) = 0,(10)
and the algebraic multiplicity of an eigenvalue equals its order as a root of (10).
For the normalization technique in Section 4, we will need normalized representations for the (gener-
alized) eigenfunctions and adjoint (generalized) eigenfunctions of the generator Aand A?, respectively.
In this section, we will consider the more general case where λis an eigenvalue of algebraic multiplicity
kNand geometric multiplicity 1. Although in Section 4 we will only need the special case where
λ= 0 is a double eigenvalue of A, the expressions are useful when considering for example the 1:1
resonant Hopf bifurcation and the triple zero bifurcation.
Proposition 1. Let λbe an eigenvalue of the generator Awith algebraic multiplicity kNand
geometric multiplicity one, then there are (generalized )eigenfunctions φisuch that
0=λφ0, Aφi=λφi+φi1, i ∈ {1, . . . , k 1},(11)
and adjoint (generalized )eigenfunctions ψisuch that
A?ψk1=λψk1, A?ψki=λψki+ψki+1, i ∈ {2, . . . , k}.(12)
Let the ordered set (q0, . . . , qk1)of vectors be a Jordan chain for ∆(λ), i.e. q06= 0 and
∆(z)[q0+ (zλ)q1+··· + (zλ)kqk1] = O((zλ)k).
4
Similarly, let (pk1, . . . , p0)be a Jordan chain for T(λ). Then the (generalized )) eigenfunctions and
adjoint (generalized )eigenfunctions are given by
φi: [h, 0] Rn:θ7→ eλθ
i
X
l=0
qil
θl
l!,
ψi: [0, h]Rn:θ7→ pi+
k1i
X
l=0
pi+lZθ
0Zh
σ
eλ(σs)(σs)l
l!(s)dσ,
(13)
for i∈ {0, . . . , k 1}, respectively. Furthermore, the following identities hold
hψi, φji=hψi+1, φj+1i, i, j ∈ {0, . . . , k 2},
hψk1, φk1i=pk1
k1
X
l=0
(l+1)(λ)
(l+ 1)! qk1l,
hψi, φji= 0, i > j,
hψ0, φji=
k1
X
l=0
j
X
m=0
pl
(l+m+1)(λ)
(l+m+ 1)! qjm, j > 0,
(14)
which can be normalized to satisfy
hψi, φji=δij .(15)
Proof. The (generalized) eigenspace at an eigenvalue λof Aof algebraic multiplicity kand geometric
multiplicity 1 is given by
N((Aλ)k),
which leads to the expressions in (11) and similarly for (12). The representations of the (generalized)
eigenfunctions and adjoint (generalized) eigenfunctions can be found in Theorem IV.5.5 and IV.5.9
in [13], respectively.
The first identity in (14) follows directly from
hψi+1, φj+1i=h(λA?)ψi, φj+1i=hψi,(λA)φj+1i=hψi, φji,
where i, j ∈ {0, . . . , k 2}. For the second identity in (14) we notice that
hψk1, φk1i=Zh
0
k1(θ)φk1(θ) = pk1q0+Zh
0
ψ0
k1(θ)φk1(θ)
=pk1q0+Zh
0
pkiZh
θ
eλ(θs)(s)eλθ
k1
X
l=0
qil
(1)lθl
l!
=pk1q0+pki
k1
X
l=0 Zh
0Zh
θ
eλs (1)lθl
l!(s)dθqil
=pk1q0+pki
k1
X
l=0 Zh
0Zs
0
eλs (1)lθl
l!(s)qil
=pk1q0+pki
k1
X
l=0 Zh
0
eλs (1)lsl+1
(l+ 1)! (s)qil=pk1
k1
X
l=0
(l)(λ)
(l+ 1)! qil,
where we used Fubini’s theorem to change the order of integration. The last equality holds since
0(z) = zI +Zh
0
θe(θ)
5
摘要:

BifurcationanalysisofBogdanov-Takensbifurcationsindelaydierentialequations*M.M.Bosschaert„Yu.A.Kuznetsov…October7,2022Abstract:Inthispaper,wewillperformtheparameter-dependentcentermanifoldreductionnearthegenericandtranscriticalcodimensiontwoBogdanovTakensbifurcationinclassicaldelaydierentialequat...

展开>> 收起<<
Bifurcation analysis of Bogdanov-Takens bifurcations in delay differential equations_2.pdf

共128页,预览5页

还剩页未读, 继续阅读

声明:本站为文档C2C交易模式,即用户上传的文档直接被用户下载,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私,请立即通知玖贝云文库,我们立即给予删除!
分类:图书资源 价格:10玖币 属性:128 页 大小:9.12MB 格式:PDF 时间:2025-04-27

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务
/ 128
客服
关注