ON THE BAUTIN BIFURCATION FOR SYSTEMS OF
DELAY DIFFERENTIAL EQUATIONS
by
Anca–Veronica Ion
Abstract. For systems of delay differential equations the Hopf bifurcation was investigated by several authors. The problem we solve here is that of the possibility of emergence of a codimension two bifurcation, namely the Bautin bifurcation, for some such systems.
Keywords: bifurcation theory, Bautin, delay differential equation
1. Introduction
The existence of periodic solutions for evolution equations is of certain interest for both pure and applied mathematicians. Even for bidimensional systems of differential equations the detection of limit cycles by theoretical means is difficult. The bifurcation theory offers a strong tool for finding limit cycles, namely the theory concerning the Hopf bifurcation (when there is a varying parameter)[2], [6]. Several authors studied the Hopf bifurcation for delay differential equations (e.g. [4], [7], [1], [5] ). We are interested to find sufficient conditions for the Bautin bifurcation for a class of such systems.
2. Setting of the problem, theoretical frame
Consider a system of the form
( ) ( ) ( ) ( )
α α(
( ) ( )
, ,α)
)
(t A x t B x t r f x t x t r
x = + − + −
•
, (1)
( )
,[ ]
,0)
(s s s r
x =φ ∈ − , (2) where x=
(
x1,..., xn)
∈Rn, α =(
α1,α2)
∈R2,A( )
α ,B(α) are nxn matrices over R, f =(
f1,..., fn)
is continuously differentiable on its domain of definition, ⊂ 2n+1D R . Moreover, f
(
0,0,α)
=0and the differential of f in the first two vectorial variables, calculated at(
0,0,α)
is equal to zero. φis an element of the Banach space B=C(
[ ]
−r,0,Rn)
(column vectors). In order to write eq. (1) as a differential equation in a Banach space, the space[ ]
[
)
{
n n}
R
R ∃ ∈
→ −
= : ,0 , iscontinuouson -r,0 and lim→ ( ) B
-0 s
is considered in section 8.2 of [4]. Its elements are ψ =ϕ +X0σ , with
n
R
∈
∈ σ
ϕ B, (column vector) and
= < ≤ − =
, 0 ,
0 ,
0 ) ( 0
s I
s r s
X
n
where 0,Inare the zero and, respectively, the unity nxn matrix. The norm of ψ is defined as the sum of the norm of ϕ in B and the norm of σ in n
R . The complexifications B ,C B0C of B , respectivelyB , are used below. 0
Considerδ(⋅)- the Dirac function, and the nxn matrix valued function
n
I s
s) ( )
( =δ
∆ . Also consider the bounded linear operator Lα :B→Rn,
∫
−= 0
) ( ) (
rd s s
Lαϕ ηα ϕ ,
with )ηα(s)=A(α)∆(s)−B(α)∆(s+r . By denoting
[ ]
,0. ),( )
(s x t s s r
xt = + ∈ − ,
we have in the spirit of [4], the following relations, equivalent with (1), (2):
( )
0, ( ), ), () ( ) 0
( Lα x f x x r α dt
dx
t t t
t = + −
(3)
), ( )
( s
ds dx s dt
dxt t
= (4) .
0 =ϕ
x (5) Define (see also [4]) the linear operator, 0 ( )
( )
0 ,~
−
+
=ϕ• ϕ ϕ•
ϕ α
α X L
A
[ ]
(
r)
BC BCC
A : 1 − ,0,Rn ⊂ →
~
α . Now we can rewrite the above problem as
), ), ( ), 0 ( (
0 ~
α
α x X f x x r
A dt dx
t t t
t = + −
, (6) .
0 =ϕ
x (7) The last term of (6) may be written as
∫
∫
− ∆ − ∆ +0 0
0f( rd (s)xt(s), rd (s r)xt(s), )
X α .
We define
∫
∫
−
− ∆ ∆ +
= 0 0
) ), ( ) ( ,
) ( ) ( ( ) , (
r t
r t
t f d s x s d s r x s
x
F α α . Thus (6) and
(7) take the form
), , (
0 ~
α
α t t
t
x F X x A dt dx
+
ϕ = 0
x , (9) this being the abstract problem in B equivalent to (1),(2). 0
The eigenvalues of α
~
A are (see [4]) the roots of the equation
( )
( )
(
)
0detλI −Aα −e−λrBα =
. We assume the following hypothesis, that we denote H1.
H1. An open set U exists in the parameter plane such that for every α∈U,
there is a pair of complex conjugated simple eigenvalues
( ) ( )
α µ α ω( )
αλ1,2 = ±i , with the property that there is a α0∈U such that
( )
0( )
0 02 ,
1 α ω α ω
λ =±i =±i , with ω0 >0 and for every α∈U, all other
eigenvalues have strictly negative real parts, uniformly bounded from above by
a negative number.
By a simple eigenvalue we mean an eigenvalue having the algebraic multiplicity equal to 1.
We remark that H1 implies the existence of a neighborhood of α0such that each eigenvalue different from λ1,2
( )
α has real part strictly less than µ( )
α . The eigenvectors corresponding to λi( )
α , i=1,2, are elements ofB -the C complexification of B , namely( )( )
s e ( ) i( )( )
0, s[ ]
r,0s i
i ∈ −
= ϕ α
α
ϕ λ α ,
where ϕi
( )( )
α 0 is a solution of(
λi( )
α I −A( )
α −e−λi( )α rB( )
α)
ϕ =0. Obviously,( )
1( )2 α ϕ α
ϕ = .
Denote by {λ ( )α}
2 , 1
M the linear subspace of B , spanned by C {ϕ1
( ) ( )
α ,ϕ2 α } and Φ( )
α the matrix having as columns the vectors( ) ( )
α ϕ αϕ1 , 2 . Let {ψ1
( ) ( )
α ,ψ2 α } be two eigenvectors for the adjoint problem ([3], [4]), corresponding to the eigenvalues −λ1,2( )
α of the infinitesimal generator of the adjoint problem. They are elements of B*C- the complexification of B* =C(
[ ]
0,r,Rn)
-row vectors, and we assume that they are selected such that, Ψ( )
α being the matrix having as rows the vectors( ) ( )
α ψ αψ1 , 2 , the relation (Ψ
( ) ( )
α ,Φα )= I2 holds, where( ) ( )
0 0 ( ) ( ) ( ) , B , B . ),
( *C C
0
0
∈ ∈
− −
=
∫∫
− ψ ξ θ η θ ϕ ξ ξ ψ ϕ
ϕ ψ ϕ
ψ θ α
r
d d
In [4] a projection π {λ ( )α}
2 , 1
0C B
: →M , is defined by
(
ϕ α) ( ) ( )
α[
(
α ϕ) ( )
α]
π +X0 =Φ Ψ , +Ψ 0 . With this projection the space B is 0C decomposed as B {λ ( )α} Kerπ
2 , 1
0C =M ⊕ . Since the solution xtof (8) and (9)
belongs to C1([−r,0],Rn), it is decomposed as )
( ) ( )
( u t y t
xt =Φ α + , (10) with u(t)=
(
Ψ(α),xt)
and y(t)=(I−π)xt, where ( ) ( ( ), ( ))_
t z t z t
u = -column
vector,z(t)∈C. Let us define
=
) ( 0
0 ) ( ) (
1 1
α λ α λ α
B .
The projection of eq. (8) on {λ ( )α}
2 , 1
M is
( )
α u=Φ( ) ( )
α Bα u+Φ( ) ( )
α Ψ α (0)F(Φ(α)u(t)+ y(t),α)Φ • ,
and since Φ(α) is invertible, this is equivalent to
( )
α u( )
α (0)F( (α)u(t) y(t),α)B
u• = +Ψ Φ + . (11) By projecting the initial condition we find
(
(α),ϕ)
) 0
( = Ψ
u .
3. Existence of the invariant manifold and the restricted equation
If α∈U \
{ }
α0 , and Reλ1,2( )
α >0 then, since these are the only two eigenvalues with positive real part (and they are simple), there is a local invariant manifold for the problem, namely the local unstable manifold, tangent to the space {λ ( )α}2 , 1
M , [3], [7].
For α =α0, since Reλ1,2
( )
α0 =0, there is a local invariant manifold for the problem, namely the local center manifold, tangent to the space { ( )}0 2 , 1 α
λ
M ,
[7].
Hence, for every α∈U with µ
( )
α ≥0, there is a neighborhood( )
αgraph of a C1function. That is, the local invariant manifold may be expressed as
( )
α{
ϕ wα( )
ϕ ϕ {λ ( )α} V( )
α}
Wloc = + ∈ ∩
2 , 1
; M ,
where wα {λ ( )α}→Kerπ 2
, 1
:M is a C1 function, wα
( )
0 =0, and it has zero differential at 0. Since ϕ {λ ( )α}2 , 1
M
∈ , we have ϕ
( )
α =zϕ1( )
α +zϕ1( )
α , withC
∈
z . This relation induces a dependence of z,z to wα
( )
ϕ that justifies the notation wα( )
ϕ =wα( )
z,z .Equation (11) implies
( )
α(
ϕ α ϕ α α)
ψ α
λ( ) ( ) (0) ( ) ( ) ( ) ( ) α( ( ), ( )),
)
(t 1 z t 1 F z t 1 z t 1 w z t z t
z = + + +
•
, (12)
(
ψ (α),ϕ)
) 0
( = 1
z . (13) Let Sα(t)φ be the solution of eq. (1) corresponding to the initial condition φ, at the moment t .
If φ∈Wloc(α), then
)) ( ), ( ( ) ( ) ( ) ( ) ( )
(t z t 1 z t 1 w z t z t
Sα φ = ϕ α + ϕ α + α . (14) By using again the function f, (12) becomes
( )
(0) ( ( ) (0), ( ) ( ), ) )( ) ( )
(t λ1 α z t ψ1 α f Sα t φ Sα t φ r α
z• = + − ,
(15) or,
), , ) ( ), ( ( ) ( ) ( )
(t λ1 α z t g z t z t α
z• = + (16) by denoting
( )
(0) ( ( ) (0), ( ) ( ), ) ), ) ( ), (
(z t z t α ψ1 α f Sα t φ Sα t φ r α
g = − . (17)
4. The equations for the invariant manifold
The following proposition is a natural consequence of the invariance of
( )
αloc
W . A similar result is given in [7], on the center manifold. We give the proof for the sake of completeness.
Proposition 1. Let φ∈Wloc
( )
α be the initial value for the problem (1). Then the( )( )
( )( )
[ ]
,0 ,), ))( ( ), ( ( ) ), ( ), ( ( ) ), ( ), ( ( ) ))( ( ), (
( 1 1
r s s t z t z w s s t z t z g s t z t z g s t z t z w t − ∈ ∂∂ = = + + ∂∂ α α α ϕ α α ϕ α (18)
( )( )
( )( )
) ), ( ) ( ), 0 ( ) ( ( ) ))( ( ), ( ( ) ( ) 0 ))( ( ), ( ( ) ( 0 ) ), ( ), ( ( 0 ) ), ( ), ( ( ) 0 ))( ( ), (( 1 2
α φ φ α α α ϕ α α ϕ α α α α α α r t S t S f r t z t z w B t z t z w A t z t z g t z t z g t z t z w t − + − + = = + + ∂∂ (19)
with z(t) solution of the Cauchy problem (16),(13) and g defined by (17).
Proof Since φ∈Wloc
( )
α and Wloc( )
α is invariant, Sα(t)φ∈Wloc(α). Let us denote, for t ≥0 and s∈[ ]
−r,0 , )Sα(t)φ(s)=x(φ)(t+s . Obviously ) ( ) ( ) ( ) ( s t s x s t t x + ∂ ∂ = + ∂ ∂ φ φ .This and (14) imply
( )
( )
( )
( )
( )( )
( ) ( )( )
( ) )) ( ), ( ( ) ( ) ( ) ( ) ( )) ( ), ( ( 1 1 1 1 s t z s t z s t z t z w s s t z s t z s t z t z w t α ϕ α ϕ α ϕ α ϕ α α • • • • + + ∂∂ = = + +∂∂ (20)
(here 1
( )
( ) 1( )
(s)ds d s ϕ α
α
ϕ• = ) and thus
( )
( )
w z t z t( )
ss s t z s t z s t z t z w
t ( ( ), ( )) ( ) ( ) ( )( ) ( ) ( ) ( ) ( ( ), ( ))
_ 1 1 1 1 _ α α λ α ϕ α λ α ϕ α = ∂∂ − + − + ∂∂ • •
. (21) With (16) we obtain (18).
On another side, since Sα(t)φis a solution of equation (1), we have
and, by taking s=0, we obtain (19).
This proposition allows the determination of the coefficients of the series of powers in z and
_
z of the functionwα. Indeed, let us write
( )
, ! ! 1 ) ), ( ) ( ), 0 ( ) ( ( 2 k j k jjk z z
F k j r t S t S f
∑
≥ + = − α α φ φ αα (22)
( )
( )
j kk j
jk z z
g k j t z t z g
∑
≥ + =2 ! !
1 )
, ), (
( α α ,
( )
j kk j
jk z z
w k j z z w
∑
≥ + = 2 , ! ! 1 ) , , (θ θ α α (23) where gjk(α)=ψ1(0)Fjk(α).By replacing (22) and (23) in (18) and by matching the obtained series, we get first order linear differential equations for wjk.
Thus, equation (18) implies
( )
( )
( )( )
( )
( )( )
( )
, . ! ! 1 ! ! 1 ! ! 1 , ! ! 1 1 1 2 1 2 1 2 2 + + + + = • − • − ≥ + ≥ + ≥ + ≥ +∑
∑
∑
∑
z z kz z z jz s w k j s z z g k j s z z g k j z z s w ds d k j k j k j k j jk k j k j jk k j k j jk k j k j jk α α ϕ α α ϕ α α (24) In this equality z• and•
z will be replaced with the right hand side of (16) to obtain
( )
( )
( )( )
( )
( )( )
( )
(
)
( )
( )
( )
. ! ! 1 ! ! 1 , ! ! 1 ) ( ) ( , ! ! 1 ! ! 1 ! ! 1 , ! ! 1 2 _ 1 2 1 2 1 1 2 1 2 1 2 2 + + + + + + + =∑
∑
∑
∑
∑
∑
∑
≥ + − ≥ + − ≥ + ≥ + ≥ + ≥ + ≥ + m l m l lm k j m l m l lm k j k j jk k j k j jk k j k j jk k j k j jk k j k j jk z z g m l z kz z z g m l z jz s w k j z z k j s w k j s z z g k j s z z g k j z z s w ds d k j α α α α λ α λ α α ϕ α α ϕ α α (25) By matching the same order terms we obtain first order differential equations for wjk(.,α).( )
(
)
( )
( )
( )
( )
( )( )
( )
( )( )
( )
(
)
( )
. ! ! 1 , ! ! 1 ) ( , 0 ! ! 1 ) ( 0 ! ! 1 0 ! ! 1 ! ! 1 ! ! 1 , 0 ! ! 1 ) ( ) ( , 0 ! ! 1 2 2 2 1 2 1 2 2 _ 1 2 1 2 1 1 2 k j k j jk k j k j jk k j k j jk k j k j jk k j k j jk m l m l lm k j m l m l lm k j k j jk k j k j jk z z F k j z z r w k j B z z w k j A z z g k j z z g k j z z g m l z kz z z g m l z jz w k j z z k j w k j∑
∑
∑
∑
∑
∑
∑
∑
∑
≥ + ≥ + ≥ + ≥ + ≥ + ≥ + − ≥ + − ≥ + ≥ + + − + = = + + + + + + + α α α α α α ϕ α α ϕ α α α α α λ α λ α (26)The relations obtained by equating the terms with similar powers of z, _
z in (26) are used as conditions to determine the constants that appear in the general form of wjk obtained above.
In this theoretical form the calculus is very lengthy and we do not make it explicitly here.
We firstly remark that the coefficients of the second order terms in z and _
z in the expansion of fi(Sα(t)φ(0),Sα(t)φ(−r),α),i=1,...,n, are independent on those of w ,α they depend only on the coefficients of the Taylor series of
(
x,y,α)
fi . The similar assertion holds for the coefficients of g(z(t),z(t),α).
Hence g20
( ) ( ) ( )
α ,g11 α ,g02 α are known, given fi and ψ1( )
α . The following algorithm to determine wjk( )
α must be used.- w20
( ) ( ) ( )
α ,w11 α ,w02 α are determined from the equations obtained by identifying the terms containing z2,zz,z2respectively, in (25), with initial conditions obtained by the same method from (26); they depend on g20( ) ( ) ( )
α ,g11 α ,g02 α .- w20
( ) ( ) ( )
α ,w11 α ,w02 α are used to compute gjk( )
α , j+k=3, (from (14), (17), (23) ) .- wjk
( )
α , j+k=3, are determined from the equations (25) and- wjk
( )
α , j+k≤3 are used to compute gjk( )
α , j+k=4 (from (14), (17), (23) ).- wjk
( )
α , j+k=4, are determined from the equations (25) andconditions (26); they depend on gjk
( )
α , j+k ≤4. - wjk( )
α , j+k ≤4 are used to compute gjk( )
α , j+k=5.We do not need so many terms to have an accurate form of the invariant manifold, but we need them in order to discuss the behaviour of the solution z
of (16) that determines the solution of (1) on the invariant manifold.
5. The Bautin bifurcation
After applying the above algorithm, equation (16) has the form
( )
( )
65 2 1
! !
1 )
( ) ( )
( g z z O z
k j t
z t
z j k
k j
jk +
+
=
∑
≤ + ≤ •
α α
λ . (27)
Now, let l1(α),l2(α)be the first and the second Lyapunov coefficient respectively, associated to (27), defined in [6]. They are functions of gkl(α). More specific, l1(α)is function of gkl(α) with 2≤k +l ≤3, and l2(α) is function of gkl(α) with 2≤k+l ≤5.
Let us define, with [6],
( )
( )
α ωµ αν1 = , ν2 =l1
( )
α , )ν =(ν1,ν2 . Let us consider the following hypothesis, denoted H2.H2. l1
( )
α0 =0, l2( )
α0 ≠0, and the map(
α1,α2)
→(ν1,ν2) is regular at α0. The value in α0of the first Lyapunov coefficient is(
( ) ( ) ( ))
Re 2
1 )
( 20 0 11 0 0 21 0
2 0 0
1 α α ω α
ω
α ig g g
l = + .
The value in α0of the second Lyapunov coefficient is much more complicated and we do not reproduce it here (see [6]).
Th. 8.2. of [6] asserts that, if H2 is satisfied for eq. (16), then eq. (16) may be transformed, by smooth invertible changes of the complex function z and time reparametrizations, into
(
)
( )
4( )
62 2 2
1 i z zz L zz O z
where L2
( )
ν =l2( )
α( )
ν .We suppose that H2 is satisfied for our problem.
Problem (28) is then (see [6]) locally, near z=0, topologically equivalent to the following complex normal form of the Bautin bifurcation
(
1 i)
z 2zz2 szz4,z• = β + +β + (29) where β1 =ν1,β2 = L2
( )
ν ν2, )s=signl2(α0. .Assume the third important fact denoted by H3. H3. l2
( )
α
0 >0 (hence s=1).Consider the polar form of (29), in the hypothesis H3
(
)
=
+ +
=
• •
, 1
,
4 2 2 1
η
ρ ρ β β ρ ρ
(30) with (ρ,η) the polar co-ordinates.
Due to H2, there is a C1bijection T between a neighbourhood U1 of α0, in the )
,
(α1 α2 plane and a neighbourhood W of (0,0) in the (β1,β2) plane.
We study the behaviour of the solutions only for β1 ≥0 (β1 and µ
( )
α have the same sign and we proved the existence of the invariant manifold only for µ( )
α ≥0).For thoseβ∈Wwith 0β1 ≥0,β2 ≥ , (30) has an repulsive focus in 0, and so does (29). Then, since the dynamical system generated by (29) is topologically equivalent to that generated by (16), this one has an unstable equilibrium in 0 (a repulsive focus or node, since they cannot be distinguished by topological equivalence).
In order to transport these conclusions to the solution of eq. (1), we remind relation (14):
)) ( ), ( , ( ) )( ( ) ( ) )( ( ) ( ) ( ) (
_
1
1 z t w z t z t
t z t
Sα φ θ = ϕ α θ + ϕ α θ + α θ .
In the intersection of the quadrant β1 ≥0,β2 <0 and Wthere is a zone, situated between the axisβ2 =0and the curve Γ=
{
(
β1,β2)
;β2 =−2 β1}
where eq. β1+β2ρ2 +ρ4 =0 has no positive solutions and thus there are no limit cycles for (30). The local phase portrait for (1) on the invariant manifold is as described above (unstable equilibrium point), for the corresponding (through −1T ) zone in the α plane, let us denote it by V2. Now let us put 2
1 V
V
V = ∪ .
In the zone situated between the curve Γ=
{
(
β1,β2)
;β2 =−2 β1}
and the axis β1 =0 eq. β1 +β2ρ2 +ρ4 =0 has two positive solutions,( ) ( )
β ρ βρ1 , 2 . Thus two concentric closed orbits for (30), hence for (29), exist. Since, for the corresponding zone of the αplane (let us denote it by *
V ), the problems (29) and (16) are topologically equivalent, eq. (16) will also have two periodic solutions. Relation (14) shows that, in this case, on the unstable manifold there exist two limit cycles for (1).
As we cross Γ, leaving the zone in the β plane described above, the two limit cycles collide and disappear.
6. Conclusions
The facts discussed above lead to the following result.
Theorem. If H1, H2, H3 are satisfied for eq. (1), then at α0a Bautin type
bifurcation takes place. There is a neighbourhood U of 1 α0in the α plane
having a subset V (with α0∈V ) with the property that for every α∈V , 0 is
an unstable equilibrium point (focus or node) for the problem (1) restricted to the bidimensional invariant manifold defined above.
There is also a subset V* of U , (having 1 α0as a limit point) with the
property that for every α∈V*, the restriction of problem (1) to the unstable
manifold has two concentric limit cycles. In this case also 0 is an unstable
equilibrium point.
References:
[1] R. Culshaw, S. Ruan, G. Webb, A mathematical model of cell to cell spread
of HIV that includes a time delay, J. Math. Biology, 46(2003), 425-444;
[2] Georgescu, A., M. Moroianu, I. Oprea, Bifurcation theory. Principles and
applications, Series of Applied and Industrial Mathematics of University of
Piteşti, 1999 (in Romanian);
[3] J. Hale, Functional Differential Equations, Applied Mathematical Sciences, 3, Springer, New York, 1971;
[4] J. Hale, L. Magalhães, W. Oliva, Dynamics in Infinite Dimensions,
Springer, New York, 2002;
[5] J. C. Ji, Stability and Hopf bifurcation of a magnetic bearing systems with
time delays, J. of Sound and Vibration, 259(4) (2003), 845-856;
[6] Y. Kuznetsov, Elements of Applied Bifurcation Theory, Applied Mathematical Sciences,112, Springer, New York, 1995;
[7] G. Mircea, M. Neamţu, D. Opriş, Dynamical systems from economy,
mechanics, biology, described by delay equations,Mirton, Timişoara, 2003 (in
Romanian).