Lecture notes for 12.006J/18.353J/2.

050J, Nonlinear Dynamics: Chaos

D. H. Rothman, MIT
October 24, 2022

Contents
1 Fluid dynamics and Rayleigh-Bénard convection 1
1.1 The concept of a continuum . . . . . . . . . . . . . . . . . . . 2
1.2 Mass conservation . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Momentum conservation . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 Substantial derivative . . . . . . . . . . . . . . . . . . . 5
1.3.2 Forces on fluid particle . . . . . . . . . . . . . . . . . . 7
1.4 Nondimensionalization of Navier-Stokes equations . . . . . . . 9
1.5 Rayleigh-Bénard convection . . . . . . . . . . . . . . . . . . . 11
1.6 Rayleigh-Bénard equations . . . . . . . . . . . . . . . . . . . . 16
1.6.1 Dimensional form . . . . . . . . . . . . . . . . . . . . . 16
1.6.2 Dimensionless equations . . . . . . . . . . . . . . . . . 16
1.6.3 Bifurcation diagram . . . . . . . . . . . . . . . . . . . 17
1.6.4 Convection in the Earth . . . . . . . . . . . . . . . . . 18

1 Fluid dynamics and Rayleigh-Bénard convection

Reference: Tritton [1]

In these lectures we derive (mostly) the equations of fluid dynamics. We

then show how they may be generalized to the problem of Rayleigh-Bénard
convection—the problem of a fluid heated from below. Later we show how
the RB problem itself may be reduced to the famous Lorenz equations.

Some topics to be discussed:

• The Navier-Stokes equations

• Reynolds number

1
 • RB convection
• Rayleigh number

Thus far we have dealt almost exclusively with the temporal behavior of a
few variables.

In these lectures we digress, and discuss the evolution of a continous field in

space and time.

Aside from the central role played by research in fluid turbulence and RB
convection in the development of the theory of chaos, we have another mo-
tivation: an appreciation of a hierarchy of mathematical descriptions of dy-
namical systems, ranging from pde’s to ode’s to discrete maps.

1.1 The concept of a continuum

Real fluids are made of atoms or molecules. We could in principle write ordi-
nary differential equations for the position and momentum of each particle.
But then we’d have ∼1023 equations! The concept of a continuum allows us
to write a partial-differential equation instead.

We proceed to describe the essential assumption that makes this possible.

Consider the following macroscopic length scales in a flow:

U
l3
l1 l2

In adddition to the length scales li above, we define

• Lhydro : the smallest characteristic length scale of macroscopic motions.

2
 • the mean free path `mfp : the characteristic length scale between molecular
collisions.

Fluids may be regarded as continuous fields if

Lhydro  `mfp .
When this condition holds, the evolution of the macroscopic field may be
described by continuum mechanics, i.e., partial differential equations.

To make this idea clearer, consider a thought experiment in which we measure

the density of a fluid over a length scale ` using some particularly sensitive
device. We then move the device in the x-direction over a distance of roughly
10`.

Suppose ` ∼ L1 ∼ `mfp . Then we expect the density to vary greatly in space

as in Figure (a) below:
density

x/L1 x/L 2 x/Lhydro

(a) (b) (c)

We expect that the fluctuations in (a) should decrease as ` increases. (Statistics

tells us that these fluctuations should decrease like 1/N 1/2 , where N ∝ `3 is the average number of
molecules in a box of size `. )

On the other hand, if ` ∼ Lhydro (see (c)), variations in density should reflect
density changes due to macroscopic motions (e.g., a rising hot plume), not
merely statistical fluctuations.

Our assumption of a continuum implies that there is an intermediate scale,

` ∼ L2 , over which fluctuations are small. Thus the continuum hypothesis

3
implies a separation of scales between the molecular scale, L1 ∼ `mfp , and the
hydrodynamic scale, Lhydro .

The motion of the continuum is expressed by partial differential equations for

evolution of conserved quantities. We begin with the conservation of mass.

1.2 Mass conservation

Let 
ρ = density
of a macroscopic fluid particle
~u = velocity

Consider a volume V of fluid, fixed in space:

dS

V u

d~s is an element of the surface, |d~s| is its area, and it points in the outward
normal direction.

~u is the velocity.

The outward mass flux through the element d~s is

ρ~u · d~s.
Therefore, Z
rate of mass loss from V = ρ~u · d~s.
s
The total mass in V is Z
ρdv
V
Thus the rate of mass loss may be rewritten as
Z Z Z
d ∂ρ
− ρdv = − dv = + ρ~u · d~s
dt V V ∂t s

4
Shrinking the volume, we eliminate the volume integrals and obtain
Z 
∂ρ
= − lim ρ~u · d~s/V .
∂t V →0

Recall that the RHS above is the definition of the divergence operator. We
thus obtain
∂ρ ~ · (ρ~u)
= −∇
∂t
We see that to conserve mass, a net divergence creates a corresponding change
in density.

For incompressible fluids,

ρ ∼ constant.
(This result is not an assumption, but instead derives from the assumption that the Mach number,
the square of the ratio of the fluid velocity to the speed of sound, is much less than unity.)

Then
~ · ~u = 0.
∇
which is the equation of continuity for incompressible fluids.

1.3 Momentum conservation

We seek an expression of Newton’s second law:

d
(momentum of fluid particle) = force acting on fluid particle (1)
dt

1.3.1 Substantial derivative

We first focus on the LHS of (1).

There is a conceptual problem: d (particle momentum) cannot be given at a

dt
fixed location, because

• the momentum field itself changes with respect to time; and

5
 • fluid particle can change its momentum by flowing to a place where the
velocity is different.

To better understand this problem physically, consider how a scalar property—

the temperature T —of a fluid particle changes in time.

A small change δT is produced by small changes δt in time and δx, δy, δz in

the position of the fluid particle:
∂T ∂T ∂T ∂T
δT = δt + δx + δy + δz
∂t ∂x ∂y ∂z
Divide by δt to obtain the rate of change:
δT ∂T ∂T δx ∂T δy ∂T δz
= + + +
δt ∂t ∂x δt ∂y δt ∂z δt
In the limit δt → 0,
δx δy δz
→ ux , → uy , → uz
δt δt δt
The rate of change of T of a fluid particle is then
DT ∂T ∂T ∂T ∂T
= + ux + uy + uz
Dt ∂t ∂x ∂y ∂z
∂T ~
= + ~u · ∇T
∂t
where
D ∂ ~
= + ~u · ∇
Dt ∂t
is the substantial derivative or convective derivative operator.

Thus we see that the temperature of a fluid particle can change because

• the temperature field changes “in place” (via ∂/∂t); and

• the particle can flow to a position where the temperature is different (via
~
~u · ∇).

6
Note that the same analysis applies to vector fields such as the velocity ~u:
D~u ∂~u ~ u
= + (~u · ∇)~
Dt ∂t

Therefore the velocity ~u enters D~u/Dt in 2 ways:

• ~u changes (in place) as the fluid moves (∂/∂t)

~
• ~u governs how fast that change occurs (~u · ∇).

This dual role of velocity is the essential nonlinearity of fluid dynamics and
thus the cause of turbulent instabilities.

We can now express the rate-of-change of momentum per unit volume (i.e.,
LHS of (1)):
D~u ∂~u ~ u
ρ =ρ + ρ(~u · ∇)~
Dt ∂t
ρ is outside the differential because a fluid particle does not lose mass. Density changes thus mean
volume changes, which are irrelevant to the momentum change of that particle. Above we have
written the (rate of change of momentum) per unit volume, which need not be equal to the rate of
change of (momentum per unit volume).

1.3.2 Forces on fluid particle

To obtain the full dynamical equation, we need the RHS of

D~u
ρ = Force acting on fluid particle / unit volume.
Dt

These forces are

• body force (i.e., gravity)

• pressure
• viscous friction (internal stresses)

7
Body force: We represent the externally imposed body force (per unit volume)
by F~ .

Pressure: Fluid flows from high to low pressure. Thus

pressure force ∂p
= − in 1-D
unit volume ∂x
~
= −∇p in 3-D

Viscous friction: Viscous stresses are the source of dissipation in fluids. They
resist relative movements between fluid particles.

For example, the shear flow

y
u

is resisted more by high viscosity fluids than low viscosity fluids.

This resistance derives from molecular motions. (A nice analog is Reif’s picture of
two mail trains, one initially fast and the other initially slow, that trade mailbags.)

In the simple shear flow above, random atomistic motions result in a flux of
x-momentum in the y-direction.

In Newtonian fluids, this flux, which we call Pxy , is proportional to the ve-
locity gradient:
∂ux
Pxy = −η
∂y
where η is called the dynamic viscosity. η has units of mass/(length × time).

The shear stress can occur at any orientation. Analogous to the 1-D Newto-

8
nian condition above, we define the viscous momentum flux
∂ui
Pij = −η .
∂xj

The conservation of momentum requires that the divergence of the momen-

tum flux Pij be balanced by a change in the momentum of a fluid particle.
Loosely stated,
∂(ρui ) X ∂ X ∂2
~ · Pij = −
= −∇ Pij = η
∂t ∂x ∂x 2 ui
viscous j j j j

We thus find that

viscous force
= η∇2~u.
unit volume
(A careful derivation requires consideration of the tensorial relationship between viscous stress and
the rate of deformation.)

Newton’s second law then gives the Navier-Stokes equation for incompressible
fluids:
∂~u ~ u ~ + η∇2~u
ρ + ρ(~u · ∇)~ = −∇p + F~
∂t | {z } |{z}
body force per unit vol
stresses on fluid element per unit vol
| {z }
(mass per unit vol)×acceleration

Incompressibility arose from our negelect of compressive forces on fluid ele-

ments.

1.4 Nondimensionalization of Navier-Stokes equations

Define the characteristic length scale L and velocity scale U . We obtain the
non-dimensional quantities
x y z
x0 = , y0 = , z0 =
L L L
~u U p
~u0 = , t0 = t , p0 =
U L ρU 2

9
The dynamical equations (without body force) become
~ 0 · ~u0 = 0
∇
∂~u0 0 ~0 0 ~ 0 p0 + 1 ∇02~u0
+ (~
u · ∇ )~
u = −∇
∂t0 Re
where
ρU L
Re = Reynolds number =
η
is the dimensionless control parameter.

The Reynolds number quantifies the relative importance of the nonlinear term
to the viscous term. To see why, note the following dimensional quantities:

~ u| ∼ ρU 2
|ρ~u · ∇~ nonlinearity
L
ηU
|η∇2~u| ∼ dissipation
L2
Their ratio is
~ u|
|ρ~u · ∇~ ρU L
∼ = Reynolds number
|η∇2~u| η

High Re is associated with turbulence (i.e., nonlinearities). Low Re is asso-

ciated with laminar or creeping flows dominated by viscous friction.

Note that as long as Re remains the same, the dimensional parameters like
U and L can change but the the flow (i.e., the equation it solves) does not.
This is dynamical similarity.

An example is running vs. swimming:

   
η η
= 0.15 cm2 /sec and = 0.01 cm2 /sec
ρ air ρ water

On the other hand, comparing 100 meter world records,

104 cm
Urun ∼ = 103 cm/sec
10 sec
104 cm
Uswim ∼ ∼ 2 × 102 cm/sec
50 sec
10
Taking L ∼ 100 cm,
Re(swim) ∼ 2 × 104 and Re(run) ∼ 6 × 103
Thus for both swimming and running, Re ∼ 104 , well into the turbulent
regime. Surprisingly, despite the slower speed of swimming, Re(swim) is
somewhat greater.

Another example: bacteria swimming in water is roughly like us swimming in

molasses, since the the small size and slow speed of bacteria would correspond
to a larger and faster body in a more viscous fluid.

1.5 Rayleigh-Bénard convection

In a thermally expansive fluid, hot fluid rises.

R-B convection concerns the study of the instabilities caused by rising hot
fluid and falling cold fluid.

Typically, fluid is confined between two horizontal, heat-conducting plates:

T0
T=T 0 (cold)

pure
g d fluid conduction

T=T 0 + δ T (hot) T=T 0 + δ T

temperature

In the absence of convection—the transport of hot fluid up and cold fluid

down—the temperature gradient is constant.

Two cases of interest:

• δT small: no convective motion, due to stabilizing effects of viscous

friction.

11
 • δT large: convective motion occurs.

How large is a “large δT ” ? We seek a non-dimensional formulation.

The following fluid properties are important:

• viscosity
• density
• thermal expansivity
• thermal diffusivity (heat conductivity)

Convection is also determined by

• d, the box size

• δT (of course)

Consider a small displacement of a cold blob downwards and a hot blob

upwards:
T=T 0

T=T 0 + δ T

Left undisturbed, buoyancy forces would allow the hot blob to continue rising
and cold blob to continue falling.

There are however damping (dissipation) mechanisms:

• diffusion of heat

12
 • viscous friction

Let DT = thermal diffusivity, which has units

length2
[DT ] =
time
The temperature difference between the two blobs can therefore be main-
tained at a characteristic time scale
d2
τth ∼
DT

We also seek a characteristic time scale for buoyant displacement over the
length scale d.

Let
ρ0 = mean density
∆ρ = −αρ0 ∆T, α = expansion coefficient
Setting ∆T = δT ,
buoyancy force density = |~g ∆ρ|
= gαρ0 δT.
Note units:
mass
[gαρ0 δT ] =
(length)2 (time)2

The buoyancy force is resisted by viscous friction between the two blobs
separated by ∼ d.

The viscous friction between the two blobs diminishes like 1/d (since viscous
stresses ∝ velocity gradients). The rescaled viscosity has units
hη i mass
=
d (length)2 (time)
Dividing the rescaled viscosity by the buoyancy force, we obtain the charac-
teristic time τm for convective motion:
η/d η
τm ∼ = .
buoyancy force gαρ0 d δT
13
Convection (sustained motion) occurs if
time for motion < diffusion time for temperature difference
τm < τth
Thus convection requires
τth
> constant
τm
or
ρ0 gαd3
δT ≡ Ra > constant
ηDT
Ra is the Rayleigh number. A detailed stability calculation reveals that the
critical constant is 1708.

Our derivation of the Rayleigh number shows that the convective instability
is favored by

• large δT , α, d, ρ0 .
• small η, DT .

In other words, convection occurs when the buoyancy force ρ0 gαd3 δT exceeds
the dissipative effects of viscous drag and heat diffusion.

Note that box height enters Ra as d3 . This means that small increases in box
size can have a dramatic effect on Ra.

For Ra sufficiently large, the flow becomes turbulent. Some examples (from
Prof. Jun Zhang, NYU):

Here the gray scale is related to the thermal

gradient.

Image courtesy of Prof. Jun Zhang, NYU. Used with permission.

14
 Here the viscous flow moves the floating
boundary and the the boundary affects the
flow, an interplay roughly analogous to fluid
motions beneath tectonic plates.

Image courtesy of Prof. Jun Zhang, NYU. Used with permission.

A close-up (red is cool,

blue is warm).

Image courtesy of Prof. Jun Zhang,

NYU. Used with permission.

And here’s a picture of downgoing cold plumes (red) plunging from the upper
thermal boundary layer into the warm (blue) fluid below:

Zocchi et al. [2]

Courtesy Elsevier, Inc., http://www.sciencedirect.com. Used with permission.

15
1.6 Rayleigh-Bénard equations

1.6.1 Dimensional form

We employ the Boussinesq approximation: density perturbations affect only

the gravitational force.

The momentum equation is therefore the Navier-Stokes equation augmented

by the buoyancy force:
∂~u ~ u = − 1 ∇p
~ + ν∇2~u − ~g α(T − T0 )
+ ~u · ∇~
∂t ρ0
Here we have written the kinematic viscosity

ν = η/ρ0

The mass conservation equation is again

~ · ~u = 0.
∇

We now additionally require an equation for the convection and diffusion of

heat:
∂T
+ (~u · ∇)T = DT ∇2 T.
∂t

1.6.2 Dimensionless equations

The equations are nondimensionalized using

length scale = d
time scale = d2 /DT
temperature scale = δT /Ra.

An additional dimensionless parameter arises:

Pr = Prandtl number = ν/DT ,

which is essentially the ratio of momentum diffusion to thermal diffusion.

16
We also use the dimensionless temperature fluctuation

θ = deviation of dimensionless T from the simple conductive gradient

Dropping primes, the mass conservation equation is

~ · ~u = 0.
∇

Momentum conservation yields (ẑ is the unit vector pointing up)

 
1 ∂~u ~ u = −∇p ~ + θẑ + ∇2~u.
+ ~u · ∇~
Pr ∂t
The heat equation becomes
∂θ ~ = Ra(~u · ẑ) + ∇2 θ
+ ~u · ∇θ
∂t

Note that there are two nonlinear terms:

~u
• ~u · ∇~
~
• ~u · ∇θ

Their relative importance depends on Pr:

~ u dominates. Instabilities are “hydrodynamic.”

• small Pr ⇒ ~u · ∇~
~ dominates. Instabilities are thermally induced.
• large Pr ⇒ ~u · ∇θ

1.6.3 Bifurcation diagram

For Ra < Rac , there is no convection.

For Ra > Rac , but not too large, a regular structure of convection “rolls”
forms, with hot fluid rising and cold fluid falling:

17
 T = T0

T = T0 + δ d

Now imagine placing a probe that measures the vertical component v of

velocity, somewhere in the box midway between the top and bottom. A plot
of v(Ra) looks like
v v+
convection (stable)

0 conduction conduction Ra
rest (unstable)

Ra c v−

The transition from conduction to convection is therefore a supercritical

pitchfork bifurcation.

Note that at any particular location we cannot know in advance whether the
symmetry is broken by an upgoing or downgoing velocity.

1.6.4 Convection in the Earth

The Earth’s radius is about 6378 km. It is layered, with the main divisions
being the inner core, outer core, mantle, and crust.

The Earth’s crust—the outermost layer—is about 30 km thick.

The mantle ranges from about 30–2900 km.

The mantle is widely thought to be in a state of thermal convection. The

18
source of heat is thought to be the radioactive decay of isotopes of uranium,
thorium, and potassium. Another heat source is related to the heat deriv-
ing from the gravitational energy dissipated by the formation of the Earth
roughly 4.5 Ga.

At long time scales mantle rock is thought to flow like a fluid. However its
effective viscosity is the subject of much debate.

One might naively think that the huge viscosity would make the Rayleigh
number quite small. Recall, however, that Ra scales like d3 , where d is the
“box size”. For the mantle, d is nearly 3000 km!!!

Consequently Ra is probably quite high. Current estimates suggest that

3 × 106 . Ramantle . 109

which corresponds to roughly

103 × Rac . Ramantle . 106 Rac

The uncertainty derives principally from the viscosity, and its presumed vari-
ation by a factor of about 300 with depth.

Regardless of the uncertainty, we can conclude that Ra for the mantle is

more than sufficient for convection, and therefore that convection is likely
the driving force of plate tectonics and volcanism.

It turns out that volcanism is, over the long-term, responsible for the CO2
in the atmosphere, and thus the source of carbon that is fixed by plants.
(Weathering reactions remove C from the atmosphere.)

Thus in some sense thermal convection may be said to also sustain life.

That is, without convection, there probably would be no CO2 in the atmo-
sphere, and therefore we wouldn’t be around to discuss it...

19
References
1. Tritton, D. J. Physical Fluid Dynamics, 2nd edition (Clarendon Press,
Oxford, 1988).
2. Zocchi, G., Moses, E. & Libchaber, A. Coherent structures in turbulent
convection, an experimental study. Physica A: Statistical Mechanics and
its Applications 166, 387–407 (1990).

20
MIT OpenCourseWare
https://ocw.mit.edu

12.006J/18.353J/2.050J Nonlinear Dynamics: Chaos

Fall 2022

For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.