Lecture notes #10 Nonlinear Dynamics YFX1560

Lecture 10: 1-D maps, the Lorenz map, the logistic map,
sine map, period doubling bifurcation, tangent bifurca-
tion, transient and intermittent chaos in maps, orbit di-
agram (or the Feigenbaum diagram), the Feigenbaum con-
stants, universals of unimodal maps, universal route to
chaos
Contents
1 The Lorenz map 2
1.1 Cobweb diagram and map iterates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Comparison of fixed point x∗ in 1-D continuous-time systems and 1-D discrete-time maps . 3
1.3 Period-p orbit and stability of period-p points . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 A proper introduction to 1-D maps 6

3 The logistic map 6

3.1 The Lyapunov exponent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Bifurcation analysis and period doubling bifurcation . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Orbit diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Tangent bifurcation and odd number period-p points . . . . . . . . . . . . . . . . . . . . . . 14

4 Sine map and universality of period doubling 16

5 Universal route to chaos 17

Handout: Orbit diagram of the logistic map.

D. Kartofelev 1/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

1 The Lorenz map

In this lecture we continue to investigate the possibility that the Lorenz attractor might be long-
term periodic. As in previous lecture, we use the one-dimensional Lorenz map in the form:
zn+1 = f (zn ), (1)
where f was incorrectly assumed to be a continuous function, to gain insight into the continuous time three-
dimensional flow of the Lorenz attractor given by


ẋ = d(y − x),
ẏ = rx − y − xz, (2)


ż = xy − bz,
where parameters r = 28, d = 10, and b = 8/3 are the control parameters that result in the system being
in a chaotic regime—more accurately, these parameters lead to the strange attractor introduced last week.

1.1 Cobweb diagram and map iterates

The cobweb diagram, also called the cobweb plot, introduced in the previous lecture is a graphical way
of thinking about and determining the iterates of maps including the Lorenz map, see Fig. 1. In order to
construct a cobweb plot move vertically to function f (z) and register the function value, move
horizontally to the diagonal and repeat, see Lecture 9 for basic examples.

Zz

it iii

it
z

Figure 1: (Left) Cobweb diagram of the Lorenz map. (Right) The Lorenz map iterates corresponding to
the cobweb diagram shown on the left.

I
In the previous lecture we showed, using the linearisation procedure, that the fixed point z ∗ (period-1
point) satisfying
I f (z ∗ ) = z ∗ , (3)
A IFF
where f (z) is the function defining the Lorenz map, is unstable. The general conditions for fixed point
stability were shown to be dependant on the absolute value of map function slope evaluated at given fixed
point z ∗ . For
|f 0 (z ∗ )| < 1, (4)
the fixed point is stable. For
|f 0 (z ∗ )| = 1, (5)
the fixed point is participating in a bifurcation. An example of this is shown below in Sec. 3.2. For
|f 0 (z ∗ )| > 1, (6)
the fixed point is unstable.
The following interactive numerical file shows the Lorenz map, its cobweb diagram and iterates. It can
be used to confirm the fact that the Lorenz map’s fixed point z ∗ is indeed unstable.

D. Kartofelev 2/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Numerics: nb#1
Cobweb diagram and iterates of the (generalised) Lorenz map. Lyapunov exponent λ(r) of the logistic
map.

1.0 1.0

0.5 0.5

zmax
zn+1

0.0 0.0
Zz

-0.5

it
-0.5

-1.0
-1.0 -0.5 0.0
zn
0.5
iii
1.0
it
-1.0
0 2 4
n
6

Figure 2: (Left) Cobweb diagram of the normalised Lorenz map showing the first ten iterates for z0 ≈ z ∗
8 10

z since |f 0 (z ∗ )| > 1. The Lorenz map has property |f 0 (z)| > 1, ∀z in

where the fixed point z ∗ is unstable,
the map basin. (Right) Lorenz map iterates corresponding to the cobweb diagram shown on the left.

I
A
I
IFF

Figure 3: (Top-left) Stable fixed point x∗ of a one-dimensional system given by ẋ = f (x) where f 0 (x∗ ) < 0 is
shown with the filled bullet. (Top-right) A family of time-series solutions shown for several initial conditions
positioned close to x∗ . The solutions correspond to the phase portrait shown on the left. (Bottom-left)
Unstable fixed point x∗ shown with the empty bullet where f 0 (x∗ ) > 0. (Bottom-right) A family of time-
series solutions shown for several initial conditions positioned close to x∗ . The solutions correspond to the
phase portrait shown on the left.

1.2 Comparison of fixed point x∗ in 1-D continuous-time systems and 1-D discrete-
time maps
The fixed points of one-dimensional continuous-time systems or in other words ordinary differential equa-
tions (ODEs) and of discrete-time one-dimensional maps are clearly analogous. Below we’ll start to distin-
guish fixed points of maps by their type. The type of the fixed point mentioned here will be referred to as
so called period-1 point. Figures 3 and 4 show that analogy and other similarities.

D. Kartofelev 3/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

iti i

i
i
iti
T
j

ii
i t
i

Figure 4: (Top-left) Stable fixed point x∗ (a period-1 point) of a one-dimensional map given by xn+1 =
f (xn ) where |f 0 (x∗ )| < 1 is shown with the filled bullet. (Top-right) Three sets of map iterates xn where

i i
n ∈ [0, 3] shown for initial conditions x01 , x∗ and x02 corresponding to the cobweb diagram shown on the left.
∗ o 0 ∗
(Bottom-left) Unstable fixed point T x (a period-1 point) shown with the empty∗ bullet where |f (x )| > 1.
j iterates xn shown for initial conditions x01 , x and x02 corresponding to

t
(Bottom-right) Three sets of map
the cobweb diagram shown on thei left. t

i
i

1.3 Period-p orbit and stability of period-p points 1

PERIOD PERIOD
We ended the previous lecture with two open ended questions: Can trajectories in a cobweb diagram of the
Lorenz map close onto themselves? And, how does a closed trajectory in the cobweb plot translate into the
three-dimensional continuous-time Lorenz flow?

i t i

1
PERIOD PERIOD

Figure 5: (Left) Period-4 orbit, shown with the blue graph, where zn+4 = zn . (Right) The Lorenz map
iterates zn where n ∈ [0, 8] corresponding to the cobweb diagram shown on the left. The map iterated
values repeat every four iterates.

So, can trajectories in a cobweb diagram of the Lorenz map close onto themselves? We could imagine a
trajectory of the Lorenz map’s cobweb plot closing onto itself in a manner shown in Fig. 5. Visual inspection
of the cobweb diagram shown in Fig. 5 reveals that the iterates zn (local maxima of the three-dimensional
Lorenz flow) repeat themselves such that z4 = z0 or more generally

zn+4 = zn , ∀n, (7)

D. Kartofelev 4/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

if this dynamics can be shown to be possible, then it would strongly suggest that a limit-cycle might be
possible in the three-dimensional Lorenz attractor. This conclusion can be generalised further:

zn+p = zn , ∀n, (8)

where p ∈ Z+ is the period of a limit-cycle. This type of fixed point is called the period-p point and
it represents the period-p orbit of the map. Period-p points are a new type of fixed points or in
other words a new type of dynamics. A period-1 point coincides with our old friend—fixed point z ∗ . The
period-p points where p > 1 don’t have corresponding analogies in one-dimensional continuous systems in
the manner as was discussed and shown in Sec. 1.2. This is because the period-p points represent oscil-
latory limit-cycle solutions which are not possible in one-dimensional systems, cf. Fig. 3 (see Lecture 2:
Impossibility of oscillations in 1-D systems).
It is natural to assume that period-p points or orbits, very much like fixed points z ∗ (a period-1 point),
can be either stable (attracting trajectories) or unstable (repelling trajectories). If we can show
that stable period-p orbits are possible in the Lorenz map, then that would strongly suggest a possibility
of periodic solutions in the continuous-time Lorenz attractor.
Let’s find the analytical expression for the relationship between zn+p and zn in (8). The n-th iterate of a
selected initial condition z0
zn = f (f (f (. . . f (z0 ) . . .))) ≡ f n (z0 ), (9)
| {z }
n times

where fn is the n-th iterate map—the map applied to itself n times. Don’t confuse this notation with
raising the map function to the n-th power. The subsequent iterates of the closed period-p orbits, similar
to the one shown in Fig. 5 with the blue trajectory, expressed analytically are

z1 = f (z0 ) ≡ f 1 (z0 )
z2 = f (z1 ) = f (f (z0 )) ≡ f 2 (z0 )
z3 = f (z2 ) = f (f (f (z0 ))) ≡ f 3 (z0 )
(10)
z4 = f (z3 ) = f (f (f (f (z0 ))) ≡ f 4 (z0 )
..
.
zn+p = f p (zn ),

where z is the period-p point in the period-p orbit and f p is the p-th iterate map.
Definition: z is a period-p point if equation

f p (z) = z, (11)

where p is minimal, is satisfied.

The stability of a period-p point is determined via linearisation. For simplicity we consider stability of
period-2 point
f 2 (z) ≡ f (f (z)) = z. (12)
Note that z is a period-2 point for map f but a fixed point (a period-1 point) for map f 2 . In the previous
lecture we showed that the stability of fixed point z ∗ depends on the slope of the map evaluated at that
point. Small perturbations |η|  1 evolve according to

ηn+1 ≈ |f 0 (z ∗ )|ηn . (13)

We find
 
chain rule +
d  orbit points 
|(f 2 (z))0 | = |f (f (z))| =   0
 zn+2 = zn ,  = |f (f| (z
0 0 0
0 )) · f (z0 )| = |f (z1 )| · |f (z0 )| > 1, (14)
dz {z }
z1
and we select z0

D. Kartofelev 5/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

because for the Lorenz map |f 0 (z)| > 1, ∀z in the map basin. The period-2 point is thus unstable. The
evolution of perturbations defined by (13) generalised to all period-p points in any closed period-p orbit are
the following:
p−1
Y
ηn+p ≈ f 0 (zn+k ) ηn , (15)
k=0
here, again, by the Lorenz map property:
p−1
Y
f 0 (zn+k ) > 1. (16)
k=0

Thus, all period-p points are unstable. The above analysis of the Lorenz map has strongly demonstrated
(not proven) that periodic solutions of the Lorenz system are not possible and that the flow is indeed long-
term aperiodic.

2 A proper introduction to 1-D maps

This section deals with a new class of dynamical systems, introduced in Lecture 9 and used up to now
without a proper introduction. In maps time is discrete, rather than continuous. These systems are
known variously as recursion relations, iterated maps, or simply maps. The Lorenz map is such a
system. When we say “map,” do we mean the function f or the recursion relation in the following form:

xn+1 = f (xn )? (17)

Following common usage, we’ll call both of them maps. If you’re disturbed by this, you must be a pure
mathematician... or should consider becoming one! Fixed point x∗ of one-dimensional map (17) satisfies
Eq. (3) and period-p point x satisfies Eq. (11) for minimal p.
Maps arise in various ways:
1. As tools for analysing differential equations. We have already encountered maps in this role. For
instance, the Lorenz map provided evidence that the Lorenz attractor is aperiodic, and is not just a
long-period limit-cycle. In future lectures the Poincaré maps will allowed us to prove the existence
of a periodic solutions, and to analyse the stability of periodic solutions in general. Maps will prove
to be superb tools for studying and analysing chaotic systems.
2. As models of natural phenomena. In some scientific contexts it is natural to regard time as
discrete. This is the case in digital electronics, in parts of economics and finance theory, in impulsively
driven mechanical systems, and in the study of certain animal populations where successive generations
do not overlap, i.e., where a generation lives only for one season.
3. As simple examples of chaos. Maps are interesting to study in their own right, as mathematical
laboratories for chaos. Indeed, maps are capable of much wilder behaviour than differential equations
because iterates xn hop along their orbits or trajectories rather than flow continuously. Continuity
is very much a restriction on possible dynamics, cf. Lecture 6: The Poincaré-Bendixson theorem.

3 The logistic map

In a fascinating and influential review article (linked below), Robert May (1976–2020) emphasised that even
simple nonlinear maps could have very complicated dynamics. May illustrated his point with the logistic
map given by:
xn+1 = rxn (1 − xn ), (18)
a discrete-time analog of the logistic equation for population growth, where xn is the dimensionless mea-
sure of the population in the n-th generation and parameter r is the intrinsic growth rate. As shown in
Fig. 6, the graph of map (18) is a parabola with a maximum value of r/4 at x = 1/2. We restrict the control
parameter r to the range (−2 or) 0 ≤ r ≤ 4 so that Eq. (18) maps the interval 0 ≤ x ≤ 1 into itself.

D. Kartofelev 6/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Xh4 xhn xu

naif.gg xn
O 12
Figure 6: The logistic map shown with the red curve and the diagonal where xn+1 = xn .

Slide: 3

i
The logistic map

iii
1
The logistic map has the following form:

xn+1 = rxn (1 − xn ), x0 ∈ [0, 1], r ∈ [0, 4],

i

n ∈ Z+ , (1)

where r is the control parameter.

Read: Robert M. May, “Simple mathematical models with very

complicated dynamics,” Nature 261, pp. 459–467, 1976.
doi:10.1038/261459a0
i
1
See Mathematica .nb file (cobweb diagram and orbit diagram) uploaded to
the course webpage.
D. Kartofelev
INITIAL
I
YFX1560

l
no

TRANSIENT PERIOD
l l l l
4
l l
PERIOD
l
3 / 19

l l
in
41
N htt 42 h13 14
n
Reading suggestion
Link File name Citation
Paper#1 paper2.pdf Robert M. May, “Simple mathematical models with very complicated dynam-
ics,” Nature, 261, pp. 459–467, (1976).
doi:10.1038/261459a0

3.1 The Lyapunov exponent

A precise estimation or calculation of the Lyapunov exponents of differential equations is not a trivial task.
In the case of iterated maps it is much easier. We remind that the positive Lyapunov exponent λ is a sign of
chaos, see Lecture 9, Sec. 2.

Slides: 4–6
The Lyapunov exponent of the logistic map The Lyapunov exponent of the logistic map
Chaos is characterised by sensitive dependence on initial For small values of η0 , the quantity inside the absolute value signs is
conditions. If we take two close-by initial conditions, say x0 and just the derivative of f n with respect to x evaluated at x = x0 :
y0 = x0 + η with η  1, and iterate them under the map, then the 1 df n
difference between the two time series ηn = yn − xn should grow λ = ln . (5)
n dx x=x0
exponentially
|ηn | ∼ |η0 eλn |, (2) Since, f n (x) = f (f (f (. . . f (x))) . . .), by the chain rule:
where λ is the Lyapunov exponent. For maps, this definition leads to df n
= f 0 (f n−1 (x0 )) · f 0 (f n−2 (x0 )) · . . . · f 0 (x0 )
a very simple way of measuring the Lyapunov exponents. Solving (2) dx x=x0
for λ yields: n−1
Y (6)
1 ηn = |f 0 (xn−1 ) · f 0 (xn−2 ) · . . . · f 0 (x0 )| = f 0 (xi ) .
λ = ln . (3)
n η0 i=0
n n
By definition ηn = f (x0 + η0 ) − f (x0 ). Thus Our expression for the Lyapunov exponent takes the form:
n−1
Y n−1
1 f n (x0 + η0 ) − f n (x0 ) 1 1X
λ= ln . (4) λ= ln f 0 (xi ) = ln |f 0 (xi )|. (7)
n η0 n i=0
n i=0

D. Kartofelev YFX1560 4 / 19 D. Kartofelev YFX1560 5 / 19

D. Kartofelev 7/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

The Lyapunov exponent of the logistic map

n−1
Y n−1
1 1X
λ= ln f 0 (xi ) = ln |f 0 (xi )|.
n i=0
n i=0
The Lyapunov exponent is the large iterate n limit of this expression,
and so we have:
n−1
1X
λ = lim ln |f 0 (xi )|. (8)
n→∞ n
i=0

This formula can be used to study the Lyapunov exponent2 as a

function of control parameter r:
n−1
1X
λ(r) = lim ln |f 0 (xi , r)|. (9)
n→∞ n
i=0

2
See Mathematica .nb file uploaded to the course webpage.
D. Kartofelev YFX1560 6 / 19

The obtained algorithm is used in the following numerical file to calculate the Lyapunov exponent of the
logistic map as a function of the growth rate r.

Numerics: nb#1
Cobweb diagram and iterates of the (generalised) Lorenz map. Lyapunov exponent λ(r) of the logistic
map.

-1
λ

-2

-3

-2 -1 0 1 2 3 4
r

Figure 7: The Lyapunov exponent of the logistic map as a function of the growth rate r. Calculation
uses 5000 iterations for every r value plotted.

3.2 Bifurcation analysis and period doubling bifurcation

Next, let’s consider only the stable fixed point and stable period-p points as we incrementally increase the
value of control parameter r ≥ 0. We do that in order to simplify our analysis and to save some lecture
time. The stable fixed point (a period-1 point) of the logistic map given by (18) and satisfying condition
(3) is the following:
f (x∗ ) = x∗ ⇒ rx∗ (1 − x∗ ) = x∗ | ÷ x∗ , (19)
r(1 − x∗ ) = 1, (20)
r − rx∗ = 1 | ÷ r, (21)
1
1 − x∗ = , (22)
r
1
x∗ = 1 − . (23)
r
Additionally, there are the trivial solutions x∗ = 0 and x∗ = 1 (for initial condition x0 = x∗ = 1, and for
n ≥ 1, xn → x∗ = 0). Fixed point (23) is stable for |f 0 (x∗ )| < 1. Using map definition (18) we write

|f 0 (x∗ )| = [rx(1 − x)]0 x=x∗

= |r − 2rx∗ | < 1. (24)

D. Kartofelev 8/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Using condition (24) with the trivial fixed point x∗ = 0 gives us

|r − 0| < 1, (25)
|r| < 1. (26)
The other trivial fixed point x∗ = 1 yields the same result
|r − 2r| < 1, (27)
| − r| < 1. (28)
Thus, the fixed points x∗ = 0 and x∗ = 1 are stable for |r| < 1. For the non-trivial fixed point x∗ = 1 − 1/r
and for condition (24) we find  
Xh4 xhn r xu
− 2r 1 −
1
< 1, (29)
r

naif.gg
|r − 2r + 2| < 1, (30)
|2 − r| < 1, (31)
−xn
1 < 2 − r < 1 | − 2, (32)
O 12 − 3 < −r < −1 | · (−1), (33)
1 < r < 3. (34)
The fixed points x∗ = 1 − 1/r exist and is stable for 1 < r < 3.

iii i
i
Figure 8: (Left) Map slope at the fixed points x∗ = 1 − 1/r (same as the trivial cases x∗ = 0 = 1) for r = 1.
(Right) Map slope at the fixed points x∗ = 1 − 1/r for r = 3.

It seems that the found intervals (26), (28) and (34) excluded r = 1 (obviously it does not satisfy |f 0 (x∗ )| <
1). Let’s find the value of the map slope |f 0 (x∗ )| for r = 1 using (24) in the case of the trivial solutions
x∗ = 0 i

in
|f 0 (x∗ )| = |1 − 0| = 1, (35)
x∗ = 1 no
INITIAL |f 0 (x∗ )| = |1 − 2| = 1, (36)
TRANSIENT PERIOD 4 l PERIOD l
41
l l
and in the non-trivial case forI x∗l= 1l − l1/rl = 1l − l1 =n 0. Which should obviously generate the same result
13 14 
N htt 42 h
1
|f 0 (x∗ )| = 1 − 2 1 − = |1 − 0| = 1. (37)
1
Below, it will also be beneficial to know what happens for r = 3, the r value just after the interval (34).
We consider the non-trivial fixed point x∗ = 1 − 1/r and find the slope
 
0 ∗ 1
|f (x )| = 3 − 2 · 3 1 − = |3 − 4| = | − 1| = 1. (38)
3
Usually, slope |f 0 (x∗ )| = 1 corresponds to the period doubling or flip bifurcation point (dynamics
explained below). Values r = 1 and r = 3 are the bifurcation points. Figure 8 shows the map and map
slopes |f 0 (x∗ )| evaluated at the non-trivial fixed point x∗ = 1 − 1/r for r = 1 and r = 3. The following
interactive numerical file show the dynamics of the logistic map for 0 ≤ r < 1 and 1 < r < 3.

D. Kartofelev 9/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Numerics: nb#2
Cobweb diagram of the logistic map. Orbit diagram of the logistic map. Period doubling bifurcation.

1.0 1.0

0.8 0.8

0.6 0.6
xn+1

xn
0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20
xn n
1.0 1.0

0.8 0.8

0.6 0.6
xn+1

xn
0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20
xn n
Figure 9: (Top-left) Cobweb diagram of the logistic map shown for r = 0.98 < 1. The fixed point
x∗ = 0. (Top-right) Map iterates corresponding to the cobweb diagram shown on the left. (Bottom-
left) Cobweb diagram of the logistic map shown for interval 1 < r < 3 with r = 2.82. The fixed point
x∗ = 1 − 1/r. (Bottom-right) Map iterates corresponding to the cobweb diagram shown on the left.

What happens for r ≥ 3? What happens after the period doubling or flip bifurcation at r = 3? The
name “flip” refers to the fact that the map trajectories start to flip between two values—the period-2 points
in a period-2 orbit. Let’s see this dynamics play out using a computer.

Numerics: nb#2
Cobweb diagram of the logistic map. Orbit diagram of the logistic map. Period doubling bifurcation.
After the initial transient behaviour has decayed the dynamics of the √ map settles to a stable
period-2 orbit. It can be showed that period-2 orbits exist for 3 ≤ r < 1 + 6.

1.0 1.0

0.8 0.8

0.6 0.6
xn+1

xn

0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20 25 30
xn n
Figure 10: (Left) Cobweb diagram of the logistic map shown for r = 3.19 ≥ 3 featuring the stable
period-2 point. (Right) Map iterates corresponding to the cobweb diagram shown on the left.

Let’s try to think about and unpack this outcome some more...

D. Kartofelev 10/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Slides: 7–9
The logistic map, period-2 window The logistic map, period-2 window
√
Period-2 window for 3 ≤ r < 1 + 6. √
Period-2 window for 3 ≤ r < 1 + 6.
(
f (p) = rp(1 − p) = q,
(10)
f (q) = rq(1 − q) = p,
here period-2 point values p and q are the f.p.s of f (x) map.

On the other hand it also holds

(
f (p) = f (f (q)) ≡ f 2 (q) = r[rq(1 − q)][1 − (rq(1 − q))] = q,
⇒
f (q) = f (f (p)) ≡ f 2 (p) = r[rp(1 − p)][1 − (rp(1 − p))] = p,
Figure: The logistic map where r = 3.18 and x0 = 0.35. Fixed points (11)
(f.p.s) in the case where r < 3 are shown with the grey bullets. Period-2 ⇒ f 2 (x) = r[rx(1 − x)][1 − (rx(1 − x))] = x, (12)
points of f (x) map for r ≥ 3 are shown with the blue bullets. The fixed
points of f 2 (x) map for r ≥ 3 are shown with the red bullets. where period-2 point values p and q are the f.p.s of f 2 (x) map.

D. Kartofelev YFX1560 7 / 19 D. Kartofelev YFX1560 8 / 19

If you want to study the dynamics of the attractors of second iterate map analytically consider one f2
of the equations in (11, slide numbering). The fourth order polynomial defined by Eq. (11)

r[rx(1 − x)][1 − (rx(1 − x))] = x, (39)

− r3 x4 + 2r3 x3 − r2 (1 + r)x2 + r2 x = x. (40)

Intersections between the graph of second iterate map f 2 and the diagonal correspond to the solutions
of f 2 (x) = x (40).

Stability of f.p.s of f 2 map in period-2 orbit

We need to know the slopes of period-2 points
(
f (p) = rp(1 − p) = q,
f (q) = rq(1 − q) = p.

According to the chain rule it holds that

(f 2 (x))0 ≡ (f (f (x))0 = f 0 (f (x)) · f 0 (x). (13)

In our case:
)
(f 2 (p))0 = f 0 (f (p)) · f 0 (p) = f 0 (q) · f 0 (p)
⇒ (f 2 (p))0 = (f 2 (q))0 .
(f 2 (q))0 = f 0 (f (q)) · f 0 (q) = f 0 (p) · f 0 (q)
(14)
Above follows from the commutative property of multiplication.
D. Kartofelev YFX1560 9 / 19

The slopes of map f 2 at its fixed points p and q are equal and they are the products of map f slopes
at its respective period-2 points.
√
What happens for r ≥ 1 + 6? Once again we use a computer...

Numerics: nb#2
Cobweb diagram of the logistic map. Orbit diagram of the logistic map. Period doubling bifurcation.
√
A period-4 orbit has emerged. It can be shown that period-4 orbits exist for 1 + 6 ≤ r < 3.54409...

D. Kartofelev 11/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560
Xh4 xhn xu
1.0

0.8
naif.gg xn
1.0

0.8

0.6 O 12 0.6
xn+1

xn
0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20 25 30

iii i
xn n
√
Figure 11: (Left) Cobweb diagram of the logistici map shown for r = 3.535 ≥ 1 + 6 featuring the stable
period-4 point. The blue graph shows the fourth iterate map f 4 . (Right) Map iterates corresponding
to the cobweb diagram of map f shown on the left.

Figure 12 counts the period-4 points or iterate values in the period-4 orbit of the logistic map. Compare
Fig. 12 to Figs 5 and 11.

I
INITIAL
TRANSIENT PERIOD
l l l l l
no

4
l l
PERIOD
l l l
in
41
N htt 42 h13 14
n

Figure 12: Period-4 orbit with its repeated iterates as they should appear in the logistic map.

What have we seen so far? As we have incrementally increased the value of control parameter r the
periods of the fixed points have increased as well—(from period-1) to period-2 and finally to period-4. The
period is clearly doubling starting from period-2. This type of bifurcation is called the period doubling
bifurcation. Now the name “period doubling” introduced above alongside “flip” bifurcation should make
more sense. Further period-doublings to orbits of period-8, -16, -32, -64, etc., also occur as r is increased.

Slide: 10
Specifically, let rn denote the value of r where a stable 2n -orbit first appears. Then computer experi-
ments reveal that:

The logistic map, period doubling

Even number periods.
rn – bifurcation point, onset of a stable period-2n orbit.

r1 = 3.0 √ period-2
r2 = 1 + 6 ≈ 3.44949 period-4
r3 ≈ 3.54409 period-8
r4 ≈ 3.56441 period-16
r5 ≈ 3.56875 period-32
r6 ≈ 3.56969 period-64
.. ..
. .
r∞ ≈ 3.569946 period-2∞
r∞ – onset of chaos (the accumulation point).
rn−1 − rn−2
δ = lim ≈ 4.669201609... (15)
n→∞ rn − rn−1

D. Kartofelev YFX1560 10 / 19

D. Kartofelev 12/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Note that the successive bifurcations come faster and faster. Ultimately, the rn converge to
a limiting value r∞ . The convergence is (essentially) geometric (it’s geometric near r∞ ): in the limit
of large n, the distance between successive transitions shrinks by a constant factor of δ. This ratio is
called the Feigenbaum constant. After the accumulation point r > r∞ the dynamics becomes
chaotic as we’ll see below.

3.3 Orbit diagram

The orbit diagram is also called the fig tree diagram (Feigenbaum in German means “fig tree”) or incor-
rectly the Feigenbaum diagram. You might guess that the system would become more and more chaotic
for r > r∞ as r increases, but in fact the dynamics are more subtle than that. To see the long-term be-
haviour for all values of r at once, we plot the orbit diagram a special kind of bifurcation diagram.
Orbit diagram plots the system’s attractor (stable fixed points and stable period-p points) as a function of
control parameter r.
To generate the orbit diagram for yourself, you’ll need to write a computer program with two loops.
First, choose a value of r. Then generate an orbit starting from some random initial condition x0 Iterate for
300 cycles or so, to allow the system to settle down to its eventual stable behaviour. Once the transients
have decayed, plot many points, say x301 , . . . , x900 above that r. Then move to an adjacent value of r and
repeat, eventually sweeping across the whole diagram. The following interactive numerical file shows the
orbit diagram for the logistic map.

Numerics: nb#2
Cobweb diagram of the logistic map. Orbit diagram of the logistic map. Period doubling bifurcation.

The connection between the cobweb diagram, orbit diagram and the Lyapunov exponent is shown in the
following interactive numerical file.

Numerics: nb#3
The logistic map: cobweb diagram, orbit diagram and map iterates, the Lyapunov exponent.

Slides: 11, 12
Orbit diagram and period doubling Zooming into the logistic map, self-similarity

No embedded video files in this pdf

D. Kartofelev YFX1560 11 / 19 D. Kartofelev YFX1560 12 / 19

The period doubling is driven by the subsequent flip bifurcations or supercritical pitchfork
bifurcations (if we use the nomenclature introduced in Lecture 2). As mentioned above, unstable
fixed point and period-p points are omitted from the orbit diagram. The Feigenbaum diagram
shows both stable and unstable fixed points and period-p points or simply diagram branches (not shown
here).
The orbit diagram of the logistic map features self-similarity. In mathematics, a self-similar object
is (exactly or) approximately similar to a part of itself, i.e., the whole has the same shape or quality
as one or more of its sub-parts.

D. Kartofelev 13/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

3.4 Tangent bifurcation and odd number period-p points

Can an odd number period appear in the logistic map—e.g., a period-3 point? One of the most intriguing
features of the orbit diagram is the occurrence of periodic windows for r > r∞ .

Slide: 13
Orbit diagram, period-3 window

D. Kartofelev YFX1560 13 / 19
√
The period-3 window that occurs for 1 + 8 ≤ r ≤ 3.8415... is the most conspicuous. On the above
slide it is shown within the red rectangle. Suddenly, against a backdrop of chaos, a stable period-3
orbit appears out of the blue.

Let’s see if we can find the period-3 orbit and window of the logistic map using a computer.

Numerics: nb#2
Cobweb diagram of the logistic map. Orbit diagram of the logistic map. Period doubling bifurcation.
√
The tangent bifurcation is a new type of bifurcation for us, that occurs at r = 1 + 8.

1.0 1.0

0.8 0.8

0.6 0.6
xn+1

xn

0.4 0.4

0.2 0.2

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20 25 30
xn n
Figure 13: (Left) Period-3 orbit of the logistic map can be found in the period-3 window of the orbit
diagram, see Slide 11. Here, showing the trajectory for r = 3.83 and x0 & 0. (Right) Map iterates xn
corresponding to the cobweb diagram of the logistic map shown on the left.

The mechanism responsible for the occurrence of the odd number periods-p points is shown in Fig. 14.
The intersections between the graph of f 3, shown √ in red, and the diagonal correspond to solutions of
3
f (x) = x. There are eight solutions for r ≥ 1 + 8, six of interest to us are marked with the smaller blue
dots, and two imposters that are not genuine period-3; they are actually fixed points, or period-1 points for
which f (x∗ ) = x∗. The blue filled dots in Fig. 14 correspond to a stable period-3 cycle; note that the slope
of f 3 (x) is shallow at these points, consistent with the stability of the cycle. In contrast, the slope exceeds
one at the cycle marked by the empty blue dots; this period-3 orbit is therefore unstable.

D. Kartofelev 14/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

h
xnn
yf.iyjo.IT
ipl xwo
king

Hagi

Xu
Figure 14: A period-3 orbit can be found in the period-3 window of the orbit diagram. The hollow bullets
correspond to the unstable period-p points, the filled bullets to the stable ones. The dashed and continuous
curves shown within the grey rectangular outline is a smaller distorted copy of map f in its entirety.

√
Now suppose we decrease r toward the chaotic regime r < 1 + 8. Then the red dashed graph in Fig. 14
changes shape—the hill moves down and the √ valleys rise up. The curve therefore moves towards the
diagonal. Figure 14 shows that when r = 1 + 8, the six blue intersections have merged to form three black
filled period-3 points by becoming tangent to the diagonal. At this critical value of r, the stable and
unstable period-3 cycles coalesce and annihilate in the tangent bifurcation. This transition defines the
beginning of the period-3 window discussed above.
All odd number periods will also undergo the period doubling. This means that all number periods [1, ∞)
are eventually represented in the orbit diagram.
Figure 14 also explains the self-similarity present in the orbit diagram as shown on Slide 12 for dif-
ferent scales of magnification. The peaks (shown within the grey rectangle) and valleys of map f 3 are
smaller distorted copies of the original map f . Thus, the local dynamics of the cobweb trajectories for map
f 3 must be similar to the dynamics of the original map f .

Slide: 14
3
Intermittency and period-3 window
Transient chaos and intermittency
√ in dynamical systems. Tangent
bifurcation occurs at r = 1 + 8 ≈ 3.8284 (onset of period-3 orbit).
xn
1.0                  
     
  
0.8   

     
0.6   
                 

 
0.4


0.2                                              

n
50 100 150

Iterates of the logistic map shown for r = 3.8282 and x0 = 0.15.

3
See Mathematica .nb file uploaded to the course webpage.
D. Kartofelev YFX1560 14 / 19

The dynamics closely related to the tangent bifurcation is intermittency. The logistic map exhibits
intermittent chaos for r values just before the period-3 window, i.e., just before the onset of the
period-3 orbit.

The following numerical file was used to calculate and create the graph shown on the above slide.

Numerics: nb#4
Period-3 window and intermittency in the logistic map.

D. Kartofelev 15/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

4 Sine map and universality of period doubling

It can be shown that in all unimodal maps same dynamics of period doubling occurs. For example we
consider the sine map:

Slide: 15
Universality of period doubling in unimodal maps

1-D sine map4 . The sine map has the form:

xn+1 = r sin(πxn ), x0 ∈ [0, 1], r ∈ [0, 1], n ∈ Z+ , (16)

where r is the control parameter.

Read: Mitchell J. Feigenbaum, “Quantitative universality for a class

of nonlinear transformations,” Journal of Statistical Physics 19(1),
pp. 25–52, 1978, doi:10.1007/BF01020332
Read: Mitchell J. Feigenbaum, “Universal behavior in nonlinear
systems,” Physica D: Nonlinear Phenomena 7(1–3), pp. 16–39,
1983, doi:10.1016/0167-2789(83)90112-4
4
See Mathematica .nb file (cobweb diagram and orbit diagram) uploaded to
the course webpage.
D. Kartofelev YFX1560 15 / 19

The graph of the sine map has the same basic shape as the graph of the logistic map. Both curves
are smooth, concave down, and have a single maximum. Such maps are called unimodal.
Sine is also a transcendental function opposed to an algebraic function as is the polynomial that
defines the logistic map. A transcendental function is an analytic function that does not satisfy a
polynomial equation. In other words, a transcendental function “transcends” algebra in that it cannot
be expressed in terms of a finite sequence of the algebraic operations of addition, multiplication, and
root extraction.
The transcendental nature of the sine map must make the underlying higher-dimensional physics
represented by the sine map fundamentally different from, e.g., the three-dimensional Lorenz flow
sampled by the Lorenz map.

Let’s study the dynamics of the sine map using a computer:

Numerics: nb#5
The sine map iterates, cobweb and orbit diagrams where xn ∈ [0, 1].

In the following numerical file the basin of attraction is widened to span [−1, 1]. In effect, here two
unimodal maps are placed side-by-side. Sine map takes the following form:

xn+1 = r sin(πxn ), x0 ∈ [−1, 1], r ∈ [0, 1], n ∈ Z+ , (41)

where r is the control parameter.

Numerics: nb#6
The sine map iterates, cobweb and orbit diagrams where xn ∈ [−1, 1].

The dynamics of sine map is surprisingly similar to the dynamics of the logistic map. Mitchell
J. Feigenbaum was one of the first researchers1 to discover the quantitative laws that are independent
of unimodal map functions f . By that we mean that the algebraic form of f (x) is irrelevant, only its
overall shape matters, i.e., unimodality.
The following slide shows the scaling constants that are related to these universal laws that are present
in all the unimodal maps:
1
The first published works on period doubling and related phenomena were authored by a Finnish mathematician Pekka
Juhana Myrberg (1892–1976).

D. Kartofelev 16/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Slide: 16

1-D unimodal maps and the Feigenbaum constants

∆n−1 rn−1 − rn−2

δ = lim = lim ≈ 4.669201609... (17)
n→∞ ∆n n→∞ rn − rn−1

dn−1
α = lim ≈ −2.502907875... (18)
n→∞ dn
D. Kartofelev YFX1560 16 / 19

Here xm = max f (x) is the maximum of the map graph. The Feigenbaum constants are valid up to
the onset of chaos at accumulation point r∞ and inside each periodic window for r > r∞ .
In addition to scaling law in control parameter r direction, shown earlier, Feigenbaum also found
a scaling law for the vertical x-direction of the orbit diagram. The Feigenbaum constants are
universal the same convergence rate appears no matter what unimodal map is iterated! They are
mathematical constants, as basic to period doubling as π is to circles.

Reading suggestion
Link File name Citation
Paper#2 paper3.pdf Mitchell J. Feigenbaum, “Quantitative universality for a class of nonlinear trans-
formations,” Journal of Statistical Physics, 19(1), pp. 25–52, (1978).
doi:10.1007/BF01020332
Paper#3 paper4.pdf Mitchell J. Feigenbaum, “Universal behavior in nonlinear systems,” Physica D:
Nonlinear Phenomena, 7(1–3), pp. 16–39, (1983).
doi:10.1016/0167-2789(83)90112-4

5 Universal route to chaos

We showed that the qualitative dynamics of the logistic and sine maps are identical. They both undergo
period doubling routes to chaos, followed by periodic windows interwoven with chaotic bands.
Even more remarkably, the periodic windows occur in the same order, and with the same relative sizes. For
instance, the period-3 window is the largest in both cases, and the next largest windows preceding it are
period-5 and period-6. But there are quantitative differences. For instance, the period doubling bifurcations
occur later (for greater parameter r value) in the logistic map, and the periodic windows are thinner.
Turns out that the onset of chaos via period doubling is predominant in nature and in artificial chaotic
systems. The Feigenbaum constants have real predictive power in various scientific applications. The
period doubling bifurcation is the “route” taken by nonlinear systems to reach chaotic solution.
In term of the bifurcations introduced in Lecture 2, the period doubling bifurcation can also be seen as a
series of subsequent or succeeding supercritical pitchfork bifurcations, see Slide 11.

D. Kartofelev 17/18 K As of November 8, 2024

Lecture notes #10 Nonlinear Dynamics YFX1560

Revision questions
1. What is cobweb diagram?
2. What is recurrence map or recurrence relation?
3. What is 1-D map?
4. How to find fixed points of 1-D maps?
5. What is the Lorenz map?
6. What is the logistic map?
7. What is sine map?
8. What is period doubling?
9. What is period doubling bifurcation?
10. What is tangent bifurcation?
11. Do odd number periods (period-p orbits) exist in chaotic systems?
12. Do even number periods (period-p orbits) exist in chaotic systems?
13. Can maps produce transient chaos?
14. Can maps produce intermittency?
15. Can maps produce intermittent chaos?
16. What is orbit diagram (or the Feigenbaum diagram)?
17. What are the Feigenbaum constants?

D. Kartofelev 18/18 K As of November 8, 2024