You are on page 1of 236

Hydrodynamics

Rami Vainio

Spring 2010

Hydrodynamics – Spring 2010


53376 Hydrodynamics
10 credit points

Lectures: Mon 10–12 and Fri 12–14 in Physicum D117


Lecturer: Rami Vainio (phone: 191 50676; email: rami.vainio@helsinki.fi)
Exercises: Thu 14–16 in Exactum C220. ∼4 problems per week. Submit to the box in 2nd floor A-wing
lobby by Tue, 6 pm.
Assistant: Heli Hietala (heli.hietala@helsinki.fi)
Student seminar: Each student has to give a seminar presentation about a topic related to the course
(20+5 min) write an essay (of 5–8 pages including figures) on it. The list of topics with links to
additional information will become available at the course website on the third week of the course;
the students should reserve their topic from the lecturer by Fri, 19.02.2010. Topics are given out
on “first come first served” basis. If you want to propose a topic of your own, contact the lecturer.
Grading: final exam 50% + exercises 25% + essay 15% + seminar presentation 10%.

Hydrodynamics – Spring 2010 1


Course outline

The course will cover the basic theory of hydrodynamics and present a number of examples especially
related to astrophysics. However, the scope of the course is general and aimed at atmospheric physicists
and geophysicicts alike.
We will follow the textbook (Ch. 1–9 in full, 14–16 selectively)
A.R. Chourdhuri: The Physics of Fluids and Plasmas – An Introduction for Astrophysicists. Cambridge
University Press, 1998
Additional material on non-astrophysical applications will be added, where appropriate.
Preliminary Outline
1. Introduction
2. Boltzmann equation
3. Derivation of the hydrodynamic equations
4. Ideal fluids
5. Viscous flows
6. Gas dynamics
7. Linear waves and instabilities
8. Turbulence
9. Rotation in hydrodynamics
10. Magnetohydrodynamics

Hydrodynamics – Spring 2010 2


Introduction

Hydrodynamics – Spring 2010 3


Scope of hydrodynamics

Hydrodynamics = theory of fluid dynamics


Fluids (fluidit):

• Liquids (nesteet)
• Gases (kaasut)

Usually plasmas (ionized gases) are treated separately from neutral fluids. However,

• hydrodynamics is a valid macroscopic theory for weakly magnetized plasmas as well.


• magnetohydrodynamics (MHD) covers the macroscopic evolution of magnetized plasmas.
• microscopic theories of neutral gases and plasmas are very different, as charged particles interact via
long-range forces.

On this course we will not address the microscopic treatment of plasmas but refer to the course Plasma
physics (5 cp, fall term) as the appropriate introduction to plasmas.
Hydrodynamics is a continuum theory: material treated at a macroscopic level without regard to the
motion of individual particles.
Intuitively clear : liquids (like water) and gases (like air) under normal conditions can be treated as
continua. But what about dilute gases like the solar wind and the interstellar medium (a few particles
per cc)? What are the necessary assumptions for the continuum hypothesis to be valid?
To shed light on that, we will derive the hydrodynamic equations from a more general theory.
Hydrodynamics – Spring 2010 4
Dynamical theories

Our aim: develop dynamical theory for fluids.


Dynamical theory: physical theory allowing the study of time evolution of a system.
Familiar examples: Classical mechanics, classical electrodynamics, quantum mechanics.

For a dynamical theory, we need a way to describe the state of the system. E.g.,

• classical mechanics: generalized coordinates qi and momenta pi ∀i = 1, ..., N


• classical electrodynamics: electric and magnetic field E(x) and B(x) ∀x
• quantum mechanics: wave function ψ(x) ∀x

Thus, the state is prescribed by giving the numerical values of a set of variables.
We also need a set of equations which tells us how the variables change with time. Once we
know the state at some instant, we shall be able to calculate the state at any other instant. E.g.,

• classical mechanics: Hamilton’s (or Lagrange’s or Newton’s) equations


• classical electrodynamics: Maxwell’s equations
• quantum mechanics: time-dependent Schrödinger equation

Thus: what are the appropriate state variables in fluid mechanics and how do they evolve in time?
The answer depends on the level of theory we need to apply.

Hydrodynamics – Spring 2010 5


Different levels of dynamical theories

Looking at a fluid consisting of N particles (with log N  1), we may study the system at different
levels of theory.
Level 0: N quantum particles. State described by ψ(x1, x2, ..., xN ), time-evolution by

∂ψ ~2 XN ∂2
i~ = Ĥψ; Ĥ = − + V ({xi})
∂t 2m i=1 ∂x2i

Level 1: N classical particles. State described by {qs, ps}, time-evolution by

∂H


q̇s =




∂ps



∂L




H({qs, ps}, t) = q̇r pr − L; L({qs, q̇s}, t) = T − V ; ps =
X

 r ∂ q̇s
∂H




ṗs = −




∂qs

Level 2: Kinetic description. State described by f (x, u), time-evolution by

∂f ∂f F (x, t)
+ ẋ · ∇f + u̇ · = C(f ); ẋ = u; u̇ =
∂t ∂u m

Level 3: Continuum description. State described by, e.g., ρ(x), T (x) and v(x), time-evolution by the
hydrodynamic equations.

Hydrodynamics – Spring 2010 6


Level 0 → Level 1

Is it always possible to replace the quantum description by the classical description? Only if particle
density is low enough so that quantum interference is unimportant.
Size of the wave packet ∼ de Broglie wavelength = λ = h/p
r √
Typical particle momentum p ∼ m κBT /m = mκBT
Interparticle distance ∼ n−1/3
Thus, λ  n−1/3 if
hn1/3 −7 n[cm−3]1/3
1√ = 4.36 · 10 r
mκBT m[amu] T [K]
Clearly, the condition is well satisfied for the dilute gases in space.
How about denser fluids? For air in standard temperature (273 K) and pressure (1013 hPa), the right-
hand side is about 0.015. For water at 20◦C it is about 0.19, so the condition is met reasonably well.
However, with liquids you need to consider the molecular size as well.
If the condition is met, then the wave packets of the particles evolve according to their respective
Schrödinger equations in an isolated manner,

i~∂ψi(xi, t)/∂t = Ĥiψi, Ĥi = −(~2/2m)∇2i + V̄ (xi)

and the expectation values hxii(t) and hpii(t) evolve according to the classical equations of
motion (Ehrenfest’s theorem).
Hydrodynamics – Spring 2010 7
Level 1 → Level 2
Level 1: prescribe the system of N classical particles with their positions,(x1, ..., xN ), and velocities,
(u1, ..., uN ). Evolution governed by Newton’s laws.
Phase space Γ of the system composed of 6N dimensions, (x1, ..., xN , u1, ..., uN ) giving the coordi-
nates. Each point in Γ corresponds to a different state of the system, evolution of the system traced by
a curve in Γ. For large N (∼ NA), solving 6N first-order (or 3N second order) equations of motion not
possible in practice. → Statistical description required.
Level 2: introduce the distribution function f (x, u, t) = density of particles in a six-dimensional
“µ-space” (x, u). Single point in Γ is mapped to N points in µ (each particle corresponding to one
point given by its position and velocity coordinates). Mapping illustrated below by an analogue of three
particles in a 6-D Γ-space (x1, x2, x3, u1, u2, u3) and 2-D µ-space (x, u) (i.e., particles in 1-D motion).
x2 u2
Γ
Instead of following
the motion of all the u µ
N points, we fol- u1
low the evolution u3
of the distribution u1
function. It is given u2
u3
by the Boltzmann
equation. x1 x3 x1 x2 x
x3

Hydrodynamics – Spring 2010 8


Level 2 → Level 3

A single-component gas in thermodynamic equilibrium fully described by two thermodynamic


variables.
A fluid in motion is not in TE as a whole.
However, a small element of fluid in its own rest frame can be in approximate TE . → State of
the fluid element prescribed by two thermodynamic variables (e.g., density ρ and temperature T ) and
velocity v .
Prescribe all fluid elements → continuum: ρ(x), T (x) and v(x) prescribe the whole system.
[Other choices often used as well, e.g., ρ(x), pressure p(x) and v(x).]
Hydrodynamic equations govern the time-evolution of these variables as a function of position and,
hence, consitute a dynamical theory of a fluid as continuum. They can be derived from the Boltzmann
equation.
Note: Each number needed to prescribe the system corresponds to one dimension of phase space of
the system. As the five variables ρ, T and vi prescribing the system need to be given in all points of
configuration space, the phase space of the hydrodynamic system has actually an infinite number of
dimensions.

Hydrodynamics – Spring 2010 9


Hamiltonian description of Level 1

Let us study the transition “Level 1 → Level 2” in more detail. Before that, let us recall the Hamiltonian
formulation of classical mechanics.
Consider a dynamical system prescribed by (qs, ps; s = 1, ..., n) and governed by the Hamiltonian
H(qs, ps, t). Thus, ∂H
ṗs = −
∂qs
∂H
q̇s = .
∂ps

For example, for a conservative system of N classical particles

H = T + V = (2m)−1 p2s + V (qs),


X

where qs are the (n = 3N ) Cartesian spatial coordinates of the particles, so


∂H
q̇s = = ps/m ⇒ ps = mq̇s = component of mechanical momentum
∂ps
∂H ∂V
ṗs = − =− Newtonian equation of motion
∂qs ∂qs

Note: generalised coordinates qs do not have to have dimensions of length. If they don’t, dimension of
ps are not those of mechanical momentum. (Always, however, [qs][ps] = [h]).
Hydrodynamics – Spring 2010 10
Ensembles in phase space

Ensemble = a set of many replicas of the same system, which are identical in all other respects apart
from being in different states at an instant of time.

⇒ Each member of an ensemble can be represented by a point in phase space Γ at an instant


of time and their evolution corresponds to different trajectories in Γ.

Ass. Ensemble points are distributed densely and smoothly in Γ

⇒ A density of ensemble points, ρens(qs, ps, t), can be defined (at any location in Γ).

Example. Consider a set of N non-interacting identical 1-D harmonic oscillators. The system
has a Hamiltonian
N 1 2 mω02 2
H= Hs, where Hs(qs, ps) := p + q = Es .
X

s=1 2m s 2 s
As the oscillators do not interact, each one conserves its total energy Es. The microstate of the system,
described by the phases of the oscillators, is unknown. A statistically representative ensemble of
this system, therefore, has the density of ensemble points given by

ω0 !N Y
N
ρens(qs, ps, t) = δ(Hs(qs, ps) − Es),
2π s=1
n
´ n
normalized so that s d qs d ps ρens = 1, i.e., interpreted as a probability density for finding the
Q

system in a given microstate (qs, ps).

Hydrodynamics – Spring 2010 11


Liouville’s theorem

Liouville’s theorem:
Dρens
= 0,
Dt
where D/Dt denotes a time-derivative along the path in phase space of any member of the
ensemble.
Proof :
Dρens ρens(qs + δqs, ps + δps, t + δt) − ρens(qs, ps, t)
= lim ,
Dt δt→0 δt
where (qs, ps) and (qs + δqs, ps + δps) represent points occupied by one member of the ensemble on the
phase-space trajectory of the system at t and t + δt, respectively. Taylor-expand:

∂ρens X ∂ρens ∂ρens


ρens(qs + δqs, ps + δps, t + δt) = ρens(qs, ps, t) + δqs + δps + δt
X

s ∂qs s ∂ps ∂t
to get

Dρens ∂ρens X ∂ρens X ∂ρens


= + q˙s + p˙s
Dt ∂t s ∂qs s ∂ps
∂ρens X ∂ X ∂ X ∂ q̇s ∂ ṗs 
 

= + (q˙sρens) + (p˙sρens) − ρens  +


∂t s ∂qs s ∂ps s ∂qs ∂ps
∂ρens X ∂ X ∂ ∂ ∂H ∂ ∂H 
 

= + (q˙sρens) + (p˙sρens) − ρens  −


X

∂t s ∂qs s ∂ps s ∂qs ∂ps


| {z
∂ps ∂qs }
=0

Hydrodynamics – Spring 2010 12


Since the number of members in the ensem- p
ble is conserved we must have, for an arbitrary
phase space volume VΓ,
ˆ ˆ
∂ vΓ
ρensdV = − ρensv Γ · ds,
∂t VΓ ∂VΓ

where v Γ = (q̇s, ṗs) is the 2n-dimensional ds


velocity vector in Γ space and ds is a surface
element (with outward normal) of the 2n − 1
dimensional hypersurface ∂VΓ bounding VΓ. q

The surface integral can be converted to volume integral, so


ˆ ˆ
∂ρens ∂ ∂
 

dV = − (q̇sρens) + (ṗsρens) dV
X

VΓ ∂t VΓ s ∂qs ∂ps

and as the volume is arbitrary, we have

∂ρens X  ∂ ∂ Dρens
 

0= + (q̇sρens) + (ṗsρens) = . 
∂t s ∂qs ∂ps Dt

Hydrodynamics – Spring 2010 13


Conservation of phase space volume

Ass. Ensemble points lie initially the phase


space volume
dn q s dn p s . p
Ass. As the system evolves in time, the ensem-
ble points move to another volume

dnqs0 dnp0s.

The number of points is conserved, i.e.,

ρensdnqs dnps = ρ0ensdnqs0 dnp0s

and since according to Liouville’s theorem


ρens = ρ0ens, we get

dnqs dnps = dnqs0 dnp0s,


q
i.e., the phase space volume occupied by the
points is conserved.

Hydrodynamics – Spring 2010 14


Distribution function in µ-space

Let us now return to the statistical description (level 2) of the fluid consisting of N similar classical
particles. We noted that

• the state of the system is represented by a point in the 6N -D phase space Γ and its time-
evolution there is traced by a curve.
• the state of the system can be mapped to a 6-D µ-space (x, u), where it is represented by N
points with coordinates (xi, ui) and its evolution traced by N curves.
The distribution function is defined as f

δN δN
f (x, u, t) = lim + , δV
δV →0 δV
where δN is the number of particles in a small µ-
space volume δV around the point (x, u).
x x+dx x
The limit δV → 0+ means that δV is very small . .
.
. . . . .
. . . . . .
. . . .
compared to the overall volume of the fluid but still u . . . . ... .. . .
. . . . . . . . δV,. δN
. . .. ... . . . . .
u+du . .. . .. .
large enough so that a large number of particles is . .. . .. . .. . . . . .. . . .
. . ... . . .
.
.
.. .
. . . . . .. . . .
.
inside the volume. Only if this limit exists, can . .
. . . . . . . .. . .. .. . . .
u . . . . . . . . . . . .
we pass from Level 1 to Level 2. .

Hydrodynamics – Spring 2010 15


Collisionless Boltzmann equation

Our N particles admit Hamiltonian description. Can we obtain an equation similar to Liouville’s equation
for the distribution function f in µ-space?
If the N particles are each governed by a Hamiltonian of form H = H(x, p, t) with p = mu, then

∂H ∂H
mu̇i = − ; mẋi = ; i = 1, 2, 3
∂xi ∂ui

and
Df ∂f ∂f ∂f
= + ẋi + u̇i = 0. (summation over i = 1, 2, 3implied!)
Dt ∂t ∂xi ∂ui
This case corresponds to N non-interacting particles moving in an external potential and the
equation is termed collisionless Boltzmann equation.
If particles of the fluid get in close distances so that they interact, this treatment does not apply. Then,
the potential energy of each particle is not only a function of the particle position but also of the distance
to the neighboring particles. In that case, the Hamiltonian formulation in µ-space becomes impossible.
(No problems arise, of course, in the Hamiltonian formulation in Γ-space.)
In neutral gases, however, the treatment of interactions is relatively easy, as the particles interact
only when at close distances, i.e., only when they collide. The effect of the collisions can be added
on the RHS of the equation. → Boltzmann equation.

Hydrodynamics – Spring 2010 16


Boltzmann equation

Hydrodynamics – Spring 2010 17


Collisions in a dilute gas

Collisionless Boltzmann equation satisfied by f (x, u, t) if interactions between particles are neglected.
Consider a dilute fluid , in which the total volume of the fluid particles is much smaller than the volume
occupied by the fluid,

na3  1; n = number density, a = radius of a particle.

Neutral particles → no long-range forces → interactions between the particles assumed to occur
only when they collide, i.e., when the separation between the particles is not much larger than d = 2a.
A particle moves freely (in a straight line) between two collisions. The average distance between two
collisions is the mean free path,
1
λ=√ .
2 nπd2
Clearly, in a dilute fluid, λ  d. Thus, binary collisions are much more common than collisions with
three or more particle interacting at the same time. Also, particle trajectories become zig-zag paths.
Collisions produce changes in f (x, u, t):
1. some particle with initial velocity u may have other velocities after collisions ⇒ δf (u) < 0
2. some particles with other initial velocities may have velocity u after collisions ⇒ δf (u) > 0
Thus,
Df 3 3
d x d u = −Cout + Cin
Dt
Hydrodynamics – Spring 2010 18
Binary collisions

Consider two particles with the same mass colliding.


Ass. Initial velocities of the particles are u and u1 and final velocities are u0 and u01.

Conservation of momentum and energy: u u1


u + u1 = u0 + u01
1
2 |u|
2
+ 12 |u1|2 = 21 |u0|2 + 12 |u01|2 u’1
In addition, assuming a central force field between
the particles, the relative velocity after the collision,
u0rel = u0 −u01, lies in the plane of the initial relative
velocity, urel = u − u1, and initial radius vector u’
between the particles, r = x − x1.

Five conditions, six unknowns (primed vectors). The sixth condition has to come from the nature of
interaction between the particles.
A statistical description of collisions sufficient. Thus, the sixth condition comes from the prescription
of the differential scattering cross-section.

Hydrodynamics – Spring 2010 19


Differential scattering cross-section

Consider two beams of similar particles colliding; densities n1 and n and velocities u1 and u,
respectively.
Particles in the second beam experience a flux of

I = |u − u1|n1

of particles from the first beam.

What is the number of collisions, δnc, per unit time and u1 u


unit volume, which deflect the particles from the second u’
dΩ
beam to the solid angle dΩ?

δnc = σ(u, u1|u0, u01) n |u − u1|n1 dΩ,


u’1
where σ(u, u1|u0, u01) is the differential scattering
cross section. Conservation of momentum and energy
and the requirement that the scattering is in the solid
angle dΩ determine the (primed) final velocities.

If molecular processes are assumed reversible, we can write for the reverse cross-section

σ(u0, u01|u, u1) = σ(u, u1|u0, u01).


Hydrodynamics – Spring 2010 20
Collision integral

Consider a beam with velocities in d3u around u. Its number density is n = f (x, u, t) d3u. Similarly,
a beam with velocities in d3u1 around u1 has a number density of n1 = f (x, u1, t) d3u1.

Thus,
δnc = σ(u, u1|u0, u01)|u − u1| f (x, u, t) f (x, u1, t) dΩ d3u d3u1
gives the number of particles scattered to the solid angle dΩ (around u0) per unit volume
and unit time.

Since Cout is the total number of collisions per unit time in µ-space volume d3x d3u, it corresponds
to ˆ ˆ
Cout = d3x d3u d3u1 dΩ σ(u, u1|u0, u01)|u − u1| f (x, u, t) f (x, u1, t).
For Cin, we need to consider reverse collisions between particles with velocities in d3u0 and d3u01, which
scatter into d3u and d3u1, respectively. Analogously,

δn0c = σ(u0, u01|u, u1)|u0 − u01| f (x, u0, t) f (x, u01, t) dΩ d3u0 d3u01.

Conservation laws ⇒ |u0 − u01| = |u − u1|. Volume occupied by the particles conserved in scattering;
thus, conservation of 2-particle phase space volume implies

d3u0 d3u01 = d3u d3u1,

provided that the scattering can be described by a potential allowing a Hamiltonian treatment.
Hydrodynamics – Spring 2010 21
Thus, assuming microscopic reversibility,

δn0c = σ(u, u1|u0, u01)|u − u1| f (x, u0, t) f (x, u01, t) dΩ d3u d3u1
ˆ ˆ
and
Cin = d3u d3x d3u1 dΩ σ(u, u1|u0, u01)|u − u1| f (x, u0, t) f (x, u01, t).

Recall that
Df 3 3
d x d u = −Cout + Cin
Dt
Hence, we finally have
ˆ ˆ
∂f F ∂f
+ u · ∇f + · = d3u1 dΩ σ(Ω)|u − u1| (f 0f10 − f f1) ≡ C[f ],
∂t m ∂u

where
f = f (x, u, t); f1 = f (x, u1, t); f 0 = f (x, u0, t); f10 = f (x, u01, t),
ẋ = u and F = mu̇ have been substituted. Here, F includes the external forces (like gravity) and
not the interparticle forces modeled by collisions.
Here, it has also been assumed that σ = σ(Ω), which holds if the differential cross section only depends
on the angle between the relative particle velocities u − u1 and u0 − u01.
Note that when performing the integrations over Ω (specifying the direction of u0) and u1, one needs
to use the conservation laws to give u0 and u01 as functions of u (fixed argument) and u1 (integration
variable).
Boltzmann equation is a non-linear integro-differential equation for the distribution function.
Hydrodynamics – Spring 2010 22
Evaluation of the collision integral
Collision integral contains an integral over u1 measured in the laboratory frame. The integral is
easier to evaluate in a coordinate system centered at u. Let us denote urel ≡ u1 − u. Thus,
ˆ ˆ
C[f ](x, u, t) = d3urel dΩ σ(θ)|urel|[f (u + u0)f (u + u01) − f (u)f (u + urel)],

where dΩ = sin θ dθ dϕ is measured in the same frame, so that the differential cross-section becomes
a function of scattering angle θ, only. In this frame, the conservation laws read

urel = u0 + u01
1
2 |urel |
2
= 12 |u0|2 + 12 |u01|2

and, therefore, 12 |urel|2 = 21 |u0 +u01|2 = 12 |u0|2 +u0 ·u01 + 12 |u01|2 ⇒ u0 ·u01 = 0, so unless u01 = 0 (head-on
collision), the angle between the particle trajectories is 90◦. Thus, also |u0rel| = |u01 −u0| = |urel|
and
u0 = |urel| cos θ(e3 cos θ + e⊥ sin θ); u01 = |urel| sin θ(e3 sin θ − e⊥ cos θ),
where e3 k urel, e⊥ = e1 cos ϕ + e2 sin ϕ and we take, e.g., e1 k urel × u, e2 = e3 × e1.
e ϕ
urel u’ u’ e1
θ u’ e
u’1 u’1 2 e3
u’1
Hydrodynamics – Spring 2010 23
Note, finally, that π − 2θ = (urel, u0rel) = θCM so cos θCM = − cos 2θ. Since

σ(θ) dΩ = σCM(θCM) dΩCM,

σ can also be evaluated in the center-of-mass (CM) frame, where it is easiest to calculate.
As urel k u measured in the CM frame, ϕ remains the same in the two coordinate systems. Thus,

σ(θ) sin θ dθ = σCM(θCM) sin θCMdθCM = σCM(θCM) d cos θCM = σCM(θCM) d cos 2θ
= σCM(θCM) 2 sin 2θ dθ = 4 cos θ σCM(θCM) sin θ dθ

giving σ(θ) = 4 cos θ σCM(π − 2θ). Clearly, the integral over Ω extends over the forward half-space,
only, and this gets mapped to the full 4π sr in the CM frame.
In summary:
ˆ ˆ π/2 ˆ 2π
C[f ] = 3
d urel dθ dϕ 2 sin 2θ σCM(π − 2θ)|urel|[f (u + u0)f (u + u01) − f (u)f (u + urel)],
0 0

u0 = |urel| cos θ(e3 cos θ + e⊥ sin θ); u01 = |urel| sin θ(e3 sin θ − e⊥ cos θ),
e3 k urel; e⊥ = e1 cos ϕ + e2 sin ϕ; e1 k urel × u, e2 = e3 × e1

Hydrodynamics – Spring 2010 24


Maxwellian distribution

Maxwellian distribution was first derived by Maxwell using the isotropy argument.
´ +∞
Consider velocities in the x-direction, distributed according to F (ux) normalized as −∞ F (ux) dux = 1.
However, as there’s nothing special about the x-direction, velocities in the y - and z -direction are also
distributed according to the same function. Thus, the probablity of finding a particle in duxduy duz is

1
f (u) duxduy duz = F (ux) dux · F (uy ) duy · F (uz ) duz
n
However, if f is an isotropic distribution, it should only depend on the magnitude u (or, e.g.,
its square). Thus,
1
f (u) := f¯(u2x + u2y + u2z ) = F (ux)F (uy )F (uz ).
n
Thus, a sum of u2i ’s in the argument of f¯ has to correspond to a product of F (ui)’s and this is possible
only if F (ux) is an exponential function of u2x, i.e.,
2 2 2 2 2
F (ux) = A1/3e−Bux ⇒ f (u) = nAe−B(ux+uy +uz ) = nAe−Bu .

Constant A (as a function of B > 0) can be found by normalizing the F -distribution to unity. To find
out the value of B , we need to apply the average kinetic energy of the particles, h 12 mu2i = 32 κBT ,
which then gives
3/2 2
 
m mu

f (u) = n  −
exp 
 

2πκBT 2κBT

Hydrodynamics – Spring 2010 25


Maxwellian as a solution of Boltzmann equation
Maxwellian is a steady-state solution of BE , as will be demonstrated below.
Consider a uniform gas without external forces. Thus, BE in steady-state reads
ˆ ˆ
0 = d3u1 dΩ σ(Ω)|u − u1| (f 0f10 − f f1).

In order for f to be a solution of this equation, we require

f f1 = f 0f10 ⇔ log f (u) + log f (u1) = log f (u0) + log f (u01). ?

Let χ(u) be a quantity conserved in collisions. Then

χ(u) + χ(u1) = χ(u0) + χ(u01).

Thus, the most general form of f satisfying (?) is

log f (u) = C0 + χr (u),


X

where χr (u) are all the independent quantities conserved in collisions. Conservation of energy
and all three components of momentum implies

log f (u) = C0 + C1u2 + C2xux + C2y uy + C2z uz = −B(u − u0)2 + log A


∴ f (u) = A exp{−B(u − u0)2} "streaming" Maxwellian

Hydrodynamics – Spring 2010 26


Boltzmann’s H-theorem

Microscopic processes at molecular level presumably reversible, but macroscopic processes not.
When a velocity distribution relaxes to a Maxwellian as a result of collisions, that is an irreversible
process. (Reverse process extremely improbable.)

When deriving BE, we assumed reversibility of microscopic processes. Yet, an equation leading to an
irreversible evolution of the distribution equation resulted. How come?

Let us construct a quantity ˆ


H= d3u f log f. (H)
H is a dimensionless quantity per unit volume (log is dimensionless.)

H -theorem (Boltzmann). If the distribution function f appearing in (H) evolves according to BE,
then H can never increase with time, i.e.,

dH
≤ 0.
dt

Proof. Differentiating (H) with respect to time


ˆ
dH ∂f
= d3u (1 + log f )
dt ∂t
Hydrodynamics – Spring 2010 27
For a uniform gas with F = 0, we have
ˆ ˆ ˆ
dH
= d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1) (1 + log f ). (1)
dt

Integration wrt. both u and u1 is performed, these dummy variables can be interchanged :
ˆ ˆ ˆ
dH
= d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1) (1 + log f1). (2)
dt
1 (1)+ 1 (2)
ˆ ˆ ˆ
dH 1
2
⇒ 2
= d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1) [2 + log(f f1)]. (3)
dt 2
Reversibility of molecular interactions assumed, so for collisions{u0, u01} → {u, u1}
ˆ ˆ ˆ
dH 1
= d3u0 d3u01 dΩ σ(Ω) |u0 − u01| (f f1 − f 0f10 ) [2 + log(f 0f10 )],
dt 2

and since |u0 − u01| = |u − u1| and d3u0 d3u01 = d3u d3u1 we have
ˆ ˆ ˆ
dH 1
=− d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1) [2 + log(f 0f10 )]. (4)
dt 2
1 1
ˆ ˆ ˆ
2 (3)+ 2 (4)
dH 1 f f1
⇒ = d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1) log 0 0 .
dt 4 f f1
For any positive α and β , we have (α − β) log(α/β) ≤ 0, which proves the theorem. 

Hydrodynamics – Spring 2010 28


Clearly, since a Maxwellian has C[f ] = 0, it also has dH/dt = 0. If we begin with a non-equilibrium distri-
bution, H keeps decreasing until the gas relaxes to the equilibrium distribution (Maxwellian)
when H attains a minimum value.
We started by assuming reversibility, but ended up with the quantity H , which has a non-symmetric
evolution in time. The arrow of time has been picked up.
H -theorem is not the same thing as the increase of entropy S . Here, H is defined only for a uniform
gas (entropy can be defined for more complicated systems); the usual definition of entropy is valid only
for systems in TE, whereas H is defined for non-equilibrium systems.
For dilute gases in TE we can, however, relate the two by

S = −κBH + constant.

Finally, we note a paradox associated with the H -theorem: suppose, at one instant of time, we reverse
all the velocities of the particles in a gas. Then, the particles will follow their original paths backwards.
If we had dH/dt < 0 before, now we should have dH/dt > 0! (Loschmidt’s paradox or reversibility
paradox.)
The resolution of the paradox lies in the assumption that the molecules before the collisions are always
uncorrelated. (This is the so-called assumption of molecular chaos, and it is hidden in the statistical
formulation of the interactions through the scattering cross-section.) We cannot, however, assume that
the molecules would be uncorrelated after the collisions. Thus, in the reverse situation, the molecules
are not uncorrelated before the collisions, and the assumptions behind the BE do not apply. So, in fact,
the arrow of time in BE is picked up because of the assumption of uncorrelated velocities
before the collisions.
Hydrodynamics – Spring 2010 29
Conservation equation

Consider a quantity χ(x, u), carried by the particles, conserved in binary collisions, i.e.,

χ + χ1 = χ0 + χ01 (X)

with the usual abbreviations. The total amount of χ per unit volume in the gas is
ˆ ˆ
f (x, u, t)
nχ(x, t) = d3u χ(x, u) f (x, u, t) = n(x, t) d3u χ(x, u) = n(x, t)hχi(x, t).
n(x, t)

Aim: find an evolution equation for nχ, when f evolves according to BE. Multiply BE by χ and integrate
over velocities: ˆ ˆ
Df 3
χ d u = C[f ] · χ d3u.
Dt
Consider, first, the RHS. Perform the same tricks as for 1 + log f in place of χ when deriving the
H -theorem to get
ˆ ˆ ˆ ˆ
1
C[f ] · χ d3u = d3u d3u1 dΩ σ(Ω) |u − u1| (f 0f10 − f f1)(χ + χ1 − χ0 − χ01)
4
´
so, because of (X), C[f ] · χ d3u = 0. Thus, for any quantity conserved in binary collisions,
ˆ ˆ
Df 3 ∂f ∂f Fi ∂f  3
 

0= χd u =  + ui + χd u
Dt ∂t ∂xi m ∂ui
Hydrodynamics – Spring 2010 30
yielding
ˆ ˆ ˆ ˆ
∂ ∂ ∂χ 3 1 ∂
0= f χ d3u + f χui d3u − ui f d u+ (f χFi) d3u
∂t ∂xi ∂xi m ∂ui
ˆ ˆ
1 ∂Fi 1 ∂χ
− f χ d3u − f Fi d3u.
m ∂ui m ∂ui

The last term on the first line = 0, because it can be converted to a surface integral at |u| → ∞,
ˆ ˆ

(f χFi) d3u = f χFiui u dΩ = 0,
∂ui |u|→∞

where it has been assumed that f tends to zero more rapidly than χFiui u increases, as u → ∞.
´
Thus, recalling that d3u Qf = nhQi for any quantity Q, we get

∂ ∂ *
∂χ + n * ∂χ + n * ∂Fi +
(nhχi) + (nhuiχi) − n ui − Fi − χ =0
∂t ∂xi ∂xi m ∂ui m ∂ui

This tells us how the density nχ = nhχi of any quantity χ, conserved in binary collisions, evolves in
time. It will be used repeteadly when deriving the hydrodynamic equations.

Hydrodynamics – Spring 2010 31


Derivation of HD equations

Hydrodynamics – Spring 2010 32


The moment equations

The equation

∂ ∂ *
∂χ + n * ∂χ + n * ∂Fi +
(nhχi) + (nhuiχi) − n ui − Fi − χ =0
∂t ∂xi ∂xi m ∂ui m ∂ui

prescribes the recipe of how to move from microscopic (molecular quantity χ) theory to macroscopic
(amount of χ per unit volume, nhχi) theory. Below, we will limit ourselves to forces independent of
velocity.

Conservation of mass. First, take χ = m. Thus,

∂ ∂
(nm) + (nmhuii) = 0, i.e.,
∂t ∂xi
∂ρ
+ ∇ · (ρv) = 0,
∂t
where ρ = nm is the mass density and v = hui is the average velocity of the particles.

Conservation of momentum. Next, consider χ = muj . Thus,

∂ ∂ *
∂uj + ∂ ∂
0 = (nmhuj i) + (nmhuiuj i) − n Fi = (ρvj ) + (ρhuiuj i) − nFj
∂t ∂xi ∂u
| {z i}
∂t ∂x i
=δij

Hydrodynamics – Spring 2010 33


Noting that u − v is the particle velocity with respect to the average flow, we define
ˆ
Pij ≡ m d3u (ui − vi)(uj − vj )f = ρh(ui − vi)(uj − vj )i

= ρ(huiuj i − huivj i − hviuj i + hvivj i)


= ρ(huiuj i − vivj − vivj + vivj ) = ρ(huiuj i − vivj ).

Thus, ρhuiuj i = Pij + ρvivj and

∂ ∂ ∂Pij ρ
(ρvj ) + (ρvivj ) = − + Fj , i.e.,
∂t ∂xi ∂xi m
∂(ρv) ρ
+ ∇ · (ρvv) = −∇ · P + F ,
∂t m
where P = Pij eiej is a second-order tensor. (This is clear, since it is defined through a dyadic product
of vectors.) Using the conservation of mass, we can simplify the LHS

∂(ρv) ∂v ∂ρ ∂v
     

+ ∇ · (ρvv) = ρ  + v · ∇v  + v  + ∇ · (ρv) = ρ  + v · ∇v 
∂t ∂t ∂t ∂t

Note: on this course, we mark the dyadic product of vectors without any symbol, just placing the vectors
side by side. Thus, ˆ
P=m d3u (u − v)(u − v)f.

Hydrodynamics – Spring 2010 34


Conservation of energy. In case the gas is mono-atomic, translational kinetic energy is conserved in
collisions, so we take χ = 21 m|u − v|2, i.e., the kinetic energy in the rest frame of the gas. Thus,

∂ ∂ ∂qi
(ρ) + (ρvi) + + Pij Λij = 0, i.e.,
∂t ∂xi ∂xi
∂(ρ)
+ ∇ · (ρv) + ∇ · q + P : Λ = 0
∂t
where
 = 12 h|u − v|2i internal energy per unit mass
q = 12 ρh(u − v)|u − v|2i energy flux
 
1 ∂vi ∂vj 
Λ = Λij eiej with Λij = 
 +  rate of strain tensor
2 ∂xj ∂xi

Again, we can use


∂ ∂ ∂
 

(ρ) + (ρvi) = ρ  + v · ∇ .


∂t ∂xi ∂t

Note: we denote the double inner product of two tensors with a colon:

A : B = Aij Bji .

In the case of P and Λ, both tensors are symmetric, so Pij = Pji and Λij = Λji. Thus,

P : Λ = Pij Λij .
Hydrodynamics – Spring 2010 35
Summary of moment equations
We have derived the moment equations (termed so because they are obtained as velocity moments
of the Boltzmann equation)

∂ρ ∂(ρvi)
+ =0
∂t ∂xi
∂vj ∂vj 1 ∂Pij Fj
+ vi =− + ; j = 1, 2, 3
∂t ∂xi ρ ∂xi m
 
∂ ∂  ∂qi 1 ∂vi ∂vj 
 

ρ  + vi =− − Pij Λij ; Λij = 


 + 
∂t ∂xi ∂xi 2 ∂xj ∂xi

so there are altogether five equations. The unknowns are velocity moments of f :
ˆ ˆ ˆ
1
ρ = m d 3 u f ; vj = d3u uj f ; Pij = m d3u (ui − vi)(uj − vj )f ;
n
ˆ ˆ
1 1
= d3u (ui − vi)(ui − vi)f ; qj = d3u (uj − vj )(ui − vi)(ui − vi)f
2n 2

so there are altogether 1+3+6+1+3=14 unknowns. Thus, the moment equations do not yet
constitute a dynamical theory of fluids.
We could, of course, introduce more moment equations by taking higher-order moments of BE. However,
they would always introduce higher order moments of the distribution function because of the term u·∇f
in BE. Thus, a way of truncating the chain has to be developed in order to obtain a dynamical theory.

Hydrodynamics – Spring 2010 36


Excursion to stellar dynamics

Boltzmann equation describes the evolution of point-like masses interacting via binary collisions. Could
BE be applied to the evolution of a large group of stars, like a cluster or a galaxy?
In short, no. Stars interact via gravity, which is a long-range force.
However, in many systems, collisions between stars are extremely rare. For galaxies, for example,
the collisional relaxation time can be estimated to be longer than the age of the universe.
If collisions are unimportant, the collisionless Boltzmann equation can be applied. In this case,
contributions from the collision integral vanish identically, and moment equations derived for the
BE apply (Jeans 1922). In this case, of course, the force F is not external, but should include the
contribution from self-gravity of the system.
The moment equations can be applied to astronomical data. As an example, consider the matter in
the neighborhood of the Sun.
Assume:
1. that the system is in a steady state and
2. that the variations of quantities is much faster in the direction perpendicular to the galactic plane
than within the directions in the plane.
Thus, the uz -moment equation gives

∂ ∂ d 1 d
0= (ρ?vz ) + (ρ?huiuz i) − n?Fz = (ρ?hu2z i) − gz ρ? ⇒ gz = (n?hu2z i)
∂t ∂xi dz n? dz

Hydrodynamics – Spring 2010 37


Oort limit

Oort (1932) estimated the gravitational field at different


heights from the galactic plane using statistics on stars and
equation
1 d
gz (z) = (n?hu2z i)
n? dz
along with the assumption hu2z i = constant.
Now, g fulfills the Poisson equation,
dgz
−4πGρ = ∇ · g = ,
dz
so the amount of matter producing the gravitational field in
the solar neighborhood can be estimated. It turns out to be

ρ ≈ (7 − 10) × 10−21 kg m−3.

Calculating the the density by estimating the amount of material in the visible stars gives

ρ? ≈ (4 − 6) × 10−21 kg m−3.

The difference,
ρ − ρ? ≈ (1 − 6) × 10−21 kg m−3,
gives, therefore, an upper limit to the density of the interstellar gas, known as the Oort limit.

Hydrodynamics – Spring 2010 38


Towards the dynamical theory of fluids

Recall the moment equations:

∂ρ ∂(ρvi)
+ =0
∂t ∂xi
∂vj ∂vj 1 ∂Pij Fj
+ vi =− + ; j = 1, 2, 3
∂t ∂xi ρ ∂xi m
∂ ∂  ∂qi
 

ρ  + vi =− − Pij Λij ,
∂t ∂xi ∂xi
 
1 ∂vi ∂vj 
where Λij =  +  and
2 ∂xj ∂xi
ˆ ˆ ˆ
1
ρ = m d 3 u f ; vj = d3u uj f ; Pij = m d3u (ui − vi)(uj − vj )f ;
n
ˆ ˆ
1 1
= d3u (ui − vi)(ui − vi)f ; qj = d3u (uj − vj )(ui − vi)(ui − vi)f
2n 2

Our task is to derive enough relations between the fourteen unknowns, ρ, vj , Pij ,  and qj so that the
system of equations becomes closed and a dynamical theory for fluids is established.

Hydrodynamics – Spring 2010 39


Zero-order approximation
Collisions are the only way of transmitting momentum and energy in a fluid consisting of neutral particles
and, thus, essential for fluid-like behaviour.
We noted that collisions drive the distribution towards a Maxwellian equilibrium distribution, with a
possible non-zero mean velocity. Thus, start by assuming that the distribution function is of form
3/2
2
  
m  m{u − v(x, t)} 
f (x, u, t) = n(x, t) 


 exp − ,
2πκBT (x, t) 2κBT (x, t)

i.e., a Maxwellian with local values of the mean velocity, density and temperature.
Start by calculating Pij (x, t)
ˆ 3/2 ˆ 
2

m  mU 

U =u−v
Pij = m d3u (ui − vi)(uj − vj )f = mn  d3U UiUj exp 
−  = p δij ,
2πκBT 2κBT

where p = nκBT is the pressure of the fluid. Likewise,


ˆ ˆ
1 3 κB T 1
= d3U U 2 f = ; qj = d3U Uj U 2f = 0,
2n 2 m 2

where  is the internal energy (per unit mass) for a monoatomic gas. Finally,

∂vi
Pij Λij = pΛii = p = p∇ · v.
∂xi
Hydrodynamics – Spring 2010 40
Thus, the zero-order moment equations can be written in the simplified form as:

∂ρ
+ ∇ · (ρv) = 0
∂t
∂v 1 F
+ v · ∇v = − ∇p +
∂t ρ m
∂
 

ρ  + v · ∇ = −p∇ · v,
∂t
i.e., as five equations for the six quantities ρ, v , p and . However, the three thermodynamic quantities
can be expressed as a function of number density and temperature, i.e., ρ = mn, p = nκBT and
 = 23 κBT /m. Thus, we conclude that the number of independent variables equals the number of
equations. We have ended up with a dynamical theory of a fluid.
This theory, however, lacks some important features of a theory for real fluids
1. Since q = 0, there is no transport of internal energy by other means than convection. I.e., heat
conduction is missing from the theory.
2. Since P is diagonal, there is no viscosity in our fluid (proof will follow later). Thus, momentum
transport inside the fluid is incomplete and relative motions of fluid layers are not damped out.
In short, our description of the fluid lacks a proper description of transport phenomena.
What went wrong in the derivation? The starting point, a local Maxwellian distribution, is too restrictive:
if there is a temperature gradient in the fluid, particles arriving to a position from the direction of
the gradient have slightly higher energies than those arriving from the opposite direction. Transport
phenomena are clearly associated with (small) departures from Maxwellian distribution.
Hydrodynamics – Spring 2010 41
Approximating C[f ]
Consider
f (x, u, t) = f (0)(x, u, t) + g(x, u, t),
where f (0) is a Maxwellian distribution constructed with the moments
ˆ ˆ ˆ
1 m
n = d3 u f ; v = d3u uf ; T = d3U U 2 f (U = u − v)
n 3nκB

and g is a small correction. Start from the collision integral


ˆ ˆ
C[f ] = d3u1 dΩ |u − u1|σ(Ω)(f 0f10 − f f1).

Assume that those distributions over which we will integrate, i.e., f 0, f 01 and f1, are all well represented
0 (0) 0 (0) 0 (0) 0 (0)
by Maxwellians f (0) , f1 and f1 , and use the property of a Maxwellian that f (0) f1 = f (0)f1 .
Thus,
ˆ ˆ ˆ ˆ
(0) (0) (0)
C[f ] ≈ d3 u 1
dΩ |u − u1|σ(Ω)(f (0)f1 − f f1 ) = (f (0) − f ) d3u1 dΩ |u − u1|σ(Ω)f1
ˆ ˆ
= −g(x, u, t) dΩ σ(Ω) d3u1|urel| f (0)(x, u1, t) = −σtot nh|urel|i g(x, u, t),

where h|urel|i(|u|, T ) is the average velocity between the colliding particles and h|urel|inσtot, thus,
represents the average scattering rate of particles of speed |u|.

Hydrodynamics – Spring 2010 42


BGK Equation

We got
C[f ] ≈ −σtot nhureli (f − f (0))
This motivates replacing collision integral by −(f − f (0))/τ , giving (Bhatnagar, Gross & Krook 1954)
(0)
∂ F ∂ f − f
 
 +u·∇+ · f = − .
∂t m ∂u τ
The BGK equation describes a distribution relaxing towards a Maxwellian with a relaxation time τ taken
to be constant.
The relaxation time can, for example, be replaced by the mean scattering time for hard-sphere molecules
(of diameter d),
1/2
1 πκ T

B
= 4nd2   ,
τ m
but actually it is a parameter of the theory.
We shall see that transport phenomena can be qualitatively described the BGK equation, but the values
of the transport coefficients are not exact. The reason lies in the fact that Boltzmann collision integral
actually demands 1/τ ∝ h|urel|i, which is a function of u. However, even if values of the coefficients
are not accurate, the model provides us with the framework under which the transport processes can be
identified.

Hydrodynamics – Spring 2010 43


Departure from Maxwellian
First, estimate the order of magnitude of the departure from Maxwellian. If the system is in an approxi-
mate steady state and there are no external forces in the system, then

|u|f (0) |g| |u|τ (0) λ (0)


≈ ⇒ |g| ≈ f ≈ f ,
L τ L L
where L is the gradient scale height of the system. Thus, departures from a Maxwellian will be small,
if mean free path  gradient scale height of the system. By using α = λ/L as a small expansion
parameter, we can write
f = f (0) + αf (1) + α2f (2) + ...
where f (i)’s are of the same order. Substituting this expansion to BGK equation allows the estimation
of the successive terms. Here, we consider only the first-order expansion:

∂ F ∂  (0)
 
(1)
g = αf = −τ  + u · ∇ + · f ,
∂t m ∂u
where, e.g.,
∂f (0) ∂n ∂f (0) ∂T ∂f (0) ∂vi ∂f (0)
= + + .
∂t ∂t ∂n ∂t ∂T ∂t ∂vi
Here, many terms can be eliminated by using the zero-order moment equations (recalling that p = nκBT ).
This gives (exercise)

1 ∂T  m 5 m 1
    

g = −τ  Ui U2 −  + Λij UiUj − δij U 2 f (0)


T ∂xi 2κBT 2 κBT 3
Hydrodynamics – Spring 2010 44
Heat flux

(0)
´ 3
Consider
´ 3now the moments calculated
´ 3 using f = f + g . Recall, that by construction n = d u f,
nv = d u uf and 3nκTT = m d U U 2f , but other moments, in particular q and the off-diagonal
elements of Pij are of interest. Let us start from the heat flux
ˆ
m
qi = d3 U U i U 2 g
2
ˆ  
m 1 ∂T  m 5 m 1
  

= −τ d3U UiU 2  Uj U2 −  + Λjk Uj Uk − δjk U 2 f (0)


2 T ∂xj 2κBT 2 κBT 3
ˆ
m 1 ∂T m 5
 

= −τ d3U Ui2 U 2  U 2 −  f (0) (no summation over i)


2 T ∂xi 2κBT 2

where odd terms of the integrand have been omitted. Since the integral has the same value for all
i = 1, 2, 3, we can replace it with third of the sum over i. Thus,

q = −K∇T
ˆ
τm m 5 5 κ2BT
 
3 4 2 (0)
K= d UU U − f = τn
 ,
6T 2κBT 2 2 m

where K is the coefficient of thermal conductivity. Thus, Fourier’s law of heat conduction is recovered.

Hydrodynamics – Spring 2010 45


P tensor
´
Consider next the P tensor, defined by Pij = m d3u UiUj f = nκBT δij + πij , where
ˆ
πij = m d3U UiUj g
ˆ
1 ∂T m 5 m 1
    

= −τ m d3U UiUj  Uk  U2 −  + Λkl Uk Ul − δkl U 2 f (0)


T ∂xk 2κBT 2 κBT 3
ˆ
τ m2 1
 

=− Λkl d3U UiUj Uk Ul − δkl U 2 f (0),


κBT 3

and odd terms have again been eliminated. Clearly, πii = 0 (integrand odd for k 6= l and terms equal
for k = l), so πij is a traceless tensor whose elements depend linearly on Λkl . Thus,

πij = −2µ(Λij − 13 δij tr Λ) = −2µ(Λij − 13 δij ∇ · v),

where the proportionality constant −2µ can be obtained, e.g., from


ˆ
τ m2 1
 

−2µΛ12 = π12 = − Λkl d3U U1U2 Uk Ul − δkl U 2 f (0)


κBT 3
ˆ
τ m2
=− (Λ12 + Λ21) d3U U12U22f (0) = −2Λ12τ nκBT
κBT

giving µ = τ nκBT . We shall see later, that µ is the coefficient of (shear) viscosity.
Hydrodynamics – Spring 2010 46
Comparison of µ with experiment

Let us first consider the value of viscosity, µ = τ nκBT , obtained by using the mean hard-sphere scattering
rate 1/2
1 πκ T

B 
= 4nd2 
τ m
as the inverse relaxation time in BGK model. Thus,

1  mκBT 1/2
 

µ= 2 .
4d π

(Note the error d ← a in Chourdhuri.) A rigorous derivation utilizing the full collision integral yields

5 πmκBT 5  mκBT 1/2
 

µ= = ,
16 σtot 16d2 π

so the BGK model underestimates the rigorous result by one part in 16, only!
Notes:

• Theory: µ independent of n. Experiment: confirmed (for gases).


Why? More molecules → more agents of momentum transport but shorter mean free path.

• Theory: µ ∝ T . Experiment: µ increases faster in gases.
Why? Higher temperature → effective radius of molecule smaller → higher µ.
Liquids: µ usually decreases with temperature.

Hydrodynamics – Spring 2010 47


Comparison of K with experiment

Heat conduction in the BGK model is


5 κB
K= µ .
2 m
3
The internal energy per unit mass is  = 2 κBT /m giving the specific heat per unit mass as

3 κB
cV = .
2m
Thus,
K 5
= .
µcV 3
A more rigorous model (full Boltzmann treatment) gives this number as 5/2 for hard-sphere molecules.
Experimental values for monoatomic inert gases are

Gas He Ne Ar Kr Xe

K
2.45 2.52 2.48 2.54 2.58
µcV

Hydrodynamics – Spring 2010 48


First-order moment equations

After now expressing Pij and qi as functions of n, T and vi, we may finally write down the moment
equations incorporating the transport phenomena. Thus, treating µ as a constant,

∂Pij ∂p ∂  1
 

= − 2µ Λij − δij ∇ · v 
∂xi ∂xj ∂xi 3
   
∂p ∂  ∂vj ∂vi  2µ ∂ ∂vk ∂p 2 1 ∂
= −µ  + + = − µ
∇ vj + ∇ · v

∂xj ∂xi ∂xi ∂xj 3 ∂xj ∂xk ∂xj 3 ∂xj
so
∂v ρF
 
 
2 1
ρ  + v · ∇v = −∇p + µ ∇ v + 3 ∇(∇ · v) +
 (P)
∂t m
Likewise,
1
 

Pij Λij = pδij Λij + πij Λij = p∇ · v − 2µ Λ : Λ − (∇ · v)2


3
giving
∂ 1
   

ρ  + v · ∇ = −p∇ · v + ∇ · (K∇T ) + 2µ Λ : Λ − (∇ · v)2 . (E)


∂t 3
Along with the conservation of mass
∂ρ
+ ∇ · (ρv) = 0 (M)
∂t
equations (P) and (E) form a closed system, i.e., a dynamical theory of fluids.

Hydrodynamics – Spring 2010 49


Hydrodynamic equations

Typically, we make some simplifications to the first order moment equations, when treating fluids

• Neglect the spatial variation of µ, as done already when deriving the equation of motion (P)
• Neglect the effects of compressibility (∇ · v ) in viscous forces in (P)
• Neglect the effect of viscous production of heat in the energy equation (E)
• Simplify the notation by writing F ← F /m.

Thus, we get the following hydrodynamic equations:

∂ρ
+ ∇ · (ρv) = 0
∂t
∂v 1 µ
+ v · ∇v = − ∇p + F + ∇2v (Navier–Stokes equation)
∂t ρ ρ
∂
 

ρ  + v · ∇ = −p∇ · v + ∇ · (K∇T )


∂t
We finally recall the starting point of our derivation, that the distribution remains close to a local
Maxwellian. This was shown to be the case when the mean free path was much shorter than the gradient
scales of the system. Thus, we expect the hydrodynamic equations to break down if this condition is not
met.

Hydrodynamics – Spring 2010 50


Ideal fluids

Hydrodynamics – Spring 2010 51


Macroscopic derivation of the HD equations

Above, we derived the hydrodynamic equations (for dilute gases) starting from molecular dynamics. Next,
we will consider the macroscopic derivation of hydrodynamic equations, treating fluids as continua, which
establishes that HD equations are valid for dense fluids as well.
We shall, again, be aiming at finding a dynamical theory for fluids. Thus, taking time derivatives is in a
key role. When treating continua, one introduces two different kinds of time derivatives:
Eulerian: This is the (partial) time derivative ∂/∂t taken at a fixed point in space, i.e., regarding x
constant.
Lagrangian: This is the (total) time derivative d/dt taken at a position following a fluid element moving
at velocity v .
Consider a quantity Q(x, t) related to the fluid (e.g., temperature, density, component of velocity). In
an infinitesimal time δt, fluid element that is at x at time t will move to x + v δt. Thus,

dQ Q(x + v δt, t + δt) − Q(x, t)


≡ lim
dt δt→0 δt
∂Q
Taylor: Q(x + v δt, t + δt) ≈ Q(x, t) + δt (x, t) + v δt · ∇Q(x, t)
∂t
dQ ∂Q
∴ = + v · ∇Q.
dt ∂t

Hydrodynamics – Spring 2010 52


Equation of continuity

Consider the amount of mass enclosed in a fixed volume V :


ˆ
MV (t) = d3x ρ(x, t).
V
ˆ ˆ
Thus, dMV d ∂ρ
= d3x ρ(x, t) V fixed
= d3 x .
dt dt V V ∂t
On the other hand, if mass is neither created nor destroyed, the temporal change of MV can only be due
to the flow of mass through the surface ∂V bounding the volume. Thus,
˛ ˆ
dMV
=− ρv · ds = − d3x ∇ · (ρv)
dt ∂V V

so ˆ
∂ρ
 

d3x  + ∇ · (ρv) = 0
V ∂t
and since V is arbitrary
∂ρ
+ ∇ · (ρv) = 0. (M)
∂t
Thus,
∂ρ dρ
0= + v · ∇ρ + ρ∇ · v ⇒ = −ρ∇ · v.
∂t dt
Clearly, the fluid is incompressible (i.e., dρ/dt = 0) iff ∇ · v = 0.
Hydrodynamics – Spring 2010 53
The equation of motion

Consider a fluid element of volume δV and, thus, mass ρ δV . The v


acceleration of the fluid element is dv/dt so Newton’s second law
applied to the fluid element states that

dv −Pij ds j e i ds
ρ δV = δF = δF body + δF surf . n
dt
The force is broken in two components:
Body force δF body = ρ δV F contains forces that affect all the δ Fbody
molecules of the fluid element; usually an external force in a
neutral fluid.
Surface force dF surf = −P · ds contains forces that are excerted on the fluid element through its
surface element ds. As both dF surf and ds are vectors, the proportionality relation requires P to
be a second-rank tensor. The total surface force acting on the element is
˛ ˆ
∂Pij 3 ∂Pij
(δFsurf )i = − Pij dsj = − d x=− δV.
∂δV δV ∂x j ∂x j

Thus, dvi ∂Pij


ρ =− + ρFi (P)
dt ∂xj
which is identical to the result from the microscopic derivation (but without explicit relation of Pij to
the fluid properties). The symmetry of Pij follows from the conservation of angular momentum.
Hydrodynamics – Spring 2010 54
Surface forces for static fluids. Euler equation

If the fluid is in static equilibrium, then it is experimentally established that

dF surf k n ⇒ dF surf = −p ds,

where p is a scalar pressure. Thus,


Pij = p δij . (SE)
Although this simple law holds for fluids at rest, it is no longer valid for fluids in motion (unless v is the
same constant everywhere).
This becomes clear if one considers a shear flow, v = y/τ ex, and takes the fluid element to have a
surface in the xz plane. If (SE) holds, then the surface force is in ey direction and it is unable to damp
out the relative motion between the fluid above and below the xz plane, contrary to experimental reality.
We will first, however, restrict our attention to fluids where (SE) holds regardless of the motional state
of the fluid. Thus, the equation of motion becomes

dvi ∂p
ρ =− + ρFi,
dt ∂xi

∂v 1
i.e., + v · ∇v = − ∇p + F .
∂t ρ

This equation is called the Euler equation and the fluids obeying it ideal fluids.
Hydrodynamics – Spring 2010 55
First law of thermodynamics

The first law of thermodynamics states

dQ = dU + p dV, where
dQ is the amount of heat added to the system
dU is the change of the internal energy of the system
p dV is the work done by the system (V is the system volume)

Thermodynamic variables can be classified to extensive and intensive. If the variable V obeys V1+2 =
V1 + V2 when combining two distinct thermodynamic systems into one, it is called extensive. If not, it’s
called intensive. Clearly, Q, U and V are extensive, but p is intensive.
Although the fluid is not in general in TE, the first law of TD can be applied to a small fluid element of
mass δm. Thus, define the extensive variables as quantities per unit mass multiplied by δm:

dQ = δm dq
dU = δm d
1 dρ
 

dV = δm d   = −δm 2
ρ ρ
p
∴ dq = d − 2 dρ
ρ
Hydrodynamics – Spring 2010 56
Energy equation
Divide dq = d − (p/ρ2) dρ by dt,
d p dρ dq
= 2 + ,
dt ρ dt dt
and use dρ/dt = −ρ∇ · v to get
d
ρ = −p∇ · v − L,
dt
where L = −ρ dq/dt is the heat loss rate per unit volume.
Heat flows from hotter to colder parts of the system at a rate proportional to the temperature difference.
Thus, the heat flux can be given as
q = −K∇T,
where K is the coefficient of thermal conductivity. Rate of heat loss by conduction of the fluid element
through its surface is, therefore,
˛ ˆ
q · ds = (∇ · q) d3x.
∂ δV δV

The corresponding heat loss rate per unit volume is

L = ∇ · q = −∇ · (K∇T )

giving
∂
 

ρ  + v · ∇ = −p∇ · v + ∇ · (K∇T ) (E)


∂t

Hydrodynamics – Spring 2010 57


Macroscopic derivation of HD – a summary
The continuum hypothesis states that matter is distributed continuously inside a body. The body, which
in HD consists of a fluid, can be broken into small fictious (fluid) elements, each having a velocity v and
carrying physical quantities like density ρ and temperature T . Continuum hypothesis yields immediately

∂ρ
+ ∇ · (ρv) = 0.
∂t

Newton II applied on fluid elements


dv
ρ = −∇ · P + ρF ,
dt
where P = Pij eiej is a second-rank tensor representing the surface forces (per unit area) excerted on
the fluid element by the fluid surrounding it. In a (local) thermal equilibrium, Pij = p δij is known
empirically to apply.
By assuming LTE and considering the first law of thermodynamics, the energy equation for the fluid
could be derived
d
ρ = −p∇ · v − ∇ · q,
dt
where  is the internal energy per unit mass and q gives the heat flux.
We need to be able to express all the fourteen variables ρ, vi, Pij ,  and qi as functions of five variables,
in order to obtain a dynamical theory for fluids. In the macroscopic theory, Pij ,  and qi have to be linked
to vi and two thermodynamic variables via constitutive equations, which are essentially empirical
relations. The relations Pij = p δij and qi = −K∂T /∂xi are examples of such relations, but depending
on a material and the physical setup, more complicated relations may have to be applied.

Hydrodynamics – Spring 2010 58


Vorticity

Consider the Euler equation for a system, where the body force is conservative, i.e., F = −∇ϕ:

∂v 1
+ v · ∇v = − ∇p − ∇ϕ.
∂t ρ

Since for an arbitrary vector field A,

A · ∇A = 12 ∇(A · A) − A × (∇ × A),

we have
∂v 1
= v × (∇ × v) − ∇p − ∇(ϕ + 12 v 2).
∂t ρ
Taking now a curl of both sides we get

∂ω ∇ρ × ∇p
= ∇ × (v × ω) + ,
∂t ρ2

where ω ≡ ∇ × v is the vorticity of the fluid.

Hydrodynamics – Spring 2010 59


Vorticity equation in incompressible fluids

For fluids with constant density (incompressible fluids), the equation for vorticity simplifies to

∂ω
= ∇ × (v × ω). (V)
∂t
The assumption incompressibility is clearly a good one for liquids like water. However, it applies very
well also for gases, provided that they move much slower than compressions spread around. The spread-
ing of compressions occurs at sound speed, so for fluid velocities much less than the speed of sound,
incomressibility is a very good approximation.
Incompressible fluids obey
∇·v =0 (I)
Any vector field can be solved, if its divergence and curl are specified. Thus, the vorticity

ω =∇×v (W)

prescribes the velocity field in an incompressible fluid and equations (V), (I) and (W) constitute a
dynamical theory of incompressible ideal fluids.

Hydrodynamics – Spring 2010 60


Redundancy of energy equation. Barotropic fluids

For incompressible ideal (real) fluids, Euler (Navier-Stokes) equation along with ∇ · v = 0 and ρ =
constant provides a closed set of equations. (Four variables, i.e., v and p, and four equations.) Thus,
for incompressible fluids, energy equation is redundant.
In many astrophysical situations, however, incompressibility is too restrictive an assumption: many such
systems
• have large density variations because of the large system size and strong gravitational fields, and/or
• exhibit flow speeds comparable to or exceeding the speed of sound.
Thus, the energy equation has to be solved together with the momentum and continuity equations. The
study of such fluids is referred to as gas dynamics.
In certain problems, however, it is possible to assume a functional relation between p and ρ:

p = p(ρ).

This a called a barotropic relation and fluids obeying it barotropic fluids. Energy equation is
redundant for barotropic fluids—the full dynamical theory now consists of the continuity equation and
Euler (Navier–Stokes) equation. As ∇ρ k ∇p, the vorticity equation for ideal barotropic fluids becomes

∂ω
= ∇ × (v × ω).
∂t

Hydrodynamics – Spring 2010 61


Hydrodynamic equations in conservative form

Equation of continuity has the form

∂(mass density)
+ ∇ · (mass flux) = 0.
∂t
It is in a standard conservative form, expressing the fact that mass in a fixed region of space can only
change because of its flux through the boundary of the region.
As established before (using the equation of continuity)

∂(ρv) dv
+ ∇ · (ρvv) = ρ = −∇ · P + ρF .
∂t dt
In the absense of (external) body forces, conservation of momentum can be stated as

∂(ρv)
+ ∇ · (ρvv + P) = 0,
∂t
where ρv is the momentum density of the fluid and T = ρvv + P can, thus, be regarded as the
momentum flux. Using this momentum equation in conservative form, it is easy to show that in
the absense of external forces, the total amount of momentum in a fluid is conserved (excerise).

Hydrodynamics – Spring 2010 62


The energy equation can also be cast into a conservative form. To do that, we first note that energy
density is a sum of kinetic and internal energy densities, i.e.,

ρ( 12 v 2 + ).

In the absense of external forces (and neglecting work done by viscous forces) we get (exercise)

∂  1 2
  
1 2
 
ρ + 2 ρv + ∇ · ρv 2 v + w − K∇T = 0,
∂t
establishing that energy flux is  
1 2
ρv + w − K∇T,
2v
where w =  + p/ρ is enthalpy per unit mass.
Clearly, by integrating the conservative energy equation over the whole system, one can establish that
the total energy of the fluid is conserved (under the assumptions used to derive the energy equation).
The conservative form of HD equations is especially useful in numerical work.

Hydrodynamics – Spring 2010 63


Hydrostatics

In hydrostatics, we are interested in finding the static equilibrium solutions of the HD equations. Thus,
we set v = 0 and ∂/∂t = 0 to get

0 = −∇p + ρF
0 = ∇ · (K∇T )

from the momentum and energy equations, respectively.


Note that we have three unknowns (ρ, p, T ) and four equations here, so the system is actually overde-
termined. This means that the momentum equation does not have a solution for an arbitrary force field
F (x).
For an incompressible (ρ constant) or a barotropic fluid with ρ = ρ(p), we can use ρ−1∇p = ∇P ,
where ˆ
dp
P(p) ≡ .
ρ
Thus, the force balance states ∇P = F . Clearly, the solution exists only if the force is conserva-
tive, i.e.,
∇P = F = −∇ϕ ⇒ P = P0 − ϕ
where P0 is a constant. For incompressible or barotropic fluids the energy equation is redundant.

Hydrodynamics – Spring 2010 64


Examples

Incompressible fluid :
ˆ
dp p
P0 − ϕ(x) = P = = ⇒ p(x) = p0 − ρ ϕ(x)
ρ ρ

Consider a homogeneous gravitational field F = −gez , i.e., ϕ = gz . Thus,

p(z) = p0 − gρz.

As we know from experience in everyday life, hydrostatic pressure increases linearly with depth (−z ) in
an incompressible fluid.
Isothermal ideal gas: consider p = RρT , where T is constant and R = κB/m. Thus,
ˆ  
dp ϕ(x) 
P0 − ϕ(x) = RT = RT ln p ⇒ p(x) = p0 exp 
− 
p RT

so for ϕ = gz
z! RT κBT
p(z) = p0 exp − , where H = =
H g mg
Here, H is the pressure scale height of an isothermal atmosphere. Note that the density obeys the
same law of stratification,
ρ = ρ0e−z/H .

Hydrodynamics – Spring 2010 65


Principle of Archimedes

Consider an incompressible fluid in static equilibrium in a homogeneous gravitational field g = −gez .


Thus,
p(z) = p0 − gρ z.
Assume that a solid body of volume V and mass M is immersed in the fluid, and that the body is at
rest with respect to the fluid. What is the total force acting on the body?
The force acting on the surface of the body resulting from the pressure of the fluid is1
˛ ˛ ˆ ˆ ˆ
∂p 3
F surf = − p ds = − (pI) · ds = − ∇ · (pI) d3x = − ∇p d3x = −ez dx
∂V ∂V V V V ∂z
ˆ
= ez gρ d3x = gρV ez = −Mfluid g
V

This states the principle of Archimedes: A body immersed in a fluid experiences a buoyant force
equal to the weight of the fluid it displaces.
In addition, the body is affected by the gravitational force M g so the resultant force acting on the body
is
R = (M − Mfluid) g.
Note that once the body body starts to move as a result of this force, the force becomes modified.
1 When transforming the surface integral into volume integral, the pressure field of the fluid is continued to the region inside the solid body.

Hydrodynamics – Spring 2010 66


Polytropic atmosphere
The assumption of constant T does not generally hold throughout the entire Earth’s atmosphere. Thus,
replace this assumption by a polytropic equation of state,

p = p0(ρ/ρ0)γ ,

where γ is the polytropic index of the gas. Isothermal state corresponds to γ = 1.


Adiabatic state corresponds to γ = cp/cV , i.e., to γ = 35 for a monoatomic gas and γ = 7
5 for a diatomic
gas. However, other values, such as γ = 1.2, are often used in practical calculations.
We have ρ = ρ0(p/p0)1/γ so ˆ
dp p0γ  p 1−1/γ
 

P= =
ρ ρ0(γ − 1) p0
Using ϕ = gz gives, thus,
p0γ  p 1−1/γ
 

+ gz = const.
ρ0(γ − 1) p0
At z = 0, take p = p0 (sea-level pressure). Thus, const. = γp0/[ρ0(γ − 1)] and
γ 1
"
z # γ−1 "
z # γ−1 p "
z#
p = p0 1 − , ρ = ρ0 1 − and T = = T0 1 −
h h Rρ h
γ p0 γ kB T0
where now h = = gives the atmospheric thickness (for γ > 1).
γ − 1 gρ0 γ − 1 mg

Hydrodynamics – Spring 2010 67


Courtesy of Windows to the Universe, http://www.windows.ucar.edu

Hydrodynamics – Spring 2010 68


Solar corona

In case of the atmosphere of the Earth, the assumption of homogeneous gravitational field is not a bad
one. However, for extended atmospheres of stars like the Sun, we have to discard this assumption.

Eclipse photos ⇒ Sun has a thin outer atmosphere


called the corona.
Corona extends to large distances from the Sun ⇒
global hydrostatic balance considered.
Corona has very little mass ⇒

GM
F =− er .
r2
Despite the relatively non-spherical appearance, the
assumption of spherical symmetry used as a starting
point. Thus,

dp GM GM p
= −ρ 2 = − 2
dr r r RT
1 d  2 dT 
 

0 = ∇ · (K∇T ) = 2 Kr
r dr dr
Hydrodynamics – Spring 2010 69
Consider, first, the energy equation. Observations tell us that corona is very hot (1–2 MK) so it is in a
plasma state. Thus, it can be shown that K ∝ T 5/2. Therefore,

d  2 5/2 dT  dT C1
 

0= r T ⇒ r2 T 5/2 = −C1 ⇒ 2
7 T 7/2 = C0 +
dr dr dr r

so, assuming that T → 0 as r → ∞, we get

r0 !2/7
T (r) = T0 .
r

This derivation assumes that nothing is heating the gas above the coronal base at r = r0, so the
temperature is obtained simply by applying heat conduction from the coronal base kept at T = T0.
Now, the momentum equation becomes

dp GM  r 2/7 p
 

=− 2
dr r r0 RT0

yielding, using p = p0 at r = r0,

7GM  r0 !5/7
  
 
p = p0 exp  − 1 .
5RT0r0 r
 
This has an asymptotic value p∞ = p0 exp − 75 (GM /RT0r0)
, which is much larger than the interstellar
pressure. Thus, Parker (1958) concluded that corona cannot be in a static eqilibrium but has to expand
as a solar wind .
Hydrodynamics – Spring 2010 70
Steady flows. Streamlines

After considering hydrostatics we turn to the next simplest HD problem, the study of steady flows.2

For that purpose, introduce the concept of a streamline.

Streamline. A streamline is defined as a line everywhere tangential to the velocity field v(x, t). Thus,

dx dy dz
dx = v(x, t) dτ ⇒ dτ = = = ,
vx vy vz

where τ is a parameter, gives the equations of the streamline, when integrated regarding t as a
constant.

Note that the path of a fluid element, obtained by integrating

dx = v(x, t) dt,

is given by a streamline in a steady flow, i.e., if ∂v/∂t = 0.

2 By steady flows we mean flows that are independent of time.

Hydrodynamics – Spring 2010 71


Example. Calculate the streamlines of the velocity field v = ω(xey − yex) + V ez .

dx dy x = R cos φ


x dx + y dy = 0 ⇒ x2 + y 2 = R2 ⇒ 

− = ⇒
ωy ωx y = R sin φ

x = R cos ω z−z

dy dφ dz φ z − z0 
 0
V

= = ⇒ = ⇒
ωx ω V ω V 
y = R sin ω z−z
V
0

The streamlines are, thus, helixes around the z -axis. This is, of course, motion consisting of rigid
√ around and translation along the z -axis (Λ = 0). The speed of each fluid element is constant
rotation
|v| = ω 2R2 + V 2.
Example. What about v = az(xey − yex + Lez ) ?

dx dy x = R cos φ



− = ⇒ x dx + y dy = 0 ⇒ 
azy azx y = R sin φ

x = R cos z−z

dy dφ dz z − z0 
 0
L

= = ⇒ φ= ; ⇒
azx az azL L y = R sin z−z0

L

Streamlines
√ still the same helixes, but now the fluid element has an accelerating speed |v| =
|az| R2 + L2. This is not rigid-body motion, as although Λxx = Λyy = Λxy = 0,

∂vz 1 ∂vz ∂vx  1 ∂vz ∂vy  1


   

Λzz = = aL; Λxz =  + = − 12 ay; Λyz =  + = 2 ax.


∂z 2 ∂x ∂z 2 ∂y ∂z
Hydrodynamics – Spring 2010 72
Bernoulli’s principle for steady flows

Consider conservative external forces, F = −∇Φ. Thus,


1
− ∇p − ∇Φ = v · ∇v = ∇( 12 v 2) − v × (∇ × v).
ρ

Take an integral along a streamline


ˆ  
1
dl · ∇( 12 v 2) − |v × (∇
 
× v) + ∇p + ∇Φ = 0

{z } ρ 
⊥dl
ˆ
1 2 dp
⇒ 2 v + P + Φ = constant P =
ρ

which states Bernoulli’s principle for steady flows.


For example, considering an incompressible fluid in homogeneous gravitational field, we get

1 2 p
2v +
ρ
+ gh = constant,

where h is the height from some horizontal level.


Let us consider some examples of the application of Bernoulli’s principle.

Hydrodynamics – Spring 2010 73


Flow in a horizontal pipe with variable cross section
Consider incompressible flow in a horizontal pipe, whence h = constant. Thus,

1 2
2v + p/ρ = constant.

Furthermore, conservation of mass implies ∇ · v = 0. Integrate the equation over an arbitrary part of
the pipe, allowing the ends of the volume to have a different cross-section. Thus,
˛
v · ds = 0.
S

The only non-zero contributions to the integral come from the ends (as v · n = 0 elsewhere), whence
ˆ ˆ
0= (v · n) ds + (v · n) ds.
S1 S2

The flow is, to a good approximation, perpendicular to the cross-sectional area of the tube, so this yields
v1A1 = v2A2,
where Ai is the area of the surface S1. Thus,

p2 − p1 = 12 ρ(v12 − v22) = 12 ρ(A22/A21 − 1)v22 A1 A2

and the pressure is, therefore, the higher the larger the
local cross-sectional area of the tube.
Hydrodynamics – Spring 2010 74
Flow through a small valve at the bottom of a tank

Consider a tank with a small valve at the bottom. Let the level A represent the free surface in the tank
and level B be the bottom level. Let the height of the free surface with respect to the bottom be H .
Bernoulli:
2 2
pA + 21 ρvA + ρgH = pB + 21 ρvB .
Consider first the case, when the vent is closed and there’s no flow. Thus,

pA + ρgH = pB,

the standard result for hydrostatic pressure.

Now, open the valve. Both pA and pB are now equal to the atmospheric
pressure. For a small valve, the velocity is extremely small at the free surface
A
(as vAAA = vBAB and AA  AB), so we can take vA = 0 to a good
approximation. Thus,

ρgH = 21 ρvB2 or vB2 = 2gH,

which is known as Torricelli’s theorem. B


Note that the velocity is independent of the direction of the outlet, so if that points upwards, it is easy
to see that according to Torricelli’s theorem the water will reach the height H again. In reality, because
of viscosity, the water will not quite reach the level of the surface.
Hydrodynamics – Spring 2010 75
Example. Use Torricelli’s theorem to estimate, how long it takes for a tank of 1 m height to get empty
through a valve in the bottom, if the diameter of the valve is 1% of the diameter of the tank.

Let H(t) be the height of the surface from the bottom of the tank. The
vertical component of velocity at the surface is dH/dt < 0. Conservation
of mass implies

dH Torricelli q
A
AA = −ABvB ≈ −AB 2gH
dt
dH AB √ √ √ AB √
⇒ √ =− 2g dt ⇒ 2( H0 − H) = 2g t(H)
H AA AA

Thus, for the tank to get empty (H = 0) it takes


v v
B
A 2H0 2m
u u
Au
u
2u
t(0) = = 100
u
t t ≈ 1 h 15 min.
AB g 9.81 m s−2

Hydrodynamics – Spring 2010 76


Kelvin’s theorem on vorticity

Consider vorticity ω = ∇ × v in an ideal fluid. Thus,

∂ω
= ∇ × (v × ω).
∂t
Circulation is defined as ˛
Γ(t) = v · dl,
C
where C is a closed material curve, i.e., a closed curve traversing of adjacent fluid elements and
following their motion. Using Stokes’s law, we can write
ˆ ˆ
Γ(S(t), t) = (∇ × v) · ds = ω · ds,
S S

where S is (any) surface enclosed by C . Thus, circulation is equal to the flux of vorticity.

Kelvin’s theorem on vorticity states that in an ideal fluid,

dΓ Γ(S1, t1) − Γ(S, t)


≡ lim = 0,
dt t1 →t t1 − t

where S1 is the surface consisting of the positions, at time t1, of those fluid elements that were on
surface S at time t.
Hydrodynamics – Spring 2010 77
Proof. We will consider a slightly more ´general theorem, stating that any solenoidal (∇ · Q = 0) vector
field, Q, fullfills dΓ/dt = 0, where Γ = S Q · ds, if Q evolves in time as

∂Q
= ∇ × (v × Q).
∂t
Calculate dΓ Γ(S1, t1) − Γ(S, t)
= lim ,
dt δt→0 δt
where t1 = t + δt. For that, consider
ˆ ˆ 
∂Q

Γ(S1, t1) = Q(x, t + δt) · ds ≈ Q(x, t) + δt (x, t) · ds


S1 S1 ∂t
ˆ ˆ
∂Q
≈ Q(x, t) · ds + δt (x, t) · ds,
S1 S ∂t

where terms of higher than first order in δt have been neglected. Thus,
ˆ ˆ ˆ 
Γ(S1, t + δt) − Γ(S, t) 1  ∂Q
≈  Q(x, t) · ds + Q(x, t) · (−ds) + (x, t) · ds

δt δt S1 S S ∂t
˛ ˆ ˆ ˆ ˆ
[]= Q(x, t) · ds − Q(x, t) · ds = (∇ · Q) d3x − Q(x, t) · ds = − Q(x, t) · ds,
∂V CC1 V CC1 CC1

where CC1 is the surface drawn by the material curve as it moves from the position of C to the position
of C1, and V is the volume enclosed by ∂V = S1 ∪ S ∪ CC1.

Hydrodynamics – Spring 2010 78


´
Calculate, next, CC1 Q · ds. Here,
ds = −δt v(x, t) × dl.
Thus,
ˆ ˆ
Q · ds = Q · (−δt v(x, t) × dl)
CC1 CC 1 S1
˛
= δt (v × Q) · dl C1
C
ˆ
= δt [∇ × (v × Q)] · ds
S

so ˆ S
dΓ ∂Q
 

=  − ∇ × (v × Q) · ds. C
dt S ∂t
v(x,t) δt
This holds for any solenoidal vector field. Since we
dl
assumed the temporal evolution of Q to satisfy

∂Q ds
= ∇ × (v × Q),
∂t
the theorem is proven.

Hydrodynamics – Spring 2010 79


Ramifications of Kelvin’s theorem – Helmholtz’s theorems

Like for any solenoidal field (cf. magnetic induction B ),


we can introduce the concept of flux tubes of vorticity.
These vortex tubes are tubes consisting of bundles of
field lines of ω . The flux of vorticity across the cross- C3
section of the tube is, therefore, constant and equal to the
circulation calculated along any closed curve surrounding ω
the vortex tube.
C2
Helmholtz’s theorems state
1. The strength of a vortex tube (i.e., the circulation
associated with it) does not vary in time
2. Fluid elements lying on a vortex line at some instant
continue to lie on that vortex line. More simply, vortex
lines move with the fluid.
3. Fluid elements initially free of vorticity remain free of
vorticity.
C1
All these theorems are corollaries of Kelvin’s theorem.
Also, because of the solenoidal property of vorticity, ∇ · ω = 0, vortex lines and tubes must appear as
closed loops, extend to infinity or start/end at solid boundaries.

Hydrodynamics – Spring 2010 80


Potential flow
Consider an incompressional ideal fluid with ω = 0 everywhere at some instant of time. Because of
Helmholtz theorems, we then know that ω = 0 everywhere at all instants of time! The fluid is said to
be irrotational .
Thus, as ∇ × v = 0 everywhere at all times, we know that v has to be expressible as a gradient of a
potential,
v = −∇φ.
Furthermore, as the fluid was assumed incompressible, we have

0 = −∇ · v = ∇ · (∇φ) = ∇2φ,

i.e., velocity potential φ fulfills the Laplace equation.


Since the normal component of velocity, vn = −en · ∇φ, has to vanish at a fixed boundary of the fluid,
the appropriate boundary condition is

∂φ
en · ∇φ = = 0.
∂n
It is well known that Laplace equation with such Neumann-type boundary conditions has a unique solution.
Notice that as potential theory (i.e., Laplace’s equation) is applied also in electro- and magnetostatic
problems, many familiar results of electrodynamics can be immediately borrowed to fluid mechanics of
irrotational flows.
Below we will consider some of the classical solutions of the irrotational, inviscid and incompressible flows
solved by potential theory.
Hydrodynamics – Spring 2010 81
Sources and sinks

In analogy with the electrostatic potential of a point charge, φES =


(q/4πε0)r−1, we express the velocity potential as

φ = (m/4π)r−1,

where r denotes the distance from the point ◦, and m is constant. Clearly,
∇2φ = 0, when r 6= 0. Thus,
m
v = −∇φ = er ,
4πr2

i.e., the flow is radially outwards from ◦, and the mass flow [kg s−1]
through any spherical surface centered around ◦ is

4πr2ρ0vr = mρ0.

Thus, the singular point ◦ is a source and m represents its strength.


A sink is a negative source (denoted by •). If a source is a point of
creation of the fluid, then sink is a point of annihilation. The flow is
directed radially inwards and the potential (for a sink of strength m) is
given by
φ = −(m/4π)r−1.
Hydrodynamics – Spring 2010 82
Flow near an infinite plate

The method of images, familiar from electrostatics, can be used


to determine the velocity potential near an infinite plate. Con-
sider a point source of strength m at x = −aex and the plate
at x = 0. The effect of the plate can be obtained by placing
an image source of the same strength at x = aex. Thus, r1 r2
m 1

1

x
φ= + m m
4π  r1 r2  O
source image
with

r1 = [(x + a)2 + y 2 + z 2]1/2 and r2 = [(x − a)2 + y 2 + z 2]1/2

being the distance from the source and the image, respectively.
Therefore,  
∂φ m dr1 m dr2 m x + a x − a
vx = − = + = + 3 
∂x 4πr12 dx 4πr22 dx r13

4π r2
Thus, at x = 0, vx = 0 and there’s no flow along the x direction. The surface of the plate is a stream
surface.

Hydrodynamics – Spring 2010 83


Flow past a cylinder

Consider the flow past a fixed infinitely long cylinder (of radius a) with its axis perpendicular to the
flow direction. At large distances from the cylinder, assume the flow to be uniform, i.e., v = −U ex.
Determine the velocity field around the cylinder.
Clearly, we have a translational symmetry (along the axis of the cylinder) in the problem. Thus, it
is two-dimensional. Furthermore, the boundary condition at the surface of the cylinder is most easily
formulated in polar coordinates (r, θ):

∂φ
= 0 at r = a.
∂r
It is, therefore, natural to make use of polar coordinates, where the Laplacian can be given as

1 ∂  ∂φ  1 ∂ 2φ
 
2
∇ φ(r, θ) = r + 2 2.
r ∂r ∂r r ∂θ
Thus, write
∂  ∂φ  ∂ 2φ
 

r r =− 2
∂r ∂r ∂θ
and separate the variables, i.e., write φ = R(r)Θ(θ). Thus,

r d  dR  1 d2 Θ
 

r = −K = − .
R dr dr Θ dθ2
Hydrodynamics – Spring 2010 84
Solve, first, the latter equation
√ √
00 −θ K θ K
Θ − KΘ = 0 ⇒ Θ = a−e + a+ e

and note that periodicity Θ(θ) = Θ(θ + 2π) has to be required. Thus, K = i n, where n is an integer,
needs to be adopted so K = −n2. Hence,

Θn(θ) = Cn cos nθ + Dn sin nθ

is the general solution for the angular part. Thus, the general solution for the Laplace equation is

φ(r, θ) = Rn(r)[Cn cos nθ + Dn sin nθ],
X

n=0

where
d dRn 
 

r r − n2Rn = 0.
dr dr

Proceeding to the radial part, we note first that n = 0 gives Θ0 = C0 and

dR0
r = B0 = constant.
dr
which satisfies the boundary condition R0(a) = 0 only for B0 = 0 ⇒ R0(r) = A0 = constant. As
constants can always be added to potentials without changing their gradients, we can set A0 = 0.
Hydrodynamics – Spring 2010 85
For n > 0, writing y = ln r we get d/dr = (dy/dr) d/dy = (1/r) d/dy ⇒ r d/dr = d/dy so

d 2 Rn Bn
2
− n2Rn = 0 ⇔ Rn = Aneny + Bne−ny = Anrn + .
dy rn

This gives
∞ Bn 
 

φ(r, θ) = r n + (Cn cos nθ + Dn sin nθ)


X

n=1 rn
and
∂φ ∞ Bn
 

= n rn−1 − n+1  (Cn cos nθ + Dn sin nθ).


X

∂r n=1 r
The inner boundary condition then gives

Bn = a2n ∀n = 1, 2, 3, ...

so 
2n

∞ a 
φ(r, θ) = rn +  (Cn cos nθ + Dn sin nθ)
X 
rn

n=1

At r → ∞, v = −U ex so φ = U x = U r cos θ there. Hence, Dn = 0 ∀n, C1 = U and Cn = 0 for


n = 2, 3, ... and finally
2
a2
 
a 
φ(r, θ) = U cos θ r +  = U x + U cos θ,

r r
with ∇ = er ∂r + eθ r−1∂θ ,
a2
v = −U ex + U 2 (cos θ er + sin θ eθ ).
r
Hydrodynamics – Spring 2010 86
Streamlines of the flow past a cylinder

Hydrodynamics – Spring 2010 87


Velocity field around a moving cylinder

We showed that for a liquid streaming uniformly around a fixed cylinder, the velocity is

a2
v = −U ex + U 2 (cos θ er + sin θ eθ )
r
Make a Galilean transformation into a coordinate system moving with constant velocity −U ex. In that
frame of reference, the fluid is at rest at r → ∞ and the cylinder moves at velocity U ex. Thus,

0 a2
v = v + U ex = U 2 (cos θ er + sin θ eθ )
r
is the fluid velocity in that frame. Here the radial distance r and the angle θ are measured from and
around the axis of the moving cylinder. Thus, |v 0| = U (a/r)2 is independent on the angle measured
around the cylinder.
This equation actually applies even if the cylinder is in a non-uniform 2-D motion (i.e., if U is a function
of time but stays perpendicular to the axis of the cylinder). This is because we rassumed the fluid to be
incompressible. An incompressible fluid has a an infinite sound speed (given by ∂p/∂ρ), implying that
the fluid reacts to the changes in the motional state of the cylinder instantaneously.

Hydrodynamics – Spring 2010 88


Equation of motion of the cylinder. d’Alembert’s paradox

The kinetic energy of the fluid per unit length of the cylinder is
ˆ ∞
2
Kfluid = 12 ρ v 0 2πr dr = 12 πρa2U 2 = 12 M 0U 2,
a

where M 0 = πρa2 is the mass of fluid displaced by a unit length of the cylider. If M is the mass of the
cylinder per unit length, then the total kinetic energy associated with the unit length of the system is

Ktot = 12 (M + M 0)U 2.

It is now possible to to deduce the equation of motion of the cylinder. If F is the force acting on the
unit length of the cylinder, then the rate of work done by the cylinder is

dU dU
F · U = (M + M 0) · U ⇒ F = (M + M 0)
dt dt
Looks just like Newton II for the cylinder in vacuum, but with M replaced by M + M 0, which is referred
to as the effective mass of the cylinder.
One also notes that if dU /dt = 0 there is no resultant force acting on the cylinder. This is contrary
to our everyday experience and known as d’Alemert’s paradox. The result arises, of course, since we
neglected viscosity from our calculation, and the fluid may flow freely on the surface of the cylinder in
our model.
Hydrodynamics – Spring 2010 89
Cylinder immersed in an ideal fluid in a gravitational field

Consider an infite cylinder immersed in a fluid with its axis perpendicular to the vertical direction. Assume
that a gravitational field of g = −gez is acting in the system.
Since the fluid is ideal, we can use the hydrostatic result for the force

F = −(M − M 0) g ez .

Assume that the cylinder is intitally at rest. Thus, U = −U ez and the equation of motion for the
cylinder is
dU
(M + M 0) = (M − M 0) g
dt
giving a result that the cylinder experiences a constant acceleration,

dU M − M0
= g.
dt M + M0
This, again, is against our everyday experience that states that a body falling in a fluid will approach
asymptotically a terminal velocity. The discrepancy is again, of course, due to the neglect of viscosity.

Hydrodynamics – Spring 2010 90


Stream function

The equation of continuity for an incompressible fluid in two dimensions gives

∂vx ∂vy
+ = 0.
∂x ∂y

In general, a dx + b dy is an exact differential iff ∂a/∂y = ∂b/∂x. Thus, vy dx − vx dy is an exact


differential, say dψ , and ˆ ˆ
ψ = dψ = (vy dx − vx dy)
is called the stream function with
∂ψ ∂ψ
= vy ; = −vx.
∂x ∂y

Recall that the streamlines are given by

dx dy
= ⇒ 0 = vy dx − vx dy = dψ,
vx vy

i.e., ψ is constant along streamlines.

Hydrodynamics – Spring 2010 91


Relation between stream function and velocity potential

If the flow is irrotational,


∂vy ∂vx 
 

0 = ∇ × v = ez  − ,
∂x ∂y
then using vx = −∂ψ/∂y and vy = ∂ψ/∂x gives

∂ 2ψ ∂ 2ψ
0= 2
+ 2 = ∇2ψ,
∂x ∂y

i.e., ψ satisfies Laplace’s equation3 just as φ in v = −∇φ. We get

∂ψ ∂φ ∂ψ ∂φ
= vy = − and = −vx =
∂x ∂y ∂y ∂x

so the functions ψ and φ fulfill the Cauchy–Riemann conditions. As

∇ψ · ∇φ = −vy vx + vxvy = 0

the curves ψ = const. (stream lines) and φ = const. (equipotential lines) always intersect at right
angles. Because of the Cauchy–Riemann conditions, the functions ψ and φ are conjugate functions,
i.e., the real and imaginary parts of a single analytic complex function, called the complex potential .

3 Functions satisying Laplace’s equation are called harmonic.

Hydrodynamics – Spring 2010 92


Complex potential

Because they are conjugate function, the stream function and the velocity potential can be combined to
a complex potential by treating the xy plane as the complex plane:

f (z) = φ(x, y) + iψ(x, y), z = x + iy

The derivative of this analytic function,

df ∂φ ∂ψ
= +i = −(vx − ivy ) ≡ −w
dz ∂x ∂x
where w is called the complex velocity. Thus, given the complex potential one can easily determine the
the complex velocity and the flow velocity.
The fact that a 2-D irrotational flow can be obtained as a derivative of an analytic complex potential is
important from the point of view of applications, because it allows us to produce solutions to complicated
flow problems from simple solutions using conformal mapping.

Hydrodynamics – Spring 2010 93


Some simple flows
Consider, first, f (z) = Az n = A(reiϕ)n = Arn(cos nϕ + i sin nϕ), where A ∈ R is a constant.
df
Thus, φ = Arn cos nϕ; ψ = Arn sin nϕ; w = − dz = −nAz n−1.
Examples:
n = 1; A = U ; φ = U r cos ϕ = U x; ψ = U r sin ϕ = U y; w = −U
rectilinear flow in the − x direction

n = 2 : φ = Ar2 cos 2ϕ = A(x2 − y 2);


n=1 n=2
2
ψ = Ar sin 2ϕ = 2xy;
vx − ivy = w = −2Az = −2A(x + iy)
⇒ vx = −2Ax; vy = 2Ay
hyperbolic flow around a rectangle

n = −1; A = U a2 : a2 a2 a2
φ = U cos ϕ; ψ = −U sin ϕ; w = U 2
r r z
a2 
 

φ+1 + φ−1 = φcyl ⇒ f (z) = U z +  = complex potential of a flow around a cylinder



z
a2 
 

ψ = U sin ϕ r −  = stream function of a flow around a cylinder



r
Hydrodynamics – Spring 2010 94
Conformal mapping

Assume that an analytic complex function

g(z) = u(x, y) + iv(x, y)

is given, and that the derivative of g is non-zero in a region of the xy plane. This function serves as
a mapping between two complex regions, i.e., (x, y) ↔ (u, v). Such a mapping is called a conformal
mapping.4
Assume that the potential φ̄(u, v) is a known solution of a simple flow problem, i.e., that it is a harmonic
function of u and v and satisfies certain boundary conditions in a region of the uv plane. Thus,

φ(x, y) = φ̄{u(x, y), v(x, y)}

is a harmonic function of x and y and represents a solution to flow problem in the region of the xy plane
that maps to the region of the flow in uv plane.
This may be applied, of course, also to the stream function. Thus, if the streamlines in the uv -plane are
given by
ψ̄(u, v) = const.
the relation
ψ(x, y) = ψ̄{u(x, y), v(x, y)} = const.
gives them in the xy -plane.
4 Note that u and v are just a transformed coordinates in the complex plane and have nothing to do with velocity.

Hydrodynamics – Spring 2010 95


Example of conformal mapping

Consider the rectilinear flow in the region u ∈ R, v ∈ [−π, π], described by

φ̄(u, v) = −V u; ψ̄(u, v) = −V v.

Consider a conformal mapping given by the analytic function g(w) as

z = g(w) = w + ew = u + iv + eu(cos v + i sin v) = u + eu cos v + i(v + eu sin v)





 x = u + eu cos v
∴
  y = v + eu sin v

The stream lines in the uv plane are given by v = const. Thus, in xy plane the streamlines are obtained
in parametric form by fixing the value of v and regarding u as a parameter in the equations of the
conformal mapping.
E.g., take v = 0 ⇒ y = 0 and x = u + eu, i.e., the x axis. v = ±π gives y = ±π and x = u − eu, i.e.,
a horizontal flow line but with x → −∞ with both u → ±∞. As ∂x/∂u = 1 − eu = 0 ⇔ u = 0 ⇔
x = −1, the streamlines thus make a 180◦ turn at x = −1. At other values of v , we get

x u + eu cos v
= → cot v, as u → ∞,
y v + eu sin v

i.e., the streamlines become straight lines at large u. This solution describes the flow from a channel.
Hydrodynamics – Spring 2010 96
The equipotential lines are obtained by keeping u fixed and letting v run from −π to π in



 x = u + eu cos v
y = v + eu sin v


For −u  1 they are just vertical lines at x = u between y = −π and π . At u  1, they become
circles.

v y
π π

u x
−π
−π

Hydrodynamics – Spring 2010 97


Viscous flows

Hydrodynamics – Spring 2010 98


Newtonian fluids

In our microscopic derivation of the pressure tensor, we found

Pij = p δij + πij ,

where πij was found to be a traceless tensor linearly dependent on Λij = 21 (∂ivj + ∂j vi).

In ideal fluids, we assumed πij = 0. This assumption was pointed out on several occasions to be erroneous
and lead to unphysical results. Derive, next, the physical result for πij from macroscopic considerations.

Internal friction of a fluid, called viscosity, opposes relative motion of


adjacent layers of fluid . For a velocity field with shear, like the one in y
the Figure to the right, we know that the slower-moving fluid below the
xz plane will act on the faster moving fluid above the xz plane (and vice
versa) to slow down (speed up) its motion. The force has to act across
the xz plane (normal in the y direction), but the force has to be in the x
direction to level out the speed differences. As the surface force through
dS = dS ey is
x
(dFsurf )i = −p dSi − πij dSj ,
we note that the x component of the force has to be due to πxy . The shear
force is expected to be larger for a larger velocity gradient.
Hydrodynamics – Spring 2010 99
Newton postulated that the shear force is proportional to the velocity gradient, i.e.,

∂vx
πxy = −µ , (N)
∂y

where µ is the coefficient of viscosity. Fluids obeying the proportionality relation between the shear stress
and the velocity gradient are known as Newtonian fluids.

Shear stress in the general situation has to be more complicated than the simple relation above. This
can be seen, e.g., by considering a fluid rotating rigidly. Consider, for example, a body rotating with
angular velocity Ω = Ωez :
v = Ω × x = Ω(xey − yex)
and −µ∂y vx = µ∂xvy = µΩ, so clearly the expression (N) is non-zero although there is no shear involved
with rigid rotation. However, µ(∂y vx + ∂xvy ) = 0. In general, we can write
   
∂vi 1 ∂vi ∂vj  1  ∂vi ∂vj  1
= 
 + +  −  = Λij + ijk ωk
2
∂xj 2 ∂xj ∂xi 2 ∂xj ∂xi

where ωk = krs∂r vs is the vorticity. Here, Λij is related to the rate of deformation and 12 ijk ωk to
rotation. Thus, the shear stress of a Newtonian fluid should depend on the symmetric part of the
velocity gradient, Λij , rather than the velocity gradient itself. The most general second rank tensor
depending linearly on symmetric combinations of velocity gradients is
 
∂vi ∂vj  ∂vk
a
 + +b δij = 2aΛij + bΛkk δij .
∂xj ∂xi ∂xk
Hydrodynamics – Spring 2010 100
Therefore, a Newtonian fluid is a fluid obeying the constitutive equation
 
∂vk  ∂vi ∂vj 2 ∂vk 
 
 
Pij = p δij − 2µ Λij − 13 δij Λkk − ζΛkk δij = p − ζ δij − µ 
 + − δij 
∂xk ∂xj ∂xi 3 ∂xk
i.e., P = p̄I + π where p̄ = p − ζtr Λ and π = −2µ(Λ − 13 I tr Λ)

Here π is a traceless tensor corresponding to viscous stress due to velocity shear. The coefficient µ is the
dynamic viscosity (or shear viscocity or viscosity) already obtained through our microscopic derivation.
The first term, i.e., p̄I, is the spherical part of the P tensor. Here

1 dρ
p̄ = 31 Pkk = p − ζ∇ · v = p + ζ .
ρ dt

The coefficient ζ is called the bulk viscosity (or volume viscosity or second viscosity). It is clearly related
to viscous forces involving compressive motions. Bulk viscosity is difficult to determine experimentally,
and important only in situations when compression is important (e.g., for shock waves and damping of
sound waves).
According to our microscopic derivation, bulk viscocity vanishes for dilute gases, but large values are
experimentally obtaind for many liquids (Bhatia & Singh).

Liquid Glycerol Water Methyl alcohol Toluene Benzene


ζ/µ 1.1 2.5 3.2 13 100

Hydrodynamics – Spring 2010 101


Navier–Stokes equation

The surface force density for P = p̄I + π, where p̄ = p − ζtr Λ and π = −2µ(Λ − 13 I tr Λ), is
 
∂Pij ∂p ∂ ∂  ∂vi ∂vj 2 ∂vk 
− =− +ζ (∇ · v) + µ  + − δij 
∂xj ∂xi ∂xi ∂xj ∂xj ∂xi 3 ∂xk
∂p ∂
=− + µ∇2vi + (ζ + 13 µ) (∇ · v)
∂xi ∂xi

i.e., −∇ · P = −∇p + (ζ + 31 µ)∇(∇ · v) + µ∇2v . Here, the viscocity coefficients are assumed to be
constants. Thus, the equation of motion for a Newtonian fluid is

dv
ρ = ρF − ∇p + (ζ + 13 µ)∇(∇ · v) + µ∇2v.
dt
This is, of course, the Navier–Stokes equation. The only difference with respect to the microscopic
result derived for dilute gases is the inclusion of the bulk viscocity term.
As noted already in connection with the microscopic derivation, for many problems it is a reasonable
approximation to neglect compressibility effects. Hence, a simpler version of the Navier–Stokes equation
is obtained as
∂v 1
+ v · ∇v = F − ∇p + ν∇2v,
∂t ρ
where ν = µ/ρ is called the kinematic viscosity.

Hydrodynamics – Spring 2010 102


Navier–Stokes vs. Euler equation

The N–S equation in form


∂v 1
+ v · ∇v = F − ∇p + ν∇2v
∂t ρ
differs from Euler equation only by virtue of one extra term, ν∇2v , but this changes everything!
The additional term contains second derivatives in spatial variables, whereas the Euler equation only
contains first-order terms. Thus, more boundary conditions are required to solve the N–S equation than
the Euler equation.
For Euler equation, one takes the normal component of the velocity to vanish at fixed solid boundaries
and obtains a unique solution. It is impossible, in general, to impose more boundary conditions, which
led to the several unphysical results highlighted in the previous section.
The N–S equation with the second derivatives now allows us to impose the viscous-fluid boundary
condition, v = 0 at the solid surfaces, which allows one, for example, to consider the effect of friction
on bodies immersed in fluids.
But why do we actually impose v = 0 at solid surfaces? Experience tells us that this is the way to go.
For example, blades of a fan collect dust. It is also very difficult to sweep surfaces clean from fine dust
by blowing from one’s mouth.

Hydrodynamics – Spring 2010 103


Vorticity equation for Newtonian fluids

Taking a curl of the N–S equation gives

∂ω
= ∇ × (v × ω) + ν∇2ω
∂t
for incompressible or barotropic fluids with conservative external forces. Kelvin’s theorem on vorticity
does not apply anymore and instead we get
ˆ  ˆ ˛
dΓ ∂ω

=  − ∇ × (v × ω) · ds = ν (∇2ω) · ds = ν (∇2v) · dl


dt S ∂t S C

where C = ∂S and the last form holds only for incompressible fluids.
Thus, a finite viscosity allows the production and decay of vorticity. This allows one to overcome some
problems that are associated with ideal fluids. For example, if a stick is moved through a fluid, like water,
we can clearly see the generation of vortices behind the moving stick. This can only be understood if the
fluid has non-zero viscosity.
For incompressible viscous fluids, energy equation is redundant. A dynamical theory of the fluid is provided
by the vorticity equation, definition of vorticity, ω = ∇ × v , and the continuity equation, ∇ · v = 0.

Hydrodynamics – Spring 2010 104


Navier–Stokes equation in curvilinear coordinates

In many flow problems, the symmetries of the boundaries or solid bodies immersed in the flow suggest the
use of some curvilinear cooridnates system, e.g., cylindrical or spherical coordinates. Thus, it is necessary
to express the terms of the N–S equation in these coordinate systems.
Differential operators acting on vectors and tensors are quite complicated in curviliear cooridnates. The
N–S equation contains the terms v · ∇v and ν∇2v , which need special care. It is usually easiest to use
the identities

v · ∇v = 12 ∇v 2 − v × (∇ × v)
∇2v = ∇(∇ · v) − ∇ × (∇ × v)

before writing down the N–S equation in component form.


In cylindrical coordinates, for example, one gets
∂vr vθ2 ∂ 2 vr 1 ∂ 2 vr ∂ 2 vr 1 ∂vr
 
∂vr ∂vr vθ ∂vr 1 ∂p 2 ∂vθ vr 
+ vr + + vz − = Fr − +ν 2 + 2 2 + + − − 2
∂t ∂r r ∂θ ∂z r ρ ∂r ∂r r ∂θ ∂z 2 r ∂r r2 ∂θ r
∂ 2v 1 ∂ 2v ∂ 2 vθ 1 ∂vθ
 
∂vθ ∂v v ∂v ∂v v vr 1 ∂p 2 ∂vr vθ 
+ vr θ + θ θ + vz θ − θ = Fθ − + ν  2θ + 2 2θ + + + − 2
∂t ∂r r ∂θ ∂z r ρr ∂θ ∂r r ∂θ ∂z 2 r ∂r r2 ∂θ r
∂ 2 vz 1 ∂ 2 vz ∂ 2 vz 1 ∂vz 
 
∂vz ∂vz vθ ∂vz ∂vz 1 ∂p
+ vr + + vz = Fz − +ν 2 + 2 2 + + .
∂t ∂r r ∂θ ∂z ρ ∂z ∂r r ∂θ ∂z 2 r ∂r

Hydrodynamics – Spring 2010 105


Flow through a circular pipe

Consider the steady-state flow in a straight pipe with circular cross-section (radius a). Take the flow to
be incompressible and external forces to vanish.

Take the flow velocity to be along the axis of the tube, i.e., v = vez . For symmetry reasons v cannot
depend on the angle measured around the cylider and because of the incompressibility (uniform cross
section), it cannot depend on z .

Thus, v = v(r)ez and


dv ∂v ∂
= + v · ∇v = v (v ez ) = 0
dt ∂t ∂z
Therefore,
µ ∂  ∂v 
 

∇p = µ∇2v = r ez .
r ∂r ∂r

Since ∇p k ez , pressure is a function z , only. Since v depends on r, only, we have

dp µ d  dv 
 

= r ,
dz r dr dr
which is of course possible only if both sides are equal to a constant. From the left-hand side, this
constant is seen to be (p2 − p1)/l = −∆p/l, where p1 and p2 are the pressure at z = 0 and z = l,
respectively.
Hydrodynamics – Spring 2010 106
We get
d  dv  ∆p dv ∆p 2 ∆p 2
 

r = − r ⇒ r = c1 − r ⇒ v = c0 + c1 ln r − r,
dr dr µl dr 2µl 4µl
where ci are constants of integration. As the velocity must remain finite at r = 0, c1 = 0, and because
v = 0 at r = a, we get
∆p 2 ∆p 2
c0 = a ⇒ v= (a − r2)
4µl 4µl
The maximum speed is, therefore, vmax = ∆p a2/4µl. The average speed is
´a ´a 2 2
v 2πr dr v max 0 r(a − r )dr vmax
vav = ´ a ´a
0 z
= 2 =
0 2πr dr a 0 r dr 2

and the rate of mass flow through the cross-sectional area of the tube, or the discharge, is
ˆ a
ρvmax πa2 πa4∆p
Q= ρvz 2πr dr = = .
0 2 8νl

Thus, the discharge is proportional to the fourth power of the radius of the tube. This result is known
as the Hagen–Poiseuille formula.
The Hagen–Poiseuille formula provides means to determine the viscosity of a fluid by measuring Q vs.
∆p.
Note that the result is experimentally valid only up to a certain velocity vmax after which the flow becomes
turbulent. When this occurs depends on the Reynolds number of the flow.

Hydrodynamics – Spring 2010 107


Couette flow

Consider incompressible fluid between two parallel plates at x = 0 and x = h. Assume that the upper
plate is moving at speed V along the z axis while the lower plate is at rest. Now, the flow between the
plates can be assumed to be along the z axis and independent on coordinates y and z . Thus, v = v(x) ez
and
dv ∂v
= + v · ∇v = 0
dt ∂t
so
2 ∂ 2v
∇p = µ∇ v = µ 2 ez ,
∂x
where, again, the pressure can depend only on z . Thus,

dp d2v
=µ 2
dz dx
where both sides must be equal to a constant. This integrates to give

x2 dp
v = c0 + c1 x + ,
2µ dz

where dp/dz is a constant.

Hydrodynamics – Spring 2010 108


Since v = 0 at x = 0, c0 = 0, and since v = V at x = h, we get the result

h2 dp
hc1 = V −
2µ dz

V x x2 − hx dp
⇒ v= +
h 2µ dz

known as the generalized Couette flow. (Simple Couette flow is obtained by setting p = const.)
Exercise: sketch the velocity profile for dp/dz < 0 and dp/dz > 0. What is the condition for zero
discharge?
The drag force excerted by the fluid on the plates (per unit area of the plate) is given by

dv V µ 2x − h dp 
 

∓µ = ∓ + ,
dx x=(h±h)/2 h 2 dz x=(h±h)/2

where the upper [lower] signs correspond to the upper [lower] plate. The force per unit area on the lower
plate is, then,
V µ h dp

h 2 dz
and on the upper plate
V µ h dp
− − .
h 2 dz
In simple Couette flow (dp/dz = 0), the drag forces are the same but oppositely directed.

Hydrodynamics – Spring 2010 109


Scaling and Reynolds number

Suppose a new submarine is designed and we want to find out the flow pattern around it when it is
moving through water. Is it possible to do laboratory experiments with a miniatyre model to find out,
e.g., the drag force excerted by the water on the submarine moving at different speeds? In short, yes.
Consider the fluid flow around geometrically similar objects. Let L be a typical size of the object
and V be a typical velocity of the fluid flow. Thus, L/V can be taken as the typical time scale of the
problem. Let us scale all variables by their typical values,

x = x0L; v = v0V ; t = t0L/V ; ω = ω 0V /L

We can see that the primed quantities are all dimensionless. The vorticity equation can be written
in a dimensionless form as
∂ω 0 0 0 0 1 02 0
= ∇ × (v × ω ) + ∇ ω
∂t0 Re

where ∇0 = ∂/∂x0 = L∇ and


LV
Re =
ν
is a dimensionless number known as the Reynolds number of the flow. For two geometrically similar
flows, the dimensionless flow patterns are identical, if their Reynolds numbers are the same.

Hydrodynamics – Spring 2010 110


Examples of scaling

• Uniform flow past a cylinder or a sphere: take L = 2a and V = U (relative speed of the fluid
and the object)
2aU
⇒ Re = .
ν
Thus, for miniature modeling of submarines we need to consider faster-than-real flows or fluids less
viscous than water (at ∼ 10◦C) in order to achieve a Reynolds number representative of the true
situation.
• Hagen–Poiseuille flow: take L = 2a and V = vave = ∆p a2/8µl

∆p a3
⇒ Re = .
ρl 4ν 2

Experimentally, laminar flow is obtained as long Re . 2400. If geometrical similarity is maintained


(i.e., a/l kept constant) and the Reynold’s number is kept constant, then

∆p a2
ρ 4ν 2

is also constant. This dimensionless quantity equals Eu/(4Re)2, where Eu ≡ ∆p/ρV 2 is the Euler
number of the flow. Thus, Hagen–Poiseuille flow is characterised by two dimensionless numbers
picked from Re, Eu and a/l.

Hydrodynamics – Spring 2010 111


Slow viscous flow past solid bodies

For incompressional (and barotropic) fluids, we derived the dimensionless equation

∂ω 0 0 0 0 1 02 0
= ∇ × (v × ω ) + ∇ ω.
∂t0 Re
Clearly, at the limit of Re  1, in steady state the equation reduces to

∇2 ω = 0

which is a much simpler equation than N–S because it is linear. Assuming incompressibility, ∇ · v = 0,
we can write v = ∇ × ψ , where ψ is the vector potential of the flow.
In a two-dimensional flow, we can take ψ = −ψ(x, y)ez , i.e.,

∂ψ ∂ψ
v=− ex + ey
∂y ∂x

showing that ψ is just the streamfunction. Use, next, ω = ∇ × (∇ × ψ) = ∇(∇ · ψ) − ∇2ψ , where
the first term vanishes in the 2-D case. Thus, the stream function fullfils the bi-harmonic equation

∇4 ψ = 0

for a viscous fluid. The correct boundary condition at the surfaces is ∇ψ = 0.


Hydrodynamics – Spring 2010 112
Stokes flow past a sphere
For axisymmetric meridional flows, like the flow past a sphere, we can use the Stokes streamfunction
Ψ defined, in spherical coordinates, by ψ = (Ψ/r sin θ) eϕ:

1 ∂(sin θ ψϕ) 1 ∂Ψ 1 ∂(rψϕ) 1 ∂Ψ


vr = = 2 ; vθ = − =−
r sin θ ∂θ r sin θ ∂θ r ∂r r sin θ ∂r
Thus, (exercise)

2 2
 ∂ sin θ ∂  1 ∂ 

2
Ê Ψ ≡  2 + 2  Ψ = 0.
∂r r ∂θ sin θ ∂θ
Far from the sphere, the velocity components for a uniform flow along the z axis are

1 ∂Ψ ∂Ψ
−U sin θ = vθ = − ⇒ = rU sin2 θ ⇒ Ψ = c(θ) + 12 U r2 sin2 θ
r sin θ ∂r ∂r
1 ∂Ψ c0(θ) + U r2 sin θ cos θ c0(θ)
U cos θ = vr = 2 = 2
= 2 + U cos θ ⇒ c0(θ) = 0
r sin θ ∂θ r sin θ r sin θ
⇒ Ψ = 12 U r2 sin2 θ at r ∼ ∞

The boundary conditions on the sphere are vr = vθ = 0 at r = a, and the boundary conditions at
infitinity are Ψ ∼ 21 U r2 sin2 θ. Let us try a solution of form Ψ = f (r) sin2 θ. Thus,
2 2
2 2
 
2f d 2 d 2
 
00 2 2 2 2
ÊΨ = f − 2  sin θ ⇒ Ê Ψ =  2 − 2  f (r) sin θ ⇒ D̂ f ≡  2 − 2 
  
 f (r) = 0.
r dr r dr r
Hydrodynamics – Spring 2010 113
Seek for solutions of form f = rn. Thus,

D̂f = [n(n − 1) − 2]rn−2 and D̂2f = [n(n − 1) − 2][(n − 2)(n − 3) − 2]rn−4

so n must fulfill
√ √
1± 1+8 1± 1+8
[n(n − 1) − 2][(n − 2)(n − 3) − 2] = 0 ⇔ n = = −1, 2 ∨ n = 2 + = 1, 4
2 2
Thus, n = −1, 1, 2, 4 and
f (r) = Ar−1 + Br + Cr2 + Dr4.
Now, at infinity, f ∼ 21 U r2 so D = 0 and C = 12 U . At the surface, f 0(a) = 0 = f (a) so
 
−2


 −Aa + B + Ua =0 

 A = 41 U a3
⇒
Aa−1 + Ba + 12 U a2 = 0 B = − 43 U a

 
 

Hence, finally,
Ψ = 12 U (r2 − 32 ar + 12 a3r−1) sin2 θ
and
1 ∂Ψ
vr = 2
= U (1 − 23 ar−1 + 12 a3r−3) cos θ
r sin θ ∂θ
1 ∂Ψ
vθ = − = −U (1 − 43 ar−1 − 41 a3r−3) sin θ
r sin θ ∂r

Hydrodynamics – Spring 2010 114


Streamlines of Stokes flow

In each meridional plane, ϕ = constant,

dr r dθ 1 ∂Ψ 1 ∂Ψ dΨ
= ⇒ dr + 2 r dθ = 0 ⇒ =0
vr vθ r sin θ ∂r r sin θ ∂θ r sin θ

so the streamlines are contours of the Stokes streamfunction, i.e.,

(r2 − 32 ar + 12 a3r−1) sin2 θ = b2,

where b is the distance of the streamline from the z axis at r → ∞. Thus, the flow is symmetric with
respect to equator, θ = 12 π .

Hydrodynamics – Spring 2010 115


Stresses. Stokes’s law

Stresses (incompressible flow):


−P = −pI + 2µΛ,
where Λ = 21 [∇v + (∇v)T], ∇v is the velocity gradient tensor, and where p can be solved from the
linearized equation of motion (consistent with Stokes flow),

0 = −∇p + µ∇2v = −∇p − µ∇ × ω.

Vorticity:  
1 ∂(rvθ ) 1 ∂vr  3 sin θ
ω= −  eϕ = − U a eϕ
r2

r ∂r r ∂θ 2
Pressure distribution: Write the equation of motion in component form as

∂p 1 ∂ 3U aµ cos θ 3U aµ cos θ
= −µ (sin θωϕ) = ⇒ p = p 0 (θ) −
∂r r sin θ ∂θ r3 2r2
1 ∂p 1∂ 3U aµ sin θ
=µ (rωϕ) = 3
⇒ p00(θ) = 0
r ∂θ r ∂r 2r
3U aµ cos θ
⇒ p = p0 −
2r2

Hydrodynamics – Spring 2010 116


Viscous stresses: The components of the viscous stress −π = 2µΛ = µ[∇v + (∇v)T] are most
easily given in spherical coordinates. For axisymmetric meridional flows, v = vr (r, θ)er + vθ (r, θ)eθ , so
the gradient operator can be given as ∇ = er ∂r + r−1eθ ∂θ . Thus,

∇v = (er ∂r + r−1eθ ∂θ )[vr (r, θ)er (θ) + vθ (r, θ)eθ (θ)]


= er (er ∂r vr + eθ ∂r vθ ) + r−1eθ (er ∂θ vr + eθ ∂θ vθ + vr ∂ e +vθ ∂
| θ{z r}
e)
| θ{z θ}
=eθ =−er

= er er ∂r vr + er eθ ∂r vθ + eθ er r−1(∂θ vr − vθ ) + eθ eθ r−1(∂θ vθ + vr )
∂vr 1 ∂vθ vr  1 ∂vθ 1 ∂vr vθ 
   

∴ Λ = er er + eθ eθ  + + 2 (er eθ + eθ er )  + −
∂r r ∂θ r ∂r r ∂θ r

Drag force: Total force exerted by the fluid on the sphere is


ˆ ˆ
F D = (−P) · ds = (−pI − π) · ds,
S S

where the integral covers the whole surface of the sphere. At the surface of the sphere, r = a and
ds = er a2 sin θ dθ dϕ so
ˆ π ˆ 2π
F D = −a2 dθ sin θ dϕ [(p + πrr ) er + πθr eθ ] .
0 0

Hydrodynamics – Spring 2010 117


The relevant components of π are, thus,
∂vr
πrr = −2µ =0
∂r
1 ∂v ∂v v 3U µ
 
r θ θ
πθr = −µ  + − = sin θ.
r ∂θ ∂r r 2a
Symmetry ⇒ non-zero component of force only in the z -direction. Therefore,
ˆ π ˆ 2π
2
F D = −a ez dθ sin θ dϕ[(p + πrr ) e · e +πrθ e
| r {z z}
·e ]
| θ {z z}
0 0 =cos θ =− sin θ
ˆ π
2
= −2πa ez dθ sin θ(p cos θ − πrθ sin θ) = 6πU µa ez Stokes’s law
0

Thus, a drag proportional to viscosity, velocity and circumference of the sphere is obtained.
Stokes’s law is obtained from ∇2ω = 0, i.e, neglecting the non-linear term v · ∇v from N–S equation
because of its assumed smallness relative to the linear viscous term ν∇2v for small Re. However, far
from the sphere v ≈ U so v · ∇v ≈ U · ∇v = U · ∇(v − U ), and |v − U | ∼ U ar−1. Thus,
|v · ∇v| ∼ U 2ar−2. On the other hand,

2 U aν 1 U 2 a2
2 1 a
ν|∇ v| = ν|∇ (v − U )| ∼ 3 = ∼ |v · ∇v|
r Re r3 Re r
so the neglected term is larger than the viscous term at a/r . Re. This leads to a correction term in
the drag force at Reynolds numbers not much smaller than unity.

Hydrodynamics – Spring 2010 118


Flow at larger Reynolds numbers

One could expect that at large Re, the viscocity term could be
neglected and the flow around a solid body, like a cylinder, would
be identical to that of an ideal flow. The real situation, found
experimentally, is quite different. When Re & 20, two vortices
appear behind the cylinder; when Re is further increased, the vor-
tices on the two sides appear alternately, and a string of vortices,
called Kármán vortex street (von Kármán 1911), appears in the
wake of the cylinder. Finally, when Re & 104 the wake becomes
turbulent.

Hydrodynamics – Spring 2010 119


Drag at large Reynolds numbers

In general, drag on a solid body can be written as

FD = 12 CDAρU 2

where A is the reference area and CD is the drag coefficient of the body that for Stokes flow around
a sphere are given by
12µ 24
A = πa2; CD = =
aρU Re
Here the Reynolds number is taken to be Re = 2aU/ν . There is a positive correction for a steady
laminar flow, as mentioned earlier. For non-steady wake flow, the drag coefficient increases even further.

Hydrodynamics – Spring 2010 120


Boundary layers

Even when increasing Re  1, the flow past an object does not become potential. Why?
The reason is that the potential flow has to allow for large tangential velocities on the surface of the
object, whereas in real fluids, v = 0 at solid boundaries. Next to the boundary, there should be a
boundary layer where the velocity increases from zero to large values somewhat away from the surface.
Thus, in the boundary layer, velocity increases rapidly because of the term ν∇2v and viscosity cannot
be neglected however small it is. This was first realized by Ludwig Prandtl in 1905.
Consider a horizontal flow in the xy plane above a semi-infinite horizontal plate at y = 0, x > 0.
Without viscosity, a potential velocity field of v = V ex would be a solution. Thus, outside the boundary
layer, this can be taken as the form of the velocity field.
Consider the transport of vorticity generated y
at the tip of the plate. We have
δ V
∂ω
0= = ∇ × (v × ω) + ν∇2ω
∂t
2 ∂ω ∂ 2ω
= −v · ∇ω + ν∇ ω ' −V +ν 2
∂x ∂y L x
Thus, denoting τ = x/V and assuming that vorticity is generated at τ = y = 0

v
∂ω ∂ ω 2
ω0 y2
− 4ντ
√ u
u 2νL
'ν 2 ⇒ω'√ e ⇒ δ(L) ' 2ντ =
u
t
∂τ ∂y 4πντ V

Hydrodynamics – Spring 2010 121


Boundary layer equations

The N–S equation and the equation of continuity in planar flow read
2
∂ 2vx 
 
∂vx ∂vx 1 ∂p  ∂ vx
vx + vy =− +ν 2 +
∂y 2

∂x ∂y ρ ∂x ∂x
2 2
 
∂vy ∂vy 1 ∂p ∂ vy ∂ vy 
vx + vy =− +ν

2
+ 2

∂x ∂y ρ ∂y ∂x ∂y
∂vx ∂vy
+ =0
∂x ∂y

When applying N–S to boundary value problems, we make use of the fact that ∂y ∼ 1/δ , ∂x ∼ 1/x.
Equation of continuity then implies that vx/x ∼ vy /δ .
It is useful to render the equation into a dimensionless form. We use

• a typical linear length scale, l, which makes x0 = x/l of unit order. Thus, δ 0 = δ/l is small.
• the asymptotic flow velocity V as the scale of velocities making vx0 = vx/V of unit order and
vy0 = vy /V ∼ δ 0 small.
• the dynamic pressure ρV 2 as the scale of pressure. Thus, p0 = p/ρV 2 is of order Eu.

Finally, since δ ∼ νx/V , we note that ν 0 = Re−1 = ν/V l ∼ δ 0−2.


r

Hydrodynamics – Spring 2010 122


We can now write down the equations in dimensionless form as
0 0
∂p0 1  ∂ 2vx0 ∂ 2vx0 
 
0 ∂vx 0 ∂vx
vx 0 + vy =− 0+ +
Re ∂x02 ∂y 02
 
∂x ∂y ∂x
1 1 Eu 2 1 1
1 · δ0 · δ0 2
1 δ0 1 1 δ0
0 0
∂p0 1  ∂ 2vy0 ∂ 2vy0 
 
0 ∂vy 0 ∂vy
vx 0 + vy 0 = − 0 + +
Re ∂x02 ∂y 02
 
∂x ∂y ∂y
δ0 δ0 Eu 2 δ0 δ0
1 · δ0 · δ0 2
1 δ0 δ0 1 δ0

∂vx0 ∂vy0
+ =0
∂x0 ∂y 0
1 δ0
1 δ0

Below the equations, the order of each term has been indicated. Keeping only the leading order terms
in δ 0 we get from the y -component that ∂y0 p0 = 0 ⇒ p0 = p0(x0). Thus,
0 0
0 ∂vx 0 ∂vx dp0 1 ∂ 2vx0
vx 0 + vy =− 0+
∂x ∂y dx Re ∂y 02
∂vx0 ∂vy0
+ =0
∂x0 ∂y 0

The pressure gradient is determined from the potential part of the flow. Thus, −dp0/dx0 ' V 0 dV 0/dx0.

Hydrodynamics – Spring 2010 123


One can then apply the so-called boundary-layer transformation,

0

ȳ = y Re

v̄y = vy0 Re

to get the boundary layer equations


0
0 ∂vx ∂vx0 0 dV
0
∂ 2vx0
vx 0 + v̄y = −V +
∂x ∂ ȳ dx0 ∂ ȳ 2
∂vx0 ∂ v̄y
+ =0
∂x0 ∂ ȳ

with the boundary conditions


ȳ = 0 : vx0 = 0; v̄y = 0
ȳ → ∞ : vx0 = V 0(x).
For the plate flow example, V 0 = 1. Thus,
0
0 ∂vx ∂vx0 ∂ 2vx0
vx 0 + v̄y =
∂x ∂ ȳ ∂ ȳ 2
∂vx0 ∂ v̄y
+ =0
∂x0 ∂ ȳ
ȳ = 0, x0 > 0 : vx0 = 0; v̄y = 0
with
ȳ → ∞ : vx0 = 1

Hydrodynamics – Spring 2010 124


Solution of the plate flow BL problem

Boundary layer problem, being an incompressible 2D N–S problem, can be most easily solved by intro-
ducing a streamfunction:
∂ψ ∂ψ
vx0 = ; v̄y = − 0 .
∂ ȳ ∂x
r r √
We deduced that δ ' 2νx/V = 2x0/Re, i.e., δ̄ ' 2x0. Thus, when solving the boundary layer
equations, use of a similarity variable √
η = ȳ/ 2x0
will be made. It turns out that a streamfunction of form

ψ= 2x0f (η)

is useful. Thus,
√ ∂η df df
vx0 = 2x0 = ≡ f˙
∂ ȳ dη dη
f √ ∂η df 1  ˙ 
v̄y = − √ − 2x 0 0 = √ ηf − f
2x0 ∂x dη 2x0

with
... η=0: f˙ = 0; f =0
f + f f¨ = 0 with Blasius equation
η→∞: f˙ = 1

Hydrodynamics – Spring 2010 125


http://faculty.virginia.edu/ribando/modules/xls/

√ √
Note: some sources (like the one above) use the similarity transformation η = ȳ/ x and ψ = x0f (η),
0

which yields the Blasius equation in form


... 1
f + 2 f f¨ = 0; f (0) = f˙(0) = 0; f˙(∞) = 1
Hydrodynamics – Spring 2010 126
Aerodynamic lift

We will, next, try to understand the aerodynamic lift, which allows birds, airplanes and helicopters to fly
and contributes to the down-force of the race cars.
For simplicity, we shall limit the discussion
to airfoils. One can think of an airfoil as
the cross-section of a translationally symmet-
ric wing that has a much longer span than
chord and a leading edge perpendicular to
the flow direction. The flow around the wing
is, thus, two-dimensional.
The lift of an airfoil is based on the principle of Bernoulli: if the upper surface of the wing is longer than
the lower surface and/or if the angle of attack is positive, the flow above the wing is faster than below.
Thus, there is underpressure above the wing that generates the lift.
We shall see that the lift of an airfoil is proportional to the
circulation Γ around the airfoil, just as for the ideal flow
around a cylinder considered in the exercises. But what
produces the circulation around the airfoil? The answer
is viscosity. The flow pattern around the airfoil without viscous forces
very large
any circulation contains extremely large velocity gradients
near the pointed end of the airfoil, which are relaxed by the
generation of vorticity, i.e., circulation around the airfoil.

Hydrodynamics – Spring 2010 127


Calculation of the lift on an airfoil

While viscosity is responsible for the initial generatation of circulation around the airfoil, we expect its
effect to be small outside the boundary layer around √ the airfoil. The thickness of a (laminar) boundary
layer near the trailing edge of the airfoil is δ ∼ 5L/ Re. The kinematic viscosity of air near the ground
is ν ∼ 10−5 m2s−1. If the speed of a model aircraft is V ∼ 10 ms−1 and the chord of the wing is
L ∼ 0.1 m, we get Re ∼ 105 (still laminar). Thus, δ/L ∼ 0.2% and the velocity field around the airfoil
is well represented as a potential flow outside the thin boundary layer attached to the airfoil.

Consider the two-dimensional velocity field in the complex xy plane, i.e., take

w(z) = vx − i vy , z = x + i y.

Because the flow is potential, w is an analytic function. Thus, it can be expanded as a Laurent series
around the origin (taken to be inside the airfoil) as
˛
a−1 a−2 1
w(z) = v∞ + + 2 + ...; a−k = z k−1w(z) dz,
z z 2π i C

where the integral can be taken along any (counter-clockwise) closed curve C enclosing the origin, and
use has been made of w → v∞ at |z| → ∞. Thus,
˛ ˛ ˛ ˛
2πi a−1 = w dz = (vx − i vy ) (dx + i dy) = (vxdx + vy dy) − i (vy dx − vxdy).
C C C C

Hydrodynamics – Spring 2010 128


Choosing now the curve to follow the streamline just above the thin boundary layer, we find that the
second integral vanishes, since dx/vx = dy/vy along a streamline.5 Thus,
˛ ˛
2πi a−1 = (vxdx + vy dy) = v · dl = Γ.
C C

The net force on the airfoil per unit span can be ob-
y
tained by integrating the force exerted by pressure, C ey
˛
F = pn dl, x
dl
ex
where n is the inward normal of the surface. Taking
n
θ = (dl, ex), we have n = − sin θ ex + cos θ ey .
Since p is not a function of the normal distance through the boundary layer, we can again take the
integral along the streamline enclosing the airfoil and the boundary layer. Bernoulli’s principle states that

p + 21 ρv 2 = constant

along a streamline (or along any line in irrotational flow), so


˛
F = − 12 ρ v 2 n dl.
C
5 There are two sections of C of length ∼ δ above the stagnation points, where the integration cannot be taken along a streamline, but as
δ/L ≈ 0, these can be neglected.

Hydrodynamics – Spring 2010 129


Introduce a complex force
˛ ˛
Z = Fy + i Fx = − 12 ρ v 2 (ny + i nx) dl = − 21 ρ v 2 (cos θ − i sin θ) dl
C C
˛ ˛
= − 21 ρ v 2 (cos θ dl − i sin θ dl) = − 12 ρ v 2 (dx − i dy)
C C
˛ ˛
= − 21 ρ ww∗ dz ∗ = − 12 ρ w(w dz + w∗ dz ∗ − w dz)
C C

and since w∗ dz ∗ − w dz = −2i =(w dz) = −2i ={(vx − i vy )(dx + i dy)} = −2i(vxdy − vy dx) vanishes
along any streamline, we have ˛
Z = − 21 ρ w2 dz.
C
Now, using the Laurent expansion of

2v∞a−1 v∞ Γ
w2 = (v∞ + a−1/z + ...)2 = v∞
2
+ 2
+ ... = v∞ + + ...
z πi z
˛ ˛
v∞Γ 1
⇒ = w2 dz ⇒ w2 dz = 2v∞Γ
πi 2π i
˛
we get
Z = − 12 ρ w2 dz = −ρv∞Γ,
C
i.e., Fx = 0 and
Fy = −ρΓv∞. Kutta–Joukowski

Hydrodynamics – Spring 2010 130


Joukowski transform. Kutta condition

Lift is

Fy = −ρv∞Γ

regardless of the airfoil shape.


But what is the value of Γ?
This can be deduced by confor-
mal mapping of a cylinder to an
airfoil shape. For example, by
1
 
−iα 
Joukowski transform: z = e ζ + .
ζ
r
Cylinder centered at µ, radius a = (1 − µx)2 + µ2y

ζ = µ + ae−iβ ; β ∈ [0, 2π]

Thus, ζ = 1 when sin β = µy /a. This point maps to the cusp.


The stagnation point is at sin(α+β) = −Γ/4πv∞a (exercise).
Mapping it to the cusp gives

−1 µy !
µ = −0.12 + 0.08 i; α = 10◦ Γ = −4πv∞a sin α + sin Kutta condition
a

Hydrodynamics – Spring 2010 131


Accretion of mass

Accretion of mass on compact massive objects is one of the most efficient energy release mechanisms.
For example, the loss of potential energy of a particle of mass m falling through the gravitational field
of a neutron star (from infinitely far to the surface) is

GM?m GM?
= 2 mc2,
a ca

where the neutron star mass M? ' M and the neutron star radius a ' 10 km. The factor GM?/c2a '
0.15, so the gravitational energy available in such accreting material is more than what can be obtained
from a fusion process converting hydrogen to heavier elements!
If accreting matter around a massive compact object possesses angular momentum, it forms an accretion
disk around the object. Accretion disks are important ingedients in many astrohysical objects such as
X-ray binaries and active galactic nuclei.
In general, accretion is a complicated phenomenon to model. We shall limit our discussion to only
two forms of accretion, thin accretion disks (to be studied now) and spherical accretion (to be studied
later together with stellar winds). Furthermore, we shall also neglect self-gravity in all discussion, thus
assuming that the disk is much less massive than the accreting object.

Hydrodynamics – Spring 2010 132


Accretion disks

Consider matter orbiting an object of mass M in a nearly circular orbit of radius r. Thus, balancing
the gravitational acceleration −GM er /r2 with the centrifugal acceleration vθ2er /r gives for the angular
velocity Ω = vθ /r
1/2
GM

Ω(r) =  3 
r
which is often referred to as Keplerian motion (for obvious reasons).
As the angular velocity is a function of radius, the rotation of matter is not rigid. Thus, a velocity shear
is produced. It is the viscous forces related to this shear that allows the dissipation of mechanical energy
to occur. In the absence of viscosity, particles on circular orbits conserve their total mechanical energy

GM GM
 = T + V = 21 r2Ω2 − =− = 12 V
r 2r
Particles losing mechanical energy cannot stay on fixed circular orbits. A slow loss of mechanical energy
means that ṙ = −r/
˙ < 0 as ˙ < 0,  < 0 and, thus, particles are slowly spiraling inwards while orbiting
the massive object. Thus, a particle starting at infinity ( = 0) and spiralling to the surface of the object
( = − 12 GM/a) has lost half of the potential energy difference on the way in. This energy goes to heat
and, ultimately, to radiation. As ˙ = Fθ vθ and Fθ ∝ ν , we expect luminosity L = 12 GM ṁ/a ∝ ν ,
where ṁ ∝ ν is the accretion rate on the object.
Thus, although there were no fixed solid bodies in the accretion disk and although the Reynolds number
of the system is enormous, viscosity is still in a central role in the physics of accretion.
Hydrodynamics – Spring 2010 133
The basic disk dynamics

Consider a thin accretion disk in cylindrical coordinates with Ω k ez . Assume azimuthal symmetry and
vz = 0. Thus, equation of continuity and the azimuthal component of the N–S equation are

∂ρ 1 ∂
+ (rvr ρ) = 0
∂t r ∂r
∂v ∂v v v
 
θ θ r θ  = {∇ · [2µ(Λ − 1 I tr Λ)]} ,
ρ + vr + 3 θ
∂t ∂r r
where we have assumed that bulk viscocity vanishes. Note that because we are now considering the
effects of viscocity at large scales, we cannot assume that µ is constant. Rather than writing down the
viscous terms in terms of the components of Λ, we´ shall derive the term from first principles. First,
however, integrate the equation over z writing Σ = dz ρ:

∂Σ 1 ∂
+ (rvr Σ) = 0
∂t r ∂r
ˆ
∂v ∂v v v
 
θ θ r θ
Σ + vr +  = dz {∇ · [2µ(Λ − 13 I tr Λ)]}θ ,
∂t ∂r r

where we have assumed that vθ and vr do not depend on z . Multiply the upper equation by rvθ and add
to the lower multiplied by r to get

∂ 1∂ vθ
(Σr2Ω) + (rvr Σr2Ω) = G; Ω=
∂t r ∂r r
Hydrodynamics – Spring 2010 134
Note that Σr2Ω is the surface density of angular momentum and vr Σr2Ω is its flux in the radial direction.
As it does not depend on any other coordinate than r, the second term is clearly the divergence of the
angular momentum flux. The equation, therefore, states the conservation of angular momentum, and G
must be related to the viscous torque.

Multiply the angular momentum transport equation by 2πr. Thus,

∂ ∂
(2πr Σr2Ω) + (vr 2πr Σr2Ω) = 2πr G
∂t ∂r

As dr L(r, t) = 2πr dr Σr2Ω is the angular momentum carried by an annulus of width dr of the disk at
r, we have  
∂L ∂(v r L)
dr  +  = G(r + dr) − G(r),


∂t ∂r
where G(r) is the viscous torque at radius r. It is the torque exerted on the material inside r by the
material outside r through viscous forces. Thus, 2πr G = ∂G/∂r and so we need to find G.

The viscous stress is τθr = µr ∂Ω/∂r so the viscous torque per unit area of a cylindrical surface is r τθr .
The viscous torque is, thus,
ˆ ˆ
∂Ω ∂Ω
G(r, t) = r dθ dz µr2 = 2πνΣr3
∂r ∂r

and
∂ 1∂ 1∂  ∂Ω
 

(Σr2Ω) + (rvr Σr2Ω) = νΣr3 


∂t r ∂r r ∂r ∂r
Hydrodynamics – Spring 2010 135
The basic equations governing the disk dynamics are, thus,

∂Σ 1 ∂
+ (rvr Σ) = 0
∂t r ∂r
∂ 1∂ 1∂  3 ∂Ω 
 
2 2
(Σr Ω) + (rvr Σr Ω) = νΣr .
∂t r ∂r r ∂r ∂r
In order to obtain a dynamical theory based on these two equations, one of the unknowns Σ, vr and Ω
needs to be prescibed. Considering the angular velocity to be nearly Keplerian, we can regard Σ and vr
as the dynamical variables of the theory and solve the equations.
As Ω ∝ r−3/2, we can write the second equation (LHS first) as

1 ∂  3 1/2
 ∂Σ 1 ∂ 3/2 ∂Σ 1 ∂ vr Σ
− 2 νΣr = + (r vr Σ) = + (rv r Σ) +
r3/2 ∂r ∂t r3/2 ∂r |∂t r ∂r
{z } 2r
=0
∂ 
1/2 1/2

⇒ rvr Σ = −3r νΣr
∂r
∂Σ 3 ∂  1/2 ∂ 
 

∴ = r νΣr1/2  = ∇ · (D∇Σ − V Σ),
∂t r ∂r ∂r

where D = 3ν and V = −3r−1/2∂r (r1/2ν) er . The diffusion term tries to spread the disk whereas
convection term drives the accretion on the central mass. Note that the mass influx, −2πrvr Σ is
positive as long as r1/2νΣ increases with with distance. The disk equation can be solved analytically,
e.g., for a constant ν and Σ(r, 0) = (m/2πr0) δ(r − r0) (see Choudhuri).

Hydrodynamics – Spring 2010 136


Near-Keplerian orbits

Consider the two remaining components of the N–S equation:

∂vr vθ2 1 ∂p GM
vr − =− − 2
∂r r ρ ∂r r
1 ∂p GM z
0=− − ,
ρ ∂z r3

√ component of the gravitational force components are approximated assuming z  r,


where the radial
i.e., taking r2 + z 2 ≈ r, and where viscosity has been neglected as the other terms are much larger.
The thin disk approximation immediately yields another simplification. Since the thickness of the is disk
h, then |∂p/∂z| ∼ p/h so that from the z component of N–S we get

p GM h 1 ∂p p h2 GM GM


∼ ⇒ ∼ ∼ 


ρh r3 ρ ∂r ρr r2 r2 r2

so that the pressure term can be neglected from the radial component of N–S. Once we again realize
that vr  vθ ,we see that the radial component yields the balance between centrifugal and gravitatinal
force that led to the Keplerian angular velocities.
To summarize, near-Keplerian motion is valid for thin disks, for which pressure forces can be neglected.
If pressure terms are important, the disk is no longer thin and the analysis complicates considerably.

Hydrodynamics – Spring 2010 137


Thin disk in steady state

In steady state
1d
(rvr Σ) = 0
r dr
1d 1 d ∂Ω
 

(rvr Σr2Ω) = νΣr 3 


r dr r dr ∂r
Both equations integrate immediately to give

rvr Σ = C1
dΩ
Σr3Ωvr − νΣr3 = C2
dr
where C1 and C2 are constants of integration. For a steady disk, the mass inflow rate ṁ = −2πr vr Σ.
Thus,

C1 = − .

The second integration constant is obtained at the inner radius of the disk, r = r?, where we impose the
condition of rigid rotation, dΩ/dr = 0 and Ω = Ω?. Thus,

ṁ 2
C2 = C1r?2Ω? = − r Ω? .
2π ?
Hydrodynamics – Spring 2010 138
In summary,

3 ∂Ω ṁ 2 ṁ  2 
3 2
r νΣ = Σr Ωvr + r? Ω? = r Ω? − r Ω
∂r 2π 2π ?
ṁ r?2Ω? − r2Ω
νΣ = ,
2π 3 ∂Ω
r
∂r

i.e., ṁ ∝ ν , as expected. The viscous dissipation per unit volume is µr2(dΩ/dr)2. This gives the energy
emitted from the unit area of the disk as
ˆ 2 2 2 2
dE dΩ dΩ ṁ r Ω − r Ω ∂Ω
 
2 2 ? ?
− = µr  dz = νΣr   =
dt dr dr 2π r ∂r

Integrating this over the disk area gives the luminosity


ˆ ∞ ˆ ∞
ṁ r?2Ω? − r2Ω ∂Ω ∂Ω
L= 2πr dr = ṁ (r?2Ω? − r2Ω) dr
r? 2π r ∂r r? ∂r
ˆ ∞
∂Ω
= ṁr?2Ω?(−Ω?) − ṁ r2Ω dr
r? ∂r
ˆ ∞  ˆ ∞ 

= −ṁr?2Ω2? + 21 ṁr?2Ω2? + ṁ rΩ2 dr = ṁ  1 2 2


− r Ω +
2 ? ? rΩ2 dr

r? r?

which equals 21 GM ṁ/r?, as expected, if the Keplerian formula for Ω is used.

Hydrodynamics – Spring 2010 139


The Keplerian formula does not apply in the region close to the inner radius of the disk, where the matter
is rotating rigidly. Thus, one should write
ˆ ∞
vθ2
 
1 2 2
− r Ω +
L = ṁ  2 ? ? dr

r? r
ˆ ∞
∂vr 1 ∂p GM 

= − 12 ṁr?2Ω2? + ṁ v
r + + 2 dr
r? ∂r ρ ∂r r
GM  GM 1 2
   

= −ṁ  12 r?2Ω2? + 12 vr,?


2
+ P? − = −ṁ − + 2 v? + P?
r? r?

where ˆ ∞ ˆ p(r)
1 ∂p dp
P(r) = − dr = = ∆w − ∆q,
r ρ ∂r p∞ ρ(p)
∆w = w(r) − w∞ is the change in specific enthalpy, w =  + p/ρ, and ∆q the deposited heat per unit
mass. Here, we have used the first law of thermodynamics in form

dp
dw = dq + .
ρ

As we assumed a detailed balance between radiative losses and viscous heating, we should take ∆q = 0,
so we get the luminosity as
L = −ṁ ∆( 12 v 2 + w + Φ),
where Φ = −GM/r is the gravitational potential. Thus, the gravitational potential energy is converted
to kinetic energy, enthalpy and heat, which in turn is converted to radiation.

Hydrodynamics – Spring 2010 140


Gas dynamics

Hydrodynamics – Spring 2010 141


Compressible fluids

Problems so far studied neglected the effects of compressibility of the fluid. Next, we will turn to the
dynamics of compressible fluids.
When comparing the Euler and N–S equations, we found that the second derivative in the N-S equation,
however small in most regions of space, changed the nature of the differential equation completely
allowing solutions with the no-slip boundary conditions to be obtained.
In a similar manner, compressibility affects the fundamental nature of the solutions of fluid dynamical
equations. Introducing compressibility gives a finite speed of propagation for mechanical disturbances in
the fluid, i.e., the speed of sound cs. At low speeds (v  cs) the effect of finite speed of information
is negligible, and incompressibility is a good approximation. As long as v < cs, the fundamental nature
of the solution does not change – the HD equations will be of the elliptic type.
At high flow velocities (v > cs), however, information cannot propagate against the flow. This changes
the HD equations from elliptic to hyperbolic type. The solutions are then fundamentally different from
the incompressible solutions, as we shall see.
Compressibility effects are most pronounced for problems involving dilute gases rather than liquids (with
the important exception of sound propagation in liquids). Thus, we will consider the fluid as a perfect
gas.

Hydrodynamics – Spring 2010 142


Thermodynamics of a perfect gas
The perfect gas law can be stated as
p = RρT,
where R = κB/m is the gas constant. The internal energy per unit mass is

 = cV T,

where cV = 12 f R is the specific heat per unit mass. Here, f is the number of degrees of freedom (DoF)
of the molecule, i.e., f = 3 (3 translational DoF) for monoatomic gas and f = 5 for a diatomic gas (3
translational and 2 rotational DoF). It is typical, instead of f , to use γ = (f + 2)/f = cp/cV , i.e.,

R RT
cV = ⇒ = .
γ−1 γ−1

Consider the the usual thermodynamic definition of entropy as δQ = T dS . Introduce then the
extensive quantities as given per unit mass, δQ = ρ δq and dS = ρ ds, and write down the first law of
thermodynamics in form
1 dρ dT dρ
   

T ds = d + p d   = d − p 2 = cV T  − (γ − 1) 
ρ ρ T ρ
dT dρ dρ dp dρ p
   

= cV T  + − γ  = cV T  − γ  = cV T d ln γ
T ρ ρ p ρ ρ
p
⇒ s = s0 + cV ln γ
ρ
Hydrodynamics – Spring 2010 143
Adiabatic flows

For an adiabatic process, 0 = δQ = ρT ds giving

ds d p
 

=0 ⇒ =0
dt dt ργ

which replaces the energy equation.


The enthalpy per unit mass is defined through

p γ 1 dp dp
 

w =+ = RT ⇒ dw = d + p d   + = T ds +
ρ γ−1 ρ ρ ρ

so for adiabatic flows (ds = 0)


ˆ
dp dp γ
dw = ⇒ =w= RT.
ρ ρ γ−1

Thus, the principle of Bernoulli for adiabatic flows states that

1 2 γ
2v + RT + Φ = constant
γ−1

along a streamline, Φ being the potential of the conservative external force field, F = −∇Φ.
Hydrodynamics – Spring 2010 144
Acoustic waves

In order to concentrate on the effects of compressibility, we shall neglect transport phenomena (i.e., set
µ = 0 = K ) for a while.
Consider a homogeneous perfect gas at rest, and denote the quantities in this equilibrium (in the absence
of external forces) by
ρ = ρ0; p = p0; v = v 0 = 0.
Suppose that pressure is perturbed to p = p0 + p1(x, t) so that the corresponding density and velocity
are ρ = ρ0 + ρ1(x, t) and v = v 1(x, t). Assuming that the perturbations are adiabatic, we can relate

d p 1 dp1 p dρ1
 

0 =  γ = γ − γ γ+1 ⇒ dp1 = c2s dρ1


dt ρ ρ dt ρ dt

where c2s = γp/ρ.


Assuming that the perturbations are small, ρ1  ρ0, p1  p0, we can replace the quantities in c2s by
their equilibrium values, i.e., write
γp0
c2s = ,
ρ0
which linearizes the equation and gives p1 = c2s ρ1. Using the same assumption in the equation of
continuity, we get
∂ρ1 ∂ρ1
0= + ∇ · [(ρ0 + ρ1)v 1] ≈ + ρ0∇ · v 1
∂t ∂t
providing an equation between ρ1 and v 1.
Hydrodynamics – Spring 2010 145
Next, apply the same technique of linearization to the Euler equation

∂v 1 ∂v 1 ∂v 1
 

0 = (ρ0 + ρ1)  + (v 1 · ∇)v 1 + ∇p1 ≈ ρ0 + ∇p1 = ρ0 + c2s ∇ρ1.


∂t ∂t ∂t
Combining this with the linearised equation of contiunuity gives

∂ 2ρ1
2
− c2s ∇2ρ1 = 0
∂t

which is the equation for acoustic waves showing that

v
u
u γp0 q
cs = u
t = γRT is the sound speed.
ρ0

Newton (1689) first derived the expression for sound speed assuming that the perturbations were isother-
mal, corresponding to γ = 1 in our equation. This expression gives cs ≈ 280 m/s for a gas at 0◦C, much
lower than the experimental value. It was Laplace (1816) who realised that the perturbations should be
regarded adiabatic and taking γ = 1.4, appropriate for air, gives cs ≈ 332 m/s, which agrees well with
the experimental value.
We note that assuming that p1 = (dp/dρ)0ρ1 gives c2s = (dp/dρ)0 which applies for all fluids. Liquids,
which are hard to compress, yield a small ρ1 with a large p1 meaning that the sound speed in liquids is
large. (For water, cs ≈ 1430 m/s.)

Hydrodynamics – Spring 2010 146


Harmonic analysis

Taking ρ1 = ρ̄1 exp{i(k · x − ωt)} and substituting this into the equation for acoustic waves gives

(−ω 2 + c2s k 2)ρ̄1 exp{i(k · x − ωt)} = 0 ⇒ ω 2 = c2s k 2

as the dispersion relation for acoustic waves.


The phase and group velocities of the acoustic waves are given by

ωk k ∂ω k
vp ≡ = c s ; vg ≡ = cs
k2 k ∂k k
The phase velocity is always perpendicular to and gives the velocity of the wave fronts (i.e., planes with
constant values of the phase, φ = k · x − ωt). The group velocity gives the velocity of individual wave
packets (and, thus, the propagation velocity of information and energy).
Acoustic waves in a homogeneous medium are non-dispersive, i.e., have phase and group speeds,
vp = cs = vg , independent of ω . Waves in more complicated media, like gravitationally stratified
atmospheres, have more complicated dispersion relations and turn out to be dispersive.
It follows from the Euler equation,
−iωρ0v 1 + c2s ikρ1 = 0
that v 1 k k. Thus, acoustic waves are longitudinal waves.

Hydrodynamics – Spring 2010 147


Non-linearity. Steepening into shock waves

When deriving the dispersion relation of acoustic waves, we assumed that non-linear terms in the HD
equations are vanishingly small. What happens if we relax this assumption?
For simplicity, let us consider a 1-D situation, where a purely longitudinal wave propagates in the x-
direction. Thus,

∂ρ ∂
+ (ρv) = 0
∂t ∂x
∂v ∂v 1 ∂p c2s ∂ρ
+v =− =−
∂t ∂x ρ ∂x ρ ∂x

where a polytropic relation between the pressure and the density is assumed and where c2s = dp/dρ.
Clearly, there are several non-linear terms in this system of equations. However, assuming that the Mach
number of the flow,
v
M≡
cs
is large (cs ≈ 0), we obtain
∂v ∂v
+v = 0,
∂t ∂x
which already contains non-linearity but is much simpler to solve than the full set of equations.
Let us, next, consider the solution of the simplified Euler equation, containing only the inertial force, in
the xt-plane.
Hydrodynamics – Spring 2010 148
Thus,
∂v ∂v
+v = 0.
∂t ∂x
Consider the values of v along the characteristic curves given by

dx
= v(x, t).
dt
Since
dv ∂v ∂v dx ∂v ∂v
= + = +v = 0,
dt ∂t ∂x dt ∂t ∂x
the value of the velocity v does not change along the characteristics, so the curves are actually straight
lines in the xt-plane. The exact solution can be formally written as

v = g(x − vt),

where g(x) = v(x, 0) is the initial velocity profile. E.g.,



v0, x < −L








g(x) = 

−v0, x>L


−v0x/L, |x| < L




v0, x < v0t − L








⇒ v(x, t) = 

−v0, x > L − v0t


−v0x/(L − v0t), |x| < L − v0t



Hydrodynamics – Spring 2010 149


At t = L/v0 the solution

v0 , x < v0t − L








v(x, t) = 

−v0, x > L − v0 t


−v0x/(L − v0t), |x| < L − v0t



has reached a state where it becomes discontinuous, i.e., a step from v = −v0 to v = v0 at x = 0.
Clearly, the solution only applies for t < L/v0 since for larger values of time the function becomes
multivalued.
At times t < L/v0 the velocity profile is becoming steeper and steeper around x = 0. This phenomenon,
caused by the term v ∂v/∂x and called non-linear steepening, leads to the formation of a shock
wave for any large-amplitude acoustic wave packet.
For a sinusoidal wave, i.e., g = −v0 sin kx, the region near x = 0 has g ≈ −v0kx. Thus, for |x|  k −1

v ≈ −v0k(x − vt)
v0kx
⇒v≈−
1 − v0kt

so the time needed for shock formation through non-linear steepening can be estimated as t ≈ (kv0)−1,
where k is the wavenumber and v0 is the amplitude of the wave.
The physics of the shocks, once they have formed, can be analysed using the conservation laws for mass,
momentum and energy.

Hydrodynamics – Spring 2010 150


Structure of shock waves

A shock wave is a region of small thickness over which the


dynamical variables of the fluid change rapidly. Shocks are
always related to compression, as steepening to shocks will only
occur in regions with ∂v/∂x < 0. (A region of fluid with
rarefaction, ∂v/∂x > 0, will get wider as a result of the inertial
term.) At the limit of very thin transition, shocks can be treated
as discontinuities in density, velocity and pressure. ex
As we treat shocks as discontinuities, we can regard them as
locally planar fronts with unit normal n. Observing the shock vs
from the rest frame of the ambient fluid, we can take the ve-
n
locity of the shock front as v s = vsn. Transforming to the
coordinate system co-moving with the shock at this velocity,
we can treat the flow near the shock as being independent of
time. In this frame of reference, the gas is flowing in from the
ambient medium to the shock at velocity −vsn. It is custom- vx = v s
ary to choose one of the coordinate axes, say x, parallel to
the shock normal so that the velocity of the gas through the
x
shock is positive, i.e., ex = −n. Thus, vx = vs at x < 0.
This region is called the upstream region of the shock. In the
downstream region (i.e., x > 0) the velocity is also in the x
direction because of symmetry.
Hydrodynamics – Spring 2010 151
Notation and jump conditions

It is customary to denote the values of the hydrodynamic quan-


tities (ρ, v and p) measured in the upstream region by subscript
p1, ρ 1 p2, ρ 2
1 and those measured in the downstream region by subscript
2. The ideal hydrodynamic equations in conservation form can v1 v2
be written as
∂ρ ∂
+ (vρ) = 0
∂t ∂x
x
∂ ∂
(ρv) + (ρv 2 + p) = 0
∂t ∂x
∂ 1 2 ∂
( 2 ρv + ρ) + [( 12 v 2 + w)ρv] = 0.
∂t ∂x
In the shock frame the system is in a steady state and the conservation laws can be integrated accross
the discontiunuity, using wρ = γp/(γ − 1), to give

v1ρ1 = v2ρ2
ρ1v12 + p1 = ρ2v22 + p2
1 2 γ p1 1 2 γ p2
v
2 1 + = v + .
γ − 1 ρ1 2 2 γ − 1 ρ2

Regarding the upstream quantities as known, the downstream quantities can be solved from these jump
conditions (or Rankine–Hugoniot conditions).
Hydrodynamics – Spring 2010 152
Solution of the jump conditions

The jump conditions can be solved to give the ratios v2/v1, ρ2/ρ1 and p2/p1 as a function of the Mach
number
v1 v1
M= =r
cs1 γp1/ρ1
of the shock. The solution is (exercise)

v1 ρ 2 (γ + 1)M2
= =
v2 ρ1 2 + (γ − 1)M2
p2 2γM2 − (γ − 1)
=
p1 γ+1

Clearly, the upstream and downstream quantites agree if M = 1. For values M < 1, the compression
ratio ρ2/ρ1 < 1, and because shocks are always compressive, the equation does not apply. Furthermore,
the compression ratio of the shock increases monotonically as a function of the Mach number and
approaches a finite limit at infinity:

ρ2 γ+1 γ+1
→ , as M → ∞. For γ = 35 , = 4.
ρ1 γ−1 γ−1

The pressure ratio, on the other hand, increases without a limit as M → ∞. It is straightforward to
show that the specific entropy s = κB ln(p/ργ ) increases over the shock as long as ρ2 > ρ1, consistent
with the second law of thermodynamics.
Hydrodynamics – Spring 2010 153
More shock properties

The downstream temperature is

p2/ρ2 p2 ρ 1 [2γM2 − (γ − 1)][(γ − 1)M2 + 2]


T2 = T1 = T1 = T1
p1/ρ1 p1 ρ 2 (γ + 1)2M2

which also increases as a function of Mach number. Thus, shock waves heat the gas. For strong shocks

RT1 2γ(γ − 1) 2 4 1 2
2 ≈ M = v
γ − 1 (γ + 1)2 (γ + 1)2 2 1

showing that strong shocks efficiently convert upstream kinetic energy to downstream thermal energy.
The Mach number of the downstream flow is

v22 ρ2v22 ρ1v12 ρ1 p1 2 ρ1 p1 (γ − 1)M2 + 2


M22 = 2 = = =M =
cs2 γp2 γp1 ρ2 p2 ρ2 p2 2γM2 − (γ − 1)

which yields M2 < 1 for all M > 1. Thus, shock waves are transitions from supersonic to subsonic
flow. This property is important, as it shows that shock waves forming ahead of obstacles in a supersonic
flow provide the means for information about the obstacle to propagate to the gas.
The thickness of a shock is, of course, finite in reality. The steepening of an acoustic wave is limited
by the transport processes (viscosity and heat conduction), which gives the shock thickness roughly as
L ∼ ν2/∆v .
Hydrodynamics – Spring 2010 154
Blast waves
Consider a sudden release of energy within of a localized region of gas, e.g., a supernova explosion or
a nuclear detonation in the atmosphere. We expect the explosion to generate a (spherical) shock wave
traveling through the gas. The task is to find out how the radius of the shock evolves with time and how
the hydrodynamic quantities behave inside the spherical shock.
Suppose that the amount of energy released is E and that the density of the ambient medium is ρ1.
Suppose further, that the temperature of the ambient medium is cold enough so that the pressure of
the ambient medium is negligible compared to the pressure inside the blast wave. Thus, the ambient
temperature can be assumed to be insignificant for the evolution of the blast wave.
Let λ be a length scale giving determining radius of the blast wave at a given time t. Let us use
dimensional analysis to determine λ. Thus, assume that

λ = E a ρb1 tc.

Dimensionally L = (M L2T −2)a (M L−3)b T c = M a+b L2a−3b T c−2a


 
1
a+b=0 a=

 

  1/5
 
5 
2
 Et 

 

 
b = − 51
 
⇒ 2a − 3b = 1 ⇒ ∴λ= 
 

 

2
ρ1
c − 2a = 0 c=

 

 
5
 

Thus, the radius of the blast wave rS(t) ∼ (Et2/ρ1)1/5 within a factor of unit order. The expansion of
a blast wave has been confirmed to follow this law by observations of nuclear explosions. Thus, it gives
a valid model of the expansion of supernova remnants.
Hydrodynamics – Spring 2010 155
Similarity variable. Fluid variables at the shock

Now, consider the expansion of the gas behind the spherical blast wave. Let us consider the solution
of the hydrodynamic equations under the assumption of self-similarity, i.e., that the hydrodynamic
quantities behind the shock depend on a dimensionless similarity variable

r ρ1 !1/5
ξ= =r .
λ(t) Et2

Since the radius of the shock scales like λ, we assume that the position of the shock in terms of the
similarity variable is ξ0, i.e., that
2 1/5
 
Et
rS(t) = ξ0λ = ξ0 


 .
ρ1
Thus,
drS 2rS 2ξ0  E 1/5
 

vS(t) = = = .
dt 5t 5 ρ1t3
The shock can be treated at the limit M → ∞. Thus, the values of the hydrodynamic variables
immediately behind the shock front are

ρ2 γ + 1 vS 2
= = ⇒ v2 = vS
ρ1 γ − 1 vS − v2 γ+1
2γM2 2
p2 = p1 = ρ1vS2 .
γ+1 γ+1

Hydrodynamics – Spring 2010 156


Similarity solution

Let us, then, try to find a solution behind the shock in form

γ+1 0
ρ(r, t) = ρ2ρ0(ξ) = ρ1 ρ (ξ)
γ−1
ξ r 4 r 0
v(r, t) = v2 v 0(ξ) = v2 v 0(ξ) = v (ξ)
ξ0 rS 5(γ + 1) t
2 2
ξ r 8ρ r !2
   
0 0 1
p(r, t) = p2   p (ξ) = p2   p (ξ) = p0(ξ),
ξ0 rS 25(γ + 1) t

where ρ0, v 0 and p0 are dimensionless functions of the similarity variable.

Subsitute, next, these into to the ideal HD equations,

∂ρ 1 ∂ 2
+ (r ρv) = 0
∂t r2 ∂r
∂v ∂v 1 ∂p
+v =−
∂t ∂r ρ ∂r
∂ ∂ p
 
 + v  ln γ = 0
∂t ∂r ρ
Hydrodynamics – Spring 2010 157
dρ0 2  0 0 d
 

−ξ + 3ρ v + ξ (ρ0v 0) = 0
dξ γ + 1 dξ
0 0 0
   
2 dv 4  02 0 dv  2(γ − 1) 1  0 dp 
∴ −v 0 − ξ + v + v ξ  = −
0
2p + ξ 
5 dξ 5(γ + 1) dξ 5(γ + 1) ρ dξ
d  p0  5(γ + 1) − 4v 0
 

ξ ln 0γ  =
dξ ρ 2v 0 − (γ + 1)

Thus, as r and t are eliminated from the system of equations, our trial form of the dependent variables
works. The task is now to solve this system of first-order ordinary differential equations with the boundary
conditions at the shock
ρ0(ξ0) = v 0(ξ0) = p0(ξ0) = 1.
The value of ξ0 can be found by requiring that the total energy of the blast wave remains constant:
ˆ r 
S 1 p

E=  ρv 2 +  4πr 2 dr
0 2 γ−1
ˆ ξ0
32π 0 0 02 4
⇒ 1= (p + ρ v ) ξ dξ.
25(γ 2 − 1) 0

Solving the equations for the primed quantities gives the similarity solution for the blast wave that can
be scaled to any values of E and ρ1.
Obviously, the solution is not valid (1) at the earliest times, as it predicts v2, vS → ∞ with t → 0, and
(2) at the latest times, when radiative cooling renders the assumption of constant energy invalid.

Hydrodynamics – Spring 2010 158


Plot of the similarity solution

Hydrodynamics – Spring 2010 159


One-dimensional gas flow

Let us consider, next, steady-state solutions involving supersonic flows.


Consider a one-dimensional, steady, adiabatic gas flow through a pipe with variable cross section. The
governing equations are:

1 d
(Avρ) = 0
A dx
d p dp 2 dρ γp
 

= 0 ⇒ = c s with c2s =
dx ργ dx dx ρ
dv 1 dp c2s dρ
v =− =−
dx ρ dx ρ dx

Thus,

c2s dρ 1 dA 1 dv 
 

− = c2s  +
ρ dx A dx v dx
1 dv 1 dA v
⇒ (1 − M2) =− , where M = is the local Mach number.
v dx A dx cs

If the flow is subsonic, then dv/dx and dA/dx have opposite sign and decreasing cross-sectional area
leads to an accelerating flow (as intuitively expected). However, if the flow is supersonic, increasing area
yields (counterintuitively) an accelerating flow.
Hydrodynamics – Spring 2010 160
de Laval nozzle

We found that
1 dv 1 dA
(1 − M2) =− .
v dx A dx
Consider a case, where the flow starts as
subsonic and ends up supersonic. This
requires that the sonic point M = 1 is
reached at dA/dx = 0 and that the pipe
has to first decrease and then increase in
cross section. This kind of a pipe is called
de Laval nozzle and it is used in rocket
engines. The exit velocity of the gas is
v
 (γ−1)/γ 
2γ RTi 
u
p
u 
e
1 −
u
ve = u  .


γ−1
t
pi

Hydrodynamics – Spring 2010 161


Extragalactic jets

Radio galaxies often


appear with huge jets.
These can be modeled as
being channeled through
de Laval nozzle -like
geometry. (Blandford &
Rees 1974)

Hydrodynamics – Spring 2010 162


Hydrodynamics – Spring 2010 163
Spherical accretion and winds

Let us, finally, treat the problem of mass flow onto or from a massive object in spherical symmetry, i.e.,
spherical accretion or spherical wind. These problems were first treated by Bondi (1952) and Parker
(1958), respectively. While the problem of spherical accretion is mainly of academic interest, a spherical
stellar wind is a qualitatively reasonable first approximation to the problem of the expansion of a non-
confined hot stellar atmosphere.
Consider a steady flow in spherical geometry under the effect of grqvitational field of a massive central
object. The governing equations are:

1 ∂ 2 2 1 dρ 1 dv
(r ρv) = 0 ⇒ + + =0
r2 ∂r r ρ dr v dr
dv 1 dp GM c2s dρ GM 1 dv 2  GM
 

v =− − 2 =− − 2 = c2s  + − 2
dr ρ dr r ρ dr r v dr r r

For simplicity, consider the isothermal expansion/accretion, where c2s = RT is a constant. (The case of
polytropic expansion/accretion is left as an exercise.) Thus,

c2s 
dv 2c2s GM
 

v− 

 = − 2 .
v dr r r

Clearly, v = cs can be realized only at


GM
r = rc ≡
2c2s
Hydrodynamics – Spring 2010 164
The equation is easily integrated as

cs !2 cs ! 2 r 2GM
− ln = 4 ln + + C,
v v rc r c2s

where C is a constant of integration. The solutions are depicted below.

Solutions I and II are double valued


and, hence, unphysical. Solution III
would correspond to supersonic flow
everywhere, and this does not corre-
spond to either the wind or the accre-
tion problem. Solution IV is every-
where subsonic. It would correspond
to a stellar breeze in the outflow case.
Transsonic solutions are obtained
setting C = −3. Solution V is
Parker’s solar (or stellar) wind and
the other transsonic solution corre-
sponds to Bondi’s spherical accre-
tion.

Hydrodynamics – Spring 2010 165


Waves and instabilities

Hydrodynamics – Spring 2010 166


Perturbation analysis

Consider a ball at rest on a horizontal point of a surface, as depicted below. All cases represent equilibrium
states.
In the leftmost case, the equilibrium is clearly unstable: even a slight perturbation of the position of
the ball will lead to a destruction of the equilibrium and the ball will roll away from the hill top.
In the case in the middle, the equilibrium is stable: if displaced from its equilibrium position and let to
move, the ball will eventually return to the bottom of the valley. Before that, it will most likely overshoot
to the other side of the equilibrium position and oscillate aroud the equilibrium position until frictional
forces bring it back to rest.
The equilibrium in the rightmost case is stable against small perturbations but unstable against large
perturbations. Such a state is called conditionally stable or metastable.
Obviously, it is not enough for a system to be an equilibrium solution of the equations of motion: in order
for the system to represent a state to be found in Nature by any reasonable probability, the solution has
to be stable against perturbations. The most robust equilibria are those that are stable against arbitrary
perturbations.
A stability analysis is also necessary in hydrodynamics to find out if any found mathematical equilibrium
state is going to represent a physical equilibrium or not.

Hydrodynamics – Spring 2010 167


Perturbation analysis for fluids

We already introduced the linearization technique, when discussing the acoustic waves. The same
technique can be applied in analysing the possible unstable growth of perturbations around any equilibrium
solution of the HD equations.
As for the acoustic waves, one can then apply Fourier analysis (i.e., write down the perturbed part of
the solution as ∝ exp{i(k · x − ωt)}) and obtain, using the HD equations, a dispersion relation for the
perturbations in form
ω = ω(k).
If the solution contains a positive imaginary part, i.e.,

ω(k) = ωr (k) + iωi(k) with ωi > 0 for some values of k

then the perturbations at that particular wavenumber grow exponentially as ∝ exp{ωi(k)t} and the
underlying equilibrium solution is unstable. If ωi < 0, then the perturbations are damped and the
equilibrium is stable against perturbations at that wavenumber.
In some cases we find that the equilibrium is stable or unstable, depending on the values of some
parameters of the equilibrium (conditional stability). As an example, we mention the onset of convection
in a vessel heated from below, once the temperature gradient exceeds a threshold value.
Of course, one can justify the linearization of HD equations only if the perturbations are small. Thus,
the linear perturbation analysis can only be applied to the initial evolution of unstable perturbations.
Eventually, the perturbations grow and we may finally end up with an irregular state (in space and time)
usually referred to as turbulence.
Hydrodynamics – Spring 2010 168
Acoustic waves in a viscous fluid

Let us consider a viscous Newtonian fluid, for which the complete N–S equation in the absence of external
forces reads
dv
ρ = −∇p + (ζ + 13 µ)∇(∇ · v) + µ∇2v.
dt
Consider the evolution of adiabatic perturbations on top of a static equilibrium solution: ρ = ρ0 +ρ1(x, t),
p = p0 + p1(x, t) and v = v 1(x, t). Linearizing the N–S equation and eliminating the pressure using
p1 = c2s ρ1 and the density using the equation of continuity gives

∂ 2v1 ζ + 13 µ  ∂v 1  µ 2 ∂v 1
 
2
= cs ∇(∇ · v 1 ) + ∇ ∇ · + ∇
∂t2 ρ0 ∂t ρ0 ∂t

where c2s = γp0/ρ0. Seeking for longitudinal solutions in form v = v1ei(kx−ωt)ex gives

2 ζ + 43 µ 2
ω = c2s k 2 −i k ω, i.e.,
ρ0
v
u 2
4
ζ + 43 µ 2

ζ + 3µ 
u
u
ω = ±csk 1 −  k − i k
u
u
2ρ0cs 2ρ0
t

Clearly, the equilibrium is stable (ωi < 0 at all k ). Acoustic waves also become dispersive in a viscous
fluid and cease to propagate (ωr = 0) at k > 2ρ0cs/(ζ + 43 µ). At large wavelengths (low k ) the results
derived from Euler equation are recovered.
Hydrodynamics – Spring 2010 169
Convective instability

Consider a perfect gas in hydrostatic equilibrium in a uniform gravitational


field, g = −gez . Thus, p(z) and ρ(z) decrease upward with z . Assume
that a blob of gas is displaced vertically. Let the original pressure and ρ *, p’ ρ’, p’
denisity of the blob (and the surrounding gas) be p and ρ, respectively.
Let the external pressure and density at the new position be p0 and ρ0.
Because acoustic waves transmit momentum in the fluid quite fast, it is
reasonable to assume that the pressure inside the blob has equilibriated
with the surroundings. However, heat transport via conduction is usually
a slower process, so it is reasonable to assume that the density of the blob
ρ, p ρ, p
behaves adiabatically. Denote the density at the new position by ρ∗.

0 1/γ
 
p dp
ρ∗ = ρ 
  and p0 = p + ∆z.
p dz
ρ dp
Thus, ρ∗ = ρ + ∆z
γp dz

On the other hand,


dρ ρ dp ρ dT 
 
ρ=p/RT
ρ0 = ρ + ∆z = ρ+ − ∆z
dz p dz T dz
1 ρ dp ρ dT 
   
0 ∗
∴ ρ − ρ = − 1 −  + ∆z
γ p dz T dz
Hydrodynamics – Spring 2010 170
We got
1 ρ dp ρ dT 
   

ρ0 − ρ∗ = − 1 −  + ∆z
γ p dz T dz
Clearly, if the difference is positive, the blob will be buoyant and the system is unstable. Thus, we derive
the stability criterion of the atmosphere as

dT  1  T dp
 

< 1− Schwarzschild criterion.


dz γ p dz

If the temperature profile is steeper than this limit, then the atmosphere will be unstable to convection,
i.e., up- and downward movement of gas. The Schwarzschild criterion is important to consider when
discussing models of the internal structure of stars.
An eq. of motion for the blob is
2
∗ d ∆z
ρ = −(ρ∗ − ρ0)g
dt2
where we neglect any viscous forces arising from the movement of the blob relative to the surrounding
gas. Thus,
d2∆z
2
+ N 2∆z = 0,
dt
where v
u g dT 1 T dp
u    

N =u − 1 −  

t 
T dz γ p dz
is the Väisälä–Brunt frequency. Clearly, if the atmosphere is stable, the blob ends up in oscillatory motion.
A full analysis of this motion will reveal that the stably stratified atmosphere will support internal gravity
waves in addition to ordinary acoustic waves.

Hydrodynamics – Spring 2010 171


Rayleigh–Bénard convection

Experimentally (Bénard 1900) it is found that a layer of liquid heated from below becomes unstable to
convection, which occurs in cells with dimensions of the order of the depth of the layer, as the temperature
gradient increases above a certain value. Let us, next, try to understand this theoretically. This requires
us to analyze the problem using full, nearly incompressible N–S equation (Rayleigh 1916).

Hydrodynamics – Spring 2010 172


Rayleigh–Bénard convection - set up

Consider a layer of liquid between z = 0 and z = d, in a gravitational field g = −gez , heated from below.
Assume that the liquid is nearly incompressible, i.e., that its density does not change in reaction to
varying pressure, but decreases as a result of increasing temperature.
The energy equation for a nearly incompressible fluid reads

∂
 

ρ  + v · ∇ = ∇ · (K∇T ),
∂t
where the internal energy is  = cpT and cp is the specific heat. Thus,

∂T K
+ v · ∇T = κ∇2T, where κ = is the thermometric conductivity.
∂t ρcp

Let Tb and Tt be the temperatures at the bottom and the top of the layer. Thus, the equilibrium solution
is
Tb − Tt
T0(z) = Tb − βz, where β = .
d
The corresponding density is
ρ0(z) = ρb(1 + αβz),
where α is the coefficient of volume expansion with temperature. The equilibrium pressure satisfies the
hydrostatic balance equation
dp0
= −ρ0(z) g.
dz
Hydrodynamics – Spring 2010 173
Next, introduce perturbations around the equilibrium state. If T = T0 + T1, then ρ = ρ0 − ρbαT1. Write
p = p0 + p1 and subsitute in the N–S equation to get

∂v 1
 

(ρ0 − ρbαT1)  + v 1 · ∇v 1 = −∇(p0 + p1) + (ρ0 − ρbαT1)g + µ∇2v 1.


∂t

Clearly, ∇p0 = ρ0g . We may also neglect the non-linear terms v 1 · ∇v 1 and ρbαT1 ∂v 1/∂t, and replace
ρ0 by the constant ρb on the left-hand side, as it multiplies a small term (Boussinesq approximation).
With these simplifactions, we get

∂v 1
ρb = −∇p1 − ρbαT1 g + µ∇2v 1 (LNS)
∂t
Substituting the perturbed variables in the temperature equation and linearising we get

∂T1
= βv1z + κ∇2T1. (T)
∂t
There are five variables (v 1, p1 and T1) and four equations in (LNS–T), but the system could be closed
by using the kinematic constraint ∇ · v 1 = 0. However, it is also possible to eliminate extra variables by
taking a curl of (LNS) twice and then considering the z component:
2
∂ 2T1 
 
∂ 2  ∂ T1 4
(∇ v1z ) = αg  2 +  + ν∇ v1z . (V)
∂t ∂x ∂y 2

Equations (T) and (V) form a closed system with v1z and T1 as the only unknowns.

Hydrodynamics – Spring 2010 174


Rayleigh–Bénard convection – marginal stability

We derived
∂T1
= βv1z + κ∇2T1
∂t
2 2
 
∂ 2 ∂ T1 ∂ T1 4
(∇ v1z ) = αg  +  + ν∇ v1z .
∂x2 ∂y 2

∂t

We can try to find solution of these equations in form

v1z = W (z) exp(σt + ikxx + iky y)


T1 = θ(z) exp(σt + ikxx + iky y)

We expect the solution to be periodic in the xy plane and allow for a general time dependence (σ can, in
principle, be complex). The functions W (z) and θ(z) then satisfy ordinary differential equations obtained
by substituting the trial forms into the equations. Of course, if σr > 0, then the perturbation grows and
if σr < 0 it decays. The limiting case is σr = 0, which is called marginally stable. In the following, we
shall assume that σ = 0 to find a criterion for the onset of the instability. This gives
2
 
 d
βW + κ  2 − k 2
θ = 0
dz
2
2

 d
−αgk θ + ν  2 − k 2
2
 W = 0, where k 2 = kx2 + ky2.
dz
Hydrodynamics – Spring 2010 175
Thus, 3
2

d
− k2
νκ 
2
  W = −αβgk W.
2
dz
Substituting z = z 0d and k = k 0/d, i.e., using the layer depth as the unit of distance, gives
3
2

d 02 02

− k  W = −Rk W,
dz 02

αβgd4
where R = is the (dimensionless) Rayleigh number of the system.
κν
For a correct physical solution of this sixth-order equation, one needs six boundary conditions. Four
of them are immediately obtained as both W = 0 and θ = 0 (i.e., [(d/dz 0)2 − k 02]2W = 0) have to
be required both at z = 0 and z = d. The remaining two boundary conditions are obtained from the
requirement that the tangential stresses at a free surface have to vanish, leading6 to d2W/dz 02 = 0,
and that the no-slip condition has to be maintained at a solid boundary, leading to dW/dz 0 = 0. The
simplest solution is obtained, when both boundaries are considered as free surfaces (e.g., liquid floating
on top of a heavier liquid), whence d2W/dz 02 = 0 is taken to apply at both surfaces. Such boundary
conditions are satisfied only by
W (z 0) = W0 sin πnz 0, n ∈ N,
with R and k 0 related by
(n2π 2 + k 02)3
R= .
k02

6 see Chandrasekhar’s Hydrodynamic and hydromagnetic stability, Dover 1981, p. 21–22.

Hydrodynamics – Spring 2010 176


We obtained as a condition of marginal stability that

(n2π 2 + k 02)3 αgd3∆T


R= where R = .
k02 κν
Intuitively, an increase of the temperature difference yieds to instability, so values of R that fall above
this value for given n and k correspond to an instability. Clearly, the most unstable situation corresponds
to n = 1. Thus, the curve
(π 2 + k 02)3
R=
k02 R
separates the unstable parameter range (above the curve)
from the stable range (below the curve) in the k 0R-plane.
The separating curve has a minimum value at
Unstable
4
π 27π
k 0 = kc0 = √ with R = Rc = ≈ 657.5.
2 4
Rc
For a Rayleigh number above this limit, there always exists Stable
an unstable range of wavenumbers.
One can find that for two rigid boundaries, the critical
k
Rayleigh number is 1707.8 and for one rigid and one free kc
boundary, it is 1100.7. Most experiments yield Rc ≈
1700.

Hydrodynamics – Spring 2010 177


Perturbations at a two-fluid interface

Consider a horizontal, planar interface of two fluids (which may be in uniform horizontal motion) in a
vertical gravitational field. (Example: the surface of a lake or a pond.) Let us study the evolution of
small perturbations of the bounding surface.
We will consider the following simplifications of the problem:

• Both fluids are assumed incompressible


• Both fluids are assumed ideal
• Surface tension is neglected

As we assume incompressible and ideal fluids, Kelvin’s theorem of vorticity applies. Thus, assuming that
the steady-state flow is irrotational also the perturbations of the fluid velocity have to remain such. We
can, therefore, assume that the velocity on both sides of the interface is obtained from a scalar potential,
v = −∇φ. Thus,
∂φ v2 p
−∇ + ∇ = −∇ − ∇Φ,
∂t 2 ρ
where Φ = gz is the gravitational potential. Integrating the equation gives

∂φ 1 2 p
− + v + + Φ = F (t),
∂t 2 ρ

where F (t) is an arbitrary function of time. Because of the assumption of irrotational velocity (at t = 0)
this equation, resembling Bernoulli’s principle, applies throughout the two fluids.
Hydrodynamics – Spring 2010 178
Consider a two-fluid interface (densities ρ and ρ0)
U’, ρ’ with uniform flows in the x direction (speeds U and
U 0), as depicted in the figure. This configuration
satisfies the steady-state ideal HD equations, as long
z
ξ1 as pressure is continuous at the interface, i.e.,
x 


 p0(0) + gρ0z, z > 0
p0(z) =  ,
 p0(0) + gρz, z<0

where z = 0 has been chosen as the position of the


U, ρ interface in steady state.
Consider small perturbations of the interface and denote its perturbed position by

z = ξ1(x, t).

The aim is to find out whether ξ1 grows with time, decays or oscillates. Velocity potential below the
surface is
φ = −U x + φ1(x, z, t),
where the first term gives the uperturbed part. The perturbed part satisfies ∇ · v 1 = 0, i.e.,

∇2φ1 = 0.

Above the surface,


φ0 = −U 0x + φ01(x, z, t) with ∇2φ01 = 0.
Hydrodynamics – Spring 2010 179
For fluid elements at an infinitesimally small distance from the interface, we can write

∂ξ1 ∂ξ1
vz = dξ1/dt = + vx .
∂t ∂x
Thus, to the linear order in small quantities,

∂φ1 ∂ξ1 ∂ξ1 ∂φ01 ∂ξ1 ∂ξ1


− = +U and − = + U0 at z = 0.
∂z ∂t ∂x ∂z ∂t ∂x
Consider the following Fourier components

ξ1 = Aei(kx−ωt)
φ1 = Cei(kx−ωt)+lz
0
φ01 = C 0ei(kx−ωt)−l z

Apply the Laplace equations to the velocity potentials to get

−k 2 + l2 = 0
2
−k 2 + l0 = 0

which give l = l0 = |k|, where the signs have been chosen so that the perturbations vanish at infinity.

Hydrodynamics – Spring 2010 180


Consider k > 0. Applying the boundary conditions at the interface then gives,

−kC = i(kU − ω)A


kC 0 = i(kU 0 − ω)A

Dividing the equations gives


C kU − ω
− =
C 0 kU 0 − ω
so we clearly need one more equation to find out the ratio of the amplitudes of the velocity potentials
in order to arrive to a dispersion relation for the perturbations. This is provided by the requirement that
pressure must be continuous accross the interface.
We have
2
 
∂φ1 v
p = −ρ 
− + + gz 
 + ρF (t)
∂t 2
02
 
∂φ01 v
p0 = −ρ0 
− + + gz  0 0
 + ρ F (t)
 

∂t 2

Comparing with the steady-state solution (that needs to be recovered far from the interface where
v 1 = 0), we get
2
ρF (t) = p0(0) + 21 ρU 2 and ρ0F 0(t) = p0(0) + 12 ρ0U 0 .

Hydrodynamics – Spring 2010 181


Thus, applying p = p0 at the boundary, we have

02 02
 
2 2
∂φ01
 
∂φ1 v − U 0 v −U
−
ρ + + gξ1 −
 = ρ  + + gξ1

∂t 2 ∂t 2

and, using v 2 = (U − ∂φ1/∂x)2 + (−∂φ1/∂z)2 ≈ U 2 − 2U (∂φ1/∂x), this gives


0 0
 
∂φ1 ∂φ1 0  ∂φ1 0 ∂φ1
 

ρ −
 −U + gξ1 = ρ −
 −U + gξ1
 at z = 0.
∂t ∂x ∂t ∂x

Thus,
ρ[−i(kU − ω)C + gA] = ρ0[−i(kU 0 − ω)C 0 + gA].
Substituting C and C 0 as a function of ω , k and A from above gives

ρ(kU − ω)2 + ρ0(kU 0 − ω)2 = kg(ρ − ρ0)

as the dispersion relation for the perturbations. Solving for the phase speed gives

v
0 0
g ρ − ρ0 ρρ0(U − U 0)2
u
ω ρU + ρ U u
± −
u
= u
k ρ + ρ0 k ρ + ρ0 (ρ + ρ0)2
t

This general equation can be applied to several special cases of interest to derive the appropriate dispersion
relations for several two-fluid-interface phenomena.

Hydrodynamics – Spring 2010 182


Surface gravity waves

Consider two fluids at rest with a lighter fluid lying above a heavier one. Thus, U = U 0 = 0, ρ > ρ0 and
v
ω g ρ − ρ0
u
u
=± u
t .
k k ρ + ρ0

Clearly, the perturbation of the surface moves as a wave (surface wave) with a phase velocity that depends
on the wavenumber. Thus, surface waves are dispersive such that the longest waves have the highest
phase speeds.
If one considers a case where ρ0  ρ, like air on ρ’ < ρ
top of water, the equation simplifies to t 1 > t0
ω/k
r
z
ω = ± gk.
x
This expression neglects two effects that you are
requested to take into account as exercises: ρ

• Surface tension will somewhat increase the propagation speed. Intuitively this is clear, as the restoring
force (trying to restore the “flatness“ of the surface) will be greater as a result of this extra tension
acting on the separating surface.
• A finite depth h of the lower fluid will modify the phase speed of waves with k −1 & h. This provides
an upper limit to the phase speed.

Hydrodynamics – Spring 2010 183


Rayleigh–Taylor instability

Let us still consider fluids at rest. Now assume,


however, that the heavier fluid is on top of the
ρ’ > ρ
lighter one, i.e., ρ < ρ0. For a perfectly flat, hori-
zontal separating surface, the configuration is still t 1 > t0
an equilibrium solution of the HD equations. We, z
however, know from experience that the heavier
fluid should fall below the lighter one, so that the x
equilibrium with the heavy fluid on top should be
ρ
unstable. This we also find from the dispersion
relation:
v v
ρ − ρ0 |ρ0 − ρ|
u u
u u
ω = ± gk = ±i gk
u u
t u
ρ + ρ0 ρ + ρ0
t

which corresponds to purely imaginary frequency, i.e., a purely growing (and a decaying), non-propagating
perturbation. This instability is called Rayleigh–Taylor instability.
We note that because of the equivalence principle, a situation corresponding to Rayleigh–Taylor instability
can also be generated in an accelerating frame of reference. If there is no gravitational field but the
interface is propagating in the −z direction with a decelerating speed in the inertial frame, we find a
heavy fluid pushing against a light one, and the same instability analysis applies.

Hydrodynamics – Spring 2010 184


The Crab nebula (and many other supernova remnants) provide us with examples of such a two-fluid
interface. The ejected gas from the explosion tends to pile up in a thin layer just behind the interface
between the ejecta and the surrounding (shocked) gas. As the interface is decelerating, this creates an
effective outward g -field in the frame of the interface. We, thus, have a situation, where the lighter fluid
is below the heavy on in this effective g -field and the situation is Rayleigh–Taylor unstable.

Hydrodynamics – Spring 2010 185


Kelvin–Helmholtz instability

Finally, let us consider the case with non-zero velocities. Let us restrict our attention to the Rayleigh–
Taylor stable case (ρ > ρ0). Thus, in the coordinate system co-moving with the lower fluid
v
0
ρ − ρ0 ρρ0k 2∆U 2
u
ρ ∆U u
u
ω= k ± gk u
− ,
ρ + ρ0 ρ + ρ0 (ρ + ρ0)2
t

where ∆U = U 0 − U is the velocity of the upper fluid with respect to the lower fluid. (Note that this
relation applies also if U and U 0 are not co-aligned, whence ∆U = |U 0 − U |.)
It is clear that if

ρρ0k∆U 2 > g(ρ2 − ρ0 )


2
ρ’ < ρ U’ − U

ω is complex. The perturbation is both propagat- t 1 > t0


z
ing at speed ωr /k = ρ0∆U/(ρ + ρ0) and growing
exponentially at rate x

ρ
v
ρρ0∆U 2 g ρ − ρ0
u
u
u
ωi = k u

(ρ + ρ0)2 k ρ + ρ0
t

This instability is called Kelvin–Helmholtz instability. Note that in the absence of gravitational field, an
ideal two-fluid interface with a tangential velocity difference (i.e., a tangential discontinuity) is always
Kelvin–Helmholtz unstable.
Hydrodynamics – Spring 2010 186
Beyond the linear theory of waves and instabilities

Our analysis of waves and instabilities has been restricted by the assumption that perturbation are small
so that the equations can be linearised around the equilibrium solution. Only acoustic waves were treated
with a simplified account of non-linearity, and shock waves, i.e., thin transition fronts were shown to be
produced as a result of non-linear steepening.

Dispersive waves (such as surface waves) will behave oppositely: their wave packets will be dispersed
as the different Fourier components have different velocities. Sometimes a balance between dispersive
effects and non-linear effects may be found. The resulting waves, called solitary waves, retain their shape
during propagation. If solitary waves are, in addition, able to survive collisions they are called solitons.

Consider a surface wave propagating in a shallow water of depth h  k −1. In the linear limit, it has the
disperison relation (exercise)

1 2 2
r r   r
ω = gk tanh kh ≈ k gh 1 − k h ≡ c0k − σk with c0 = gh and σ = 16 c0h2
 3
6
A wave fulfilling this dispersion relation would have to satisfy the differential equation

∂v ∂v ∂ 3v
+ c0 + σ 3 = 0.
∂t ∂x ∂x
Adding the non-linear term v(∂v/∂x) and transforming to a coordinate system moving with the linear
Hydrodynamics – Spring 2010 187
wave speed c0 [i.e., using X = x − c0t] gives

∂v ∂v ∂ 3v
+v +σ = 0,
∂t ∂X ∂X 3
which is the celebrated Korteweg–de Vries equation that admits a soliton solution of form

3v0
v= .
cosh2[ 12 (v0/σ)1/2(X − v0t)]

Such soliton solutions are localized


r
wave pulses
r
that retain their shapes. Their amplitude, 3v0, propagation
speed, (c0+) v0, and width, 2 σ/v0 = h 2c0/3v0, are all related to each other. More on solitons during
the student seminar.
In the non-linear regime of instabilities two kinds of behavior may be expected. The instability may be
limited by non-linear effects so that the amplitude of the perturbation stops growing after a certain limit,
often depending on the system parameter driving the instability. In this case a new stable laminar solution
may be achieved, replacing the original equilibrium. Rayleigh–Benard convection at Rayleigh numbers
close to the critical value provides an example. It is found that the amplitude of the convection velocity
increases linearly with R − Rc so when the temperature gradient is increased, more and more energy is
transported by convection.
Another possibility is that no laminar state will be achieved but that the unstable system develops an
irregular flow (in space and time) that can be analysed only in a statistical sense (e.g., flow past a body
at the largest Reynolds number or convection at the largest Rayleigh numbers). These kind of systems
are referred to as being turbulent.

Hydrodynamics – Spring 2010 188


Turbulence

Hydrodynamics – Spring 2010 189


The need for a statistical theory

P Let us consider the trajectories of a dynamical


q2 system in phase space around an unstable equi-
librium point P . If the system is excactly at P
at time t = 0 it will stay there. If, on the other
hand, the system is even slightly displaced from
this point, the distance from the point will start
to grow exponentially and after a finite amount
of time the system lies in a very different region
of phase space. It will, therefore, be practically
q1 impossible to predict the state of the system
deterministically after a finite amount of time.
Thus, for an unstable fluid system, we expect that
the growth of perturbations may lead to a loss of
q3 our prediction capability.
One often encounters in practice systems in which fluid velocities seem to vary randomly in space and
time. Such a state of a fluid is called turbulence. In addition to hydrodynamically unstable fluids,
stable fluids stirred with random forces may develop turbulence.
Since it is impossible to develop a deterministic theory of turbulence, we wish to have a statistical de-
scription of its properties based on averages. Ensemble averiging would be most appropriate theoretically,
observationally temporal averaging is most easily realised. For ergodic systems, the two are equivalent.
We shall assume ergodicity and denote statistical averaging by an overbar.
Hydrodynamics – Spring 2010 190
Correlation length

The velocity in each point can be written as

v = v̄ + v 0,

where v̄ and v 0 are the statistical average and the fluctuating part. Obviously,

v 0 = 0.

Thus, a more complicated average of the fluctuating field is needed to describe its statistical properties.
Maybe the simplest quantity to consider, proposed by Taylor (1935), is

v 0(x, t) · v 0(x + r, t).

At r = 0, this equals v 02(x, t) which is proportional to the average kinetic energy density of the fluctu-
ations. At r → ∞, the fluctuating velocities are uncorrelated, so we get

lim v 0(x, t) · v 0(x + r, t) = v 0(x, t) · v 0(x + r, t) = 0


r→∞

Thus, v 0(x, t) · v 0(x + r, t) has a substantial value only if r . a finite distance. This distance is called
the correlation length of the turbulence. Thus, correlation function like v 0(x, t) · v 0(x + r, t) contain
the information on the strength and correlation length of the turbulence and are appropriate measures
of its properties.
Hydrodynamics – Spring 2010 191
Correlation tensor

The velocity correlation tensor of the turbulent flow is defined as

R(r; x, t) = v 0(x, t)v 0(x + r, t); i.e., Rij (r; x, t) = vi0 (x, t)vj0 (x + r, t)

In general, it has nine components but symmetries can reduce the number of independent components
drastically. From now on, we will not denote the depence on time explicitly if both velocities are evaluated
at the same time.
We already concluded that in case of unstable equilibria, it is impossible to prescribe the instanta-
neous values of the fluctuating fields. But what about the velocity correlations? A proper dynamic
theory of turbulence would give the values of Rij (r; x) (and higher-order correlation function such as
vi(x1)vj (x2)vk (x3)) from given initial and boundary conditions for the mean variables and statistics of
the fluctuations. Such theory does not exist yet.
In the following, we will consider incompressible, homogeneous and isotropic turbulence. The second
assumption means that we assume the correlation tensor components not to depend on x. The third
assumption means that the medium has no preferred direction, i.e., that the only vector the correlation
tensor can depend on is r . It implies also that v̄ = 0. Homogeniety and incompressibility imply that

Rij (r) = Rij (r; x) = Rij (r; x − r) = vi(x − r)vj (x) = Rji(−r)
∂Rij ∂vj (x + r) ∂Rji(r)
= vi(x) = vi(x) (∇ · v)(x + r) = 0 ⇒ =0
∂rj ∂rj ∂rj
Hydrodynamics – Spring 2010 192
The form of the correlation tensor in isotropic turbulence can be derived as follows. Consider a Cartesian
coordinate system K̃ , where one of the base vectors, say ẽ3, is directed along r = rẽ3. Denote
the correlation function R̃33 = ṽ3(x)ṽ3(x + rẽ3) = 31 v 2f (r) with f (0) = 1. Because of symmetry,
the correlation functions of the velocity components perpendicular to r have to agree, i.e., R̃ij =
ṽi(x)ṽj (x + rẽ3) = 31 v 2g(r) with i = j = 1, 2 and g(0) = 1. Off-diagonal components of R̃ij vanish in
this coordinate system as we may assume that the fluctuating velocity components are independent of
each other. Thus,  
g(r) 0 0 
v 2 
(R̃ij ) =  0 g(r) 0  .
3 
0 0 f (r)

Transforming to a Cartesian coordinate system with another orientation of axes ei, the tensor components
transform as
Rij = Mir MjsR̃rs,
where Mij = ei · ẽj is an orthogonal tensor, i.e., M · MT = I ⇔ Mik Mjk = δij . Thus,

Rij = 13 v 2[(Mi1Mj1 + Mi2Mj2)g(r) + Mi3Mj3f (r)]


= 13 v 2{Mik Mjk g(r) + Mi3Mj3[f (r) − g(r)]} | Mik Mjk = δij
= 13 v 2{δij g(r) + Mi3Mj3[f (r) − g(r)]}

As Mi3 = ei · ẽ3 = ei · (r/r) = ri/r, we have

1 2
(
rirj )
1 2
(
rr )
Rij = 3v δij g(r) + 2 [f (r) − g(r)] i.e., R = 3v g(r)I + [f (r) − g(r)] 2 .
r r

Hydrodynamics – Spring 2010 193


We can derive a connection between f (r) and g(r) using the incompressibility condition, i.e,

∂Rij 1 2  dg ∂r rirj  df dg  ∂r ∂ rirj !


   

0= = 3v  δij + 2 − + [f (r) − g(r)]


∂ri dr ∂ri r dr dr ∂ri ∂ri r2 
 dg ri r r df dg r ∂ r r
   !
i j i k j
= 31 v 2 

δij + 2  −  + [f (r) − g(r)] δki 2 
dr r r dr dr r ∂ri r
 dg rj r r df dg r ∂ r r
   !
1 2 i i j k j 
= 3v  + 2  −  + [f (r) − g(r)]δki
dr r r dr dr r ∂ri r2 
 df rj δ r + r δ r r r
  
ik j k ij k j i 
= 13 v 2  + [f (r) − g(r)]δki  − 2
dr r r2 r3 r 
  
1 2 df rj  (δki δik + 1)rj rirj ri 

 
= 3v  + [f (r) − g(r)] − 2
r2 r3 r 
 
 dr r 

    
 df rj [(I · I)kk + 1]rj rj 
 rj df 2[f (r) − g(r)]
 
= 31 v 2  = 13 v 2
  
+ [f (r) − g(r)]  − 2 +
r2 2 
 
 dr r r  r 
 dr r 

giving f
r df 1.0
g(r) = f (r) +
2 dr
r df  rirj df 
   

∴ Rij (r) = 13 v 2 δij f (r) +



− .
2 dr 2r dr 
r
Thus, the correlation tensor in incompressible, homogeneous, isotropic turbulence is fully determined by
the longitudinal correlation function, f (r). It should look as sketched above.

Hydrodynamics – Spring 2010 194


Turbulent diffusion

Convection sets on in a fluid heated from below to ensure efficient transport of heat through the fluid.
Under normal conditions, convection is orders of magnitude more efficient method of heat transport than
conduction. We should ask, therefore, what is the role of the turbulent velocity fields in heat transport.
More generally, it is of interest to study the effect of turbulent velocity fields to transport phenomena.
Let us look at the problem of heat transport again. The energy equation in an incompressible fluid could
be written in form
∂T K
+ v · ∇T = κ∇2T, where κ = is the thermometric conductivity.
∂t ρcp

If κ ≈ 0 (its smallness will be discussed later), we find that temperature is a passively advected scalar
in the fluid. This means that its value for each fluid element is fixed. Thus, if the fluid elements should
diffuse as a result of turbulent velocity fields, we find that turbulence would provide a diffusion term in
the energy equation even without heat conduction.
Let us consider the displacements of the fluid elements,
ˆ t
ξ(t) = v L(t0) dt0,
0

where v L(t) = v(x0 + ξ(t), t) is the velocity of the fluid element, intially at x = x0, in the Lagrangian
description and v(x, t) is the turbulent velocity field. (For simplicity, we’ll assume v̄ = 0.)
Hydrodynamics – Spring 2010 195
Obviously, the mean displacement is zero, ξ = 0. However,
ˆ t ˆ t ˆ t ˆ t
ξ2 = vL (t0) dt0 · vL (t00) dt00 = dt0 dt00 v L(t0) · v L(t00)
0 0 0 0
ˆ t ˆ t
0
= dt dt00 v L(t0) · v L(t00)
0 0

is non-zero. Here, v L(t0) · v L(t00) is a measure of correlation between the velocities at the positions of
a fluid element at times t0 and t00. If the turbulent velocity field is in a steady state, the value of the
correlation may depend only on the time difference, τ = t00 − t0. Thus, we may assume

v L(t0) · v L(t00) = v 2R(τ ).

Clearly, R(0) = 1 and R(τ ) should have a substantial value only if τ is smaller than or of the order of a
correlation time, τ . τcor. We may also take R(τ ) = R(−τ ) if we assume homogeniety of turbulence.
Thus,
ˆ t ˆ t ˆ t ˆ t−t0
ξ 2(t) = dt0 v 2 dt00 R(τ ) = dt0 v 2 dτ R(τ ).
0 0 0 −t0

For t  τcor, R(τ ) ≈ 1 and ξ 2 ≈ v 2t2. Now, consider times t  τcor. Thus, for overwhelmingly most
values of t0, the limits of the inner integral may be replaced with ±∞ and we get
ˆ t ˆ +∞ ˆ +∞
0
ξ 2(t) = dt v 2 dτ R(τ ) = 2t v 2 dτ R(τ ).
0 −∞ 0

Hydrodynamics – Spring 2010 196


We have found that for times t  τcor
ξ 2(t) = 6DTt,
where ˆ +∞
1
DT = 3 v2 dτ R(τ ) =: 13 v 2 τcor
0
The linear time dependence of the mean square displacements is a signature of a diffusive process.
Consider a diffusion process of test particles with density n, which all lie in the origin at time t = 0.
Thus,
∂n D ∂ ∂n
 

= D∇2n = 2 r2  (spherical symmetry assumed)


∂t r ∂r ∂r
´∞ 2
r n(r, t) d3r
2
ξ (t) = ∞ ´
0 .
n(r, t) d 3r
0

Multiplying the diffusion equation by r2 and integrating over all space gives
ˆ ˆ ˆ ∞
∂ ∂ ∂n ∂ ∂n
   
2 3 2 3 2 2
r nd r = D r  d r = D 4πr r  dr
∂t ∂r ∂r 0 ∂r ∂r
ˆ ∞ ˆ ∞ ˆ
∂n
= −2D 4πr3 dr = 6D 4πr2n dr = 6D n d3r
0 ∂r 0
ˆ ˆ
∴ r2n d3r = 6Dt n d3r i.e., ξ 2(t) = 6Dt

We may, therefore, identify DT = 31 v 2τcor as the turbulent diffusion coefficient.

Hydrodynamics – Spring 2010 197


Properties of turbulent diffusion

Let us analyze turbulent diffusion more closely.


In turbulent diffusion, diffusion coefficient does not depend on molecular properties or on the scalar
quantity being advected (passively), but only on the statistics of the velocity fluctuations. Thus, the
same coefficient may be used as the effective thermometric conductivity (related to energy transport)
and kinematic viscosity (related to momentum transport). However, this statement does not quite hold
in a general case, if the quantity being transported can react back on the fluid.
Molecular diffusion coefficients in a dilute gas are all of the order

D, ν, κ ∼ λhui

where hui is the averge molecular speed in the rest frame of the fluid, λ = 1/( 2σn) is the mean free
path of the molecules and σ and n are the collision cross-section and the density of the collision targets
(other molecules of the gas). However, turbulent diffusion is often the dominant mode of diffusion, up
to the extent that in a turbulent flow molecular diffusion may often be completely negligible.
Thus, for air in room temperature, for example, λ ∼ 3 µm and hui ∼ 470 m s−1 giving D ∼ 10 cm2 s−1.
The time it takes for heat to diffuse across a room of L = 10 m via molecular diffusion is, thus,

L2
τD ∼ ∼ 1 day.
D
This would make the design of heating systems quite difficult. Fortunately, convection by turbulent
velocity fields is a much faster process.
Hydrodynamics – Spring 2010 198
The mean equations

Let us, then, estimate the effect of turbulent diffusion on momentum transport. Consider an incompress-
ible fluid obeying N–S equation
 
∂ ∂  ∂vi 
(ρvi) = ρFi + −p δij − ρvi vj + µ 
∂t ∂xj ∂xj

Break up the quantities into mean part and fluctuating part, i.e.,

vi = v̄i + vi0 ; p = p̄ + p0; Fi = F̄i + Fi0.

Substitute in N–S and average. All linear terms in fluctuating quantities vanish in averaging. The only
non-zero contribution from fluctuations comes from

vivj = (v̄i + vi0 )(v̄j + vj0 ) = v̄iv̄j + vi0 vj0 .

Thus,  
∂ ∂  0 0 ∂ v̄i 
(ρv̄i) = ρF̄i + −p̄ δij − ρv̄i v̄j − ρvi vj + µ ,
∂t ∂xj ∂xj
which was first derived by Reynolds (1895). This Reynolds equation is very similar to N–S equation,
containing only the extra term ∝ vi0 vj0 . This term is known as the Reynolds stress.

Hydrodynamics – Spring 2010 199


One option for evaluating the Reynolds stress is to use

∂ 0 0 ∂vi0 0 0 ∂vj0
vv = v + vi
∂t i j ∂t j ∂t
and
∂vi0 ∂vi ∂ v̄i
= −
∂t ∂t ∂t
and substitute the time derivatives from the N–S and Reynolds equations, respectively. This procedure,
however, clearly leads to an equation that involves the three-correlations vi0 vj0 vk0 . If one tries to derive
an equation for their evolution, that involves four-correlations etc. Thus, we again meet a closure
problem: this chain of correlation function needs to be truncated by applying some sort of consitutive
relation.
One option is to apply  
∂ v̄i ∂ v̄j 
vi0 vj0 = −DT 
 + 
∂xj ∂xi
Thus, for constant DT
 
∂ ∂  ∂ v̄i 
(ρv̄i) = ρF̄i + −p̄ δij − ρv̄i v̄j + ρ(DT + ν) ,
∂t ∂xj ∂xj

and, thus, DT is the turbulent kinematic viscocity of the fluid. As already deduced, in practice ν  DT
usually holds. It has to be kept in mind that the adopted crude closure scheme does not always lead to
qualitatively correct conlusions, but that the Reynolds stresses sometimes have to be modeled by other
types of terms than diffusive.

Hydrodynamics – Spring 2010 200


Energy spectrum of turbulent fluctuations
Let us, then, return to the problem of developing a statistical description of turbulent fluctuations.
We already introduced the correlation tensor
 
v2 f (r) − g(r) r df
Rij (r) = vi(x)vj (x + r) = g(r) δij +


2
rirj  ; with g(r) = f (r) + .
3 r 2 dr

Let us, instead of Rij , consider its Fourier transform


ˆ
1
Φij (k) = Rij (r)eik·r d3r.
(2π)3

The inverse transform gives ˆ


Rij (r) = Φij (k)e−ik·r d3k.
The incompressibility condition ∂Rij /∂ri = ∂Rij /∂rj = 0 implies that

kiΦij = kj Φij = 0.

Symmetry considerations, similar to those applied for Rij , imply that the most general form of Φij in
isotropic turbulence is
Φij (k) = C(k)kikj + D(k)δij
and on applying the incompressibility conditions we find that
k 2C(k) = −D(k).
Hydrodynamics – Spring 2010 201
Thus, Φij (k) is given by one scalar function E(k) as

E(k)  kikj 
 

Φij (k) = δij − .


4πk 2 k2

Consider, then, the kinetic energy of the turbulence (per unit mass)
ˆ ˆ
1 1 E(k)  kiki  3
 
1 2
2 v = 1
R
2 ii (0) = Φii(k) d3k = δii − dk
2 2 4πk 2 | {z
k 2
}
=3−1=2

ˆ ˆ ∞
E(k) 3 1 2
= dk⇒ 2v = E(k) dk.
4πk 2 0

Thus, E(k) is the energy spectrum of turbulence, E(k) dk giving the amount of turbulent kinetic energy
in the wavenumber range [k, k + dk]. Specifying E(k) is enough for a complete description of Rij in
isotropic homogeneous turbulence.
In the following, we will present the celebrated Kolmogorov’s universal equilibrium theory for isotropic
homogeneous turbulence, which obtains an expression for E(k) from dimensional arguments.

Hydrodynamics – Spring 2010 202


Turbulent fluid in equilibrium

Our next task is to derive an expression for the kinetic energy distribution E(k) corresponding to a
dynamic equilibrium of the fluid equations. The procedure differs from that used for finding the steady-
state distribution function of the molecules, because the fluid system is dissipative. Thus, when left
alone, turbulence in a fluid typically decays.
We shall assume that in a steady state, a turbulent fluid is constantly stirred, i.e., kinetic energy of
turbulent fluctuations are constantly fed into the system. As turbulent kinetic energy is constantly
dissipated by viscous forces, this implies that in steady state the rates of energy input and dissipation
agree. We shall assume also that the stirring of the fluid is performed so that the turbulence produced is
homogeneous and isotropic. Kolmogorov (1941) was the first to propose a theory for such a turbulence.
Turbulent velocity field can be throught of as being made of eddies
P
of different sizes. The input energy is typically fed into the largest
scales. Kolmogorov’s idea was that the largest eddies can feed en-
ergy to smaller eddies and these eddies to even smaller ones, etc., P
producing a cascade of energy from the largest scales to the smallest,
where the dissipation occurs.
The cascade may be understood as a consequence of Kelvin’s theo-
rem. Consider a vortex tube that extends from point Q to point P .
Q
As a result of the random velocity field (generated by stirring), the
distance of the points tends to get larger. Because of incompress-
ibility, the volume of a vortex tube stays constant; thus, its diameter
Q
will tend to decrease. Thus, energy cascades to smaller scales.
Hydrodynamics – Spring 2010 203
Turbulent cascade
Eddies of size ` are expected to have a certain velocity v associated with them. The corresponding
Reynolds number Re` = v`/ν is expected to be large for large eddies. Thus, we may assume that
energy is fed into the system at rate  per unit mass per unit time at the largest eddy size, L, and
velocity, V , for which
VL
Re = 1
ν
Energy is then cascaded to smaller and smaller eddies. For eddies of sufficiently small size, `d, we expect
Red ∼ 1, i.e.,
` d vd ∼ ν
and the energy in these eddies is dissipated by viscosity at rate  per unit mass per unit time in order to
maintain equilibrium. The eddies at intermediate scales merely transmit the energy from scales L to d.
The intermediate eddies are characterised by their velocity, v , and scale, `. Dimensionally, the only way
to express  in terms of v and ` is
v3
∼ ⇒ v ∼ (`)1/3.
`
This applies down to the smalles scales so

3 1/4
vd3 vd3`3d 3
vd4
 
ν ν
∼ = ∼ ∼ ⇒ `d ∼  
, vd ∼ (ν)1/4.
`4d `4d
 
`d ν 

As  ∼ V 3/L, we have
1/4 1/4
L  L4  3 3
 
L V  V
∼ 3 = 3  = Re3/4; ∼ Re1/4.
`d ν ν vd
Hydrodynamics – Spring 2010 204
We can now construct the approximate form of the energy
spactrum E(k). As the largest scale eddies have sizes ∼ L,

log E
there is a cutoff in the spectrum at wavenumbers smaller than
kL ∼ 1/L. On the other hand, the smallest eddies have sizes ∼ d E ~ k −5/3
so there is also a cutoff in the spectrum at wavenumbers larger
than kd ∼ 1/d. The range of wavenumbers

kL  k  kd ∼ Re3/4kL
is called the intertial range. Within this range, the energy spec-
trum is expected to depend only on  and k , so dimensionally

E(k) = C2/3k −5/3, kL  k  kd kL kd log k

is the only possible form of the spectrum. Here, C is a dimen-


sionless number called the Kolmogorov constant. This is the
famous “5/3-law” of Kolmogorov.
We already noted that v ∼ (`)1/3 = (/k)1/3. Thus, E(k) dk ∼
v 2 ∼ (/k)2/3 giving the same form of the spectrum once we
recognize that we should take dk ∼ k in this equation.
Kolmogorov spectrum was not obtained above through a rigorous
analysis. It has, however, been confirmed by many observations
since the 1960s.

Hydrodynamics – Spring 2010 205


Rotation and hydrodynamics

Hydrodynamics – Spring 2010 206


Differential rotation

Start rotating a bucket filled with water. Initially, the solid wall of the rotating bucket transmits angular
momentum to the outer edge of the water but the central parts of the water inside the bucket are still
at rest. The water, therefore, rotates differentially, i.e., with a non-constant angular velocity.
As time goes on, viscous forces tend to even out the differences in the angular velocity and after a while,
the water rotates with the same angular velocity everywhere. This kind of rotation is called rigid-body
(or solid-body) rotation (although the body rotating is not rigid).
Differentially rotating systems in astrophysics are ubiquitous. We already met thin accretion disks, which
have negligible pressure gradients and, thus, a balance between gravity and centrifugal force leading to
r
Ω(r) = vθ /r ≈ |gr |/r

The effect of viscosity is to produce a slow inflow of material rather than to produce rigid-body rotation.
Inside a slowly rotating star, gravitational force and the pressure gradient nearly balance each other. The
pressure can slightly adjust itself to counter the effect of (small) centrifugal force. Viscous forces could,
thus, be expected to turn the rotation finally into a solid-body rotation. Molecular viscosity, however, is
completely negligible inside a star and angular momentum is transported via turbulent convection.
The effective turbulent viscosity inside a star may not be isotropic, and some remaining differential
rotation as a result of this anisotropy may be expected. The Sun demonstrates a difference in angular
velocities between the poles and the equator by & 10 percent.

Hydrodynamics – Spring 2010 207


Stability of differential rotation

Consider a fluid of uniform density rotating differentially around an axis of symmetry. Suppose that a
ring of fluid at distance r0 from the axis rotating at velocity v0 is interchanged with a ring at distance
r1 > r0 rotating at v1. The system is stable if the interchanged rings tend to move towards their original
positions.
Assume that the angular momentum is conserved. Thus,

rv = constant

and the displaced ring originally at r0 now rotates at velocity

v0(r0/r1).

The ring originally at r1 had a centripetal acceleration of v12/r1, which was provided by pressure gradient
forces left after balancing gravity in the equilibrium. The ring moved to r1 now requires the centripetal
force [v0(r0/r1)]2/r1 = v02r02/r13 to remain at its position, and if v12/r1 is greater than this, the system is
stable. Thus,
r02v02 v12
3
< ⇔ (r02Ω0)2 < (r12Ω1)2
r1 r1
is the condition for stability. This reads

d 2 2
[(r Ω) ] > 0 Rayleigh’s criterion.
dr
Hydrodynamics – Spring 2010 208
N–S in a rotating frame of reference

Consider a coordinate system rotating at constant angular velocity Ω = Ωe3 in the inertial frame. This
rotating frame has the Cartesian base vectors

E 1 = cos Ωt e1 + sin Ωt e2; E 2 = − sin Ωt e1 + cos Ωt e2; E 3 = e3 ,

where ei are the fixed base vectors of the inertial frame. A position vector of a point with respect to an
origin at the axis of rotation is
x = xiei = XiE i = X.
Using dE i/dt = Ω × E i and denoting V = ẊiE i and V̇ = ẌiE i we get

dx d dXi dE i
v= = (XiE i) = E i + Xi = ViE i + XiΩ × E i = V + Ω × X
dt dt dt dt
dv d  dx  d2Xi dXi dE i d2 E i
 

f= = = Ei + 2 + Xi 2 = V̇ + 2Ω × V + Ω × (Ω × X)
dt dt dt dt2 dt dt dt

Here, V and V̇ are the velocity and acceleration as measured by an observer in the rotating frame and
the incompressible N–S equation in the rotating frame of reference is, therefore,

∂V 1
+ V · ∇V = V̇ = − ∇p + F ext + ν∇2V − 2Ω × V − Ω × (Ω × X).
∂t ρ

Note that the del operator is given by ∇ = E i ∂X .
i

Hydrodynamics – Spring 2010 209


Centrifugal and Coriolis forces. Effective potential. Rossby number

Typically, F ext is of gravitational origin and can given by F ext = −∇Φ. Then, the centrifugal force,
−Ω × (Ω × X) = 21 ∇(|Ω × X|2), can be combined with the external force introducing an effective
gravitational potential ,
Φeff = Φ − 12 |Ω × X|2.
The effect of the Coriolis force, −2Ω × V , is more complicated but it is non-zero only if there are
motions with respect to the rotating frame. Thus, the pressure in a static incompressible equilibrium in
the rotating frame is obtained as
p/ρ + Φeff = constant.
Clearly, the effect of centrifugal force needs to be taken into account when it modifies the effective
potential significantly. The importance of Coriolis force is determined by the Rossby number ,

U 2/L U
= = ,
ΩU ΩL
which compares the two terms V · ∇V and 2Ω × V in a dynamic situation, where they do not vanish.
Here, U and L are the typical velocity and length scales of the flow, respectively.
At large Rossby numbers, Coriolis force is unimportant, but for slow, large-scale flows ( . 1), like those
of the atmosphere and oceans, it plays a crucial role in the dynamics of the fluid. In geophysical fluid
dynamics, one considers the dynamics of a thin spherical shell of fluid with small Rossby number.

Hydrodynamics – Spring 2010 210


Geostrophic approximation
From now on, we will denote the velocity and position in the rotating frame by v and r .
Fluid flows associated with large scale circulations (atmospheres, oceans) are nearly horizontal. If the
flows are changing only slowly and have large Rossby numbers, dv/dt ≈ 0. The Reynolds number for a
global circulation is also very large so the viscosity term can be neglected. Furthermore, if the centrifugal
correction to the effective potential is small (like for the Earth), then −∇Φeff ≈ −ger and the N–S
equation becomes
∇p
− − ger − 2Ω × v = 0.
ρ
Consider the horizontal (⊥ er ) and vertical (k er ) components of the equation. Gravity dominates over
the Coriolis force for typical circulation velocities, so the vertical component of the Coriolis force may be
neglected. Thus, the vertical component of the force balance states that

1 ∂p
= −g.
ρ ∂r

Since gravity is absent in the horizontal direction, Coriolis force is the one balancing pressure gradients
in the horizontal spherical surface. Thus,

∇hp = −2ρ(Ω × v)h.

Thus, the horizontal pressure gradient is expected to be much smaller than the vertical gradient. These
equations contitute the geostrophic approximation. Note that ∇hp ⊥ v ! A familiar example of the
use of this approximation is the explanation of the circulation of air around a cyclone.

Hydrodynamics – Spring 2010 211


Vorticity in the rotating frame. Bjerknes’s theorem

From now on, consider ideal incompressible fluids. Thus, set ν = 0 and write ρ−1∇p = ∇(p/ρ). Making
use of
v · ∇v = 12 ∇v 2 − v × (∇ × v),
and ω ≡ ∇ × v , we can write the Euler equation in the rotating frame in form

v2
 
∂v p 1
− v × ω = −∇  + + Φ − |Ω × v|2
 − 2Ω × v.
∂t ρ 2 2

On taking the curl of this equation, we again get rid of the “potential part” of the force, but the Coriolis
term does not vanish. Thus, in a rotating coordinate system,

∂ω
= ∇ × (v × ω − 2Ω × v) = ∇ × [v × (ω + 2Ω)].
∂t
Since Ω does not depend on time, this can be written as

∂Q
= ∇ × [v × Q]
∂t
with Q = ω + 2Ω. Thus, we can immediately conclude that Kelvin’s theorem on vorticity is generalized
to ˆ
d
(ω + 2Ω) · ds = 0 Bjerknes’s theorem.
dt S

Hydrodynamics – Spring 2010 212


Taylor–Proudman theorem. Taylor column

Consider a steady flow in the rotating frame of ref-


erence. Thus,

∇ × [v × (ω + 2Ω)] = 0.

If the flow is slow (or the system is large) so that


ω  Ω, i.e., the Rossby number is small, then

0 = ∇ × (v × Ω)
= Ω · ∇v − Ω(∇ · v) − v · ∇Ω + v(∇ · Ω)
= Ω · ∇v

for incompressible flow. This means that v does not


change in the direction of Ω. Thus, slow steady
flows are invariant in the direction of Ω, which is
known as the Taylor–Proudman theorem.
If a flow meets an obstacle in a rapidly rotating
frame, a Taylor column is generated above it. The
flow does not only flow around the obstacle but also
around its extension in the direction of Ω.

Hydrodynamics – Spring 2010 213


Self-gravitating rotating masses

Consider a rigidly rotating fluid of constant density under self-gravity. It may be expected that the shape
of the body differs from a sphere being flatter at the poles.
In the rotating frame
p
 

∇  + Φ − 12 |Ω × r|2 = 0
ρ
where Φ is the gravitational potential. This implies that
p
+ Φ − 12 |Ω × r|2 = constant.
ρ

On the outer surface of the body, pressure is contant. Choose the z -axis co-aligned with the axis of
rotation. Thus,
Φ(x, y, z) − 21 Ω2(x2 + y 2) = constant
on the outer surface. Knowing Φ would give us an equation for the outer surface.
We want to establish that the shape of the body is an ellipsoid. For this purpose, we need the gravitational
potential of an ellipsoid. For an ellipsoid of uniform density inside the bounding surface

x2 y 2 z 2
+ + =1
a2 b 2 c 2
the gravitational potential inside the surface is given by

Φ = πGρ(α0x2 + β0y 2 + γ0z 2 − χ0),


Hydrodynamics – Spring 2010 214
where
ˆ ∞ ˆ ∞
dλ dλ
α0 = abc ; β0 = abc
0 (a2 + λ)∆ 0 (b2 + λ)∆
ˆ ∞ ˆ ∞
dλ dλ
γ0 = abc ; χ0 = abc ,
0 (c2 + λ)∆ 0 ∆
r
and ∆ = (a2 + λ)(b2 + λ)(c2 + λ). If the rotating mass of fluid takes the shape of an ellipsoid, then

Φ = πGρ(α0x2 + β0y 2 + γ0z 2 − χ0),

must satisfy

Φ − 12 Ω2(x2 + y 2) = constant ⇒
(2πGρα0 − Ω2)x2 + (2πGρβ0 − Ω2)y 2 + 2πGργ0z 2 = constant

on the surface given by


x2 y 2 z 2
+ + = 1.
a2 b2 c2
Thus,
a−2 ÷ b−2 ÷ c−2 = (2πGρα0 − Ω2) ÷ (2πGρβ0 − Ω2) ÷ 2πGργ0.
If we can find solutions of this equation, then we would have established that rotating fluids take up
ellipsoidal configurations.

Hydrodynamics – Spring 2010 215


Maclaurin spheroids

Intuitively, one expects a the rtating fluid to be flattened at the poles and symmetric with respect to the
rotation axis. Thus, one adopts
a = b > c,
i.e., a spheroidal configuration with eccentricity

c2
e = 1 − 2.
a
Such solutions of the problem are called Maclaurin spheroids in honor of Maclaurin, who demonstrated
these solutions in 1740. For a = b, the integrals can be solved as

1 − e2 −1 1 − e2
α0 = β 0 = sin e −
e3 e2
√ −1
 
2 sin e
γ0 = 2  1 − 1 − e2 2  
e e

and substituting these to


a−2 2πGρα0 − Ω2
1 − e = −2 =
c 2πGργ0
gives √
Ω2 2 1 − e2 2 −1 6 2
= (3 − 2e ) sin e − (1 − e ).
πGρ e3 e2

Hydrodynamics – Spring 2010 216


0.5 (sketch) 1.0 (sketch)
Ω2
0.4 0.8
πGρ L
0.3 0.6

0.2 0.4

0.1 0.2

0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
e e
Ω2/πGρ reaches a maximum value of 0.499 at e ≈ 0.930. Thus, after the maximum, less rotation makes
the system more flattened. But if, instead of√Ω, one looks at the angular momentum L = 25 M a4Ω of
the spheroid of mass M = 43rπa2cρ = 43 πa3 1 − e2ρ, the situation looks different. L can be made
dimensionless dividing it by GM 3(a2c)1/3, which is constant for all spheroids (of different e) with
constant mass. Thus,
√ v √ v
!2/3 u
Ω2 Ω2
u
L 3 a u
3 −1/3
u
(1 − e)
u u
L̄ = r = u
t = u
t
GM 3(a2c)1/3 5 c πGρ 5 πGρ

is a measure of the angular momentum of the body, and that increases without a limit as e → 1. The
increase, of course, comes from the increase of the moment of inertia of the body.
Hydrodynamics – Spring 2010 217
Jacobi ellipsoids

Jacobi (1834) established that also ellipsoidal solution with three unequal axes are admitted under some
circumstances. Such solutions are called Jacobi ellipsoids. It can be shown that for L̄ < 0.304 (i.e.,
e < 0.813), Maclaurin spheroids are the only possible ellipsoidal solutions. However, at L̄ > 0.304 a
Jacobi ellipsoid with three unequal axes becomes possible.
When Jacobi ellipsoids are admitted, we must ask which of the two possible solutions is more probable
in Nature. For a given angular momentum, it can be shown that a Jacobi ellipsoid has less rotational
energy than a Maclaurin spheroid. Thus, an internal dissipation mechanism (like viscosity) may make a
Maclaurin spheroid unstable and relax to a Jacobi ellipsoid after some dissipation of rotational energy.
Thus, the solution has spontaneously broken the symmetry which was present in the formulation of the
problem.
Are there any astronomical objects that would be examples of Jacobi ellipsoids? Some elliptic galaxies
show tri-axial shape, so one might be tempted to call these stellar-dynamics analogues of Jacobi ellipsoids.
Elliptical galaxies, however, have very little angular momentum, so the explanation is probably incorrect.
Rather, the shape stems from the anisotropic velocity distribution of the stars inside the galaxy.

Hydrodynamics – Spring 2010 218


Magnetohydrodynamics (MHD)

Hydrodynamics – Spring 2010 219


Scope of magnetohydrodynamics

So far we have considered neutral fluids, in which collisions are the main means of transporting momentum
and energy between the molecules of the gas. The only force acting at large distance in neutral fluids is
gravity. We introduced some fluid effects, where gravitational field acted as the restoring force, mainly
involving surface phenomena (recall, however, also the internal gravity waves).
For fluids made of charged particles, the situation is completely different. Besides gravitational fields,
also the electromagnetic fields now affect the motion of the molecules. This inevitably leads to a more
complicated dynamical theory than ordinary hydrodynamics, since fluids consisting of charged particles
not only react to external electromagnetic forces but act as sources of the fields as well. Therefore, our
former usual convention of treating body forces as external forces no longer apply.
Magnetohydrodynamics is primarily a macroscopic dynamical theory of plasmas, which are quasineutral7
gases containing enough free charges to make collective electromagnetic phenomena important for their
physical behaviour. The latter means that plasmas are dense enough so that a large number of particles
are interacting with each other through Coulomb interactions at the same time. For a more quantitative
definion of a plasma state, see the lectures of “Plasma physics”. For our purposes it is enough to
understand that overwhelmingly largest part of the ordinary matter in the unverse is in the plasma state.
(All stars are made of plasma; inside the orbit of Pluto, 0.1% of mass (mainly in Jupiter) and 10−15% of
volume is in non-plasma state.)
Besides plasmas MHD can also be applied to electrically conducting liquids like mercury.
7 Quasineutral ionized gases consist of equal amounts of positive and negative charges. Thus, the net charge of all the particles of a quasineutral
gas is zero.

Hydrodynamics – Spring 2010 220


Equation of continuity and momentum equation

Equations governing the motion of the fluid in MHD are very similar to the HD equations. We will justify
them here very heuristically and refer the student to the lectures of “Plasma physics” for a more rigorous
derivation.
The equation continuity (conservation of mass)

∂ρ
+ ∇ · (ρv) = 0
∂t
of course applies in MHD as well.
The momentum equation has to be supplemented with forces that come from the interactions of the
molecules with the electric (E ) and magnetic (B ) fields permeating the fluid. A charged particle (charge
q ) feels the Lorentz force,
q(E + v × B)
so a large number N of charged particles (indexed by i) inside a volume ∆V feels the resultant force per
unit volume of
1 X N
f= qi(E + v i × B).
∆V i=1
In a quasineutral situation, the sum i qi = 0, so the first term vanishes. The second term corresponds
P

to
f = j × B,
where j = (∆V )−1 i qi v i is the electric current density (SI unit A m−2) carried by the fluid.
P

Hydrodynamics – Spring 2010 221


Thus, the momentum equation of MHD reads

dv
ρ = ρF − ∇p + j × B + ρν∇2v,
dt
where all other body forces than the Lorentz force (e.g., gravity) have been included in F .
The next task is to find out, how the current density and the magnetic field evolve. In MHD, one limits
the theory to describe only “slow” phenomena, i.e., one takes Ampère’s law to apply. Thus,

µ0j = ∇ × B,

where µ0 is the permeability of vacuum. (In plasmas, all charges can be regarded as free.) Thus, the
current density is immediately eliminated from the equation of motion and one gets

dv 1
ρ = ρF − ∇p + (∇ × B) × B + ρν∇2v.
dt µ0

Before turning to the evolution of the magnetic field, let us examine the Lorentz force somewhat more
closely. First, use the vector identity
2
 
B 
(∇ × B) × B = (B · ∇)B − ∇   .
2

Comparing the second term with the pressure gradient shows that the magnetic field introduces a pressure
of B 2/2µ0. The first term represents a tension along the magnetic field lines.

Hydrodynamics – Spring 2010 222


Magnetic force in tensorial form

In fact, the magnetic body force can be written in a form of a surface force with the aid of the identity

∇ · (BB) = (∇ · B)B + (B · ∇)B.

As magnetic field is always source free (∇ · B = 0), we get for the Lorentz force
2
 
1 B BB 
(∇ × B) × B = −∇ · 
 I−  = −∇ · M.
µ0 2µ0 µ0

Thus,
dvi ∂
ρ= ρFi − (Pij + Mij ),
dt ∂xj
where Mij = (B 2/2µ0)δij − BiBj /µ0. The second term is a tension along the field lines, which can be
seen by writing the tensor in matrix form in a (local) coordinate system with ez k B :
   
2
 B /2µ0 0 0   0 0 0 
B 2/2µ0
   
(Mij ) = 

 0 0 

 − 

 0 0 0 


2 2
0 0 B /2µ0 0 0 B /µ0
   

The first term corresponds to an isotropic pressure (generating expansion) and the second term to a
tension along the magnetic field (trying to keep the lines of force as short as possible).

Hydrodynamics – Spring 2010 223


Faraday’s and Ohm’s laws. Induction equation

The evolution of the magnetic field is governed by Faraday’s law

∂B
= −∇ × E.
∂t
Thus, although the electric field drops out from the equation of motion, it re-enters the problem through
Faraday’s law.
Electric field is related to the current density via the Ohm’s law. In the plasma rest frame, we write
η
E 0 = ηj = ∇ × B,
µ0

where η is the resistivity of the fluid. The electric field transforms from the plasma rest frame to the
laboratory frame as
E = E0 − v × B
giving
∂B η
= ∇ × (v × B) + ∇2B,
∂t µ0
where resistivity has been assumed constant and the vector identity ∇ × (∇ × B) = ∇(∇ · B) − ∇2B
along with ∇ · B = 0 has been used. This is the induction equation of MHD.

Hydrodynamics – Spring 2010 224


MHD equations. Energy equation

The MHD system of equations is

∂ρ
+ ∇ · (ρv) = 0
∂t
∂v ∇p (∇ × B) × B
+ v · ∇v = F − + + ν∇2v
∂t ρ µ0 ρ
∂B η
= ∇ × (v × B) + ∇2B
∂t µ0

appended with an appropriate equation of state or the energy equation. The MHD energy equation is

∂
 

ρ  + v · ∇ = −p∇ · v + ∇ · (K∇T ) + ηj 2,


∂t
where the last term is due to Ohmic heating of the fluid, i.e., due to dissipation of the electromagnetic
fields. For highly conducting plasmas, the last two terms are often neglected, and the internal energy
written as  = cV T = p/ρ(γ − 1). Then, as for neutral fluids, one obtains

d p
 

= 0.
dt ργ

Instead of an energy equation, it is also often customary to use a barotropic equation of state or simply
assume the fluid to be incompressible. In those cases, of course, the energy equation becomes redundant.
Hydrodynamics – Spring 2010 225
Magnetic Reynolds number

Consider the MHD induction equation

∂B
= ∇ × (v × B) + λ∇2B,
∂t
where λ = η/µ0 is the magnetic diffusivity. Taking the ratio of the two terms on the right-hand side,
i.e.,
V B/L LV
ReM = =
λB/L2 λ
defines a dimensionless number known as the magnetic Reynolds number .
The direct proportionality of ReM on L means that it is much larger for astrophysical plasmas compared
to laboratory plasmas. For example, for a hydrogen plasma at T = 104 K the magnetic diffusivity is

λ ≈ 103 m2 s−1.

Thus, for a typical laboratory system with L ∼ 1 m and V ∼ 0.1 m s−1,

ReM ∼ 10−4.

Instead, for example, for a convection cell at the solar surface L ∼ 106 m and V ∼ 103 m s−1 we find

ReM ∼ 106.

Hydrodynamics – Spring 2010 226


Ideal MHD. Alfvén’s theorem on flux-freezing

Thus, under typical conditions:

∂B
Laboratory: ≈ λ∇2B
∂t
∂B
Astrophysics: ≈ ∇ × (v × B)
∂t
Of course, these are very crude simplifications: in many laboratory conditions ∇ × (v × B) cannot be
neglected and in astrophysical plasmas λ∇2B is important in localized regions with large currents.
The limit of λ → 0, i.e.,
∂B
= ∇ × (v × B)
∂t
is known as the ideal MHD limit. Our general proof of Kelvin’s theorem on vorticity again becomes
handy, because it shows immediately (since ∇ · B = 0) that
ˆ
d
B · ds = 0
dt S

for any material surface S following the fluid flow. This result was first pointed out by Alfvén (1942) and
it is called Alfvén’s theorem on flux-freezing. It states that the flux inside a magnetic flux tube
remains constant in an ideal MHD flow. Alfvén’s theorem is the magnetic analogue of Kelvin’s theorem
on vorticity.
Hydrodynamics – Spring 2010 227
Consequences of Alfvén’s theorem

In astrophysical systems with high magnetic Reynolds number we can then imagine the magnetic field
lines to be frozen-in to the plasma flow. Suppose we have a straight plasma column with homogeneous
magnetic field. If the plasma column is bent or twisted, then the magnetic field lines become bent or
twisted as well.
In general, if two plasma elements are at some time connected with a magnetic field line, they remain
forever connected with a magnetic field line in ideal MHD.

Hydrodynamics – Spring 2010 228


Magnetic field amplification/generation by plasma motions
Consider, first, the evolution of a magnetic field in a collapsing magnetized plasma cloud, i.e., a star
forming from a cloud of interstellar plasma. As magnetic flux is frozen to the plasma, the magnetic
field increases in inverse proportion to the cross-sectional area of the plasma cloud. Starting from typical
values of interstellar density and magnetic field, the solar magnetic field would be of the order of ∼ 106 T
in case no mechanism would be available for annihilating magnetic energy (actually, B ∼ 1 mT).
Solar magnetic field, however, follows a 22-year activity quasi-cycle involving polarity changes. This is
thought to occur because of the continuous re-generation of the field through a dynamo mechanism
involving differential rotation and convection inside the Sun.

Hydrodynamics – Spring 2010 229


MHD waves

MHD contains a wealth of wave phenomena that modified with respect to or not observed at all in neutral
fluids. By linearizing the MHD equations around a homogeneous static equilibrium with ρ = ρ0 +ρ1(x, t),
v = v 1(x, t), p = p0 + p1(x, t) and B = B 0 + B 1(x, t), we get

∂ρ1
+ ρ0 ∇ · v 1 = 0
∂t
∂v 1 ∇p1 (∇ × B 1) × B0
=− +
∂t ρ0 µ0 ρ0
∇p1 = c2s ∇ρ1
∂B 1
= ∇ × (v 1 × B 0)
∂t

with c2s = γp0/ρ0 for an adiabatic fluid with zero viscosity and resistivity. Taking the time derivative
of the equation of motion and using the other three equations to eliminate ρ1, p1 and B 1 we get the
MHD wave equation,

∂ 2v1 B0
2
= c2s ∇(∇ · v 1) + {∇ × [∇ × (v 1 × v A)]} × v A, where v A = √ is the Alfvén velocity
∂t µ0ρ0

⇒ ω 2v 1 = c2s k(k · v 1) − {k × [k × (v 1 × v A)]} × v A MHD dispersion relation.

Hydrodynamics – Spring 2010 230


Applying the vector identity a × (b × c) = b(a · c) − c(a · b) (i.e., the “BACCAB formula”) three times,
the dispersion relation can be written in form

0 = −ω 2v 1 + (c2s + vA
2
)k(k · v 1) + (v A · k)[(v A · k)v 1 − k(v A · v 1) − v A(k · v 1)]
= {[(v A · k)2 − ω 2]I + (c2s + vA
2
)kk − (v A · k)[kv A + v Ak]} · v 1 ≡ D · v 1.

The non-trivial (v 1 6= 0) solutions of this equation are obtained as

det D = 0,

where Dij = [(v A·k)2 −ω 2]δij +(c2s +vA


2
)kikj −(v A·k)(kivAj +vAikj ) is the (symmetric) dispersion tensor
of the MHD fluid. Choosing the coordinate system so that v A = vAez and k = k(sin θex + cos θez )
we can write
  
2
 −ω + k 2 vA
2
+ k 2c2s sin2 θ 0 c2s k 2 sin θ cos θ  v1x 
2 2 2
k cos2 θ
  
0 = D · v1 = 

 0 −ω + vA 0 

 v1y 


c2s k 2 sin θ cos θ 0 −ω 2 + k 2c2s cos2 θ v1z
  

Setting the determinant of this matrix to zero gives three solutions, i.e., three different MHD wave modes

v 1 k ey ; ω 2/k 2 = vA
2
cos2 θ shear Alfvén wave
v 1 ⊥ ey ; 2
ω 2/k 2 = 12 (vA + c2s ) + [ 41 (vA
2
+ c2s )2 − c2s vA
2
cos2 θ]1/2 fast MHD wave
v 1 ⊥ ey ; 2
ω 2/k 2 = 21 (vA + c2s ) − [ 41 (vA
2
+ c2s )2 − c2s vA
2
cos2 θ]1/2 slow MHD wave

Hydrodynamics – Spring 2010 231


Propagation parallel to the field

Let us, first, investigate the MHD wave modes at the limit θ → 0, i.e., when the waves are propagating
parallel to the field lines. Longitudinal waves (v 1 k k k v A) are simple sound waves with

ω = ±csk.

Either the fast (if cs > vA) or the slow (if vA > cs) MHD mode reduces to a sound wave at θ → 0.
The dispersion relation of the shear Alfvén wave becomes

ω = ±vAk, v1 ⊥ k

which shows that these waves are transverse waves. The magnetic field of the wave is obtained as

B0
−iωB 1 = ik × (v 1 × B 0) = ikB0v 1 ⇒ B1 = − v1 ⊥ B 0
vA

showing that the magnetic field fluctuations are also transverse and bend the magnetic field lines. As
v v
B 2/µ0
u
u
u
u
u magnetic tension
vA = u
t = u
t ,
ρ0 mass density

shear Alfvén waves are a direct analogue of transverse waves propagating along a string.
Either the fast (if vA > cs) or the slow (if vA < cs) MHD mode reduces to a shear-wave at θ → 0.
Hydrodynamics – Spring 2010 232
Propagation perpendicular to the field

If θ → π/2, then both the shear Alfén wave and the slow MHD wave have zero phase speed, i.e., these
waves do not propagate perpendicular to the magnetic field. The dispersion relation of the fast mode is
r
2 + c2 k,
ω = ± vA s
r
where the speed vA2 + c2 is called the magnetosonic speed. Correspondingly, this wave is referred to as
s
the magnetosonic wave.
Magnetosonic waves compress the magnetic field, i.e.,
v1
B1 = B 0.
vA

The restoring force in this wave is the gradient of total pressure. This can be seen by writing

2 B02 γp0 γB pB γp0


vA + c2s = + = + ,
µ0 ρ0 ρ0 ρ0 ρ0

where pB = B02/2µ0 is the magnetic pressure and γB = 2 is the adiabatic index corresponding to a two-
dimensional gas. This is natural, since the magnetic force j × B 0 is always in the plane perpendicular to
the magnetic field (i.e., two-dimensional) and any motions of the fluid in the direction of the magnetic
field leave the field unchanged.

Hydrodynamics – Spring 2010 233


Magnetohydrostatics

Static solutions of MHD, in the strict sense, would need to fulfill λ∇2B = ∂B/∂t = 0 in addition to
fulfilling the force balance equation
∇p = ρF + j × B.
However, it is customary to consider any configuration that fulfills the force balance equation a magne-
tohydrostatic equlibrium. This, of course, assumes implicitly that the transport of momentum via MHD
waves has a much shorter times scale than the decay of the magnetic field via diffusion.

In the absence of the external forces, the force balance equation reduces to

(∇ × B) × B
∇p = .
µ0

In astrophysics, we often deal with plasmas, for which the plasma beta,

p 2µ0p
β= =
pB B2

is small. Thus, the force balance equation reduces to

(∇ × B) × B = 0 ⇒ ∇ × B = α(x)B,

where α(x) is a scalar. Magnetic fields fulfilling this equation are called force-free magnetic fields.
Hydrodynamics – Spring 2010 234
As ∇ · (∇ × B) = 0 and ∇ · B = 0 identically, the force-free equation implies B · ∇α = 0, i.e., that
α is constant along magnetic field lines. If α is a constant over all space, the equation

∇ × B = αB

is linear. Linear force-free fields form an important category of MHD equilibrium solutions in solar and
heliospheric physics, for example. The force-free magnetic field solutions include also the potential fields,
for which α = 0.
We note that a force-free field may be a valid solution of the magnetostatic problem even if the plasma
beta is not small, if there is an external force field (like gravitational field) present in the system. In that
case, the balance is obtained so that the hydrostatic balance and the force-free condition are satisfied
simultaneously:

∇p = ρF
0 = (∇ × B) × B

Finally, we note that in a general equilibrium in the absence of external forces, the pressure gradient has
to be balanced by the magnetic force. Such solutions are considered, for example, when studying the
magnetic confinement of plasma, which is important for fusion research. In this respect, also the analysis
of the stability of the MHD equilibria is extremely important, and it turns out that many equilibrium
solutions of plasmas confined by magnetic fields are actually unstable.
More on MHD on a special course next fall. Welcome!

Hydrodynamics – Spring 2010 235

You might also like