Solutions 7

8.1.1
a) The stable fixed points are (0, 0) for µ ≤ 0 and (µ, 0) for µ > 0. The eigenvalues are
−|µ|, −1. Fixed points exchange stability through the bifurcation.
x ’ = m x − x2 m=−1 x ’ = m x − x2 m=1
y’=−y y’=−y

3 3

2 2

1 1

0 0
y

y
−1 −1

−2 −2

−3 −3

−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
x x

b) For µ < 0, the origin (0, 0) is a stable fixed point; the other two (unstable) fixed
√
points are (± −µ, 0). For µ > 0 the only fixed point is the origin, which is unstable. The
√
eigenvalues at (0, 0) are µ, −1, and the eigenvalues at (± −µ, 0) are −2µ, −1. The origin
loses stability through the bifurcation.
3
x ’ = mu x + x3 mu = 1
x’=mx+x m=−1 y’=−y
y’=−y

3
3

2
2

1
1

0
y

0
y

−1 −1

−2 −2

−3 −3

−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
x x

1
2

8.1.6
a) The nullclines are drawn for µ = −2, 0, and 2.
5 y

1
2
y = x2x −2
0
x

−1

−2

−3

−4

−5
−5 −4 −3 −2 −1 0 1 2 3 4 5

b) The fixed points are the intersections of the nullclines. From the above picture it’s
clear that as µ increases, two fixed points first collide then disappear. The collision occurs
when the parabola y = x2 + µ is tangential to the line y = 2x. Since the slope of the
tangent line of the parabola at point (t, t2 + µ) is 2t, if the tangent line is y = 2x, then
t = 1. Therefore y = x2 + µ is tangential to y = 2x at point (1, 1 + µ), which implies that
bifurcation occurs at µ = 1. The motion of the fixed points described above also shows that
this is a saddle-node bifurcation.
√
c) For µ < 1 we have two fixed points with x∗ = 1 ± 1 − µ, for µ > 1 there are no fixed
points. Sketched below are phase portraits for µ = 0 < 1 and µ = 2 > 1. On the left, the
origin is a stable node, and (2, 4) is a saddle.
x’=y−2x mu = 0 x’=y−2x mu = 2
y ’ = mu + x2 − y y ’ = mu + x2 − y

10
8

5
4

2
y

0
0

−2

−4
−5

−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
x x
 3

8.2.3
We start with a negative µ very close to 0. The figure below shows the phase portrait
for µ = −0.04, there is an unstable cycle around the origin:
1.5 y

0.5

0
x

−0.5

−1

−1.5
−1 −0.5 0 0.5 1 1.5

As µ increases to −0.02, the cycle gets closer to the origin:

1.5 y

0.5

0
x

−0.5

−1

−1.5
−1 −0.5 0 0.5 1 1.5

Now if µ becomes positive, the unstable limit cycle disappears and the origin becomes
unstable:
1.5 y

0.5

0
x

−0.5

−1

−1.5
−1 −0.5 0 0.5 1 1.5

Therefore our computer experiment suggests that this Hopf bifurcation is subcritical.
4

8.2.6
When µ is negative, say µ = −0.04, the phase portrait looks like

1.5 y

0.5

0
x

−0.5

−1

−1.5
−1 −0.5 0 0.5 1 1.5

The origin is a stable spiral. Now let µ = 0.4 be a small positive number, then a stable
limit cycle appears, and the origin becomes an unstable spiral:

1.5 y

0.5

0
x

−0.5

−1

−1.5
−1 −0.5 0 0.5 1 1.5

This experiment suggests that the Hopf bifurcation is supercritical.

8.2.8 a,c) We show below the nullclines in the first quadrant (left) and the phase portrait
for a = 1.2 (right). We can see that all the predators go extinct (y → 0).
 5

x ’ = x (x (1 − x) − y) a = 1.2 x ’ = x (x (1 − x) − y) a = 1.5
y ’ = y (x − a) y ’ = y (x − a)

2 4

1.8
3.5

1.6
3
1.4

2.5
1.2

1 2
y

y
0.8
1.5

0.6
1
0.4

0.5
0.2

0 0

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.5 1 1.5 2 2.5 3 3.5 4
x x

! !
2x − 3x2 − y −x 0 0
b) The Jacobian A = evaluates to for the origin.
y x−a 0 −a
The eigenvalues are 0, −a and it is a non-isolated fixed point.
!
−1 −1
A evaluates to for (1, 0). The eigenvalues are −1, 1 − a. It is a stable
0 1−a
node for a > 1 and a saddle for a < 1. !
√
a − 2a2 −a 2 ). The eigenvalues are a(1−2a± 4a2 −3)
Finally, A evaluates to for (a, a−a 2
a − a2 0
√
3
and the fixed point
√
is an unstable spiral when 0 ≤ a < 12 , a stable spiral when 1
2 <a< 2 ,
3
a stable node for 2 < a < 1 and a saddle when a > 1.
d) A supercritical Hopf bifurcation occurs at ac = 12 , when the real parts of the eigenvalues
simultaneously cross the imaginary axis into the right half plane. See the phase portraits
in part f).
√
1 3−4a2 1
e) Near ac = 2 , the frequency of the limited cycle oscillation is about ac 2 c = 2√ 2
.
f) We show the phase portraits for a = 0.4 (left), a = 0.6 (middle), and a = 0.9 (right).
The qualitatively different phase portraits correspond to the cases when the fixed point
(a, a − a2 ) is (a) an unstable spiral surrounded by a stable limit cycle, (b) a stable spiral,
and (c) a stable node.
6

x ’ = x (x (1 − x) − y) a = 0.9
x ’ = x (x (1 − x) − y) a = 0.4 x ’ = x (x (1 − x) − y) a = 0.6 y ’ = y (x − a)
y ’ = y (x − a) y ’ = y (x − a)

2 2
0.6

1.8 1.8
0.5
1.6 1.6

1.4 1.4
0.4

1.2 1.2

0.3

y
1 1
y

y
0.8 0.8
0.2
0.6 0.6

0.4 0.4
0.1

0.2 0.2

0 0 0

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.5 1 1.5
x x x

8.4.2
The radial and angular dynamics are uncoupled and so can be analyzed separately. Treat-
ing ṙ = r(µ − sin r) as a vector field on the line, we see that for |µ| > 1, r ∗ = 0 is the only
fixed point and it is stable for µ < −1 and unstable for µ > 1. At µ = −1, the sys-
tem undergoes a saddle node bifurcation of cycles at r ∗ = (2n + 23 )π. For −1 < µ <
0, r ∗ = 0, (2n − 1)π − arcsin µ, 2nπ + arcsin µ. The stabilities of fixed points alternate
with increasing r ∗ , starting with r ∗ = 0 being stable. At µ = 0 the zero fixed point
undergoes a supercritical Hopf bifurcation and a new cycle r ∗ = arcsin µ emerges. For
0 < µ < 1, r ∗ = 0, arcsin µ, (2n − 1)π − arcsin µ, 2nπ + arcsin µ. The stabilities of fixed
points alternate with increasing r ∗ , starting with r ∗ = 0 being unstable. Subsequently,
the system undergoes another saddle node bifurcation of cycles when µ = 1 in which pairs
of adjacent fixed points [arcsin µ, π − arcsin µ], [2π + arcsin µ, 3π − arcsin µ] . . . collide at
π π
2 , 2π + 2 , . . .. Since the motion in the θ-direction is simply rotation at constant angular
velocity, we see that all trajectories spiral asymptotically to or from limit circles at r = r ∗ .

9.2.1
a) Recall that the fixed points C + and C − are
p p
(x∗ , y ∗ , z ∗ ) = (± b(r − 1), ± b(r − 1), r − 1),

where r > 1. In the following we’ll write (x, y, z) instead of (x∗ , y ∗ , z ∗ ).

 
−σ σ 0
The Jacobian A =  r − z −1 −x  has characteristic polynomial
 

y x −b

det(λI − A) = λ3 + (σ + 1 + b)λ2 +
[b(σ + 1) + x2 + σ(z + 1 − r)]λ + [bσ(z + 1 − r) + σ(x2 + xy)].
 7

Notice that at C + and C − , we have x2 = xy = b(r − 1), z = r − 1, therefore the eigenvalues

satisfy
λ3 + (σ + 1 + b)λ2 + b(r + σ)λ + 2σb(r − 1) = 0.

b) Hopf bifurcation occurs when two eigenvalues are pure imaginary (cf. Figure 8.2.4).
In this case λ = iω, where ω is real and nonzero. Thus the characteristic equation becomes

−iω 3 − (σ + 1 + b)ω 2 + ib(r + σ)ω + 2σb(r − 1) = 0.

We then seperate the real part and the imaginary part of the above and obtain that
2σb(r − 1)
ω 2 = b(r + σ) = .
σ+1+b
Solving for r, we get
σ+b+3
r = rH = σ( ).
σ−b−1
It is required that r must be positive, so Hopf bifurcation can only occur if rH is positive,
i.e. σ > b + 1.
c) If r = rH , then the two imaginary roots are
p
λ1,2 = ±i b(rH + σ).

It’s well known that all 3 roots should add up to −(σ + b + 1), since the two imaginary ones
cancel, we have λ3 = −(σ + b + 1).

9.2.2
Let C(t) = rx(t)2 + σy(t)2 + σ(z(t) − 2r)2 be the value of C at time t. Then

C ′ (t) = 2rxx′ + 2σyy ′ + 2σ(z − 2r)z ′

= 2rxσ(y − x) + 2σy(rx − y − xz) + 2σ(z − 2r)(xy − bz)
= −2σ(rx2 + y 2 + bz 2 − 2brz)
= −2σ(rx2 + y 2 + b(z − r)2 − r 2 b)
2σ x2 y2 (z − r)2
=− ( + + − 1).
r 2 b br br 2 r2
It is then clear that if at time t, (x, y, z) is outside the ellipsoid
x2 y2 (z − r)2
K: + 2+ ≤ 1,
br br r2
then C(t) will decrease. We can pick C so large that the above ellipsoid is contained in

E: rx2 + σy 2 + σ(z − 2r)2 ≤ C,

which is another ellipsoid, then eventually all trajectories will enter E.

8

The smallest possible value of C is obtained when the ellipsoids E and K are tangent.
The picture is that one shrinks E by decreasing C until the surface of E touches K. This
is equivalent to the following problem:
Given condition
x2 y2 (z − r)2
+ 2+ = 1,
br br r2
find the maximum of
rx2 + σy 2 + σ(z − 2r)2 .
The maximum is the smallest possible C. This problem can be solved using Langrange
multipliers. In practice, given a pair of parameters r and σ, one can use a computer to
handle the extreme value problem.

9.2.6
a) Let f = (−νx + zy, −νy + (z − a)x, 1 − xy) be the instantaneous velocity, then
∂ ∂ ∂
∇·f = (−νx + zy) + (−νy + (z − a)x) + (1 − xy)
∂x ∂y ∂z
= −ν − ν + 0 = −2ν < 0,

therefore the system is dissipative (cf. Figure 9.2.1).

b) A fixed points (x, y, z) satisfies that

(1) zy = νx
(2) (z − a)x = νy
(3) 1 = xy.

From (1) and (3) we have z = νx2 , then use (2)

(νx2 − a)x2 = νxy = ν.

Since xy = 1, x 6= 0, so
a = νx2 − νx−2 = ν(k2 − k−2 ),
where x2 = k2 , xy = 1, and z = νx2 . This is exactly the parametric form described in the
problem.
   
−ν z y −ν νk 2 k−1
c) The Jacobian A =  z − a −ν x  evaluates to  νk2 − a −ν k  at fixed
   

−y −x 0 −k−1 −k 0
point (k, k−1 , νk2 ). The characteristic polynomial is

f (λ) = λ3 + 2νλ2 + [ν 2 − νk 2 (νk2 − a) + k2 + k−2 )λ + 2ν(k2 + k−2 ).

 9

Using the relation a = ν(k2 − k−2 ) to eliminate a, it follows that

f (λ) = λ3 + 2νλ2 + (k2 + k−2 )λ + 2ν(k2 + k−2 ) = (λ + 2ν)(λ2 + k2 + k−2 ).

√
Therefore the eigenvalues are −2ν and ± k2 + k−2 i. Since there are two imaginary eigen-
values, the fixed points are centers.