MATH 467-3 Fall 2003

Dynamical Systems

Homework Set 3 Some Solutions

Course Web Site: http://www.math.sfu.ca/∼ralfw/math467/

1. 3.3.1

ṅ = GnN − kn (1)
Ṅ = −GnN − f N + p (2)

is a model for a laser, more realistic than that treated in class; N is the number of excited atoms,
n the number of laser photons, and we let the bifurcation parameter be the pump strength p. We
simplify this to a one-dimensional system by the quasistatic approximation:
(a) Assuming Ṅ ≈ 0, we find N (t) ≈ p/(Gn(t) + f ); substituting, we derive the first-order system
for n(t),
Gpn
ṅ = − kn ≡ h(n).
Gn + f
(b) The fixed points (found from ṅ = 0) are given by n∗ = 0 and n∗ = (Gp − kf )/kG. The
stability of the fixed point n∗ = 0 is given from


 
∂h

Gpf


Gpf Gp
= − k = − k = − 1 k.
∂n
n=0 (Gn + f )2

n=0 f2 kf

Hence n∗ = 0 becomes unstable when p > pc = kf /G.

(c) We expect a transcritical bifurcation, since n∗ = 0 is always a fixed point, and there is no
symmetry; also, the two fixed points coincide when p = pc .
Check by expanding near n∗ = 0:
 −1
−1 Gpn Gn
h(n) = Gpn(f + Gn) − kn = 1+ − kn
f f
 
Gpn G G2
= 1 − n + 2 n2 − . . . − kn
f f f
 
Gp G2 p 2
= − k n − 2 n + O(n3 ),
f f

which yields the transcritical bifurcation normal form near n∗ = 0, and the bifurcation value
pc = kf /G.
(d) A typical time scale for decay of n is 1/k, for decay of N is 1/f . We expect that the adiabatic
approximation is valid if N equilibrates rapidly to changes in n, that is, if N relaxes over much
shorter time scales (or decays much more quickly), so we require 1/f  1/k, or f  k. This
can be confirmed by a careful nondimensionalization.
2. 3.4.8
x
ẋ = rx − ≡ f (x, r)
1 + x2
(symmetric under the map x → −x).
∗ 2
 x = 0 exists for all r. There are nontrivial fixed points satisfying r −1/(1+x ) = 0,
The fixed point
∗
or x = ± 1/r − 1; these exist provided 1/r − 1 > 0, or 0 < r < 1.
 We determine the stability of the fixed point at the origin by computing



∂f

1 − x2


= r− = r − 1,
∂x
x=0 (1 + x2 )2
x=0

indicating that x∗ = 0 is linearly stable for r < 1 and unstable for r > 1, so we expect a subcritical
pitchfork bifurcation at r = 1, x = 0. We can verify this by expanding near the bifurcation point
x = 0, r = 1, and comparing with the normal form:

f (x, r) = rx − x(1 + x2 )−1 = (r − 1)x + x3 + O(x5 ),

confirming the presence of a subcritical pitchfork bifurcation at r = 1, x = 0.

10

0
x

−5

−10
−0.5 0 0.5 1 1.5
r

Figure 1: Bifurcation diagram r = 1/(1 + x2 ), showing the subcritical pitchfork bifurcation; solid lines
indicate stable branches, dashed curves are unstable.

What happens to the (unstable) nontrivial fixed points x∗ = ± 1/r − 1 at r = 0 (they collide with
x∗ = 0 at r = 1)? Note that these fixed points satisfy |x∗ | → ∞ as r → 0; and in fact they collide
with the “point at infinity”:
To study the behaviour near infinity, set y = 1/x, then |x| → ∞ corresponds to y → 0. We find the
differential equation for y using ẏ = −ẋ/x2 , and substituting x = 1/y; after some algebra we find
y
ẏ = (1 − r)y − ;
1 + y2

expanding near y = 0 (that is, near the “point at infinity”),

ẏ = −ry + y 3 + O(y 5 ).

Thus y = 0 is unstable for r < 0, stable for r > 0; as r decreases through 0, y = 0 undergoes
 a
∗
subcritical pitchfork
 bifurcation, so for r > 0, there are two unstable fixed points y = ± r/(1 − r)
(that is, x∗ = ± (1 − r)/r), which bifurcate from y = 0, that is, from the “point at infinity” in
the original dynamical system.
3. 3.4.11

ẋ = rx − sin x
(a) For r = 0, we have ẋ = − sin x, with fixed points at x = nπ, n = 0, ±1, ±2, . . . . The fixed
points with n even are stable, n odd are unstable.
(b) When r > 1, the unique fixed point is an unstable fixed point at x = 0.
 1

0.5

−0.5

−1

−15 −10 −5 0 5 10 15

Figure 2: Vector field for r = 0; solid circles indicate stable fixed points, open circles show unstable
equilibria.

−1

−2
−15 −10 −5 0 5 10 15

Figure 3: Plot of rx and sin x for r = 2, showing that for r > 1 there is a unique fixed point.

(c) The easiest way to study this problem is to look at intersections of the functions f (x) = rx
and f (x) = sin x, which yield the locations of the fixed points. As r decreases through 1,
the origin stabilizes in a subcritical pitchfork bifurcation, and two new unstable fixed points
are created. As r decreases further, tangencies between rx and sin x occur successively at the
second, third, fourth, . . . peak of sin x; each of these is a saddle-node bifurcation generating a
stable (for smaller |x|) and unstable (larger |x|) fixed point. Due to the symmetry x → −x,
each bifurcation occuring for positive x values is twinned to a simultaneous bifurcation at
negative x values. In summary, there are infinitely many pairs of saddle-node bifurcations as
r decreases from ∞ to 0.

−1

−2
−15 −10 −5 0 5 10 15

Figure 4: Plot of rx and sin x for successively decreasing values of r; as the straight line becomes
tangent to and then intersects successive peaks of the sine function, there are infinitely many saddle-node
bifurcations (in pairs).

(d) The nth bifurcation (n > 1) involves the tangency of rx with the nth positive peak of the sine
function, which is centered at x = (2n − 3/2)π (for x > 0; with a simultaneous bifurcation for
x < 0). For small r, or large n, the line rx has slope near 0, and the tangency occurs near
the maximum of the peak, that is, near xn ≈ (2n − 3/2)π, where sin x ≈ 1. Thus the nth
bifurcation value of r satisfies, for large n, rn xn = sin xn ≈ 1, so rn ≈ 1/xn = 1/(2n − 3/2)π.
 1.5

0.5

−0.5

−1

−1.5
−30 −20 −10 0 10 20 30

Figure 5: Plot of rx and sin x for r = r5 = 1/(2 · 5 − 3/2)π, showing that this is the (approximate) 5th
bifurcation value of r.

(e) As decreases from 0 to −∞, pairs of fixed points collide in (infinitely many) saddle-node
bifurcations and are annihilated, until for sufficiently negative r (less than about -0.22) the
only remaining fixed point is the stable fixed point at x = 0.

−1

−2
−15 −10 −5 0 5 10 15

Figure 6: Plot of rx and sin x for negative and decreasing values of r; pairs of fixed points are successively
annihilated in saddle-node bifurcations.

(f) The bifurcation diagram is given by the curves x = 0 and r = sin x/x.

20

10

0
x

−10

−20
−0.5 0 0.5 1 1.5
r

Figure 7: Bifurcation diagram r = sin x/x; solid lines indicate stable branches, dashed curves are unstable.

4. 3.4.12 Quadfurcation
A saddle-node bifurcation has the normal form ẋ = r − x2 : no fixed points for r < 0, two fixed
points for r > 0, bifurcating from x = 0.
A pitchfork “trifurcation”, with ẋ = rx − x3 = x(r − x2 ), has one branch of fixed points for r < 0,
three branches for r > 0, bifurcating from x = 0.
One possibility to construct a “quadfurcation” is to have multiple saddle-node bifurcations occuring
simultaneously: for example, for

ẋ = r − (x − 1)2 (x + 1)2 ,
 for r < 0 there are no fixed points, while for r > 0 there are four branches of fixed points; at r = 0,
saddle-node bifurcations occur simultaneously at x = +1 and x = −1.
We can extend this idea to arbitrarily many branches of fixed points: for n = 1, 2, 3, . . . , let x1 ,
x2 , . . . , xn be n distinct real numbers. Then the (even-order) dynamical system

ẋ = r − (x − x1 )2 (x − x2 )2 . . . (x − xn )2 = r − j = 1n (x − xj )2

has no fixed points for r < 0, and 2n fixed points for r > 0, born out of n simultaneous saddle-node
bifurcations at r = 0.
Similarly, the (odd-order) system
 
ẋ = x r − (x − x1 )2 (x − x2 )2 . . . (x − xn )2

has a fixed point at x = 0 for all r, and 2n new fixed points created (in saddle-node bifurcations)
as r increases through 0.
Here is another approach to obtaining a “quadfurcation”, and by extension, multiple branches:
Consider for instance
ẋ = −(r − x2 )(2r − x2 ),
√
which √has no fixed points for r < 0, and four branches of fixed points for r > 0, at x√= ± r and
x = ± 2r; all emerge√from x = 0 at r = 0. Note that the upper fixed point at x = + 2r is stable,
as is the point x = − r.
We can generalize this as follows: let a1 , a2 , . . . , an be n distinct positive real numbers, aj > 0;
without loss of generality, choose 0 < a1 < a2 < · · · < an . Then the dynamical system
  
ẋ = (−1)n+1 a1 r − x2 a2 r − x2 . . . an r − x2

has no fixed points for r < 0, and 2n fixed points for r > 0, all created in a bifurcation at r = 0,
√
x = 0; with the given choice of sign, the largest fixed point, at x = + an r, is stable for r > 0.
Similarly, the system
  
ẋ = (−1)n+1 x a1 r − x2 a2 r − x2 . . . an r − x2

has a unique (stable) fixed point at x = 0 for r < 0, and (2n + 1) fixed points for r > 0, born in
a generalized pitchfork bifurcation from x = 0, r = 0. If n is even, then x = 0 remains stable for
r > 0 beyond the bifurcation.
5. 3.4.14 Subcritical Pitchfork Bifurcation

ẋ = rx + x3 − x5
(a) The fixed points are x∗ = 0 and the solutions of r + x2 − x4 = 0 (an algebraic equation in x2 ),
with solutions

∗ 1 1
x =± ± r+
2 4
(there are four nontrivial fixed points corresponding to four choices of sign ±/±). When
r < −1/4, none of the four nontrivial fixed points exists; they are all created in a saddle-node
bifurcation at r = −1/4.
(b) See Figure 8, below.
(c) The (double) saddle-node bifurcation is at r = −1/4, as seen from the formula for the nontrivial
fixed points in a) above; or from writing the dynamical system in the equivalent form
 
3 5 1 2 1 2
ẋ = rx + x − x = x (r + ) − (x − ) ,
4 2
√
from which it is clear that the fixed points are born at x = ±1/ 2 when r = 1/4.
 r < −1/4 r = −1/4
0.3 0.3

0.2 0.2

0.1 0.1

dx/dt

dx/dt
0 0

−0.1 −0.1

−0.2 −0.2

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
x x

(a) (b)
−1/4 < r < 0 r=0
0.3 0.3

0.2 0.2

0.1 0.1
dx/dt

dx/dt
0 0

−0.1 −0.1

−0.2 −0.2

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
x x

(c) (d)
r>0
0.3 1.5

0.2 1

0.1 0.5
dx/dt

0 0

x
−0.1 −0.5

−0.2 −1

−1.5
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
x r

(e) (f)

Figure 8: (a)–(e): The vector field ẋ = rx − x3 + x5 and its fixed points (solid circle: stable; open circle:
unstable) for (a) r < −1/4 (one stable fixed√ point); (b) r = −1/4 (saddle-node bifurcations: x = 0 is
stable, the new fixed points at x = ±1/ 2 are semistable); (c) −1/4 < r < 0; (d) r = 0 (subcritical
pitchfork bifurcation); (e) r > 0 (one unstable and two stable fixed points). (f) Bifurcation diagram for
subcritical pitchfork bifurcation, with saddle-node bifurcations at r = −1/4 and the pitchfork bifurcation
at r = 0; solid curves denote stable branches, dashed curves are unstable.

6. Additional problem: Numerical Bifurcation Diagrams

The solution to (a) is given in the text and lecture notes. Following the description on the problem
sheet, here is a set of Matlab commands that produces the desired plots for (b):

% This file collects a sequence of commands to compute the bifurcation

% picture for the imperfect bifurcation problem of func3.

% Note that this is a set of commands that is obtained by some Matlab

% trial and error, and is a summary of the commands that worked... These
% commands shouldn’t be run as a single M-file; in particular, at some
% points the value of b in the file func3 was changed.

% First, following (more-or-less) the instructions in the problem set,

% let’s compute a nice bifurcation diagram for b = 0.1:
x = -1:0.02:1.5;
figure(1);
clf;
plot(x, func3(x,1), ’b’, [-1 1.5], [0 0], ’:’)
x1 = fzero(’func3(x,1)’, [-1 -0.5])
x2 = fzero(’func3(x,1)’, [-0.5 0.5])
x3 = fzero(’func3(x,1)’, [0.5 1.5])
% We find x1 = -0.9456, x2 = -0.1010, x3 = 1.0467
figure(2)
clf; hold on;
bifurcationplot(’vfield3’,x1,1,-0.9,4);
bifurcationplot(’vfield3’,x3,1,-0.8,3);
% Now save, print and annotate this figure
% Now compute similar bifurcation diagrams for b = 0 and b < 0.
% Change the 4th line of func3.m to read b= 0.0; (and save func3.m)
% We notice that we need a larger range of values of x:
x = -1.5:0.02:1.5;
plot(x,func3(x,1),’b’,[-1.5 1.5],[0 0], ’:’);
x1 = fzero(’func3(x,1)’, [-1 -0.5])
x2 = fzero(’func3(x,1)’, [-0.5 0.5])
x3 = fzero(’func3(x,1)’, [0.5 1.5])
% We find x1 = -1, x2 = 0, x3 = 1, which we could have written down
% immediately. However, we notice that for b = 0 this is a pitchfork
% bifurcation (which we also notice when we try to plot the bifurcation
% diagram), and the curves of fixed points cannot be integrated past the
% bifurcation point. Thus in order to obtain a branch for a < 0, we need
% to find a fixed point for negative a.
plot(x,func3(x,-1),’b’,[-1.5 1.5],[0 0], ’:’);
% This plot shows that there is only a single fixed point for a < 0, at
% x4 = 0, which we can find either analytically or numerically.
x4 = fzero(’func3(x,-1)’,[-1 1])

% Now plot the bifurcation diagram (the values of si and sf used here were
% obtained by trial and error to get the best graphs).
figure(2); clf; hold on;
bifurcationplot(’vfield3’,x1,1,-1,4);
bifurcationplot(’vfield3’,x2,1,-3,2);
bifurcationplot(’vfield3’,x3,1,-1,4);
bifurcationplot(’vfield3’,x4,-1,-4,1.2);
% Save and print...

% Lastly, set b = -0.1 in func3.m

x = -1.5:0.02:1;
plot(x,func3(x,1),’b’,[-1.5 1],[0 0], ’:’);
x1 = fzero(’func3(x,1)’, [-1 -0.5])
x2 = fzero(’func3(x,1)’, [-0.5 0.5])
x3 = fzero(’func3(x,1)’, [0.5 1.5])
% We find x1 = -1.0467, x2 = 0.1010, x3 = 0.9456 (symmetry with b=0.1)

figure(2); clf; hold on;

bifurcationplot(’vfield3’,x1,1,-0.8,3);
bifurcationplot(’vfield3’,x3,1,-0.9,4); % ... and save, print, ...
 Bifurcation curve : vfield3 b = 0.1
2.5

1.5

1 Stable

0.5

x
Unstable
saddle−node
−0.5 bifurcation

−1

−1.5
Stable

−2

−2.5
−2 −1 0 1 2 3 4 5 6
r

(a)
Bifurcation curve : vfield3 b = 0.0
3

2 Stable

1
pitchfork
bifurcation

Stable Unstable
0
x

−1

Stable

−2

−3
−4 −2 0 2 4 6 8
r

(b)
Bifurcation curve : vfield3 b = −0.1
2.5

2
Stable
1.5

saddle−node
0.5 bifurcation

Unstable
0
x

−0.5

Stable
−1

−1.5

−2

−2.5
−2 −1 0 1 2 3 4 5 6
r

(c)

Figure 9: Bifurcation diagrams for ẋ = b + ax − x3 giving the fixed points x∗ as functions of a for (a)
b = 0.1; (b) b = 0; (c) b = −0.1.