https://www.math.uni-rostock.de/~wj/projects/Chaos_Lorenz/lorenz_equations.

pdf (20-July-2023)

THE LORENZ EQUATIONS

MARIANNA KATERINA PELEKANOS

MSc MATHEMATICS

SCHOOL OF MATHEMATICAL SCIENCES

QUEEN MARY UNIVERSITY

September 10, 2009

1
ABSTRACT

The project aims to analyse the work of Edward Lorenz. The way dynamical systems
with complex behaviour, such as the weather behave after a long period of time. The
project will mention the Lorenz equations and what they represent. What happens to
the system if we vary one parameter having the other parameters fixed will also be
discussed here. In addition it will be discussed about what happens to the dynamical
system if we slightly change the initial conditions. Finally it will talk about what is
meant by sensitive dependence on initial conditions and the butterfly effect and how
these to concepts are related to chaos theory.

2
TABLE OF CONTENTS

Acknowledgements 4

1. Introduction 5

2. Derivation of the Lorenz Equations 7

2.1 The Bernard Rayleigh Experiment 7
2.2 Derivation of the Lorenz Equations 9

3. The Parameter r 18
3.1 Solutions of the Lorenz Equations 18
3.2 Different Values of r 19

4. Nontrivial Fixed Points 21

4.1 The Region r>1 21
4.2 Time Dependent non Stationary Solutions 23
4.3 Period Doubling Scenarios 24

5. The Butterfly Effect 27

6. Conclusion 29

Bibliography 30

3
ACKNOWLEDGEMENTS

This dissertation was an interesting experience for me, but would not have been
possible without the support of my supervisor Doctor Wolfram Just which I thank
very much for all the help and support he have given me.

My special thanks to my parents, who made this Masters possible and also to my
friends for their support.

4
CHAPTER 1
INTRODUCTION

Could anyone think that a tiny error in an input number of a computer program can
have huge consequences in the long run behaviour of a complex dynamical system?

Edward Lorenz was a meteorologist and he devoted his work to try to give reasons
why it is difficult to give accurate weather forecasting for more than two or three
weeks.

Lorenz studied Mathematics at Dartmouth Collage and then had a Masters degree in
Harvard University also in Mathematics. He, as well received a SM and a ScD in
Meteorology from MIT. Furthermore, he served for the US Army Air Corps in World
War II as a weather forecaster and it is then when he chose to do a graduate job in
Meteorology. Soon after he took his doctorate in Meteorology. Lorenz was also
awarded with many prized such as the Crafoord prize and the Kyoto prize. (MIT,
2008, April 16).

The atmosphere can be modelled by mathematical equations. A small incident during

a survey to model the atmosphere led Lorenz to discover chaos. After this event,
Lorenz was searching for complex mathematical systems. Such system is the Lorenz
attractor or Lorenz equations which will be studied in the project. At the beginning
the Lorenz equations were derived to model the atmosphere mathematically.

Firstly, the project will concentrate on deriving the Lorenz equations. What Lorenz
considered before the derivation, what mathematical and physical notations were used
is something it will also be mentioned in this chapter. It will be explained what the
variables and what the parameters of these equations represent.

The next two chapters will look at the different parameter range of these equations.
Basically we will adjust one parameter having the other parameters fixed and observe
the different behaviours the system takes. If it has stable or unstable points, if it has
periodic orbits and even if the system has chaotic behaviour. It will also be discussed

5
why the invariant set of the Lorenz equations is also called Lorenz attractor and about
the nontrivial fixed points of the attractor.

After that, the project will talk about the butterfly effect, a term first conceived by
Lorenz himself in a paper he wrote in 1972. This chapter will also mention sensitive
dependence on initial conditions and how this is related to chaos theory. This will be
also explained by the means of some plots.

Finally, chapter 6 presents the conclusions of the project. Who else is involved with
chaos theory and where the Lorenz equations are used.

6
CHAPTER 2
DERIVATION OF THE LORENZ EQUATIONS

Before deriving Lorenz equations, let us first say a few things about what Lorenz did,
some mathematical details he considered and about the small experiment he prepared
that led him to find the equations the project will talk about.

2.1 The Benard Rayleigh experiment

Lorenz (1963) considered a two-dimensional fluid element, such as water, placed

between two plates. The plates are separated by a distance H . The lower plate is then
heated uniformly and the upper plate was cooled. The difference of the temperature
between the two plates is ΔΤ and this difference is a fixed parameter. The x and y
coordinates are parallel to the two plates and the z coordinate is perpendicular to the
plates. The equations of motion are given by the Navier Stokes equations for an
incompressible fluid (∇ ⋅ u = 0 ) , and the equation of the temperature is given by the
heat transfer equation. It is assumed that motion variations are in one direction and are
parallel to the two plates. This means that the temperature and the velocity of the fluid
depend only on x and z and independent of the y direction. So there are no
variations in the y direction.

In his paper, Lorenz (1963), considered systems of deterministic equations used in

hydrodynamics. By saying deterministic equation we mean an equation that has the
power over a dynamical system without arbitrary forces. He was mostly interested in
periodic solutions and we will discuss this later on in the project. To derive his
equations, he coped with a phase space Γ in which there was only one trajectory that
passed through every point and where time was continuous. In addition the trajectory
was bounded with a uniform way. This means that there is a bounded area, say R in
which all trajectories eventually stayed in this area. Lorenz organised the trajectories
in three ways. The first way was according to the existence of transient properties.
Regarding the stability of the trajectories was the second way and the third way was
according to the periodicity of the trajectories.

7
Lorenz (1963) used Saltzman’s (1962) equations of free convection to derive his
system of three ordinary differential equations. When we say convection we mean the
heat flow of a fluid from a hot area to a cold area. Saltzman (1962) rewrote the
equations of motion in terms of θ and ψ .He derived his equations using double
Fourier transforms in x and z , but we will not discuss this here.

Saltzman’s two equations are:

∂ 2
∇ψ =−
(
∂ ψ , ∇ 2ψ )
+ ν∇ 4ψ + gα
∂θ
,
∂t ∂ ( x, z ) ∂x (1)

∂θ ∂ (ψ , θ ) ΔΤ ∂ψ
= + + κ∇ 2θ , (2)
∂t ∂ ( x , z ) Η ∂x

where ψ is a stream function for a two-dimensional motion, θ is the leaving

temperature in the state of no convection, g is a constant representing the acceleration
of gravity, α is the coefficient of thermal expansion, ν is the kinematic viscosity, κ

is the thermal conductivity and

(
∂ ψ , ∇ 2ψ) =
(
∂ψ ∂ ∇ 2ψ
−
) (
∂ ∇ 2ψ ∂ψ )
.
∂ ( x, z ) ∂x ∂z ∂x ∂z

These equations are introduced, taking into account that motion is parallel to the
x − z plane and there is no motion in the y direction.

At both the boundaries, the vertical velocity is equal to zero. Furthermore at a free
boundary the tangential stress is also zero and at a rigid boundary the tangential
velocity is zero. At the boundaries this means that

ψ = 0, ∇ 2ψ = 0, ∂ψ ∂z = 0 .

For the derivation of the Lorenz equations we will only useψ = ∇ 2ψ = 0 . Here, we
note that these boundary conditions apply to z = 0 and z = H , where H is the height
of the fluid.

8
2.2 Derivation of Lorenz Equations

The nonlinear system of the partial differential equations (1) and (2) with the
boundary conditions mentioned above is difficult to solve. To make things easier, one
should choose some approximations to reduce the partial differential equations
approximately to nonlinear ordinary differentia equations. For this reason Lorenz
(1963) introduced solutions of the following form, as these expressions satisfy the
boundary conditions.

(
a 1+ a2 )
−1
κ −1ψ = Χ 2 sin (πaH −1 x )sin (πH −1 z ),
(3), (4)
πRc−1 Ra ΔΤ −1θ = Y 2 cos(πaH −1 x )sin (πH −1 z ) − Z sin (2πH −1 z ).

Here, X , Y , Z are only functions of time t , Ra is the Rayleigh number, after the one

(
who invented it. This is equal to Ra = gαH 3 ΔΤν −1κ −1 and Rc = π 4 a −1 1 + a 2 )
3
is the
critical value of the Rayleigh number.

Lorenz (1963) substituted equations (3) and (4) into equations (1) and (2) and with
some calculations he ended up with the following set of ordinary differential
equations, which are known as the Lorenz equations.

⋅
X = −σX + σY
.
Y = − XZ + rX − Y
.
Z = XY − bZ .

σ , r, b are constants to be determined and the dot represents derivative with respect to
time.
It will now be discussed in some detail how the Lorenz equations are obtained from
Saltzman’s (1962) partial differential equations.

Now we will substitute (3) into (1). First we will compute the LHS of (1).

9
 ∂ 2ψ ∂ 2ψ ∂ 2ψ
∇ 2ψ = + +
∂x 2 ∂y 2 ∂z 2

∂ 2ψ
From equation (3) we see that nothing depends on y so = 0 and hence
∂y 2

∂ 2ψ ∂ 2ψ
∇ 2ψ = + .
∂x 2 ∂z 2

∂ 2ψ
=−
(
1+ a2 ) ( )
κX 2 πaH −1 sin πaH −1 x sin πH −1 z
2
( ) ( )
∂x 2
a

∂ 2ψ
=−
(
1+ a2 ) ( )
κX 2 πH −1 sin πaH −1 x sin πH −1 z
2
( ) ( )
∂z 2
a

and hence, taking common factors

∇ 2ψ = −
(1 + a )κX
2
( ) (
2 sin πaH −1 x sin πH −1 z πaH −1 )([ ) + (πH ) ]
2 −1 2

∂ 2
To compute ∇ ψ we take into account that the only variable that depends on t is
∂t
X.

∂ 2
∇ψ =−
1+ a2 ( ) (
κ 2 sin πaH −1 x sin πH −1 z πaH −1 ) ( )([ ) + (πH ) ]X
2 −1 2
.

∂t a

For the RHS of (1) we compute the following quantities

∂θ
∂x
R ΔΤ
=− c
Ra π
[ ( ) (
Y 2 πaH −1 sin πaH −1 x sin πH −1 z . ) ( )]

10
 ∂ 4ψ ∂ 4ψ ∂ 4ψ
∇ ψ =∇ ∇ ψ = 4 + 4 +2 2 2
4 2 2

∂x ∂z ∂x ∂z

∂ 4ψ
= 2
1+ a2 ( ) ( )
κX πaH −1 sin πaH −1 x sin πH −1 z
4
( ) ( )
∂x 4
a
∂ψ4
= 2
1+ a2 ( ) ( )
κX πH −1 sin πaH −1 x sin πH −1 z
4
( ) ( )
∂z 4
a
∂ψ4
=
∂ ⎛ ∂ 2ψ ⎞ 1 + a 2
2
⎜ 2 ⎟⎟ =
( ) (
κ 2 X πaH −1 πH −1 sin πaH −1 x sin πH −1 z
2 2
)( ) ( ) ( )
2 ⎜
∂x ∂z
2 2
∂z ⎝ ∂x ⎠ a

∇ 4ψ = 2
(1 + a )κX sin (πaH x )sin (πH z )([ πaH ) + (πH )
2
−1 −1 −1 4 −1 4
(
+ 2 πaH −1 ) (πH ) ].
2 −1 2

The final term of the RHS of (3) is

(
∂ ψ , ∇ 2ψ )
which according to Saltzman (1962)
∂ ( x, z )
this, as mentioned above, equals to

(
∂ ψ , ∇ 2ψ
=
)
∂ψ ∂ ∇ 2ψ
−
(
∂ ∇ 2ψ ∂ψ
=
) ( )
∂ ( x, z ) ∂x ∂z ∂x ∂z

=2
(1 + a ) 2 2
[ ]
κ 2 X 2 (πH −1 )(πH −1 ) + (πH −1 ) cos(πaH −1 x )sin (πH −1 z )sin (πaH −1 x )cos(πH −1 )z +
2 2

2
(1 + a ) 2 2
[ ]
κ 2 X 2 (πaH −1 )(πH −1 )(πH −1 ) + (πaH −1 ) cos(πaH −1 x )sin (πH −1 z )sin (πaH −1 x )cos(πH −1 z ).
2 2
2
a

Lorenz (1963), in his paper omitted all other trigonometric terms besides the ones
∂ (ψ , ∇ 2ψ )
occurring in equations (3) and (4). So we ignore the term as it doesn’t
∂ ( x, z )
contain any of the trigonometric terms in the two equations.

So putting all the terms together in

∂ 2
∇ψ =−
∂ ψ , ∇ 2ψ (
+ ν∇ 4ψ + gα
∂θ )
we get
∂t ∂ ( x, z ) ∂x

11
−
(1 + a )κ2
( ) (
2 sin πaH −1 x sin πH −1 z πaH −1 )([ ) + (πH ) ]X =
2 −1 2
.

ν
(
1+ a2 )
κX 2 sin πaH −1 x sin πH −1 z πaH −1 ( ) ( )([ ) + (πH )]
2 −1
2
−
a
gα
Rc ΔΤ
Ra π
[ (
Y 2 πaH −1 sin πaH −1 x sin πH −1 z ) ( ) ( )]

Cancelling the terms that occur in both sides and using the definitions of Ra and Rc ,

this simplifies to

2
π2 ⎛π2 ⎞ π4
( ) ( ) ( )
.
− 2 1 + a X = ν ⎜⎜ 2 1 + a 2
⎟⎟ − ν 4 1 + a 2
2 2

H ⎝H ⎠ H
π2 π2
( dX
) ( )
.
⇒ X = −ν 2 1 + a 2 X + ν 2 1 + a 2 Y =
H H dt

Lorenz (1962), in his paper usedτ = π 2 H −2 1 + a 2 κt . So using the chain rule ( )

dX dX dt dt H 2 1
= ⋅ = 2
dτ π (1 + a 2 )κ
and with we get
dτ dt dτ

dX ⎡ π2 π2 ⎤ H2
(
= ⎢− ν 2 1 + a 2 X + ν 2 1 + a 2 Y ⎥ ⋅ 2
dτ ⎣
1
. ) ( ) ( )
⎦ π 1+ a κ
2
H H

⋅
This reduces to X = −νκ −1 X + νκ −1Y and with νκ −1 = σ we obtain the first required
equation, that is

⋅
X = −σX + σY .

In order to obtain the other two equations we consider equation (2). We substitute
equations (3) and (4) into this equation to get the following relationships.

To calculate ∂θ on the LHS of (2), we take into consideration that the only
∂t
variables depending on t are Y and Z . Therefore we have

12
∂θ ΔΤRc ⋅ ΔΤRc ⋅
∂t
=
πRa
(
Y 2 cos πaH −1 x sin πH −1 z −
πRa
) (
Z sin 2πH −1 z . ) ( )

Now we calculate the RHS of (2). Here we have

ΔΤ ∂ψ ΔΤ (1 + a 2 )
= κX 2 (πaH −1 )cos(πaH −1 x )sin (πH −1 z ) .
H ∂x H a

∂ 2ψ ∂ 2ψ ∂ 2ψ
Next we calculate ∇ 2θ = + + . As before, nothing depends on y so we
∂x 2 ∂y 2 ∂z 2

∂ 2ψ ∂ 2ψ
only need to consider ∇ 2θ = + . After taking common factors, this equals to
∂x 2 ∂z 2

∇ 2θ =
ΔΤRc
πRa
[ { }
− Y 2 cos(πaH −1 x )sin (πH −1 z )(πaH −1 ) + (πH −1 ) + Z (2πH −1 ) sin (2πH −1 z )
2 2 2
]
∂ (ψ ,θ )
The last term we need to calculate is which as before this equals to
∂ ( x, z )

∂ (ψ ,θ ) ∂ψ∂θ ∂θ∂ψ
= − = (5)
∂ ( x, z ) ∂x∂z ∂x∂z

(1 + a ) 2κ (πaH )(πH ) ΔΤR

2
−1 −1 c
( ) ( ) ( ) (
XY cos πaH −1 x sin πH −1 z cos πH −1 z cos πaH −1 x − )
a πR a

(1 + a ) 2κ 2 (πH )XZ cos(πaH x )sin (πH z )cos(2πH z ) +

2
−1 −1 −1 −1

a
2ΔΤR (1 + a )
κ (πaH )(πH )XY sin (πaH x )sin (πH z )sin (πaH x )cos(πH z )
2
c −1 −1 −1 −1 −1 −1

πRa a

Now if we take the trigonometric term from the second line from equation (5) we
observe that this can be reduced to

13
 ( ) ( )
cos πH −1 z sin πH −1 z cos 2 πaH −1 x = ( ) 1
( ⎛ 1 − 2 sin 2πaH −1 x
sin 2πH −1 z ⎜⎜ ) ( ) ⎞⎟ =
⎟
2 ⎝ 2 ⎠

1
4
( 1
) (
sin 2πH −1 z − sin 2πH −1 z sin 2πaH −1x
2
) ( )

Noticing that the second term of the above equation does not appear in equations (3)
and (4) we neglect this term as Lorenz (1963) did in his paper.

The trigonometric term in the third line becomes

( ) ( ) ( ) (
cos πaH −1 x sin πH −1 z cos 2πH −1 z = cos πaH −1 x sin πH −1 z 2 cos 2 πH −1 z − 1 = ) ( )( ( ))
( ) ( ) ( )
2 cos πaH −1 x sin πH −1 z cos 2 πH −1z − cos πaH −1 x sin πH −1 z ( ) ( )

As explained before, Lorenz (1963) omitted terms not involving trigonometric terms
as those appear in (3) and (4). So we omit the first term of the above equation.

Finally, we take the term in the last line of equation (5), we see that this is

⎛ 1 − cos(2πaH −1
x)⎞
sin 2
(πaH −1
x ) sin (πH −1
z ) cos(πH −1
z) =
1
(
sin 2πH −1
z )⎜
⎜
⎟ =
⎟
2 ⎝ 2 ⎠

=
1
4
( 1
) (
sin 2πH −1 z − sin 2πH −1 z cos 2πaH −1 x .
4
) ( )

As Lorenz (1963) did, we omit the second term of this equation.

Now putting all the terms together, we obtain

∂θ ∂ (ψ ,θ ) ΔΤ ∂ψ
=− + + κ∇ 2θ
∂t ∂ ( x, z ) H ∂x

14
 ΔΤRc ΔΤRc
( ) ( ) ( )
⋅ ⋅
⇒ 2 cos πaH −1 x sin πH −1 z Y − sin 2πH −1 z Z =
πRa πRa

−
(1 + a ) 2κ (πaH )(πH ) ΔΤR
2
−1 −1 c 1
(
sin 2πH −1 z XY − )
a πR a 4
(1 + a )κ 2 2 (πaH )(πH ) ΔΤR 1 [cos(πaH x )sin (πH z )]XZ −
2
−1 −1 c −1 −1

a πR 4 a

2
ΔΤRc 1 + a
κ
( 2
) (πaH )(πH ) 1 sin (πH z )XY +
−1 −1 −1

πRa a 4
(
ΔΤ 1 + a 2 ) (
κ 2 πaH −1 cos πaH −1 x sin πH −1 z X − ) ( ) ( )
H a
κ
ΔΤRc
πRa
[ {( ) ( )} (
2 πaH −1 + πH −1 cos πaH −1 x sin πH −1 z Y + 2πH −1 sin 2πH −1 z Z
2 2 2
) ( ) ( ) ( ) ]

Comparing the coefficients of the trigonometric term cos πaH −1 x sin πH −1 z we ( ) ( )

have

ΔΤRc
2Y =
⋅ (
1 1+ a2
κ 2 πaH −1
)
ΔΤRc
(
2πH −1 XZ +
ΔΤ 1 + a 2
)
κ 2 πaH −1 X − ( ) ( ) ( )
πRa 2 a πRa H a

κ 2
ΔΤRc
πRa
πaH −1 [( ) + (πH ) ]Y .
2 −1 2

After some simplifications we get to

⋅
Y=
dY
dt
π2
( π2 R
)
= − 1 + a 2 κ 2 XZ + 1 + a 2 κ 2 a X − κ πaH −1
H H Rc
( ) [( ) + (πH ) ]Y .
2 −1 2

H2 1 dt H2 1
As we did before we use t = τ =
κπ 2 (
1+ a 2
and
)
dτ κπ 1 + a 2
2
we obtain, using
( )
the chain rule

dY dY dt
= =
dτ dt dτ
⎡ π2 π 2 Ra π2 ⎤ ⎡ H2 ⎛ 1 ⎞⎤
⎢ − (
1 + a 2
κ XZ )
+ 1 + a 2
κ X − κ 1 (
+ a 2
Y ⎥⋅⎢ 2 ) ( ) ⎜⎜
( )
⎟⎟⎥
H2 H 2 Rc H2 ⎦ ⎣ κπ ⎝ 1+ a
2
⎣ ⎠⎦

15
 Ra
With = r we finally obtain
Rc

⋅
Y = − XZ + rX − Y .

This is the second equation of the Lorenz equations.

(
Finally we compare the coefficients of sin 2πH −1 z . This is )

ΔΤRc ⋅
Z =−
1 1+ a2 (
κ πaH −1 πH −1
) (
ΔΤRc
XY −
1 ΔΤRc
)( )
κ πaH −1 πH −1 XY + ( )( )
πRa 2 a πRa 2 πRa
ΔΤRc
κ
Ra
(
2πH −1 Z
2
)

This simplifies to

π2 π2
( )
⋅ dZ
Z= = 1 + a 2 κ 2 XY − 4κ 2 Z .
dt H H

H2 1
As we did before, using the chain rule with t = τ
κπ 2
(
1+ a2 )
we have

dZ dZ dt ⎡ π2 π2 ⎤
= ( )
= ⎢ 1 + a 2 κ 2 XY − 4κ 2 Z ⎥ ⋅ 2
H2
= XY − bZ ,
dτ dt dτ ⎣ H H ⎦ κπ 1 + a
2
( )

(
where b = 4 1 + a 2 )
−1
. So we find the third equation we required. This is
⋅
Z = XY − bZ .

The system of the three equations is:

⋅
X = −σX + σY (6)
.
Y = − XZ + rX − Y
.
Z = XY − bZ . 16
The variables X , Y , Z represent respectively the rate of convective overturning, how
the temperature varies in the horizontal direction and how temperature varies in the
vertical direction.

The three real and positive parameters are:

σ which is called the Prandtl number and it represents the ratio of the viscosity of
the fluid of a material to its thermal conductivity, r is the Rayleigh number and it
represents the temperature difference between the top and bottom of the system and
b represents a ratio of width to height of the container being considered.

Lorenz used σ = 10, r = 28 and b = 8 . In the next chapter we will study what
3
happens if we change the values of these parameters

17
CHAPTER 3
THE PARAMETER r

After deriving the equations, the next step is to analyse what happens when we vary
one of the parameters in the Lorenz equations. We will investigate what happens
when we fix the parameters b and σ and vary r . As we said in the previous chapter
this parameter is the Rayleigh number. Here we will study what we can observe for
small and large r , say r >> 1 . We will also examine if the system has stable or
unstable solutions and in result if periodic orbits exist. Finally, this chapter will state
if this system can ever reach chaotic behaviour and if so for what value of r .

3.1 Solutions of the Lorenz Equations

Here we consider a solution of the Lorenz equations, say X (t ) , Y (t ) , Z (t ) and we are

going to study how a nearby solution, say X (t ) + x0 (t ) , Y (t ) + y 0 (t ) , Z (t ) + z 0 (t ) will

behave. x0 (t ) , y 0 (t ) and z 0 (t ) are small perturbations which depend on τ and so we

end up with the following equation.

⋅
⎡ x0 ⎤ ⎡ − σ σ 0 ⎤⎡ x0 ⎤
⎢ y ⎥ = ⎢− Z + r − 1 − X ⎥⎥⎢⎢ y 0 ⎥⎥ , (7)
⎢ 0⎥ ⎢
⎢⎣ z 0 ⎥⎦ ⎢⎣ Y X − b ⎥⎦⎢⎣ z 0 ⎥⎦

Equation (7) allows us to study stability properties of the solution X (t ) , Y (t ) , Z (t ) .

The coefficients of these equations are not all constants. Some of them are variables
and change as time changes.

To calculate the (change of phase space) volume of the system we need to calculate
the divergence of the vector field. That is, we calculate the trace of the Jacobean

18
 ∂
(− σx + σy ) + ∂ (rx − y − xz ) + ∂ (− bz + xy )
∂x ∂y ∂z
= −σ − 1 − b = −(σ + 1 + b ).

Because σ and b are positive numbers, the change in volume is negative and so the
volume is contracting. This means all attractors have zero volume. We will discuss
this later on in the chapter in more detail.

3.2 Different Values of r

In this section we will investigate what happens for different values of r . At this point
we want to study simple stationary. That is solutions that do not depend on time. We
shall work within a simply connected and closed region R . This region contains a set
A = ∩ t ≥0 φt (R ) . All trajectories will eventually enter this set and will never leave.
Therefore this set is an attracting set and has bounded solutions.

We now study the stability of the trivial fixed point X = Y = Z = 0 . For a steady state
solution of Lorenz equations we require X = Y = Z = 0 . This is the state where there
is no convection. That is, as we said previously, there is no heat flow from hot areas to
cold areas.

So for this situation the characteristic equation of the matrix (7) is

⎛ ⎡− σ σ 0 ⎤ ⎡λ 0 0 ⎤ ⎞
⎜ ⎟
det⎜ ⎢⎢ r − 1 0 ⎥⎥ − ⎢⎢ 0 λ 0 ⎥⎥ ⎟ = 0
⎜⎢ 0 0 − b⎥⎦ ⎢⎣ 0 0 λ ⎥⎦ ⎟⎠
⎝⎣

19
⎛−σ − λ σ 0 ⎞
⎜ ⎟
⎜ r −1− λ 0 ⎟=0
⎜ 0 − b − λ ⎟⎠
⎝ 0

− (σ + λ )[(1 + λ )(b + λ − 0)] + σr (b + λ ) + 0 = 0

[ ]
− (b + λ ) λ2 + λ (σ + 1) + σ (1 − r ) = 0

(b + λ )[λ2 + λ (σ + 1) + σ (1 − r )] = 0 . (8)

Let us take equation (8) and we shall fix σ and b and vary r . First we shall take the
case when r > 0 . So we have

(b + λ )[λ2 + λ (σ + 1) + σ (1 − r )] = 0 . This is

− (σ + 1) ± (σ + 1)2 − 4σ (1 − r ) − (σ + 1) ± (σ − 1)2 + 4σr

λ = −b or λ1, 2 = = .
2 2

So for r > 0 the equation has three solutions and all of them are real. The solutions
are

− (σ + 1) + (σ − 1)2 + 4σr − (σ + 1) − (σ − 1)2 + 4σr

λ1 = −b, λ2 = , λ3 =
2 2

λ1 = −b is negative as b is a positive constant. λ3 is also negative as the square root

is positive and (σ + 1) is also positive. Since they have a negative sign the expression

is negative. λ 2 is positive as (σ − 1)2 + 4σr > (σ + 1) .

Now, if r < 1 the three roots are all negative and the origin (0,0,0 ) is a stable point.

In the next chapter we will see what is happening to the system in the range r > 1 .

20
CHAPTER 4
NONTRIVIAL FIXED POINTS

In this chapter we will mostly consider the region r > 1 and the non trivial fixed
points occurring in this region. We note that the nontrivial fixed points correspond to
stationary convection. Also, as a remark, a fixed point is a time independent solution.

4.1 The Region r >1

If we take the region r > 1 then the matrix (7) has a characteristic equation

λ3 + (σ + b + 1)λ2 + (r + σ )bλ + 2σb(r − 1) = 0 (9)

Here, we first have to compute the nontrivial fixed point, say X ∗ , Y ∗ , Z ∗ . These
numbers are computed from the equations of motion, i.e. equations (6), letting
⋅ ⋅ ⋅
X = Y = Z = 0 . Then equations (6) give three equations for X ∗ , Y ∗ , Z ∗ , namely
−σ *X * + σ *Y * , − X * Z * + r * X * − Y * , X *Y * − b * Z * . Solving these equations for X ∗ ,
Y ∗ , Z ∗ we obtain two extra solutions for the Lorenz equations. We will call them s1

(
and s 2 . These solutions are fixed points and are ± b(r − 1),± b(r − 1), r − 1 . )

Now we investigate the stability of this fixed point using equation (7) and the
associated characteristic equation which is equation (9), for the nontrivial fixed point.

In this case the origin is a saddle point. Such points are unstable but they have a
stable and unstable direction.

σ (σ + b + 3)
At r = we have
(σ − b − 1)
⎡ σ (σ + b + 3) ⎤ ⎡σ (σ + b + 3) ⎤
λ3 + (σ + b + 1)λ2 + ⎢ + σ ⎥bλ + 2σb ⎢ − 1⎥ = 0
⎣ (σ − b − 1) ⎦ ⎣ (σ − b − 1) ⎦
(σ − b − 1)λ3 + (σ − b − 1)(σ + b + 1)λ2 + (σ (σ + b + 3) + σ )bλ + 2σb(σ (σ + b + 3) − 1) = 0.

21
At this r value we obtain an imaginary pair of eigenvalues, which indicates the
second instability. The eigenvalues are

λ1 = −(σ + b + 1), λ2,3 = ±i 2σ (σ + 1) / (σ − b − 1) .

We can call this critical value of r , rh . Then for r = rh we have the so-called Hopf

bifurcation (Guckenheimer, Holmes, 1983, p.93). For r = rh the fixed points become
unstable.

This value of r is a critical value because it shows us if we have stability or instability

of steady convection. If σ < b + 1 there exists a stable steady convection and for
σ > b + 1 there exists unstable steady convection.

If we now take σ = 10 and b = 8 then we have

3

r = rh = =
(
σ (σ + b + 3) 10 10 + 3 + 3
8 )
≈ 24.74 .
σ − b −1 10 − 8 − 1
3

For 1 < r < 24.74 the two solutions s1 and s 2 are stable. For r > 1.346 with σ = 10

and b = 8 the eigenvalues become complex and their real parts are negative.
3

For r > 24.74 , s1 and s 2 are not stable. For this condition there is one real eigenvalue
which is negative and two complex eigenvalues. The real parts of the complex
eigenvalues are both positive.

Lorenz (1963) chose a slightly larger value of r . He fixed r = 28, σ = 10, b = 8 and
3
then used numerical integration for equations (5) with initial condition the origin
(0,0,0) .

22
4.2 Time Dependent non Stationary Solutions

This section will concentrate mostly on the time dependent solutions of the Lorenz
equations at the standard parameter values, namely the so called Lorenz attractor. As
we have seen in the previous sections the Lorenz equations acknowledge stable fixed
point solutions for r < rh . If r > 1 we obtain a non convective state, for 1 < r < rh the
recently generated stable fixed point corresponds to time independent convection. At
r = rh a Hopf bifurcation takes place and in our case no stable time periodic state is

generated for r > rh . This is a so called sub critical Hopf bifurcation. In fact for r > rh
aperiodic time dependence is observed and the corresponding attractor has some quite
‘strange’ figures.

Let us first say a few more details about this so called strange attractor. According to
Lorenz (1979) an attractor has basically two sheets and following an orbit these sheets
seem to join together. From this, take as a accurate statement, it would follow that
pairs of orbits should join together as well and this is not possible. Therefore one
sheet is a combination of two sheets which are very close to each other. Consequently,
repeating this reasoning, two sheets that are joint together are composed of four sheets
four sheets have eight sheets and so on. Hence it appears that there are an infinite
number of sheets. The closed set of this kind of sheets is called a strange attractor.

Now, let us examine in slightly more detail how such a strange attractor can occur in a
dynamical system. For a value r ≈ 13.926 there is a global bifurcation a so called
homoclinic explosion. This means that the system generates an orbit which links the
saddle point at the origin with itself. Roughly speaking the period of such an orbit is
infinite. (Sparrow, 1982, p.26).

One can show that for r < ra , because of the homoclinic explosion the equations of
motion create an infinite number of periodic orbits which are all unstable.

We note that the two stationary points s1 and s 2 are still stable at r ≈ 24.06 and any
trajectory starting near s1 and s 2 will tend to s1 or s 2 . However if we cross the

23
threshold r = rh no longer a stable fixed point exists and the time dependent solution
will ‘bounce’ between all the unstable periodic points

Let us now investigate what happens to the strange attractor in the area r. > ra . The

study will concentrate near the area r = 28 . This is the r value Lorenz (1963)
considered in his paper and from this value he discovered the chaotic behaviour of the
Lorenz attractor but this will be discussed in more detail later on. To study aperiodic
time dependence it is convenient to use a kind of symbolic coarse grained description,
a so called symbolic dynamics. One associates a symbol sequence to the orbit under
consideration. For the Lorenz system it is convenient to generate the symbol sequence
according to the revolution of the two different lopes of the chaotic attractor. One uses
two symbols for the coarse grained dynamics. The first is when we have a ' x' when
the trajectory goes around the point s 2 (x > 0 ) and we have a ' y ' when the trajectory
goes around s 2 (x < 0 ) . At r = 28 we observe that for the sequences ' x' and ' y ' the
periodic orbits and trajectories are not in the set. The periodic orbits and trajectories
are missing because they are separated from the attractor in homoclinic explosions.
These homoclinic explosions occur at a large number of r values. In addition there is
an infinite number of different attractors in all the neighbourhoods of all the values
of r . This situation concerns the trajectories that stay forever in the attractor of the
orbit under consideration. In this strange attractor the orbits and trajectories disappear
in homoclinic explosions from the set, they do not leave. This does not happen for all
the aperiodic trajectories which disappear. So near the value r = 28 there are no
stable periodic orbits.

4.3 Period Doubling Scenarios

This section will give attention to stable periodic orbits of the Lorenz attractor. We
will conceder values of r much larger than r = 28 . Here we will see three intervals of
r values and investigate their periodic behaviour taking in mind that there are other
such intervals smaller than these considered here.

24
The first interval to analyse is 99.524 < r < 100.795 . According to Sparrow (1982,
pp.56-58) there exists a stable periodic orbit in 99.98 ≤ r ≤ 100.795 . We still consider
the two sequences with ' x' and ' y ' where we have mentioned before. Because of the
symmetry of the Lorenz attractor there are stable orbits and these orbits seem to
attract trajectories.

For 99.629 < r < 99.98 there is another stable periodic orbit. (Sparrow, 1982, p.56).
As r tends to 99.98, the loops of the orbit will come together at the end. Here we
have a period doubling bifurcation. That is when the stable periodic orbit becomes
unstable but still exists and it is replaced by a stable periodic orbit with a double
period compared to the previous one. There is another stable orbit in the
interval 99.547 < r < 99.629 . As the value of r decreases more we observe more
period doubling bifurcations.

Moreover, we observe that just below r = 100.795 there is a pair of non-symmetric

stable periodic orbits and above this value these orbits cannot be seen any more and it
is suggested that they disappear in saddle-node bifurcations. (Sparrow, 1982, p.58).
This bifurcation happens when on one side of r there is a non-stable periodic orbit
and a stable periodic orbit. When the parameter increases or decreases in the direction
of r , the two orbits move very close and their periods rends towards the same value.
(Sparrow, 1982, p.52). For r < 99.524 there exists an infinite number of non-stable
periodic orbits.

The second interval is 145 < r < 166 (Sparrow, 1982, p.59). Here things happen in a
similar manner as the above situation but at r = 160 there is a symmetric periodic
orbit which is stable in the range 154.4 < r < 166.07 . At r ≈ 154.4 there is a symmetric
saddle-node bifurcation. This kind of bifurcation has occurred in the Lorenz system
when two saddle non-symmetric orbits and a non-symmetric orbit loses its stability to
a pair of non-symmetric orbit combined together to create a stable symmetric orbit. At
this value of r the orbit loses its stability to a pair of non-symmetric periodic orbits.
(Sparrow, 1982, p.53). These orbits are stable in 148.2 < r < 154.4 . Under the value of
r = 154.4 this orbit also exists and it is non-stable.

25
At r ≈ 148.2 there is period doubling bifurcation. At a value of r greater than
r = 166.07 we have intermittent chaos (Sparrow, 1982, pp.62-63). Here, although the
stable symmetric orbits does not exist, trajectories look to move close to it and then
they wonder off and have chaotic behaviour for a short time and then return to
periodic behaviour. The length of the chaotic intervals gets bigger as the value of r
increases.

The final interval we will mention is r > 214.364 (Sparrow, 1982, pp.66-69). Here we
have the same circumstances as the previous interval with the difference that the last
stable symmetric orbit continues to exist for r > 313 . Because this time r goes up to
infinity and there is no upper limit we cannot observe chaotic behaviour.

In the next chapter we will talk about what we mean by sensitive dependence on
initial conditions and how the Lorenz attractor posses chaotic behaviour.

26
CHAPTER 5
THE BUTTERFLY EFFECT

Lorenz equations were an attempt to describe the atmosphere and weather conditions
in a simple manner but then realised that these equations can be formulated to
describe a laboratory water wheel. He soon discovered that the Lorenz equations
possessed chaotic behaviour as we have seen earlier.

The term ‘butterfly effect’ was first conceived by Edward Lorenz in his effort to
describe sensitive dependence on initial conditions and chaos theory. At first the term
butterfly effect was used only in meteorology to describe the chaotic behaviour of the
weather. Later on chaos theory and the butterfly effect were used in a greater range in
science.

Edward Lorenz accidentally discovered the chaotic behaviour of the atmosphere when
he was working on a computer program for weather forecasting. Lorenz wished to
look again at a specific sequence. Instead of starting from the beginning he decided to
start from somewhere in the middle of this sequence because he didn’t want to waste
much time. He ran the program and left for a while. When he came back he
discovered that the sequence didn’t progress in the same way as the previous one. At
the beginning he thought that there was a problem with the hardware but he soon
discovered what was going on. The first sequence was run with initial conditions
0.506127 . In the second sequence Lorenz typed only 0.506 . The difference between
the two initial conditions was only 0.000127 , which was something very small and
yet the two sequences came out so differently. This is how Lorenz discovered the
chaotic behaviour of the atmosphere and he started a new search to discover chaotic
systems.

It is so amazing that such a tiny difference can cause so huge changes in the long run
behaviour. Following this, Lorenz wrote a paper in 1972 with the title ‘Predictability:
Does the Flap of a butterfly’s wings in Brazil set off a Tornado in Texas?’

27
It is surprising that something that seems so small such as the flapping of a butterfly’s
wings and cannot be felt by anyone, in the long run behaviour can create something as
big as like a tornado in the other side of the earth. In the world of meteorologists and
other scientists, this phenomenon is also known as sensitive dependence on initial
conditions. This is related to chaos theory. A very small variation in initial conditions
may alter the behaviour of a complex system, such as the weather in the far future as
it is not possible to measure the initial conditions of the atmosphere accurately. The
fact that such a miniscule event can create huge differences is one of the reasons we
cannot predict weather forecasting accurately for more than one or two weeks.

Figure 4.1: The Lorenz attractor. Lorenz equations are always following a spiral. The
equations do not settle down to one point and they do not have periodic behaviour as the
system never repeats itself. (Valbonesi, I.,2008 April 17).

28
CHAPTER 6
CONCLUSION

We have seen and discussed about the Lorenz equations. These equations where
studied by many scientists such as Haken, Knoblock, Malkus, Yorke and many more
(Sparrow, 1982, p.4). We have also seen that the Lorenz equations behave chaotically
if we increase the value of the parameter r . At this point, we note that all the results
for stable or unstable points, periodic orbits and bifurcations are made using
numerical integration with the use of a computer.

Lorenz work to try to find the reason we cannot predict the weather accurately for a
long period of time and the butterfly effect was at first only considered by
meteorologists. He could never believe that the butterfly effect and chaos theory were
impacted by other fields of science such as Mathematics, Physics, Biology. (MIT,
2008, April 16). Lorenz (1972) said that we cannot blame the structure of the weather
if we cannot have precise forecasts for more than three weeks. It is our ‘incomplete
knowledge’ of physical principles and some approximations in order to formulate
these principles.

29
Bibliography

Carter, S. (1996). The Lorenz Model. Retrieved 05/06/2009 from:

http://www.taygeta.com/perturb/node2.html

Ho, A. (1995). Lorenz Attractor. Retrieved 05/06/2009 from:

http://www.zeuscat.com/andrew/chaos/lorenz.html

MIT (2008, April 16). Edward Lorenz, father of chaos theory and butterfly effect, dies
at 90. MIT TechTalk. 52 (24). Retrieved 05/06/2009 from:

http://web.mit.edu/newsoffice/2008/obit-lorenz-0416.html

Guckenheimer, J. Holmes, P. (1983). Nonlinear Oscillations Dynamical Systems and

Bifurcations of Vector Fields. New York: Springer-Verlag

Lorenz, E, N. (1963). Deterministic Nonperiodic Flow. Journal of the Atmospheric

Sciences, 20, pp.130-141

Lorenz, E, N. (1972). Predictability; Does the Flap of a Butterfly’s wings in Brazil Set
Off a Tornado in Texas? American Association of the Advancement of Science

Lorenz, E, N. (1979). On the Prevalence of Aperiodicity in Simple Systems.

Massachusetts Institute of Technology. Berlin: Springer, 755, pp.53-75

Saltzman, B. (1962). Finite Amplitude Free Convection as an Initial Value Problem.

Journal of the Atmospheric Sciences, 19, pp.329-341

School of Mathematics and Statistics, University of St Andrews, Scotland.

(2008). Edward Norton Lorenz. Retrieved 10/08/2009 from:

http://www.gap-system.org/~history/Biographies/Lorenz_Edward.html

Sparrow, C. (1982). The Lorenz Equations: Bifurcations, Chaos, and Strange

Attractors. New York: Springer- Verlag

30
Thompson, S, M, T, Stewart, H, B. (1986). Nonlinear Dynamics and Chaos.
Chichester: John Wiley and Sons

Tucker, W. (NA). The Lorenz Attractor Exists: An auto-Validated Proof, Cornell

University. New York: Ithaca

Valbonesi, I. (2008 April 17). Edward Lorenz: Does the Flap of a Butterfly’s Wings
in Brazil Set Off a Tornado in Texas. Ecopolis. Retrieved 10/08/2009 from:

http://www.ecopolis.org/edward-lorenz-does-the-flap-of-a-butterflys-wings-in-brazil-
set-off-a-tornado-in-texas/

Williams, R.F. (1979). The Structure of Lorenz Attractors. Math IHES, 50, pp.101-
152

31
32