Dynamical systems 2019 kristian.gustafsson@physics.gu.

se

7 Two-dimensional bifurcations
As in one-dimensional systems: fixed points may be created, de-
stroyed, or change stability as parameters are varied (change of ’topo-
logical equivalence’). In addition closed orbits may undergo these
changes.

7.1 Saddle-node, transcritical, and pitchfork

bifurcations
Assume that a saddle point and an attracting node collide as a pa-
rameter µ is varied. The mechanism of why the collision occurs at
all (instead of the fixed points moving past each other): Fixed points
are formed at intersections of nullclines. As µ is varied, the nullclines
deform continuously. If they slip through each other the fixed points
collide:

Change coordinates to the local eigenframe of the saddle point. Let

the unstable direction of the saddle be v̂ u = (1, 0) and the stable
direction v̂ s = (0, 1). When the node comes closeby, it must merge
along the unstable manifold of the saddle [otherwise trajectories could
not remain continuous and linear as the fixed points merge].

1
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

• The bifurcation is essentially one-dimensional (in any dimen-

sion). Normal form (in unstable|stable directions of saddle):
ẋ = µ + x2 (same as 1D)
ẏ = −y

• Along the interconnecting manifold, the eigenvalues have oppo-

site signs ⇒ at bifurcation (at least) one eigenvalue must vanish.
• After bifurcation a slow region remains (ghost of fixed points)
[before bifurcation Tpass = ∞ (along interconnecting manifold),
√
after bifurcation Tpass decreases smoothly : Tpass ∼ 1/ µ − µc]
(see Strogatz Sec. 4.3 and Lecture 2).
• Repelling node? ⇒ reverse the arrows!
• The sum of all indices of the fixed points involved in a two-
dimensional bifurcation in a smooth flow must be conserved (as-
suming that no value of the bifurcation parameter gives rise to a
line of fixed points). After the saddle-node bifurcation no fixed
points remain and the index must be zero. ⇒ only fixed-points
with opposite signs may annihilate. Nodes, degenerate nodes,
spirals, centers, stars: I = +1, Saddles: I = −1 ⇒ bifurcations
where two fixed points merge and annihilate consist of one sad-
dle and one fixed point with I = +1. Moreover, since one of
the eigenvalues smoothly crosses zero at the bifurcation (⇔ ∆
crosses zero) the second fixed point is typically a node (unless
also τ passes zero), hence the name saddle-node bifurcation.

2
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

Similarly, the other bifurcations discussed in Lecture 2 (transcritical,

subcritical pitchfork, supercritical pitchfork), occur in one-dimensional
subspaces in higher-dimensional systems. Transversal directions are
simply attracting or repelling. The bifurcations are summarized in
the Table on the last page. The dynamics along the x-axis is that of
1D flows (x-component of flow plotted as black) and blue shows flow
in 2D. The bifurcation diagrams show that the index is preserved.

7.2 Hopf bifurcation

A stable fixed point has Re[λ1,2] < 0. A bifurcation to an unstable
fixed point occurs if the maximal eigenvalue crosses zero. Consider the
three possible bifurcations from stable to unstable in a linear system:
a b c

b c

Cases a and b have Im[λ1,2] = 0, while case c has Im[λ1,2] 6= 0. Case

a corresponds to saddle-node, transcritical, and pitchfork bifurcations
above. Case b is marginal and therefore not so interesting. Case c
is a Hopf bifurcation: a new type of bifurcation that does not exist in
1D systems. Consider the transition with Im[λ1,2] 6= 0:

3
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

Hopf bifurcations often lead to the formation of limit cycles. The

bifurcation can be either supercritical or subcritical.

7.2.1 Supercritical Hopf bifurcation

If a small-amplitude limit cycle is formed to ’catch’ the unstable tra-
jectories after the bifurcation, the bifurcation is supercritical.

Example

ṙ = r(µ − r2)
θ̇ = ω

Radial equation is on the form of a supercritical pitchfork.

When µ < 0 the radial equation has a stable fixed point at r∗ = 0
(stable spiral in x,y-space).
When µ > 0 the√origin becomes unstable spiral, but √ trajectories are
∗
caught at r = µ, i.e. limit cycle of radius r = µ. ⇒ stable
small-amplitude oscillations.

4
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

To verify that the eigenvalues cross Re[λi] = 0 with nonzero Im[λi]

we need to convert to cartesian coordinates x = r cos θ, y = r sin θ.
Some algebra gives
ẋ = µx − ωy + cubic terms
ẏ = ωx + µy + cubic terms .
Stability matrix at fixed point in origin
 
µ −ω
J=
ω µ
Eigenvalues λ = µ ± iω, become unstable as µ becomes positive.

7.2.2 Subcritical Hopf bifurcation

If no stable limit cycle is formed when the fixed point becomes un-
stable, trajectories must run away to a distant attractor: fixed point,
limit cycle, infinity (or strange attractor for d > 2).

Example
ṙ = µr + r3 − r5
(1)
θ̇ = ω
√
Non-negative zeroes at r0∗ = 0 for all µ, [r±∗ ]2 = (1 ± 1 + 4µ)/2
if −1/4 ≤ µ ≤ 0. When µ passes 0: r0∗, r−∗ and −r−∗ merge in a
subcritical pitchfork bifurcation. System settles in the distant limit

5
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

cycle at r = r+∗ .

This jump is similar to the catastrophes in lecture 3. The system

exhibits hysteresis, when µ becomes positive, the system jumps to the
distant attractor r = r+∗ . To go back to the original stable spiral, it
is not enough to reduce µ below zero, it must be reduced below the
saddle-node bifurcation at µ = −1/4. The bifurcation at µ = −1/4
is an example of a global bifurcation, to be discussed in Section 7.3.1.
Whether one obtains a stable limit cycle after a subcritical Hopf
bifurcation depends on the global properties of the flow. Before the
bifurcation the system always has an unstable limit cycle.

7.3 Global bifurcations

The bifurcations mentioned above are local, they happen locally as
fixed points collide or change stability. It is also possible to cre-
ate/destroy limit cycles in non-local regions of flow.

7.3.1 Bifurcations of cycles

Consider once again Eq. (1)
ṙ = µr + r3 − r5
θ̇ = ω
This time, consider the bifurcation as µ passes µc = −1/4. The
one-dimensional system for r undergoes a saddle-node bifurcation at
a non-zero value of r ⇒ limit cycles in the two-dimensional system

6
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

bifurcate:

Limit cycles are structurally stable, they typically remain after small
perturbations to the system. However, they can be destroyed|created
in collisions with other limit cycles (as shown here), or with a fixed
point (see Section 7.3.3 below). They can also be destroyed by the
formation of fixed points on the limit cycle.

7.3.2 Infinite-period bifurcation

Example: Saddle-node bifurcation on existing limit cycle.
ṙ = r(1 − r2)
θ̇ = µ − sin θ
Uncoupled equations.
r-equation ’usual’ equation for attracting limit cycle at r = 1.

• When µ > 1 we have a stable limit cycle with a bottleneck (slow

velocity) at θ = π/2.

7
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

• When µ = 1 a half-stable fixed point appears at (r, θ) = (1, π/2)

⇒ it takes an infinite time to pass θ = π/2 along the homoclinic
orbit (former limit cycle). At θ = π/2 the flow must be vertical
towards r = 1.
• When µ < 1 a saddle-node pair is formed, joined by heteroclinic
trajectories.

As shown in lecture 2 the dynamics is slow close to the saddle-node

p
bifurcation (the time scale along the limit cycles scales as 1/ |µ − 1|
for both sides of the bifurcation).
The scaling of the period with the control parameter µ is important
in order to investigate oscillating systems in numerical or real-life ex-
periments. Observing amplitude and period time as µ is varied allows
to identify or rule out what kind of system we have.
Another infinite-time bifurcation (with another scaling in period time,
T ∼ ln µ, see Problem set 2) is the bifurcation of homoclinic orbit.

7.3.3 Bifurcation of saddle connections

Bifurcation of homoclinic orbit
Consider a collision between a limit cycle and a saddle point to form

8
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

a homoclinic orbit.

ẋ = y
ẏ = µy + x − x2 + xy

Non-local bifurcation ⇒ need to use computer! The result for some

values of µ:

For µ < µrmc ≈ −0.8645 a saddle point and a limit cycle are isolated.
As µ is increased, the limit cycle expands until it eventually collides
with the saddle point at µ = µc, forming a homoclinic orbit.
When µ > µrmc the homoclinic orbit breaks.
An example is a Josephson junction which is equivalent to a forced
pendulum with friction (Lecture 9). The homoclinic bifurcation is
another example of an infinite-period bifurcation (the homoclinic orbit
has an infinite period time).
Bifurcation of heteroclinic trajectory
Note that the upper right unstable manifold in the bifurcation above
lies inside of stable manifold when µ < µc and lies outside when
µ > µc. This property changes the topology of the solution (see pur-
ple trajecotires), i.e. a bifurcation occurs as µ passes µc also in a

9
Dynamical systems 2019 kristian.gustafsson@physics.gu.se

homoclinic bifurcation without collision with a limit cycle (the homo-

clinic orbit instead surrounds a band of closed orbits). A similar but
simpler example is the following system that has a bifurcation of a
heteroclinic trajectory:

ẋ = µ + x2 − xy
ẏ = y 2 − x2 − 1

This system has two saddle √ points (det J = −2(x − y)2 < 0) at:
(x∗±, y±∗ ) = ±(µ, 1 − µ)/ 1 − 2µ. When µ = 0, they lie on the y-
axis, (x∗±, y±∗ ) = (0, ±1), and since ẋ = 0 along the y-axis, they must
be connected by a heteroclinic trajectory. When |µ| is small but
non-zero, the fixed points move to either side of the y-axis, (x∗±, y±∗ ) ≈
±(µ, 1). Since ẋ = µ along the y-axis, the connection between the
stable and unstable manifolds must be broken as seen in the following
phase portrait:

The topology changes after the bifurcation: purple trajectories shows

solutions in x from +∞ to −∞ for µ < µc = 0, while after the
bifurcation the system shows solutions from −∞ to +∞.

10
 Type saddle-node transcritical supercrit. pitchfork subcrit. pitchfork
Normal ẋ = µ + x2 ẋ = µx − x2 ẋ = µx − x3 ẋ = µx + x3
form ẏ = −y ẏ = −y ẏ = −y ẏ = −y

µ<0 Similar to super

crit.
Dynamical systems 2019

µ=0

11
µ>0
kristian.gustafsson@physics.gu.se

Diagram
Index