Representation of Period Doubling by

Digraphs and Characteristic Polynomials

arXiv:0909.0335v1 [math.DS] 2 Sep 2009

Richell O. Celeste
Institute of Mathematics
University of the Philippines
Diliman Quezon City
e-mail: [email protected]

Yoshifumi Takenouchi
Institute of Mathematics
University of the Philippines
Diliman Quezon City
e-mail: [email protected]

Abstract
A general procedure which defines a partial ordering of cyclic permutations
induced by continuous maps is known for constructing immediate successors
to a cycle. We expound on this procedure in terms of labelled digraphs and
characteristic polynomials then apply this study to period doubling, the most
common route to chaos for a nonlinear dynamical system.

Key words: forcing relation; digraph; spectrum; period doubling.

1 Introduction
Li and Yorke [10] showed that the presence of a periodic orbit with least period 3
forces the presence of periodic orbit with all possible least periods. However before
[10], Sharkovsky [6, 9] had already obtained the results which include the statement
above as a corollary. Baldwin [3] extended this idea to get a partial order on the set
of cycles in which the type of the periodic orbit is taken into account. Bernhardt
[5] showed how to construct immediate successors to a given cycle.
First we reexamine [5] to obtain graph theoretical explanation of the relation
between a cycle and its immediate successors. With this view point, we then obtain
the representation of the period-doubling bifurcation phenomena by the digraphs

1
consisting of signed vertices and directed edges and by the characteristic polynomials
of the adjacency matrices. Obtaining a discrete expression of the period-doubling
bifurcation phenomena implies that we may define the period-doubling bifurcation
phenomena on the graph itself without referring to the original one-dimensional
continuous maps.
In section 2 we present basic definitions and concepts on the forcing relation on
cycles in dimension one and on graph theory. In section 3 we obtain the relation
between the digraph and the characteristic polynomial of any cycle and that of its
immediate successor. In section 4 we describe the period-doubling phenomena of
the family of the logistic curves with their digraphs and characteristic polynomials.

2 Preliminaries
Definition 2.1. Let f : I → I be a continuous map of a compact interval I to itself.
We define f 0 (x) = x and for n ∈ N \ {0}, f n (x) = f (f n−1 (x)). If there exists k ∈ N
such that f k (x) = x, then we say x is periodic for f and x has (least) period s,
where s is smallest element of N such that f s (x) = x. For x ∈ I, the orbit of x is
the set {f n (x)|n ≥ 0}. If x is periodic with least period s, then the orbit of x is the
finite set O(x) = {x, f (x), f 2 (x), . . . , f s−1 (x)}.

Definition 2.2. A cycle (cyclic permutation) of length n is a bijection θ : {1, 2, . . . , n}

→ {1, 2, . . . , n} such that θk (1) 6= 1 for 1 ≤ k < n. We denote a cycle θ by
θ = (k1 , k2 , . . . , km ), where θ(ki ) = ki+1 and θ(km ) = k1 . We assume, without loss
of generality, that k1 = 1. Write the elements of a periodic orbit O in increasing
order: x1 < x2 < · · · < xs . We say O has orbit type θ if θ is a cycle of length s
and for each i ∈ {1, 2, . . . , s}, f (xi ) = xθ(i) .

(Sn , ◦) denotes the group of permutations on n objects, Cn denotes the subset

S of
Sn consisting of cycles (cyclic permutations) of length n. We also denote n≥1 Cn
by C.

Definition 2.3. The forcing relation on cycles is defined as follows: β ∈ C

forces α ∈ C if and only if every continuous map of the interval that has a periodic
orbit of type β has a periodic orbit of type α.

Baldwin [3] showed that this relation induces a partial order on the set of cycles
and provided an algorithm to decide when one cycle forces another.

Definition 2.4. [5] We say β ∈ C is an immediate successor to α ∈ C, or α is

an immediate predecessor to β, if β 6= α and the set of all cycles forced by β is
same as the set of all cycles forced by α and α itself.

2
Definition 2.5. For θ ∈ Sn , the canonical θ-linear map, or the connect-the-
dots map of θ is defined by Lθ : [1, n] → [1, n] where Lθ = θ on {1, 2, . . . , n} and
Lθ is linear on Ii := [i, i + 1] for i ∈ {1, 2, . . . , n − 1}.

It is known that β ∈ C forces α ∈ C if and only if Lβ has a periodic orbit of

cycle type α [2].

Definition 2.6. For θ ∈ Cn , n ≥ 3 we say θ is +k-modal cycle or −k-modal

cycle if the number of all the local extrema of Lθ is equal to k and θ(1) < θ(2) or
θ(1) > θ(2), respectively. In particular +1-modal cycle is called unimodal cycle.

The forcing relation defined by Baldwin induces a total order on the set of
unimodal cycles [7].

Definition 2.7. From Lθ , we define the labelled digraph (directed graph) of

θ, denoted by G(θ), as follows: G(θ) = (V, E) consists of the set of n − 1 signed
±
vertices V = {v1± , v2± , . . . , vn−1 } where we take vi+ if θ(i) < θ(i + 1) or vi− otherwise,
and the set of directed edges E = {eij } where eij a directed edge from vi to vj exists
if Ij ⊂ Lθ (Ii ).

Conversely, from G(θ) we can obtain Lθ . In this sense, we say they are equivalent.
Indeed Baldwin’s algorithm [3] which decides whether one cycle θ forces another or
not, is derived from G(θ).

Definition 2.8. Theh induced

i matrix of θ or the adjacency matrix of G(θ) [8],
denoted by M (θ) = ai,j , is the (n − 1) × (n − 1) matrix with (i, j)-th entry given
by 
1, if Ij ⊂ Lθ (Ii ),
ai,j =
0, otherwise,
where i, j ∈ [1, n − 1].

Definition 2.9. [8] A graph H = (W, F) is said to be a subgraph of the graph

G = (V, E) if W ⊂ V and F ⊂ E. In particular if F consists of all the edges from E
which connect the vertices from W, then H is called an induced subgraph.

Definition 2.10. [8] A cycle of length n in the digraph G is a subgraph with the
vertex set {v1 , . . . , vn } having directed edges from vi to vi+1 where i = 1, . . . , n − 1
and a directed edge from vn to v1 . In particular if v1 < v2 < · · · < vn then this cycle
induces the cyclic permutation θ = (12 · · · n).

3
3 Theorem
Definition 3.1. [5] Let θ ∈ Cn , then θ∗ ∈ S2n is defined by

θ∗ (2k) := 2θ(k), θ∗ (2k − 1) := 2θ(k) − 1,

where k ∈ {1, 2, . . . , n}.

Definition 3.2. [5] Let ρs denote the transposition (2s − 1, 2s), where s ∈ N. Then
for θ ∈ Cn define η as follows:

η := θ∗ ◦ ρi1 ◦ ρi2 ◦ · · · ◦ ρim ,

| {z }
m

where 1 ≤ i1 < i2 < · · · < im ≤ n.

If m is odd, then η ∈ C2n and it is an immediate successor to θ. If m is even,
then η ∈ S2n \ C2n . There are 2n−1 immediate successors η ∈ C2n to θ, and 2n−1
non-cyclic permutations η ∈ S2n . (See [4] and [5])

For θ and η above, denote their labelled digraphs by G(θ) and G(η), respectively.
±
Then G(θ) = (V, E) consists of the set of n−1 signed vertices V = {v1± , v2± , . . . , vn−1 },
+ −
where we take vi if θ(i) < θ(i + 1) or vi otherwise, and the set of directed edges
E = {eij } where eij a directed edge from vi to vj exists if Ij ⊂ Lθ (Ii ). G(η) = (W, F)
consists of the set of 2n − 1 signed vertices W = {w1± , w2± , . . . , w2n−1±
}, where we
+ −
take wi if η(i) < η(i + 1) or wi otherwise, and the set of directed edges F = {fij }
where fij a directed edge from wi to wj exists if Ij ⊂ Lη (Ii ). Let us divide W into
±
two subsets W1 = {w2± , . . . , w2n−2 ±
} and W2 = {w1± , w3± , . . . , w2n−1 }.

Then we have the following observations:

(i) G(η)|W1 , the induced subgraph of G(η) = (W, F) determined on W1 is iso-

morphic to the digraph G(θ) = (V, E). Indeed, the isomorphism ϕ : V → W1 is
given by
ϕ(vi± ) = w2i
±

where i ∈ {1, 2, . . . , n − 1}. Observe that the sign of vertex vi is always consistent
with that of vertex w2i .

(ii) In G(η) = (W, F) there does not exist any directed edge from W2 to W1 .

(iii) G(η)|W2 , the induced subgraph of G(η) = (W, F) determined on W2 , con-

sists of an n-cycle induced by θ. Indeed, this n-cycle is given by

4
 t
? - t - t - ......... - t
w1 w2θ(1)−1 w2θ2 (1)−1 w2θn−1 (1)−1

and the type of this n-cycle is also θ.

Together, (i), (ii), and (iii) provide a graph theoretical explanation of the reason
why η is an immediate successor to θ (see Lemma 2.4 in [5]). As a consequence, we
have the following theorem.

Theorem 3.1. Denote the characteristic polynomials of the adjacency matrices of

G(θ), G(η), G(η)|W1 and G(η)|W2 by Pθ (λ), Pη (λ), PW1 (λ) and PW2 (λ), respectively.
Then Pη (λ) is given by the product of PW1 (λ) and PW2 (λ), namely

Pη (λ) = Pθ (λ) · (λn − 1).

As a corollary of this theorem, we have the following.

Corollary 3.1. The spectrum of G(η) consists of the spectrum of G(θ) and the n
distinct roots of unity.

Assume that the adjacency matrix of G(θ) is given by

" #
M (θ) = ai,j ,
(n−1)×(n−1)

where 1 ≤ i, j ≤ n − 1, and the adjacency matrix of G(η) is given by

" #
M (η) = bi,j ,
(2n−1)×(2n−1)

where 1 ≤ i, j ≤ 2n − 1.
Then we have the following Lemma 3.1, and 3.2.

Lemma 3.1. G(η)|W1 is isomorphic to G(θ) by ϕ, namely

b2i,2j = ai,j ,

where 1 ≤ i, j ≤ n − 1.

5
 Proof. For any θ ∈ Cn , θ∗ ∈ S2n is defined by
 
∗ ··· 2i − 1 2i 2i + 1 2i + 2 ···
θ := .
· · · 2θ(i) − 1 2θ(i) 2θ(i + 1) − 1 2θ(i + 1) · · ·
If η := θ∗ ◦ · · · ◦ ρi ◦ · · · , then
 
· · · 2i − 1 2i 2i + 1 2i + 2 ···
η= ,
· · · 2θ(i) 2θ(i) − 1 2θ(i + 1) − 1 2θ(i + 1) · · ·
and 
[2θ(i) − 1, 2θ(i + 1) − 1] if θ(i) < θ(i + 1)
Lη ([2i, 2i + 1]) = .
[2θ(i + 1) − 1, 2θ(i) − 1] if θ(i) > θ(i + 1)
If η := θ∗ ◦ · · · ◦ ρi+1 ◦ · · · , then
 
··· 2i − 1 2i 2i + 1 2i + 2 ···
η= ,
· · · 2θ(i) − 1 2θ(i) 2θ(i + 1) 2θ(i + 1) − 1 · · ·
and 
[2θ(i), 2θ(i + 1)] if θ(i) < θ(i + 1)
Lη ([2i, 2i + 1]) = .
[2θ(i + 1), 2θ(i)] if θ(i) > θ(i + 1)
If η := θ∗ ◦ · · · ◦ ρi ◦ ρi+1 ◦ · · · , then
 
· · · 2i − 1 2i 2i + 1 2i + 2 ···
η= ,
· · · 2θ(i) 2θ(i) − 1 2θ(i + 1) 2θ(i + 1) − 1 · · ·
and 
[2θ(i) − 1, 2θ(i + 1)] if θ(i) < θ(i + 1)
Lη ([2i, 2i + 1]) = .
[2θ(i + 1), 2θ(i) − 1] if θ(i) > θ(i + 1)
If η := θ∗ ◦ ρi1 ◦ ρi2 ◦ · · · ◦ ρim does not include ρi nor ρi+1 , then
 
··· 2i − 1 2i 2i + 1 2i + 2 ···
η= ,
· · · 2θ(i) − 1 2θ(i) 2θ(i + 1) − 1 2θ(i + 1) · · ·
and 
[2θ(i), 2θ(i + 1) − 1] if θ(i) < θ(i + 1)
Lη ([2i, 2i + 1]) = .
[2θ(i + 1) − 1, 2θ(i)] if θ(i) > θ(i + 1)
Under the assumption θ(i) < θ(i + 1), we have
ai,j = 1
⇔ Ij ⊂ Lθ (Ii )
⇔ [j, j + 1] ⊂ [θ(i), θ(i + 1)]
⇔ θ(i) ≤ j and j + 1 ≤ θ(i + 1)
⇔ 2θ(i) ≤ 2j and 2j + 1 ≤ 2θ(i + 1) − 1
⇔ [2j, 2j + 1] ⊂ [2θ(i), 2θ(i + 1) − 1].

6
Here we note that

[2θ(i), 2θ(i + 1) − 1] ⊂ [2θ(i) − 1, 2θ(i + 1) − 1], [2θ(i), 2θ(i + 1)] ⊂ [2θ(i) − 1, 2θ(i + 1)],

and
[ 2j , 2j + 1] 6= [2θ(i) − 1, 2θ(i)], [2θ(i + 1) − 1, 2θ(i + 1)],
even odd odd even odd even

thus,
[2j, 2j + 1] ⊂ [2θ(i), 2θ(i + 1) − 1]
⇔ [2j, 2j + 1] ⊂ [η(2i), η(2i + 1)]
⇔ I2j ⊂ Lη (I2i )
⇔ b2i,2j = 1.
Similarly, under the assumption θ(i) > θ(i + 1), we can also show that

ai,j = 1 ⇔ b2i,2j = 1.

This concludes the proof of the lemma. 

Lemma 3.2. In G(η) = (W, F), there does not exist any directed edge from W2 to
W1 , that is
b2i−1,2j = 0,
where 1 ≤ i ≤ n, 1 ≤ j ≤ n − 1. Moreover, G(η)|W2 consists of an n-cycle induced
by θ, namely 
b2i−1,2j−1 = 1 if j = θ(i),
b2i−1,2j−1 = 0 otherwise,
where 1 ≤ i, j ≤ n.
In other words
···
 
b12 b14 b1(2n−2)

 b32 b34 ··· b3(2n−2) 


 ··· ··· ··· ···  = On×(n−1) ,

 ··· ··· ··· ··· 
b(2n−1)2 b(2n−1)4 ··· b(2n−1)(2n−2)

and
···
   
b11 b13 b1(2n−1) c11 c12 ··· c1n

 b31 b33 ··· b3(2n−1)  
  c21 c22 ··· c2n 


 ··· ··· ··· ··· =
  ··· ··· ··· ··· 

 ··· ··· ··· ···   ··· ··· ··· ··· 
b(2n−1)1 b(2n−1)3 ··· b(2n−1)(2n−1) cn1 cn2 ··· cnn

7
 ⇐⇒ b2i−1,2j−1 = cij ,
where 
cij = 1 if j = θ(i),
cij = 0 otherwise.

Proof.
j = θ(i)
⇔ θ∗ (2i − 1) = 2j − 1 and θ∗ (2i) = 2j
⇔ [2j − 1, 2j] = [θ∗ (2i − 1), θ∗ (2i)].
For any θ ∈ Cn , θ∗ ∈ S2n is defined by
 
∗ ··· 2i − 1 2i ···
θ := .
· · · θ∗ (2i − 1) θ∗ (2i) · · ·

If η := θ∗ ◦ · · · ◦ ρi ◦ · · · , then
 
··· 2i − 1 2i ···
η= ,
··· θ∗ (2i) θ∗ (2i − 1) · · ·

and thus [2j − 1, 2j] = [θ∗ (2i − 1), θ∗ (2i)] = [η(2i), η(2i − 1)].
If η := θ∗ ◦ ρi1 ◦ ρi2 ◦ · · · ◦ ρim does not include ρi , then
 
··· 2i − 1 2i ···
η= ,
· · · θ∗ (2i − 1) θ∗ (2i) · · ·

and thus [2j − 1, 2j] = [θ∗ (2i − 1), θ∗ (2i)] = [η(2i − 1), η(2i)].
In either event, we have

[2j − 1, 2j] = [θ∗ (2i − 1), θ∗ (2i)]

⇔ 
I2j−1 = Lη (I2i−1 )
b2i−1,2j−1 = 1 ,
⇔
b2i−1,k = 0 if k 6= 2j − 1,

ending the proof of the lemma. 

The following statement is obvious:

Lemma 3.3.
Al×l Ol×m
= |Al×l ||Bm×m |.
Cm×l Bm×m

We are now ready to prove the theorem.

8
Proof of Theorem 3.1. The characteristic polynomial of the adjacency matrix of
G(θ) is given by
λ − a11 −a12 ··· −a1(n−1)
−a21 λ − a22 ··· −a2(n−1)
··· ··· ··· ···
Pθ (λ) = |λI − M (θ)| = .
··· ··· ··· ···
··· ··· ··· ···
−a(n−1)1 −a(n−1)2 ··· λ − a(n−1)(n−1)
Let
 
1 2 3 ··· n n + 1 n + 2 n + 3 ··· 2n − 1
σ2n−1 := .
1 3 5 ··· 2n − 1 2 4 6 ··· 2n − 2
Using Lemmas 3.1, 3.2, and 3.3, the characteristic polynomial Pη (λ) of η is thus
λ − b11 −b12 ··· ··· −b1(2n−2) −b1(2n−1)
−b21 λ − b22 ··· ··· −b2(2n−2) −b2(2n−1)
··· ··· ··· ··· ··· ···
=
··· ··· ··· ··· ··· ···
−b(2n−2)1 −b(2n−2)2 ··· ··· λ − b(2n−2)(2n−2) −b(2n−2)(2n−1)
−b(2n−1)1 −b(2n−1)2 ··· ··· −b(2n−1)(2n−2) λ − b(2n−1)(2n−1)

λ − b11 −b12 ··· ··· −b1(2n−2) −b1(2n−1)

··· ··· ··· ··· ··· ···
−b(2n−2)1 −b(2n−2)2 ··· ··· λ − b(2n−2)(2n−2) −b(2n−2)(2n−1)
= sgn(σ2n−1 )
−b21 λ − b22 ··· ··· −b2(2n−2) −b2(2n−1)
··· ··· ··· ··· ··· ···
−b(2n−1)1 −b(2n−1)2 ··· ··· −b(2n−1)(2n−2) λ − b(2n−1)(2n−1)

λ − b11 ··· −b1(2n−1) −b12 ··· −b1(2n−2)

··· ··· ··· ··· ··· ···
−b(2n−1)1 ··· λ − b(2n−1)(2n−1) −b(2n−1)2 ··· −b(2n−1)(2n−2)
= (sgn(σ2n−1 ))2
−b21 ··· −b2(2n−1) λ − b22 ··· −b2(2n−2)
··· ··· ··· ··· ··· ···
−b(2n−2)1 ··· −b(2n−2)(2n−1) −b(2n−2)2 ··· λ − b(2n−2)(2n−2)

λ − b11 · · · −b1(2n−1) λ − b22 · · · −b2(2n−2)

= ··· ··· ··· ··· ··· ···
−b(2n−1)1 · · · λ − b(2n−1)(2n−1) −b(2n−2)2 · · · λ − b(2n−2)(2n−2)

= (λn + sgn(θ)(−1)n )Pθ (λ)

9
since (sgn(σ2n−1 ))2 = 1,
 
−b12 ··· −b1(2n−2)
 ··· ··· ···  = On×(n−1) ,
−b(2n−1)2 · · · −b(2n−1)(2n−2)

λ − b22 · · · −b2(2n−2) λ − a11 · · · −a1(n−1)

··· ··· ··· = ··· ··· ··· = Pθ (λ),
−b(2n−2)2 · · · λ − b(2n−2)(2n−2) −a(n−1)1 · · · λ − a(n−1)(n−1)
and
λ − b11 · · · −b1(2n−1)
··· ··· ···
−b(2n−1)1 · · · λ − b(2n−1)(2n−1)

λ − c11 · · · −c1n 
cij = 1 if j = θ(i),
= ··· ··· ··· where
cij = 0 otherwise.
−cn1 · · · λ − cnn

= sgn(ι) · (λ − 0)n + sgn(θ) · (−1)n

= λn + sgn(θ)(−1)n .
Here we note that sgn(θ)(−1)n is always equal to −1 since

n (+1) · (−1) if n is odd,
sgn(θ)(−1) =
(−1) · (+1) if n is even.

This completes the proof of the theorem. 

4 Immediate Successors and Period Doubling

4.1 The logistic map
The logistic map [1, 9] is written as

xn+1 = axn (1 − xn ) (1)

where the current value xn , is mapped onto the next value xn+1 . Here, we restrict
our considerations to 0 ≤ xn ≤ 1 and 0 ≤ a ≤ 4. The corresponding function to the
logistic map, called the logistic function or the logistic curve, is given by

fa (x) = ax(1 − x) (2)

10
By repeatedly iterating the logistic map forward through time, we may observe
different behaviors of the iterated solutions. The sequence of iterated solutions to
the map is called an orbit. The behavior of orbits, which originate from typically
chosen initial condition used for the iterations, depends on the control parameter a.
The logistic curve is parabolic like the quadratic function with fa (0) = fa (1) = 0
and a maximum at x = 1/2. Being a smooth curve with only one critical point,
that is, a local maximum, it is classified as a unimodal (single humped) map.

Figure 1. Bifurcation diagram of the logistic map

4.2 Period-doubling bifurcations

The logistic map has two fixed points: 0 and (a − 1)/a. If a < 1, then
|fa0 (0)| < 1 and we say that the fixed point x = 0 is stable for a < 1. If a > 1, then
|fa0 (0)| > 1 and so the fixed point x = 0 is now unstable and a new stable fixed point
x = (a − 1)/a emerges. The fixed point x = (a − 1)/a is stable for 1 < a < 3. At
a = 3, the fixed point x = (a − 1)/a√ becomes unstable and a stable
√ 2-cycle (period
2 2
2 attractor) {x, y} = {(a + 1 − a − 2a − 3)/2a, (a + 1 + a − 2a − 3)/2a} is
said to bifurcate from this fixed point. The above-mentioned 2-cycle is the solution
of the system
√ of equations given by y = ax(1 − x) and x = ay(1 − y). When
a = 1 + 6 ≈ 3.44949, this 2-cycle loses its stability and a stable 4-cycle begins
followed by a stable 8-cycle (between a ≈ 3.54409 and a ≈ 3.56440). The period
continues doubling over shorter intervals until a ≈ 3.56995
√ where the chaotic regime
takes over. A period 3 attractor is born at a = 1 + 8 ≈ 3.82843 [11], causing a
notably large window in the bifurcation diagram (see Figure 1). This 3-cycle also
undergoes a period doubling cascade in which period 3·2` attractors are successively
produced until chaos. This mechanism by which each (k2` )-cycle loses its stability
and is replaced by a stable (k2`+1 )-cycle is called period-doubling bifurcation
(see Figure 2).

11
 Figure 2. Magnification of the first few period doubling
To determine period 4, period 8, and in general, period n solutions of the logistic
map, we solve the system of difference equations, x1 = x2 , x2 = x3 , . . . , and xn = x1 .
We note that these computations become very complicated even for relatively small
values of n. Finding the location of these bifurcation points on the bifurcation
diagram also becomes very difficult. Even so, examining the types of orbits and
their characteristic polynomials turns out to be less tortuous. This is illustrated in
the following sections.

4.3 Period-doubling bifurcation beginning with nontrivial fixed point

The attracting period 2 orbit of type θ1 = (12) bifurcates from the fixed point of
type θ0 = (1) at a = 3. θ1 is the only one immediate successor to θ0 , and it follows
that
Pθ1 (λ) = λ − 1
= (λ − 1) · Pθ0 (λ),
if we define Pθ0 (λ) := 1 for θ0 = (1). Indeed the graph G(θ1 ) = (V1 , E1 ) consists
of a vertex v11 with the minus sign and a directed edge from v11 to itself (a loop
associated with v11 ):

t ?
−

v11

The attracting period 4 √ orbit of type θ2 = (1324) bifurcates from the period 2
orbit of type θ1 at a = 1 + 6 ≈ 3.44949. θ2 with modality +1 is one of 22−1 = 2
immediate successors to θ1 . Here we note that (1423) with modality −1 is another
immediate successor to θ1 , however, the logistic map can realize only +1-modal
(unimodal) cycles. Now it follows that
Pθ2 (λ) = λ3 − λ2 − λ + 1
= (λ2 − 1)(λ − 1)
= (λ2 − 1) · Pθ1 (λ).

12
Indeed the graph G(θ2 ) = (V2 , E2 ) is given as follows:

+
v21
t









t


t ?
− −
v23 
v22

+ − − − + −
where V2 = {v21 , v22 , v23 }, V21 = {v22 }, and V22 = {v21 , v23 }. Here we note that
(i) G(θ2 )|V21 is isomorphic to G(θ1 ), the sign of vertex v11 is consistent with that of
vertex v22 , (ii) there are no directed edges from V22 to V21 , (iii) G(θ2 )|V22 consists
of a 2-cycle induced by θ1 .
The attracting period 8 orbit of type θ3 = (15472638) bifurcates from the period
4 orbit of type θ2 at a ≈ 3.54409. θ3 with modality +1 is one of 24−1 = 8 immediate
successors to θ2 . There are other 7 immediate successors to θ2 , however, their
modalities are not equal to +1. Now it follows that

Pθ3 (λ) = λ7 − λ6 − λ5 + λ4 − λ3 + λ + λ − 1
= (λ4 − 1)(λ2 − 1)(λ − 1)
= (λ4 − 1) · Pθ2 (λ).

Indeed the graph G(θ3 ) = (V3 , E3 ) is given as follows:

− +
v37 v31
t - t
Y
HHH
o
SS HH
S  HH
−  HH v +
v36 t
S
 S - t 32
I
@
@  S
@  S
 S
−  @ −
v35 t
/
 @ -S t v33
YH @
H
 t
HH @
−
v34 ?

+ + − − − − − + − − + − − −
where V3 = {v31 , v32 , v33 , v34 , v35 , v36 , v37 }, V31 = {v32 , v34 , v36 }, and V32 = {v31 , v33 , v35 , v37 }.
Here we note that (i) G(θ3 )|V31 is isomorphic to G(θ2 ), the sign of vertex v3(2j) is
consistent with that of vertex v2j (j = 1, 2, 3), (ii) there are no directed edges from
V32 to V31 , (iii) G(θ3 )|V32 consists of a 4-cycle induced by θ2 .

13
 Inductively, for the attracting period 2` orbit of type θ` with modality +1, which
`−1
is one of 2(2 −1) immediate successors to θ`−1 , and bifurcated from the period 2`−1
orbit of type θ`−1 , we obtain G(θ` ) and Pθ` (λ).

4.4 Period-doubling bifurcation beginning with period 3

Next consider the period-doubling bifurcation beginning with period 3 orbit of
type θ0 = (123). The graph G(θ0 ) = (V0 , E0 ) consists of two vertices with a 2-cycle
and a loop:


t - t
+ −
v01 
v02 6

and thus Pθ0 (λ) = λ2 − λ − 1.

The attracting period 6 orbit of type θ1 = (135246) bifurcates from the period 3
orbit of type θ0 = (123) at a ≈ 3.8415. θ1 with modality +1 is one of 23−1 = 4 im-
mediate successors to θ0 . (146235), (136245), (145236) are also immediate successors
to θ0 , however, their modalities are not equal to +1. Now it follows that

Pθ1 (λ) = λ5 − λ4 − λ3 − λ2 + λ + 1
= (λ3 − 1)(λ2 − λ − 1)
= (λ3 − 1) · Pθ0 (λ)

Indeed the graph G(θ1 ) = (V1 , E1 ) is given as follows:

− +
v15 v11
t - t
 CO 


C
C 
−  +
v14 t C  - t v12

6 Q C 
Q C 
Q
Q C 
s t
Q C 
Q
+
v13

+ + + − − + − + + −
where V1 = {v11 , v12 , v13 , v14 , v15 }, V11 = {v12 , v14 }, and V12 = {v11 , v13 , v15 }. Here
we note that (i) G(θ1 )|V11 is isomorphic to G(θ0 ), the sign of vertex v1(2j) is consistent
with that of vertex v0j (j = 1, 2), (ii) there are no directed edges from V12 to V11 ,
(iii) G(θ1 )|V12 consists of a 3-cycle induced by θ0 .

14
 Inductively, for the attracting period (3 · 2` ) orbit of type θ` with modality +1,
`−1
which is one of 2(3·2 −1) immediate successors to θ`−1 , and bifurcated from the
period 2`−1 orbit of type θ`−1 , we obtain G(θ` ) and Pθ` (λ).

4.5 Period-doubling bifurcation beginning with period k

It is known that periodic attractors exist for all odd integers k. Therefore, in
general, for the attracting period (k · 2` ) orbit of type θ` with modality +1, which is
`−1
one of 2(k·2 −1) immediate successors to θ`−1 , and bifurcated from the period 2`−1
orbit of type θ`−1 , it follows that
`
i−1
Y
Pθ` (λ) = Pθ0 (λ) (λ(k·2 ) − 1),
i=1

where Pθ0 (λ) is the characteristic polynomial of the induced matrix of θ0 , and θ0 is
the type of the original period k orbit.
Indeed the graph G(θ` ) consists of the set of vertices V` and the set of directed
edges E` where V` = {v`1 , v`2 , v`3 , . . . , v`(k·2` −2) , v`(k·2` −1) }, V`1 = {v`2 , . . . , v`(k·2` −2) },
and V`2 = {v`1 , v`3 , . . . , v`(k·2` −1) }. Here we note that (i) G(θi )|Vi1 ∼ = G(θi−1 ) for
i = 1, 2, . . . , `, and the signs of the vertices are preserved by each of the above iso-
morphisms, (ii) there are no directed edges from Vi2 to Vi1 for i = 1, 2, . . . , `, (iii)
G(θi )|Vi2 consists of a (k · 2i−1 )-cycle induced by θi−1 for i = 1, 2, . . . , `, and G(θ0 )
consists of k − 1 vertices and some directed edges.
In other words, G(θ` ) essentially consists of k − 1 vertices with some directed
edges, a k-cycle of type θ0 , a 2k-cycle of type θ1 , · · · , a (k · 2`−1 )-cycle of type θ`−1 .
That is the reason why Pθ` (x) can be expressed as a product of Pθ0 (λ), λk − 1,
`−1
λ2k − 1, · · · , λ(k·2 ) − 1. Thus the spectrum of G(θ` ) consists of the spectrum of
i−1
G(θ0 ) and (k · 2i−1 ) distinct roots of λ(k·2 ) − 1 = 0 where i = 1, 2, . . . , `.
Remark 4.1. Conversely, we may define G(θ` ), the doubling of the graph G(θi−1 )
by properties (i), (ii), and (iii).

5 Acknowledgement
The authors wish to thank Professor Evelyn L. Tan for referring to us [8].

References
[1] Paul S. Addison, Fractals and Chaos An Illustrated Course, IOP Publishing Ltd
(1997).

15
[2] L. Alseda, J. Llibre, M. Misiurewicz, Combinatorial Dynamics and Entropy in
Dimension One, 2nd ed., World Sci., (2000).

[3] S. Baldwin, Generalizations of a theorem of Sarkovskii on orbits of continuous

real-valued functions, Discrete Mathematics, Vol 67 Issue 2 (1987), 111-127.

[4] C. Bernhardt, Simple permutations with order a power of two, Ergod Th. &
Dynam. Sys., 4, 179 - 186 (1984).

[5] C. Bernhardt, The ordering on permutations induced by continuous maps of the

real line, Ergod Th. & Dynam. Sys., 7, 155 - 160 (1987).

[6] K. Burns, B. Hasselblatt, Sharkovsky’s Theorem, (preprint), (2007).

[7] P. Collet and J.-P. Eckmann, Iterated Maps on the Interval as Dynamical Sys-
tems, Progress in Physics, vol. 1, Birkhauser, Boston, Mass., (1980)

[8] D. M. Cvetković, M. Doob and H. Sachs, Spectra of Graphs: Theory and Appli-
cations, (3rd rev. enl. ed.), New York, Wiley, (1998).

[9] P. G. Drazin, Nonlinear systems (Cambridge Texts in Applied Mathematics),

Cambridge University Press (1997).

[10] T.-Y. Li, J.A. Yorke, Period three implies Chaos, Amer. Math. Monthely, 82
no.10 (1975), 985-992.

[11] P. Saha, S.H. Strogaz, The Birth of Period Three, Mathematics Magazine, Vol.
68 No 1 Feb. (1995), 42-47.

16