5 Bifurcation and centre manifold

For the general ODE ẋ = f (x) near its stationary point x∗ , we learned early that if none of
the eigenvalues of the Jacobian Df (x∗ ) has a real part zero, then the behaviour of ẋ = f (x)
is determined by its linearised system ẏ = Df (x∗ )y with y ≈ x − x∗ . What happens if the
Jacobian matrix Df (x∗ ) has eigenvalues with zero real part?
If some eigenvalues have zero real part, nonlinear terms are expected to play a role,
and the behaviour could change accordingly. The study of these qualitative changes in the
behaviours (mainly stability/instability of stationary points and periodic orbits), subject to
changes in certain parameters, is call bifurcation theory. Since the stability/instability of
fixed points is indicated precisely by the real part of the eigenvalues, we are going to see how
these eigenvalues pass the imaginary axis, as the parameter changes.

Example 5.1. Consider the following two systems

(
ẋ = µx + ωy,
(a) ẋ = µx, (b)
ẏ = −ωx + µy.

It is easy to see that, the eigenvalue λ(µ) = µ in (a), and λ(µ) = µ ± iω in (b). The stability
is changed when µ crosses zero. More general scenario is shown in Figure 5.1.

λ(µ) Imλ
λ(µ)
iω

µ Reλ

−iω
λ∗ (µ)

Figure 5.1: Left figure: real eigenvalue passing through zero as a function of µ; Right
figure: complex eigenvalues passing through the imaginary axis (think of the eigenvalues as
parametrised curves in the complex plane).

5.1 Centre manifold theorem

We learned Stable Manifold Theorem earlier, which states that the structure of the system
near a hyperbolic fixed point does not change when nonlinear terms are added. Consider
the system
ẋ = −x, ẏ = y + x2
and its linearised system as shown in Figure 5.2.
The stable manifold E s and the unstable manifold E u for the linearised system (in normal
form) are easy to calculate, which is just the horizontal and vertical axis. Therefore, the

61
5.1 Centre manifold theorem

Eu Wu

Es Es

Eu Ws Wu Ws
ẋ = −x, ẏ = y ẋ = −x, ẏ = y + x2

Figure 5.2: The nonlinear system and its linearised system.

notation E s and E u (instead of E s and E u ) is used for the linearised system, to emphasize
that they are linear vector spaces. The corresponding stable and unstable manifold for the
nonlinear system are usually curved. For the system ẋ = −x, ẏ = y + x2 , the unstable
manifold is still the vertical axis (show this!), but the stable manifold is different, and can
be approximated as a local series expansion

W s = (x, y) | y = M(x) = a2 x2 + a3 x3 + · · · . (5.1)

The constant term in M(x) vanishes because W s passes the origin, and the linear term
vanishes because W s should be tangent to E s (the horozontal axis), the stable manifold of
the linearised system. Now the coefficients a2 , a3 , · · · representing W s can be obtained by
taking the derivative of both sides of y = M(x). On one hand

ẏ = y + x2 = (a2 + 1)x2 + a3 x3 + · · · .

On the other hand,

d


M(x) = 2a2 x + 3a3 x2 + · · · ẋ = (−x) 2a2 x + 3a3 x2 + · · · .
dt
Matching the two expressions for ẏ = dtd M(x), we get a2 = −1/3, a3 = a4 = · · · = 0. In
other words, the stable manifold is exactly y = −x2 /3.
Because the real parts of the eigenvalues are away from zero, the nonlinear system is stable
under changes in the parameters or nonlinear terms. However, if there is any eigenvalue with
zero real part, we expect some qualitative changes in the property when certain parameter
changes, which precisely why bifurcation theory and Centre Manifold Theorem are studied
together.

Theorem 5.1 (Centre Manifold Theorem). Given ẋ = f (x), x ∈ Rn , f smooth and suppose
x = 0 is a stationary point. Suppose the Jacobian matrix Df (0) has eigenvalues in sets σu
with Re(λ) > 0, σs with Re(λ) < 0 and σc with Re(λ) = 0 and corresponding generalized
linear eigenspaces E u , E s and E c respectively. Then there exist unstable and stable manifolds
W u , W s of the same dimension as E u , E s and tangential to E s and E u at x = 0; and an
invariant centre manifold W c tangential to E c at x = 0.
So in general, locally Rn = W c ⊕ W u ⊕ W s with the approximate governing equations on

62
5.2 Calculating the centre manifold W c

each manifold

ẋ = g(x) on W c
ẏ = By on W s (stable directions)
ż = Cz on W u (unstable directions) ,

where g(x) is quadratic (or higher order) in x, all eigenvalues of B have negative real parts,
and all eigenvalues of C have positive real parts.

Es Ws

Wc
Wc

Ec

Ws

Figure 5.3: Behaviour on W c depends on nonlinear terms, behaviour off W c is dominated

by exponential contraction in the E s direction.

In Figure 5.3, there is no unstable direction and in the stable direction the dynamics
is attracting, so solutions tend to the centre manifold very quickly. The dynamics on W c
depends on nonlinear terms, is usually much slower and characterise the dynamics of the
whole system in the long time. So the question is how this decomposition can be useful in
general, and how the centre manifold can be approximated or computed.

5.2 Calculating the centre manifold W c

Suppose that after a change of coordinate transformation, the hyperplane (or line, if x is one
dimension) (x, 0) is spanned by E c and (0, y) by E s , then the centre manifold is tangential
to y = 0 at (0, 0) and we may assume that

W c = (x, y) | y = h(x), h(0) = 0, Dh(0) = 0 .

In this coordinate, the system can be written as

ẋ = Ax + f1 (x, y), ẏ = Cy + f2 (x, y),

where all eigenvalues of A have real parts zero and those of C have real parts non-zero, fi
contain only nonlinear terms. So on the centre manifold W c ,

ẋ = Ax + f1 (x, h(x))

63
5.2 Calculating the centre manifold W c

and ẏ can be calculated on the centre manifold in two ways: directly from the ẏ equation
above, or by differentiating y = h(x), i.e.
d
ẏ = Ch(x) + f2 (x, h(x)) and ẏ = h(x) = Dh(x)ẋ = Dh(x) [Ax + f1 (x, h(x))] .
dt
where Dh(x) is the (matrix) partial derivatives of h(x), in one dimension simply h′ (x).
Expanding h as a Taylor series (noting that the constant and linear terms vanish), the two
equations for ẏ provide two different polynomials and the coefficients of different monomials
can be equated to determine the coefficients of the Taylor expansion.

For a specific problem, here is the general procedure to calculate the centre manifold
(which is very similar if you want to find the stable/unstable manifold):

(a) Change the system into normal form (if needed), such that the linearised system is a
diagonal matrix
(b) Identify the centre manifold E c of the linearised system, which is the linear space spanned
by the eigenvectors associated with the zero eigenvalues.
(c) Parameterise the centre manifold. You can parameterised E c first, and then for W c .
For instance, if E c is the y-axis, then E c is parameterised as x = 0 (and z = 0 if in
three dimension), and W c (also a line!) is parameterised by x = a2 y 2 + a3 y 3 + · · · and
z = b2 y 2 + b3 y 3 + · · · . If E c is the xy-plane in three dimension, then E c is parameterised
by z = 0 and W c is parameterised by

z = ax2 + bxy + cy 2 + · · · .

(d) Finally determine the coefficients in the parmaterisation by differentiation on both sides.
Example 5.2. Consider the system

ẋ = xy, ẏ = −y − x2 .

The linear normal form (based on the linearisation at the origin) has the constant matrix
 
0 0
A= = diag (0, −1).
0 1
Then the eigenpairs are
   
1 0
λ1 = 0, e1 = , λ2 = −1, e2 = .
0 1
Since the matrix A is already in normal form, no coordinate transformation is needed.
Now the centre manifold takes the form

y = h(x) = ax2 + bx3 + cx4 + O(x5 ), (5.2)

There is no constant term, because the centre manifold passes through the origin; there is
no linear term, because this manifold should be tangent to e1 (or equivalently E c , the centre
manifold of the linearised system).

64
5.2 Calculating the centre manifold W c

We can determine the coefficients by comparing two ways for calculating ẏ. Directly from
the ẏ equation of the system
ẏ = −y − x2 = −(ax2 + bx3 + cx4 ) − x2 + O(x5 ). (5.3)
On the other hand, differentiating (5.2) w.r.t t gives ẏ = dh(x)/dt = ẋh′ (x), i.e.,
x(ax2 + bx3 + cx4 + · · · )(2ax + 3bx2 + 4cx3 + · · · ) = 2ax4 + · · · . (5.4)
Equating coefficients of x2 , x3 and x4 in (5.3) and (5.4) gives
−a − 1 = 0, −b = 0, −c = 2a2 ,
i.e. a = −1, b = 0 and c = −2.
Thus the centre manifold is y = −x2 − 2x4 + O(x5 ) and the dynamics on the centre
manifold is
ẋ = xh(x) = −x3 − 2x5 + O(x7 ).
Thus ẋ < 0 if x > 0 and ẋ > 0 if x < 0. So the origin is stable and the solutions look like a
stable node, but the motion onto the centre manifold in the y-direction is much faster than
the motion on the centre manifold, leading to a phase portrait as shown in Figure 5.4.

W c : y = −x2 + · · ·

Figure 5.4: Phase portrait showing exponential collapse onto the centre manifold and then
slow motion towards (0, 0) on the centre manifold.

Remark. If you try higher order terms, you get

y = −x2 − 2x4 − 12x6 − 112x8 − 1360x10 − 19872x12 + · · · .
The fast increasing of the coefficients implies that this approximation is valid only in a small
neighbourhood of the origin.
Example 5.3. Consider the system
ẋ = y − x + xy, ẏ = x − y − x2 .
First we have to convert the linearised system
     
ẋ x −1 1
=A , A=
ẏ y 1 −1

65
5.2 Calculating the centre manifold W c

into normal form. It is easy to calculate the eigenpairs of A,

   
1 −1
λ1 = 0, e1 = , λ2 = −2, e2 = .
1 1
Let  −1  
−1 1 −1 1 1 1
U = [e1 , e2 ] = = ,
1 1 2 −1 1
the change of variable from (x, y) to (u, v) (in normal form) is
             
u x (x + y)/2 x −1 u u u−v
=U = , or =U = [e1 , e2 ] = .
v y (y − x)/2 y v v u+v
The new system in (u, v) is
u̇ = uv − v 2 , v̇ = −2v + uv − u2 .
The centre manifold is parametrised by v = h(u) = au2 + bu3 + cu4 + · · · , then
v̇ = (2au + 3bu2 + 4cu3 + · · · )u̇


= (2au + 3bu2 + 4cu3 + · · · ) u(au2 + bu3 + cu4 + · · · ) − (au2 + bu3 + cu4 + · · · )2
= 2a2 u4 + · · ·
and on the other hand
v̇ = −2v + uv − u2
= −2(au2 + bu3 + cu4 + · · · ) + u(au2 + bu3 + cu4 + · · · ) − u2
= (2a + 1)u2 + (2b − a)u3 + (3c − b)u4 + · · · .
Comparing the coefficients of u2 , u3 and u4 of the two expressions of v̇, we get
1 1 3 1 1 3
a = − , b = − , c = − , or v = − u2 − u3 − u4 + · · · .
2 4 8 2 4 8
The dynamics on the centre manifold is
1 1 3
u̇ = uv − v 2 = − u3 − u4 − u5 + · · · ,
2 4 8
which is stable if u is small. Going back to the original coordinates, the centre manifold is
approximately
1 1 3
y − x = − (x + y)2 − (x + y)3 − (x + y)4 + · · ·
4 16 64
Remark. The calculation could be quite involved if you are calculating unnecessary higher
order terms than needed in the end. In general, the lowest power appears in equations with
stable linear part. For instance, the second expression above starts with u2 , while the first
expression starts with u4 .
Remark. Strictly speaking, the change of variables from (x, y) to (u, v) is not necessary, but
we need to know that the centre manifold is represented as
y − x = a2 (x + y)2 + a3 (x + y)3 + · · · .
Take the derivative of both sides (w.r.t t),


ẏ − ẋ = 2a2 (x + y) + 3a3 (x + y)2 + · · · (ẋ + ẏ).
After substituting ẋ and ẏ, we compare the coefficients of powers of (x + y) on both sides,
and we should get the same answer. (Probably it is worth the effect to perform the change
of variable at the very beginning).

66
5.3 Extended centre manifold

5.3 Extended centre manifold

As it stands, the CMT does not allow us to deal with parameters. To include the effect of
parameters and hence to treat bifurcations, we work on the extended centre manifolds by
augmenting the equation with the apparently trivial equation µ̇ = 0:

ẋ = Ax + f1 (x, y, µ),
ẏ = Cy + f2 (x, y, µ),
µ̇ = 0.

The additional equation allows us to parametrise the centre manifold as y = h(x, µ) instead
of the form y = h(x) considered in the last section (hence the extended centre manifold).
The trivial equation µ̇ = 0 adds one more dimension to the centre manifold and allows
us to work in a neighbourhood of both (x, y) = (0, 0) in phase space and µ = 0 in parameter
space, where µ = 0 is the value at which the bifurcation occurs. So A has the zero real
part eigenvalues and C has stable and unstable manifolds, and f1 , f2 contain only nonlinear
terms. The CMT gives the motion on the stable and unstable manifolds,W s and W u in y,
and there is a nc + 1 dimensional centre manifold (where nc is the dimension of x), valid for
|x| and |µ| small.
This time, if coordinates are chosen so that the central motion is in normal form, the
extended centre manifold can be parametrised by y = h(x, µ), with

h(0, 0) = 0, Dh(0, 0) = 0.

Notice that Dh = [Dx h, Dµ h], which is the partial derivative w.r.t both variables. Then
ẋ = Ax + f1 (x, h(x, µ), µ) is the equation on the (extended) centre manifold.
There are three typical equations (to leading order) on the extended centre manifold if
A = 0 and x is a scalar:

ẋ = µ − x2 (saddlenode bifurcation)
ẋ = µx − x2 (transcritical bifurcation) (5.5)
ẋ = µx − x3 (pitchfork bifurcation)

x x x

µ µ µ
Figure 5.5: (µ, x) plane for local bifurcations of stationary points: saddlenode, transcritical
and pitchfork (supercritical).

Typical behaviour is sketched in the (x, µ) plane: these are called bifurcation diagrams.
By convention, dotted lines are used to show unstable solutions and continuous lines for

67
5.3 Extended centre manifold

stable solutions. The pitchfork illustrated her is supercritical, meaning that the non-trivial
stationary points are stable; if they were unstable then it would be a subcritical pitchfork
bifurcation (a subcritical pitchfork bifurcation when the solid line is dashed, and the dashed
line is solid). More details will be given in the next subsection.
Example 5.4. Consider the second order ODE

ü + u̇ − µu + u2 = 0

with a parameter µ. Setting v = u̇, we get the equivalent system of ODEs

u̇ = v, v̇ = −v + µu − u2 . (5.6)
 
0 1
At the origin, the matrix for the linear part is . The eigenvalues satisfies s(s + 1) −
µ −1
µ = 0, so there is a eigenvalue with real part zero, if and only if s = 0, or µ = 0. Therefore
we expect a bifurcation at the origin if µ = 0.
Rough Explanation of what happens for µ small : The stationary points are governed by
v = 0, µu − u2 = 0. That is u = 0, or u = µ, and we expect a transcritical bifurcation
(exchange of stability).
The general procedure: (a) Transform to normal form (including in the µ̇ = 0 equation);
(b) Expand extended CM; (c) Calculate dynamics on CM.
(a). Transformation: The linear part (5.6) at the origin if µ = 0 is not in normal
(diagonal) form. We have the eigenpairs,
   
1 1
λ1 = 0, e1 = , λ2 = −1, e2 = .
0 −1

Hence the change of coordinate uses the matrix of eigenvectors and the NEW coordinates
x, y are defined by
            
u 1 1 x x −1 −1 u 1 1 u
= or =− =
v 0 −1 y y 0 1 v 0 −1 v

Hence the coordinate transform is

x = u + v, y = −v or u = x + y, v = −y.

In terms of these new coordinates, the system (5.6) becomes

ẋ = u̇ + v̇ = v − v + µu − u2 = µ(x + y) − (x + y)2 ,
ẏ = −v̇ = v − µu + u2 = −y − µ(x + y) + (x + y)2 .

(b). Extended Centre Manifold Now the extended system in the new coordinates is

ẋ = µ(x + y) − (x + y)2 , ẏ = −y − µ(x + y) + (x + y)2, µ̇ = 0.

The extended centre manifold should be tangential to the (x, µ) plane (or y = 0) at
(x, y, µ) = (0, 0, 0), and is parametrised by

y = h(x, µ) = ax2 + bxµ + cµ2 + h.o.t.

68
5.3 Extended centre manifold

From the ẏ equation

ẏ = −(ax2 + bxµ + cµ2 + . . . ) − µ(x + . . . ) + x2 + · · · = (1 − a)x2 − (b + 1)xµ + · · ·

From the definition of the extended centre manifold

∂h ∂h
ẏ = ẋ + µ̇ = (2ax + bµ)ẋ = · · · ,
∂x ∂µ

where all the terms are at least cubic. So equating coefficients of the quadratic terms (of
which there are none in the second equation!) gives a = 1, b = −1, and the extended centre
manifold is
y = x2 − xµ + . . . .

(c). Dynamics on the centre manifold. Locally on the extended Centre Manifold µ̇ = 0
is trivial so it is the ẋ equation that is interesting:

ẋ = µ(x + x2 − µx + . . . ) − (x2 + 2x(x2 − µx) + . . . )

= µx − x2 + O(x3 )

Substituting back into the equation for ẋ we get (to leading order)

ẋ = µx − x2 +O(x3 ).
| {z }
Standard Form for transcritical

The phase portrait for the reduced dynamics for x is shown in Figure 5.6 and the phase
portrait for the original system is in Figure 5.7.

µ<0 x

µ
µ<0

µ<0

Figure 5.6: Phase portraits on the (one-dimensional) centre manifold and the bifurcation
diagram.

Remark. If the stable manifold is of higher dimension, then y1 = h1 (x, µ), y2 = h2 (x, µ) and
we need to find h1 , h2 using the same method. For example, for the system

ẋ = µx − yz, ẏ = −y + x2 , ż = −z + x3 .

Add µ̇ = 0 to this system, then the stable manifold expanded by y and z is parameterised
by x and µ, that is

y = h1 (x, µ) = a1 x2 + a2 xµ + a3 µ2 + · · · , z = h2 (x, µ) = b1 x2 + b2 xµ + b3 µ2 + · · · .

69
5.4 Classifications of bifurcations

Figure 5.7: Full phase portraits of the dynamics in µ < 0 and µ > 0.

Then a1 = 1, a2 = a3 = b1 = b2 = b3 = 0. That is y = x2 + · · · , but we have to go to

cubic polynomials to find the stable manifold for z, which gives z = x3 + · · · . Therefore, the
reduced dynamics on the stable manifold is

ẋ = fµ (x) = µx − x5 .

If µ < 0, x = 0 is the only stable fixed point. If µ > 0, there are three fixed point 0, µ1/4
and −µ1/4 . Since

fµ′ (0) = µ > 0, fµ′ (±µ1/4 ) = µ − 5(±µ1/4 )4 = −4µ < 0,

the fixed point 0 is unstable, and the fixed points ±µ1/4 are stable.

5.4 Classifications of bifurcations

Suppose x = 0 ∈ Rn is a stationary point of the system of ODEs ẋ = f (x, µ) if µ = 0,
and Dx f (0, 0) has a single zero eigenvalue. (If the stationary point is x∗ at µ∗ , then we
simply work in shifted coordinates x − x∗ and µ − µ∗ ). Now we consider the extended centre
manifold for the system governed by (x ∈ R)

ẋ = f (x, µ), µ̇ = 0

where f satisfies f (0, 0) = 0 and fx (0, 0) = 0. Consider the Taylor series expansion of f for
|x|, |µ| small:
1
ẋ = f (0, 0) + fx (0, 0)x + fµ (0, 0)µ + (fxx x2 + fµµ µ2 + 2fxµ xµ) + O(|x|3 , |µ|3)
2!
where all partial derivatives are evaluated at (0, 0).
By the assumption that f (0, 0) = 0 (x = 0 is the stationary point on the centre manifold
for µ = 0) and fx (0, 0) = 0 (there is a zero eigenvalue), the above Taylor series is simply
1

ẋ = fµ (0, 0)µ + fxx (0, 0)x2 + fµµ (0, 0)µ2 + 2fxµ (0, 0)xµ + · · · .
2
Different bifurcations could occur, depending on whether the partial derivatives vanish or
not.

70
5.4 Classifications of bifurcations

Saddle-node Bifurcation ẋ = µ − x2

If both fµ (0, 0) and fxx (0, 0) are non-zero, then

1 x2
ẋ = fµ (0, 0)µ + fxx (0, 0)x2 + O(|xµ|, |µ|2, · · · ) ≈ µfµ (0, 0) + fxx (0, 0).
2 2
The stationary points are s
2fµ (0, 0)
x∗± ≈ ± − µ (5.7)
fxx (0, 0)
if µfµ (0, 0)/fxx (0, 0) ≤ 0. So the stability is determined for sufficiently small |x| and |µ| by
the sign of fxx and fµ : there is no solution, if µfµ /fxx > 0, and there are two solutions given
by (5.7) if µfµ /fxx ≤ 0. Since
s
∂ 2fµ (0, 0)
f (x, µ) ≈ x∗± fxx (0, 0) = ± − µfxx (0, 0).
∂x x=x∗ fxx (0, 0)
±

Therefore, if µfµ /fxx > 0, x+ is stable, x− unstable if fxx < 0 and x− is stable, x+ unsta-
ble if fxx > 0 This is a saddle-node bifurcation, also called tangential bifurcation or fold
bifurcation.

Transcritical Bifurcation ẋ = µx − x2

If in addition to f (0, 0) = fx (0, 0) = 0, fµ (0, 0) is zero, but fxx (0, 0) 6= 0, the ODE equation
becomes
1 
ẋ ≈ fxx (0, 0)x2 + 2fxµ (0, 0)xµ + fµµ (0, 0)µ2
2
Then the possible stationary points are x∗± = k± µ, where
p 2
−fxµ ± fxµ − fxx fµµ
k± = .
fxx
2
So if fxµ − fxx fµµ > 0, there are two branches of solutions which intersect at the bifurcation
point (0, 0). This is a transcritical bifurcation. Stability is determined by looking at the
leading order terms of the derivative fx (x, µ) and a relatively simple manipulation shows
that one branch is stable and the other is unstable, with stability being exchanged as µ
passes through zero. To show the stability,

∂ q
f (x, µ) = fxx (0, 0)x∗± + fxµ (0, 0)µ = ±µ 2 −f f .
fxµ xx µµ
∂x x=x∗±

So the fixed point x+ is stable if µ < 0 and unstable if µ > 0; x− has the opposite stability
property.

71
5.4 Classifications of bifurcations

Pitchfork Bifurcation ẋ = µx − x3
If fµ (0, 0) = fxx (0, 0) = 0 then
1
1

ẋ ≈ fµµ µ2 + 2fxµ xµ + fxxx x3 + fµµµ µ3 + . . . . (5.8)
2 6
fµµ
If fxµ 6= 0, there is one branch of solutions with x ≈ − 2f xµ
µ. However there is a second set
of solutions by balancing the second term fxµ xµ and the third terms fxxx x3 :
1
fxµ xµ + fxxx x3 = 0
6
from which, provided fxxx 6= 0,
6fxµ
x2 = − µ (5.9)
fxxx
giving two new solutions in whichever sign of µ makes the right hand side positive. There
are no other ways of balancing leading order terms (by posing x ∼ µα ) so these are the only
bifurcating solutions. Since
∂ 1
f (x, µ) = fxµ µ + fxxx x2 + · · · , (5.10)
∂x 2
fµµ
we see that the solution x ≈ − 2f xµ
µ is stable (locally) if fxµ µ < 0 and unstable if fxµ µ > 0.
So the sign of fxµ determines on which side of µ = 0 this branch is stable.
The stability of second set of solutions is determined by substituting (5.9) into (5.10)
giving −2fxµ µ and so the stability is the opposite of the simple branch described above.
This is called a pitchfork bifurcation: if the non-trivial branch is stable it is called a
supercritical pitchfork bifurcation and if the non-trivial branch is unstable it is called a
subcritical pitchfork bifurcation, as shown in Figure 5.8.

Supercritical Pitchfork Bifurcation ẋ = µx − x3 Subcritical Pitchfork Bifurcation ẋ = −µx + x3

Figure 5.8: Two types of Pitchfork Bifurcation

Remark. It should be noted that the classification of bifurcation is based on the behaviour
near the bifurcation point: in saddle-node bifurcation, the number of fixed points is from
zero to two, one stable and one unstable; in transcritical bifurcation, two fixed points always
exist and exchange stability; in pitchfork bifurcation, the number of fixed points changes
from one to three, and the stability is exchanged.

72
5.5 Hopf bifurcations

5.5 Hopf bifurcations

The only ‘typical’ case not treated in the previous section is the appearance of purely imag-
inary (simple) eigenvalues ±iω, instead of zero eigenvalues. Usually, there is only one zero
eigenvalue (as seen in all examples in the previous section), and the centre manifold is only
one dimension. But in the simplest case with bifurcation with two purely imaginary eigen-
values, the centre manifold at µ = 0 is two dimensional and the extended centre manifold is
three dimensional. The equations governing the bifurcations could very complicated, but a
standard example will be enough for our purposes.
Consider the canonical example
ẋ = µx − ωy − x(x2 + y 2 ), ẏ = ωx + µy − y(x2 + y 2)
p
or in polar coordinates r = x2 + y 2, θ = arctan y/x
ṙ = µr − r 3 , θ̇ = ω. (5.11)
The linearisation about the origin is
 
µ −ω
ω µ
with eigenvalues µ ± iω. So if µ < 0, the origin is stable and if µ > 0 it is unstable. Observe
√
that the ṙ equation implies that if µ > 0, r → µ, i.e. there is a periodic orbit of radius
√
µ, which is stable (sketch the right hand side of the ṙ equation if this is not obvious).

Figure 5.9: Hopf bifurcation: when µ increases, the stable focus becomes unstable, and a
periodic solution called limit cycle appears.

This is an example of a Hopf bifurcation, also known as Poincaré-Andronov-Hopf bifurca-

tion. As the parameter is varied, a stationary point changes its stability and a periodic orbit
is created with the opposite stability (like a pitchfork bifurcation in r). If the bifurcating
periodic orbit is stable, then this is a supercritical Hopf bifurcation and if it is unstable this
is a subcritical Hopf bifurcation. The above canonical example can also be written with com-
plex numbers. That is, if we define z(t) = x(t) + iy(t), then the above system is equivalent
to
d
z = (µ + iω)z − z|z|2 .
dt

73
5.5 Hopf bifurcations

Remark. If only the first equation in (5.11) in the radial variable r is taken, a pitchfork
bifurcation happens at µ = 0. This connection (at least qualitatitively) between Hopf
bifurcation (for two variables) and pitchfork bifurcation (in radial like variable) is true in
general.
Example 5.5 (Brusselator model for autocatalytic reaction). Consider the system of ODEs

ẋ = a − (b + 1)x + x2 y, ẏ = bx − x2 y,

where a and b are two positive parameters. The unique steady state is (a, b/a) with the
Jacobian  
b − 1 a2
J= .
−b −a2
Since the determinant of J is a2 , the only possible bifurcation Hopf bifurcation, which occurs
only when the trace is zero. That is when b∗ = 1 + a2 . For b < b∗ , the steady state (a, b/a)
is stable; for b > b∗ , it is unstable and a new periodic solution (a limit cycle) appears.
Example 5.6. We considered in Example 4.3 the following model

ẋ = −x + ay + x2 y, ẏ = b − ay − x2 y

for glycolysis oscillation with b = 1/2 and a > 0. For general b, the only stationary point is
b
x∗ = b, y∗ = .
a + b2
We can find the conditions for possible bifurcations. From the Jacobian matrix
 
−1 + 2xy a + x2
J(x, y) =
−2xy −a − x2 ,

we have !
b2 −a
a + b2
J(x∗ , y ∗) = b2 +a
2b2
− a+b 2 −a − b2
Since detJ(x∗ , y ∗ ) = a + b2 > 0, the only possible bifurcation is the real parts of the complex
eigenvalues pass zero (real eigenvalues can not pass zero, which leads to zero determinant),
or Hopf bifurcation. This happens when the trace of J(x∗ , y ∗ ) is zero, that is,
b2 − a − a2 − 2ab2 − b4
trJ(x∗ , y ∗) = .
a + b2
If trJ(x∗ , y ∗ ) < 0, the fixed point (x∗ , y ∗ ) is stable; otherwise it is not stable.

Example 5.7. Consider the following system

ẋ = 1 − y 2, ẏ = −x − µy − y 2, (5.12)

for µ ≥ 0.
 The fixed points
 are (µ − 1, −1) and (−µ − 1, 1) with the Jacobian matrix is J =
0 −2y
.
−1 −µ − 2y

74
5.5 Hopf bifurcations

Figure 5.10: The phase portrait of the system (5.12) for µ = 1.8 (left figure) and µ = 2.2
(right right). As µ passes 2, the periodic solution (or the limit cycle) disappear, and the
unstable focus at (µ − 1, −1) becomes a stable focus. (Can you add arrows to indicate the
direction of the trajectories, based on the local behaviours of the stationary points?)
 
0 −2
At the fixed point (−µ − 1, 1), J = . Since detJ = −2 < 0, the two
−1 −µ − 2
eigenvalues have opposite signs, and this is always a saddle point.
 
0 2
At the fixed point (µ − 1, −1), J = and the eigenvalues are the roots of
−1 −µ + 2

λ2 + (µ − 2)λ + 2 = 0,
√
2−µ± µ2 −4µ−4
or λ± = . Since λ+ λ− = 2 > 0, the real parts of the eigenvalues pass zero if
2 √
and only if µ passes 2. At µ = 2, the two eigenvalues are ± 2i, this √ is a unstable focus
becomes a stable focus. When µ continues to increase beyond 1 + 2 2, the discriminant
∆ = µ2 − 4µ − 4 becomes positive, and the two eigenvalues are both real and negative.
Therefore, the stable focus becomes a stable node, as shown in Figure 5.10.

Figure 5.11: The global bifurcation at about µ∗ = 1.63, where there is a homoclinic orbit at
(−µ − 1, 1).

75
5.5 Hopf bifurcations

Remark. There is actually another bifurcation at about µ∗ = 1.63, with a homoclinic orbit
at the fixed point (−µ − 1, 1): the unstable manifold W u coincides with the stable manifold
W s through this fixed point. This kind of global bifurcation is much more difficult to study,
where the critical parameter µ∗ can not be determined as the local bifurcation in the previous
few examples.

76