Mapas Acoplados

Studiengang: Informatik
Prüfer: Prof. Dr. rer. nat. habil. P. Levi
Betreuer: Dr. rer. nat. Michael Schanz

Dr. rer. nat. Viktor Avrutin
begonnen am: 1. April 2004
beendet am: 30. September 2004
CR-Klassifikation: G.1.10, J.2
Studienarbeit Nr. 1943
Spatiotemporal Dynamics of
Coupled Map Lattices
Heiko Hamann
Institut für Parallele und

Verteilte Systeme (IPVS)
Universität Stuttgart
Universitätsstraße 38
D–70569 Stuttgart
Contents
1 Coupled Map Lattices - an Introduction 2
2 Applications of Coupled Map Lattices 4
3 Properties of Coupled Map Lattices 6

3.1 Couplings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 Border Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.4 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Phenomenology of Coupled Map Lattices 8

4.1 Frozen Random Pattern . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Pattern Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Zigzag Pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.4 Spatiotemporal Intermittency Type I . . . . . . . . . . . . . . . . 20
4.5 Spatiotemporal Intermittency Type II . . . . . . . . . . . . . . . . 22
4.6 Traveling Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.7 Fully Developed Chaos . . . . . . . . . . . . . . . . . . . . . . . . 26
5 Long Transients 28
6 A Symmetric System and its Coupled Map Lattice Correspon-

dence 32
6.1 The Original System . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2 A First Coupled Map System Version . . . . . . . . . . . . . . . . 32
6.3 A Second Coupled Map System Version . . . . . . . . . . . . . . . 33
7 Spatial Bifurcations 44
8 Using the AnT 4.669 Software to simulate Coupled Map Lattices 56
9 Conclusion 60
10 Acknowledgments 61
1
Chapter 1
Coupled Map Lattices - an

Introduction
When I am judging a theory,

I ask myself whether,
if I were God,
I would have arranged the world
in such a way.
ALBERT EINSTEIN
There is nothing more difficult to take in hand,

more perilous to conduct or more uncertain in its success
than to take the lead in the introduction of a new order of things.
NICCOLO MACHIAVELLI in ”The Prince”
The coupled map lattices were introduced by Kunihiko Kaneko as a simple

model of spatiotemporal chaos in the year 1983. Attracted by the fascinating
idea of Landau to regard turbulence as a product of elementary low-dimensional
systems he asked the question: What happens if we choose a low-dimensional
chaos as an elementary system? [51, 6] This led him to the construction of the
coupled map lattice model.
A coupled map lattice is a dynamic system with discrete time, discrete space and
continuous states. It consists of dynamic elements on a lattice that interact with
some other certain elements. Each element is represented by a non-linear map,
e.g. the logistic map. The interaction is caused by a coupling of the elements
normally concerning the next neighbors. That leads us to a general formula for
coupled map lattices
r
γ X i−k
xin+1 = (1 − γ)f(xin ) + f(xn ) + f(xi+k
n ), (1.1)
2r k=1
2
1
0.5
f(x)
0
-0.5
-1
-1 -0.5 0 0.5 1
Figure 1.1: Quadratic map for α = 2
where f(x) is a non-linear map, γ the coupling constant and r the coupling range.
The lower index of x (n) denotes discrete time and the upper index (i) the spa-
tial index of the site. Although two- and more-dimensional lattices could be
represented by one-dimensional lattices at least the two-dimensional lattices are
considered to be a model on their own (see section 3.2). A lattice consists of
N sites and the standard boundary condition is periodic. For couplings, two-
dimensional lattices and border conditions see chapter 3.
Throughout this work only the nearest neighbor coupling (r = 1) is used. There-
fore a less general formula can be used:
γ
xin+1 = (1 − γ)f(xin ) + (f(xi−1 i+1
n ) + f(xn )); (1.2)
2
and f is chosen to be the quadratic map (a variant of the logistic map f(x) =
αx(1 − x), see the appendix for the association between these two maps ), i.e.
f(x) = 1 − αx2 with α ∈ [0, 2] and x ∈ [−1, 1] in the first part of this text
(see figure 1.1). According to Kaneko this system is called “coupled logistic
lattice” [6]. Another system will be introduced in chapter 6.
3
Chapter 2
Applications of Coupled Map

Lattices
The trouble with facts is

that there are so many of them.
SAMUEL MCCHORD CROTHERS
Physics is like sex.

Sure, it may give some practical results,
but that’s not why we do it.
RICHARD FEYNMAN
There is a vast variety of applications for coupled map lattices. Once the his-
tory of coupled map lattices begun by serving as a model for fluid turbulence but
coupled map lattices have been used in many other fields until now. According
to Kaneko [51] some of these fields will be named here.
Since in the real neuron states are not discrete (as in discrete-state models, e.g.
McCullough-Pitts neuron) it might be a good way to use globally coupled maps
as a model in neural dynamics. However, that is an oversimplification in the
coupling instead of the state. In such models clustering with synchronized oscil-
lation are observed [11, 14, 26, 32, 33, 41, 42, 49]. A connected field are excitable
media which have important applications in biology (heart rhythm and fibrilla-
tion). Corresponding to a simple 1-dimensional map of Nagumo and Sato [2]
that is based on Harmon’s experiment on an artificial neuron [1] a CML can be
constructed which shows soliton turbulence [11, 40, 46].
There are even some very interesting applications to computer science. CMLs
can be used as memory storage when there are many attractors with different
domain structures. In the frozen-random-pattern-phase the attractors are char-
acterized by their domains that do not change with time. Thus these could be
used as memory units. Chaos can be used in optimization problems [5, 9, 31].
There is research on using a CML to search for solutions of traveling salesman
4
problems [35]. An obvious application is image data processing since both the
preservation of relevant data and smoothing can be attained by a local map-
ping and the diffusion of the coupling. The most important question concerning
CMLs that computer scientists should ask is: What kind of computations can be
performed by a CML that are not possible on conventional, digital computers?
CMLs are analogous parallel processors [39] that must have higher computational
ability than cellular automata since CMLs can create information [3].
Some other fields are: convection parameterization in ocean models [57], coupled
maps with growth and death as a model for cell differentiation [54], population
dynamics [21, 37, 44, 45, 47, 53], dripping handrail and boiling chaos [12, 19, 50],
fluid dynamics, chemical waves, and pattern information.
5
Chapter 3
Properties of Coupled Map

Lattices
I have deep faith

that the principle of the universe
will be beautiful and simple.
ALBERT EINSTEIN
If you are out to describe the truth,

leave elegance to the tailor.
ALBERT EINSTEIN
3.1 Couplings
Naturally a very influential property of coupled map lattices is the coupling of
the sites. The standard coupling is the nearest neighbor coupling (coupling range
r = 1). The other extreme is the global coupling (r = N−1 2
, for N uneven). Not
only the coupling constant γ and the range r can be varied but also the symmetry
of the coupling can be broken. That leads us away from the models of symmetric
diffusion to the asymmetric models of uni-directional coupling and open flow.
Another variation is expressed in the following equation:
r
γ X
xin+1 = (1 − γ)f(xin ) + g(xi−k i+k
n ) + g(xn ), (3.1)
2r k=1
where g(x), the so called coupling dynamics, can be chosen to be g(x) = x, the
so called ”linear-coupling”, or g(x) = f(x), the so called ”future-coupling”, which
will be used throughout this paper [12]. The future-coupling is more useful be-
cause each site value remains in the same range. Linear-coupling causes different
behavior for example in the special case of a spatially homogeneous lattice one
cannot observe the standard periods of the local quadratic map.
6
3.2 Lattices
3.2 Lattices
As already said above the standard lattice is one-dimensional. But two-dimensional
lattices are used, too. They are important for some applications, e.g. chemical
waves. A different model indeed is the concept of inhomogeneous lattices where
the map parameter has different values at adjacent sites [55].
Naturally lattices have another property: the size. Some attractors depend
strongly on the lattice size, i.e. the number of cells (see for example section 4.6
about traveling waves).
3.3 Border Conditions

There are some alternatives one can choose from: free (no special treatment
at the boundaries at all), periodic (the lattice is constructed as a ring), fixed
(bounds are set to a certain value), noise (the borders are set to random val-
ues), periodically driven (like fixed but with periodically changing values) and so
on. Throughout this paper the periodic boundary is used only. The dynamics
should be qualitatively independent of the boundary conditions given a lattice of
sufficient size.
3.4 Initial Conditions

Each site xi0 must have an initial value to start with. This can be done randomly,
for example uniformly distributed over the whole interval of allowed values (that
is the standard initialization in this work). But other distributions, e.g. Gauss,
could be used, too. Trigonometric functions like sine or cosine are sometimes very
useful. An extreme example is the spatial homogeneous initialization because
the coupling has then no effect anymore. Since we have xi−1 n = xin = xi+1
n for a
homogeneous lattice we get for the nearest neighbor coupling:
γ
xn+1 = (1 − γ)f(xn ) + (2f(xn )) = f(xn ). (3.2)
2
The observed behavior would be that of the standard quadratic map since the
system will always stay spatially homogeneous.
7
Chapter 4
Phenomenology of Coupled Map

Lattices
I want to know how God created this world.

I am not interested in this or that phenomenon,
in the spectrum of this or that element.
I want to know His thoughts; the rest are details.
ALBERT EINSTEIN
I have no special talents.

I am only passionately curious.
ALBERT EINSTEIN
In a simulation with a CML there are many buttons to push and many
switches to turn. Therefore we need some instruments to order the many pa-
rameters like the system parameter, coupling constant, lattice size and number
of iterations according to their influences to the system dynamics. These instru-
ments will be provided by the universality classes in pattern dynamics in CMLs
introduced by Kaneko. Qualitative dynamical behaviors that occur in many CML
systems are described by universality classes. They form the pattern sequence of
CMLs that basically can be observed by starting at a low value for the system pa-
rameter α and steadily increasing it to its maximum. However, depending on the
coupling constant and even the lattice size some patterns may be observed only
fuzzy or even not. In the same system one could observe the pattern sequence:
frozen random, traveling wave, spatiotemporal intermittency (STI) type-II, fully
developed chaos (FDC); or with different parameters: frozen random, pattern
selection, zigzag and fully developed chaos.
See figure 4.1 for an overview of the different regions of the patterns. Note
that the borders between the regions cannot be determined accurately since the
behavior of the system depends on other parameters, too (e.g. lattice size, ini-
tialization, long transient times). Especially the wavy line at 1.9 between STI
8
Figure 4.1: Schematic overview of the different patterns dependent on the system
parameter α and the coupling γ for the coupled logistic lattice
type II and FDC depends strongly on the number of iterations. The problem is
that transient times of several millions even for small lattice sizes are common.
Thus the area of occurrence of a certain pattern can not be described by the
system parameter only but by an at least two-dimensional, better three- or even
four-dimensional parameter space. The needed parameter settings and the spe-
cial phenomenon of each universality class will be described in the next sections.
Various types of diagrams will be used throughout this paper and especially
in this chapter. The orbit of a certain system run can be presented as a two-
dimensional space-time plot using grey scale to indicate the site values or as a
three-dimensional space-time plot using a third axes and lines. This type of di-
agram can be projected to a plane - the amplitude plot. To show the degree of
chaotic motion Lyapunov spectra, that consist of all Lyapunov exponents (the
amount depends on the lattice size), are presented. The results of a so called “pa-
rameter scan” can be plotted as a diagram with the system parameter α against
the maximum Lyapunov exponent for example or against the sites’ values at their
asymptotic dynamics - the bifurcation diagram. Besides these classical diagrams
spatial return maps are used that are plots of each site xi for example against its
nearest right neighbor xi+1 . These diagrams show characteristic properties for
each pattern but show only little information concerning the system dynamics.
Before we start with the first pattern a very important term has to be introduced
and explained. To describe the spatiotemporal dynamics the term “domain” is
used. A domain is a part of the lattice whose sites are correlated in space and
time to some certain degree. The different domains in a lattice are separated
9
4.1 Frozen Random Pattern
by kinks that are caused by initial values near unstable fixed points of the local
dynamics. While in big domains the motion is chaotic it is (nearly) periodic in
smaller ones.

The frozen random pattern has domains with different sizes at the same time.
Some are smaller and nearly periodic others are bigger and exhibit chaotic behav-
ior. Small domains have period two, larger period four and much larger period
eight... With the increasing system parameter α the number of domains increases
and the motion gets more chaotic (see figure 4.4). Period doubling of kinks with
spatial bifurcation can be observed (see figure 4.3). It is possible that the orbits
of such domains are only nearly periodic, i.e. that there are small deviations.
While for period doubling more domains are essential an increase of domains
does not automatically effect a period doubling. For example in the first five
space-time plots in figure 4.4 the number of domains is increasing but all of them
have period two (see chapter 7, too).
The building of domains is very sensitive to the initial conditions. These can be
chosen so that the attractors have arbitrarily large domains. That is why it is
believed that the number of attractors increases exponentially with the system
size.
The spatial return map for a system with parameter settings for the frozen ran-
dom pattern can be seen in figure 4.5(c) and 4.5(d).
The Lyapunov spectrum (see figure 4.5(e)) in the frozen-random-pattern phase
has a stepwise structure which reflects the existence of the separated domains.
There is chaotic motion in larger domains only which effects the positive Lya-
punov exponents. The number of such exponents is proportional to the number
of large domains.
The approximate system parameter interval given in figure 4.5(f) is schematic
(that applies also to the next sections). For some initial conditions the system
is in space and time homogeneous for α < 1.0 while others already work for
α = 0.8. There is a dependence on the coupling constant at least for approxi-
mately α > 1.5: couplings with γ > 0.5 may lead to the traveling wave pattern.
A special property of some parameter settings in the frozen random regime is self
similarity in space. The best method to present this self similarity is provided by
the spatial return maps. Figure 4.6 shows an example and zooms into the upper
right pattern (figures b and c). The self similarity cannot be seen that good in
the amplitude plot since groups of neighboring sites seem nearly always to build
the basic pattern only in one dimension of magnitude and other dimensions are
seen in different groups. Therefore its difficult to find good examples that really
express self similarity. However, that is tried to be done in figure 4.2. The left
diagram shows the pattern over the full range and the pattern is one dimension
10
0.9 0.9
0.8
0.88
0.7
0.86
xi
xi
0.6
0.84
0.5
0.82
0.4
0.3 0.8
4695 4700 4705 4710 4715 4720 4725 4730 4735 4740 715 720 725 730 735
i i
(a) 50 sites, range is 0.3 to 0.9 (b) 20 sites, range is 0.8 to 0.9
Figure 4.2: Amplitude plots, α = 1.40116, γ = 0.3, N=30000
of magnitude smaller in the right diagram.

Since patterns for small values of α show no chaotic motion a more sophisti-
cated distinction can be done by the introduction of the universality class simply
called “patterns with kinks”. These patterns occur after the first bifurcation
(α > 0.75) and show at least two regions separated by kinks and oscillating in
different phases between the two values of the period two attractor (see chapter 7
which focuses on patterns with kinks).
For more information about the frozen random pattern see [10, 24, 25, 51].
11
1 1
0.9 0.8
0.8 0.6
0.7 0.4
xi
0.6 xi 0.2
0.5 0
0.4 -0.2
0.3 -0.4
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
i i
(a) corresponds to the 1st plot of figure 4.4 (b) corresponds to the 6th plot of fig-
(period 2) ure 4.4 (period 4)
1 1
0.8 0.8
0.6 0.6
0.4 0.4
xi
xi
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
i i
(c) corresponds to the 7th plot of fig- (d) corresponds to the 8th plot of fig-
ure 4.4 (nearly period 8) ure 4.4 (nearly period 16)
Figure 4.3: Amplitude plot, same parameter settings as in figure 4.4
12
n
xin
n
xin
n
xin
n
xin
n
xin
n
xin
n
xin
n
xin
Figure 4.4: Space-time plots indicating the increasing number of domains with
increasing system parameter α (from top to bottom: 0.8, 0.82, 0.9, 0.95, 1.0, 1.3,
1.38, 1.4); γ = 0.3, N = 200, starting with iteration 3000 every 32nd time step
is shown, uniform random initial condition
13
n
n
xin
xin
(a) Space-time plot with grey scale, N = (b) Space-time plot with lines, 128 sites of
500, starting at the top with iteration the run of figure 4.5(a) are shown
3000, every 32nd time step is shown
1 1
0.8 0.8
0.6 0.6
0.4 0.4
xi
xi
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
(c) Spatial return map at iteration 5000 (d) Spatial return map at iteration 5000
with N = 500 with N = 50000
0.2
0
0
-0.2
λmax
-0.5
λj
-0.4
-0.6
-1
-0.8
-1.5
-1
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
(e) Lyapunov spectrum, N=64 and tran- (f) The unmarked region is approximately
sient is 2000 the interval where the frozen random
pattern can be found (α ∈ [0.9, 1.6]),
depends also on the coupling parame-
ter γ
Figure 4.5: Characteristics of the frozen random pattern, α = 1.43, γ = 0.3
14
0.6
0.55
0.5
xi
0.45
0.4
0.35
0.3
0.3 0.35 0.4 0.45 0.5 0.55 0.6
xi+1
(a) Complete map
0.564
0.56
0.562
0.55
0.56
0.54
0.558
0.53
xi
0.556
xi
0.52
0.554
0.51
0.552
0.5
0.55
0.49
0.548
0.48
0.546
0.47
0.47 0.48 0.49 0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.546 0.548 0.55 0.552 0.554 0.556 0.558 0.56 0.562 0.564
xi+1 xi+1
(b) Zoomed into the upper right part (c) Deeper zoomed into the upper right
part
Figure 4.6: Spatial return maps, α = 1.40116, γ = 0.3, N=200000, uniform

random initialization
15
4.2 Pattern Selection
n
n
xin xin xin xin
(a) α = 1.570 (b) α = 1.575 (c) α = 1.578 (d) α = 1.579
Figure 4.7: Space-time plots, splitting of larger domains into smaller ones, γ = 0.4

If the system parameter α is further increased the larger domains get unstable
and split into smaller ones (see figure 4.7). This small domains are “selected”.
This results in a lattice with only few different and small domain sizes. The
initial conditions have no notable influence any more. The position of the do-
mains is fixed and does not move (in difference to the traveling wave pattern, see
section 4.6). Since there are only small domains the chaos is suppressed which
effects a small positive or even close to zero maximum Lyapunov exponent (see
figure 4.8(e)). In CMLs the diffusion of the coupling which tends to homogenize
the system fights always against the upcoming chaos that leads to an inhomo-
geneous state. The chaos in a certain large domain increases until chaos is so
strong that the domain splits into smaller ones to stabilize the domain structure.
Obviously a pattern with smaller Lyapunov exponent is selected. [51]
The transition from the pattern-selection-phase to the spatiotemporal-intermit-
tency-type-II-phase is fluid in both the classification and according to the param-
eter α (compare figure 4.8(b) to 4.13(b)). If the coupling constant γ is increased
that may lead to a traveling wave pattern. If it is decreased the domains are
much smaller and the transient process is shorter.
See section 4.7 and [52] for another pattern selection regime with high non-
linearity.
For more information about the pattern selection see [24, 25].
16
n
n
xin
xin
1 1
0.8 0.8
0.6
0.6
0.4
0.4
0.2
xi
xi
0.2
0
0
-0.2
-0.2
-0.4
-0.4 -0.6
-0.6 -0.8
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
(c) Spatial return map at iteration 5000, (d) Spatial return map at iteration 5000
N = 500 with N = 50000
0.2
-0.2 0
-0.4
λmax
-0.6
-0.5
λj
-0.8
-1
-1.2 -1
-1.4
-1.6
-1.5
-1.8
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
sient is 2000 the interval where pattern selection
can be found (α ∈ [1.6, 1.7]), depends
also on the coupling parameter γ
Figure 4.8: Characteristics of pattern selection, α = 1.7, γ = 0.4
17
4.3 Zigzag Pattern
n
xin
Figure 4.9: Space-time plot with grey scale, α = 1.79, γ = 0.1, N = 500, showing
every 32nd time step starting at the top with iteration 3000 to 4984
4.3 Zigzag Pattern

The zigzag pattern is the simplest example of pattern selection. The domain
size is 1. There can be two or more neighboring regions with a different phase
of oscillation. They are separated by defects whose dynamics is described by
Brownian motion. [51]
The spatial return map of the ideal zigzag pattern would show only two points
at the accumulation regions seen in figures 4.10(c) and 4.10(d). In the Lyapunov
spectra of zigzag patterns there can be positive exponents as described for pattern
selection above and as seen in figure 4.10(e). Other spectra may have a more
stepwise structure and no positive Lyapunov exponents.
For more information about the zigzag pattern see [51, 13, 20].
18
4.3 Zigzag Pattern
n
n
xin
xin
500, starting at the top with iteration the run of figure 4.10(b) are shown
1 1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
xi
xi
0
0.2
-0.2
0
-0.4
-0.2
-0.6
-0.4 -0.8
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
N = 500 with N = 50000
0.3
0.2
0.1 0
0
λmax
-0.1
-0.5
λj
-0.2
-0.3
-0.4 -1
-0.5
-0.6
-1.5
-0.7
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
sient is 2000 the interval where the zigzag pattern
can be found (α ∈ [1.748, 1.88]), de-
pends also on the coupling parameter
γ
Figure 4.10: Characteristics of the zigzag pattern, α = 1.79, γ = 0.1
19
4.4 Spatiotemporal Intermittency Type I
n
xin
(a) α = 1.7520
n
xin
(b) α = 1.7527
Figure 4.11: Space-time plot with grey scale, γ = 0.001, N = 500, showing every
45th time step of the first 5000 iterations

The transition from an ordered pattern to fully developed chaos goes through
spatiotemporal intermittency (STI). There are laminar motion, i.e. periodic or
weakly chaotic motion and a regular structure, and turbulent bursts in space-
time, i.e. chaotic motion and no regular structure. Two types are distinguished:
type-I without spontaneous creation of bursts and type-II with spontaneous cre-
ation of bursts. In this section STI type-I is described. That bursts do not arise
spontaneous means that a laminar site whose neighbors are also laminar will stay
laminar in the next time step. A typical example is the CML with α in the
period-3 window of the local map. There is a strong connection to probabilis-
tic cellular automata with two states (one for laminar and one for burst) that
show very similar dynamics. There may be a connection to directed percolation
(see [15, 16, 17]).
The Lyapunov spectrum of STI type-I is split into a part with positive exponents
of the relatively same value and a part with negative exponents of the relatively
same value (see figure 4.12(e)). The stepwise structure of the above patterns
disappears thus we have only one big step. The spatial return map of STI type-I
is very characteristic and cannot be mixed up with other patterns. The interval
where this pattern can be found in the coupled logistic lattice is comparatively
very small and strongly restricted.
For more information about the STI type-I see [4, 6, 7, 15, 16, 17].
20
n
n
xin
xin
3000, every 45th time step is shown
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
xi
xi
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
(c) Spatial return map at iteration 5000, (d) Spatial return map at iteration 5000,
N = 500 N = 50000
0.4
0.3
0
0.2
λmax
0.1
-0.5
λj
-0.1
-1
-0.2
-0.3
-1.5
-0.4
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
(e) Lyapunov spectrum, N=64 and tran- (f) The marked region is approximately
sient is 2000 the interval where spatiotemporal in-
termittency of type-I can be found
(α ∈ [1.7516, 1.7529]), depends also on
the coupling parameter γ
Figure 4.12: Characteristics of spatiotemporal intermittency type-I, α = 1.752,

γ = 0.001
21
4.5 Spatiotemporal Intermittency Type II

In difference to STI type-I spontaneous creations of bursts are possible here. For a
site in laminar state whose neighbors are also in laminar state there is a chance of
the creation of a burst. In the coupled logistic lattice the STI type-II is observed
at the transition from pattern selection to fully developed chaos. In the typical
case we start with a spatially ordered pattern (selected pattern) with some defects
that show chaotic motion and extend over some domains of the ordered pattern.
These chaotic regions vanish and are created again and again. However, even
if the whole lattice was in chaotic motion it can be observed that the selected
pattern is build again at least in parts. Thus the temporal change has a very
long memory. For smaller values of α (about α < 1.8) in the STI-type-II-phase
the chaotic motion dies out completely and no new creation of chaotic regions
could be observed. Since there are no bursts anymore the result is a motion like
pattern selection or for higher values of the coupling constant γ a traveling wave.
Thus for these values of α the STI type-II can be seen as a transient motion. For
bigger values of α this could not be observed.
The STI type-II has been observed in many experiments, for example Benard
convection [18, 22, 30], Faraday instability of wave [38], a 2-dimensional electric
convection of liquid crystal [28, 43], and viscous rotating fluid [29] according to
Kaneko [51].
Note that there is no clear border between the pattern selection and STI type-II
as mentioned above since the STI type-II can be a transient phenomenon for some
parameter settings and the distinction between strictly local and already global
chaos (chaotic motion inside a domain or over more domains) is difficult (compare
figure 4.8(b) to figure 4.13(b)). The typical Lyapunov spectrum of STI type-II has
many positive exponents and some are even big. The whole spectrum is smooth
which indicates much chaotic motion and is in clear difference to the split spectra
of STI type-I. For the parameter settings of the run shown in figure 4.13(e)
all chaotic regions vanish after about 12000 iterations (N = 64, for N = 500
that could not be observed). Therefore the Lyapunov Spectrum presented in
figure 4.13(e) shows the values of the transient region. After all chaotic motion
has vanished a spectrum typical for pattern selection (see section 4.2) would
be the result. For a deeper discussion on transients and their influence to the
maximum Lyapunov exponent see chapter 5. Besides the system parameter α
also the lattice size seems to be a critical parameter that controls the length of
the transient phenomenon.
For more information about the STI type-II see [4, 7, 51].
22
n
n
xin
xin
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
xi
xi
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
N = 500 with N = 50000
0.4
0.2
0
0
-0.2
λmax
-0.4 -0.5
λj
-0.6
-0.8
-1
-1
-1.2
-1.5
-1.4
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
sient is 2000 the interval where spatiotemporal in-
termittency of type-II can be found
(α ∈ [1.704, 1.93]), depends also on the
coupling parameter γ and the number
of iterations
Figure 4.13: Characteristics of spatiotemporal intermittency type-II, α = 1.77,

γ = 0.3
23
4.6 Traveling Wave
n
xin xin
(a) α = 1.7, showing every 128th (b) α = 1.5, showing every 256th
time step of the first 15000 it- time step of the first 25000 it-
erations erations
Figure 4.14: Regular and irregular motion depending on the α parameter; γ =

0.7, N = 64
4.6 Traveling Wave

For large coupling constants (about γ > 0.45) the frozen random and the selected
patterns are not fixed in space anymore and move with some certain velocity. Ac-
cording to Kaneko [51] it depends on the parameter α (whether it is in the regime
of the frozen random pattern or the pattern selection) if the motion of domains is
irregular or regular. See figure 4.14 that shows an example for the frozen random
regime with irregular motion (a) and one for the pattern selection regime (b).
The occurrence of traveling waves depends strongly on the lattice size. While for
small lattices it is easy to find a traveling wave it is quite difficult for larger lat-
tices. That leads to the conjecture that traveling waves are more a phenomenon
of small lattice sizes and are few in CMLs with large lattices.
The spatial return maps of figure 4.15 are just another example for spatial re-
turn maps of the pattern-selection-phase since the motion of domains cannot be
represented by this type of diagram. The Lyapunov spectrum (figure 4.15(e))
shows also a typical example for the stepwise spectra of pattern selection without
positive exponents.
To define the parameter settings that lead to traveling waves is very wired since
there are strong dependencies between the system parameter, the coupling con-
stant and the lattice size. That is why the interval given in 4.15(f) has only
schematic meaning. For more information about the traveling wave pattern and
about the relations to chaotic itinerancy see [23, 26, 32, 36, 33, 41, 42, 48, 49].
24
4.6 Traveling Wave
n
n
xin
xin
5000, every 256th time step is shown
1 1
0.8 0.8
0.6
0.6
0.4
0.4
0.2
xi
xi
0.2
0
0
-0.2
-0.2
-0.4
-0.4 -0.6
-0.6 -0.8
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
(c) Spatial return map at iteration 5000, (d) Spatial return map at iteration 5000,
N = 500 N = 50000
-0.5
0
-1
-1.5
λmax
-2 -0.5
λj
-2.5
-3
-1
-3.5
-4
-1.5
-4.5
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
(e) Lyapunov spectrum, same parameter (f) The unmarked region is approximately
settings as in figure 4.15(a) except the interval where the traveling wave
N=64 and transient is 2000 can be found (α ∈ [1.5, 1.8]), depends
also on the coupling parameter γ and
the lattice size
Figure 4.15: Characteristics of the traveling wave, α = 1.69, γ = 0.9
25
4.7 Fully Developed Chaos
n
xin
Figure 4.16: Space-time plot with grey scale, α = 1.9, γ = 0.3, N = 500, showing
every 32nd time step starting at the top with iteration 0 to 4992

In low-dimensional systems chaos is structurally unstable, i.e. there are periodic
windows. For fully developed chaos (FDC) of CMLs no window structures could
be observed. The spatiotemporal chaos of CMLs is stable. For a small coupling
constant and very small lattices the system gets out of FDC into the spatially
homogeneous, periodic state. However, the transient time increases exponentially
with the lattice size and cannot be observed for larger lattices than 10 sites in
practice. [51]
Note that the spatial return map (figure 4.17(c) and 4.17(d)) still shows some
structure and that there are untouched regions in the diagram. The Lyapunov
spectrum (figure 4.17(e)) is smooth and there are many positive exponents. Note
the linear run of the maximum Lyapunov exponent in the regime of the FDC
(figure 4.17(f)).
However, things are not that simple. In [52] F. H. Willeboordse reports that there
is another pattern selection regime for values up to α = 1.9. Therefore high non-
linearity should not be regarded as a guarantee for fully developed chaos.
For more information about fully developed chaos see [27, 34].
26
n
n
xin
xin
(a) Space-time plot with grey scale, α = (b) Space-time plot with lines, 128 sites of
1.99, γ = 0.6, N = 500, starting at the run of figure 4.17(a) are shown
the top with iteration 3000, every 32nd
time step is shown
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
xi
xi
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
i+1 i+1
x x
N = 500 with N = 50000
0.5
0
-0.5
-1
λmax
-1.5 -0.5
λj
-2
-2.5
-1
-3
-3.5
-1.5
-4
0 10 20 30 40 50 60 0.8 1 1.2 1.4 1.6 1.8 2
j α
sient is 2000 the interval where fully developed
chaos can be found (α ∈ [1.8, 2]), de-
pends also on the coupling parameter
γ
Figure 4.17: Characteristics of fully developed chaos, α = 1.99, γ = 0.6
27
Chapter 5
Long Transients
We must use time as a tool,

not as a crutch.
JOHN F. KENNEDY
The only reason for time is

that everything doesn’t happen at once.
ALBERT EINSTEIN
As already mentioned in section 4.5 the STI type-II is sensitive to the number
of iterations for α < 1.8. For these values the STI type-II leads after a tran-
sient time to pattern selection or a traveling wave. If the chaotic regions have
vanished the Lyapunov spectrum looks different than the spectrum of the STI
type-II: there are fewer positive Lyapunov exponents and the maximum exponent
is smaller, too.
To see the overall development of the maximum Lyapunov exponent another
type of diagram is used: the plot of the maximum exponent against the system
parameter α. The data of such a plot is the result of intensive computations.
For figure 5.2(a) the maximum Lyapunov exponent had to be computed for 2000
different values of α. In this diagram one can see that there is obviously no
chaotic motion in the frozen random patterns before α ≈ 1.4. Then the maxi-
mum exponent increases in the beginning very fast later slower but steadily. From
α ≈ 1.55 to α ≈ 1.60 it stays at one level. That is a region where the system
tends to pattern selection, i.e. the domains get smaller but there is still chaotic
motion, because of the parameter setting (coupling constant γ = 0.3) it cannot
be reached. From α ≈ 1.60 to α ≈ 1.64 the domains get larger but at α ≈ 1.64
they are smaller again.
At α ≈ 1.69 the area interesting for this chapter begins. Suddenly the maximum
Lyapunov exponent gets down to values of λmax ≈ 0.06 or as seen in the other
plots (5.2(b), 5.2(c)) even down to zero and below. This is the transition from the
frozen-random-pattern-phase to the pattern-selection-phase although this could
28
already be considered to be the STI-type-II-phase since that begins at least at
α = 1.7. This transition is visualized by figure 5.1 which shows three space-time
plots (in figure 4.7 pattern selection is reached already at α = 1.579 because
of the different coupling constant γ = 0.4). In the first two plots some chaotic
motion in comparatively larger domains can be seen (maximum Lyapunov expo-
nents are λ1max ≈ 0.247 and λ2max ≈ 0.240), so they are representatives of the
frozen random pattern, while in the third one the domains are slightly smaller
and there is no chaotic motion anymore (λ3max ≈ −0.0119).
While in figure 5.2(a) λmax steadily jumps between lower values and higher ones
that are the perfect continuation of the linear run from α ≈ 1.8 to α = 2, al-
ready in the next figure (5.2(b)) with 50000 iterations there are no jumps up to
α ≈ 1.76 and in figure 5.2(c) there are nearly no jumps up to α ≈ 1.8. This
behavior indicates that there is no coexistence of attractors and the STI type-II
is really a transient process that leads to pattern selection for these parameters,
although the degree of chaos in the transient process increases linear with α.
The further run of the maximum exponent is interesting since it does not stay at
about zero but increases in a first effort forming a first hump with its maximum
value at α ≈ 1.7 and then a much bigger one with maximum at α ≈ 1.76 (what
can only be seen in figure 5.2(d)). All of these CMLs show small domains with
little chaotic motion so that these can be considered to belong to pattern selec-
tion although the step to the frozen random pattern is very small.
The small and the big hump as well as the many downward peaks may indicate
self similarity in the run of the maximum Lyapunov exponent between α = 1.69
and α = 1.8.
That a longer and longer part of the interval starting at α = 1.69 gets rid of
high values of exponents the more iterations are performed shows that the tran-
sient time increases with increasing α. That and the peaks in figure 5.2(d) after
α = 1.8 open the question how far the border can be driven.
29
n
xin xin xin

(a) α = 1.688 (b) α = 1.689 (c) α = 1.690
Figure 5.1: Space-time plots in grey scale, showing the transition from frozen
random pattern to pattern selection, γ = 0.3
30
0.4 0.4
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
λmax
λmax
0.15 0.15
0.1 0.1
0.05 0.05
0 0
-0.05 -0.05
-0.1 -0.1
-0.15 -0.15
1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
α α
(a) 5000 iterations, transient is 2000, ran- (b) 50000 iterations, transient is 2000,
dom uniform initialization random uniform initialization
0.35 0.35
0.3 0.3
0.25 0.25
0.2 0.2
λmax
λmax
0.15 0.15
0.1 0.1
0.05 0.05
0 0
-0.05 -0.05
1.65 1.7 1.75 1.8 1.85 1.65 1.7 1.75 1.8 1.85
α α
(c) 100000 iterations, transient is 95000, (d) 1000000 iterations, transient is
random uniform initialization 990000, random uniform initialization
Figure 5.2: Effect of more iterations to the maximum Lyapunov exponent in the
STI type-II region, plot of the maximum Lyapunov exponent, γ = 0.3, N = 64
31
Chapter 6
A Symmetric System and its

Coupled Map Lattice
Correspondence
The mathematical sciences particularly exhibit

order, symmetry, and limitation;
and these are the greatest forms of the beautiful.
ARISTOTLE
6.1 The Original System

The local map used in this chapter instead of the quadratic map is defined by

 fl (xn , α) = α(1 − xn ) if xn < 12
1
xn+1 = f(xn , α) = fc (xn ) = 2 if xn = 12 , (6.1)

fr (xn , α) = αxn (xn − 1) + 1 if xn > 21
with x ∈ [0, 1] and α ∈ [0, 4]. This system was introduced in [56] and will
be called “symmetric map” here. There are two characteristic properties: first,
the symmetry of the system function f with respect to its discontinuity point
f(x, a) = 1 − f(1 − x, a) and as a consequence, second, the asymptotic dynamics
are symmetric attractors or pairs of symmetric coexisting attractors. This sys-
tem shows the border collision period doubling scenario, i.e. starting with two
coexisting limit cycles with period 2j we get after a border collision bifurcation
one limit cycle with period 2j+1 followed by a pitchfork bifurcation.
6.2 A First Coupled Map System Version

Putting the symmetric map mentioned above into equation 1.2 (diffusive nearest
neighbor coupling) results in the CML that is investigated here. As it can be
32
6.3 A Second Coupled Map System Version
seen in the bifurcation diagram (figure 6.1(a)) the behavior is very different to
the dynamics of the local symmetric map for α > 3 but for α < 3 the outer lines
are nearly the same as those of the local map. That there is a strong connection
to the local symmetric map can be seen in figure 6.1(b) (maximum Lyapunov
exponent). At least for α = 0 there is a simple explanation for the six groups
of lines in the interval α ∈ [0, 2]: f(x, 0) is 0 for x < 21 , 21 for x = 21 and 1 for
x > 12 ; since 21 does not occur there are only two alternatives for f(x, 0). In the
system function. This results in xin = (1 − γ)a + γ2 (b + c) with a, b, c ∈ {0, 1}.
Therefore we have two possible values for a and three for b + c which gives us
2 ∗ 3 = 6 possible values. For γ = 0.3 that are: 0, 0.15, 0.3, 0.7, 0.85, 1 (compare
to figure 6.1(a) and 6.2).
Small domains are highly attractive at a comparatively high value of the coupling
constant (γ = 0.3) in this system (see figure 6.5 as a counter example with big
domains in this system). At all values for α the lattice is dominated by small
domains (see figure 6.3). For α < 2 the lattices are homogeneous (fixed) in time
in the asymptotic dynamics. At α = 2 the lattices are homogeneous in both
time and space. For 2 < α < 3.8 there is the period 2 region. So we have a
huge pattern with kinks region starting at α = 0 and ending at α ≈ 3.7 except
α = 2. Around α ≈ 3.7 there is the zigzag pattern region (see figure 6.3(e)) and
around α ≈ 3.83 the STI-type-II region (see figure 6.3(f)). At α ≈ 3.85 the fully
developed chaos begins (see figure 6.3(g)).
The domain sizes get smaller with increasing α (see figure 6.3). However, the
smaller the domains get the more they build global waves and regions with dif-
ferent values of the zigzag domains (see figure 6.4).

Another Coupled Map Lattice can be formulated oriented at the original equation
of the symmetric map. This variant is simpler in its dynamics but maybe not in
its notation:
γ
xin+1 = (1 − γ)g(xin , xin ) + (g(xi−1 i i+1 i
n , xn ) + g(xn , xn )) (6.2)
2
with  1
 α(1 − x) if y < 2
1 1
g(x, y) = 2 if y = 2
. (6.3)
 1
αx(x − 1) + 1 if y > 2
The difference of this system to the above one is that at each site only one func-
tion of the three alternatives is used to compute xin and which one is determined
by the site itself. The effect is that this system shows mainly typical behavior:
It is more similar to the behavior of the symmetric map like the coupled logistic
lattice is to the quadratic map and it shows the typical pattern sequence. As
33
0.8
0.6
x
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
α
(a) Superimposed bifurcation diagram
-1
-2
λmax
-3
-4
-5
-6
-7
0 0.5 1 1.5 2 2.5 3 3.5 4
α
(b) maximum Lyapunov exponent
Figure 6.1: γ = 0.3, N = 64, 5000 iterations, random uniform initialization
34
0.8
0.6
x
0.4
0.2
0
0 0.5 1 1.5 2
α
Figure 6.2: Superimposed bifurcation diagram, γ = 0.3, N = 64, 5000 iterations,

random uniform initialization
35
n
xin
(a) α = 1.0
n
xin
(b) α = 2.5
i
xin
(c) α = 3.1
n
xin
(d) α = 3.4
n
xin
(e) α = 3.7
n
xin
(f) α = 3.83
n
xin
(g) α = 3.9
Figure 6.3: Space-time plots with grey scale, γ = 0.3, N = 500, random uniform
initialization
36
0.8
0.75
0.7
0.65
0.6
xi
0.55
0.5
0.45
0.4
0.35
0.3
0 10 20 30 40 50 60
i
(a) α = 3.1
0.8
0.7
0.6
xi
0.5
0.4
0.3
0.2
0 10 20 30 40 50 60
i
(b) α = 3.4
0.8
0.7
0.6
0.5
xi
0.4
0.3
0.2
0.1
0 10 20 30 40 50 60
i
(c) α = 3.7
Figure 6.4: Amplitude plots for different values for α, γ = 0.3, N = 64, 5000
iterations, random uniform initialization
37
n
xin
Figure 6.5: Amplitude plot, α = 3.1, γ = 0.7, N = 64, 5000 iterations, random
uniform initialization
it can be seen in figure 6.6 the system shows nearly the same behavior like the
local map up to α = 3. The next peak can be recognized but is interrupted by
exponents with smaller absolute values. Figure 6.8 shows a closer view of the
next part and at least the run of the Lyapunov exponent from α = 3.75 to α = 4
reminds one of the region between α = 1.6 and α = 2 of the coupled logistic
lattice.
In the bifurcation diagram of the CML (figure 6.7) it seems to be so that the
typical coexisting attractors of the symmetric map have vanished, however, this
is caused by the high sensibility of the system to the initialization and the random
initialization itself. Changes in the initialization cause the system to jump be-
tween the different attractors. Because of the high resolution of this scan one can
not see the gaps in the “lines”. Up to α = 2 there is no period two attractor and
the border collision period doubling really leads to a period two cycle at α = 3.
That can be shown by a different initialization (here: sine, see figure 6.10).
Indeed the coupled symmetric map shows the typical pattern sequence. Starting
with patterns with kinks (α ∈ [2, 3], see figure 6.9(a)) and the frozen random
regime (α ∈ [3.6, 3.8], figure 6.9(b)), continuing with STI-type-II (α ∈ [3.8, 3.9],
figure 6.9(d)) and leading to fully developed chaos (α ∈ [3.9, 4], figure 6.9(e)).
There is also a STI-type-I regime (around α = 3.8304 for low γ, figure 6.9(c)).
Some difficulties brings the fact that large domain sizes like in the coupled map
lattice have not been observed for any value of the system parameter α or the
coupling constant γ. Therefore the identification of the pattern selection regime
is difficult and no patterns with only few different domain sizes have been ob-
served.
That high attraction of small domain sizes is characteristically for this system
(as for the above version of the symmetric map) shows figure 6.11 where one can
see regions with the zigzag patterns even for this comparatively high coupling
constant (γ = 0.3).
38
-1
-2
λmax
-3
-4
-5
-6
-7
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 6.6: Maximum Lyapunov exponent, γ = 0.3, N = 64, 5000 iterations,

transient is 3000, random uniform initialization, using the linearized system func-
tion
39
0.9
0.8
0.7
0.6
x
0.5
0.4
0.3
0.2
0.1
0
2.8 3 3.2 3.4 3.6 3.8 4
α
Figure 6.7: Bifurcation diagram, γ = 0.3, N = 64, 5000 iterations, random

uniform initialization; the bifurcation diagrams of all 64 sites are printed into
one
40
0.4
0.3
0.2
0.1
λmax
-0.1
-0.2
-0.3
3.5 3.55 3.6 3.65 3.7 3.75 3.8 3.85 3.9 3.95 4
Figure 6.8: Maximum Lyapunov exponent, γ = 0.3, N = 64, 5000 iterations,

transient is 3000, random uniform initialization, using the linearized system func-
tion
41
n
xin
(a) α = 2.5, γ = 0.3
n
xin
(b) α = 3.2, γ = 0.3
n
xin
(c) α = 3.8304, γ = 0.001
n
xin
(d) α = 3.81, γ = 0.3
n
xin
(e) α = 3.9, γ = 0.3
Figure 6.9: Space-time plots with grey scale, N = 500, random uniform initial-
ization
42
0.7 0.8
0.65 0.75
0.6 0.7
0.55 0.65
0.5 0.6
0.45 0.55
x
x
0.4 0.5
0.35 0.45
0.3 0.4
0.25 0.35
0.2 0.3
3 3.05 3.1 3.15 3.2 3 3.05 3.1 3.15 3.2
α α
(a) offset is 0.65 (b) offset is 0.35
Figure 6.10: Bifurcation diagram, initialization with sine (amplitude is 0.1, one
period), 1000 iterations
n
xin
Figure 6.11: Space-time plot with lines, α = 3.71, γ = 0.3, N = 128
43
Chapter 7
Spatial Bifurcations
The secret to creativity

is knowing how to hide your sources.
ALBERT EINSTEIN
As already stated coupled map lattices show not only qualitative changes in
time but also in space, e.g. the increasing number of domains with increasing
non-linearity (see for example figure 4.4). This chapter focuses on spatial changes
that can be indicated by bifurcation diagrams.
Although the following phenomenon occurs in both the second variant of the
coupled symmetric map and the coupled logistic lattice we get back to the lat-
ter one because the transients in the examined region are much smaller there.
The region that is investigated here is the interval α ∈ [0.75, 1.25]. Crutchfield
and Kaneko presented a site histogram for this region that shows a high proba-
bility for the two outer curves that follow the period two structure of the local
quadratic map (see figure 7.1) and a cross-hatching inside the envelope caused by
kinks [12]. However, these are transient dynamics (only 500 iterations). A much
more intensive computation with 200000 iterations and 5000 values of α results
in the superimposed bifurcation diagram of figure 7.2. Note the vertical broken
lines before α ≈ 0.85 and the funnels at seven positions starting at α ≈ 0.874.
Again the outer curves have the highest probability, the parallel running curves
are less probable and the funnels are least probable.
The open ends of the funnels are positions of spatial bifurcations. This bifurca-
tion is an increase of the number of domains. The computed maximum Lyapunov
exponent shows some numerical artefacts (peaks above zero) that correspond to
the positions of the funnels (see figure 7.3). The numerical problems might be
caused by a transient phenomenon. The correctly computed maximum exponent
is awaited to be zero or less since there is no chaotic motion. Although figure 7.2
gives a good overview of the whole coexisting attractors with its random uniform
initialization it is technically nearly impossible to investigate one single attractor
only. Therefore the sine with many periods (above nine) is used as initializa-
44
tion. That leads to figure 7.4 showing only two funnels: the first indicating the
bifurcation from one domain to two and the second one from two to ten. Again
both spatial bifurcations have an impact to the maximum Lyapunov exponent
(figure 7.5). With the series of diagrams in figure 7.6 the cause for the funnels
can clearly be seen. Before the funnel there are two domains. At the opening of
the funnel suddenly the big domain is divided into nine new and smaller ones.
There are two types of new domains: bigger ones with seven sites (here) and
a flat top and smaller ones of only size five (here) with sharp tops. In general
one can distinguish into bigger and smaller domains. The dependency of these
domains to the lattice size is unclear since many variations have been observed:
for example N = 20, 3 and 5; N = 32, 4 and 6; N = 100, 14 and 18; N = 512, 4
and 15, or 6 and 19. Both types of the domains cause the funnels. All the sites
of the smaller types have new values that did not occur before the bifurcation.
The values of the sites of the bigger domain types depend on the neighboring do-
main and can be different in their behavior according to which half of the domain
the sites belong to. If the neighboring domain is an old one that was existent
already before the bifurcation then the neighboring half of the new domain has
the same values as the sites of the old domain at the corresponding positions. If
the neighboring domain is a new one and therefore one of the smaller type the
sites of the neighboring half have values that did not occur before. The values of
the smaller type build always those halves of the funnels that are next to the mid
(the mid is a virtual axes at x ≈ 0.6 here). While the values of the bigger type
build always those halves of the funnels that are averted from the mid. With
increasing non-linearity the new values of the smaller types get bigger for x > 0.6
while the values of the bigger ones get smaller and for x < 0.6 vice versa until the
former new values are equal to those of the old domains. This behavior causes
the funnels.
See figures 7.7 and 7.8 as further examples for different initializations and fig-
ure 7.8 as an example for smaller emerging domain sizes.
Now we focus on the vertical broken lines before α ≈ 0.85. Already the fact
that with fewer iterations the empty space at α < 0.77 can be filled with vertical
lines indicates that this seems to be a transient phenomenon. This is supported
by another fact that for the most sine initializations these vertical lines die out
pretty fast. But this will be discussed later again. The transient times seem to
increase in principle with increasing non-linearity as already seen in chapter 5
concerning the run of the maximum Lyapunov exponent.
Of course these vertical lines are actually no lines. This is caused by the small
intervals this phenomenon occurs in. So that even with high resolutions of the
parameter scans they are touched only by one pixel. Therefore it is amazing
that so many occur anyway and it seems to be so that there are infinitely many
of them. Figure 7.9 shows a closer look at this phenomenon. The series of the
diagrams shows that with increasing time the phenomenon contracts (the sizes
of the intervals are approximately: 1.4 ∗ 10−3 , 4 ∗ 10−5 , 2.4 ∗ 10−12 ) but it stays
45
1
0.8
0.6
0.4
x
0.2
-0.2
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
Figure 7.1: Bifurcation diagram of the quadratic map
qualitatively unchanged. Clearly that is a negative exponential decrease (see fig-

ure 7.10). Although from a mathematical point of view this will lead to size zero
the transient time could be infinite and therefore it could be hardly regarded as
a transient phenomenon in praxis. Finally this leads to the conjecture that there
might be infinitely many such phenomena at each time in the system.
Figure 7.11 shows some amplitude plots to show what happens on the lattice.
With increasing non-linearity the following happens: At first the amplitudes of
the sites in the mid get smaller until the middle site is stable in time (because
of discrete space there might be no exactly stable site of course). After that
two kinks emerge and move to the sides (because of the ring topology there are
actually no sides of course) while the amplitudes in the mid grow again. At some
point there is only one stable site left and also there the amplitudes grow again.
At the end the situation is exactly the same as in the beginning.
46
1
0.8
0.6
x
0.4
0.2
-0.2
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
α
Figure 7.2: Superimposed bifurcation diagram (bifurcation diagrams of all, N=64

sites in one), γ = 0.3, N=64, 200000 iterations, 5000 points computed, random
uniform initialization
47
0.2
-0.2
λmax
-0.4
-0.6
-0.8
-1
-1.2
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
α
Figure 7.3: Maximum Lyapunov exponent with numerical artefacts, same pa-
rameter settings as in figure 7.2
48
1
0.8
0.6
x
0.4
0.2
-0.2
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
α
Figure 7.4: Superimposed bifurcation diagram, γ = 0.3, N = 64, 50000 iterations,

5000 points computed, sine initialization (period=24)
0.01
-0.01
-0.02
λmax
-0.03
-0.04
-0.05
-0.06
-0.07
-0.08
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
α
Figure 7.5: Maximum Lyapunov exponent with numerical artefacts, same pa-
rameter settings as in figure 7.4
49
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
xi
xi
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
i i
(a) α = 0.971 (b) α = 0.973
1 1
0.8
0.8
0.6
0.6
xi
0.4
0.4
0.2
0.2
0
-0.2
0 10 20 30 40 50 60 0.85 0.9 0.95 1 1.05 1.1
i α
(c) α = 1.015 (d) Superimposed bifurcation diagram
with marked positions
Figure 7.6: Amplitude plots of different values for α, γ = 0.3, N = 64, 50000
iterations, sine initialization (period=24)
50
1
0.9
0.8
0.7
0.6
xi
0.5
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60
i
(a) α = 0.97
0.8
0.6
xi
0.4
0.2
0
0 10 20 30 40 50 60
i
(b) α = 1.0
0.8
0.6
xi
0.4
0.2
-0.2
0 10 20 30 40 50 60
i
(c) α = 1.015
iterations, random uniform initialization
51
1 1
0.8 0.8
0.6 0.6
xi
xi
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i i
(a) α = 1.03 (b) α = 1.04
1 1
0.8 0.8
0.6 0.6
xi
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
0 5 10 15 20 25 30 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
i α
(c) α = 1.06 (d) Superimposed bifurcation diagram
with marked positions
iterations, sine initialization (period=19)
52
0.85
0.8
0.75
0.7
0.65
x
0.6
0.55
0.5
0.45
0.767951 0.767952 0.767952 0.767952 0.767952 0.767952 0.767953
α
(a) 5000 iterations
0.85
0.8
0.75
0.7
0.65
x
0.6
0.55
0.5
0.45
0.76793 0.767935 0.76794 0.767945 0.76795 0.767955 0.76796 0.767965 0.76797 0.767975
α
(b) 10000 iterations
0.85
0.8
0.75
0.7
0.65
x
0.6
0.55
0.5
0.45
0.767952 0.767952 0.767952 0.767952 0.767952
α
(c) 35000 iterations
Figure 7.9: Superimposed bifurcation diagrams, γ = 0.3, N = 32, sine initializa-

tion (period=19)
53
0.01
0.0001
1e-06
Interval Size
1e-08
1e-10
1e-12
1e-14
1e-16
5000 10000 15000 20000 25000 30000 35000 40000 45000 50000
Iterations
Figure 7.10: Decrease of the interval size with increasing number of iterations
54
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
xi
xi
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i i
(a) α = 0.7674 (b) α = 0.7675
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
xi
xi
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i i
(c) α = 0.7676 (d) α = 0.7687
0.85 0.85
0.8 0.8
0.75 0.75
0.7 0.7
xi
xi
0.65 0.65
0.6 0.6
0.55 0.55
0.5 0.5
0.45 0.45
0 5 10 15 20 25 30 0 5 10 15 20 25 30
i i
(e) α = 0.76875 (f) α = 0.7689
Figure 7.11: Development of the orbits with increasing α, plotted are the itera-
tions 5000 to 5015, parameters as in figure 7.9(a)
55
Chapter 8
Using the AnT 4.669 Software to

simulate Coupled Map Lattices
If you put tomfoolery into a computer,

nothing comes out of it but tomfoolery.
But this tomfoolery, having passed through a very expensive machine,
is somehow ennobled and no one dares to criticize it.
PIERRE GALLOIS
Insanity: doing the same thing over and over again

and expecting different results.
ALBERT EINSTEIN
AnT 4.669 is a software package to simulate and investigate dynamical sys-

tems. Due to its generic architecture many different classes of dynamical systems
can be simulated, e.g. maps, ordinary and delay differential equations, many
sub-classes of these and of course coupled map lattices. A main feature of AnT is
the variety of provided investigation methods like period analysis, Lyapunov ex-
ponents, generalized Poincare section analysis and much more. Another feature
are the so called “scan runs”, i.e. the ability to investigate a dynamical system by
varying some quantities with relevant influence such as the system parameters,
initial values, or even some parameters of the investigation methods.
To simulate and analyze a CML one has to implement the system function (which
is quiet easy and can be managed with little knowledge of programming) and
adapt the initialization file to one’s purpose. After that the simulation can begin.
The investigation methods of AnT relevant to simulate CMLs are the Lyapunov
exponent calculation, the space-time plots in grey scale in PGM format, the pe-
riod analysis and the trajectory output. At the time this text is written amplitude
plots, spatial return maps and space-time plots with lines are not provided but
can be produced by converting the orbit output files, e.g. by programming tiny
tools in C, and using for example gnuplot.
56
Normally all these investigation methods are reliable except the Lyapunov expo-
nent calculation that suffers from numerical failures sometimes. However, this
concerns only one variant of the implemented Lyapunovs calculations. The ap-
proach of Wolf et al. [8] is implemented. There are two variants of the method,
the first one, based on the original approach, which uses the linearized system
function, and the second one, based on a modified approach, which uses the un-
changed system function only. This method uses the temporal averaged local
divergence rates for an approximation of the Lyapunov exponents.
Using the linear method the ith Lyapunov exponent is obtained as
1
λi = lim ln[ith eigenvalue of (Jn−1 Jn−2 ...J0 )], (8.1)
n→∞ n
where the Jacobi matrix at time n is given by
∂xin+1
(Jn )i,j = . (8.2)
∂xjn
As an example the Jacobi matrix for the CML with quadratic map and nearest
neighbor coupling at time n is given by
γ 2 γ N 
β(x1n )

2 (xn ) 0 ... 0 0 0 2 (xn )
 γ (x1n ) β(x2n ) γ 3
(x n) 0 ... 0 0 0 
2 γ 2
2
γ 4
β(x3n )

2 (xn ) (x n)
 0 0 ... 0 0 
2
γ 3 γ 5
β(x4n )
 
2 (xn ) (x n)
 0 0 0 ... 0 
 2 
Jn = −2α  ... .. ..  ,

 . .  
γ N−4 γ N−2
 0
 ... 0 2 (x n ) β(xN−3
n ) 2 (xn ) 0 0  
γ N−3 γ N−1
 0
 0 ... 0 2 (x n ) β(xN−2
n ) 2 (xn ) 0  
γ N−2 γ N 
 0 0 0 ... 0 2 (xn ) β(xN−1
n ) 2 (xn )
γ 1 γ N−1
2 (xn ) 0 0 0 ... 0 2 (xn ) β(xN n)
(8.3)
with β = (1 − γ).
To use the modified approach of the Lyapunov exponent calculation for CMLs
is not recommended using AnT 4.669 Release 3.1.5, since some severe problems
occurred, e.g. unsorted Lyapunov spectra (see figure 8.1). Some scans over the
length of the deviation vectors (see figure 8.3) show by their stepwise structure
and/or the independence to that there are numerical problems. As a compari-
son see figure 8.2 which is an example of a valid scan using the single quadratic
map.
Therefore an investigation of that problems would be interesting although the
linearized calculation is running fine and produces good results. Since the lin-
earized approach seems to run faster than the modified one actually no efforts
have to be done except a warning for users to use the method with care. Please
note that this method produces mainly correct results (especially for high values
of non-linearity) and bad results occur seldom.
Other good improvements would be status reports (percentage finished) even for
57
0.4 0.3
0.2
0.3
0.1
0.2
0
0.1
-0.1
λj
λj
0 -0.2
-0.3
-0.1
-0.4
-0.2
-0.5
-0.3
-0.6
-0.4 -0.7
0 10 20 30 40 50 60 0 10 20 30 40 50 60
j j
(a) α = 1.752, γ = 0.001 (b) α = 1.79, γ = 0.1
Figure 8.1: Lyapunov spectra using the modified approach of calculation, coupled
logistic lattice, N = 64, 5000 iterations and transient 2000, random uniform
initialization (full range, seed=5)
single runs with the calculation of the Lyapunov spectrum (causing intensive
computation) and better formats of the orbit and Lyapunov exponent outputs
to plot them with gnuplot easily (e.g. no header, every value in a new line). A
more sophisticated improvement would be a period analyzer that analyses only
parts of the lattice, i.e. it should detect the domains and their period. The part
of the lattice to be analyzed could be specified by the user, too.
58
0.23958
0.23956
0.23954
0.23952
λmax
0.2395
0.23948
0.23946
0.23944
0.23942
1e-10 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001
Figure 8.2: An example of a scan over the length of the deviation vectors of
the single quadratic map
0.01 0.12
0 0.1
−0.01
0.08
−0.02
0.06
−0.03
λmax
λmax
−0.04 0.04
−0.05 0.02
−0.06
0
−0.07
−0.02
−0.08
−0.09 −0.04
−0.1 −0.06
1e−10 1e−09 1e−08 1e−07 1e−06 1e−05 0.0001 1e−10 1e−09 1e−08 1e−07 1e−06 1e−05 0.0001

(a) α = 1.3 (b) α = 1.43
0.25 0.336
0.334
0.2
0.332
0.33
0.15
λmax
λmax
0.328
0.1 0.326
0.324
0.05
0.322
0.32
0
0.318
−0.05 0.316
1e−10 1e−09 1e−08 1e−07 1e−06 1e−05 0.0001 1e−10 1e−09 1e−08 1e−07 1e−06 1e−05 0.0001

(c) α = 1.55 (d) α = 1.88
Figure 8.3: Scans over the length of the deviation vectors using the modified
approach, coupled logistic lattice, γ = 0.3, N = 64, 5000 iterations and transient
3000, random uniform initialization
59
Chapter 9
Conclusion
We know nothing at all.

All our knowledge is the knowledge of school children.
The real nature of things we shall never know.
ALBERT EINSTEIN
You see things; and you say, ’Why?’

But I dream things that never were;
and I say, ”Why not?”
GEORGE BERNARD SHAW
in ”Back to Methuselah”, part 1, act 1
This work provides an introduction to the basics of coupled map lattices,

gives some references to applications and focuses on some issues that have not
been treated much in the literature (e.g. increasing transients with increasing
non-linearity parameter, superimposed bifurcation diagrams of all sites) maybe
because of limited computational power or the lack of appropriate simulation
software in the past. These issues provide much work for the future.
A short investigation of the new symmetric map in two coupled map versions
has been done, however, much more is left open. Note also that the vast fields of
globally and unidirectional coupled lattices are not treated here.
The AnT 4.669 software package has proved to be a very well instrument to
investigate coupled map lattices although some improvements can be done.
Since this paper is written at a computer science institute one thing needs to
be mentioned: Besides the many challenges the coupled map lattices provide to
classical non-linear dynamics scientists there are deep questions that should be
treated by computer scientists like: What kind of computation can be done by
coupled map lattices and what can they do that discrete computers cannot?
60
Chapter 10
Acknowledgments
Thank you for sending me a copy of your book.

I’ll waste no time reading it.
MOSES HADAS
Science is a wonderful thing

if one does not have to earn one’s living at it.
ALBERT EINSTEIN
For what I have received,

my Lord make me truly thankful.
And more truly for what I have not received.
STORM JAMESON
This text is the product of good team work. Therefore all my thanks go to
Michael Schanz and Viktor Avrutin, who spent so many hours of their worthy
time. Their effort in supporting me was much more than the performance of
their duty. During our discussions the essential ideas for this work emerged. We
became really a team because with our different personal backgrounds everybody
was able to contribute his share. It’s a great thing to meet nice people and do
some scientific work together.
All the simulations were done by the AnT 4.669 software package which was
developed by V. Avrutin, R. Lammert, M. Schanz, G. Wackenhut and many of
my fellow students.
For some computationally very intensive parameter scans some machines of the
IPVS were used.
Finally I would like to thank my parents for supporting me and giving me the
chance to focus on my work as a student.
61
Appendix
Convert the system parameter α from the logistic map

to the quadratic map
Do not worry about your difficulties in mathematics,

I assure you that mine are greater.
ALBERT EINSTEIN
The following computations will result in a homeomorphism (also called a con-

tinuous transformation, do not mix up with homomorphism) which is an equiva-
lence relation and a one-to-one correspondence between points in two geometric
figures or topological spaces that is continuous in both directions [58].
Given is the quadratic map
xn+1 = 1 − αx2n . (1)
To convert this into the logistic map given by yn+1 = ayn (1 − yn ) we start with
a general polynomial of forth degree to represent α by the system parameter of
the logistic map a:
α = da4 + ea3 + fa2 + ga + h. (2)
For xn we are satisfied by a linear formulation:
xn = byn + c. (3)
Putting equation 3 into 1 results in
byn+1 + c = 1 − α(byn + c)2 ; (4)

1
yn+1 = (1 − α(byn + c)2 − c). (5)
b
Putting equation 2 into 5 results in
1 d a4 c 2 ea3 c2
yn+1 = −bda4 y2 −2da4 yc− −bea3 y2 −2ea3 yc− −bfa2 y2 −2fa2 yc
b b b
fa2 c2 gac2 hc2 c
− − bgay2 − 2acgy − − bhy2 − 2hyc − − . (6)
b b b b
62
Now we consider three terms; at first the constant term in equation 6 con-
cerning y:
1 d a4 c 2 e a 3 c 2 f a 2 c 2 g a c 2 h c 2 c
T0 := − − − − − − ; (7)
b b b b b b b
at second the linear coefficients:
T1 := −2 d a4 c − 2 e a3 c − 2 f a2 c − 2 g a c − 2 h c; (8)
and at last the quadratic coefficients:
T2 := −b d a4 − b e a3 − b f a2 − b g a − b h. (9)
Using some simple conditions a system of equations can be formulated and solved.
In the logistic map the linear and the quadratic coefficients are equal except their
sign. We set T1 = −T2 and compute b:
b = −2c. (10)
Since the linear coefficients have to be a we set T1 = a and compute c:

1 a
c=− . (11)
2 d a + e a + f a2 + g a + h
4 3
Since there is no constant in the logistic map we set T0 = 0 and compute d:

1 4 g a + 4 e a 3 − a2 + 4 f a 2 + 2 a + 4 h
d=− . (12)
4 a4
Since d should be independent of α and the degree of the numerator of equation 12
is smaller than the degree of the denominator we have to set d = 0 and compute
e now:
1 4 g a + 4 h − a 2 + 4 f a2 + 2 a
e=− . (13)
4 a3
Since e should be independent of α we set e = 0 and compute f:
1 4 g a + 4 h − a2 + 2 a
f=− . (14)
4 a2
Again f should be independent of α. Since the degree of the numerator and
the denominator are equal we have additional to f = 0 the choice to set f = 41 .
Therefore we set the numerator equal to zero and a2 = 0: 4 g a + 4 h − a2 + 2 a =
0 and compute g:
1 2h+a
g=− . (15)
2 a
Since g should be independent of α we get h:
h = 0. (16)
63
2
1.5
α 1
0.5
0
2 2.5 3 3.5 4
Figure 1: The connection between the system parameter of the quadratic map α
and the system parameter of the logistic map a
Now all variables except a are computed and equation 6 results in the logistic
map. For α we get:
a a
α = ( − 1), (17)
2 2
which is the formula to compute the system parameter of the quadratic map on
a restricted interval α ∈ (0, 2] by the given system parameter of the logistic map
a ∈ (2, 4] (see figure 1).
The stretching of the state space of the logistic lattice into the bigger interval
of the quadratic map is done by the linear transformation yn = b1 (xn − c) with
4 −2
b = a−2 and c = α−2 .
64
Erklärung / Declaration
Hiermit versichere ich, diese Arbeit

selbstständig verfasst und nur die
angegebenen Quellen benutzt zu haben.
/
With this I assure that this work was
written independently and using the
stated sources only.
(Heiko Hamann)
65
Bibliography
[1] L. D. Harmon, “Studies with artificial neurons”, Kybernetik, 1, 1961, 89

[2] J. Nagumo and S. Sato, “On a response characteristic of a mathematical
neuron model”, Kybernetik, 10, 1972, 155
[3] R. Shaw, “Strange attractors, chaotic behavior and information flow”, Z.
Naturforschung, 36 (a), 1981, 80
[4] K. Kaneko, “Period-doubling of kink-antikink patterns, quasi-periodicity in
antiferro-like structures and spatial intermittency in Coupled Map Lattices -
toward a prelude to a ’field theory of chaos”’, Prog. Theor. Phys., 72, 1984,
480-6
[5] W. Freeman and C. A. Skarda, “Spatial EEG patterns, non-linear dynamics
and perception: the neo-Sherringtonian view”, Brian Res. Rev., 10, 1985, 147
[6] K. Kaneko, “Spatiotemporal Intermittency in Coupled Map Lattices”, Prog.
Theor. Phys., 74 (5), 1985, 1033-1044
[7] K. Kaneko, “Kinks and spatially complex behavior in Coupled Map Lattices”,
edited by S. Takeno, Springer, Berlin, 1985, 272-7
[8] A. Wolf, J.B. Swift, H.L. Swinney, J.A. Vastano, “Determining Lyapunov
exponents from a time series”, Physica D, 16, 1985, 285-317
[9] W. Freeman, “Petit mal seizure spikes in olfactor bulb and cortex caused
by runaway inhibition after exhaustion of excitation”, Brain. Res. Rev., 11,
1986, 259
[10] K. Kaneko, “Lyapunov analysis and information flow in Coupled Map Lat-
tices”, Physica D, 23, 1986, 436-47
[11] K. Kaneko, talk at the First International Conference on Form, Tsukuba,
Japan, 1986
[12] J.P. Crutchfield and K. Kaneko “Phenomenology of spatiotemporal chaos”
in Directions in Chaos (edited by H. Bai-Lin, World Scientific Publishing Co.,
Singapore), 1987
66
BIBLIOGRAPHY
[13] K. Kaneko, “Pattern competition intermittency and selective flicker noise in

spatiotemporal chaos”, Physic. Lett. A, 125, 1987, 25-31
[14] K. Aoki and N. Mugibayashi, “Onset of turbulent pattern in a Coupled Map

Lattice. Case for soliton-like behavior”, Phys. Lett. A, 128, 1988, 349
[15] H. Chate and P. Manneville, “Spatiotemporal intermittency in Coupled Map

Lattices”, Physica D, 32, 1988, 409
[16] H. Chate and P. Manneville, “Structure of clusters generated by spatiotem-

poral intermittency and directed percolation in 2-space dimension”, Phys.
Rev. A, 38, 1988, 4351
[17] H. Chate and P. Manneville, “Continuous and discontinuous transition to

spatiotemporal intermittency in 2-dimensional Coupled Map Lattices”, Eu-
rophys. Lett., 6, 1988, 591
[18] S. Ciliberto and P. Bigazzi, “Spatiotemporal intermittency in Rayleigh-

Benard convection”, Phys. Rev. Lett., 60, 1988, 286
[19] J. P. Crutchfield and K. Kaneko, “Are attractors relevant to turbulence?”,

Phys. Rev. Lett., 60, 1988, 2715-18
[20] K. Kaneko, “Chaotic diffusion of turbulence defect and pattern selection in

spatiotemporal chaos”, Europ. Physc. Lett., 6, 1988, 193-9
[21] M. Rotenberg, “Life on a regular polygon: maps induced by spatially ex-

tended prey-predator equations”, Physica D, 30, 1988, 192
[22] F. Daviaud, M. Dubois and P.Berge, “Spatio-temporal intermittency in

quasi-1-dimensional Rayleigh Benard convection”, Europhys. Lett., 9, 1989,
441
[23] K. Ikeda, K. Matsumoto and K. Ohtsuka “Maxwell-Bloch turbulence”, Prog.

Theor. Phys. Suppl., 99, 1989, 295
[24] K. Kaneko, “Pattern dynamics in spatiotemporal chaos”, Physica D, 34,

1989, 1-41
[25] K. Kaneko, “Spatiotemporal chaos in one- and two-dimensional Coupled

Map Lattices”, Physica D, 37, 1989, 60-82
[26] K. Kaneko, “Chaotic but regular posi-nega switch among coded attractors
by cluster size variation”, Phys. Rev. Lett., 63, 1989, 219-24
[27] K. Kaneko, “Towards thermodynamics of spatiotemporal chaos”, Prog.

Theor. Phys. Suppl., 99, 1989, 263-87
67
BIBLIOGRAPHY
[28] S. Nasuno, M. Sano and Y. Sawada, “Defect turbulence in EHD convec-

tion of liquid crystal”, in Cooperative Dynamics in Complex Systems (ed. H.
Takayama, Springer, Berlin), 1989
[29] Y. Couder, reported at the summer school “Pattern, Order, and Chaos”
(Como), 1990
[30] F. Daviaud, M. Bonetti and M. Dubois, “Transition to turbulence via spa-
tiotemporal intermittency in 1-d Rayleigh Benard convection”, Phys. Rev.
Lett. A, 42, 1990, 3388
[31] P. Davis, “Applications of optical chaos to temporal pattern search in a
non-linear optical resonator”, Japan J. Appl. Phys. Lett., 29, 1990, 1238
[32] K. Kaneko, “Clustering, coding, switching, hierarchical ordering, and control
in network of chaotic elements”, Physica D, 41, 1990, 137-72
[33] K. Kaneko, “Globally coupled chaos violates the law of large numbers”,
Phys. Rev. Lett., 65, 1990, 1391-4
[34] K. Kaneko, “Supertransients, spatiotemporal intermittency, and stability of
fully developed spatiotemporal chaos”, Phys. Lett. A, 149, 1990, 105-12
[35] J. F. Magnan, J.M. Bertille and J.C. Perez, “Global optimization with a dy-
namical system”, talk at the SIAM conference on dynamical systems, Florida,
1990
[36] I. Tsuda, “Chaotic neural networks and thesaurus”, Neurocomputers and
Attention, (ed. A. V. Holden and V. I. Kryukov, Manchester University Press),
1990
[37] G. Cocho and G. Martinez-Mekler, “On a Coupled Map Lattice formulation
of the evolution of genetic sequences”, Physica D, 51, 1991, 119
[38] J. Gollub and R. Ramshankar, “Spatiotemporal chaos in interfacial waves”,
in New Perspectives in Turbulence (ed. S. Orszag and L. Sirovich, Springer,
Berlin), 1991
[39] A. V. Holden, J. V. Tucker and B. C. Thompson, “Can excitable media be
considered as computational systems?”, Physica D, 49, 1991, 240
[40] H. Ito and L. Glass, “Spiral breakup in a new model of discrete excitable
media”, Phys. Rev. Lett., 66, 1991, 671
[41] K. Kaneko, “Globally coupled circle maps”, Physica D, 54, 1991, 5-19
[42] K. Kaneko, “Partition complexity in network of chaotic elements”, J. Phys.
A, 24, 1991, 2107-19
68
BIBLIOGRAPHY
[43] S. Nasuno and S. Kai, “Instabilities and transition to defect turbulence in

electrohydrodynamic convection to nematics”, Europhys. Lett., 14, 1991, 779
[44] Y. Liu, “A non-linear diffusion system on a discrete space-time lattice”,
Physica D, 54, 1991, 147
[45] R. V. Sole and J. Valls, “Order and chaos in a 2D Lotka-Volterra Coupled
Map Lattice”, Phys. Lett. A, 153, 1991, 330
[46] “Waves and Patterns in Chemical and Biological Media”, Physica D, 49,
1991, ed. H. L. Swinney and V. I. Krinsky
[47] T. Ikegami and K. Kaneko, “Coupled Map Lattice for homeochaotic sym-
bionet in evolution”, CHAOS, 2, 1992, 397-408
[48] K. Kaneko, “Global traveling wave by local phase slips”, Phys. Rev. Lett.,
69, 1992, 905-908
[49] K. Kaneko, “Mean field fluctuation in network of chaotic elements”, Physica
D, 55, 1992, 368
[50] T. Yanagita, “Coupled Map Lattice for boiling”, Phys. Lett. A, 165, 1992,
405
[51] “Theory and Applications of Coupled Map Lattices”, edited by K. Kaneko,
Wiley, New York, 1993
[52] F. H. Willeboordse, “Pattern selection at high non-linearity in a diffusively
coupled logistic lattice”, Physical Review E, 47, 1993, 1419-22
[53] J. Bascompte et al., “Spatially induced Bifurcations in Single-Species Pop-
ulation Dynamics”, Journal of Animal Ecology, 63, 1994, 256-264
[54] K. Kaneko, “Coupled Maps with Growth and Death: an Approach to Cell
Differentiation”, Physica D, 103, 1997, 505-527
[55] A. Sharma, N. Gupte, “Spatiotemporal intermittency and scaling laws in
inhomogeneous Coupled Map Lattices”, Phys. Rev. E, 66, 2002
[56] V. Avrutin and M. Schanz, “On the Border Collision Period Doubling Sce-
nario”, accepted for publication in Phys. Rev. E
[57] P. Lind et al., “Coupled Bistable Maps: A Tool to study Convection Pa-
rameterization in Ocean Models”, International Journal of Bifurcation and
Chaos, 14 (3), 2004, 999-1015
[58] E. W. Weisstein. “Homeomorphism.” From MathWorld - A Wolfram Web
Resource. http://mathworld.wolfram.com/Homeomorphism.html
69

Mapas Acoplados

Uploaded by

Copyright:

Available Formats

Mapas Acoplados

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mapas Acoplados

Uploaded by

Copyright:

Available Formats

Studiengang: Informatik

Prüfer: Prof. Dr. rer. nat. habil. P. Levi

Betreuer: Dr. rer. nat. Michael Schanz

begonnen am: 1. April 2004

beendet am: 30. September 2004

CR-Klassifikation: G.1.10, J.2

Studienarbeit Nr. 1943

Institut für Parallele und

1 Coupled Map Lattices - an Introduction 2

2 Applications of Coupled Map Lattices 4

3 Properties of Coupled Map Lattices 6

4 Phenomenology of Coupled Map Lattices 8

6 A Symmetric System and its Coupled Map Lattice Correspon-

8 Using the AnT 4.669 Software to simulate Coupled Map Lattices 56

Coupled Map Lattices - an

When I am judging a theory,

There is nothing more difficult to take in hand,

The coupled map lattices were introduced by Kunihiko Kaneko as a simple

Figure 1.1: Quadratic map for α = 2

Applications of Coupled Map

The trouble with facts is

Physics is like sex.

Properties of Coupled Map

I have deep faith

If you are out to describe the truth,

3.3 Border Conditions

3.4 Initial Conditions

Phenomenology of Coupled Map

I want to know how God created this world.

I have no special talents.

4.1 Frozen Random Pattern

Figure 4.2: Amplitude plots, α = 1.40116, γ = 0.3, N=30000

of magnitude smaller in the right diagram.

Figure 4.3: Amplitude plot, same parameter settings as in figure 4.4

Figure 4.5: Characteristics of the frozen random pattern, α = 1.43, γ = 0.3

Figure 4.6: Spatial return maps, α = 1.40116, γ = 0.3, N=200000, uniform

4.2 Pattern Selection

Figure 4.8: Characteristics of pattern selection, α = 1.7, γ = 0.4

4.3 Zigzag Pattern

Figure 4.10: Characteristics of the zigzag pattern, α = 1.79, γ = 0.1

4.4 Spatiotemporal Intermittency Type I

Figure 4.12: Characteristics of spatiotemporal intermittency type-I, α = 1.752,

4.5 Spatiotemporal Intermittency Type II

Figure 4.13: Characteristics of spatiotemporal intermittency type-II, α = 1.77,

Figure 4.14: Regular and irregular motion depending on the α parameter; γ =

4.6 Traveling Wave

Figure 4.15: Characteristics of the traveling wave, α = 1.69, γ = 0.9

4.7 Fully Developed Chaos

Figure 4.17: Characteristics of fully developed chaos, α = 1.99, γ = 0.6

We must use time as a tool,

The only reason for time is

xin xin xin

A Symmetric System and its

The mathematical sciences particularly exhibit

6.1 The Original System

6.2 A First Coupled Map System Version

6.3 A Second Coupled Map System Version

Figure 6.1: γ = 0.3, N = 64, 5000 iterations, random uniform initialization