Introduction to

Dynamical Systems
A HANDS-ON APPROACH WITH MAXIMA

ISBN 972-99396-0-8

Jaime E. Villate
9 789729 939600

Introduction to Dynamical Systems

A Hands-on Approach with Maxima

Jaime E. Villate
University of Porto
College of Engineering
Porto, Portugal

Introduction to dynamical systems: a hands-on approach with Maxima

c 2006 Jaime E. Villate
Copyright
E-mail: [email protected]

This work is licensed under the Creative Commons Attribution-Share Alike 2.5 License. To view a copy of
this license, visit http://creativecommons.org/licenses/by-sa/2.5/ or send a letter to Creative
Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.
ISBN: 972-99396-0-8
This is a partial translation of Portuguese version 1.2 of February 27, 2007.

The cover figure is the Julia set for the complex number 0.75 + i 0.1, with 48 iterations, as explained in
chapter 12.

Contents
Preface
1

vii

Introduction

1.1

Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Solving physics problems with Maxima . . . . . . . . . . . . . . . . . . . . . .

1.3

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.4

Multiple-choice questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.5

Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Discrete dynamical systems

2.1

Discrete systems evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

2.2

Graphical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

2.3

Fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

2.4

Periodic points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

2.5

Solving equations numerically . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

2.5.1

Iteration method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

2.5.2

Newtons method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.6

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.7

Multiple-choice questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

2.8

Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Index

26

Bibliography

27

iv

CONTENTS

List of Figures
1.1

Power dissipated in a resistor . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Trajectory of a particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.1

Evolution of yn+1 = cos(yn ) with y0 = 2. . . . . . . . . . . . . . . . . . . . . . .

13

2.2

Staircase diagram for xn+1 = cos(xn ) with x0 = 2. . . . . . . . . . . . . . . . . .

13

2.3

Solutions of the system yn+1 = y2n 0.2. . . . . . . . . . . . . . . . . . . . . . .

14

2.4

Solutions of the discrete logistic model . . . . . . . . . . . . . . . . . . . . . . .

15

2.5

Newtons method for finding roots of an equation. . . . . . . . . . . . . . . . . .

22

vi

LIST OF FIGURES

Preface
This book is updated very often. The number of the current version can be found on the second
page and the most recent version can always be found on the Web at http://fisica.fe.up.pt/
maxima/dynamicalsystems. This version has been written to be used with Maximas version
5.11. (http://maxima.sourceforge.net).
This book started as the lecture notes for a one-semester course on the physics of dynamical
systems, taught at the College of Engineering of the University of Porto, since 2003. The subject
of this course on dynamical systems is at the borderline of physics, mathematics and computing,
and it substituted a course on classical mechanics that we used to teach to students majoring in
computing engineering.
Since dynamical systems is usually not taught with the traditional axiomatic method used in other
physics and mathematics courses, but rather with an empiric approach, it is more appropriate to
use a practical teaching method based on projects done with a computer.
The study of dynamical systems advanced very quickly in the decades of 1960 and 1970, giving
rise to a whole new area of research with an innovative methodology that gave rise to heated debates within the scientific community. The innovative boost was fueled by the rapid development
of computers.
A new generation of researchers rose, who used their computers as laboratories for experimenting
with equations discovering new phenomena. The traditional mathematicians criticized that approach for the lack of a rigorous mathematical foundations for those new results. Many of those
results were found within the framework of physical problems: non-linear dynamics, condensedmatter physics and electromagnetism. However, many physicists regarded that new research field
simply as a computer simulation of old physical concepts long established, without any new
physics in it. A usual comment would be: this is all very nice, but where is the physics in
it?.
Thus, the pioneers in this new field of dynamical systems would be often confronted with rejected
publications in scientific journals and negative assessments of their work. On the other hand, their
activities awakened a strong interest that increased very quickly and were viewed by some as a
refreshing new trend; the methods used in the study of dynamical systems match well with the
working environment of modern scientists.
The new paradigm spread to teaching and the traditional courses on physics and mathematics
have been gradually infected with this new experimental/ computational methodology, contrasting with the traditional axiomatic method. As it happened in the scientific community, the new
methodology has also led to some debate among teachers; at the same time, it has awakened big
interest as a better way to motivate todays students. Subjects such as chaos and fractals are very

viii

Preface

appealing to them.
In this book we intend to explore some topics on dynamical systems, using an active teaching
approach, supported by computing tools and trying to avoid too may abstract details. The use of a
Computer Algebra System (CAS) does not eliminate the need for mathematical analysis from the
student; using a CAS to teach an engineering course does not turn it into a purely technical subject
either. One of the difficulties inherent to any Computer Algebra System is the fact that there are
no unique solutions to the problems it solves. Different methods to solve a problem may lead to
solutions that look very different but might be equivalent. Or the solutions can be really different
and only equivalent within some domain. In some cases the system does not give any answer or it
might even give a wrong answer.
It is necessary to gain some experience to be able to use CAS tools successfully and to be able to
test the validity of its results. In the process of gaining that experience, the user will also gain a
better insight into the mathematical methods implemented in the system.
Nowadays the great majority of engineering and exact sciences professionals depend on a calculator to calculate the square root of a real number, for instance, 3456. Some of us were taught in
School how to do that with pencil an paper, in a time when there were no calculators. I do not
believe that this new dependency is a serious handicap, and Im not in favor of teaching kids how
to calculate square roots with paper and pencil before they are allowed to use the calculator. What
I find very important is that the algorithm we used to calculate square roots with paper and pencil
remains available and well documented in the literature; it is an valuable piece in our legacy of
algorithms.
On the other hand, now that students have calculators to compute square roots, they can move
faster into other topics such as the study of quadratic equations; and in doing so, they might even

gain a deeper insight of the function x, which they did not attain when they had to spend a
lot of time learning the algorithm to calculate square roots. In the case of differential equations
and difference equations, with the help of Computer Algebra Systems students can advance faster
into subjects such as chaos and fractals, instead of dedicating a whole semester to learn several
algorithms to obtain analytical solutions for a few types of equations.
I would like to acknowledge the help of my colleagues Helena Braga and Francisco Salzedas, with
whom I have taught the course on Physics of Dynamical Systems; I would also like to thank our
students in that course throughout the last 3 years; their positive comments have encouraged me
to undertake the task of writing this book. The students have been asked to make projects for
that course, and some of those projects were very interesting and helped me learn some of the
subjects covered in this book. Special thanks go to the student Pedro Martins and to my colleague
Francisco Salzedas, who made a careful review of the manuscript.

Jaime E. Villate
Porto, February 2007

Chapter 1

Introduction
1.1

Differential equations

Differential equations play a very important role in Engineering and Science. Many problems lead
to one or several differential equations that must be solved. Most attention has been given to linear
equations in the literature; several analytical methods have been developed to solve that type of
equations.
Non-linear differential equations are much harder to analyze and there are no general solution
techniques for those equations. Problems that lead to linear equations are easier to study. From
the last half of the 20th century, the rapid development of the computer made it possible to solve
non-linear problems using numerical methods. Non-linear systems lead to a wealth of new and
interesting phenomena not present in linear systems.
A new approach, that relies more on geometric interpretation rather than analytical analysis, has
gained popularity for the study of non-linear systems. Many of the concepts used in that geometrical approach, such as the phase space, have long be used in dynamics to study the motion of a
mechanical system.
In order to give a short introduction to that methodology to study differential equations, in the
next chapters we will consider several problems specific dynamics and electrical circuit theory.
Before we begin, we will introduce a Computer Algebra System (CAS), Maxima, which will be
used extensively throughout the book.

1.2

Solving physics problems with Maxima

Maxima is a software package in the category of Computer Algebra Systems (CAS), namely, a
system that can be used not just for numerical calculation but also to deal with algebraic equations with abstract variables. There are various CAS packages available; we have decided to use
Maxima because it is Free Software. That means that it can be installed and used by our students
without having to obtain a license for it, and they can even study its source code to get a better
understanding of how that system works. Another important advantage is the possibility of modifying the original package to better suit our needs; we took advantage of that facility to add new

Introduction

features needed for this book.

Maxima includes several functions to manipulate mathematical functions, including differentiation, integration, power series approximation, Laplace transforms, solving ordinary differential
equations and graph plotting in 2 and 3 dimensions. It can also work with matrices and vectors. Maxima can be used to solve problems numerically and write down programs as done with
traditional programming languages.
The following examples should serve to give a first glimpse at the way Maxima can be used. In
the next chapters we will go deeper into the subject, but readers who are not familiar with Maxima
and would like to have a general overview from the beginning can start by going through appendix
A. The examples that we will solve in this section are in the area of dynamics of a particle and
direct-current circuits, which are the main subjects in this book. A minimum knowledge in those
two subjects will be necessary in order to follow those examples.
Example 1.1
A battery is connected to an external resistor with resistance R and the voltage across the resistor
is measured with a voltmeter V. To find the electro-motive force and the internal resistance r of
the battery, two external resistors of 1.13 k and 17.4 k were used. The voltage drop in both
cases were 6.26 V e 6.28 V. Find the intensity of the current in both cases. Obtain the values of
and r. Plot a graph of the power dissipated in the external resistance, as a function of R, for values
of R within 0 and 5r.

, r

Solution: The current through R is found from Ohms law:

I=

V
R

(1.1)

With the values given for the potential difference, V , and the resistance, R, we can use Maxima
to find the currents:
(%i1) 6.26/1.13e3;
(%o1)

.005539823008849558

When Maximas console is started, the (%i1) label appears indicating that the system is ready
to accept a command; the letter i stands for input. The expression 1.13e3 is the form used to
represent the number 1.13 103 in Maxima. Each command must end with a semi-colon. When
the Enter key is pressed, the system responds with a label (%o1) followed by the result of the
first command (%i1); o stands for output.
The current in the second case is computed in a similar way:

1.2 Solving physics problems with Maxima

(%i2) 6.28/17.4e3;
(%o2)

3.609195402298851E-4

Thus, the current in the 1.13 k resistor is 5.54 mA, and in the 17.4 k resistor is 0.361 mA.
To obtain the batterys electro-motive force and internal resistance we should use the voltagecurrent characteristic for a battery:
(1.2)
V = rI
replacing the two set of values given for V and R we will get a system of two equations with two
variables. We will save those two equations in two Maxima variables that we will dub as eq1 and
eq2
(%i3) eq1: 6.26 = emf - r*%o1;
(%o3)
6.26 = emf - .005539823008849558 r
(%i4) eq2: 6.28 = emf - r*%o2;
(%o4)
6.28 = emf - 3.609195402298851E-4 r
notice that the symbol used to save a value in a variable is a colon and not an equal sign. A maxima
variable can store a numerical value or something more abstract as a mathematical equation in this
case. The equal sign makes part of the equation that is being stored. To avoid having to write the
numerical values of the currents obtained previously, we used the symbols %o1 and %o2 that stand
for the value of those previous results.
The last two equations constitute a linear system of equations with two variables. That kind of
system can be solved in Maxima, using the command solve:
(%i5) solve([eq1,eq2]);
(%o5)

983100
79952407
[[r = ------, emf = --------]]
254569
12728450

(%i6) %,numer;
(%o6)
[[r = 3.861821352953423, emf = 6.281393806787158]]
The syntax [eq1,eq2] was used to create a list with two elements, which is what the command
solve expects when there are more than one equation to be solved. Some warning messages
given by Maxima were omitted above. The command solve gives an exact result, in the form
of two rational numbers. The command in %i6 was used to approximate those rational numbers
with fixed-point numbers. The symbol % stands for the output of the last command executed; in
this case it is equivalent to %o5. We thus conclude that the electro-motive force is approximately
6.2814 V and the internal resistance is 3.8618 .
The electric power dissipated in the resistance R is
P = RI 2
the current I across the external resistor can be calculated in terms of the electro-motive force and
the resistances r and R

I=
R+r

Introduction

therefore, the power dissipated in the external resistor is



P=R
R+r

2

to draw the plot of P, as a function of R, we can use the following command

(%i7) plot2d(R*(6.2814/(R+3.8618))2, [R, 0, 5*3.8618]);
the result is shown in figure 1.1. Moving the cursor in the graphic window, it is possible to read the
coordinates of the point where the cursor is. We can check that the power dissipated in an external
resistor reaches its maximum value when the external resistance equals the internal one.

Figure 1.1: Power dissipated in the external resistor as a function of the external resistance.

Two frequent errors should be avoided:

An expression such as
a = 3;
will not assign any value to the variable a. After that expression is written, variable a remains
as an undefined variable. To assign a value of 3 to the variable a, the following command
must be used:
a : 3;
New users of Maxima should become aware of the distinction between equations and expressions. An example of an equation is

1.2 Solving physics problems with Maxima

x2 - 3*x = 2*x + 5
while an expression is something like
2*x + 5

Some commands in Maxima accept only equations or expressions as their input values. For
instance, the command plot2d used in the previous example accepts only expressions and
not equations. The command solve requires one equation, or a list with several equations,
but it will also accept expressions instead of equations: each expression given will be automatically converted into an equation by equating it to zero; for instance, the command
solve(x2 - 5*x + 5);
will find the two values of x that will solve the equation
x2 5 x + 5 = 0

Example 1.2
The position vector of a particle, as a function of time t, is given by the equation:




~r = 5 t 2 et/5 ~ex + 3 et/12 ~ey
in MKS units. Find the vectors for the position, velocity and acceleration at t = 0, t = 15 s, and
when time approaches infinity. Plot the trajectory of the particle during the first 60 seconds of its
motion.
Solution: We start by representing the position vector as a list with two elements; the first element
will be the x coordinate and the second one will be the y coordinate. We will save that list in a
variable named r, so we can use it later on.
(%i8) r: [5-t2*exp(-t/5),3-exp(-t/12)];
2
- t/5
- t/12
(%o8)
[5 - t %e
, 3 - %e
]
the vector velocity equals the derivative of the position vector and the vector acceleration is the
derivative of the vector velocity. The command used in Maxima to find the derivative of an expression is diff. The input to that command can also be a list with several expressions; thus, the
velocity and acceleration are:
(%i9) v: diff(r,t);

(%o9)
(%i10) a: diff(v,t);

2
- t/5
- t/12
t %e
- t/5 %e
[---------- - 2 t %e
, --------]
5
12

Introduction

(%o10)

2
- t/5
- t/5
- t/12
t %e
4 t %e
- t/5
%e
[- ---------- + ----------- - 2 %e
, - --------]
25
5
144

the constant %e in Maxima represents the Euler number, e. To find the position, velocity and
acceleration at t = 0, we use the following commands
(%i11) r, t=0, numer;
(%o11)
(%i12) v, t=0, numer;
(%o12)
(%i13) a, t=0, numer;
(%o13)

[5, 2]
[0, .08333333333333333]
[- 2, - .006944444444444444]

The argument numer, was used to get the result in floating-point form. In vector notation, the
results we obtained are:
~r(0) = 5~ex + 2~ey
~v(0) = 0.08333~ey
~a(0) = 2~ex 0.006944~ey
For t = 15 s the results are obtained in a similar way
(%i14) r, t=15, numer;
(%o14)
[- 6.202090382769388, 2.71349520313981]
(%i15) v, t=15, numer;
(%o15)
[.7468060255179592, .02387539973834917]
(%i16) a, t=15, numer;
(%o16)
[0.0497870683678639, - .001989616644862431]
The limiting values when times goes to infinity can be calculated with Maximas command limit;
the symbol used in Maxima to represent infinity is inf
(%i17) limit(r,t,inf);
(%o17)
(%i18) limit(v,t,inf);
(%o18)
(%i19) limit(a,t,inf);
(%o19)

[5, 3]
[0, 0]
[0, 0]

Thus, a particle will approach the point 5~ex + 3~ey , where it will remain at rest.
To plot the graph of the trajectory we will use the option parametric of the command plot2d.
the x and y components of the position vector will be given separately; the command plot2d will
not accept them inside a list as we have been using them. To get the first element of the list r (x
component) is labelled as r[1] and the second element r[2].

1.3 References
(%i20) plot2d([parametric,r[1],r[2],[t,0,60],[nticks,100]]);

The time domain, from 0 to 60, is defined with the notation [t,0,60]. The option nticks was
used to increase the number of intervals of t, because the default value of 10 intervals would render
a broken curve instead of a continuous trajectory. The graph obtained is shown in figure 1.2.

Figure 1.2: Trajectory of the particle during the first 60 seconds, from the initial instant when t
was equal to 5.2 .

1.3

References

For more information about Maxima, see appendix A and the Maxima Book (de Souza et al.,
2003).

1.4

Multiple-choice questions

1. Only one of the following Maxima com- 2. Newtons second law was defined in Maxima
mands is correct. Which one?
with:
(%i6) F = ma;
A. solve(t-6=0,u-2=0,[t,u]);
which Maxima command should be used to
B. solve(t+4=0,u-4=0,t,u);
compute the value of the force corresponding to a mass of 7 with an acceleration of 5
C. solve([x3+4=2,y-4],x,y);
(SI units).
D. solve(x-6=0,y-2=0,[x,y]);
E. solve([t+3,u-4],[t,u]);

A. solve(F, m=7, a=5);

Introduction
B.
C.
D.
E.

solve(F, [m=7, a=5]);

solve(%o6, m=7, a=5);
%o6, m=7, a=5;
solve(F: m=7, a=5)

(%i2) x=5$
(%i3) x;
which will be the output (%o3)?
A. 5
B. x
C. 3

3. If we input the following commands in Maxima:

D. true
E. 0

(%i1) x:3$

1.5

Problems

1. An ammeter was used to measure the current at points D and F in the circuit which diagram
is shown in the figure. At point D the value obtained for the current was 0.944 mA, in the
direction ADC, and at point F it was 0.438 mA, in the direction CFE. (a) Store the equation
for Ohms law, V = IR, in a Maximas variable ohm. (b) Give a value to variable I equal to
the current at point D, and substitute the resistance in Ohms law with each of the values 2.2
k and 6.8 k, to compute the potential difference in each of the resistors; repeat the same
procedure to calculate the potential difference in each resistor.
2.2 k

4.7 k

6V
3V

9V
1.0 k

6.8 k

3.3 k

2. The position of a particle moving along the x axis is given by the equation x = 2.5t 3 62t 2 +
10.3t, where x is measured in meters and t in seconds. (a) Find the expressions for the velocity
and the acceleration as a function of time. (b) Find the values of the time, position and acceleration in all the instants when the particle is at rest (v = 0). (c) Draw the plots for the position,
velocity and acceleration as a function of time, for t in the interval between 0 and 20 seconds.
3. The position vector of a particle, as a function of time, is given by the equation:


~r = 5.76 et/2.51 ~ex + et/2.51 cos(3.4t)~ey
in SI units. (a) Compute the position, velocity and acceleration in the instants t=0, t=8 s, and
when time goes to infinity. (b) Plot the graphs of the x and y components of the position, as
a function of time, for t in the interval between 0 and 15 seconds. (c) Plot the graph of the
trajectory, on the plane xy, in the interval of t between 0 and 15 s.

Chapter 2

Discrete dynamical systems

A discrete dynamical system is a system with a state that only changes at a discrete sequence of
instants {t0 , t1 , t2 , . . .}. In the interval between two of those instants the state remains constant.
In this chapter we will only consider first-order discrete systems. In the following chapters the
methods used in this chapter will be extended to the case of continuous dynamical systems. In the
last two chapters of the book we will resume the study of discrete systems and we will introduce
second-order systems and systems on the complex plane.
A first-order system is a system in which only one variable y is needed to describe its state. The
value of that variable at the instants {t0 , t1 , t2 , . . .} will be a sequence {y0 , y1 , y2 , . . .}. The interval
of time between a consecutive pair of instants tn and tn+1 does not have to be constant, for different
values of n.
The evolution equation allows us to compute the state yn+1 , at an instant tn+1 , from the value of
the state yn at the previous instant tn :
yn+1 = F(yn )
(2.1)
where F(y) is a known function. The equation above is a first-order difference equation. Given
an initial state y0 , successive applications of the function F will generate the sequence of states yn
which determine the evolution of the system. In some cases it might be possible to find a general
expression for yn as a function of n.
Example 2.1
Find the first four terms in the evolution of the system xn+1 = cos xn , with initial state x0 = 2
Solution: Applying the difference equation three times, we obtain the first four terms in the sequence:
{2, cos(2), cos(cos(2)), cos(cos(cos(2)))}

(2.2)

10

Discrete dynamical systems

Example 2.2
A loan of $ 500 is obtained from a bank, which charges a 5% yearly interest rate. The loan is to
be paid in 20 months, with monthly payments of $ 26.11. What will be the amount in debt after
10 months?
Solution: During the month n the amount in debt, yn , will be equal to the amount in debt in the
previous month, yn1 , plus the interests due for that month, minus the monthly payment p:
yn = yn1 + j yn1 p

(2.3)

where j is the monthly interest rate (in this case, 0.05/12). Using Maxima, the sequence of amounts
in debt yn can be obtained by applying the above recurrence relation several times:
(%i1) j: 0.05/12$
(%i2) y: 500$
(%i3) y: y + j*y - 26.11;
(%o3)
(%i4) y: y + j*y - 26.11;

475.9733333333333

(%o4)
(%i5) y: y + j*y - 26.11;

451.8465555555555

(%o5)

427.619249537037

it would be necessary to repeat the command (%i3) ten times. The answer can be obtained in an
easier way if we define a Maxima function depending on an integer variable, using the recurrence
relation, and we use that function to calculate y10 directly:
(%i6) y[0]: 500$
(%i7) y[n] := y[n-1] + j*y[n-1] - 26.11;
(%o7)

y := y
+ j y
- 26.11
n
n - 1
n - 1

(%i8) y[10];
(%o8)

255.1779109580579

Some care should be taken in Maxima when using functions of an integer argument. In the previous example, when we calculated y[10], the values of y[9], y[8],. . ., y[1], were also implicitly
calculated and stored in memory. If we changed the recurrence relation, those values that were

11
already calculated and stored would not be updated. Thus, before we change the recurrence relation, or the initial value y[0], it is necessary to erase the previously calculated sequence, by using
the command kill.
For example, if the value of the loan was duplicated to $ 1000, and the monthly payment was also
duplicated, will the amount in debt after the tenth payment would also duplicate? let us see:
(%i9) kill(y)$
(%i10) y[0]: 1000$
(%i11) y[n] := y[n-1] + j*y[n-1] - 52.22;
(%o11)

y := y
+ j y
- 52.22
n
n - 1
n - 1

(%i12) y[10];
(%o12)

510.3558219161157

thus, the amount in debt also doubles.

Another question that we might ask in the original example is: what will the monthly payment
should be if instead of 20 months we would want to pay the loan in 40 months?
To answer that question, we use a variable p to represent the monthly payment, we calculate the
amount in debt after 40 months, as a function of p, and we equal that expression to zero to calculate
the value of p.
(%i13) kill(y)$
(%i14) y[0]: 500$
(%i15) y[n] := expand(y[n-1] + j*y[n-1] - p)$
(%i16) solve(y[40] = 0, p);
(%o16)

72970398
[p = --------]
5366831

(%i17) %, numer;
(%o17)

[p = 13.59655222979818]

The monthly payment should be $ 13.60. The function expand was used to force Maxima to
calculate the products in every step, avoiding large expressions with several parenthesis in the
calculation of yn . Some additional messages written by Maxima were omitted, which explain that
some floating point numbers were represented as fractions, to prevent numerical errors.

12

Discrete dynamical systems

2.1

Discrete systems evolution

The evolution of a first-order discrete system:

yn+1 = F(yn )

(2.4)

is obtained by applying successively a function F to the initial state y0 = c:

{c, F(c), F(F(c)), F(F(F(c))), . . .}

(2.5)

{c, F(c), F 2 (c), F 3 (c), . . . yn = F n (c)}

(2.6)

or in a more compact form:

2.2

Graphical analysis

A graphical method to represent the evolution of a system consists on plotting a point for each step
in the sequence, with x-coordinate equal to the index n and y-coordinate equal to yn . In Maxima,
the function evolution in the additional package dynamics will plot that kind of graph1 .
Three arguments should be given to the program. The first argument must be an expression that
depends only on the variable y; that expression will specify F(y) from the right hand side of the
difference equation 2.1. The second argument should be the initial value y0 , and the third argument
is the number of sequence elements that should be plotted.
For instance, using y to identify the state variable in example 2.1, we have F(y) = cos y with initial
value y0 = 2. To plot a graph with the first 20 terms, we use the commands:
(%i18) load("dynamics")$
(%i19) evolution(cos(y), 2, 20)$
The graph obtained in (%i19) is shown in figure 2.1.
Another type of diagram which will be very useful to analyze first-order discrete dynamical systems is the so-called staircase diagram,2 which consists in plotting the functions y = F(x) and
y = x, as well as an alternating sequence of horizontal and vertical segments joining the points
(y0 ,y0 ), (y0 ,y1 ), (y1 , y1 ), (y1 ,y2 ), and so on. For example, figure 2.2 shows the staircase diagram for
the sequence represented in figure 2.1
The function staircase, included in the additional package dynamics, plots staircase diagrams.
That function needs the same three arguments as the function evolution; namely, function F(y)
from the right-hand side of the difference equation 2.1, the initial value y0 and the number of steps
in the sequence. The independent variable in the expression for F should always be y. You might
1 Maximas package dynamics was added in version 5.10;

to use that package.

2 Also know as cobweb diagram.

if you have an older version, you must upgrade it in order

13

2.2 Graphical analysis

yn
2

1
n
5

0.5

10

15

Figure 2.1: Evolution of yn+1 = cos(yn ) with y0 = 2.

yn+1
2

1
yn
1

1
1

Figure 2.2: Staircase diagram for xn+1 = cos(xn ) with x0 = 2.

14

Discrete dynamical systems

need to make the appropriate change if the state variable is something different from y in your
problem.
The graph 2.2 was obtained with the command
(%i20) staircase(cos(y),2,8)$
Notice that we did not have to load the package dynamics again, because it was already loaded in
(%i18). the staircase diagram allows us to understand when a sequence will converge or diverge.
For instance, consider the system yn+1 = y2n 0.2. If the initial value is y0 = 1.1, we obtain
the graph on the left side of figure 2.3; we see that the sequence will converge to a negative
value
y, which is at the intersection of the functions F(y) = y2 0.2 and G(y) = y, namely, y =
(5 3 5)/10.

The two functions intersect in another point, positive, equal to y = (5 3 5)/10. In the graph we
can see that even though the initial value was close to the second intersection point, the sequence
moved away from it and towards the second intersection point, due to the fact that between the
two intersection points the function y2 0.2 is under G(y) = y. if we used an initial value to the
right of the second intersection point, for instance, y0 = 1.5, the sequence grows quickly towards
infinity (right-hand side in figure 2.3). To make the sequences converge to the second intersection
point, it would have been necessary that between the two intersection points F(x) > G(y); that is
to say, the slope of F(y) should be less than 1, rather that greater than 1, at the second intersection
point.
yn+1

yn+1
4

2
yn
1

yn
2

Figure 2.3: Solution to the system yn+1 = y2n 0.2 with initial values 1.1 (left) and 1.5 (right).

Example 2.3
Analyze the solutions to the logistic model, which consists in considering a population P with
constant natality rate, a, and a mortality rate bP, directly proportional to the population, where a
and b are constants.
Solution: The population under study could be a group of specimens from some animal species,
where the sequence {P0 , P1 , P2 , . . .} represents the number of specimens in several consecutive
years.

15

2.3 Fixed points

Let Pn represent the number of specimens at the beginning of period n. during that period of time,
an average aPn new specimens are born, and bPn2 specimens die. Thus, in the beginning of the next
period, n + 1, the population would be

Pn+1 = (a + 1) Pn

b
Pn
1
a+1


(2.7)

It is convenient to define a new variable yn = b Pn /(a + 1). Thus, we obtain an equation with a
single parameter c = a + 1
(2.8)
yn+1 = c yn (1 yn )
Figure 2.4 shows the solutions obtained with an initial value y0 = 0.1, in the cases c = 2 and c = 4.
With c = 2, the solution converges quickly to the fixed point y = 0.5.
With c = 4, the state of the system goes through many different values between 0 and 1, and
it does not seem to follow any regular pattern. That type of behavior is called chaotic. The
state in any given period is perfectly determined from the state in the previous period, but a very
small modification of the state in an initial period will lead to a completely different state in the
next period. The state of the system takes different values within a small interval, without ever
repeating a previous value.
yn+1

yn+1

0.5

0.5

yn
0.5

yn

0.5

Figure 2.4: Solutions of the logistic model with an initial value 0.1. With c = 2 (left) the
sequence converges, while with c = 4 (right) it becomes chaotic.

2.3

Fixed points

A fixed point of the system 2.1 is a value y0 which, if used as initial value of the state it would
lead to a constant state in later periods. That will only be possible if
F(y0 ) = y0

(2.9)

this implies that successive application of function F will not change the initial value. The solution
of the system with initial value y0 is a constant sequence, {y0 , y0 , y0 , . . .}.
From the point of view of the graphical analysis, the fixed points are all those points where the
curve F(x) intersects the right line y = x in the stairway diagram. For example, in the case of he

16

Discrete dynamical systems

logistic model, figure 2.4 shows that in the two cases c = 2 and c = 4 there are two fixed points,
one of them at y = 0. We can use Maximas command solve to find the fixed points; in the case
c = 4, the fixed points can be found in this way:
(%i21) flogistic: 4*y*(1-y);
(%o21)
4 (1 - y) y
(%i22) fixed: solve(flogistic - y);

(%o22)

3
[y = -, y = 0]
4

The two fixed points are 0 and 0.75.

Let us consider a fixed point, where the curve F(x) intersects the straight line y = x, such that the
derivative F 0 (y) is bigger than 1 at that point. Namely, at the intersection point between the curve
F(x) and the line y = x the curve F crosses from under the line, on the left, to over the line, on the
right. Thus, if we plot the staircase diagram starting from a point near the fixed point, the sequence
will move away from the fixed point, describing a staircase in the staircase diagram. We call that
kind of fixed point a repulsive node.
If the derivative is negative and less than -1, the sequences will also move away from the fixed
point, but alternating from side to side, describing a cob web in the staircase diagram. We call
that kind of fixed point a repulsive focus.
If the derivative of the function F near the fixed point has a value between 0 and 1, the sequences
that start near the fixed point will approach it describing a staircase. That kind of fixed point is
called an attractive node (an example of this case was already found in the left-hand side of 2.4).
If the derivative of the function F near the fixed point has a value between 0 and -1, the sequences
starting near the point will approach it, describing a cob web that alternates from side to side in
the staircase diagram. That kind of point is called an attractive focus (an example of that kind of
point was already encountered in figure 2.2).
In summary, we have the following kinds of fixed points y0 :
1. Attractive node, if 0 F 0 (y0 ) < 1
2. Repulsive node, if F 0 (y0 ) > 1
3. Attractive focus, if 1 < F 0 (y0 ) < 0
4. Repulsive focus, if F 0 (y0 ) < 1
If F 0 (y0 ) equals 1 or -1, the situation is more complex: the fixed point could be either attractive or
repulsive, or even attractive from one side and repulsive from the other side.
Returning to our previous example of the logistic model (see (%i21) through (%o22) above), the
value of the derivative of F at the fixed points is:

17

2.4 Periodic points

(%i23) dflogistic: diff(flogistic, y);
(%o23)
(%i24) dflogistic, fixed[1];

4 (1 - y) - 4 y

(%o24)
(%i25) dflogistic, fixed[2];

- 2

(%o25)

Thus, in the case c = 4, both fixed points are repulsive. At y0 = 0 there is a repulsive node, and
there is a repulsive focus at y0 = 0.75.

2.4

Periodic points

If the sequence {y0 , y1 , y2 , . . .} is a solution of the dynamical system

yn+1 = F(yn )

(2.10)

any element in the sequence can be obtained from y0 , applying the composed function F n
yn = F n (y0 ) = F(F(. . . (F (y))))
| {z }
n times

(2.11)

A solution is dubbed a cycle of period 2, if it is a sequence of only two alternating values:

{y0 , y1 , y0 , y1 , . . .}. The two points y0 and y1 are periodic points with period equal to 2. Since
y2 = F 2 (y0 ) = y0 , it is necessary that F 2 (y0 ) = y0 . Furthermore, since y3 = F 2 (y1 ) = y1 we also
have F 2 (y1 ) = y1 . Finally, since F(y0 ) = y1 6= y0 , it is also necessary that F(y0 ) 6= y0 , and since
F(y1 ) = y0 6= y1 , we also have F(y1 ) 6= y1 .
Those conditions can be summarized by saying that two points y0 and y1 form a cycle of period
two if both of them are fixed points of the function F 2 (y), but none of them is a fixed point of the
function F(y). Explained in a different way, if we calculate the fixed points of F 2 (y), all of the
fixed points will appear, plus the periodic points of period 2 of functio F.
The cycle of period two will be attractive or repulsive according to the value of the derivative of
F 2 at each point in the cycle.
To calculate the derivative of F 2 at y0 , we use the chain rule
(F 2 (y0 ))0 = (F(F(y0 )))0 = F 0 (F(y0 ))F 0 (y0 ) = F 0 (y0 )F 0 (y1 )

(2.12)

thus, the derivative of F 2 takes the same value in the two points y0 and y1 of the cycle, and it is
equal to the product of the derivatives of F in those two points.
Generalizing the definition a point y0 is part of a cycle of period m, if F m (y0 ) = y0 , but F j (y0 ) 6= y0
for any j < m. The complete set of m points that make part of the cycle are
y0

18

Discrete dynamical systems

y1 = F(y0 )
y2 = F 2 (y0 )
..
.
ym1 = F m1 (y0 )

All those points are fixed points of F m but they cannot be fixed points od F j , with j < m.
If the absolute value of the product of the derivative at the m points of the cycle:
m1

F 0 (y j )

(2.13)

j=0

is greater than 1, then the cycle is repulsive; if the product is less than 1, the cycle is attractive, and
if the product is identical to 1, the cycle could be either attractive or repulsive in different regions.
Example 2.4
Find the cycles with period 2 for the logistic system
yn+1 = 3.1 yn (1 yn )
and say whether they are attractive or repulsive.
Solution: We start by defining function F(y) and the composed function F 2 (y)
(%i26) flog: 3.1*y*(1-y)$
(%i27) flog2: flog, y=flog;
(%o27)

9.610000000000001 (1 - y) y (1 - 3.1 (1 - y) y)

The periodic points of period 2 will be among the solutions of the equation F 2 (y) y = 0
(%i28) periodic: solve(flog2 - y);

(%o28)

sqrt(41) - 41
sqrt(41) + 41
21
[y = - -------------, y = -------------, y = --, y = 0]
62
62
31

The last two points, namely 0 and 21/31 are fixed points (the proof of that is left as an exercise for
the reader and can be done by solving the equation F(y) = y, or simply by showing the validity of
that equation in each case).
The other two points must then form a cycle of period two; if we use any of them as initial value,
the sequence will oscillate between those two points.
To find out whether the cycle is attractive or repulsive, we compute the product of the derivative
at the two points in the cycle

19

2.5 Solving equations numerically

(%i29) dflog: diff(flog, y);
(%o29)

3.1 (1 - y) - 3.1 y

(%i30) dflog, periodic[1], ratsimp, numer;

(%o30)

- .3596875787352

(%i31) dflog, periodic[2], ratsimp, numer;

(%o31)

- 1.640312421264802

(%i32) %o30*%o31;
(%o32)

.5900000031740065

The absolute value of the product is less than 1, which implies that the cycle is attractive.

2.5

Solving equations numerically

Discrete, first-oder dynamical systems can be used for solving one-variable equations numerically.
The problem to be solved consists on finding the roots of a real function f , namely, the values of
x that satisfy the equation
f (x) = 0

(2.14)

For example, suppose we want to find the values of x that solve the equation:
3x2 x cos(5x) = 6
That kind of equation cannot be solved analytically; it must be solved by numerical methods. The
numerical methods to solve that equation consist on defining a dynamical system with convergent
sequences which approach the solutions of the equation. In the following sections we will study
two of those methods.

2.5.1

Iteration method

If the equation 2.14 can be written in the form

x = g(x)

(2.15)

Its solutions are the fixed points of the dynamical system:

xn+1 = g(xn )

(2.16)

20

Discrete dynamical systems

To find a fixed point, we choose an arbitrary initial point and calculate the evolution of the system.
Example 2.5
Find the solution of the equation x = cos x
Solution: This equation is already given in a form that allows us to use the iteration method. We
use the dynamical system with recurrence relation
xn+1 = cos(xn )
To find a fixed point, we choose an arbitrary initial point and calculate the evolution of the system
(%i33) x: 1$
(%i34) for i thru 15 do (x: float(cos(x)), print(x))$
0.54030230586814
0.85755321584639
0.65428979049778
0.79348035874257
0.70136877362276
0.76395968290065
0.72210242502671
0.75041776176376
0.73140404242251
0.74423735490056
0.73560474043635
0.74142508661011
0.73750689051324
0.74014733556788
0.73836920412232
The solution of the equation is approximately 0.74. This method was successful in this example,
because the fixed point of the dynamical system chosen happened to be attractive. If the point
were repulsive, the iteration method would have failed.

Example 2.6
Find the square root of 5, using only additions, multiplications and divisions.
Solution: The square root of 5 is the positive solution of the equation
x2 = 5

21

2.5 Solving equations numerically

which can be written as:
x=

5
x

we solve the dynamical system associated to the function

f (x) =

5
x

It can be easily seen that for any initial value x0 , different from
will always be a cycle with period 2:

5, the solution of that system



5
5
x0 , , x0 , , . . .
x0
x0

To escape from that cycle, and approach the fixed point at 5, we can try to use the middle point:


1
5
xn+1 =
xn +
2
xn
That new system will converge quickly to the fixed point at

5:

(%i35) x : 1$
(%i36) for i thru 7 do (x: float((x + 5/x)/2), print(x))$
3.0
2.333333333333334
2.238095238095238
2.236068895643363
2.236067977499978
2.23606797749979
2.23606797749979

2.5.2

Newtons method

Newtons method can be used to find the roots of the equation 2.14. We start by assuming that
there is a root of the equation at approximately value x0 and we improve that approximation by
finding the point x1 where the tangent to the function at f (x0 ) intersects the x axis (see figure 2.5)
x1 = x0

f (x0 )
f 0 (x0 )

We can use the same equation to further improve our guess x1 to a new guess x2 . In general

(2.17)

22

Discrete dynamical systems

xn+1 = xn

f (xn )
f 0 (xn )

(2.18)

f(x0)

x
x1

x0

Figure 2.5: Newtons method for finding roots of an equation.

It must be noticed that the roots of a continuous function f , points where f is zero, are fixed points
of the dynamical system defined by equation 2.18 3 .
The advantage of this method, over the iteration method, can be seen by using our analysis of the
fixed points of a dynamical system. The function that generates the system 2.18 is
g(x) = 1

f (x)
f 0 (x)

(2.19)

the derivative of that function is

g0 = 1

f 00 f
( f 0 )2 f 00 f
=
( f 0 )2
( f 0 )2

(2.20)

at the fixed points, f vanishes. Thus, g0 will also vanish at the fixed points. Therefore, the fixed
points of 2.18 will always be attractive. It means that if the initial point x0 is chosen close enough
to one of the roots of f , the sequence xn will approach it. The problem consists on finding what
close enough means in each case.
To illustrate the method, we will solve example 2.6 once again, using Newtons method.
The square root of 5 is one of the solutions of the equation x2 = 5. Hence, to find the square root
of 5 we can search for the positive root of the function
f (x) = x2 5
The derivative of that function is
f 0 (x) = 2x
substituting it into the recurrence relation 2.18, we obtain


xn2 5 1
5
xn+1 = xn
=
xn +
2xn
2
xn
which is exactly the same sequence that we have already obtained and solved in the previous
section. However, in this case we did not need to introduce any clever tricks; we just applied the
standard method.
there are any regions where f and f 0 are both equal to zero, the roots will not be isolated points, but there will
be a whole interval with an infinite number of roots. In this section we will not study those kind of roots.
3 If

23

2.6 References

2.6

References

Some useful references, with a level similar to the one used here, are Chaos (Alligood et al., 1996),
A First Course in Chaotic Dynamical Systems (Devaney, 1992) and Chaos and Fractals (Peitgen
et al., 1992).

2.7

Multiple-choice questions

1. The state variable of a first-order, discrete 3. A first-order discrete dynamical system has
dynamical system takes on the values from
a single fixed point at 0.739, and no cycles.
the following sequence:
Starting with an initial value 2, the evolution
{3.4, 6.8, 7.2, 5.1, 6.8, ...}
of the system is the sequence:
what can be concluded about that system:
{2, 0.915, 0.610, 0.820, 0.683, . . .}
what can be said about that system?
A. it does not have any cycles with period
A. it is chaotic.
less than 5.
B. it has a fixed point.

B. the fixed point is attractive.

C. it is a chaotic system.

C. it has a cycle with period 2.

D. it has a cycle of period 3.

D. it has a cycle with period 3.

E. it has a cycle of period 2.

E. the fixed point is repulsive.

2. The figure shows the staircase diagram of the 4. A function F(x) has the following properties:
discrete dynamical system yn+1 = y2n 0.2,
F 2 (2) = 5
F(5) = 2
which has two fixed points y = 0.17 and
thus,
we
can
conclude
that
the discrete dyy = 1.17.
namical system xn+1 = F(xn ) has a cycle
yn+1
with period equal to:
A. 2
B. 3
C. 4

D. 5
E. 1

yn
1

what type of fixed point is y = 1.17?

A. repulsive focus.
B. attractive focus.
C. attractive node.
D. part of a cycle with period 2.
E. repulsive node.

5. The figure shows the staircase diagram for

the first 40 iterations of a discrete dynamical
system.

24

Discrete dynamical systems

thus, we can conclude that the system has:

C. a cycle with period 2.

A. an attractive focus.

D. a cycle with period 3.

B. a repulsive focus.

E. a cycle with period 40.

2.8

Problems

1. The sequence obtained in this chapter to calculate square roots,



a
1
xn +
xn+1 =
2
xn

was
already known by the Sumerians, 4000 years ago. Using that method, calculate 3, 15
and 234. Use any positive initial value and represent the number a as a floating-point number
(for instance, 3.0), to force Maxima to give its results also as a floating-point number. In each
case, compare the result with the value obtain using function sqrt() in Maxima.
2. Assume that the current whale population in the world is 1000 and that every year the normal
increase of the population (births minus deaths by natural causes) is 25%. Assuming that the
number of whales killed by fishermen every year remained constant at 300 during the next
years, how will the whale population evolve during the next 10 years?
3. Compute the first 10 terms of the sequence defined by the equation:
xn+1 = xn2 2
using the following initial values:
(a) x0 = 1

(c) x0 = 2

(b) x0 = 0.5

(d) x0 = 1.999

Discuss the behavior of the sequence in each case.

4. For each function in the following list, the point y = 0 makes part of a cycle for the system
yn+1 = F(yn ). Determine the period of the cycle for each case and calculate the derivative
of the function in order to determine whether the cycle is attractive or repulsive. Draw the
staircase diagram of the sequence with initial value 0.
(a) F(y) = 1 y2

(b) F(y) = cos y

2
1
3
(c) F(y) = y3 y2 + 1
2
2

(d) F(y) = |y 2| 1
4
(e) F(y) = arctg(y + 1)

5. Find the fixed points and the cycles with period 2 of the dynamical system:
yn+1 = F(yn )
and classify each point and cycle as attractive or repulsive, for each of the following cases:

25

2.8 Problems
y
2
2y
(b) F(y) =
10
(c) F(y) = y4 4y2 + 2
(a) F(y) = y2

(d) F(y) =

sin y
2

(e) F(y) = y3 3y
(f) F(y) = arctg (y)

In each case, start by drawing a staircase diagram using the function staircase and use it to find
out the position of the fixed points and cycles; use the option domain to get a better overview
of the position of the fixed points. Then try to find the points analytically. In some cases that
will not be possible and the result will have to be obtained approximately from the plots.
6. Considering the sequence xn+1 = |xn 2|
(a) Find all the fixed points. Show those points in a plot of the functions F(x) = |x 2| and
G(x) = x.
(b) Explain the kind of sequence that will be obtained if x0 were a integer, either even or odd.
(c) Find the solution for the initial value 8.2.
(d) Find all the cycles with period two. Show all the points in those cycles in a plot of the
functions F 2 (x) and G(x) = x.
7. Consider the function


F(x) =

2x
, if x 1
4 2x , if x 1

(a) Show that F(x) is equivalent to 2 2|x 1|.

(b) Plot, in the same graph, the functions F(x), F 2 (x), F 3 (x) and G(x) = x. What can you
conclude about the fixed points and cycles of the system xn+1 = F(xn )?
(c) Make a table or plot a graph of n against xn , between n = 0 and n = 100, for each of the
initial values 0.5, 0.6, 0.89, 0.893 and 0.1111111111. Discuss the results obtained.
(d) In the previous item, the sequence remains constant, starting at n = 55, for each of the
initial values. Compute again the sequences obtained in the last item, using the following
program, which uses higher numerical precision than the function evolution from the
dynamics module.
evolution60(f, x0, n) :=
block([x: bfloat(x0), xlist:[0, x0], fpprec: 60],
for i thru n
do (x: ev(f), xlist: append(xlist, [i, float(x)])),
openplot_curves([["plotpoints 1 nolines"], xlist])
)
what can you conclude?

Index
A
attractive focus, 16
attractive node, 16
C
chaotic, 15
cycle, 17
D
diff, 5
difference equation, 9
dynamics, 12, 14, 25
E
Euler, 6
evolution, 12, 25
evolution equation, 9
expand, 11
F
fixed point, 15
I
iteration method, 20
K
kill, 11
L
limit, 6
list, 3
M
Maxima, 1
N
Newtons method, 21
nticks, 7
numer, 6

O
Ohms law, 2
P
parametric, 6
periodic points, 17
plot2d, 5
R
repulsive focus, 16
repulsive node, 16
roots, 19
S
solve, 3, 5, 16
staircase, 12
staircase diagram, 12
V
voltage-current characteristic, 3

Bibliography
Alligood, K. T., Sauer, T. D. & Yorke, J. A. (1996), Chaos - an introduction to dynamical systems,
Springer-Verlag.
de Souza, P. N., Fateman, R., Moses, J. & Yapp, C. (2003), The Maxima book,
http://maxima.sourceforge.net.
Devaney, R. L. (1992), A first course in chaotic dynamical systems: theory and experiments,
Westview Press.
Peitgen, H.-O., Jurgens, H. & Saupe, D. (1992), Chaos and fractals: new frontiers of science,
Springer-Verlag.