Dynamical Systems

Exercises and solutions

Franco Vivaldi
School of Mathematical Sciences

c The University of London, 2018 Last updated: February 8, 2018
2
Contents

1 Asymptotics 1

2 Periodic points 9

3 Stability of periodic orbits 15

4 Renormalisation 21

5 Homeomorphisms and diffeomorphisms 27

6 Doubling map 33

7 Conjugacy 39

8 Two-dimensional mappings 43

9 Fractals 47

3
4 CONTENTS
Chapter 1

Asymptotics

Let f and g be real functions. We write f (x) ≺ g(x) to mean that

f (x)
lim =0 (1)
x→∞ g(x)

and f (x) ∼ g(x) to mean that

f (x)
lim = 1. (2)
x→∞ g(x)

Exercise 1.

(a) Prove that if f (x) ≺ g(x), then f (x) = O(g(x)), as x → ∞.

[The big-O definition requires a constant; find it using the definition of the
limit (1).]

(b) Prove that the converse is not true, by giving a counterexample.

[Is f (x) = O( f (x))?]

Exercise 2. Prove that if g(x) → ∞, then

f (x) ≺ g(x) ⇒ e f (x) ≺ eg(x) .

Is the converse implication true?

1
2 CHAPTER 1. ASYMPTOTICS

Exercise 3. Let ε and c be real numbers with 0 < ε < 1 < c. Order the following
functions according to the ≺ relation
x
xc , cx , cc , xε , ε x, 1, xlog(x) , log(x), log(log(x)), log(x)c .

[Use Hôpital’s rule; take logarithms and then use problem 2.]

Exercise 4. Which function grows faster

xlog(x) or log(x)x ?

[How are these functions defined? Use problem 2.]

Exercise 5. Find an asymptotic expression for the function

x + log(xex ) x
f (x) = 3
+ p
1 + log(x ) log(x) log(x)

as x → ∞. Namely, find a function g simpler than f , such that f ∼ g.

[Identify the dominant term of each numerator and denominator, than that of each
fraction, etc. (What’s the logarithm of a product?)]

Exercise 6. Prove that

(1 + 2x + O(x2 )) = (1 + 2x)(1 + O(x2 )) x → 0.

Exercise 7. Prove or disprove

1
1. = 1 + O(x) x→0
1 + x2
2. cos(x) sin(x) = O(x2 ) x→∞

3. cos(x) sin(x) = O(x2 ) x→0

4. cos(O(x)) = 1 + O(x2 ) all x

5. O(x + y) = O(x2 ) + O(y2 ) x, y → ∞

2
6. e(1+O(1/n)) = e + O(1/n) n→∞

7. nlog(n) = O(log(n)n ) n→∞

 3
√
Exercise 8. Multiply (log(n) + γ + O(1/n)) by (n + O( n)) and express your
answer in O-notation.

Solutions

Solution 1. Let
(a)
f (x)
lim → 0.
x→∞ g(x)
Fix an arbitrary ε > 0. By definition of limit, there is a real constant C, such
that, for all x > C, we have  
 f (x) 
 (x)  < ε .
 

This means that, for all sufficiently large x, we have

| f (x)| < ε |g(x)|,
which states that f (x) = O(g(x)), as desired.
(b) We show that the converse statement is false. Indeed, for all functions f we
have f (x) = O( f (x)). However,
f (x)
lim = 1 6= 0,
x→∞ f (x)
and hence f (x) 6≺ f (x).

Solution 2. Assume that f (x) ≺ g(x), and that g tends to infinity. We must show
that
e f (x)
g(x)
= e f (x)−g(x) → 0
e
that is, that f (x) − g(x) → −∞.
We have
 
f (x) − g(x) f (x)
f (x) − g(x) = g(x) = − 1 g(x) ∼ −g(x) → −∞.
g(x) g(x)
Alternatively, choose ε < 1. Then for all sufficiently large x, we have f (x) ≤
| f (x)| < ε |g(x)| = ε g(x) (g is positive), or f (x) − g(x) < g(x)(1 − ε ) → −∞.
The converse implication is false. We have ex ≺ e2x , but x 6≺ 2x.
4 CHAPTER 1. ASYMPTOTICS

Solution 3. We find
x
ε x ≺ 1 ≺ log(log(x)) ≺ log(x) ≺ log(x)c ≺ xε ≺ xc ≺ xlog(x) ≺ cx ≺ xx ≺ cc .
x
The only tricky relation is xx ≺ cc , which is established by taking logarithms, and
using problem 2.

Solution 4. We will show, in three different ways, that

xlog(x) ≺ log(x)x .

By definition, we have
2
xlog(x) = elog(x) log(x)x = ex log(log(x)) .

We compare the exponents

log(x)2 and x log(log(x)).

From the previous problem, we have log(x)2 ≺ x, and hence, log(x)2 ≺ x log(log(x)).
The desired result now follows from problem 2.
Alternatively, let
xlog(x)
f (x) = .
log(x)x
We take the logarithm of both sides, and then divide by log(x), to obtain

log( f (x)) log(x)2 − x log(x)

g(x) = = = log(x) − x.
log(x) log(x)

It is now clear that

lim g(x) = −∞ = lim log( f (x))
x→∞ x→∞
and hence
xlog(x)
lim =, lim f (x) = 0,
x→∞ log(x)x x→∞

as desired.
Finally, for every constant c > 1, we clearly have

cx ≺ log(x)x . (2)
 5

Furthermore, we have
log(x)2 ≺ x
and hence
log(x)2 ≺ x log(c).
Taking exponentials, we find
2
xlog(x) ≺ cx ,
which, together with (2), gives the desired result.

Solution 5. We find
2x + log(x) x
f (x) = +
1 + 3 log(x) log(x)3/2
2x 1 + log(x)/2x x
= +
3 log(x) 1 + 1/(3 log(x)) log(x)3/2
 
2x 1 + log(x)/2x 3
= + √ .
3 log(x) 1 + 1/(3 log(x)) 2 log x

As x → ∞, the quantity within square brackets tends to 1. Thus

2x
f (x) ∼ .
3 log(x)

Solution 6. We expand the product

(1 + 2x)(1 + O(x2 )) = 1 + 2x + O(x2 ) + 2xO(x2 ) = 1 + 2x + O(x2 ) + O(x3 ).

As x → 0, we have O(x2 ) + O(x3 ) = O(x2 ), and the result follows.

Solution 7.

1. TRUE. Since x → 0, we expand in Taylor series near zero

1
= 1 + x2 + x4 + · · · = 1 + O(x2 ) = 1 + O(x).
1 + x2

2. TRUE. Sine and cosine are bounded functions, and so is their product. So we
may write sin(x) cos(x) = O(1), for all x. Now O(1) = O(x2 ), as x → ∞.
6 CHAPTER 1. ASYMPTOTICS

3. FALSE. Expanding in Taylor series near 0, we have

sin(x) cos(x) = (x + O(x3 ))(1 + O(x2 ))

= x + O(x3 ) + xO(x2 ) + O(x3 )O(x2 )
= x + O(x3 ) + O(x3 ) + O(x5 ) = x + O(x3 ).

The identity x + O(x3 ) = O(x2 ) is false.

4. TRUE. As x → 0, the statement follows from the Taylor expansion

cos(O(x)) = 1 + O(x)2 = 1 + O(x2 ).

As x → ∞, we have cos(x) = O(1) = O(x2 ).

This proof is a bit sloppy. We now develop a rigorous argument, from the
definition of O. We have to show that the set L on the left is a subset of the
set R on the right. If a function f belongs to L, then we can write f (x) =
cos(g(x)), where |g(x)| < c|x|, for some c. Then
1 1
1 − f (x) = 2 sin(g(x)/2)2 ≤ g(x)2 ≤ c2 x2 .
2 2
This shows that 1 − f (x) ∈ O(x2 ), that is, that f (x) ∈ 1 + O(x2 ) = R.
Thus f ∈ L ⇒ f ∈ R, that is, L ⊂ R, as desired.

5. TRUE. We have

O(x + y)2 = O((x + y)2 ) = O(x2 + y2 + 2xy) = O(O(x2 ) + O(y2 ))

= O(O(x2 )) + O(O(y2 )) = O(x2 ) + O(y2 ).

6. TRUE. First, we take care of the exponent. As n → ∞, we find

(1 + O(1/n))2 = 1 + 2O(1/n) + O(1/n)2 = 1 + O(1/n).

Second, we expand the exponential

e1+O(1/n) = eeO(1/n) = e(1 + O(1/n)) = e + eO(1/n) = e + O(1/n).

7. TRUE. This follows from problem 1(a) and problem 4.

 7

Solution 8. We expand the product

√ √
(log(n) + γ + O(1/n))(n + O( n)) = n log(n) + γ n + O(1) + O( n log(n))
√ √
+O( n) + O(1/ n).
√
As n → ∞, the dominant term is O( n log(n)), namely
√ √ √
O(1/n) + O(1) + O( n) + O( n log(n)) = O( n log(n)).

So we find
√ √
(log(n) + γ + O(1/n))(n + O( n)) = n log(n) + γ n + O( n log(n)).
8 CHAPTER 1. ASYMPTOTICS
Chapter 2

Periodic points

Exercise 1. Let the phase space Σ be the set N of positive integers, and let the
dynamics f : N → N be given by

x/2 x even
f (x) =
3x + 1 x odd.
(a) Prove that f has no fixed points.
(b) Show that all points x ≤ 10 are (eventually) periodic, and compute their tran-
sient length, i.e., the length of the pre-periodic part of the orbit through x.
(With the help of a numerical experiment, you should be able to formulate
the ‘3x + 1 conjecture’, one of the most famous unresolved puzzles in num-
ber theory.)

Exercise 2. We represent a real number x as the sum of its integer and fractional
parts, namely
x = ⌊x⌋ + {x}.
The integer part ⌊x⌋ (also called the floor of x) is the largest integer not exceeding
x, while the fractional part has values in the range 0 ≤ {x} < 1.
Now let Σ = [0, 1] and

f (x) = {1/x} x 6= 0
n
0 x = 0.
Prove that f has infinitely many fixed points, and determine all of them.
[The tricky bit is to plot the graph of the function x 7→ {1/x}. Begin with the
restricted domain 1/3 < x < 1.]

9
10 CHAPTER 2. PERIODIC POINTS

Exercise 3. Let Σ = [0, 1] and f (x) = 4x(1 − x).

(a) Determine all fixed points.

(b) For each fixed point, determine three eventually fixed points.
[Construct pre-images of fixed points under f , and then their pre-images.]

(c) Determine all 2-cycles.

[Let Pk (x) = f k (x) − x. The 2-cycles which are not 1-cycles are the roots of
the polynomial P2 (x)/P1 (x). (Why is P2 /P1 a polynomial? Think about it.) ]

Exercise 4. Let Σ = [0, 1], and let f be the tent map, defined as

2x 0 6 x < 1/2
f (x) =
2 − 2x 1/2 6 x 6 1.

(a) Construct the function f 2 .

(b) The point x = 0 is a fixed point. Characterise the set Θ all the points in Σ that
eventually reach x = 0.

Exercise 5∗ . Let Z[x] be the set of polynomials with integer coefficients, and let
f : Z[x] → Z[x] be defined by the formula

p(x) 7→ xn p(1/x) n = deg(p)

where deg(p) is the degree of p. Prove that all points of Z[x] are periodic with
period dividing 2. Characterise the fixed points.

Solutions
 11

Solution 1.
(a) If x is even, then f (x) < x; if x is odd, then f (x) > x. So we cannot have
f (x) = x.
(b) We compute each orbit, until it reaches a point already computed.

orbit transient length

[1, 4, 2, 1, . . .] (a 3-cycle) 0
[2, . . .] 0
[3, 10, 5, 16, 8, 4, . . .] 5
[4, . . .] 0
[5, . . .] 3
[6, 3, . . .] 6
[7, 22, 11, 34, 17, 52, 26, 13, 40, 20, 10, . . .] 14
[8, . . .] 1
[9, 28, 14, 7, . . .] 17
[10, . . .] 4

So all points x such that 1 ≤ x ≤ 10 are (eventually) periodic with period 3.

The ‘3x + 1 conjecture’ states that all orbits end up in the above 3-cycle.

Solution 2. The point x = 0 is a fixed point. For x 6= 0, the fixed point

equation x = f (x) reads
   
1 1 1
=x ⇐⇒ − = x.
x x x
Let m be the floor of 1/x. We write the fixed-point equation explicitly:
1 1 1
− m = x, where ≤ x < , m = 1, 2, . . . .
x m+1 m
2
This gives x + mx − 1 = 0, and hence
√
−m ± m2 + 4
x= .
2
The negative sign does not yield points of Σ. One verifies that (do it!)
√
−m + m2 + 4
0< <1 m = 1, 2, . . . ,
2
giving infinitely many fixed points.
12 CHAPTER 2. PERIODIC POINTS

Solution 3.
(a) The fixed-point equation yields
P1 (x) = x(4x − 3) = 0,
giving two fixed points: x∗ = 0 and x∗ = 3/4.
(b) Letting y = f (x), we find
√
−1 1± 1−y
x= f (y) = .
2
Using f −1 , we compute pre-images of fixed points. For x∗ = 0 we find
0 = x∗ √


+

 2 2
x∗ = 0 ← 1 ← 1 ←

4√

 2  2− 2
4
Note that the point 1 has a single pre-image, which is the critical point of the
function f . For x∗ = 3/4 we obtain

 3 = x∗
4 √


3

∗
x = ← 1  2+ 3
4 ← 4√
4

 2− 3


4
(c) For period 2 we we must solve the equation f 2 (x) = x, which yields
P2 (x) = 64 x4 − 128 x3 + 80 x2 − 15 x = 0.
We eliminate the fixed points. Long division gives
P2 (x)
= 16 x2 − 20 x + 5
P1 (x)
and this polynomial has roots
√
5± 5
∗
= x± ,
8
which constitute the desired 2-cycle, since they are distinct (they need not
be).
 13

Solution 4.

(a) Since f is a piecewise-defined, so is f 2 . It has four pieces. (Why?)

(b) We have
f −1 ({0}) = {0, 1} f −1 ({1}) = {1/2}.
Let Ω ⊂ Σ be the set of all rational points in the unit interval whose denom-
inator is a non-negative power of 2. We claim that Ω = Θ, the set of all
pre-images of the origin.
We proceed by induction on the power n of 2 at denominator. The points
0, 1, 1/2 are in Ω, which is the base case. Assume now that for some n > 1,
all rational of the form x = m/2n are pre-images of the origin, where m is an
odd integer in the range 0 < m < 2n . Then we have

f −1 ({m/2n }) = {m/2n+1 , 1 − m/2n+1 } = {m/2n+1 , (2n+1 − m)/2n+1 } ⊂ Ω.

This completes the inductive step, and we have shown that Θ ⊂ Ω.

Conversely, let x = m/2n ∈ Ω be given. If x < 1/2 then f (x) = m/2n−1 , and
if x > 1/2 then f (x) = (2n−1 − m)/2n−1 . Furthermore in each case the power
of 2 decreases by 1. So f n (x) = 1, and x is a pre-image of the origin. This
shows that Ω ⊂ Θ, and hence Ω = Θ.

Hint 5. Experiment with quadratic polynomials.

14 CHAPTER 2. PERIODIC POINTS
Chapter 3

Stability of periodic orbits

Exercise 1. Explain concisely what is a superstable periodic orbit. Try not to use
symbols.
Exercise 2. Let fλ (x) = x + λ sin(x), with λ > 0.
(a) Show that this map has infinitely many fixed points.
(b) Determine the range of values of λ for which f has stable fixed points. (They
are subdivided into two families.) For what value of λ are they superstable?
(c) Determine the nature of the fixed points at λ = 2.

Exercise 3. Let Σ = [0, 1] and f µ (x) = µ x(1 − x) with µ ≥ 1.

(a) Show that f has a superstable periodic orbit if such an orbit contains the point
x = 1/2.
(b) Show that at µ = 1, the map has a single fixed point, and two fixed points for
µ > 1.
(c) Determine the µ -range for which f has a stable 1-cycle. When is such a cycle
superstable?
(d) Determine the nature of the fixed point at µ = 3. [Hint: look at f 2 , near the
fixed point.]
(e) Determine the µ -range for which f has a stable 2-cycle. When is such a cycle
superstable?

15
16 CHAPTER 3. STABILITY OF PERIODIC ORBITS

Exercise 4. Let x∗ be such that g(x∗ ) = 0, where g is a differentiable real function.

(a) Show that if g′ (x∗ ) 6= 0, then x∗ is a superstable fixed point for Newton’s
method.

(b) Let g′ (x∗ ) = 0, and let k > 1 be the smallest integer for which

d k g(x∗ )
6= 0.
dxk
Show that x∗ is an attractor, but not a superstable one. Determine the multi-
plier at x∗ , as a function of k. Comment on your findings. [Hint: expand the
multiplier in Taylor series.]

Solutions

Solution 1. A periodic orbit is said to be superstable if its multiplier is equal

to zero, that is, if one point of the orbit is a critical point. The convergence to a
superstable orbit is faster than exponential. This means that the distance of a point
from the cycle is of the order of the square of the distance of the pre-image(s) of
this point.

Solution 2.
(a) The fixed point equation λ sin(x) = 0 gives infinitely many fixed points: xk∗ =
k π , k ∈ Z, independent of λ .

(b) The multiplier is f ′ (x) = 1 + λ cos(x), so that


′ ∗ 1+λ k even (type I)
f (xk ) = 1 + λ cos(k π ) =
1−λ k odd (type II).

Type I fixed points are always unstable, since λ is positive. Type II fixed
points are stable provided that 0 < λ < 2, and superstable for λ = 1.

(a) For λ = 2, linear analysis fails to establish the nature of type II fixed points,
so we need to include nonlinear terms. We expand f (x) near x = π in Taylor
series, letting x = π + δ . We find

f ′ (x) = 1 + 2 cos(x) f ′′ (x) = −2 sin(x) f ′′′ (x) = −2 cos(x)

 17

giving
f ′ (π ) = −1 f ′′ (π ) = 0 f ′′′ (π ) = 2
so that
1
f (π + δ ) = π − δ + δ 3 + O(δ 5 ).
3
Thus  
1 2 4
δt+1 = δt −1 + δt + O(δt ) .
3
For sufficiently small |δt |, the quantity in parenthesis is smaller that 1 in ab-
solute value. So δ decreases, and these points are attractors.

Solution 3.
(a) We have f ′ (x) = µ (1 − 2x), so there is a unique critical point xc = 1/2, which
must belong to any superstable cycle.
(b) The map f has two fixed points, namely 0 and x∗ (µ ) = 1 − µ −1 . For µ = 1
they coincide.
(c) For µ > 0, we have f ′ (0) > 1, so 0 is unstable, whereas f ′ (x∗ ) = 2 − µ . This
means that x∗ is stable in the range 1 < µ < 3, and it is superstable for µ = 2.
(d) When µ = 3, we have x∗ = 2/3. Let x = x∗ + δ . We compute

g(δ ) = f (δ ) − x∗ = 3(2/3 + δ )(1 − 2/3 − δ ) = −δ − 3 δ 2 .

Then
g2 (δ ) = δ − 18δ 3 + 27δ 4 = δ (1 − 18 δ 2 + O(δ 3 )).
Thus, for sufficiently small |δ |, the quantity in parenthesis is smaller than 1,
and x∗ is an attractor.
(e) The 2-cycles are the roots of the polynomial
f 2 (x) − x
φ2 (x) = = µ 2 x2 − (µ 2 + µ ) x + µ + 1.
f (x) − x
Let x1 and x2 be the roots of φ2 (x). Then, looking at the coefficients of φ2 (x),
we deduce that
µ2 + µ µ +1 µ +1
x1 + x2 = 2
= x1 x2 = .
µ µ µ2
18 CHAPTER 3. STABILITY OF PERIODIC ORBITS

So the multiplier of the 2-cycle is given by

ω (µ ) = f ′ (x1 ) f ′ (x2 ) = µ 2 (1 − 2(x1 + x2 ) + 4x1 x2 ) = −µ 2 + 2 µ + 4.
The 2-cycle is born at ω (µ ) = 1, or
ω (µ ) − 1 = −(µ + 1)(µ − 3) = 0
giving µ = 3, as desired (the other root√ is out of range). The 2-cycle is super-
stable if ω (µ ) = 0, that is, if µ = 1 + 5 (we have chosen the positive sign
in front of the√radical). This cycle bifurcates to a 4-cycle if ω (µ ) = −1, that
is, if µ = 1 + 6.

Solution 4.
(a) Let f (x) = x − g(x)/g′ (x). Then
g′ (x)2 − g(x)g′′ (x) g(x)g′′ (x)
f ′ (x) = 1 − = .
g′ (x)2 g′ (x)2
Since, by assumption, g(x∗ ) = 0 and g′ (x∗ ) 6= 0, then f ′ (x∗ ) = 0, that is, x∗ is
superstable.
(b) Let δ = x − x∗ , and let a = d k g/dxk (x∗ ). From Taylor’s theorem we have
a k
g(x) = δ + O(δ k+1 ),
k!
whence
a a
g′ (x) = δ k−1 + O(δ k ) g′′ (x) = δ k−2 + O(δ k−1 ).
(k − 1)! (k − 2)!
From the above we obtain
a2
′′ δ 2k−2 + O(δ 2k )
g(x)g (x) k! (k − 2)!
f ′ (x) = f ′ (x∗ + δ ) = =
g′ (x)2 a2
δ 2k−2 + O(δ 2k )
(k − 1)!2
a2
+ O(δ 2 )
k! (k − 2)!
= .
a2 2
+ O(δ )
(k − 1)!2
 19

The value at x = x∗ is obtained by letting δ tend to zero:

(k − 1)!2 1
f ′ (x∗ ) = lim f ′ (x∗ + δ ) = = 1− .
δ →0 k! (k − 2)! k

The convergence is exponential rather than super-exponential, with rate 1 −

k−1 . Thus the convergence is maximal when k = 2, that is, when the second
derivative of g does not vanish at the root x∗ .
20 CHAPTER 3. STABILITY OF PERIODIC ORBITS
Chapter 4

Renormalisation

Exercise 1. Explain what is a bifurcation tree. Try not to use symbols.

Exercise 2. Describe the Feigenbaum-Cvitanović equation, and the significance

of its solutions. Try not to use symbols.

Exercise 3. Consider the map f : [0, 1] → [0, 1] defined by

 x + 2 x2 0 ≤ x < 21


f (x) =
1
2 − 2x 2 ≤ x ≤ 1.


(a) Prove that f has no stable fixed points.

(b) Prove that f has no stable periodic points of any period.
[Look at the multiplier, using the chain rule of differentiation. Consider the
point x = 1/2 separately.]
(c) Find a 2-cycle and compute its multiplier.

Exercise 4. Let f be as in the previous problem.

(a) Show that in the vicinity of 0, the function f is an approximate solution to the
Feigenbaum-Cvitanović equation, in the sense that

α f ( f (x/α )) = f (x) + O(x3 ).

What value do you get for the scaling α in this case?

21
22 CHAPTER 4. RENORMALISATION

(b) Improve the accuracy of the above approximation as follows. Let fˆ(x) =
f (x) + c x3 , and determine the parameter c so that

α fˆ fˆ(x/α ) = fˆ(x) + O(x4 ).

(c) Let g : R → R be given by

x
g(x) =
1−bx
where b is a real constant. Show that this function is an exact solution to
the Feigenbaum-Cvitanović equation. (This solution is not relevant to period-
doubling, since it has the normalisation g(0) = 0. It is relevant to intermit-
tency.) Relate the function g to the functions f and fˆ above.
[Expand g(x) in Taylor series about zero.]

Exercise 5. Consider the logistic map fλ (x) = 1 − λ x2 . Prove that if λ > 2, some
orbits from the interval [−1, 1] escape to infinity.
[Look at the orbit of the critical point. First draw a picture, then from it construct a
rigorous argument.]

Exercise 6. Let Σ be the space of real analytic functions f : R → R, and let D :

Σ → Σ be the differentiation operator

d f (x)
(D f )(x) = .
dx
Show that D has a one-parameter family of fixed points. Find a 2-cycle and a 4-
cycle. Can you find eventually periodic points?

Exercise 7. Let Σ be the space of real functions f analytic at x = 0 and normalised

so that f (0) = 0. Let R be the operator defined as

(R f )(x) = α f (x/α ) α > 1.

Prove that R : Σ → Σ. Specifically, prove that (R f )(0) = 0, and that if ρ is the radius
of convergence of f , then the radius of convergence of R f is αρ . What happens to
the radius of convergence under repeated iterations of R?
 23

Exercise 8. Consider the Feigenbaum-Cvitanović equation for period-doubling

(R f )(x) = α f ( f (x/α )) f (0) = 1 f (x) = f (−x) α = 1/ f (1).
Find an approximate value α for the universal constant α ∗ = 1/ f ∗ (1), where R f ∗ =
f ∗ . Specifically, let f (x) = 1 − λ x2 , and determine λ such that
(R f )(x) = f (x) + O(x4 )
whence the corresponding value of α . Compare the latter with α ∗ .

Solutions

Solution 1. We consider a real map, depending on a parameter. Its stable cycles

are points on the real line. As the parameter changes, these points also change, and
in the parameter-coordinate plane we obtain a graph, called a bifurcation tree.
The bottom branch of the tree corresponds to the fixed point, while the nodes
correspond to period-doubling bifurcations, where a cycle loses stability, giving
birth to a cycle of twice the period. The leaves correspond to the accumulation
point of the period-doubling sequence.

Solution 2. We consider a renormalisation operator, acting by function composi-

tion combined with scaling about the origin. The Feigenbaum-Citanović equation
is the fixed point equation of such operator, and its solutions are functions which are
invariant under composition and scaling. One typically looks for analytic solutions
satisfying certain normalisation conditions.
For instance, restriction to even analytic functions with a quadratic maximum at
the origin, and whose value at zero is unity, yields the universal function of the accu-
mulation point of period-doubling. The associated scaling constant (the reciprocal
of the value of the fixed point at 1), is a universal constant.

Solution 3.
(a) The fixed points satisfy the equation f (x) = x, so there are two cases:
i) 0 ≤ x < 1/2. We have x + 2x2 = x, hence x = 0. We find f ′ (x) = 1 + 4 x
so f ′ (0) = 1, and we must look at higher-order terms. Since f ′′ (0) = 4 > 0,
we have that for x > 0, f (x) > x, and the fixed point is unstable.
ii) 1/2 < x ≤ 1. We have x = 2 − 2 x, giving x = 2/3. We have f ′ (x) = −2
for all x in that range, so the fixed point is unstable.
24 CHAPTER 4. RENORMALISATION

(b) Let (x1 , . . . , xn ) be a n-cycle. We may assume n > 1, since the fixed points
have been dealt with already. No point in the n-cycle can be equal to 1/2,
since f 2 (1/2) = 0, and so 1/2 is non-periodic (it is eventually fixed). In
particular, f is differentiable at every point in the cycle. The multiplier of the
cycle is given by the product

f ′ (x1 ) f ′ (x2 ) · · · f ′ (xn ).

For all point x in the cycle we have x 6= 0, 1/2. Thus the derivative of f is
defined, and is greater than 1 in absolute value. It follows that any n-cycle is
unstable.
(c) A 2-cycle must necessarily have one point x1 in the interval (0, 1/2) and
another point x2 in the interval (1/2, 1). Let x ∈ (0, 1/2). Then f (x) = x +
2 x2 , and f 2 (x) = 2 − 2 (x + 2 x2 ). The equation f 2 (x) = x gives
√
−3 ± 41
4 x2 + 3 x − 2 = 0 =⇒ x1 =
8
and we must choose the positive sign. Then
√
19 − 41
x2 = f (x1 ) = .
16
The derivative of f at x1 and x2 is equal to 1 − 4 x1 and −2, respectively. So
the multiplier of the 2-cycle is
√
2 (1 − 4 x1 ) = 5 − 41 < −1.

Solution 4.
(a) We construct f 2 using the left branch f (x) = x + 2x2 , since if x is small,
f (x) < 1/2. We find

x2 x3 x4
α f ( f (α )) = x + 4 +8 2 +8 3.
α α α
Choosing α = 2 gives

2 f ( f (x/2)) = x + 2 x2 + O(x3 ) = f (x) + O(x3 )

as desired.
 25

(b) Letting fˆ(x) = x + 2 x2 + c x3 we find

 c 3
2 fˆ( fˆ(x/2)) = x + 2 x2 + 2 + x + O(x4 ) = fˆ(x) + O(x4 ).
2
The last equation gives
c
2+ =c
2
or c = 4. Thus the function fˆ(x) = x + 2 x2 + 4 x3 satisfies Feigenbaum-
Cvitanović equation near zero, up to order 4.

(c) We find

x/α
 
x/α 1 − b x/α x
α g(g(x/α )) = α g =α = .
1 − b x/α x/α 1 − 2 b x/α
1−b
1 − b x/α

So choosing α = 2 we obtain

2 g(g(x/2)) = g(x).

Note that, for sufficiently small x, we have

x
g(x) = = x(1 + b x + b2 x2 + · · ·) = x + b x2 + b2 x3 + b3 x4 + · · ·
1−bx

Thus the functions f and fˆ of the previous problem are two successive ap-
proximations to this solution for b = 2.

Solution 5. Let x0 = 0, the critical point. Then x2 = f 2 (0) = f (1) = 1 − λ < −1,
since λ > 2. We have f (−1) = 1 − λ = −1 − c, with c = −2 + λ > 0. Furthermore,
f ′ (x) = −2λ x > 1 for all x < −1. Therefore, for all x < −1 we have f (x) < g(x),
where g(x) = x − c. An easy induction shows that gt (x) = x − ct, and therefore, for
x < −1, we have f t (x) < x − ct.
Putting everything together, we have f t+2 (0) = f t (x2 ) < 1 − λ − ct → −∞, as
t → ∞.
26 CHAPTER 4. RENORMALISATION

Solution 6. We find that the functions f (x) = α ex , where α a real number, form
a one-parameter family of fixed points of D.
The set {e−x , −e−x } is a 2-cycle.
The set {sin(x), cos(x), − sin(x), − cos(x)} is a 4-cycle.
Let g(x) = f (x) + p(x), where f (x) is periodic and p(x) is a polynomial. Then
g(x) is eventually periodic. The degree of p(x) gives the transient length.
Solution 7. We have
(R f )(0) = α f (0 · α ) = α f (0) = 0.
Let ρ f be the radius of convergence of the series
∞
f (x) = ∑ ck x k
k=0
or
1
= lim sup(ck )1/k < ∞.
ρ k→∞
Let now fˆ = α f (x/α ) where
∞
fˆ(x) = ∑ ĉk xk .
k=0
We find
ck
ĉk =
α k−1
and therefore
 1/k
1 1/k 1 1/k 1 1 1
= lim sup(ĉk ) = lim sup k−1
ck = lim sup (k−1)/k = .
ρ̂ k→∞ k→∞ α ρ k→∞ α αρ
Thus
ρ̂ = αρ .
Under repeated applications of R, the radius or convergence diverges to infinity.
Solution 8. Let f (x) = 1 − λ x2 . Then α = 1/ f (1) = 1/(1 − λ ), and
1
(R f )(x) = f (1 − λ (1 − λ )2 x2 ) = 1 + 2λ 2 (1 − λ ) x2 + O(x4 ).
λ −1
Requiring that (R f )(x) = f (x) + O(x4 ) yields 2 λ 2 − 2λ − 1 = 0, or
√
1+ 3 √
λ= α = −1 − 3 = −2.732 . . .
2
to be compared with the exact value α ∗ = −2.50290 . . ..
Chapter 5

Homeomorphisms and
diffeomorphisms

Exercise 1. Consider the following maps f : R → R with f (x) given by

a) −2x + 7 b) −x2 c) x3
d) ex e) −2x − sin(x) f) x + sin(x)
Decide in each case if f is a homeomorphisms, a diffeomorphisms, or neither. If f
is a hom(diff)eo, decide if it is order-preserving or reversing.

Exercise 2. We represent the circle S1 as the interval [−1, 1] with the end-points
identified. Consider the following family of mappings

f λ : S1 → S1 fλ (x) = λ x + (1 − λ ) x3 , λ ∈ R.

Determine the ranges of values of λ for which fλ is i) a homeomorphism; ii) a

Cr -diffeomorphism (in which case you must determine r); iii) neither.
[Note that both domain and co-domain are circles, so the points −1 and +1 must be
identified. In all, there are seven distinct λ -ranges to be considered.]

Exercise 3. Construct orientation-preserving diffeomorphisms f : R → R with

the properties specified below. (First draw the graph of a function with the desired
properties. Then define it analytically, and prove that it satisfies such properties.)
(a) A single attractive fixed point at x = −1.

(b) A single fixed point at x = 0, which is neither attractive nor repelling.

27
28 CHAPTER 5. HOMEOMORPHISMS AND DIFFEOMORPHISMS

(c) Three fixed points —two attractors and one repeller.

Exercise 4. Construct an orientation-reversing diffeomorphisms f : R → R with

the following properties.
(a) A single unstable fixed point at x = 1.
(b) A single symmetric 2-cycle, that is, a 2-cycle of the form {x0 , −x0 }.

Exercise 5. We wish to construct an orientation-reversing diffeomorphism of a

neighbourhood of the origin, all whose points are 2-cycles (we ignore the trivial
case f (x) = −x). The strategy is to build f (x) as a power series, in such a way that
f 2 (x) = x + O(xk ), for increasing k.
(a) Let f (x) = −x + x2 . Verify that f 2 (x) = x + O(x3 ).
(b) Let f (x) = −x + x2 + α x3 , where α is a real parameter. Determine α so that
f 2 (x) = x + O(x5 ).
(c) Proceed as above to determine the coefficient of terms of higher degree, up
to the largest degree you can handle (you may need Maple). The larger the
degree, the larger the mark.
Note: the above method yields only a formal solution, since the power series
constructed in this way may have a zero radius of convergence.

Solutions

Solution 1.
a) Order-reversing diffeo.
b) Not invertible, so neither.
c) Order-preserving homeo, but not diffeo since f −1 is not differentiable at zero.
d) Not surjective, so neither.
e) Order-reversing diffeo.
f ) Order-preserving homeo, because the derivative vanishes.
 29

Solution 2. We compute

f ′ (x) = λ + 3 (1 − λ ) x2 f ′′ (x) = 6 (1 − λ ) x.

As a real function, the function f is analytic. We check its behaviour at the points
x = ±1. We find f (1) = 1 and f (−1) = −1, and since 1 and −1 are identified,
f is continuous on S1 . Furthermore, f ′ (1) = f ′ (−1) = 3 − 2λ , whereas f ′′ (1) =
− f ′′ (−1) = 6(1 − λ ). Therefore f is of class C1 but not C2 , for all values of λ ,
except for λ = 1 (see below).
Furthermore
λ
f ′ (x) = 0 ⇐⇒ x2 = .
3(λ − 1)
So f ′ (x) does not vanish at all, or does not vanish in the interval [−1, 1], respectively,
in the ranges
3
0<λ <1 1<λ < .
2
There are three critical values.
i) λ = 0. The function f0 (x) = x3 is of class C0 .
ii) λ = 1. The function f1 (x) = x is of class C∞ .
iii) λ = 3/2. The function f3/2 (x) = (3x − x3 )/2, is of class C0 .

SUMMARY:
λ class comment

(−∞, 0) − f not invertible

0 C0 f −1 not differentiable at x = 0
(0, 1) C1 f ′′ not continuous at x = 1
1 C∞ f (x) = x
(1, 3/2) C1 f ′′ not continuous at x = 1
3/2 C1 f −1 not differentiable at x = 1
(3/2, ∞) − f not invertible

Solution 3.

(a) We let Φ1 (x) = f (x) − x = α (x + 1), giving f (x) = x(α + 1) + α . Choosing

α > −1 we ensure the orientation-preserving property, while for −2 < α < 0
the point x = −1 is an attractor. Thus −1 < α < 0.
30 CHAPTER 5. HOMEOMORPHISMS AND DIFFEOMORPHISMS

(b) The function

1 x 3x2 − 1
g(x) = 2
g′ (x) = −2 g′′ (x) = 2
x +1 (x + 1)2
2 (x2 + 1)3
′′
√ at x = 0, with g(0) = 1. The second derivative g (x)
has a single maximum
vanishes at x = ±1/ 3, so that
√ 8
|g′ (x)| ≤ |g′ (1/ 3)| = √ < 1
9 3
Moreover g′′ (0) < 0. Therefore the function
f (x) = x − 1 + g(x)
has the following properties:
i) f ′ (x) > 0 ii) f (0) = 0 iii) f ′ (0) = 1 iv) f ′′ (0) < 0.
Now, the function f is differentiable (because g is). i) says that f is an
orientation-preserving diffeomorphism. ii) says that f has a fixed point at
zero. iii) and iv) say that such a point is neither an attractor nor a repeller.
(c) Let α be a positive real number, and let
1
f (x) = α x + arctan(x) f ′ (x) = α + > 0.
x2 + 1
Then f is an orientation-preserving diffeomorphism of the real line. The fixed
point equation reads
Φ1 (x) = f (x) − x = (α − 1)x + arctan(x) = 0.
We have Φ1 (0) = 0, so x = 0 is a fixed point. Furthermore
Φ1 (x) = α x + O(x2 ), x → 0; Φ1 (x) = (α − 1)x + O(1), x → ∞.
Let 0 < α < 1. Then Φ1 (x) is positive for sufficiently small x and negative
for sufficiently large x, and from the intermediate value theorem, there exists
a point x∗ > 0 for which Φ1 (x∗ ) = 0. Such a point is unique, since Φ′′1 (x) < 0
for all positive x. The third fixed point x = −x∗ then results from symmetry:
Φ1 (x) = −Φ1 (−x). Since f ′ (0) = α + 1 > 0, the origin is a repeller. Since
0 < f ′ (x∗ ) = f ′ (−x∗ ) < 1, the other points are attractors.
 31

Solution 4.
(a) We let Φ1 (x) = f (x) − x = α (x + 1), giving f (x) = x (α + 1) + α . We need
α < −1 for reversing the orientation and |α +1| > 1 for a repeller, so α < −2.
(b) Consider the analytic function
2
f (x) = −x + ε g(x) g(x) = (x2 − x02 ) e−x .
We have f (±x0 ) = ∓x0 , so {x0 , −x0 } is a 2-cycle. The function g(x) is dif-
ferentiable with bounded derivative |g′ (x)| < C, say. Then f ′ (x) > 0 for all
|ε | < 1/C. Since g(x) does not vanish at any other point, f (x) does not have
any other symmetric 2-cycle.
(Generalisation. The function
n
2
f (x) = −x + ε g(x) g(x) = ∏(x2 − xi2 ) e−x
i=1
has precisely n symmetric 2-cycles. Furthermore, for sufficiently small |ε |, f
is a diffeomorphism.)

Solution 5.
(a) We have
f 2 (x) = x − 2x3 + x4 = x + O(x3 ).
(b) Let f (x) = −x + x2 + α x3 . We find
f 2 (x) = x − x2 − α x3 + (−x + x2 + α x3 )2 + α (−x + x2 + α x3 )3
= x − 2(α + 1)x3 + (α + 1)x4 + α (−1 + 3 α )x5 + O(x6 ).
Choosing α = −1 we kill both the cubic and quartic term, giving
f 2 (x) = x + 4x5 + O(x6 ) = x + O(x5 ).

(c) The function

20
2 533 8 43981 10
f (x) = ∑ ck x k = −x + x2 − x3 + x4 − 1736 x6 + x − x
k=1 3 540 12960
1111801 12 558379369 14 527406923 16
+ x − x + x
68040 5443200 653184
10155352946783 18 17448102987228961 20
− x + x .
1306368000 193995648000
32 CHAPTER 5. HOMEOMORPHISMS AND DIFFEOMORPHISMS

has the property that

f 2 (x) = x + O(x23 ).
This expansion is almost certainly divergent, as demonstrated by the the reg-
ular growth of (ck )1/k with k.

1.8

1.6

1.4

1.2

0.8

0.6

0.4

0.2

0 2 4 6 8 10 12 14 16 18 20

Figure 5.1: Growth of the coefficients ck .

Chapter 6

Doubling map

Exercise 1. Order the integers from 20 to 30 inclusive, using Sharkowsky’s order-

ing. Then explain what it means.

Exercise 2. Consider the doubling map.

(a) By looking at periodic binary digits, show that there are three orbits of mini-
mal period 4. How many orbits are there of minimal period 6?
[To answer, you do not need to compute all 6-strings!]

(b) Determine all points of the 3-cycle with initial condition

x0 = .001 001 001 001 . . . = .001

as rational numbers.
[The number x0 is the sum of a geometric series.]

(c) Do the same for the 6 cycle

x0 = .001101 001101 001101 . . . = .001101.

(d) Divide the unit interval in four equal sub-intervals, hence determine how the
points of the periodic orbit with initial condition 1/51 are distributed among
those intervals. Do the same with the orbit with initial condition 1/13.

(e) Divide the unit interval in 16 equal sub-intervals, hence determine the binary
digits of a 16-cycle that has has one point in each sub-interval.

33
34 CHAPTER 6. DOUBLING MAP

(f) Let x be a point in an n-cycle of the doubling map

∞
x= ∑ bk 2−k xk = xk+n .
k=1

Prove that
n
B
x= n
2 −1
where B = B(b1 , b2 , . . . , bn ) = ∑ bk 2n−k .
k=1

[Write k = s n + r, with 1 ≤ r ≤ n, then split the summation into two parts.]

p
Exercise 3. Consider the mapping f (x) = 1 − 2 |x| on the interval [−1, 1].
(a) Draw the graph of f hence determine f ′ . Is this map chaotic?
[This is a tricky question. Consider the orbit with initial condition x = 0 sep-
arately. You would need a positive lower bound for the Lyapounov exponent,
but that’s hard to obtain. ]
(b) Theory shows that the average position of the points in a ‘typical’ non-periodic
orbit is equal to −1/3, namely that

1 N−1 1
x̄ = lim
N→∞ N
∑ xt = − .
3
t=0

Compute numerically the average position of the points of an orbit, using an

initial condition x0 of your choice and the largest value of N your computer
can handle1 . Verify that this result is in agreement with the theory (Do not
expect great accuracy: convergence is slow.)

Solutions

Solution 1.
24 ⊳ 28 ⊳ 20 ⊳ 30 ⊳ 26 ⊳ 22 ⊳ 29 ⊳ 27 ⊳ 25 ⊳ 23 ⊳ 21
If a continuous map f : [a, b] → R has an orbit of one of the above periods, then it
also has orbits of all the periods that are “smaller” according to this ordering. In
particular, if the period 21 is present, so are all the other periods in the list.

1 If N < 1000, then it is time to junk your computer.

 35

Solution 2.
(a)
0011, 0001, 0111.
There are 26 binary strings of length 6. Of those, 23 have minimal period 3
or 1, which must be subtracted. Furthermore 22 strings have minimal period
2 or 1, which must also be subtracted, except that in doing so we subtract the
1-strings twice. Thus the total number of strings of minimal period 6 is

26 − 23 − 22 + 21 = 64 − 8 − 4 + 2 = 54.

The total number of orbits is then given by 54/6 = 9.

(b) We have
∞
1 1 1 1 1
x0 = .001 = 3 + 6 + 9 + · · · = ∑ = .
2 2 2 k=1 8 7
Then
2 4
x1 ≡ 2x0 (mod 1) = x2 ≡ 2x1 (mod 1) = .
7 7
(c) We have

x0 = .001101 = .000001 + .000100 + .001000

= .000001 × (1 + 4 + 8)
∞  k
1 13
= 13 ∑ 6
= .
k=1 2 63

Iterating the map, we find

13 26 52 41 19 38
x0 = , x1 = , x2 = , x3 = , x4 = , x5 = .
63 63 63 63 63 63

(d) Let x0 = 1/51. The numerators of the points in the orbits are

1, 2, 4, 8, 16, 32, 13, 26, 1, . . .

So the orbit has period 8. We partition the unit interval into four sub-intervals


4
\
Ik = (k − 1)/4, k/4 k = 1, . . . , 4 Ik = [0, 1).
k=1
36 CHAPTER 6. DOUBLING MAP

We have (referring to numerators only)

{1, 2, 4, 8} ∈ I1 {13, 16} ∈ I2 {26, 32} ∈ I3 .

Likewise, if x0 = 1/13 we find

1, 2, 4, 8, 3, 6, 12, 11, 9, 5, 10, 7
so the period is 12. This time there are 3 points in each interval, so the orbit
is uniform.
(e)
x0 = 0000111100101101 = 259/4369.
The above string contains all 4-substrings. (There is more than one orbit with
this feature.)
(f) Let the binary representation of x have n periodic digits
x = .b1 · · · bn
Then ∞
x= ∑ bk 2−k bk = bk+n , ∀k > 0.
k=1
By writing k = sn + r, with 1 ≤ r ≤ n, we obtain bk = br , from the above
equation, and we can write
" #
∞ n
x = ∑ ∑ br 2−r 2−n s
s=0 r=1
n ∞
s
= ∑ br 2−r · ∑ 2−n
r=1 s=0
n
1
= ∑ br 2−r
r=1 1 − 2−n
n
1
= 2n
∑ br 2−r n
r=1 2 −1
n
1
= ∑ br 2n−r n .
r=1 2 −1
 37

Solution 3.
p
(a) Let y = f (x) = 1 − 2 |x|. We find
 2  2
1−y 1−y
|x| = x=± .
2 2

The map is not differentiable at x = 0. We find

1

p
 x<0
′ |x|
f (x) =
 − p1
 x > 0.
|x|

The orbit of 0 of pre-periodic: 0 7→ 1 7→ −1 7→ −1. Apart from ±1, the map

is differentiable with | f ′ (x)| > 1, so the map appears to be chaotic. However,
to prove it in a straightforward way, we would need a positive lower bound
for the Lyapounov exponent. However, the quantity log | f ′ (x)| is not bounded
away from zero.

(b) We compute
N−1
1
ΛN (x0 ) =
N ∑ xt .
t=0

# ---- the function f

f:=x->1-2*sqrt(abs(x)):
# ---- parameters
N:=1000:
nprint:=5:
# ---- initial condition
x:=1.0/3.0:
xbar:=x:
# ---- main loop
for s to nprint do
to N do
x:=f(x):
xbar:=xbar+x
od:
print(xbar/(s*N))
38 CHAPTER 6. DOUBLING MAP

od:
Selected output for x0 = 1/3:

N ΛN
1000 −0.3450708675
2000 −0.3573080697
3000 −0.3362268846
4000 −0.3128255468
5000 −0.3164218084

Convergence is slow.
Chapter 7

Conjugacy

Exercise 1. What is chaos? Explain it briefly, without using symbols.

Exercise 2. Write f ∼ g if f is topologically conjugate to g. Prove that ∼ is an

equivalence relation, i.e., ∼ is reflexive, symmetrical and transitive.

Exercise 3. Let f and g be conjugate via a diffeomorphism h. Let f have a fixed

point at x∗ .
(a) Prove that the multiplier of g at h(x∗ ) is the same as that of f at x∗ .

(b) Formulate and prove the analogous statement for an n-cycle.

Exercise 4. Consider the following mappings

f (x) = a x g(y) = b y (a 6= b).

(a) Determine conditions on a and b for which f and g are conjugate via y =
h(x) = xn , where n is a positive integer.

(b) Use problem 2 (a) to infer that no diffeomorphism exists which conjugates f
and g.

Exercise 5. Let f (x) = 2x (1 − x), and let a, b be distinct complex numbers. Con-
struct a map g(y) which is conjugate to f , and which has fixed points at y = a and
y = b (with the former superstable).

39
40 CHAPTER 7. CONJUGACY

Exercise 6. Consider the complex mapping

z2 + n
f (z) = n ∈ Z.
2z
(a) Show that f is the Newton’s iteration for finding the solutions of the equation
z2 − n = 0.
(b) Show that f is conjugate to
g(w) = w2
via √
z+ n
w = φ (z) = √ .
z− n
(c) Using the above conjugacy, what can you say about the periodic orbits of f ,
and their stability?

Solutions

Solution 1. Chaos is sensitive dependence on initial conditions, a mechanism that

causes the orbits of a dynamical system which are initially close to each other to
separate exponentially. Chaotic systems features a dense set of unstable periodic
orbits.
Formally, a system is defined to be chaotic if its Lyapounov exponent is positive.
This is the average of the logarithm of the multiplier along an orbit, which turns out
to be the same for almost all orbits.
A most prominent feature of chaotic behaviour is the emergence of probabilistic
laws, which exist alongside the extremely complex orbital motions.

Solution 2. Notation:
h
f ∼g ⇐⇒ h◦ f = g◦h

Reflexivity: f ∼id f .
−1
Simmetry: f ∼h g implies h−1 ◦ g = f ◦ h−1 , that is, g ∼h f . Now h−1 is a
homeo, by definition.
Transitivity: if f ∼h g and g ∼k l, then

f = h−1 ◦ g ◦ h = h−1 ◦ k−1 ◦ l ◦ k ◦ h = (k ◦ h)−1 ◦ l ◦ (k ◦ h).

 41

Now, the composition of continuous bijections is a continuous bijection, so (k ◦ h)

is a homeomorphism, and f ∼k◦h l.

Solution 3.

(a) We have g = h ◦ f ◦ h−1 . Hence, letting y = h(x), we have

g′ (y) = (h−1 )′ (y) f ′ (x) h′ ( f (x))

If x is a fixed point, f (x) = x, and therefore

g′ (y) = f ′ (x)[(h−1 )′ (y) h′ (x)]

but the product in square brackets is unity, because

1 = (h−1 ◦ h)′ (x) = h′ (x) (h−1 )′ (y).

(b) Let h be a smooth conjugacy between f and g, and let {xk∗ }nk=1 be an n cycle
for f . Then {h(xk∗ )}nk=1 is an n-cycle for g, with the same multiplier.
For an n-cycle, first note that if f ∼ g, then f n ∼ gn . (with the same conjugacy
function —see notes). Furthermore we know (see notes) that {h(xk∗ )}nk=1 is
an n-cycle for g. Then proceed as above with f n and gn in place of f and g,
respectively.

Solution 4.

(a) We have
h( f (x)) = (ax)n = g(h(x)) = bxn .

So b = an , with n odd. This is a homeomorphism, but not a diffeomorphism.

(b) It suffices to note that f and g both have a unique fixed point, but with differ-
ent multiplier.
42 CHAPTER 7. CONJUGACY

Solution 5. The mapping f as an unstable fixed point at x = 0 and a superstable

fixed point at x = 1/2.
Let h(x) = mx + q. Requiring h(0) = b and h(1/2) = a yields
y−b
h(x) = 2(a − b) + b h−1 (y) = .
2 (a − b)

The equation g(y) = h( f (h−1 (y))) gives

y2 − 2ay + ab
g(y) =
b−a
which is the desired mapping.

Solution 6.

(a) The Newton’s map for finding the roots of the equation h(z) = z2 − n = 0 is

h(z) z2 − n z2 + n
f (z) = z − = z − = .
h′ (z) 2z 2z

(b) Two mappings f : X → X and g : Y → Y are (topologically) conjugate if there

exists a homeomorphism (bi-continuous map) φ : X → Y such that φ ◦ f =
g◦φ.
The map φ is a homeomorphism of the Riemann sphere. We prove that φ ◦
f = g◦φ.
√ √ 2
z2 + n + 2z n

z+ n
(φ ◦ f )(z) = 2 √ = √ = (φ (z))2 = (g ◦ φ )(z).
z + n − 2z n z− n

(c) The restriction of g to the unit circle is the doubling map, and hence g has
infinitely many periodic orbits, of every period. All these cycles are unstable,
and they are dense on the unit circle, which is the Julia set of g. In addition, g
has two superstable fixed points, at 0 and ∞,
It can be shown that a smooth conjugacy preserves periodic orbits and their
multipliers.
Because the conjugacy φ is smooth, the map√f has the same orbit structure as
g. (One verifies that J( f ) is R if n < 0, and −1 R if n > 0.)
Chapter 8

Two-dimensional mappings

Exercise 1. Consider the following two-dimensional map:

(x, y) 7→ ( f (x, y), g(x, y)).

Write down the equations whose solutions are the points of period dividing 3.

Exercise 2. Consider the following map of R2 :

Φ(xt , yt ) = (xt+1 , yt+1 )

where

xt+1 = −yt + g(xt )

yt+1 = xt − g(xt+1 )

and g is a real function.

[Check the subscripts carefully.]

(a) Find a function g for which Φ has no fixed points.

[This means that the fixed-point equation has no solutions.]

(b) Let g(x) = λ + x2 . Determine the range of values of λ for which Φ has two
distinct fixed points.

43
44 CHAPTER 8. TWO-DIMENSIONAL MAPPINGS

Exercise 3. Consider the two-dimensional mapping

2πλ 2πλ
xt+1 = xt + = xt + ;
yt+1 yt + sin xt
yt+1 = yt + sin xt .

where x, y, λ are real numbers, λ > 0, yt 6= 0, and xt is periodic with period 2π .

(Thus the phase space is a cylinder, with the circle y = 0 removed.)

(a) Show that the mapping has two infinite families of fixed points.

(b) Show that, for fixed λ , the number of (marginally) stable fixed points is at
most finite. Determine the range of values of λ for which all fixed points are
unstable.

Exercise 4. Consider the linear map

f :C→C f (z) = λ z.

Prove that all orbits of f are periodic if and only if λ has unit modulus, and its
argument is a rational multiple of 2π .

Solutions

Solution 1.

x = f ( f ( f (x, y), g(x, y)), g( f (x, y), g(x, y)))

y = g( f ( f (x, y), g(x, y)), g( f (x, y), g(x, y))).

Solution 2.
(a) Letting xt = xt+1 = x and yt = yt+1 = y and adding the two equations, we find

x + y = −y + x

giving y = 0, hence x = g(x). So the fixed points of Φ take the form (x∗ , 0),
where x∗ is a fixed point of g. Therefore if g has no fixed point, neither has
Φ. We choose g(x) = x + 1, say.
 45

(b) The equation g(x) = x gives λ + x2 = x, and hence

p
1± 1−4λ
x= .
2
For two distinct real solutions, the discriminant must be positive, whence
λ < 1/4.

Solution 3.
(a) The Jacobian matrix is given by
2πλ cos(x) 2πλ
!
1− −
J(x, y) = (y + sin x)2 (y + sin x)2 .
cos(x) 1

One sees that Det(J(x, y)) = 1 (the mapping is area-preserving).

The fixed points are:
2πλ
= 2π k x ≡ 0, π (mod 2π ) k ∈ Z k 6= 0.
y
They form two families:

(0, λ /k) (π , λ /k) k 6= 0.

(b) We compute
2 2
!
J(π , λ /k) = 1 + 2π k + 2π k
λ λ
1 1
2π k2
Tr(J(π , λ /k)) = 2 + >2
λ
so the points (π , λ /k) are all unstable.
2 2
!
J(0, λ /k) = 1 − 2π k − 2π k
λ λ
1 1

We have
2π k2
Tr(J(0, λ /k)) = 2 − < 2. (1)
λ
46 CHAPTER 8. TWO-DIMENSIONAL MAPPINGS

For marginal stability, we require Tr(J(0, λ /k)) > −2, that is,

π k2
λ>
2
which can be satisfied for at most finitely many values of k. The inequality
(1) implies r
πλ
|y| > .
2
which shows that the (marginally) stable fixed points are located sufficiently
far away from the origin.
From (1) it also follows that for λ < π /2 there are no stable fixed points at
all.

Solution 4. We have zn = λ t z0 , so if z0 = zn , then |z0 | = |zn | = |λ ||z0 |. If z0 6= 0,

this gives |λ | = 1, as desired.
Let now
λ = e2πα .
For periodicity when z 6= 0, we need λ n = 1. This gives 2πα n = 2kπ , for some
k ∈ Z. But then α = k/n, a rational number.
Conversely, if |λ | = 1 and the argument 2πα of λ is such that α = k/n, with
k and n coprime, then n is the smallest natural number such that λ n = 1, and all
non-zero points are periodic with the same period n.
Chapter 9

Fractals

Exercise 1. Prove that the box dimension of the set

{1/nα : n ∈ N} α >0
is equal to 1/(α + 1).

Exercise 2.
(a) Explain why in the definition of Hausdorff distance between two subsets of
the plane, these sets are required to be closed and bounded.
(b) Compute the Hausdorff distance between the unit disc and a filled-in square
inscribed in it (that is, the vertices of the square belong to the unit circle).

Exercise 3. Let h(A, B) be the Hausdorff distance between two compact subsets
of R.
(a) Let C be the Cantor ternary set, let I be the closed unit interval, and let A be a
set constituted by the point 0 and the rationals 3−k , for k = 0, 1, . . ., that is
 
1 1
A = 0, 1, , 2 , . . . .
3 3
Compute h(I,C), h(I, A) and h(A,C).
(b) Determine iterated function systems whose attractors are I, C and A, respec-
tively.

47
48 CHAPTER 9. FRACTALS

Exercise 4∗ . Let H (C) be the set of all closed and bounded subset of C, with the
Hausdorff distance. Let S be the unit circle in C. Characterise the sets in H (C)
which belong to the circle of unit radius and centre C.

Exercise 5. Let A be the set of figure 9.

Figure 9.1: A fractal.

(a) After choosing coordinates, construct the iterated function system Φ whose
fixed point is A.

(b) Compute the box dimension of A, hence show that A is a fractal.

(c) Write a Maple code to generate and plot this fractal.

Solutions

Solution 1. We choose boxes of size εn , where

1 1
εn = α
− lim εn = 0.
n (n + 1)α n→∞
 49

Using the binomial theorem, we find

(n + 1)α − nα nα + α nα −1 + O(nα −2 ) − nα
εn = =
nα (n + 1)α nα (n + 1)α
α + O(n−1 ) α
= α
∼ α +1
n(n + 1) n
where, if f and g are real functions, we write f (n) ∼ g(n) to mean
f (n)
lim = 1.
n→∞ g(n)

To cover the point 1, 1/2α , . . . , 1/nα , we need n boxes of size εn . The remaining
points lie in the interval [0, 1/(n + 1)α ]. To cover such an interval, we need m
boxes, where  
1 n
m= α
∼
(n + 1) εn α
and ⌈·⌉ is the ceiling function. Thus

N(εn ) = n + m ∼ n(1 + 1/α ).

We compute the box dimension

log(N(εn )) log(1 + 1/α ) + log(n) 1
D = lim = lim = .
n→∞ − log(εn ) n→∞ − log(α ) + (α + 1) log(n) α +1
Thus, by adjusting α , the box dimension of our set can be any real number between
0 and 1.

Solution 2.
(a) Boundedness is required to avoid infinite distances. (Why? Give an example.)
Closeness ensures that two sets are the same iff the distance between them is
zero. (Why? Explain in detail.)
(b) If A is the disc and B the square, then B ⊂ A, and hence hBA = 0. An elemen-
tary geometric consideration shows that
1
hAB = h(A, B) = 1 − √ .
2
50 CHAPTER 9. FRACTALS

Solution 3.

(a) One finds, in a straightforward manner

1 1 1
h(I,C) = h(I, A) = h(A,C) = .
6 3 3

(b) We look for contraction mappings covering the sets I, C, and A with smaller
copies of themselves. One finds

I: w1 (x) = x/2 w2 (x) = x/2 + 1/2

C: w1 (x) = x/3 w2 (x) = x/3 + 2/3
A: w1 (x) = x/3 w2 (x) = 1.

Note that the IFS for the set A requires condensation.

Hint 4. Any set lying inside S and containing the origin is at unit distance from it.
Next find sets not contained in S.

Solution 5.

(a) The set A is the union of three small copies of itself, two of which are rotated
clockwise (or anti-clockwise) by π /2. It follows that the IFS for A comprises
three mappings. Placing the origin in the barycentre of the √ A, and assuming
that A has height 1, we find, using complex notation (i = −1)
z
Φ1 (z) =
2
z 3
Φ2 (z) = i +
2 4
z 3
Φ3 (z) = i − .
2 4

(Alternatively, one could use affine maps of R2 .)

(b) Because Φ consists of three maps, with the same contractivity factor 1/2, it
follows that the box dimension of A is log 3/ log 2. Since 3 is not a power of
2, this number is not an integer. Hence, by definition, A is a fractal.
 51

(c) # ---- utilities

PlotStuff:=style=point,symbol=point,scaling=constrained,axes=none:
z2ReIm:=z->[Re(z),Im(z)]:
# ---- contraction mappings and IFS
half,threq:=evalf(1/2),evalf(2/3):
f1,f2,f3:=z->z*half,z->I*z*half+threq,z->I*z*half-threq:
Phi:=S->map(f1,S) union map(f2,S) union map(f3,S):
# ---- iterations and plot
{0}:to 8 do Phi(%) od:
plot(map(z2ReIm,%),PlotStuff);