1

From linear
to non-linear

I) Description of Boussinesq equations system.
Let us consider a fluid layer placed horizontally in a gravity field. This layer is supposed to be
infinite in this plan and its thickness is noted d. The fluid can be a gas or a liquid; It is characterized
by its viscosity and its heat-storage capacity . We define its PRANDTL number

=
For ordinary liquids (water, alcohol,) 7 = , for viscous liquids (Oil, honey, ) is
greater : It reaches 1010 for the earth inner layer (manteau terrestre ??) ; Gas have diffusion
coefficients quite identical giving 7 . 0 . Mercury, which is a good heat conductor, has a coefficient
025 . 0 = . The inner part of the sun has a PRANDTL number very small.
The borders of this layer consist of two free or rigid surfaces which one fixes the temperature
or the heat flux. We consider here, the simplest case:
- Free wall: V
vertical
=
vertical
vertical
x
V

= 0 on the walls
- Free-free wall:
2
2


vertical
vertical
x
V

= 0 on the walls (unrealistic conditions in experiments)

and infinitely conducting of heat the temperature is fixed on the wall.

"As soon as" the temperature field is not uniform any more (small disturbance), it is prone to
a Archimedes force because its density is related to its temperature (state equation of the fluid).
These effects have as a characteristic time:
2
1
1

|

\
|
=
d
g
A

These forces are destabilizing and can induced fluid motions. But, this fluid is also prone to
two types of effects which are stabilizing:
The diffusion of a temperature disturbance with a characteristic time :

2
d
C
=
2

The viscosity damping of a velocity disturbance with a characteristic time :

2
d
V
=

The ratio
2
A
V C

compares the stabilizing effects over the destabilizing effects. This
dimensionless number is called RAYLEIGH number and it can be define by:

3
d g
Ra

=
A great Rayleigh number means that characteristic times of damping (c et v) are large and
thus that the disturbances are deadened weakly: They have time to grow and are at the origin of the
convective state. On the other hand, when Ra is small, the stabilizing effects can "kill" the
disturbances very quickly: the fluid remains at rest in the conductive state.

The Rayleigh number is proportional to the temperature difference applied to the fluid layer,
its a dimensionless measurement of the constraint which is applied to it.

The equations which describe the state of this system called Rayleigh-Benard System can
be written:
With:
) ( . +

Grad V
t dt
d

=
) , , , (
) , , , (
) , , , (
t z y x w
t z y x v
t z y x u
V
) , , , ( t z y x p and ) , , , ( t z y x T
- Continuity equation : 0 ) ( = + V Div
dt
d
r

[1]
- Energy equation : + = T
dt
dT
C
V
[2]
- With
V
C

= and the heat flux created by the viscous friction.

- Momentum Equation (Navier-Stokes) : V F p Grad
dt
V d
A
r r
r
+ + = ) ( [3]
- State equation of the fluid : ( ) ( )
0 0
1 T T = [4]
where is thermal dilatation coefficient.

3
The Boussinesq approximations consists to neglect:
- The fluid compressibility : 0 =
dt
d

[1] becomes 0 ) ( = V Div
r
[1]
- The heat production from viscous effects : 0 =
[2] becomes T T Grad V
t
T
= +

) ( .
v
[2]
- The Archimedes force can be written as : z g F
A
r
r
= and we can assume that
dt
V d
dt
V d
r r
0

(density variations only for the gravity forces).
[3] becomes V z g p Grad V Grad V
t
V
r
r
r r
r
+ =
|

\
|
+

) ( ) ( .
0
[3]

Theses three equations, with their boundary conditions, are the Boussinesq equations. A
stable solution is trivial solution:
0
r r
= V and ( ) b az z T + =
if ( )
1
0 T z T = = and ( )
2
T d z T = = we obtain ( )
1
1 2
T z
d
T T
z T +
|

\
|
=

Generally, we note : 0
2 1
>
|

\
|
=
d
T T

So ( ) z T z T =
1


We also obtain the hydrostatic pressure gradient: ( ) ( )
0 1 0
1 T T z g
z
P
+ =

We assume now that we disturb this stable state:
p p= ( ) t z y x , , , +
( ) z T T =
1
( ) t z y x , , , +
0
r r
= V ( ) t z y x u , , ,
r
+

4
, and u
r
are small quantities introduced in the equations:
[1] 0 ) ( = u Div
r
[1]
[2] ( )

= + +

) ( .
1
z T Grad u
t
r
[2]
[3] ( ) ( ) u z g Grad u Grad u
t
u
r r r r
r
+ + + = +

0 0
) ( ) ( . [3]

Let now rewrite these equations in a dimensionless form using as scales:

2
0
d
t =
for the time
d =
0
for the temperature
d
u

=
0
for the velocity
2
2
0 0
d
p

= for the pressure
and d for the length scale

We obtain:
[1"] 0 ) ( = u Div
r
[1]
[2"]

+ = +

z u Grad u
t
r r r
. ) ( . [2]
[3"] { } u z Ra Grad u Grad u
t
u
r r r r
r
+ + = +

) ( ) ( . [3]
where is PRANDTL number :

=
Ra is the Rayleigh number:

4
gd
Ra =


With the boundary conditions:
0 . = z u
r r
at z=0 and z=1
=0 at z=0 and z=1

5

II) Linear equations and marginal stability (linear stability)

We now assume that the quantities , and u
r
are and will stay small. Then the transport
terms ) ( . Grad u
r
and ) ( . u Grad u
r r
can be neglected for the linear stability study:

[1] 0 ) ( = u Div
r
[1]
[2]

+ =

z u
t
r r
. [2]
[3] { } u z Ra Grad
t
u
r r
r
+ + =

) ( [3]

Taking the curl of the last equation, the pressure will disappear. On the other side we now
obtain an equation for =
r
r
) (u t Ro , the vorticity. Usually its easier to take the curl twice because:

( ) ( ) u u Div Grad u t Ro t Ro
r r r
= )) ( (

The equations then become:
[1] 0 ) ( = u Div
r
(unchanged)
[2]

+ =

z u
t
r r
. (unchanged)
[3]
( )
( ) ( ) ( ) { } u z Rot Rot Ra
t
u
r r
r
+ =

. [3]

We try to find a solution for these equations as linear rolls type:

If
w
v
u
u=
r
, we find u
r
as: 0 = v and 0 =

y


The system is then reduced to:
6

[1] 0 =

z
w
x
u

[2]

+ =

w
t

[3]
( )
( )
)
`

z x
Ra u
t
u

2
. [3u]

( )
( )
)
`

+ =

2
2
.
x
Ra w
t
w

[3w]
(L System)

According the experiments, we look at solutions as straight rolls type:


( )
t ikx
z u u e e

. =
( )
t ikx
z w w e e

. =
( )
t ikx
z e e

. = (Normal modes)
where is the eigenvalue of a linear operator.
If we try to solve the problem for w and , we obtain:
[2]
|
|

\
|

+ =

2
2
2
k
z
w [II]
7
[3w]
(

|
|

\
|
+

=
|
|

\
|

. 2
2 4
2
2
2
4
4
2
2
2
Ra k w k
z
w
k
z
w
w k
z
w
[IIIw]
The functions w(z) and (z) must satisfy the boundary conditions (w(0)=0 ; w(1)=0 ; (0)=0
and (1)=0). If we take ( ) ( ) z n Sin W z w
0
= and ( ) ( ) z n Sin z
0
= , we then obtain:
[II] ( ) { } 0
0 0
2 2 2
= + + W k n [II]
[IIIw] ( ) ( ) { } 0 .
0
2
0
2
2 2 2 2 2 2
= + + +

k Ra W k n k n [IIIw]

The marginal stability is given by =0 :
[II] ( ) 0
0 0
2 2 2
= + W k n
[IIIw] ( ) 0 .
0
2
0
2
2 2 2
= + k Ra W k n

To obtain a non-trivial solution ( 0
0
et 0
0
W ), the determinant of this system must vanish.
So:
( ) 0 .
2
3
2 2 2
= + k Ra k n or
( )
2
3
2 2 2
k
k n
Ra
+
=


We can plot this set of curves in the (r,Ra) plan. They are parameterized by n :



The lower curve is obtain for n=1. The minimum of this curve corresponds to the convection
threshold. Numerically, we obtain for the critical values:
8
2214 . 2
2
=

C
k et
( )
51 . 657
4
27
4
2
3
2 2
=
+
=

C
C
C
k
k
Ra

Calculus details?
In fact, when the surfaces are not free but rigid, the functions w(z) and (z) are different and
the marginal stability curve has for minimums :
1708
C
Ra and 117 . 3
C
k

III) Lorenz attractor (non-linear approach of the problem).

This calculation, published in 1963 by LORENZ, makes it possible to reach the slightly non-
linear behaviour of the convection. In fact, this one does not behave exactly like the Lorenz model
predicts it. In reality, the non-linear effects act even with the high orders. This model however gave
birth of the deterministic chaotic system theory. It is a very good calculation exercise of the first
non-linear terms and it allows to correctly describing the bifurcation between the conductive and the
convective state.

Rewrite now the non-linear equations of the convection:

[1] 0 ) ( = u Div
r

[2]

+ = +

z u Grad u
t
r r r
. ) ( .
[3] { } u z Ra Grad u Grad u
t
u
r r r r
r
+ + = +

) ( ) ( .

where is the PRANDTL number :

=
Ra is the Rayleigh number:

4
gd
Ra =


As during the linear stability analysis, we take the curl twice and we assume for the solution a
straight roll type (
w
u
u 0 =
r
et 0 =

y
).
9

Calculus details:
If
w
u
u 0 =
r
( )
0
0
=

= =
y
u
x
v
x
w
z
u
z
v
y
w
u Rot
r
r

u
w
u

= 0
r
r

Reminder: ( ) ( )
|

\
|
+ =
2
.
.
W W
Grad W W Rot W Grad W
r r
r r r r
so ( ) ( ) ( ) ( ) ( ) W Rot W W Rot Rot W Grad W Rot
r r r r r r
= = .
then: ( ) ( ) ( ) ( ) ( ) W Rot Rot W Grad W Rot Rot
r r r r
= .
( )
( ) ( )
0
0
A
z
u
z
w
u Rot =

=
r
r
( ) ( )
( )
( ) ( )
( )
( ) ( )
2
2
2
2
1
2
2 2
0
B
z x
w
x
u
x
A
B
z
w
z x
u
z
A
u Rot Rot
=


+

=
r
r


Normal modes:
We then can express the equations using the normal modes. We chose for u, w and to take
the following functions:
( ) ( ) z Cos kx Sin u t z x u . . ) , , (
0
=
( ) ( ) z Sin kx Cos w t z x w . . ) , , (
0
=
and ( ) ( ) z Sin kx Cos t z x . . ) , , (
0
=

Remark:
The flow is bi-dimensional : Introducing the LAGRANGE stream function ( ) t z x , , , the
continuity equation is automatically verified if :
z
t z x u

= ) , , ( and
x
t z x w

+ = ) , , (
Boundary conditions:
No flux through the surfaces 0 ) , , ( =

+ =
x
t z x w
Free surface 0
) , , (
2
2
=

z z
t z x u

10
So a function like ( ) ( ) z Sin kx Sin t t z x . ). ( ) , , (
0
= is appropriate
Then ( ) ( ) ( ) ( ) z Cos kx Sin u z Cos kx Sin t
z
t z x u
u
. . . . ) ( ) , , (
0 0
0
= =

=
43 42 1

and ( ) ( ) ( ) ( ) z Sin kx Cos w z Sin kx Cos t k
x
t z x w
w
. . . . ) ( ) , , (
0 0
0
= =

+ =
3 2 1


0 ) ( = u Div
r
( ) ( ) ( ) ( ) ( ) ( ) ( ) 0 . . . . .
0 0 0 0
= + = + =

z Cos kx Cos w ku z Cos kx Cos w z Cos kx Cos ku

z
w
x
u
0
0 0
= + w ku
for the curl : ( )
0
0
=

= =
x
w
z
u
u Rot
r
r
we obtain ( ) ( ) ( ) ( ) ( ) z Sin kx Sin z Sin kx Sin kw u
x
w
z
u
. .
0 0 0
0
= + =

43 42 1

( )
( ) ( )
0
0
A
z
u
z
w
u Rot =

=
r
r
with
( ) ( )
( ) ( ) z Sin kx Sin
w
w
z Sin kx Sin
u
u

2 0 0
2 0 0
. 2
2
2 .
2

=

then
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( )
( ) ( ) ( ) 0 2 . 2 .
2
. 2 2
2
2 . 2
2
0 0 0 0 0 0 0
=
(

+
=

= z Sin kx Sin
w ku
z Sin z Cos kx Sin
w
z Sin kx Cos kx kSin
u
A

!!!
The non-linear disappears: It gives no contribution. We then obtain the same system as the
previous case (Linear analysis).
[3u]
( )
( )
)
`

z x
Ra u
t
u

2
.
[3w]
( )
( )
)
`

+ =

2
2
.
x
Ra w
t
w


On the other side, for the temperature we obtain:

+ =

w
z
w
x
u
t

Using sine functions for the normal modes, we compute at first the non-linear contributions:
( ) ( ) z Sin w
z
w
x
u

2
2
0
0
=

In fact
( ) ( ) ( ) ( ) [ ] ( ) ( ) ( ) ( ) [ ] z Cos kx Cos z Sin kx Cos w z Sin kx Sin k z Cos kx Sin u
z
w
x
u

. . . . . . . .
0 0 0 0
+ =

11

{
( ) ( ) ( ) [ ] ( ) ( ) ( ) [ ] ( ) ( ) ( ) ( ) z Sin w z Cos z Sin w z Cos z Sin kx Cos w z Sin z Cos kx Sin ku
w

2
2
. . . .
0
0
0 0
2
0 0
2
0 0
0
= = +
(
(

=
The term

+ = +

z u Grad u
t
r r r
. ) ( . generates a second harmonic in the vertical direction. Its this
harmonic which initiates the distortion of the temperature profile:

This harmonic is the only contribution of the normal modes non-linearity product. For taking
account in the equations, we have to counterbalance by a second harmonic in the temperature
expression:

( ) ( ) ( ) z Sin z Sin kx Cos t z x 2 . . . ) , , (
1 0
=

Let us recomputed the non-linear term:

( ) ( ) ( ) ( ) [ ] ( ) ( ) ( ) ( ) ( ) [ ] . 2 . 2 . . . . . . . .
1 0 0 0 0
z Cos z Cos kx Cos z Sin kx Cos w z Sin kx Sin k z Cos kx Sin u
z
w
x
u

+ =

{
( ) ( ) ( ) [ ] ( ) ( ) ( ) ( ) ( ) ( ) z Cos z Sin kx Cos w z Cos z Sin kx Cos z Sin z Cos kx Sin ku
w

2 . . 2 . . . .
0 1
2
0
2
0 0
0
+
(
(

=
[ ] ( ) ( ) ( ) ( ) ( ) z Cos z Sin kx Cos w z Cos z Sin w 2 . . 2
0 1 0 0
=
( ) ( ) ( ) ( ) z Cos z Sin kx Cos w z Sin
w


2 . . 2 2
2
0 1
0 0

=
( ) ( ) ( ) ( ) [ ] z Sin z Sin kx Cos w z Sin
w


3 . . 2
2
0 1
0 0
+
(

=
Because ( ) ( ) ( ) ( ) [ ] ( ) ( ) a Sin a Sin a Sin a Sin a Cos a Sin
3 2
2 2 1 2 = = and ( ) ( ) ( ) a Sin a Sin a Sin 3 3 4
3
=

12
The system is then the following:
[NL-I] 0 =

z
w
x
u

[NL-II]

+ =

w
z
w
x
u
t

[NL-IIIu]
( )
( )
)
`

z x
Ra u
t
u

2
.
[NL-IIIw]
( )
( )
)
`

+ =

2
2
.
x
Ra w
t
w



By injecting the expressions of the terms resulting from the normal modes in the velocity and
temperature equations then projecting on each harmonic in z (forgetting the third harmonic), we
obtain :
( ) ( ) z Cos kx Sin u t z x u . . ) , , (
0
= ( ) ( ) ( ) z Cos kx Sin k u u . .
2 2
0
+ =
( ) ( ) z Sin kx Cos w t z x w . . ) , , (
0
= ( ) ( ) ( ) z Sin kx Cos k w w . .
2 2
0
+ =
( ) ( ) ( ) z Sin z Sin kx Cos t z x 2 . . . ) , , (
1 0
= ( ) ( ) ( ) ( ) ( ) z Sin z Sin kx Cos k 2 . 4 . . .
2
1
2 2
0
+ + =

[NL-I] ( ) ( ) ( ) ( ) ( ) ( ) ( ) 0 . . . . .
0 0 0 0
= + = + =

z Cos kx Cos w ku z Cos kx Cos w z Cos kx Cos ku

z
w
x
u
0
0 0
= + w ku
[NL-II]

+ =

w
z
w
x
u
t

( ) ( ) [ ] ( ) [ ] ( ) ( ) ( ) ( ) [ ] ( ) ( ) ( ) ( ) ( ) [ ] . 2 . 2 . . . . . . . . 2 .
1 0 0 0 0
1 0
z Cos z Cos kx Cos z Sin kx Cos w z Sin kx Sin k z Cos kx Sin u z Sin
t
z Sin kx Cos
t

+ +

( ) ( ) ( ) ( ) ( ) ( ) ( ) z Sin z Sin kx Cos k z Sin kx Cos w 2 . 4 . . . . .

2
1
2 2
0 0
+ + =
( ) ( ) [ ] ( ) [ ] ( ) ( ) ( ) ( ) [ ] z Sin z Sin kx Cos w z Sin
w
z Sin
t
z Sin kx Cos
t

3 . . 2
2
2 .
0 1
0 0 1 0
+
(

( ) ( ) ( ) ( ) ( ) ( ) ( ) z Sin z Sin kx Cos k z Sin kx Cos w 2 . 4 . . . . .

2
1
2 2
0 0
+ + =
( ) ( ) ( ) ( ) ( ) ( ) ( )[ ] 0 3 4 .
2
2 . .
0 1
2
1
0 0 1 2 2
0 0 0 1
0
= +
(

+
(

+ + +

w z Sin
w
t
z Sin k w w
t
kx Cos z Sin

We then obtain for the first harmonic 1 : ( ) 0 .
2 2
0 0 0 1
0
= + + +

k w w
t
so ( )
2 2
0 0 0 1
0
.

+ + =

k w w
t

And for the second: ( ) 0 4 .
2
2
1
0 0 1
= +

w
t
so
1
2 0 0 1
4
2


=

w
t

13
[NL-IIIu]
( )
( )
)
`

z x
Ra u
t
u

2
.
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) { } z Cos kx Sin k Ra z Cos kx Sin k u z Cos kx Sin k
t
u
. . . . . .
0
2
2 2
0
2 2 0
+ = +

( ) ( ) { }
0
2
2 2
0
0 2 2
. Rak k u
t
u
k + + =

+ so ( )
( )
)
`

+
+ + =

0
2 2
2 2
0
0
.

k
Rak
k u
t
u


[NL-IIIw]
( )
( )
)
`

+ =

2
2
.
x
Ra w
t
w


( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) { } z Sin kx Cos k Ra z Sin kx Cos k w z Sin kx Cos k
t
w
. . . . . .
0
2
2
2 2
0
2 2 0
+ + = +

( ) ( ) { }
0
2
2
2 2
0
0 2 2
. Rak k w
t
w
k + =

+ so ( )
( )
)
`

+
+ + =

0
2 2
2
2 2
0
0
.

k
Rak
k w
t
w

We now only interested by the variables
0
w ,
0
et
1
.
The system becomes:
( )
( )
)
`

+
+ + =

0
2 2
2
2 2
0
0
.

k
Rak
k w
t
w

( )
2 2
0 0 0 1
0
.

+ + =

k w w
t

1
2 0 0 1
4
2


=

w
t

We use: ( )
2 2
+ = k A
( ) t A t k . .
2 2
= + =
(

=
2
0
A
w X


(
(

=
2
3
2
0
A
Ra k
Y


(

=
3
2
1
A
Ra k
Z


The equations become:
1) ( )
( )
)
`

+
+ + =

0
2 2
2
2 2
0
0
.

k
Rak
k w
t
w
( ) Y X A
A
Y
Rak
A
A
Rak
A
A
X
X
A
A
+ |

\
|
=

|
|

\
|
+ |

\
|
=

\
|

.
2
2
2
. .
2
2
3 2

so Y X
X

+ =

2) ( )
2 2
0 0 0 1
0
.

+ + =

k w w
t
Y
Ra k
A
A X
A
XZ
A
Ra k
A
Y
A
Ra k
A
|
|

\
|
|

\
|
+ |

\
|
|
|

\
|
=

|
|

\
|
2
3
2
3
2
3
2
2 2
.
2

14
Y X
A
Ra k
XZ
Y

|
|

\
|
+ =

3
2

so Y rX XZ
Y
+ =

with
( )
|
|

\
|
+
=
3
2 2
2
k
Ra k
r

3)
1
2 0 0 1
4
2


=

w
t
Z
Ra k
A
XY
A
Ra k
A
Z
A
Ra k
A
|
|

\
|
|

\
|
|
|

\
|
=

|
|

\
|
2
3
2
2
3
2
3
4
2
2
2
.

Z
A
XY
Z
|
|

\
|
=

2
4

so bZ XY
Z
=

with
( )
2 2
2
4
k
b
+
=

At the end, the system is the following:
Y X
X

+ =

Y rX XZ
Y
+ =

bZ XY
Z
=

avec
( )
|
|

\
|
+
=
3
2 2
2
k
Ra k
r

et
( )
2 2
2
4
k
b
+
=

LORENZ System
Remarks:
The linear stability theory has shown that the bifurcation between conductive-convective state
have for main characteristics:
2

=
C
k and
( )
2
3
2 2
C
C
C
k
k
Ra
+
=


for these same conditions we obtain
( ) 3
8
2
4 4
2
2
2
2 2
2
=
|
|

\
|
|

\
|
+
=
+
=

C
k
b et |

\
|
=
C
Ra
Ra
r . Adding 10 =
(PRANDTL number for air) we obtain a set of parameters called canonical parameters .

CHAPTER II: Dynamical System Analysis
LORENZ dynamical system Analysis:
Y X
X

+ =

Y rX XZ
Y
+ =

bZ XY
Z
=

15
1) Critical points finding :
The first part of the analysis consists to find the stationary solutions of the equations. We are then
finding the points like 0 =

Z Y X
, so :
0 = + Y X Y X=
0 = + Y rX XZ ( ) 0 1 = + r Z X
0 = bZ XY bZ X =
2

The second equation impose 0 = X or ( ) 1 = r Z .
If 0 = X , we obtain 0 = =X Y and
0
2
= =
b
X
Z
. The first critical point is then A(0 ;0 ;0) : Steady
state.
If ( ) 1 = r Z , we obtain ( ) = = 1 r b X (if and only if r1) ( ) = = = 1 r b X Y . We then obtain
two new points B(+ ;+ ;r-1) and C(- ;- ;r-1) if r1.

Note: The first one corresponds to the steady state, the differences with the static case are logically
null. The two others cases correspond to the convective mode, the sign of indicates the direction
of rotation of the cells (or rather of a cell in particular).
2) Critical point nature.
Point A(0 ;0 ;0) : The steady state X=Y=Z=0 is a trivial solution of the system. It is of course one
of the critical points of this dynamic system with three degrees of freedom. Let us study its linear
stability. We write:
x X + =0 Y X
X

+ =

becomes y x
x

+ =

and with the linearization : y x

x

y Y + =0 Y rX XZ
Y
+ =

becomes y rx xz
y
+ =

and with the linearization : y rx

y

z Z + =0 bZ XY
Z
=

becomes bz xy
z
=

and with the linearization : bz

z

The associated matrix of the linear system id then : [A]=
|
|

\
|

b
r
0 0
0 1
0
. Let us calculate the eigenvalues
of this matrix:
Det(A-I)= ( ) ( ) ( ) ( ) r b + + + + 1 1
2
=0
The first eigenvalue is b = (<0), the two others (corresponding to the X-Y plan) are given
by the polynomial form ( ) ( ) ( ) 0 1 1
2
= + + + r . The product des roots is (1-r). When r>1, this product
is negative. That means one of these two eigenvalues is positive and the other is negative. The
steady state is then unstable for r>1 and this, according the definition of
C
Ra Ra r= , corresponds to
the case where the convection appears Ra>Ra
C
.
16
We have p=-tr[A]=( ) + 1 >0 and q=Det[A]= ( ) r 1 .
If r<1, p>0 and q>0 : We obtain a stable node or focus.
If r>1, p>0 and q<0 : We obtain a (unstable) saddle point.
Eigenvalues study : ( ) ( ) ( ) 0 1 1
2
= + + + r
( ) ( ) ( ) ( )
43 42 1
3 2 1
?
0
2 2
1 4 1 1 4 1 + + = + =
>
r r the second term is positive if r>1 else it is negative. However in
the worst case (r=0), we obtain ( ) ( )
2
2 2
2
1 1 2 4 1 2 4 1 = + = + + = + = >0 .
Then these two eigenvalues are:
( )
2
1 +
=

and the third being : b =

Figure =f(r) for the point A(0 ;0 ;0 )

It is noticed that two of the eigenvalues are always negative (
( )
2
1 +
=

and b = ). On the
other hand, the third solution (
( )
2
1 + +
=

) is positive for r higher than 1. The disturbance goes
developing and ends up "drowning" the stationary regime. r replaces Ra here.
Thus, for a certain temperature gap, the static solution is not stable anymore, the fluid starts
moving. We find the existence of a critical Rayleigh number, beyond which the movement is
maintained. For a sufficiently high temperature gap, the convective mode is established itself
(computation shows indeed that the other solutions can be stable).
Thus when r 1, we pass from a stable state towards an unstable state: It is the passage of the
conductive state to a convective state. This steady state "forks" towards the convective state given
by another critical point: (X=Y ; Z=X
2
/b).
Points B(+ ;+ ;r-1) : With ( ) 1 = r b . This point corresponds to the state
b
r
Y X
1
= = and
1 =r Z
We pose :
17
x X + = Y X
X

+ =

becomes ( ) ( ) y x y x
x

+ = + + + =

and with the linearization :

y x
x

y Y + = Y rX XZ
Y
+ =

becomes ( )( ) ( ) ( ) y xz x z y x r z r x
y
+ = + + + + + =

1 and with the

linearization : z y x
y

( ) z r Z + = 1 bZ XY
Z
=

becomes ( )( ) ( ) bz xy y x z r b y x
z
+ + = + + + =

1 and with the

linearization : bz y x
z
+

The associated matrix of the linear system id then : [A]=
|
|

\
|

b


1 1
0
. Let us calculate the eigenvalues
of this matrix :
Det(A-I)= [ ] ( ) [ ] ( ) [ ] ( ) 0 1 2 1
2 3 2 3
= + + + =
|
|

\
|
+ + + + + + C B A r b r b b
C B A

43 42 1 43 42 1 43 42 1

Constants A, B and C are independent of : The second point C(- ;- ;r-1) will be thus as
the same nature as B. It is easier to study the eigenvalues graphically. For that we write :
[ ] ( ) [ ] ( ) [ ] 0 1 2 1
2 3
= + + + + + + r b r b b [ ] [ ] b b r b b b 2 2 1
2 3
+ = + + + + soit :
[ ]
[ ]
( )

f
b b
b b b
r =
+
+ + + +
=
2
2 1
2 3


Figure =f(r) for point B (and C).

We observe that this state is stable for r>1: (3 negatives eigenvalues) and unstable for r<1 : (2
negatives eigenvalues and one positive).

To summarize :
18
For r<1 A is Stable and B&C Unstable
For r>1 It is the opposite A becomes Unstable while B&C becomes Stable



So we observe a Stability exchange called bifurcation. Near the convection threshold we
obtain the bifurcation diagram:


CHAPTER III: Bifurcations

Note : This bifurcation is a pitchfork type bifurcation. Rayleigh-Bnard convection is a
second order phase transition, with as order parameter the velocity W. The symmetry which is
broken is the invariance with translation: The system is periodic in the plan (rolls) and no more
19
uniform, we have now a periodic solution in space for the velocity (XW
0
) and the temperature
(Y
0
).
We find the equation of the normal form of the bifurcation while eliminating from the Lorenz
system the slaves modes, i.e. those which do not have own dynamics. The linearized system at r=1
is:
[A]=
|
|
|

\
|

b 0 0
0 1 1
0

Because b>>1: Z0 and of course dt dZ then dt dZ =0=XY-bZ bZ=Y
2

Because >>1: XY
So Z and X (or Y) are the slave modes and Y (or X) are the master mode.

If we introduce the expression of two slaves modes X and Z, in the equation in Y [which is
the temperature mode due to the Archimedes force: Origin of the convection], we find:

( )
b
Y
Y r
dt
dY
3
1 =

This equation is called the amplitude equation. It describes the behavior of the convection
in the vicinity of the threshold (i.e. in the vicinity of the critical value r=1). It can be obtain from a
more subtle means and even comprise terms of spatial variation which take account of the
modulations of the convective structures:

3
2
2
A
x
A
A
dt
dA

+ =

When r, becomes >0 !! (for r25 (24.73) the eigenvalues become (-,+,+).

20
The numerical exploration of the system of the three equations, shows that for r=24.74, there
is a stability loss of the stationary regime. This instability occurs abruptly, and the system is
captured by a strange attractor which caused such an amount of passion: The LORENZ attractor.
CHAPTER IV : Transition to chaos

21

LORENZ attractor
The fractal dimension of the attractor on which the trajectories are rolled up is about 2.06.
With higher r, the system finds a limit cycle formed by an inverse Hopf bifurcation, then a new
strange attractor...

22
CHAPTER V : Chaos and dimension



Bifurcation diagram of LORENZ system.
Also let us note, that it is by exploring this dynamic system that Pomeau and Maneville
observed and understood the Type I intermittency concept, of which here some results: In the
vicinity of r=166, where a limit cycle loses its stability.

Iterated map showing the channel followed by the trajectories.
Finally let us notice, that it is because its "supercriticality" of the bifurcation, that the
Rayleigh-Bnard flow could be the subject to such detailed studies. The non-linear developments
are based on the proper modes (modes propres) of the system which are first easily accessible and
on the other hand which is not very far away from the real solution.
23

When the bifurcation is subcritical, the steady state solution abruptly loses its stability (by a
1st order transition) and the system is found in areas of the phase space very far from its basic state
and impossible to reach theoretically.

In addition, which made the Lorenz model so famous; it is its sensitivity to the initial
conditions. Lorenz had made its first simulations with five significant digits. Thinking of saving
time without harming the exactitude of its experiments, it remade them with only three digits. The
results, in spite of this tiny error, were completely different. Thus, some negligible variations of the
initial state of the system can lead to completely different states. This property is characteristic of
the chaotic states and is at the origin of all the theoretical conclusions of the study. We can illustrate
it by following simulation. Two hundred very close initial states are chosen and we look at their
evolutions in time in the phase space. If the points remain enough grouped initially, they tend to
move away more and more, so that the segment of origin is progressively transformed to a cloud
after fifteen seconds.

On the basis of these observations, Lorenz extrapolated its conclusions to the atmospheric
behaviors. The divergences between two initially close states are such that at the end of two weeks
that only the perfect knowledge of these initial states makes it possible to predict with certainty the
weather evolution. However the measurements are not sufficiently accurate to reach a such
precision, therefore weather is unpredictable beyond two weeks. In fact, the beat of a butterfly wing
can cause a sufficient disturbance to create a hurricane at the end of two weeks anywhere elsewhere
in the world. And we will never be able to know at one precise moment the location of all the
butterflies in the world. It is this image which gave its name for the "butterfly effect".


The Lorenz equations were discovered by Ed Lorenz in 1963 as a very simplified model of
convection rolls in the upper atmosphere. Later these same equations appeared in studies of lasers,
batteries, and in a simple chaotic waterwheel that can be easily built.
Lorenz found that the trajectories of this system, for certain settings, never settle down to a fixed
point, never approach a stable limit cycle, yet never diverge to infinity. What Lorenz discovered
was at the time unheard of in the mathematical community, and was largely ignored for many years.
Now this beautiful attractor is the most well known strange attractor that chaos has to offer.
A simple physical model of the Lorenz equations at work is a leaky waterwheel. A waterwheel built
from paper cups with equal sized holes in the bottom of each cup is allowed to turn freely under the
force of a steady stream of water poured into the top cup. For a slow flow of water, the water leaks
out fast enough that friction keeps the waterwheel from moving. For just a little more flow the
waterwheel will pick a direction and spin in that direction forever. If the flow is increased further
the waterwheel does not settle into a stable cycle. Instead it spins in one direction for a bit, then
slows down and start to spin the other. The waterwheel will constantly change its direction of spin,
and never in a repeating predictable manner. Here is a picture of the waterwheel. Many more
sophisticated, but similar systems have been built, and they all show the same chaotic behavior.

24