I  A

Astrophysical Fluid
Dynamics
Prof. Chris Reynolds
NST PART II A STROPHYSICS /P HYSICS L ENT 2021
 About

These notes accompany the 24 lectures in the Lent-2021 term on Astrophysical

Fluid Dynamics, a course within the Part II Physics/Astrophysics of the
Natural Sciences Tripos. They are not intended to be a complete or self-
contained discussion of course material and must be used in conjunction
with the lectures and the course textbook, Principles of Astrophysical Fluid
Dynamics by C.J.Clarke and R.F.Carswell [PAFD].
I would like to express my sincere thanks to Zihan Yan who produced these
LATEXnotes based on the hand-written handouts from the Lent-2020 run of this
course. Minor additions and modifications have been included for Lent-2021.
If you find any typo or error, please contact [email protected] for correction.

Chris Reynolds
(Cambridge, Jan 2021)

i
 Contents

About i

Contents iii

A Basic Principles 1
A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
A.2 Collisional and Collisionless Fluids . . . . . . . . . . . . . . 2

B Formulation of the Fluid Equations 5

B.1 Eulerian vs Lagrangian . . . . . . . . . . . . . . . . . . . . . 5
B.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
B.3 Conservation of Mass . . . . . . . . . . . . . . . . . . . . . . 7
B.4 Conservation of Momentum . . . . . . . . . . . . . . . . . . 8

C Gravitation 13
C.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
C.2 Potential of a Spherical Mass Distribution . . . . . . . . . . 17
C.3 Gravitational Potential Energy . . . . . . . . . . . . . . . . . 17
C.4 The Virial Theorem . . . . . . . . . . . . . . . . . . . . . . . 18

D Equations of State and the Energy Equation 21

D.1 The Equation of State . . . . . . . . . . . . . . . . . . . . . 21
D.2 The Energy Equation . . . . . . . . . . . . . . . . . . . . . . 23
D.3 Heating and Cooling Processes . . . . . . . . . . . . . . . . . 25
D.4 Energy Transport Processes . . . . . . . . . . . . . . . . . . 26

E Hydrostatic Equilibrium, Atmospheres and Stars 27

E.1 Hydrostatic Equilibrium . . . . . . . . . . . . . . . . . . . . 27
E.2 Stars as Self-Gravitating Polytropes . . . . . . . . . . . . . . 30
E.3 Isothermal Spheres (Case n → ∞) . . . . . . . . . . . . . . . 32
E.4 Scaling Relations . . . . . . . . . . . . . . . . . . . . . . . . 33

iii
Contents

F Sound Waves, Supersonic Flows and Shock Waves 37

F.1 Sound Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
F.2 Sound Waves in a Stratified Atmosphere . . . . . . . . . . . 40
F.3 Transmission of Sound Waves at Interfaces . . . . . . . . . . 44
F.4 Supersonic Fluids and Shocks . . . . . . . . . . . . . . . . . 45
F.5 The Rankine-Hugoniot Relations . . . . . . . . . . . . . . . 46
F.6 Theory of Supernova Explosions . . . . . . . . . . . . . . . . 50

G Bernoulli’s Equation and Transonic Flows 59

G.1 Bernoulli’s Equation . . . . . . . . . . . . . . . . . . . . . . 59
G.2 Rotational and Irrotational Flows . . . . . . . . . . . . . . . 61
G.3 The De Laval Nozzle . . . . . . . . . . . . . . . . . . . . . . 63
G.4 Spherical Accretion and Winds . . . . . . . . . . . . . . . . 67

H Fluid Instabilities 73
H.1 Convective Instability . . . . . . . . . . . . . . . . . . . . . . 73
H.2 Jeans Instability . . . . . . . . . . . . . . . . . . . . . . . . . 76
H.3 Rayleigh-Taylor and Kelvin-Helmholtz Instability . . . . . . 79
H.4 Thermal Instability . . . . . . . . . . . . . . . . . . . . . . . 83

I Viscous Flows 89
I.1 Basics of Viscosity . . . . . . . . . . . . . . . . . . . . . . . . 89
I.2 Navier-Stokes Equation . . . . . . . . . . . . . . . . . . . . . 91
I.3 Vorticity in Viscous Flows . . . . . . . . . . . . . . . . . . . 92
I.4 Energy Dissipation in Incompressible Viscous Flows . . . . . 93
I.5 Viscous Flow through a Pipe . . . . . . . . . . . . . . . . . . 95
I.6 Accretion Disks . . . . . . . . . . . . . . . . . . . . . . . . . 96
I.7 Steady-State, Geometrically-Thin Disks . . . . . . . . . . . . 100

J Plasmas 103
J.1 Magnetohydrodynamic (MHD) Equations . . . . . . . . . . 103
J.2 The Dynamical Effects of Magnetic Fields . . . . . . . . . . 106
J.3 Waves in Plasmas . . . . . . . . . . . . . . . . . . . . . . . . 106
J.4 Instabilities in Plasmas . . . . . . . . . . . . . . . . . . . . . 110
J.5 Magnetorotational Instability . . . . . . . . . . . . . . . . . 111

K Appendix : Typos and Corrections 115

Bibliography 117

iv
 CHAPTER A

BASIC P RINCIPLES

A.1 Introduction
Fluid Dynamics concerns itself with the dynamics of liquid, gases and (to some
degree) plasmas. Phenomena considered in fluid dynamics are macroscopic.
We describe a fluid as a continuous medium with well-defined macroscopic
quantities (e.g. density ρ, pressure p...), even though, at a microscopic level,
the fluid is composed of particles.
Most of the baryonic matter in the Universe can be treated as a fluid.
Fluid dynamics is thus an extremely important topic within astrophysics.
Astrophysical systems can display extremes of density (both low and high) and
temperature beyond those accessible in terrestrial laboratories. In addition,
gravity is often a crucial component of the dynamics in astrophysical systems.
Thus the subject of Astrophysical Fluid Dynamics encompasses but significantly
extends the study of fluids relevant to terrestrial systems and/or engineers.
In the astrophysical context, the liquid state is not very common (examples
are high pressure environments of planetary surfaces and interiors), so our focus
will be on the gas phase. Key difference is that gases are more compressible
than liquids.

Examples (Fluids in the Universe).

– Interiors of stars, white dwarfs, neutron stars;

– Interstellar medium (ISM), intergalactic medium (IGM), intracluster

medium (ICM);

– Stellar winds, jets, accretion disks;

– Giant planets.

In our discussion, we shall use the concept of a fluid element. This is a

region of fluid that is

1
A. B ASIC P RINCIPLES

(i) Small enough that there are no significant variations of any property q
that interests us
q
lregion  lscale ∼
|∇q|

(ii) Large enough to contain sufficient particles to be considered in the

continuum limit
3
nlregion 1
where n is the number density of particles.

A.2 Collisional and Collisionless Fluids

In a collisional fluid, any relevant fluid element is large enough such that the
constituent particles know about local conditions through interactions with
each other, i.e.
lregion  λ
where λ is the mean free path. Particles will then attain a distribution of
velocities that maximises the entropy of the system at a given temperature.
Thus, a collisional fluid at a given density ρ and temperature T will have a
well-defined distribution of particle speeds and hence a well-defined pressure,
p. We can relate ρ, T and p as equation of state:

p = p(ρ, T )

In a collisionless fluid, particles do not interact frequently enough to satisfy

lregion  λ. So, distribution of particle speeds locally does not correspond
to maximum entropy solution, instead depending on initial conditions and
non-local conditions.

Examples. Collisionless Fluids

– Stars in a galaxy;

– Grains in Saturn’s rings;

– Dark matter;

– ICM (transitional from collisional to collisionless)

Example (Expand example of ICM). Treat as fully ionised plasma of H(e− , p).
The mean free path is set by Coulomb collisions and an analysis gives

33/2 (kB Te )2 20

λe =
4π 1/2 ne e4 ln Λ

2
 A.2. Collisional and Collisionless Fluids

where

ne = e− number density
Λ = ratio of largest to smallest impact parameter

and for T & 4 × 105 K we have ln Λ ∼ 40. So, if Ti = Te , we have

2  −1
Te ne

λe = λi ≈ 23 kpc .
108 K 10 cm−3
−3

So we have
collisionless
z }| {
Rgalaxy ∼ λe  Rcluster ∼ 1 Mpc.
| {z }
collisional

3
 CHAPTER B

F ORMULATION OF THE F LUID

E QUATIONS

B.1 Eulerian vs Lagrangian

Two main frameworks for understanding fluid flow:
(i) Eulerian description — one considers the properties of the fluid measured
in a frame of reference that is fixed in space. So we consider quantities
like
ρ(r, t), p(r, t), T (r, t), v(r, t)

(ii) Lagrangian description — one considers a particular fluid element and

examines the change in the properties of that element. So, the spatial
reference frame is co-moving with the fluid flow.
The Eulerian approach is more useful if the motion of particular fluid
elements is not of interest.
The Lagranian approach is useful if we do care about the passage of given
fluid elements (e.g. gas parcels that are enriched by metals).
These two different pictures lead to very different computational approaches
to fluid dynamics which we will discuss later.
Mathematically, it is straightforward to relate these two pictures. Consider
a quantity Q in a fluid element at position r and time t. At time t + δt the
element will be at position r + δr. The change in quantity Q of the fluid
element is
DQ Q(r + δr, t + δt) − Q(r, t)
 
= lim
Dt δt→0 δt
but
∂Q
Q(r + δr, t + δt) = Q(r, t) + δt + δr · ∇Q + O(δt2 , |δr|2 , δt|δr|)
∂t
so
DQ ∂Q δr
 
= lim + · ∇Q + O(δt, |δr|)
Dt δt→0 ∂t δt

5
B. F ORMULATION OF THE F LUID E QUATIONS

which gives us
DQ ∂Q
= + u · ∇Q
Dt ∂t | {z }
“convective”
|{z} |{z}
Lagrangian Eulerian derivative
time derivative time derivative

B.2 Kinematics
Kinematics is the study of particle (and fluid element) trajectories.
Streamlines, streaklines and particle paths are field lines resulting from the
velocity vector fields. If the flow is steady with time, they all coincide.

(i) Streamline: families of curves that are instantaneously tangent to the

velocity vector of the flow u(r, t). They show the direction of the fluid
element.

Figure B.1: Streamline

Parameterise streamline by label s such that

dr dx dy dz
 
= , ,
ds ds ds ds
and demand dr/ds k u, we get
dr dx dy dz
×u=0 ⇒ = =
ds ux uy uz

(ii) Particle paths: trajectories of individual fluid elements given by

dr
= u(r, t)
dt
For small time intervals, particle paths follow streamlines since u can be
treated as approximately steady.

6
 B.3. Conservation of Mass

(iii) Streaklines: locus of points of all fluid that have passed through a given
spatial point in the past.
r(t) = r0
for some given t in the past.

We now proceed to discuss the equations that describe the dynamics of a

fluid. These are essentially expressions of the conservation of mass, momentum
and energy.

B.3 Conservation of Mass

Consider a fixed volume V bounded by a surface S. If there are no sources or
sinks of mass within the volume, we can say

rate of change of mass in V = − rate that mass is flowing out across S

Figure B.2: Mass flow of a fluid element

this gives

∂
Z Z
ρ dV = − ρu · dS
∂t V S
∂ρ
Z Z
⇒ dV = − ∇ · (ρu) dV
V ∂t V
∂ρ
Z  
⇒ + ∇ · (ρu) dV = 0.
V ∂t

This is true for all volumes V . So we must have

∂ρ
+ ∇ · (ρu) = 0 Eulerian Continuity Equation
∂t

7
B. F ORMULATION OF THE F LUID E QUATIONS

The Lagrangian expression of mass conservation is easily found:

Dρ ∂ρ
= + u · ∇ρ = −∇ · ρu + u · ∇ρ = −ρ∇ · u.
Dt ∂t

Thus we have

Dρ
+ ρ∇ · u = 0 Lagrangian Continuity Equation
Dt

In an incompressible flow, fluid elements maintain a constant density, i.e.

Dρ
= 0.
Dt

We can now see that incompressible flows must be divergence free, ∇ · u = 0.

B.4 Conservation of Momentum

Pressure
Consider only collisional fluids where there are forces within the fluid due
to particle-particle interactions. Thus there can be momentum flux across
surfaces within the fluid even in the absence of bulk flows.
In a fluid with uniform properties, the momentum flux through a surface is
balanced by an equal and opposite momentum flux through the other side of
the surface. Therefore, there is no net acceleration even for non-zero pressure
since pressure is defined as the momentum flux on one side of the surface.
If the particle motions within the fluid are isotropic, the momentum flux
is locally independent of the orientation of the surface and the components
parallel to the surface cancel out. Then, the force acting on one side of surface
element is
dF = p dS

In the more general case, forces across surfaces are not perpendicular to
the surface and we have
dFi = σij dSj .
X

where σij is the stress tensor — the force in direction i acting on a surface
with normal along j.
Isotropic pressure in a static fluid corresponds to

σij = pδij .

8
 B.4. Conservation of Momentum

Momentum Equation for a Fluid

Consider a fluid element that is subject to a gravitational field g and internal
pressure forces. Let the fluid element have volume V and surface S.

Figure B.3: A fluid element subject to gravity

Pressure acting on the surface element gives force −p dS. Pressure force on
element projected in direction n̂ is −pn̂ · dS. So, net pressure force in direction
n̂ is Z Z Z
F · n̂ = − pn̂ · dS = − ∇ · (pn̂) dV = − n̂ · ∇p dV .
S V V

Rate of change of momentum of fluid element in direction n̂ is the total

force in that direction:
D
 Z  Z Z
ρu dV · n̂ = − n̂ · ∇p dV + ρg · n̂ dV .
Dt V V V

In limit that dV → δV we have

R

D
(ρu δV ) · n̂ = −δV n̂ · ∇p + δV ρg · n̂
Dt

D Du
⇒ n̂ · u (ρδV ) +ρδV n̂ · = −δV n̂ · ∇p + δV ρg · n̂
Dt
| {z } Dt
=0 by mass
conservation

Du
 
⇒ δV n̂ · ρ + ∇p − ρg = 0.
Dt
This must be true for all n̂ and all δV . So,

Du
ρ = −∇p + ρg Lagrangian Momentum Equation
Dt

9
B. F ORMULATION OF THE F LUID E QUATIONS

or
∂u
ρ + ρ(u · ∇)u = −∇p + ρg Eulerian Momentum Equation
∂t
Now consider the Eulerian rate of change of momentum density ρu and
introduce more compact notation
∂
(ρui ) ≡ ∂t (ρui )
∂t
= ρ∂t ui + ui ∂t ρ
= −ρuj ∂j ui − ∂j pδij + ρgi − ui ∂j (ρuj )
where we have used notation
∂
∂j ≡
∂xj
and employed summation convention (summation over the repeated indices).
This gives
∂t (ρui ) = −∂j ( ρui uj + pδij ) + ρgi = −∂j σij + ρgi
| {z } |{z}
stress tensor stress tensor
due to bulk flow due to random
‘Ram Pressure’ thermal motions
where we have generalised the stress tensor to include the momentum flux
from the bulk flow,
σij = pδij + ρui uj .

In component free language we write

∂t (ρu) = −∇ · (ρu ⊗ u + pI) +ρg
| {z }
flux of
momentum
density

Example (Flow in a pipe in the y-direction).

Figure B.4: Flow in a pipe

10
 B.4. Conservation of Momentum

Any surface will experience a momentum flux p due to pressure. Only

surfaces with a normal that has a component parallel to flow will experience
ram pressure.
0 0
 
p
σij = 0 p + ρu2 0
 
0 0 p

The remaining equation of fluid dynamics is based on the conservation of

energy. We will defer a discussion of that until later.

11
 CHAPTER C

G RAVITATION

C.1 Basics
Define the gravitational potential Ψ s.t. the gravitational acceleration g is

g = −∇Ψ

If l is some closed loop, we have

I Z Z
g · dl = (∇ × g) · dS = − [∇ × (∇Ψ)] · dS = 0
l S S

as curl of any gradient is zero. So gravity is a conservative force — the work

done around a closed loop is zero.
As a consequence, the work needed to take a mass from point r to ∞ is
Z ∞ Z ∞
− g · dl = ∇Ψ · dl = Ψ(∞) − Ψ(r)
r r

which is independent of path.

A particular important case is the gravity of a point mass, which has
GM
Ψ=− if mass at origin
r
GM
Ψ=− if mass at location r0
|r − r0 |
For system of point masses we have
X GMi
Ψ=−
i
|r − r0i |
X GMi (r − r0 )
⇒ g = −∇Ψ = − i

i
|r − r0i |3
Replacing Mi → ρi δVi and going to the continuum limit we have
r − r0
Z
g(r) = −G ρ(r0 ) dV 0
|r − r0 |3

13
C. G RAVITATION

Take divergence of both sides

r − r0
Z  
0
∇r · g = −G ρ(r ) ∇r · dV 0
|r − r0 |3
| {z }
4πδ(r−r0 )
Z
= −4πG ρ(r0 )δ(r − r0 ) dV 0
= −4πGρ(r)

So,

∇ · g = −∇2 Ψ = −4πGρ Poisson’s Equation

We can also express Poisson’s equation in integral form: for some volume
V bounded by surface S we have

Z Z
∇ · g dV = −4πG ρ dV
ZV V

⇒ g · dS = −4πGM
S

This is useful for calculating g when the mass distribution obeys some
symmetry.

Example (Spherical distribution of mass).

Figure C.1: Spherical distribution of mass

14
 C.1. Basics

By symmetry g is radial and |g| is constant over a r = const. shell. So

Z
g · dS = −4πG M (r)
| {z }
mass
enclosed
⇒ − 4πr2 |g| = −4πGM (r)
GM (r)
⇒ |g| =
r2
GM (r)
∴ g=− r̂
r2

Example (Infinite cylindrically symmetric mass).

Figure C.2: Cylindrical distribution of mass

By symmetry, g is uniform and radial on the curved sides of the cylindrical

surface, and is zero on the flat side, then

Z Z
g · dS = −4πG ρ dV
V
⇒ − 2πRl|g| = −4πGl · M (r)
| {z }
enclosed mass
per unit length
2GM (R)
⇒ g=− R̂
R

Example (Infinite planar distribution of mass). Assume infinite and homo-

geneous in x and y, ρ = ρ(z).

15
C. G RAVITATION

Figure C.3: Planar distribution of mass

By symmetry, g is in −ẑ direction and is constant on a z = const. surface.

So, if we also have reflection symmetry about z = 0,
Z Z
g · dS = −4πG ρ dV
S
ZV z
⇒ − 2|g|A = −4πGA ρ(z) dz
−z
Z z
⇒ g = −4πGẑ ρ(z) dz
0

(For planar distribution of finite height zmax , g is constant for z ≥ zmax .)

Example (Finite axisymmetric disk).

Figure C.4: Finite axisymmetric disk

16
 C.2. Potential of a Spherical Mass Distribution

I
g · dS = ?
S
No surfaces where g vanishes by symmetry, and no easily determined
surfaces where |g| is a constant. We would need to solve Poisson’s equation
directly, e.g. using separation of variables.

C.2 Potential of a Spherical Mass Distribution

We found that, for a spherical distribution,
dΨ
Z r
G
g = −|g|r̂, |g| = 2 4πρ(r0 )r02 dr0 =
r 0 dr
so, Z r0 Z r
G

0 02 0
Ψ= 4πρ(r )r dr dr
∞ r2 0
Taking Ψ(∞) = 0 by convention, integrate this by parts:
r0
G r
Z r0
G
 Z 
Ψ=− 4πρ(r0 )r02 dr0 + 4πρ(r)r2 dr
r 0 r=∞ ∞ r
GM (r0 )
Z r0
⇒ Ψ=− + 4πGρ(r)r dr
r0 ∞

where we have made an assumption that M (r)/r → 0 as r → ∞.

We find that Ψ is affected by matter outside of r through our choice of
setting Ψ = 0 at infinity. So Ψ 6= −GM (r)/r unless there is no mass outside
of r.

C.3 Gravitational Potential Energy

For a given system of point masses,
X GMi
Ψ=−
i
|r − ri |

and the energy required to take a unit mass to ∞ is −Ψ. Energy required to
take a system of point masses to ∞ is
1 X X GMi Mj 1X
Ω=− = Mj Ψj
2 j6=i i |rj − ri | 2 j

where the half is present to avoid double counting pairs.

For a continuum matter distribution,
1
Z
Ω= ρ(r)Ψ(r) dV
2

17
C. G RAVITATION

Specialising to the spherically symmetric case gives

1
Z ∞
Ω= 4πρ(r)r2 Ψ(r) dr
2 0

Rr
Integrate by parts, choosing parts u ≡ Ψ, dv ≡ 4πρr2 so that v =
0 4πρr02 dr0 = M (r), then

1 ∞
dΨ
 Z ∞ 
Ω= M (r)Ψ(r) − M (r) dr .
2 0 0 dr

Assuming that we have a finite distribution of mass with a non-singular

behaviour at r = 0, the first term on the RHS (i.e. the boundary term) is zero.
Noting further that
dΨ GM (r)
=
dr r2
we conclude
1 GM (r)2
Z ∞
Ω=− dr .
2 0 r2

Integrate again by parts, choosing u ≡ GM (r)2 , dv ≡ 1/r2 ,

1 1 ∞
1 1 dM
Z ∞
Ω = GM (r)2 − 2GM dr
2 r 0 2 0 r dr
| {z }
=0
M (r)
Z ∞
⇒ Ω = −G dM
0 r

This is equivalent to the assembly of spherical shells of mass, each brought

from ∞ with potential energy

GM (r)
dM (r) .
r

C.4 The Virial Theorem

We now come to a powerful result that greatly helps in the understanding of
isolated gravitating systems.
Consider the motion of a cloud of particles (atoms, stars, galaxies...).
Particle with mass mi at ri is acted upon by a force

d2 ri
F = mi
dt2

18
 C.4. The Virial Theorem

Consider the 2nd derivative of the scalar moment of inertia, Ii = mi ri2

1 d2 d dri
 
(mi ri2 ) = mi ri ·
2 dt2 dt dt
d ri
2 dri 2
 
= mi ri · 2 + mi
dt dt
dri 2
 
= ri · Fi + mi
dt
| {z }
2×Kinetic Energy Ti

If I ≡ 2 then we can sum the previous equation over all particles to give
P
i mi ri

1 d2 I
= (ri · Fi ) + 2T
X
2 dt 2
i
| {z }
V , the virial
(R. Clausius)

In the absence of external forces (i.e. an isolated system), we have that

Fi = j Fij where Fij is the force exerted on the ith particle by the jth
P

particle. Consider any two particles with mi and mj at ri and rj , Newton’s

3rd Law says
Fij = −Fji
and so their contribution to the virial is Fij · (ri − rj ). We then have

V = Fij · (ri − rj )
XX

i j>i

If there are no non-gravitational interactions except for possibly when

ri = rj , all forces other than gravitational can be neglected and
Gmi mj
Fij = − 3 rij where rij = ri − rj
rij

Thus we have
X X Gmi mj
V =−
i j>i
rij
where each term is the work done to separate each pair of particles to infinity
against gravity.
And so, V = Ω and we can use above to write

1 d2 I
= 2T + Ω
2 dt2
If the system is in a steady state (“relaxed”), then I = const. and we can
say
2T + Ω = 0 The Virial Theorem

19
C. G RAVITATION

Here, the kinetic energy T has contributions from local flows and
random/thermal motions.

gravitational potential sets the “temperature”

Virial theorem ⇒
or velocity dispersion of the system

20
 CHAPTER D

E QUATIONS OF S TATE AND THE

E NERGY E QUATION

D.1 The Equation of State

In 3-dimensions, the (scalar) equation of mass conservation and the (vector)
equation of momentum conservation can be written as 4 independent scalar
equations. Given appropriate boundary conditions, these must be solved in
order to find the density (scalar field), pressure (scalar field), gravitational
potential (scalar field), and velocity components (3-d vector field); a total of
six degrees of freedom.
To close the system of equations, we need additional information.
Specifically, we need to find relations between Ψ, p and the other fluid variables
such as ρ and u.
Ψ(r) and ρ are related via Poisson’s equation (and/or we sometimes consider
an externally imposed gravitational potential).
p and the other thermodynamic properties of the system are related by the
equation of state (EoS). This is only valid for collisional fluids.
Most astrophysical fluids are quite dilute (particle separation much larger
than effective particle size) and can be well approximated as ideal gases. The
corresponding EoS is
kB
p = nkB T = ρT
µmp
where µ is the mean particle mass in units of the proton mass mp . (Exceptions,
where significant deviation from ideal gas behaviour occurs, can be found in
high density environments of planets, neutron stars and white dwarfs.)
The ideal gas EoS introduces another scalar field into the description of the
fluid, the temperature T (r, t). In general, we need to solve another PDE that
describes heating and cooling processes in order to close the set of equations.
We shall move on to this soon — see “Section D.2 : The Energy Equation”.
However, for special cases, we can relate T and ρ without the need to solve
a separate energy equation. Fluids for which p is only a function of ρ are

21
D. E QUATIONS OF S TATE AND THE E NERGY E QUATION

known as barotropic fluids.

Example (Isothermal case). T is constant so that p ∝ ρ. Valid when the fluid

is locally in thermal equilibrium with strong heating and cooling processes
that are in balance.

Example (Adiabatic case). Ideal gas undergoes reversible thermodynamic

changes such that
p = Kργ
where K, γ are constants.

Derivation. First law of thermodynamics is

đQ = dE
|{z} + p dV
|{z} | {z }
heat absorbed by change in internal work done by
unit mass of fluid energy of unit unit mass of fluid
from surronding mass of fluid

Here đ is a Pfaffian operator — change in quantity depends on the path

taken through the thermodynamic phase space. For an ideal gas, we can write
R∗
p= ρT, E = E(T )
µ
where R∗ is a modified gas constant.
So, first law reads
dE
đQ = dT + p dV
dT
R∗ T
= CV dT + dV
µV
where we define specific heat capacity at constant volume as CV ≡ dE/dT
and have noted that for unit mass we have ρ = 1/V .
For a reversible change we have đQ = 0, so
R∗ T
CV dT + dV = 0
µV
R∗
⇒ CV d(ln T ) + d(ln V ) = 0
µ
⇒ V ∝ T −CV µ/R∗
⇒ p ∝ T 1+CV µ/R∗

CV depends on the number of degrees of freedom with which the gas can
store kinetic energy, f such that
R∗
CV = f
2µ

22
 D.2. The Energy Equation

Monatomic gas has f = 3 ⇒ CV = 3R∗ /2µ; diatomic gas at a few×100 K

(two rotational modes excited) has f = 5 ⇒ CV = 5R∗ /2µ.
Returning to the ideal gas law,

R∗
p= ρT with ρ = 1/V for a unit mass of fluid
µ
R∗ T
⇒ pV =
µ
R∗
⇒ p dV + V dp = dT
µ

but
dE
đQ = dT + p dV
dT 
dE R∗

= + dT − V dp
dT µ
| {z }
specific heat capacity
at constant pressure, Cp

so,
R∗
Cp − CV =
µ
Let us define
Cp
γ≡
CV
so that, for the reversible/adiabatic processes discussed above, we have

p ∝ T 1+CV µ/R∗ ⇒ p ∝ T γ/(γ−1)

V ∝ T −CV µ/R∗ ⇒ V ∝ T −1/(γ−1)

which we can combine to give

p ∝ ργ


We say that a fluid element behaves adiabatically if p = Kργ with K =

constant. A fluid is isentropic if all fluid elements behave adiabatically with
the same value of K. K is related to the entropy per unit mass.

D.2 The Energy Equation

In general, the equation of state will not be barotropic and we will need to
solve a separate differential equation which follows the heating and cooling
processes in the gas, the energy equation.

23
D. E QUATIONS OF S TATE AND THE E NERGY E QUATION

From the first law of thermodynamics we have

đQ = dE + p dV in absence of dissipative processes

| {z }
dW =−pdV

so,
DE DW đQ
= +
Dt Dt dt
with
DW D 1 p Dρ
 
= −p =
Dt Dt ρ ρ2 Dt
and
đQ
≡ −Q̇cool rate of cooling per unit mass
dt
therefore,
DE p Dρ
= 2 − Q̇cool
Dt ρ Dt
The total energy per unit volume is
1
E = ρ( u2 + |{z}
Ψ + |{z}
E )
2
|{z} potential internal
kinetic
so,
DE Dρ E Du DΨ p Dρ
 
= +ρ u· + + 2 − Q̇cool
Dt Dt ρ Dt Dt ρ Dt
where
DE ∂E
≡ + u · ∇E
Dt ∂t

Dρ
= −ρ∇ · u
Dt

Du
ρ = −∇p + ρg = −∇p − ρ∇Ψ
Dt

DΨ ∂Ψ
≡ + u · ∇Ψ
Dt ∂t
Putting all together
DE E ∂Ψ
= − ρ∇ · u − u · ∇p − ρu · ∇Ψ + ρ
Dt ρ ∂t
p
+ ρu · ∇Ψ − ρ∇ · u − ρQ̇cool
ρ
∂E ∂Ψ
⇒ +u · ∇E = −(E + p)∇ · u − u · ∇p + ρ − ρQ̇cool
∂t ∂t

24
 D.3. Heating and Cooling Processes

which gives

∂E ∂Ψ
+ ∇ · [(E + p)u] = ρ − ρQ̇cool Energy Equation
∂t ∂t
In many settings, ∂Ψ/∂t = 0, i.e. Ψ depends on position only. If, further, we
have no cooling (Q̇cool = 0), then this equation expresses the conservation of
energy in which the Eulerian change in total energy density E is driven by the
divergence of the enthalpy flux (E + p)u.

D.3 Heating and Cooling Processes

The Q̇cool term in the energy equation describes processes that locally cool
(Q̇cool > 0) or locally heat (Q̇cool < 0) the fluid. There are many such processes
and a full discussion of them would be lengthy. Here, we discuss just a small
number of important cases.

(1) Cooling by radiation : energy carried away from fluid by photons.

(i) Energy loss by recombination of an ionized gas (line emission as
electrons cascade down energy levels);
(ii) Energy loss by free-free emission (free electrons accelerated in electric
fields of ions)
Lff ∝ ne np T 1/2
(iii) Collisionally-excited atomic line radiation (electron collides with
atom in ground state → produces excited atomic state which returns
to ground state by emitting a photon with energy χ)
√
Lc ∝ ne nion e−χ/kT χ/ T

In cold gas clouds with T ∼ 104 K, H cannot be excited so cooling

occurs through trace species (O+ , O++ , N+ ).
These are all two-body interactions ⇒ cooling rate per unit volume
proportional to ρ2 . Recalling that Q̇cool is defined per unit mass, such
processes give Q̇cool = ρf (T ).
(2) Heating by cosmic rays : heating and energy transport via high-energy
(often relativistic) particles that are diffusing/streaming through the
fluid.
– High energy particles ionise atoms in fluid, excess energy put into
freed e− which ends up as heat in fluid.

ionisation rate per unit volume ∝ CR flux × ρ

⇒ Q̇cool ∝ CR flux. (independent of ρ)

25
D. E QUATIONS OF S TATE AND THE E NERGY E QUATION

Combining these cases, we can parametrise Q̇cool as:

Q̇cool = AρT α − H
|{z}
| {z }
radiative CR heating
cooling

where α depends upon the physics of the dominant radiative cooling process.

D.4 Energy Transport Processes

Transport processes move energy through the fluid. Important examples are:

(1) Thermal conduction — transport of thermal energy by diffusion of the

hot e− into cooler regions. Relevant in, for example

– Interiors of white dwarfs;

– Supernova shock fronts;
– ICM plasma.

There is also thermal conduction associated with ions,

p but it is smaller
than the electron thermal conduction by a factor of mion /me ∼ 43.
The energy flux per unit area is

Fcond = −κ∇T

where κ is thermal conductivity (computed from kinetic theory).

The rate of change of E per unit volume is

−∇ · Fcond = κ∇2 T

(2) Convection — transport of energy due to fluctuating or circulating fluid

flows in presence of entropy gradient. Important in cores of massive stars,
or interiors of some planets, or envelopes of low-mass stars.

(3) Radiation transport — relevant in optically-thick systems (mean free

path of photon much shorter than size of system).
If scattering opacity dominates, then we have radiative diffusion. If rad
is the energy density of the radiation field, the radiative flux through the
fluid is
Frad ∝ −∇rad

The general topic of radiation transport through a fluid flow is very

complex and beyond the scope of this course.

26
 CHAPTER E

H YDROSTATIC E QUILIBRIUM ,
ATMOSPHERES AND S TARS

We now have the full set of equations describing the dynamics of an ideal
(inviscid, dilute, unmagnetized) non-relativistic fluid:

∂ρ
+ ∇ · (ρu) = 0 Continuity Equation
∂t
∂u
ρ + ρ(u · ∇)u = −∇p + ρg Momentum Equation
∂t
∇2 Ψ = 4πGρ Poisson’s Equation
∂E ∂Ψ
+ ∇ · [(E + p)u] = ρ − ρQ̇cool Energy Equation
∂t ∂t
1 2
 
E=ρ u +Ψ+E Defn of total energy
2
kB
p= ρT EoS for ideal gas
µmp
3p
E= Internal energy (monoatomic)
2ρ
We proceed to use those equations to explore astrophysically relevant
situations.
This chapter starts with the simplest, but important, case — fluid systems
that are a static equilibrium with pressure forces balancing gravity.

E.1 Hydrostatic Equilibrium

A fluid system is in a state of hydrostatic equilibrium if

∂
u = 0, =0
∂t

27
E. H YDROSTATIC E QUILIBRIUM , ATMOSPHERES AND S TARS

Then, continuity equation is trivially satisfied

∂ρ
+ ∇ · (ρu) = 0
∂t
Momentum equation gives
∂u
ρ + ρ(u · ∇)u = −∇p − ρ∇Ψ = 0
∂t
resulting in

1
∇p = −∇Ψ Equation of Hydrostatic Eqm.
ρ

Assuming a barotropic equation of state p = p(ρ), this system of equations can

be solved.

Example (Isothermal atmosphere with constant (externally imposed) g).

Suppose g = −gẑ. Then equation of hydrostatic equilibrium with isothermal
equation of state reads
1
A · ∇ρ = −∇Ψ = −gẑ
ρ
gz
⇒ ln ρ = − + const.
A
µg

⇒ ρ = ρ0 exp − z
R∗ T
i.e. exponential atmosphere.
Examples of this is the Earth’s atmosphere: T ∼ 300 K and µ ∼ 28 ⇒
e-folding ∼ 9 km. The highest astronomical observatories are at z ∼ 4 km, so
have ρ and p ∼ 60% of sea level.

If system is self-gravitating (rather than having an externally imposed

gravitational field), we also have

∇2 Ψ = 4πGρ

This must be solved together with the equation of hydrostatic equilibrium.

Example (Isothermal self-gravitating slab). Consider static, isothermal slab

in x and y which is symmetric about z = 0 (e.g. two clouds collide and generate
a shocked slab of gas between them).
R∗
Isothermal ⇒ p= ρT ⇒ p = Aρ, A const.
µ

also, ∇ = ∂
∂z due to symmetry, p = p(z), Ψ = Ψ(z).

28
 E.1. Hydrostatic Equilibrium

Then the equation of hydrostatic equilibrium becomes

1
A ∇ρ = −∇Ψ
ρ
d dΨ
⇒ A (ln ρ) = −
dz dz
⇒ Ψ = −A ln(ρ/ρ0 ) + Ψ0 (ρ0 = ρ(z = 0))
⇒ ρ = ρ0 e−(Ψ−Ψ0 )/A .
Since A ∝ T , we note that this last equation has the form of a Boltzmann
distribution.
Poisson’s equation is
d2 Ψ
= 4πGρ0 e−(Ψ−Ψ0 )/A .
dz 2
Let’s change variables to χ = −(Ψ − Ψ0 )/A, Z = z 2πGρ0 /A so that Poisson’s
p

equation becomes
d2 χ dχ
= −2eχ χ= = 0 at Z = 0
dZ 2 dZ
dχ d2 χ dχ χ
⇒ = −2 e
dZ dZ" 2 dZ
1 d dχ 2 d χ
  #
⇒ = −2 (e )
2 dZ dZ dZ
dχ
 2
⇒ = C1 − 4eχ .
dZ
But we have boundary condition dχ/dZ = 0 when χ = 0 ⇒ C1 = 4.
dχ √ dχ
Z Z
∴ = 2 1 − eχ ⇒ √ =2 dZ .
dZ 1 − eχ
Change variables eχ = sin2 θ
2 cos θ
⇒ eχ dχ = 2 sin θ cos θ dθ or dχ = dθ .
sin θ
So, we can evaluate χ integral
dχ 2 cos θ dθ
Z Z
√ = √
1 − eχ sin θ 1 − sin2 θ
2 dθ
Z
=
sin θ
1 1 + t2
Z
= 2· dθ
2 t
dt
Z
=2
t
= 2 ln t + C2

29
E. H YDROSTATIC E QUILIBRIUM , ATMOSPHERES AND S TARS

by setting
θ 1
t = tan ⇒ dt = (1 + t2 ) dθ
2 2
and by noting
θ θ 2t
sin θ ≡ 2 sin cos = = eχ/2
2 2 1 + t2
So, Poisson’s equation becomes

2 ln t = 2Z + C2

Now, χ = 0 at Z = 0 ⇒ θ = π/2, t = 1 ⇒ C2 = 0, so t = eZ

2eZ 1
⇒ sin θ = eχ/2 = =
1 + e2Z cosh Z
This gives
s 
2πGρ0 
Ψ − Ψ0 = 2A ln cosh  z
A
ρ
ρ= q0 
cosh2 2πGρ0
A z

E.2 Stars as Self-Gravitating Polytropes

Consider a spherically-symmetric self-gravitating system in hydrostatic
equilibrium; from now on we will refer to this as a “star”. We have

∇p = −ρ∇Ψ
dp dΨ
⇒ = −ρ (spherical polar)
dr dr
Now, ρ > 0 within star ⇒ p is monotonic function of Ψ. Also,
dp dp dΨ dΨ dp
= = −ρ ⇒ ρ=−
dr dΨ dr dr dΨ
So ρ is monotonic function of Ψ.

∴ p = p(Ψ), ρ = ρ(Ψ) ⇒ p = p(ρ)

i.e. non-rotating stars are barotropes!

A barotropic EoS can be written as

p = Kρ1+1/n

30
 E.2. Stars as Self-Gravitating Polytropes

where in general n = n(ρ). When n = constant, we say that we have a

polytropic EoS and the structure is called a polytrope. Real stars are in fact
well approximated as polytropes.
It is important to note that in general we will have
1
1+ 6= γ.
n
We only have 1 + 1/n = γ (i.e. p ∝ ργ ) if the star is isentropic (constant
entropy throughout) due to, for example, mixing by convective motions.
Assuming a polytropic EoS, the equation of hydrostatic equilibrium gives
1  1+1/n   
− ∇Ψ = ∇ Kρ = (n + 1)∇ Kρ1/n
ρ
ΨT − Ψ n
 
⇒ ρ= ΨT ≡ Ψ where ρ = 0, the surface.
[n + 1]K
If the central density if ρc and central potential is Ψc , we have
ΨT − Ψc
 n
ρc =
[n + 1]K
so we can write,
ΨT − Ψ
 n
ρ = ρc
ΨT − Ψc
Feeding this into Poisson’s equation gives
ΨT − Ψ
 n
∇2 Ψ = 4πGρc
ΨT − Ψc
Define θ = (ΨT − Ψ)/(ΨT − Ψc ), we then get
4πGρc n
∇2 θ = − θ
ΨT − Ψc
In spherical polars, this becomes
1 d dθ 4πGρc n
 
r2 =− θ
r dr
2 dr ΨT − Ψc

Defining a scaled radial coordinate ξ = r (4πGρc )/(ΨT − Ψc ), we finally get

p

1 d dθ
 
ξ2 = −θn Lane-Emden Eqn. of Index n
ξ 2 dξ dξ

The appropriate boundary conditions for the Lane-Emden equation are

θ = 1, dθ/dξ = 0 at ξ = 0. (Zero force at ξ = 0, enclosed mass → 0 as ξ → 0.)
The Lane-Emden equation can be solved analytically for n = 0, 1 and 5;
otherwise solve numerically.

31
E. H YDROSTATIC E QUILIBRIUM , ATMOSPHERES AND S TARS

Solution for n = 0
This is a somewhat singular case, physically corresponding to a fluid that is
constant density and incompressible.
1 d dθ
 
ξ2 = −θn = −1
ξ dξ
2 dξ
d dθ
 
⇒ ξ2 = −ξ 2
dξ dξ
dθ 1
⇒ ξ2 = − ξ3 − C
dξ 3
ξ 2 C
⇒ θ =− + +D
6 ξ
We need θ = 1 at ξ = 0 ⇒ C = 0, D = 1.
ξ2
∴ θ =1−
6
For solutions for n = 1 or n = 5 cases, see the book [PAFD] (section 5.5.2
& 5.5.3).

E.3 Isothermal Spheres (Case n → ∞)

The isothermal case p = Kρ corresponds to n → ∞. Let’s combine
dp dΨ
= −ρ and p = Kρ
dr dr
dΨ K dρ
⇒ =−
dr ρ dr
⇒ Ψ − Ψc = −K ln(ρ/ρc )

From Poisson’s equation

∇2 Ψ = 4πGρ
1 d 2 dΨ
 
⇒ r = 4πGρ
r2 dr dr
K d 2 1 dρ
 
⇒ r = −4πGρ
r2 dr ρ dr
Let ρ = ρc e−Ψ (defining Ψc = 0), and set
s
K
r = aξ, a=
4πGρc
then
1 d dΨ
 
ξ2 = e−Ψ
ξ 2 dξ dξ

32
 E.4. Scaling Relations

with Ψ = dΨ/dξ = 0 at ξ = 0.
This replaces the Lane-Emden equation in the case where the system is
isothermal.
At large radii, this has solutions of the form ρ ∝ r−2 , so the enclosed mass
∝ r. Thus, the mass of an isothermal sphere of self-gravitating gas tends to ∞
as the radius tends to ∞. This is why we cannot adopt our usual convention
of defining Ψ = 0 at ∞.
So, to be physical, isothermal spheres need to be truncated at some finite
radius. There needs to be some continuing pressure by an external medium.
These are called Bonnor-Ebert spheres, whose density profile depends on ξcut .
E.g. dense gas core in a molecular cloud is well-described by a Bonnor-Ebert
sphere.

E.4 Scaling Relations

In many circumstances, stars behave as polytropes, e.g. fully convective stars

with p – ρ close to the adiabatic relation. In such a star, assuming monatomic
gas with γ = 5/3, we have p = Kρ5/3 ⇒ n = 3/2.
Consider a set of stars which share a given polytropic index n and a given
adiabat K. They will then form a one-parameter family characterised by their
central density ρc .
Thus one can find how mass and radius vary as a function of ρc and,
eliminating ρc , obtain scaling relations relating the mass and radius.
All stars with given n have the same θ(ξ) since the Lane-Emden equation
does not depend on ρc . Recall

ΨT − Ψ
 n
ρ= ⇒ ΨT − Ψc = K(n + 1)ρ1/n
(n + 1)K c

s v
u 4πGρ1−1/n
u
4πGρc c
ξ= r ⇒ ξ= t
r
ΨT − Ψc K(1 + n)
ΨT − Ψ
 n
ρ = ρc = ρc θn
ΨT − Ψc

The surface of the polytrope is at ξ = ξmax defined as location where we

have θ(ξ) = 0. Let rmax be the corresponding physical radius. Then the total

33
E. H YDROSTATIC E QUILIBRIUM , ATMOSPHERES AND S TARS

mass of the polytrope is

Z rmax
M= 4πr2 ρ dr
0
1−1/n −3/2
4πGρc
" # Z ξmax
= 4πρc θn ξ 2 dξ
K(1 + n)
|0 {z }
same for all
polytrope of index n
1
( n3 −1)
⇒ M ∝ ρc2

From definition of ξ above, we also know that

1
( n1 −1)
rmax ∝ ρc2

Eliminating ρc gives
3−n
M ∝ R 1−n Mass-Radius Relation for Polytropic Stars

For γ = 5/3, n = 3/2 this gives M ∝ R−3 or R ∝ M −1/3 . This suggests

more massive stars have smaller radii.
This relation actually works well for white dwarfs (where the polytropic
EoS is due to e− degeneracy pressure rather than gas pressure). But for most
main-sequence stars we observe M ∝ R.
Reason is that stars do not share the same polytropic constant K. Let’s
write the temperature at the core in terms of the central density and K

p = Kρ1+1/n 

 µK 1/n
R∗ ⇒ Tc = ρ
p= ρT  R∗ c
µ


Nuclear reactions in the core tend to keep Tc similar in the cores of stars
of different masses. So we can say that

K ∝ ρ−1/n
c

Substitute this into above expression for mass when n = 3/2 gives

M ∝ ρc−1/2 , R ∝ ρ−1/2
c ⇒ M ∝R

When can the K = const. relation be applied? Answer: when new mass is
added to a star adiabatically and the nuclear processes have not had time to
adjust. For Sun we have

– Time to adjust to new hydrostatic equilibrium is

th ∼ R/Cs ∼ 1 day

34
 E.4. Scaling Relations

– Time to lose significant energy

GM 2
tth ∼ Etot /L ∼ ∼ 30 Myr
RL

So, mass loss/gain is followed by rapid re-adjustment of hydrostatic

equilibrium but true thermal equilibrium is reached after a much longer time.

Example (Spherical rotating star). Spherical rotating star with angular

velocity Ω gains non-rotating mass. How does Ω evolve?
Conservation of angular momentum gives M R2 Ω = const. So, if Ω →
Ω + ∆Ω then

M R2 ∆Ω + Ω∆(M R2 ) = 0
∆Ω ∆(M R2 )
⇒ =−
Ω M R2

But we can use

R ∝ M (1−n)/(3−n)

to say

∆Ω  
∝ −∆ M (5−3n)/(3−n)
Ω
∆Ω 5 − 3n
 
⇒ ∝− ∆M
Ω 3−n

so,

∆Ω < 0 if >0 (e.g. n = 32 ) Spin Down

(
5−3n
∆M > 0 ⇒ 3−n
∆Ω > 0 if 5−3n
3−n <0 (e.g. n = 2) Spin Up

Example (Star in a binary system). Star in a binary system loses mass to its
companion.
Donor star loses mass, ∆M < 0. So since R ∝ M (1−n)/(3−n) , the radius
will increase if 1 < n < 3.
So there is the potential for unstable (runaway) mass transfer (need to
look at evolution of the size of the Roche lobe to conclusively decide whether
process is unstable).

35
E. H YDROSTATIC E QUILIBRIUM , ATMOSPHERES AND S TARS

Figure E.1: Roche lobe overflow

36
 CHAPTER F

S OUND WAVES , S UPERSONIC

F LOWS AND S HOCK WAVES

F.1 Sound Waves

We now start discussion of how disturbances can propagate in a fluid. We
begin by talking about sound waves in a uniform medium (no gravity). We
proceed by conducting a first-order perturbation analysis of the fluid equations:
∂ρ
+ ∇ · (ρu) = 0
∂t
∂u 1
+ (u · ∇)u = − ∇p
∂t ρ
The equilibrium around which we will perturb is
ρ = ρ0 (uniform & constant)
p = p0 (uniform & constant)
u=0
We consider small perturbations and write in Lagrangian terms (Lagrangian
meaning the change of quantities are for a given fluid element)
p = p0 + ∆p
ρ = ρ0 + ∆ρ
u = ∆u
The relation between Lagrangian and Eulerian perturbations is:
δρ = ∆ρ − ξ · ∇ρ0
|{z} |{z} | {z }
Eulerian Lagrangian Element displacement dot
pert. pert. Gradient of unpert. state

In present example, ∇ρ0 = 0 and so δρ = ∆ρ. But the distinction between

Lagrangian and Eulerian perturbations will be important for other situations
that we’ll address later.

37
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Substitute the perturbations into fluid equations and ignore terms that are
2nd order (or higher) in the perturbed quantities:
Start with continuity equation:

∂
(ρ0 + ∆ρ) + ∇ · [(ρ0 + ∆ρ)∆u] = 0
∂t
0
∂ρ0
7 ∂∆ρ :0 : 2nd order : 2nd order
 + + ∆u +  · ∆u + ρ0 ∇ · (∆u) + 
∆ρ∇
· (∆u) = 0
 
⇒ ∇ρ ·
0

 
∇(∆ρ) 
∂t ∂t
∂
⇒ (∆ρ) + ρ0 ∇ · (∆u) = 0 O
1 eq.f.1.1
∂t
And similarly, the momentum equation:

∂ 1
(∆u) = − ∇(∆p)
∂t ρ0
∂ dp ∇(∆ρ)
⇒ (∆u) = − . assuming barotropic EoS O
2 eq.f.1.2
∂t dρ ρ=ρ0 ρ0

Now, take ∂/∂t of O

1:

∂2 ∂
2
(∆ρ) = −ρ0 [∇ · (∆u)]
∂t ∂t 
∂

= −ρ0 ∇ · (∆u)
∂t
dp
= ∇2 (∆ρ).
dρ ρ=ρ0

We get
∂ 2 (∆ρ) dp
= ∇2 (∆ρ). Wave Equation
∂t 2 dρ ρ=ρ0

This admits solutions of the form ∆ρ = ∆ρ0 ei(k·x−ωt) . Substituting into

the wave equation we get

dp
(−iω)2 ∆ρ0 = (ik)2
dρ ρ=ρ0
dp
⇒ ω2 = k2
dρ ρ=ρ0

The (phase) speed of the wave is vp = ω/k, so the sound wave travels at
speed

dp
s
cs = Sound Speed as the Derivative of p(ρ)
dρ ρ=ρ0

38
 F.1. Sound Waves

Consider a 1-D wave and substitute

∆ρ = ∆ρ0 ei(kx−ωt)
∆u = ∆u0 ei(kx−ωt)

into O
1 . We get

− iω∆ρ + ρ0 ik∆u = 0
ω ∆ρ ∆ρ
⇒ ∆u = = cs
k ρ0 ρ0

So we learn that

– Fluid velocity and density perturbations are in phase (since ∆u/∆ρ ∈ R);

– Disturbance propagates at a much higher speed than that of the individual

fluid elements, provided density perturbations are small, since

∆ρ0
∆u0 = cs  cs
ρ0

Sound waves propagate because density perturbations give rise to a pressure

gradient which then causes acceleration of the fluid elements, this induces
further density perturbations, making disturbances propagate.
Sound speed depends on how the pressure forces react to density changes. If
the EoS is “stiff” (i.e. high dp/dρ ), then restoring force is large and propagation
is rapid.
Examples of dp/dρ :

Example (Isothermal case).

dp
c2s =
dρ T

In this case, compressions and rarefactions are effective at passing heat to

each other to maintain constant T . Then
R∗
p= ρT
µ
s
R∗ T
⇒ cs,I =
µ

Example (Adiabatic case).

dp
c2s =
dρ S

39
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

No heat exchange between fluid elements; compressions heat up and

rarefactions cool down from p dV work. So

p = Kργ
dp γp
⇒ = γKργ−1 =
dρ S ρ
s
γR∗ T
⇒ cs,A =
µ

Notes about these two examples:

√
– We see that cs,I and cs,A differ by only γ;

– Thermal behaviour of the perturbations does not have to be the same as

that of the unperturbed structure!

Background is approximately isothermal

(
E.g. Earth’s atmosphere
Sound waves are adiabatic

– Waves for which cs is not a function of ω are called non-dispersive. The

shape of a wave packet is preserved.

F.2 Sound Waves in a Stratified Atmosphere

We now move to the more subtle problem of sound waves propagating in a
fluid with background structure. For concreteness, let’s consider an isothermal
atmosphere with constant g = −gẑ.
Horizontally travelling sound waves are unaffected by the (vertical)
structure. So let’s just focus on z-dependent terms, taking u = uẑ. Continuity
and momentum equations are:
∂ρ ∂
+ (ρu) = 0 O
1 eq.f.2.1
∂t ∂z
∂u ∂u 1 ∂p
+u =− − g, O
2 eq.f.2.2
∂t ∂z ρ ∂z
and the equilibrium is

u0 = 0
R∗ T
ρ0 (z) = ρ̃e−z/H , H≡
gµ
R∗ T
p0 (z) = ρ0 (z) = p̃e−z/H .
µ

40
 F.2. Sound Waves in a Stratified Atmosphere

Consider a Lagrangian perturbation:

u → ∆u
ρ0 → ρ0 + ∆ρ
p0 → p0 + ∆p.
Remember that δρ = ∆ρ − ξ · ∇ρ. So we have

∂ρ0 
δρ = ∆ρ − ξz 
∂z 



δp = ∆p − ξz
∂p0 Eulerian to Lagrangian perturbation relation
∂z 



δu = ∆u,



and
dξ ∂ξ : 2nd order ∂ξ
∆u = = + u· ∇ξ
 =
dt ∂t ∂t
Substituting perturbed quantities into the Eulerian continuity equation,
∂ ∂
(ρ0 + δρ) + [(ρ0 + δρ)δuz ] = 0
∂t  ∂z
∂ ∂ρ0 ∂

⇒ ρ0 + ∆ρ − ξz + (ρ0 ∆uz ) = 0 (ignoring 2nd order terms)
∂t ∂z ∂z
0
0
7 ∂∆ρ
∂ρ0 ∂ξz ∂ρ0 ∂ ∂ρ
* ∂ρ
0 0 ∂∆uz
⇒  + − − ξz + ∆uz + ρ0 =0
∂t ∂t ∂t ∂z  ∂t ∂z ∂z ∂z

∂∆ρ ∂ρ0 ∂ρ0 ∂∆uz

⇒ − ∆uz + ∆uz + ρ0 =0
∂t | {z } ∂z ∂z ∂z
∂ξz /∂t
∂∆ρ ∂∆uz
⇒ + ρ0 = 0. O
3 eq.f.2.3
∂t ∂z
A similar calculation for the momentum equation gives
∂∆uz 1 ∂∆p
=−
∂t ρ0 ∂z
∂∆uz c2 ∂∆ρ ∂p
⇒ =− u , c2u ≡ . O
4 eq.f.2.4
∂t ρ0 ∂z ∂ρ ρ0

To perform this calculation (which we leave as an exercise!), you need a

relation that is obtained from the Lagrangian continuity equation:
Dρ
+ ρ∇ · u = 0
Dt 
∂ξ

⇒ ∆ρ + ρ0 ∇ · ∆t = 0 (integrating over a short time ∆t)
∂t
⇒ ∆ρ + ρ0 ∇ · ξ = 0.

41
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Let’s now derive the wave equation and dispersion relation. Take ∂/∂t of
3:
O
∂ 2 ∆ρ ∂ ∂∆uz
 
2
+ ρ0 =0
∂t ∂z ∂t
∂ ∆ρ
!
2 ∂ c2u ∂∆ρ
⇒ 2
− ρ0 = 0,
∂t ∂z ρ0 ∂z
where the last step involved substitution from O
4 . If the medium is isothermal,
then cu is independent of z. So,
∂ 2 ∆ρ c2u ∂ 2 ∆ρ c2u ∂ρ0 ∂∆ρ
− ρ0 + ρ0 =0
∂t2 2
 
 ρ
 0 ∂z

ρ
20 ∂z ∂z
∂ 2 ∆ρ 2 ∂ ∆ρ
2 c2u ∂ρ0 ∂∆ρ
⇒ 2
− c u 2
+ =0
| ∂t {z ∂z } ρ0 ∂z ∂z
| {z }
normal sound wave equation extra piece associated
with stratification

Now,
∂ρ0 ∂  −z/H 
= ρ̃e
∂z ∂z
1
= − ρ̃e−z/H
H
ρ0
=−
H
So,
∂ 2 ∆ρ 2 ∂ ∆ρ
2 c2u ∂∆ρ
− cu − =0
∂t2 ∂z 2 H ∂z
Look for solutions of the form ∆ρ ∝ ei(kz−ωt)
ik
⇒ − ω 2 = −c2u k 2 + c2u
H
ik
 
⇒ ω 2 = c2u k 2 − Dispersion Relation
H

We can also write this as

ik ω2
k2 − − 2 =0
H cu
and solve the quadratic for k(ω):
s
i ω2 1
k= ± −
2H 2
cu 4H 2
Let’s take ω ∈ R. We have two cases to examine if we wish to understand the
implications of this dispersion relation.

42
 F.2. Sound Waves in a Stratified Atmosphere

Case I: ω > cu /2H

Examine the real and imaginary parts of k:
1
Im k =
2H
s
1
2 2
ω
 
Re k = ± −
cu 2H
So the density perturbation is
 √ 
−z/2H i ± (ω/cu )2 −(1/2H)2 z−ωt
∆ρ ∝ e| {z } |e {z }
(†) (‡)

corresponding to
(†) Exponentially decaying amplitude with increasing height;

(‡) Wave with phase velocity

s
1
2 2
ω ω

vph = , K≡± −
K cu 2H
where vph is function of ω, meaning that the wave is dispersive. Wave
packet consisting of different ω’s will change shape as it propagates.
As before, we can relate ∆u to ∆ρ:
∆ρ ω
∆uz =
ρ0 k
with

∆ρ ∝ e−z/2H
ρ0 ∝ e−z/H

giving
∆ρ
∆uz ∝ e+z/2H , ∝ e+z/2H
ρ0
Thus the perturbed velocity and the fractional density variation both
increase with height. In the absence of dissipation (e.g. viscosity), the kinetic
energy flux (∝ ∆ρ∆u) is conserved and the amplitude of the wave increases
until
∆ρ
∆u ∼ cs , ∼1
ρ0
where the linear treatment breaks down and the sound wave “steepens” into a
shock. So, in the absence of dissipation, an upward propagating sound wave
from a hand clapping would generate shocks in the upper atmosphere!

43
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Case II: ω < cu /2H

In this case, we find that k is purely imaginary. So,
∆ρ ∝ e|k|z eiωt
This is a non-propagating, evanescent wave. In essence the wave cannot
propagate since the properties of the atmosphere change significantly over one
wavelength, giving rise to reflections.

F.3 Transmission of Sound Waves at Interfaces

Consider two non-dispersive media with a boundary at x = 0. Suppose we
have a sound wave travelling from x < 0 to x > 0. Let the incident wave have
unity amplitude (in, say, the density perturbation), and denote by r and t the
amplitude of the reflected and transmitted waves, respectively:

ei(k1 x−ω1 t)

tei(k2 x−ω2 t)

rei(k3 x−ω3 t)

x=0

Figure F.1: Waves at boundary x = 0

At the boundary x = 0, variables must be single valued and the accelerations

are finite, thus oscillates in the second medium must have the same frequency
∴ ω1 = ω2 = ω3 = ω.
The reflected wave is in the same medium as the incident
∴ k3 = −k1 . (phase speed reversed)
Amplitude of sound wave continuous at x = 0
∴ 1 + r = t,

44
 F.4. Supersonic Fluids and Shocks

and the derivative of the amplitude is continuous at x = 0

∴ k1 (1 − r) = k2 t.
We can combine these relations to get
2k1 k1 − k2
t= r=
k1 + k2 k1 + k2
From these relations we can see that the reflection/transmission of sound
waves strongly depends on the relative sound speeds in the two media:
(i) If cs,2 > cs,1 ⇒ k2 < k1 ⇒ r > 0, i.e reflected wave in phase with
incident;
(ii) If cs,2 < cs,1 ⇒ r < 0 ⇒ reflected wave is π out of phase with incident
wave;
(iii) If cs,2  cs,1 ⇒ k2  k1 ⇒ t  1, i.e. wave almost completely reflected.

F.4 Supersonic Fluids and Shocks

Shocks occur when there are disturbances in the fluid caused by compression
by a large factor, or acceleration to velocities comparable to or exceeding cs
The linear theory applied to sound waves breaks down.
When thinking about the sound speed, recall that the chemical composition
of the fluid matters, cs ∝ µ−1/2
cs in atomic Hydrogen  cs in diatomic Nitrogen for given T
| {z } | {z }
µ≈1 µ=28
e.g. ISM e.g. Earth
atmosphere

Disturbances in a fluid always propagate at the sound speed relative to

the fluid itself. Consider an observer at the centre of a spherical disturbance,
watching the fluid flow past is at speed v.

cs cs
v α
O O v

subsonic case supersonic case

Figure F.2: Subsonic flow vs supersonic flow

45
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

The velocity of the disturbance relative to the observer, v 0 , is the vector

sum of the fluid velocity and the disturbance velocity relative to the fluid.
– Subsonic case: v 0 sweeps 4π steradians;
– Supersonic case: disturbance always to the right. If we continuously
produce a disturbance, the envelope of the disturbances will define a
cone with opening angle α given by
cs
sin α = Mach Cone
v
The ratio of the flow speed to the sound speed is called the Mach number
v
M≡
cs
1
∴ sin α =
M
Imagine an obstacle in a supersonic flow — disturbances cannot propagate
upstream from the obstacle so the flow cannot adjust to the presence
of obstacle. The flow properties must change discontinuously once the
obstacle is reached, giving shock!

F.5 The Rankine-Hugoniot Relations

We analyse a shock by applying conservation of mass, momentum and energy
across the shock front.
In the frame of the shock, lets assume following geometry

p1 , ρ1 p2 , ρ2

u1 u2

pre-shocked gas post-shocked gas

dx

Figure F.3: Geometry of shock front

46
 F.5. The Rankine-Hugoniot Relations

Continuity gives

∂ρ ∂
+ (ρux ) = 0
∂t ∂x !
Z dx/2
∂
⇒ ρ dx + ρux − ρux =0
∂t −dx/2 x=dx/2 x=−dx/2

where we have integrated over a small region dx around the shock.

Let’s take dx → 0 and assume that mass does not continually accumulate
at x = 0. Then
∂
Z 
ρ dx = 0
∂t
⇒ ρ1 u1 = ρ2 u2 1st Rankine-Hugoniot Relation

Apply similar analysis to the momentum equation:

∂ ∂ ∂Ψ
(ρux ) = − (ρux ux + p) − ρ
∂t Z ∂x ∂x
∂

⇒ ρux dx = − (ρux ux + p) + (ρux ux + p)
∂t dx/2 −dx/2

⇒ ρ1 u21 + p1 = ρ2 u22 + p2 2nd R-H Relation

We note that uy and uz do not change across the shock front (can be
immediately seen by looking at the y- and z-components of the momentum
equation).
Now for the energy equation. Start with the adiabatic case so that the gas
cannot cool and hence we have Q̇cool = 0. Also take gravitational potential to
have no time-dependence. Then
0
∂E : 0 ∂Ψ
+ ∇ · [(E + p)u] =  cool + ρ
 
−ρQ̇ 
∂t ∂t
∂E
⇒ + ∇ · [(E + p)u] = 0
∂t Z
∂

⇒ E dx + (E + p)ux − (E + p)ux =0
∂t dx/2 −dx/2
⇒ (E1 + p1 )u1 = (E2 + p2 )u2
 
Since E = ρ 1 2
2u + E + Ψ , this becomes

1
ρ1 u31 + ρ1 E1 u1 + ρ1 Ψ1 u1 + p1 u1
2
1
= ρ2 u32 + ρ2 E2 u2 + ρ2 Ψ2 u2 + p2 u2
2

47
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

But Ψ1 = Ψ2 and ρ1 u1 = ρ2 u2 , so terms involving Ψ cancel out. We are

left with

1 2 p1 1 p2
u + E1 + = u22 + E2 + 3rd R-H Relation
2 1 ρ1 2 ρ2

For an ideal gas, we have

E = CV T 

 CV µ p
R∗  ⇒ E=
p= ρT  R∗ ρ
µ
Cp 
γ= 
R∗

CV 
⇒ CV (γ − 1) =
R∗  µ
Cp − CV = 

µ
which combine to give

1 p
E= (internal energy per unit mass)
γ−1ρ

If we assume that γ does not change across the shock (e.g. there are no
disassociation of molecules), the 3rd R-H relation becomes
1 2 γ p1 1 γ p2
u + = u22 +
2 1 γ − 1 ρ1 2 γ − 1 ρ2
1 2 c2s,1 1 2 c2s,2
⇒ u1 + = u2 +
2 γ−1 2 γ−1
since, for adiabatic case, the sound speed is
∂p γp
c2s = =
∂ρ S ρ

Using all three R-H relations and after some algebra we get

ρ2 u1 (γ + 1)p2 + (γ − 1)p1
= =
ρ1 u2 (γ + 1)p1 + (γ − 1)p2

In the limit of strong shocks, p2  p1 , we get

ρ2 γ+1
→
ρ1 γ−1

For γ = 5/3, this gives ρ2 = 4ρ1 . So there is a maximum possible density

contrast across an adiabatic shock — with stronger and stronger shocks, the
thermal pressure of the shocked gas increases and prevents further compression.

48
 F.5. The Rankine-Hugoniot Relations

Note that, since p2  p1 , and ρ2 ≤ 4ρ1 , we have

p1 p2
6= γ i.e. K1 6= K2
ργ1 ρ2
The gas has jumped adiabats during its passage through the shock. Shocking
the gas produces a non-reversible change, due to viscous processes operating
within shock.
While the R-H conditions are symmetric in the up- and down-stream
quantities, the thermodynamic requirement that entropy increases dictates the
direction of the jump (i.e. a fast/cold upstream flow shocking to produce a
slow/fast downstream flow).
It is interesting that we can derive R-H conditions using the inviscid
equations that do not explicitly include dissipation/entropy-generating terms.
Not all shocks are adiabatic! To consider the other extreme, let’s discuss
isothermal shocks. Here we have Q̇cool 6= 0 such that the shocked gas cools
to produce T2 = T1 . Whether a shock is isothermal or adiabatic depends
on whether the “cooling length” is smaller or larger than the system size,
respectively.

T1

thickness of
shock set by
x
viscosity
cooling length

Figure F.4: Temperature profile through a shock

For isothermal shocks, the first two R-H equations are unchanged:
ρ1 u1 = ρ2 u2
ρ1 u21 + p1 = ρ2 u22 + p2
but the 3rd R-H equation is replaced by
T1 = T2

49
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Now,
s
R∗ T
cs,I = ⇒ cs,1 = cs,2
µ
p
r
= ⇒ p = c2s,I ρ
ρ

So, 2nd R-H equation becomes

ρ1 (u21 + c2s ) = ρ2 (u22 + c2s )

c2s c2
⇒ u1 + = u2 + s (since ρ1 u1 = ρ2 u2 )
u u2
 1
1 1

⇒ c2s − = u2 − u1
u1 u2
u2 − u1
⇒ c2s = u2 − u1
u1 u2
⇒ c2s = u1 u2

Thus we see that

2
ρ2 u1 u1

= = = M12
ρ1 u2 cs
where M1 is the Mach number of the upstream flow. So the density compression
can be very large.
Note that since c2s = u1 u2 and u1 > cs (condition for a shock), we must
have u2 < cs . So flow behind the shock is subsonic. In fact this is always true
for any shock and is necessary to preserve causality (the post shock gas must
know about the shock!).

F.6 Theory of Supernova Explosions

An important application of shock wave theory is to supernova explosions
in the interstellar medium (ISM). A supernova (SN) deposits about 1051 erg
(= 1044 J) of energy into the surrounding medium, then shocked medium
expands, sweeps up more gas, and creates large bubbles in the ISM.
Consider following system:

– Uniform density medium with density ρ0 ;

– Point explosion with energy E;

– Ignore temperature of the ambient ISM (T0 = 0), thus no confinement of

explosion by an external pressure.

50
 F.6. Theory of Supernova Explosions

ρ0 ρ0

ρ1 , p1
R

t=0 t>0 shock layer

thickness D

Figure F.5: Supernova explosion

Given that T0 = 0, the shock has M → ∞. Assuming an adiabatic shock,

we sweep mass into a shell with density ρ, given by
γ+1
ρ1 = ρ0
γ−1
If all mass is swept up into shell then
4π
ρ0 R3 = 4πρ1 R2 D (assuming D  R)
3
1 γ−1
 
⇒ D= R
3 γ+1
For γ = 5/3, we have D ≈ 0.08R which justifies the assumption D  R.
Assume that all gas in the shell moves with a common velocity. In the
frame of a local patch of the shock we have

ρ0 ρ1

u0 u1

Figure F.6: Situation seen in shock frame

51
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

and so

ρ0 u0 = ρ1 u1
ρ0 γ−1
⇒ u1 = u0 = u0 . O
1 eq.f.6.1
ρ1 γ+1

Thus, relative to the unshocked gas, the velocity of the shocked gas U is

2u0
U = u0 − u1 = .
γ+1

Then, the rate of change of momentum of the shocked shell is

d 4π 2u0
 
ρ0 R3
dt 3 γ+1

This momentum gain is provided by pressure acting on the inside surface of

the shell — call this pin . Let’s make the ansatz that this is related to the
pressure within the shell by
pin = αp1 ,

and we now relate p1 and u0 using the R-H jump condition: we have

p0 + ρ0 u20 = p1 + ρ1 u21
" #
ρ1 u21
⇒ p1 = ρ0 u20 1− (since ρ0 = 0 by assumption )
ρ0 u20
γ−1
 
= ρ0 u20 1 − (assuming a strong shock)
γ+1
2
= ρ0 u20 O
2 eq.f.6.2
γ+1

So, equating rate of change of momentum of the shocked shell to the

pressure acting on the inside surface of the shell, we have

d 4π 2u0
 
ρ0 R3 = 4πR2 pin
dt 3 γ+1
= 4πR2 αp1
2
= 4πR2 ρ0 u20
γ+1

d h 3 i
⇒ R u0 = 3αR2 u20
dt
d h 3 i
⇒ R Ṙ = 3αR2 Ṙ2 since u0 ≡ Ṙ
dt

52
 F.6. Theory of Supernova Explosions

This admits solutions of the form R ∝ tb :

d  3b b−1   2
⇒ t bt = 3αt2b btb−1
dt
⇒ b(4b − 1)  = 3αb2
t4b−2  (cancellation justifies assumed form of solution)
t4b−2
1
⇒ b = 0 (not physical) or b=
4 − 3α
1/(4−3α) (3α−3)/(4−3α)
⇒ R∝t , u0 ∝ t ∝ R3α−3
To determine α, we need to consider energy conservation. For an adiabatic
shock, the explosion energy is conserved and transformed into kinetic and
internal energy:
– Kinetic energy of the shell is
1 4π
· ρ0 R3 U 2
2 3
– Internal energy per unit mass is
1 p
E=
γ−1ρ
and so the internal energy per unit volume is
1
ρE = p
γ−1
Since the shell is very thin, it has small volume and so most of the
internal energy is in the central cavity which contains little mass
4π 3 pin 4π 3 p1
Int. Energy of cavity ≈ R = R α
3 γ−1 3 γ−1
So, energy conservation says that
1 4π 4π 3 p1
E= · ρ0 R3 U 2 + R α
2 3 3 γ−1
1 4π 2u0 4π 2 1
 2
= · ρ0 R3 + R3 α ρ0 u20
2 3 γ+1 3 γ+1 γ−1
| {z
O
} | {z
O
}
1 2
4π 3 2 1 4 2
 
= R u0 ρ0 + αρ0 ,
3 2 (γ + 1)2 (γ + 1)(γ − 1)
from which we conclude that
E ∝ R3 u20 ∝ t(6α−3)/(4−3α) .
But E must be conserved. So we need α = 1/2 to remove time dependence of
E. Using α = 1/2 we find

R ∝ t2/5 , u0 ∝ t−3/5 , p1 ∝ t−6/5

53
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Similarity Solutions
The above problem only has 2 parameters, E and ρ0 . Look at their dimensions
M L2 M
[E] = , [ρ0 ] =
T2 L3
These cannot be combined to give quantities with the dimension of length
or time. So, there is no natural length scale or time scale in the problem!
Given some time t, the only way to combine E, ρ0 and t to give a length
scale is !1/5
Et2
λ=
ρ0
We can define a dimensionless distance parameter
1/5
r ρ0

ξ≡ =r
λ Et2
Then, for any variable in the problem X(r, t), we will have

X = X1 (t)X̃(ξ)

i.e. X is a function of scaled distance ξ always has the same shaped scaled
up/down by the time dependence factor X1 (t).
So,

∂X dX̃ ∂ξ
= X1
∂r dξ ∂r t
∂X dX1 dX̃ ∂ξ
= X̃(ξ) + X1
∂t dt dξ ∂t r

ξ is neither a Lagrangian or an Eulerian coordinate. It labels a particular

feature in the flow (e.g. shock wave) that can move through the fluid. So we
can write  1/5
E
Rshock ∝ t2/5
ρ0
Let’s put some numbers in for the case of supernova explosions,
1/5
E

R(t) = ξ0 t2/5 (we will assume ξ0 ∼ 1)
ρ0
dR 2 2R
1/5
E

u0 (t) = = ξ0 =
dt 5 ρ0 t3 5 t
In a supernova we have

E ≈ 1044 J = 1051 erg

ρ0 = ρISM ≈ 10−21 kg m−3

54
 F.6. Theory of Supernova Explosions

So similarity solution gives

R ≈ 0.3t2/5 pc
)
where t is measured in yrs
u0 ≈ 105 t−3/5 km s−1

The original explosion injects the stellar debris at about 104 km s−1 . So
the above solution is valid for

t & 100 yr (when u0 < uinj )

t . 105 yr (after which energy losses become important)

Structure of the Blast Wave

We can, in principle, write each variable ρ, p, u, r in terms of separated functions
of t and ξ. We can then substitute into the Eulerian equation of fluid dynamics
(in spherical coordinates with ∂/∂φ = ∂/∂θ = 0, i.e. spherical symmetry).
The result is a set of ODE’s where ξ is the only dependent variable — the
time dependence cancels out! (Sedov 1946)
E.g. solution for γ = 7/5:

p̃(ξ)

ũ(ξ) ρ̃(ξ)

Figure F.7: Solution for γ = 7/5

These solutions tell us that

– Most of mass is swept up in a shell just behind the shock (from form of
ρ̃);

– Post-shock pressure is indeed a multiple of pin (from form of p̃ justifies

pin = αp1 assumption);

– Shell material is not really moving at a single velocity, but arguments

above are restored by taking some weighted average (from form of ũ).

55
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

Breakdown of the Similarity Solution

The self similar solution breaks down when the surrounding medium pressure
p0 becomes significant, p1 ∼ p0 .
From the strong shock solution, we derived

2 γp0
p1 = ρ0 u20 , c2s =
γ+1 ρ0

So if p1 ∼ p0 then

2 ρ0 c2s
ρ0 u20 ∼
γ+1 γ
⇒ u0 ∼ cs

i.e. shell not moving supersonically anymore.

The blast wave weakens to a sound wave:

ρ ρ

shock
max radius sound
of cavity wave

cavity
ρ0 ρ0

shell

r r

Figure F.8: Blast wave phase vs late phase

As a sound wave, disturbance passes into the undisturbed gas as a mild

compression followed by a rarefaction. After the sound wave passes, gas returns
to the original state.
For SN, the maximum bubble/cavity size is set by the radius when blast
wave becomes sonic and p1 ∼ p0 . We’ve just shown that this implies

γ+1 2
u20 ∼ c
2γ s

56
 F.6. Theory of Supernova Explosions

We showed above that energy conservation gives

4π 3 1 2u0 2 α 2ρ0 u20 1

"   #
E= R ρ0 + with α =
3 2 γ+1 γ−1 γ+1 2
4π 3 2 1
 
= R ρ0 u20 +
3 (γ + 1) 2 (γ − 1)(γ + 1)
4π 3 2(γ − 1) + γ + 1
 
= R ρ0 u20
3 (γ + 1)2 (γ − 1)
4π 3 3γ − 1
= R ρ0 u20
3 (γ + 1)2 (γ − 1)

(γ + 1)(γ 2 − 1) 3E γ+1 2
⇒ u20 = · ∼ c
3γ − 1 4πρ0 R 3 2γ s
| {z }
when blast
wave becomes
sonic and p1 ∼p0
4π 3 c2s 3γ − 1
⇒ E∼ ρ0 Rmax ·
3 2γ γ 2 − 1
Internal energy initially contained within Rmax is
4π 3 p0 4π 3 c2s
Einit = Rmax = Rmax ρ0
3 γ−1 3 γ(γ − 1)
So, when p0 ∼ p1 , we have E ∼ Einit . Therefore, blast wave propagates
until the explosion energy is comparable to the internal energy in the sphere!
Some numbers:

– Timescale on which the bubble reaches Rmax is roughly the sound crossing
time
Rmax
ts ∼
cs
For ISM: T ∼ 104 K, ρ ∼ 10−21 kg m−3 , giving

Rmax ∼ few × 100 pc

tmax ∼ 10 Myr

– SN rate is about 10−7 Myr−1 pc−3 . So, over a duration tmax , can find 1
SN in ∼ 106 pc3 . But
4π 3
R > 106 pc3
3 max
so filling factor of SN driven bubbles is > 1. This would seem to suggest
that the entire ISM would be heated to SN to > 106 K. Not Observed!
We need to account for cooling and the finite height of the Galactic disk
(i.e. bubble “blow out”). After 105 yrs, when R ∼ 20 pc, cooling losses

57
F. S OUND WAVES , S UPERSONIC F LOWS AND S HOCK WAVES

become important and so the bubble grows more slowly than R ∝ t2/5 .
Simulations show that R ∝ t0.3 and Rmax ∼ 50 pc, giving filling factor
< 1. Thus, due to cooling, only a small fraction of E is deposited into
ISM.

58
 CHAPTER G

B ERNOULLI ’ S E QUATION AND

T RANSONIC F LOWS

G.1 Bernoulli’s Equation

Let’s start with the momentum equation:
∂u 1
+ (u · ∇)u = − ∇p − ∇Ψ. (∗) eq.g.1.1
∂t ρ

If the fluid is barotropic, then p = p(ρ) and so

∂ dp ∂p d dp 1 ∂p
Z Z
= =
∂x ρ ∂x dp ρ ρ ∂x
1 dp
Z 
⇒ ∇p = ∇ .
ρ ρ
Also, we have the vector identity
1 2
 
(u · ∇)u = ∇ u − u × (∇ × u). (Ex. Sheet 1)
2

Using these, the momentum equation (∗) becomes

∂u 1 2 dp
  Z 
+∇ u − u × w = −∇ +Ψ (∗∗) eq.g.1.2
∂t 2 ρ
where we have defined the vorticity:

w=∇×u Definition of Vorticity

Now, assume a steady flow ( ∂u/∂t = 0) and take the dot product of (∗∗)
with velocity u. Since we have u · (u × w) = 0 always, the result is

1 2 dp
 Z 
u·∇ u + +Ψ =0
2 ρ

59
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

This gives us Bernoulli’s Principle: For steady barotropic flows, the

quantity
1 2 dp
Z
H= u + +Ψ
2 ρ
is constant along a streamline. The quantity H is called Bernoulli’s constant.
If p = 0, H = constant is the statement that kinetic + potential energy is
constant along streamlines.
If p 6= 0, pressure differences accelerate or decelerate the flow as it flows
along the streamline.
Everyday examples of Bernoulli’s equation at work:

Example (the apocryphal Aircraft wing).

u1

u2

Figure G.1: Aircraft wing

u1 > u2 ⇒ p1 < p2 from H

⇒ pressure difference
⇒ lift force.

Of course, this cannot be the whole story of how aircraft wings work or else
inverted flight would be impossible!

Example (Shower curtain).

p1 p2

u1 u2 = 0

Figure G.2: Shower curtain

60
 G.2. Rotational and Irrotational Flows

Downward flow of air on inside of curtain induced by falling water

⇒ p1 < p2
⇒ curtain blows inwards.

G.2 Rotational and Irrotational Flows

An irrotational flow is one in which ∇ × u = 0 everywhere, i.e. the vorticity
w = 0 everywhere.
For a steady irrotational flow, (∗∗) gives that

∇H = 0

so, H = constant everywhere (not just along streamlines).

For a general (not necessarily irrotational or steady state) flow, we have

∂u
= −∇H + u × w
∂t

Take curl:

∂
(∇ × u) = −∇ × (∇H) +∇ × (u × w)
∂t | {z } | {z }
w ≡0

∂w
⇒ = ∇ × (u × w) Helmholtz’s Eqn.
∂t

From Helmholtz’s equation, we observe three results:

(i) If w = 0 initially, it will stay zero thereafter. We will see later that this
is no longer true once we include viscous terms.

(ii) The flux of vorticity through a surface S that moves with the fluid is a
constant, i.e.
D
Z
w · dS = 0
Dt S

Proof. We have

D ∂w D
Z Z Z
w · dS = · dS + w· dS
Dt S ∂t Dt
|S {z } | {z }
intrinsic changes in w change in S
caused by flow

61
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

S0
• vector area of element of this envelope
dS0
is
−δtu × dl
• vector area of whole volume is
Z
dS = 0
uδt
I
dS0 − dS = δt u × dl
I
D
⇒ dS = u × dl
S Dt
dS
dl C integral around
the edge of dS

Figure G.3: Change of area element with time

So,

D
Z Z I
w· dS = w · (u × dl)
S Dt ZS I∂dS
= w × u · dl
ZS ∂dS
= w × u · dl since “internal loops” cancel out
ZC
= ∇ × (w × u) · dS
S

D ∂w
Z Z  
⇒ w · dS = dS · − ∇ × (u × w)
Dt S S ∂t
| {z }
=0 from the
Helmholtz’s eqn.

D
Z
⇒ w · dS = 0
Dt S

i.e. flux of vorticity is conserved and moves with the fluid. This is Kelvin’s
vorticity theorem.

(iii) For an irrotational flow, the fact that ∇ × u = 0 everywhere implies

that there exists a potential function Φu such that

u = −∇Φu

62
 G.3. The De Laval Nozzle

If such a flow is also incompressible, then ∇ · u = 0 and so

∇2 Φu = 0

i.e. can reduce problem of finding velocity field to that of solving Laplace’s
equation.

G.3 The De Laval Nozzle

Consider steady flow in a tube with a variable cross-section A(z);

A(z)

Figure G.4: Tube with variable cross-section

For a steady flow, mass conservation gives

ρuA = constant Ṁ (mass flow per second)

⇒ ln ρ + ln u + ln A = ln Ṁ
1
⇒ ∇ρ + ∇ ln u + ∇ ln A = 0
ρ
1
⇒ ∇ρ = −∇ ln u − ∇ ln A
ρ

and the momentum equation (with no gravity) gives

1
u · ∇u = − ∇p
ρ

Let’s further assume a barotropic equation of state. Then

1 dp
u · ∇u = − ∇ρ
ρ dρ

So, putting these pieces together gives

 
u · ∇u = [∇ ln u + ∇ ln A] c2s c2s = dp/dρ (†) eq.g.3.1

63
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

If flow is also irrotational, we have

1 2 1
   
(u · ∇)u = ∇ u = u2 ∇ ln u2 = u2 ∇ ln u
2 2
and so, from (†) we have
u2 ∇ ln u = [∇ ln u + ∇ ln A] c2s
⇒ (u2 − c2s )∇ ln u = c2s ∇ ln A
This implies that an extremum of A(z) must correspond to either
(a) Minimum or maximum in u, or
(b) u = cs .
Thus, we see that there is the potential for a transition from subsonic to
supersonic flow at a minimum or maximum of the cross-sectional area of the
tube.
To make progress, we applying Bernoulli’s equation
1 2 dp
Z
u + = H, constant [no gravity, steady, irrotational]
2 ρ
and examine the two standard barotropic cases.

Case I: Isothermal EoS

R∗ ρT
p= , T = const.
µ
dp R∗ T dρ
Z Z
⇒ =
ρ µ ρ
R∗ T
= ln ρ
µ
= c2s ln ρ
Suppose that we have a minimum or maximum in A(z) that allows a flow
to have a sonic transition. Let A = Am at this location.
Then Bernoulli gives
1 2 1
u + c2s ln ρ = c2s + c2s ln ρ
2 2 A=Am
" !#
ρ|A=Am
⇒ u =2
1 + 2 ln
c2s
ρ
uA
  
= cs 1 + 2 ln
2
cs Am
where this last step has used mass conservation, i.e. ρuA = constant. Thus,
given A(z) we can determine u(z) and ρ(z), i.e. structure of the flow everywhere
subject to given Ṁ and cs .

64
 G.3. The De Laval Nozzle

Case II: Polytropic EoS

p = Kρ1+1/n
Let’s examine the case where the sonic transition occurs at A = Am . But
now we do not know the sound speed cs since cs = cs (ρ) and ρ varies. We
need to solve for
n+1
c2s = Kρ1/n
n
Now,
dp dp dρ
Z Z
=
ρ dρ ρ
n + 1 1/n dρ
Z
= K ρ
n ρ
n+1
Z
=K ρ1/n−1 dρ
n
n + 1 1/n
=K nρ
n
= nc2s
Mass conservation:

ρuA = ρ cs Am = Ṁ
Am Am
n+1
 1/2
⇒ ρ K ρ1/2n Am = Ṁ
Am n Am
n+1
 
⇒ ρ2+1/n
K A2m = Ṁ 2
Am n
 !2 n/(2n+1)
Ṁ n
⇒ ρ =
K(n + 1)

Am Am

Knowing ρ|Am , we can now determine cs and Am . Bernoulli gives:

1 2 dp
Z
u + = const.
2 ρ
!2
1 Ṁ 1 2
⇒ + K(n + 1)ρ1/n = c + K(n + 1) ρ1/n
2 Aρ 2 s Am Am
1 n+1
 
= K ρ1/n + K(n + 1) ρ1/n
2 n Am Am
1 n+1
  
= +n K ρ1/n
2 n Am

This is an implicit equation for the density structure through the flow.

65
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

General points of physical interpretation:

(u2 − c2s )∇ ln u = c2s ∇ ln A
– In subsonic regime u < cs
A decrease ⇒ ∇ ln u positive
⇒ u accelerates along streamline
e.g. rivers flowing through narrows;
– In supersonic regime u > cs
A increases ⇒ ∇ ln u positive
⇒ u accelerates along streamline

Gas becomes very compressible. A increases, u increases, ρ is greatly

reduced. Ṁ = Aρu constant.
So, a nozzle that gets progressively narrower, reaches a minimum, and then
widens again can be used to accelerate a flow from a subsonic to a supersonic
regime.

Subsonic Supersonic

Am
u = cs

Figure G.5: De Laval nozzle

Recall momentum equation:

1 dp
(u · ∇)u = − ∇ρ = −c2s ∇ ln ρ
ρ dρ
(u · ∇)u = u2 ∇ ln u
⇒ u2 ∇ ln u = −c2s ∇ ln ρ
– If u  cs , ∇ ln u  ∇ ln ρ, this implies accelerations are important,
pressure or density changes are small — almost incompressible;
– If u  cs , ∇ ln u  ∇ ln ρ, u ≈ constant, pressure changes do not lead to
much acceleration but there is change in ρ — compressible flow.

66
 G.4. Spherical Accretion and Winds

G.4 Spherical Accretion and Winds

We find flows with a mathematical structure when we consider steady-state and
spherically-symmetric accretion flows or winds in the gravitational potential of
a central body.
Consider the spherically-symmetric accretion of gas onto a star (described
as a point of mass). We will assume

– gas is at rest at ∞ (reservior);

– steady state flow;

– barotropic EoS.

Mass conservation gives

ρuA = constant Ṁ
⇒ 4πr2 ρu = Ṁ ,

where, for convenience, we define u to be inward pointing.

Momentum equation gives
du 1 dp GM
u =− − 2
dr ρ dr r
d ln u d ln ρ GM
⇒ u2 = −c2s − 2 , (∗) eq.g.4.1
dr dr r
assuming self-gravity of the accretion gas is negligible.
Now, steady flow must have
d  
ln Ṁ = 0
dr
d d d
⇒ ln ρ + ln u + ln r2 = 0
dr dr dr
d d 2
⇒ ln ρ = − ln u − .
dr dr r
Substitute into (∗) gives
d d 2 GM
 
u2 ln u = c2s ln u + −
dr dr r r2
Therefore
d 2c2 GM
 
(u2 − c2s ) ln u = s 1 − 2 .
dr r 2cs r
There is a critical point in the flow at
GM
r = rs = Sonic Point
2c2s

67
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

where u is either a minimum/maximum or there is a sonic transition. This

is called the sonic point, somewhat similar to De Laval nozzle, except no
boundaries/tubes!
Can gain insight into the general structure of such flows by plotting possible
solutions on the (r/rs , u) plane.

Spherical Wind
(with u → −u
in above)
cs

Spherical
Accretion
Settling
Flows

r/rs

Figure G.6: Plot in (r/rs , u) plane

Back to accretion problem: progress requires the EoS.

Case I: Isothermal EoS

Equation of state is:

R∗ ρT
p= , T = const.
µ

s
R∗ T
⇒ cs = = const.
µ

and we know
GM
rs =
2c2s

68
 G.4. Spherical Accretion and Winds

Need to use Bernoulli’s equation to constrain ρ and Ṁ .

1 dp
Z
H = u2 + +Ψ = const.
2 ρ
| {z }
c2s ln ρ
1 2 GM 1 GM
⇒ u + c2s ln ρ − = c2s + c2s ln ρs −
2 r 2 rs
1 2 GM 3
 
⇒ u + c2s ln ρ − = c2s ln ρs −
2 r 2
ρ 3 2GM
   
s
⇒ u2 = 2c2s ln − +
ρ 2 r

where ρs is the density at r = rs .

Now,

as r → 0, u2 → 2GM/r, i.e. free-fall;

as r → ∞ and u → 0, ρ = ρs e−3/2 , giving

ρs = ρ∞ e3/2

Thus, for a given ρ∞ , we know ρs and hence Ṁ .

Ṁ = 4πrs2 ρs cs
πG2 M 2 e3/2 ρ∞
⇒ Ṁ =
c3s

Note.

– Ṁ proportional to M 2 , more massive stars can accrete much more gas;

– Ṁ proportional to 1/c3s , accretion very sensitive to temperature; can

accrete more effectively from a colder medium.

Case II: Polytropic EoS

Equation of state is:

dp
Z
p = Kρ1+1/n ; = K(n + 1)ρ1/n = nc2s
ρ

Bernoulli gives

1 2 GM 1 GM
u + (n + 1)Kρ1/n − = c2s + nc2s −
2 r 2 rs

69
G. B ERNOULLI ’ S E QUATION AND T RANSONIC F LOWS

with rs = GM/(2c2s ). Using the mass accretion rate Ṁ = 4πrs2 ρs cs we can

then write
!1/2
Ṁ GM
rs = =
4πρs cs 2c2s
4πρs
2/3  1/3
GM

⇒ cs =
2 Ṁ
Combine this with
n+1
c2s = Kρ1/n
s
n
to get
n+1 4πρs
4/3  2/3
GM
  
Kρ1/n =
n s
2 Ṁ
4π
4/3  2/3
GM n

⇒ ρ1/n−2/3 = ρs(3−2n)/3n =
s
2 Ṁ (n + 1)K
4π
4n/(3−2n)  2n/(3−2n)  3n/(3−2n)
GM n

⇒ ρs = .
2 Ṁ K(n + 1)
Back to Bernoulli:
1 2 GM 3
 
u + (n + 1)Kρ1/n
− = cs n −
2
2 r 2
!2
1 Ṁ 3 GM
 
⇒ + (n + 1)Kρ 1/n
= c2s n− +
2 4πr2 ρ 2 r
As r → ∞, u → 0, we have
   n
3
c2s n − 2
ρ∞ = 
(n + 1)K


n+1
c2s,∞ = Kρ1/n
∞
n
So, finally,
Ṁ = 4πrs2 ρs cs
!n
4πG2 M 2 n
= · cs ρ∞
4c4s n− 3
2
!n−3/2
πG2 M 2 n
= ρ∞ 3
c3s,∞ n− 2

Therefore,
!n−3/2
π(GM )2 ρ∞ n
Ṁ = 3
c3s,∞ n− 2

70
 G.4. Spherical Accretion and Winds

Same functional form as isothermal case, but now an additional coefficient

related to polytropic index.
This is known as Bondi Accretion.
The generalisation to the case of a star accreting from a medium that it is
moving through is called Bondi-Hoyle-Lyttleton Accretion. The result is

(GM )2 ρ∞
Ṁ ∼
(c2∞ + v∞
2 )3/2

where v∞ is the velocity of gas relative to star at ∞.

71
 CHAPTER H

F LUID I NSTABILITIES

Consider a fluid in a steady state ( ∂/∂t = 0). Thus it is in a state of

equilibrium.

– If a small perturbation of this configuration grows with time, the

configuration is unstable with respect to those perturbations;

– If a small perturbation decays with time or just oscillates around the

equilibrium configuration, the configuration is stable with respect to
those perturbations.

An awful lot of interesting astrophysics is due to the action of fluid

instabilities!

Examples.

– Convection in stars;

– Multiphase nature of the ISM;

– Mixing of fluids that have relative motion;

– Turbulence in accretion disks;

– Formation of stars and galaxies.

In this chapter, we discuss some of the most important instabilities.

H.1 Convective Instability

This concerns the stability of a hydrostatic equilibrium. We can gain insight
without doing a full perturbation analysis.
Consider following system:

– Ideal gas in hydrostatic equilibrium;

73
H. F LUID I NSTABILITIES

– Uniform gravitational field in −ẑ direction.

Now perturb a fluid element upwards, away from its equilibrium point.

p0 , ρ∗ p0 , ρ0

⇑ surrounding
medium

p, ρ p, ρ

Figure H.1: Perturbing a fluid element upwards

We assume that any pressure imbalances are quickly removed by acoustic

waves, but that heat exchange takes longer. This implies displaced element
evolves adiabatically with a pressure p0 equal to pressure at new location of
atmosphere.
Since we assume heat transfer is slow, initially perturbations will change
adiabatically. Stability depends on new value of density.
ρ∗ < ρ0 ⇒ perturbed element buoyant
⇒ system unstable;
∗ 0
ρ >ρ ⇒ perturbed element sinks back
⇒ system stable.
For adiabatic change,
p = Kργ
)  0 1/γ
∗ p
⇒ ρ =ρ .
p0 = Kρ∗ γ p
To first order,
dp
p0 = p + δz
dz

!1/γ
∗ p+ dp
dz δz
⇒ ρ =ρ
p
1 dp
 1/γ
=ρ 1+ δz
p dz
ρ dp
≈ρ+ δz
pγ dz

74
 H.1. Convective Instability

In surrounding medium,
dρ
ρ0 = ρ + δz
dz
and the system is unstable if ρ∗ < ρ0 . So instability needs
ρ dp dρ
ρ+ δz < ρ + δz
pγ dz dz
ρ dp dρ
⇒ <
pγ dz dz
d d
⇒ ln p < γ ln ρ
dz dz
d
ln pρ−γ < 0


⇒
dz
dK
⇒ <0 (instability)
dz
So, the system is unstable if the entropy of the atmosphere decreases
with increasing height. This can also be related to temperature and pressure
gradients.
dK d
<0 ⇒ ln K < 0
dz dz
But
K = pρ−γ = (const.)p1−γ T γ (p = R∗ ρT /µ)
so,
d d d
ln K = (1 − γ) ln p + γ ln T < 0
dz  dz dz
dT 1 T dp

⇒ < 1− (instability)
dz γ p dz
Hence, we have the Schwarzschild stability criterion which reads

dT 1 T dp
 
> 1−
dz γ p dz

Since hydrostatic equilibrium requires dp/dz < 0, we see that (since γ > 1)

– Always stable to convection if dT /dz > 0;

– Otherwise, can tolerate a negative temperature gradient provided

dT 1 T dp
 
< 1−
dz γ p dz
So convective instability develops when T declines too steeply with
increasing height.

75
H. F LUID I NSTABILITIES

Examples (Convectively unstable systems).

– Outer regions of low mass stars;

– Cores of high mass stars.

For stable configurations, we can example the dynamics of atmosphere:

equation of motion is
d2
ρ∗ δz = −g ρ∗ − ρ0


dt 2

d
small 2
ρ dT 1 ρ dp
   
⇒ (ρ + δρ) δz = −g − 1− δz
dt2 T dz γ p dz


d2 g dT 1 T dp
   
⇒ δz = − − 1 − δz
dt2 T dz γ p dz
So, it is simple harmonic motion with angular frequency N where
g dT 1 T dp
   
N2 = − 1− Brunt-Väisälä Frequency
T dz γ p dz
These oscillations are internal gravity waves.

H.2 Jeans Instability

This concerns the stability of a self-gravitating fluid against gravitational
collapse.
Consider following system:

– Uniform medium initially static;

– Barotropic EoS;

– Gravitational field generated by the medium itself.

So equilibrium is

p = p0 , const.
ρ = ρ0 , const.
u=0

and governing equations are

∂ρ
+ ∇ · (ρu) = 0
∂t
∂u 1
+ (u · ∇)u = − ∇p − ∇Ψ
∂t ρ
∇2 Ψ = 4πGρ

76
 H.2. Jeans Instability

Introduce a perturbation:

p = p0 + ∆p
ρ = ρ0 + ∆ρ
u = ∆u
Ψ = Ψ0 + ∆Ψ

Note. There is an inconsistency between the assumption ρ0 = constant > 0

and the assumption Ψ0 = const. We proceed anyways — this is the Jeans
swindle (1902). This is closely tied to the fact that it is impossible to
construct a model of a static infinite Universe. A more complete analysis of
perturbations against a background of a (relativistic) homogenous expanding
Universe recovers the same local instability as that found by Jeans, hence
justifying the swindle.
Linearized equations are:
∂∆ρ
+ ρ0 ∇ · (∆u) = 0 O
1 eq.h.2.1
∂t
∂∆u dp 1 ∇(∆ρ)
=− ∇(∆ρ) − ∇(∆Ψ) = −c2s − ∇(∆Ψ) O
2 eq.h.2.2
∂t dρ ρ0 ρ0
∇2 (∆Ψ) = 4πG∆ρ O
3 eq.h.2.3

Look for plane wave solutions

∆ρ = ρ1 ei(k·x−ωt)
∆Ψ = Ψ1 ei(k·x−ωt)
∆u = u1 ei(k·x−ωt) .

Substitution into the linear equations gives

O
1 ⇒ − ωρ1 + ρ0 k · u1 = 0 O
4 eq.h.2.4

O
2 ⇒ − ρ0 ωu1 = −c2s kρ1 − ρ0 kΨ1 O
5 eq.h.2.5

O
3 ⇒ − k Ψ1 = 4πGρ1 .
2
O
6 eq.h.2.6

Eliminating u1 and Ψ1 from these

4 +O
O 5 ⇒ ρ1 ω 2 = k 2 (ρ1 c2s + ρ0 Ψ1 )
= k 2 ρ1 c2s − 4πGρ0 ρ1 from O
6

4πGρ0
 
⇒ ω 2 = c2s k 2 − .
c2s
Introduce the Jeans wavenumber kJ2 = 4πGρ0 /c2s so we have the dispersion
relation
ω 2 = c2s (k 2 − kJ2 )

77
H. F LUID I NSTABILITIES

Notes:

– For k  kJ , we have normal sound waves ω 2 = c2s k 2 .

– For k & kJ , we have modified sound waves. Gravity leads to dispersion

of the wave and a slower group velocity.

– For k < kJ , ω is purely imaginary (for k ∈ R), giving

ω = iω̃, ω̃ ∈ R

and
ei(k·x−ωt) = eω̃t eik·x
leading to exponentially growing solution: Gravitational Instability.

The maximum stable wavelength is

s
2π πc2s
λJ = = Jeans Length
kJ Gρ0

The associated mass is

MJ ∼ ρ0 λ3J Jeans Mass

These are central concepts in the theory of

– Star formation (instability of giant molecular clouds);

– Cosmological structure formation (instability of the homogeneous

primordial gas).

78
 H.3. Rayleigh-Taylor and Kelvin-Helmholtz Instability

H.3 Rayleigh-Taylor and Kelvin-Helmholtz Instability

This concerns the stability of an interface with a discontinuous change in
tangential velocity and/or density.

z z

ρ0 , u0

ξ
x perturbation x
perturbed position of
interface at ξ(x, t)
ρ, u
g

Figure H.2: Perturbation of interface of discontinuity

For convenience, let’s assume:

– Constant gravity, ideal fluid;

– Pressure continuous across the interface;

– Incompressible flow ∇ · u = 0;

– Irrotational flow ∇ × u = 0 ⇒ u = −∇Φ;

– 2D problem (symmetry direction into the page of Fig. H.2).

The momentum equation (for either upper or lower fluid) is

∂u 1
+ u · ∇u = − ∇p + g
∂t ρ
∂Φ 1 2 p
   
⇒ −∇ +∇ u =− ∇ −∇Ψ
∂t 2 ρ
| {z }
since ρ=const.
within each fluid
∂Φ 1 2 p
 
⇒ ∇ − + u + +Ψ =0
∂t 2 ρ
∂Φ 1 2 p
⇒ − + u + + Ψ = F (t) (∗) eq.h.3.1
∂t 2 ρ

79
H. F LUID I NSTABILITIES

where F (t) is a function that is constant in space but not in time.

Now consider a perturbation at the interface of these two fluids. Let us
study the evolution of the perturbed position of the interface ξ(x, t).
The velocity potential u = −∇Φ, so if the unperturbed velocities in the
fluids are U and U 0 we have

Φlow = −U x + φ
Φup = −U 0 x + φ0

⇒ ∇2 φ = ∇2 φ0 = 0 (since ∇ · u = 0) O
1 eq.h.3.2

φ and φ0 are sourced by displacements of the interface. Consider an element

of the lower fluid that is at the interface. Then
Dξ
uz =
Dt
giving
∂φ ∂ξ ∂ξ 

− = +U 
∂z ∂t ∂x

0 to first order O
2 eq.h.3.3
∂φ ∂ξ ∂ξ 
− = + U0  
∂z ∂t ∂x
Now look for plane wave solutions

ξ = Aei(kx−ωt)
φ = Cei(kx−ωt)+kz z
0
φ0 = C 0 ei(kx−ωt)+kz z

where extra terms on the exponents kz z and kz0 z are there to seek solutions
where perturbed potential decays at large |z|.
But we know that

∇2 φ = 0 ⇒ − k 2 + kz2 = 0
⇒ kz = |k|

so φ → 0 as z → −∞.
2
∇2 φ0 = 0 ⇒ − k 2 + kz0 = 0
⇒ kz0 = −|k|

since φ0 → 0 as z → ∞.
For now, let’s stipulate k > 0. So

φ = Cei(kx−ωt)+kz
φ0 = C 0 ei(kx−ωt)−kz

80
 H.3. Rayleigh-Taylor and Kelvin-Helmholtz Instability

From O
2 , we have

−kC = −iωA + iU kA = i(kU − ω)A O

3 eq.h.3.4
0 0
kC = i(kU − ω)A O
4 eq.h.3.5

We need one more equation if we’re to solve for A, C, C 0 . We get that from
pressure balance across the interface.

∂φ 1 2
 
p = −ρ − + u + gξ + ρF (t)
∂t 2
∂φ 0 1 02
 
0 0
p = −ρ − + u + gξ + ρ0 F 0 (t)
∂t 2

and equality at z = 0:
! !
∂φ u2 0 ∂φ0 u02
ρ − + + gξ =ρ − + + gξ + K(t) O
5 eq.h.3.6
∂t 2 ∂t 2

where
K ≡ ρF (t) − ρ0 F 0 (t)
The perturbation vanishes for z → ±∞ at all times, so we can look at
equation (∗) for each fluid in the limit |z| → ∞, taking limit carefully so that
Ψ terms cancel, to get
1 1
ρF (t) − ρ0 F 0 (t) = U 2 ρ − U 02 ρ0
|2 {z 2 }
conditions at ∞
and so a constant

Therefore, K(t) is actually a constant.

Next in our attempt to use O 5 to match across boundary, we need to
determine u and u0 . Now

u = −∇Φ = −∇(−U x + φ) = U x̂ − ∇φ
∂φ
⇒ u2 = U 2 − 2U (dropping 2nd order terms)
∂x
and similarly
∂φ0
u02 = U 02 − 2U 0
∂x
So, O
5 reads

∂φ 1 2 ∂φ0 1 02 1 1 02
!
∂φ ∂φ0
 
+ gξ = ρ0 − + U − U0
0
ρ − + U −U + gξ + U 2 ρ − U  ρ
∂t 2 ∂x ∂t 2 ∂x |2 2
{z }
K

81
H. F LUID I NSTABILITIES

∂φ ∂φ ∂φ0 ∂φ0
   
⇒ ρ − −U + gξ = ρ0 − − U0 + gξ
∂t ∂x ∂t ∂x
⇒ ρiωC − ρU ikC + ρgA = ρ0 iωC 0 − ρ0 U 0 ikC 0 + ρ0 gA
⇒ ρ(kU − ω)C + iρgA = ρ0 (kU 0 − ω)C 0 + iρ0 gA

Now eliminate C and C 0 from O

3 and O
4 to give

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

This is the dispersion relation for our system. Let’s now look at some specific
applications.

(A) Surface gravity waves: two fluids at rest initially with ρ0 < ρ (i.e. denser
fluid on bottom). The dispersion relation gives

ω 2 (ρ + ρ0 ) = kg(ρ − ρ0 )
g(ρ − ρ0 )
⇒ ω2 = k
ρ + ρ0

So, for k ∈ R, we have that ω ∈ R and hence system displays

oscillations/waves. Phase speed is
s
ω g ρ − ρ0
=± = f (k)
k k ρ + ρ0 | {z }
waves are
dispersive

If ρ0  ρ, then ω/k = ± g/k.

p

Example. Surface waves on ocean.

(B) Static stratified fluid: two fluids at rest initially with ρ0 > ρ (i.e. denser
fluid on top). Then
g(ρ − ρ0 )
ω2 = k
ρ + ρ0

So, for k ∈ R we have ω 2 < 0 and so ω is purely imaginary.

s
ω g ρ0 − ρ
= ±i
k k ρ + ρ0

The positive root of this gives us exponentially growing solutions. This

is the Rayleigh-Taylor Instability.

82
 H.4. Thermal Instability

(C) Fluids in motion: two fluids with ρ > ρ0 (so stable to Rayleigh-Taylor)
but different velocities non-zero U and U 0 . Take full dispersion relation:

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

divide by k 2 and solve the quadratic in ω/k,

s
ω ρU + ρ0 U 0 g ρ − ρ0 ρρ0 (U − U 0 )2
⇒ = ± −
k ρ + ρ0 k ρ + ρ0 (ρ + ρ0 )2

There is instability if

g ρ − ρ0 ρρ0 (U − U 0 )2
− <0 (instability)
k ρ + ρ0 (ρ + ρ0 )2

If g = 0, then any relative motion gives instability, i.e. Kelvin-Helmholtz

Instability;
If g 6= 0, then unstable modes are those with

(ρ2 − ρ02 )g
k>
ρρ0 (U − U 0 )2

i.e. gravity is stabilising influence.

H.4 Thermal Instability

This concerns the stability of a medium in thermal equilibrium (heating =
cooling) to perturbations in temperature. Consider the following system:

– No gravitational field;

– Static thermal equilibrium

u0 = 0, Q̇0 = 0, ∇p0 = 0, ∇ρ0 = 0 where p = Kργ

| {z }
∇K0 =0

Let’s start by deriving an alternative form of the energy equation that

involves the entropy-like variable K; this will be well suited to problems of
thermal instability.

p = Kργ ⇒ dp = ργ dK + Kγργ−1 dρ
γp
= ργ dK + dρ O
1 eq.h.4.1
ρ

83
H. F LUID I NSTABILITIES

R∗ R∗ R∗
p= ρT ⇒ dp = T dρ + ρ dT
µ µ µ
p R∗
= dρ + ρ dT O
2 eq.h.4.2
ρ µ
Equate O
1 and O
2 to give

p p R∗
ργ dK + γ dρ = dρ + ρ dT
ρ ρ µ
p R∗
⇒ ργ dK = (1 − γ) dρ + ρ dT
ρ µ
p R∗
 
⇒ dK = ρ1−γ (1 − γ) 2 dρ + dT
ρ µ(1 − γ)
| {z }
−đQ

First law of thermodynamics:

dE
đQ = p dV + dT , (unit mass)
dT
and so

đQ = p d(1/ρ) + CV dT
p R∗
= − 2 dρ − dT . since we have (γ − 1)CV = R∗ /µ
ρ µ(1 − γ)
Then we have

dK = −(1 − γ)ρ1−γ đQ for fluid element

Turn this into Lagrangian energy equation by noting that Q̇ = −đQ/dt,

DK
⇒ = −(γ − 1)ρ1−γ Q̇
Dt
1 DK D ρ
⇒ ≡ (ln K) = −(γ − 1) Q̇
K Dt Dt p

1 DK ρQ̇
= −(γ − 1) Entropy Form of Energy Eqn O
3 eq.h.4.3
K Dt p

This joins our usual continuity and momentum equations

∂ρ
+ ∇ · (ρu) = 0 O
4 eq.h.4.4
∂t
∂u
ρ + ρu · ∇u = −∇p O
5 eq.h.4.5
∂t

84
 H.4. Thermal Instability

Now we look at thermal stability. Consider small perturbations to the

equilibrium
ρ → ρ0 + ∆ρ
p → p0 + ∆p
u → ∆u
K → K0 + ∆K.
Linearize the equations:
∂∆ρ
O
4 ⇒ + ρ0 ∇ · (∆u) = 0 O
6 eq.h.4.6
∂t
∂∆u
O
5 ⇒ ρ0 = −∇(∆p) O
7 eq.h.4.7
∂t
∂∆K γ−1
O
3 ⇒ = − γ−1 ∆Q̇,
∂t ρ0
where we can write
∂ Q̇ ∂ Q̇
∆Q̇ = ∆p + ∆ρ
∂p ρ
∂ρ p
so that
∂∆K
= −A∗ ∆p − B ∗ ∆ρ O
8 eq.h.4.8
∂t
with
γ − 1 ∂ Q̇ γ − 1 ∂ Q̇
A∗ = , B∗ =
ρ0γ−1 ∂p ρ ργ−1
0
∂ρ p
We also have
p0
p = Kργ ⇒ ∆p = ργ0 ∆K + γ ∆ρ O
9 eq.h.4.9
ρ0
We seek solutions of the form
∆p = p1 eik·x+qt
∆ρ = ρ1 eik·x+qt
∆u = u1 eik·x+qt
∆K = K1 eik·x+qt
so, instability if Re(q) > 0. Substituting into linearized equations gives

O
6 ⇒ qρ1 + ρ0 ik · u1 = 0
O
7 ⇒ qρ0 u1 = −ikp1
O
8 ⇒ qK1 = −A∗ p1 − B ∗ ρ1
γp0
O
9 ⇒ p1 = ργ0 K1 + ρ1
ρ0

85
H. F LUID I NSTABILITIES

We can combine these to obtain the dispersion relation:

p0 1
!
A∗ q B ∗ q2
− = − +γ
k 2 q k 2 ρ0 ργ0
p0
⇒ q 3 + A∗ ργ0 q 2 + k 2 γ q − B ∗ k 2 ργ0 = 0
ρ0
| {z }
cubic in q, call E(q)

This has at least one real root — system is unstable if that real root is
positive, q > 0.
Now E(∞) = ∞, E(0) = −B ∗ k 2 ργ0 . So the system is unstable if B ∗ > 0.

γ − 1 ∂ Q̇
∴ B∗ = >0 (condition for instability)
ργ−1
0
∂ρ p

∂ Q̇
⇒   >0
µp
∂ R∗ T p
T2 ∂ Q̇
⇒ − >0
p ∂T p

∂ Q̇
∴ unstable if <0 Field Criterion
∂T p

The system is always unstable if it’s Field unstable (named after George
Field who wrote the classice paper on thermal instability in 1965).
However,
. even a Field stable system can be unstable if A∗ < 0 ⇒
∂ Q̇ ∂T < 0. From the dispersion relation, we see that this can happen for
ρ
long wavelength modes, i.e. k small. Then

q 2 (q + A∗ ργ0 ) ≈ 0 ⇒ q ≈ −A∗ ργ0

Interpretation:

– Short wavelength perturbations are readily brought into pressure

equilibrium by the action of sound waves, therefore, thermal instability
proceeds at fixed pressure;

– Long wavelength perturbations: there is insufficient time for sound

waves to equalise pressure with surroundings, so they tend to develop at
constant density.

86
 H.4. Thermal Instability

Example. Let’s assume a specific form for Q̇

Q̇ = AρT α − H
Aµ α−1
= pT −H
R∗

∂ Q̇ Aµp α−2
⇒ = (α − 1) T
∂T p
R∗
.
This is Field unstable, ∂ Q̇ ∂T < 0 if α < 1.
p

Bremsstrahlung has α = 0.5 ⇒ Field unstable.

87
 CHAPTER I

V ISCOUS F LOWS

Thus far, we have been assuming that changes in the momentum of a fluid
element are due entirely to pressure forces (acting normal to the surface of the
element) or gravity (acting on the bulk).
This assumption is justified in the limit λ → 0, i.e. the the particles
composing the fluid have vanishingly small collisional mean-free-path.
For finite-λ, momentum can diffuse through the fluid. This brings us to a
discussion of viscosity.

I.1 Basics of Viscosity

In a viscous flow, momentum can be transferred if there are velocity differences
between fluid elements.
Continuity equation unchanged

∂ρ
+ ∂j (ρuj ) = 0
∂t

But momentum equation needs to be changed

∂
(ρui ) = −∂j σij + ρgi , gi = −∂i Ψ
∂t

with
0
σij = ρui uj + pδij − σij
|{z}
viscous
stress tensor

As we’ll see later, σij

0 is related to velocity gradients.

The connection between the viscous stress tensor and the microphysics (i.e.
the mean-free-path) is uncovered by considering a simple linear shear flow:

89
I. V ISCOUS F LOWS

∇v

Figure I.1: Linear shear flow

Microscopically, thermal/random motion of the particles can allow

momentum to “diffuse” across streamlines. Becomes more important as gas
gets less collisional.
Let’s analyse the microscopic behaviour: assume the typical (thermal)
velocity in j-direction is uj . So the momentum flux associated with this is

ρui uj
| {z }
i-cpt of momentum
carried in j-direction

Typical thermal velocity is ∼ kT /m. So, flux of ith-component of momentum

p

in the upward jth direction is

s
kT
ρui α , α∼1
m

For the element on the other side of the surface in the j-direction, the
corresponding momentum flux across surface is
s
kT
−ρu∗i α
m

where u∗i is i-velocity of that element. For a j-separation of δl we have

u∗i = ui + δl(∂j ui )

So, s
kT
net momentum flux = −ρ(∂j ui )δl
m
The relevant scale δl is the mean-free-path
1
δl ∼ λ = ,
nσ

90
 I.2. Navier-Stokes Equation

where σ is the collision cross section of the particles. If we treat the particles
as hard spheres of radius a (decent approximation for neutral gas), then

σ = πa2

So, s
m kT
net momentum flux = −ρ(∂j ui ) α
ρπa2 m
Putting this into momentum equation:

∂ α √
 
(ρui ) = −∂j (ρui uj + pδij ) + ∂j 2
mkT ∂j ui + ρgi
∂t |πa {z }
≡η, shear viscosity
√
A rigorous derivation shows that, for this hard-sphere model, α = 5 π/64.
Observations about the shear viscosity:

– η is independent of density (a denser gas has more particles to transport

the momentum but the mean-free-path is shorter);

– η increases with T ;

– Isothermal system has η = const.

For a fully ionized plasma (e.g. the ICM), the mean-free-path is set by
Coulomb collisions. Then
√
λ ∝ T 2, vth ∝ T ⇒ η ∝ T 5/2 .

Thus the viscosity has a stronger temperature dependence than found for
hard-sphere collisions.

I.2 Navier-Stokes Equation

The most general form of σij
0 which is

– Galilean invariant;

– Linear in velocity components;

– Isotropic

is given by
2
 
0
σij = η ∂j ui + ∂i uj − δij ∂k uk + ζδij ∂k uk
3
with η and ζ independent of velocity. This is a symmetric tensor which ensures
that there aren’t unbalanced torques on fluid elements.

91
I. V ISCOUS F LOWS

The term associated with η relates to momentum transfer in shear flows

(this term has zero trace).
The term associated with ζ relates to momentum transfer due to bulk
compression (∂k uk ≡ ∇ · u).
Putting this into the momentum equation gives
∂(ρui )
= − ∂j (ρui uj ) − ∂j pδij
∂t
2
   
+ ∂j η ∂j ui + ∂i uj − δij ∂k uk + ζδij ∂k uk + ρgi
3
which we can combine with continuity equation to give
∂ui
 
ρ + uj ∂j ui = − ∂j pδij
∂t
2
   
+ ∂j η ∂j ui + ∂i uj − δij ∂k uk + ζδij ∂k uk + ρgi
3
This is the general form of the Navier-Stokes equation.
Outside of shocks (ζ ≈ 0) and for isothermal fluid (η = constant) we have
∂u 1 η 1
 
+ u · ∇u = − ∇p − ∇Ψ + ∇ u + ∇(∇ · u)
2
∂t ρ ρ 3
|{z}
≡ν
kinematic
viscosity

I.3 Vorticity in Viscous Flows

Start with the Navier-Stokes equation with ζ = 0 and η = const.. Take the
curl of this, recalling definition of the vorticity w = ∇ × u:
∂w
+ ∇ × (u · ∇u)
∂t
1 η 1
  
= ∇ × − ∇p − ∇Ψ + ∇2 u + ∇(∇ · u)
ρ ρ 3
To tidy up LHS, use the vector identity and definition of vorticity:
1
u · ∇u = ∇u2 − u × (∇ × u)
2
⇒ ∇ × (u · ∇u) = −∇ × (u × w).
To tidy up RHS, assume a barotropic fluid, p = p(ρ):
1 1 1
   
⇒ ∇× ∇p = ∇ × ∇p + ∇ × ∇p
ρ ρ ρ
1
=− 2 ∇ρ × ∇p
ρ | {z }
=0 since surfaces of
constant ρ and p align

92
 I.4. Energy Dissipation in Incompressible Viscous Flows

Putting pieces together, we get

∂w η
 
= ∇ × (u × w) + ∇ × ∇2 u
∂t ρ

∂w η
⇒ = ∇ × (u × w) + ∇2 w
∂t ρ
where, in the last step, we have ignored gradients of ν = η/ρ (so strictly
assumed uniform density). So, vorticity is carried with flow but also diffuses
through flow due to action of vorticity.

I.4 Energy Dissipation in Incompressible Viscous

Flows
Viscosity leads to dissipation of kinetic energy into heat — an irreversible
process.
Let’s analyse this in the case of an incompressible flow so that we don’t
need to about about p dV work. Then the total kinetic energy is
1
Z
Ekin = ρu2 dV
2
Let’s consider the rate of change of Ekin with time

∂ 1 2 ∂
 
ρu = ui (ρui )
∂t 2 ∂t
0
= −ui ∂j (ρui uj ) − ui ∂j δij p + ui ∂j σij
0 0
= −ui ∂j (ρui uj ) − ui ∂i p + ∂j (ui σij ) − σij ∂j ui

Look at the first term of RHS:

 0
ui ∂j (ρui uj ) = ui uj ∂j (ρui ) + ρui
∂j
u
*
j

where last term is zero due to incompressible assumption,

∇·u=0 ⇒ ∂j uj = 0

Also note that

1 1 1
   
0
∂j ρuj · ui ui = ρui ui ∂j*+
u
j uj ∂j ρui ui
2 2 2
= uj ui ∂j (ρui )

1
 
∴ ui ∂j (ρui uj ) = ∂j ρuj · ui ui .
2

93
I. V ISCOUS F LOWS

So,

∂ 1 2 1
   
0 0
ρu = −∂j ρuj · ui ui − ∂i (ui p) + p∂i ui + ∂j (ui σij ) − σij ∂j ui
∂t 2 2
1 p
   
0 0
= −∂i ρui u2 + − uj σij − σij ∂j ui .
2 ρ

Integrating over the volume,

∂Ekin ∂ 1 2
Z
= ρu dV
∂t ∂t 2
Z V 
1 p
   Z
0 0
= − ∂i ρui u2 + − uj σij dV − σij ∂j ui dV
2 ρ
1 2 p
I     Z
0 0
=− ρu u + − u · σ · dS − σij ∂j ui dV
2 ρ | {z }
| {z }
Energy flux into volume including Rate of change of Ekin due
work done by viscous forces u·σ viscous dissipation

Let’s take the volume V to be the whole fluid so that the surface integral
is zero (e.g. v at bounding surface = 0, or v at ∞ = 0).
Then
∂Ekin
Z
0
= − σij ∂j ui dV
∂t
1
Z
0
=− σij (∂j ui + ∂i uj ) dV since σ 0 is symmetric
2

But σij
0 = η(∂ u + ∂ u ) for an incompressible fluid. So,
j i i j

∂Ekin 1
Z
=− η(∂j ui + ∂i uj )2 dV
∂t 2

We see that η needs to be positive in order for us to obey the 2nd law of
thermodynamics.

Note. The course book [pafd] together with its correction posted on the
course website have adopted a different sign convention, choosing to set
0
σij = ρui uj + pδij |{z}
+ σij
sign
change

2
 
0
σij = |{z}
− η ∂j ui + ∂i uj − δij ∂k uk − ζδij ∂k uk
3 |{z}
sign sign
change change

The alternative convention adopted in these notes is more standard.

94
 I.5. Viscous Flow through a Pipe

I.5 Viscous Flow through a Pipe

Now consider flow through a long pipe with a constant circular cross-section

z u

Figure I.2: A pipe

Assume

– Steady flow with uR = uφ = 0, uz 6= 0;

– Incompressible, uniform density fluid;

– Neglect gravity.

Navier-Stokes equation reads

0
∂u 1 1
 
7 :0 :0
 +
u ·{z
∇u
 = − ∇p + ν ∇2 u + ∇(∇·
u)
 
∂t ρ |3
| }

|{z} symmetry {z }
steady incompressible
state

1
⇒ ν∇2 u = ∇p
ρ
By symmetry we have
∂p ∂p
uR = uφ = 0 ⇒ = =0
∂R ∂φ
For the z-component

1 ∂p 1 ∂ ∂uz 1 ∆p
 
=ν R = −
ρ ∂z R ∂R ∂R ρ l
| {z } | {z } | {z }
function of function of R only constant, written
z only in terms of global
pressure gradient

Integrating gives
∆p 2
u=− R + a ln R + b
4ρνl
where a and b are constants. Apply boundary conditions:

95
I. V ISCOUS F LOWS

– At R = 0, u finite ⇒ a = 0;

– At R = R0 , u = 0 (no slip BC at wall).

∆p
⇒ u= (R2 − R2 )
4νρl 0
So velocity profile is parabolic.
The mass flux passing through an annular element 2πR dR is 2πRρu dR.
So, the total mass flow rate is

π ∆p 4
Z R0
Q= 2πρuR dR = R
0 8 νl 0

As η → 0, i.e. ν → 0, the flow rate → ∞ (or, in other words, an inviscid flow

can’t be in steady state in this pipe if there is a non-zero pressure gradient).
If ∆p increases sufficiently, it becomes unstable and irregular, giving
turbulent motions above critical speed.
The actual transition to turbulence is usually phrased in terms of the
Reynolds number
LV
Re ≡
ν
where L and V are “characteristic” length and velocity scales of the system.
We have turbulence when
Re > Recrit

I.6 Accretion Disks

Accretion disks are one of the most important applications of the N-S equation
in astrophysics.
Consider some gas flowing towards some central object (star, planet, black
hole...). Almost always, the gas will have significant angular momentum about
that object. If gravitationally bound to the object, the gas will settle into a
plane defined by the mean angular momentum vector. Residual motions in
other directions will be damped out on a free-fall timescale.
The gas will settle into circular orbits — the lowest energy configuration
for a given angular momentum. In the vertical direction (parallel to angular
momentum vector) the system will come into hydrostatic equilibrium with
internal vertical pressure gradient balancing the vertical component of gravity.
In the radial direction (along direction towards central object), the system will
achieve a state where the centripetal force is supplied by gravity and the radial
pressure gradient.
Very important special case is when the disk is “thin”, meaning that scale-
height in vertical direction h is much less than radius r. Then, radial pressure

96
 I.6. Accretion Disks

gradients are negligible and we can just write

s
GM GM
Ω2 R = 2 ⇒ Ω=
R R3
where Ω is the angular velocity of the flow around the central object. This
means that
dΩ
6= 0 ⇒ shear flow
dR
Viscosity will allow angular momentum to be transferred from the fast
moving inner regions to the more slowly moving outer regions. This means
the inner disk fluid elements lose angular momentum. We have
√
J = R2 Ω = GM R per unit mass

meaning that inner disk fluid elements drift inwards.

Ultimately, most of the mass flows inwards; a small amount of the mass
carries all of the angular momentum out to large radius.
Let’s set up a simple model for a geometrically-thin accretion disk. We
assume:
– Cylindrical polar coordinates (R, φ, z);

– Axisymmetric, ∂/∂φ = 0;

– Hydrostatic equilibrium in z-direction, uz = 0;

– uφ close to Keplerian velocity (i.e. thin disk);

– uR small and set by action of viscosity;

– Bulk viscosity zero.

Continuity equation in cylindrical polars is
0 0
∂ρ 1 ∂ 1 ∂ 
* ∂  *
+ (RρuR ) + (ρu
 φ) + (ρuz ) = 0
∂t R ∂R R∂φ |∂z {z }
  
| {z }
axisymmetry hydrostatic eqm.

∂ρ 1 ∂
⇒ + (RρuR ) = 0
∂t R ∂R
Define the surface density Σ by
Z ∞
Σ≡ ρ dz
−∞

Then, integrating above form of continuity equation over z we have

∂Σ 1 ∂
+ (RΣuR ) = 0 O
1 eq.i.6.1
∂t R ∂R

97
I. V ISCOUS F LOWS

We can get the same result by thinking of the disk as a set of rings/annuli:

R + ∆R
R

central
object

Figure I.3: Infinitesimal annulus element of disk

rate of change
flux into flux out
of mass in = +
annulus of annulus
the annulus

∂
⇒ (2πR∆RΣ) = 2πRΣ(R)uR (R) − 2π(R + ∆R)Σ(R + ∆R)uR (R + ∆R)
∂t
∂Σ (R + ∆R)Σ(R + ∆R)uR (R + ∆R) − RΣ(R)uR (R)
 
⇒ R =−
∂t ∆R
∂Σ ∂
⇒ R =− (RΣuR ) taking ∆R → 0
∂t ∂R
Now we look at conservation of angular momentum. Here we use the
ring/annulus approach (but we could also start with the Navier-Stokes equation
in cylindrical polars). Clearly,

ang. mtm. ang. mtm.

rate of change net torque on ring
= of mass − of mass +
of ang. mtm. (viscous, magnetic, etc.)
entering ring leaving ring

∂
⇒ (2πR∆RΣR2 Ω) = f (R) − f (R + ∆R) + G(R + ∆R) − G(R)
∂t
where
f (R) ≡ 2πRΣuR ΩR2
and G(R) is torque exerted by disk outside of radius R on the disk inside of
radius R:
∂Ω dΩ
G(R) = 2πRνΣR R = 2πR3 νΣ
∂R dR

∂ 1 ∂ 1 ∂ dΩ
 
∴ (RΣuφ ) = − (ΣR2 uφ uR ) + νΣR3 O
2 eq.i.6.2
∂t R ∂R R ∂R dR

98
 I.6. Accretion Disks

Now assume ∂uφ /∂t = 0 since gas is on Keplerian orbits. Then

∂Σ 1 ∂ 1 ∂ dΩ
 
Ruφ + (ΣR2 uφ uR ) = νΣR3
∂t R ∂R R ∂R dR
∂ 1 ∂  2 1 ∂ 3 dΩ
  
⇒ − uφ (RΣuR ) + ΣR uφ uR = νΣR
∂R R ∂R R ∂R dR
∂  uφ R ∂  ∂ 1 ∂ dΩ
 
⇒ −uφ(RΣuR ) +  (RΣuR ) + ΣuR (uφ R) = νΣR 3
∂R R ∂R ∂R R ∂R dR
∂ ∂ dΩ
 
⇒ RΣuR (R2 Ω) = νΣR3
∂R ∂R dR
 
∂ dΩ
∂R νΣR3 dR
⇒ uR = ∂
RΣ ∂R (R2 Ω)

Substitute this into O

pand specialise to the case of a Newtonian point source
1
gravitational field Ω = GM/R3 gives

∂Σ 3 ∂ ∂
 
= R1/2 (νΣR1/2 )
∂t R ∂R ∂R

So the surface density Σ(R, t) obeys a diffusion equation.

Note (Notes on accretion disks).

– In general, ν = ν(R, Σ) and so this is a non-linear diffusion equation for

Σ. It reduces to linear if ν = ν(R);

– Solutions of this equation show that an initial ring of matter will broaden
and then “slump” inwards towards the central object;

t0

t1
t3 t2

Figure I.4: The viscous evolution of a ring

99
I. V ISCOUS F LOWS

– Timescale for evolution is tν where

Σ 1 1 1/2 1 νΣ
∼ R νΣR1/2 ∼ 2
tν RR R R
R 2 R Ruφ
⇒ tν ∼ = = Ω−1 Re
ν uφ ν

where Re is the Reynolds number;

– If viscosity is due to particle thermal motions, typical values would

suggest that Re ∼ 1014 ! This means

tν  age of universe

There must be another source of effective viscosity: we now know that

there is an effective viscisty due to MHD turbulence driven by the
magnetorotational instability.

I.7 Steady-State, Geometrically-Thin Disks

Consider a steady state such that ∂/∂t = 0. Then

∂Σ 1 ∂
+ (RΣuR ) = 0
∂t R ∂R
ṁ
⇒ RΣuR = C1 = −
2π
where ṁ = −2πRΣuR is the steady state mass inflow rate. Now recall that
 
∂ dΩ
∂R νΣR3 dR
uR = ∂
RΣ ∂R (R2 Ω)
ṁ 3 ∂
⇒ − =− (νΣR1/2 ) for Ω2 = GM/R3
2πRΣ  ΣR1/2  ∂R
s
ṁ  R∗ 
⇒ νΣ = 1−
3π R

where we have taken as a boundary condition that νΣ = 0 at R = R∗ . This

amounts to saying that there are no viscous torques at R = R∗ . Physically R∗
can be:

– Surface of accreting star;

– Innermost circular orbit around a black hole.

100
 I.7. Steady-State, Geometrically-Thin Disks

Let’s now calculate the viscous dissipation neglecting p dV work and bulk
viscosity. Specifically, we will calculate the viscous dissipation per unit surface
area of the disk:

dV
Z
0
Fdiss = − σij ∂j ui
2πR dR dφ
1
Z
= η(∂j ui + ∂i uj )2 dz
2
dΩ 2
Z  
= ηR dz
2
dR
dΩ 2
 
= νΣR2
dR

Combining with our previous result for νΣ and recalling that Ω2 = GM/R3 ,
we have  s 
3GM ṁ  R∗ 
Fdiss = 1−
4πR3 R

Note (Notes on dissipation in disk).

– Total energy emitted is

Z ∞
GM ṁ
L= Fdiss 2πR dR =
R∗ 2R∗

Here, −GM/R∗ is gravitational potential at R∗ . Therefore, GM ṁ/R∗

is the rate of gravitational energy loss of flow. GM ṁ/2R∗ is radiated,
other half stays in flow as kinetic energy and is dissipated in boundary
layer on star, or carried into the black hole;

– At given location far from inner edge (R > R∗ ) we have

3GM ṁ
Fdiss ≈
4πR3

But an elementary estimate based on loss of gravitational potential energy

would give
1 ∂ GM ṁ 1 GM ṁ
 
Fdiss,est = · · =
|2πR
{zdR} ∂R |
R
{z }
2
|{z} 4πR3
area of change in grav. half converts
annulus potential of ṁ to radiation,
over annulus rest to kinetic

The extra factor of “3” in the correct formula is due to the transport of
energy through the disk by viscous torques.

101
I. V ISCOUS F LOWS

Radiation from Steady-State Thin Disks

If disk is optically-thick, all radiation is thermalised and it radiates locally as
a black body
 s 
3GM ṁ  R∗ 
2 4
· σSB Teff = 1−
|{z} 4πR3 R
top and
bottom of disk

  s 1/4
3GM ṁ  R∗ 
⇒ Teff = 1−
8πσR3 R

So, for R  R∗ , Teff ∝ R−3/4 .

The radiation emitted at a frequency f is

2h f3
Z ∞
Ff = 2πR dR
R∗ c2 ehf /kTeff − 1

So, we see that all of the observables from a steady-state disk are
independent of viscosity ν (provided it is large enough to provide necessary
angular momentum transport). In order to study/constrain ν, we need to
study non-steady disks.

102
 CHAPTER J

P LASMAS

Plasmas are fluids composed of charged particles, thus, electromagnetic fields

become important for both microphysics and large scale dynamics.

J.1 Magnetohydrodynamic (MHD) Equations

Consider a fully ionised hydrogen plasma, so contains only protons (number
density n+ , bulk velocity u+ ) and electrons (n− , u− ). Mass conservation for
each of the proton and electron fluids is
∂n+  
+ ∇ · n+ u + = 0
∂t
∂n−
+ ∇ · n− u − = 0


∂t
The mass density is ρ = m+ n+ + m− n− and the center-of-mass velocity is
m+ n+ u+ + m− n− u−
u=
m+ n+ + m− n−
So, we can combine these to give
∂ρ
+ ∇ · (ρu) = 0. Continuity Equation
∂t
The continuity equation is the same as found before.
The charge density is q = n+ e+ + n− e− and the current density is
j = e+ n+ u+ + e− n− u− . So, the above information also gives conservation of
charge equation:
∂q
+ ∇ · j = 0, Charge Conservation
∂t
When we formulate the momentum equation, we have to consider the
Lorentz force on each particle

F = e(E + v × B)

103
J. P LASMAS

So, for the two species of particles:

!
∂u+
+ +
m n + u · ∇u+ = e+ n+ (E + u+ × B) − f + ∇p
∂t
!
− − ∂u−
m n + u · ∇u− = e− n− (E + u− × B) − f − ∇p
∂t

where f ± is fraction of pressure gradient that accelerates the protons/electrons.

Summing these equations gives
∂u
 
ρ + u · ∇u = qE + j × B − ∇p
∂t
Ohm’s law lets us relate j to E and B:

j = σ(E + u × B)

where σ is the electrical conductivity. This equation is needed to close the set
of equations.
So, recapping the current set of equations
∂ρ
+ ∇ · (ρu) = 0 O
1 eq.j.1.1
∂t
∂q
+∇·j=0 O
2 eq.j.1.2
∂t
∂u

ρ + u · ∇u = qE + j × B − ∇p O
3 eq.j.1.3
∂t
j = σ(E + u × B) O
4 eq.j.1.4

We need to relate q, j, E and B — Maxwell’s equations!

∇·B=0
q
∇·E=
0
1 ∂E
∇ × B = µ0 j +
c2 ∂t
∂B
∇×E=−
∂t
where we note 0 µ0 = 1/c2 .

Simplifying MHD
Let us simplify in the case of a non-relativistic, highly conducting plasma.
Suppose fields are varying over length scales l and timescales τ . Then
(i)
∂B E l
∇×E=− ⇒ ∼ ∼u
∂t B τ

104
 J.1. Magnetohydrodynamic (MHD) Equations

(ii)
1 ∂E . 1 l 2 u2
 
|∇ × B| ∼ ∼ 2 1
c2 ∂t c2 τ c
for non-relativistic flows. Therefore, displacement current can be ignored
in non-relativistic MHD;

(iii) Look at two terms from O

3:

 2
|qE| qE 0 E/l E u
∼ ∼ ∼ u2 0 µ0 ∼ 1
|j × B| jB B/lµ0 B c

Therefore, charge neutrality is preserved to a high approximation due

to strength of electrostatic forces. If there is a charge imbalance, it will
oscillate with a characteristic frequency, the plasma frequency
s
ne2
ωp =
0 me

(iv) Neglecting displacement current in relevant Maxwell equation, we get

∇ × B = µ0 j = µ0 σ(E + u × B)

Take curl:

∇ × (∇ × B) = µ0 σ(∇ × E +∇ × (u × B))
| {z } | {z }
=−∇2 B−∇(∇·B)
0
:
 − ∂B
∂t

∂B 1
⇒ = ∇ × (u × B) + ∇2 B
∂t | {z } µ0 σ
advection of | {z }
the field dissipation of
by the flow the field
through the flow

If the fluid is a good conductor, i.e. σ is very large, then we can ignore
the diffusion term and we have an equation that is analogous to the
Helmholtz equation/Kelvin’s theorem:
∂B
= ∇ × (u × B)
∂t

We talk about the “freezing” of the magnetic flux into the plasma. In
the high σ limit we must also have

j = σ(E + u × B) is finite
⇒ E+u×B=0 as σ → ∞
⇒ E·B=0
i.e. E⊥B

105
J. P LASMAS

So, the full set of ideal MHD equations, i.e., equations describing a non-
relativistic, perfectly conducting, charge neutral plasma are:

∂ρ
+ ∇ · (ρu) = 0
∂t
∂u

ρ + u · ∇u = j × B − ∇p
∂t
E+u×B=0


 
∇·B=0

∂B


= ∇ × (u × B)


∇·E=0

∂t


⇒ ∇ × B = µ0 j
∇ × B = µ0 j 

∇·B=0

∇ × E = − ∂B/∂t 





p = Kργ 


J.2 The Dynamical Effects of Magnetic Fields

The magnetic force density appearing in the above ideal MHD equations is

1
fmag = j × B = (∇ × B) × B
µ0

So using vector identity this is

1
!
B2
 
fmag = −∇ + (B · ∇)B
µ0 2 | {z }
| {z } magnetic tension
magnetic pressure term (vanishes for
term with straight field lines)
pmag =B 2 /2µ0

Since there are new force terms in the momentum equation, this will change
the nature of the waves that are possible.

J.3 Waves in Plasmas

We can repeat the perturbation analysis that we conducted for sound waves
but now include the effects of a magnetic field. We will perturb about an
equilibrium consisting of a static (u = 0) plasma with uniform density ρ0 ,
uniform pressure p0 , and uniform magnetic field B0 .
We start by writing down the governing equations of ideal MHD, assuming

106
 J.3. Waves in Plasmas

a barotropic equation of state:

∂ρ
+ ∇ · (ρu) = 0
∂t
∂u 1

ρ + (u · ∇)u = (∇ × B) × B − ∇p (J.1)
∂t µ0
∂B
= ∇ × (u × B)
∂t
∇·B=0
p = p(ρ)
(J.2)

We now introduce perturbations and linearize the equations:

∂δρ
+ ρ0 ∇ · (δu) = 0
∂t
∂δu 1
ρ0 = (∇ × δB) × B0 − c2s ∇δρ
∂t µ0
∂δB
= ∇ × (δu × B0 ) = −B0 (∇ · δu) + (B0 · ∇)δu
∂t
∇ · δB = 0

We now adopt our usual plane wave form for the perturbations,

δρ = δρ1 ei(k·r−ωt)
δp = δp1 ei(k·r−ωt)
δu = δu1 ei(k·r−ωt)
δB = δB1 ei(k·r−ωt) .

The continuity equation becomes,

−iωδρ + iρ0 k · δu = 0
⇒ ωδρ = ρ0 k · δu.

The momentum equation becomes,

i
−iωρ0 δu = (k × δB) × B0 − ic2s δρk
µ0
1
⇒ ωρ0 δu = ([B0 · δB)k − (B0 · k)δB] + c2s δρ k.
µ0

Finally, the flux-freezing (induction) equation becomes,

−iωδB = −iB0 (k · δu) + i(B0 · k)δu

⇒ ωδB = B0 (k · δu) − (B0 · k)δu.

107
J. P LASMAS

The full dispersion relation for MHD waves is then derived from eliminating
the perturbation amplitudes from these expressions. Here, we are going to
gain insight for the physics by just focusing on some special cases.
Firstly, we consider the case of modes with wavevectors orthogonal to the
background magnetic field direction, k k B0 . The linearized equations then
become

ωδρ = ρ0 k · δu
1
ωρ0 δu = (B0 · δB)k + c2s δρ k
µ0
ωδB = B0 (k · δu)

We can immediately notice from the second of these relations that the
velocity perturbations are aligned with the wavevector, δu k k, i.e. these
are longitudinal modes. Eliminating δρ and δB from this set of equations in
favour of δu, we get

1 2
ω 2 ρ0 δu = B (k · δu)k + c2s ρ0 (k · δu)k
µ0 0

Take the dot product of this last equation with k and then cancel k · δu
throughout (since we know that this must be non-zero since modes are
longitudinal),

k 2 B02
ω 2 ρ0 = + c2s ρ0 k 2
µ0
!
B2
⇒ ω =
2
c2s + k2
µ0 ρ0
 
ω 2 = c2s + vA
2
k2 ,

where we have defined the Alfvén speed,

s
B02
vA = ,
µ0 ρ0

q
This describes dispersion-free longitudinal waves with a phase speed c2s + vA
2.

The restoring force comes from both the gas pressure and magnetic pressure
acting in phase. This is known as the fast magnetosonic wave.

108
 J.3. Waves in Plasmas

u k

Figure J.1: Fast magnetosonic wave

We now consider the case of modes with k k B0 . The linearized equations

become,

ωδρ = ρ0 k · δu
1
ωρ0 δu = [(B0 · δB)k − B0 k δB] + c2s δρ k
µ0
ωδB = B0 (k · δu) − B0 kδu.

Eliminating δρ and δB from this set of equations in favour of δu, we get

1
ω 2 ρ0 δu = (B 2 k 2 δu − (B0 · δu)B0 kk) + c2s (k · δu)k.
µ0 0

There are actually two distinct wave modes wrapped up in these expression, a
longitudinal mode and a transverse mode. To extract the longitudinal mode,
take the dot product with k

1
ω 2 ρ0 (k · δu) = (B 2 k 2 (k · δu) − (B0 · δu)B0 k 3 ) + c2s (k · δu)k 2
µ0 0

and cancel factor of k · δu to get

⇒ ω 2 = c2s k 2 .

These are simply sound waves, with the magnetic field not playing a role since
the velocity perturbations are directed along the magnetic field.

109
J. P LASMAS

u k

Figure J.2: Longitudinal wave with k k B

Return to the more general expression for the case k k B0 ;

1
ω 2 ρ0 δu = (B 2 k 2 δu − (B0 · δu)B0 kk) + c2s (k · δu)k.
µ0 0
Taking the cross product with k, we get

B02 2
ω2 = k = vA
2 2
k .
µ0 ρ0
This describes transverse waves with phase speed vA where the restoring force
is provided by magnetic tension. These are Alfvèn waves.

Figure J.3: Alfvén wave

J.4 Instabilities in Plasmas

The presence of magnetic forces can profoundly affect the nature of instabilities
in plasmas. For example, we can repeat the derivation of the Rayleigh-Taylor
instability including a magnetic field aligned with the interface.

110
 J.5. Magnetorotational Instability

ρ1 ρ1 > ρ2

interface
B

ρ2
g

Figure J.4: Configuration of fluid interface

We will not repeat the analysis here, but we find the new dispersion relation
is
ρ1 − ρ2 2 (k · B)2
ω 2 = −kg +
ρ1 + ρ2 µ0 ρ1 + ρ2
For sufficiently small wavelength (high |k|), the second term always wins,
giving stable oscillations (Alfvén waves in this case). The interpretation is that
magnetic tension forces tend to stabilise R-T modes.

J.5 Magnetorotational Instability

We end with a discussion of an MHD instability which is extremely important
for accretion disks. We examine the stability of a plasma which is in orbit
about a central object.

Ω(r) decreases
outwards.

Figure J.5: Shear flow in an accretion disk

To uncover the essence of the instability, we simplify as much as possible.

We conduct a “local analysis” meaning that we consider the dynamics in

111
J. P LASMAS

some small patch of the rotating flow at R = R0 , working in the comoving

reference frame of the equilibrium flow. We assume that the equilibrium flow
has an angular velocity about the central body Ω(R). We let our local frame
of reference rotate at Ω(R0 ) and set up a Cartesian coordinate system with
ẑ pointing “upwards” (meaning aligned with the angular velocity Ω) and
x̂ pointing outwards (i.e. away from the central body axis). Working in a
Lagrangian picture, the momentum equation is:
Du 1 1
= − ∇p + (∇ × B) × B + 2u × Ω + Ω × (Ω × r) − RΩ(R)2 R̂,
Dt ρ µ0 ρ
where the last term is an expression of gravity. Further simplifying, let us
assume that the flow is cold so that pressure forces are negligible (this assume
can be readily relaxed but will make the analysis more involved). Introducing
perturbations and assuming a plane-wave form, we have

D ∆u 1 dΩ2
− 2∆u × Ω = (B0 · ∇)∆B − ∆x R R̂
Dt µ0 ρ dR
i dΩ2
⇒ −iω∆u − 2∆u × Ω = B0 k ∆B − ∆x R R̂
µo ρ dR
The induction equation gives
∂∆B
= ∇ × (∆u × B0 ) = (B0 · ∇)∆u
∂t
⇒ −iω ∆B = ikB0 ∆u
kB0
⇒ ∆B = − ∆u,
ω
and we can easily relate ∆x and ∆ux ;
D ∆x
= ∆ux
Dt
⇒ −iω ∆x = ∆ux
i ∆ux
⇒ ∆x = .
ω
So eliminating in favour of ∆u in our perturbed form of the momentum
equation, we have

i kB0 i ∆ux dΩ2

−iω∆u − 2∆u × Ω = B0 k ∆u − R R̂
µo ρ ω ω dR

Writing this out in components and noting that B02 /ρ0 µ0 = vA

2 gives,

dΩ2
ω 2 ∆ux − 2i∆uy Ωω = (kvA )2 ∆ux + ∆ux
d(ln R)
ω 2 ∆uy + 2i∆ux Ωω = (kvA )2 ∆uy ,

112
 J.5. Magnetorotational Instability

or in matrix form.
dΩ2
! !
ω 2 − (kvA )2 − −2iωΩ ∆ux
d(ln R) = 0.
2iωΩ ω − (kvA )2
2 ∆uy

We obtain the dispersion relation by setting the determinant of the matrix

to zero. This gives
" #
dΩ2 h i
ω − (kvA ) −
2 2
ω 2 − (kvA )2 − 4Ω2 ω 2 = 0
d(ln R)

Writing as a quadratic in ω 2 gives our final form of the dispersion relation:

" # " #
dΩ2 dΩ2
4
ω −ω 2
4Ω − 2
+ 2(kvA )2 + (kvA )2 (kvA )2 + = 0.
d(ln R) d(ln R)

If we “turn off” magnetic forces by setting vA = 0, the dispersion relation

gives

dΩ2
ω 2 = 4Ω2 +
d(ln R)
1 d
= (R4 Ω2 ) ≡ κ2R
R3 dR
= Ω2 . (Keplerian)

For a Keplerian profile Ω2 = GM/R3 , or indeed any profile in which the

specific angular momentum R2 Ω increases with radius, this describes local
radial oscillations of the flow at the radial epicyclic frequency κR .
Now turn on magnetic forces, so vA > 0. There will be instability if
ω < 0. Considering the basic properties of the dispersion relation, viewed as
2

a quadratic in ω 2 , we see that there will be instability if

dΩ2
(kvA )2 + < 0.
d(ln R)

This is the magneto-rotational instability (MRI). For sufficiently weak magnetic

field or long wavelength (small k) modes, there will be instability if the angular
velocity decreases outwards,

dΩ2
<0 (instability).
dR
Magnetic tension will stabilize modes with k > kcrit where

dΩ2
(kcrit vA )2 = − (= 3Ω2 for Keplerian)
d(ln R)

113
J. P LASMAS

Specializing to the Keplerian case, we find that the fastest growing mode has
a growth rate
3
|ωmax | = Ω
2
and wavenumber given by

kmax vA ≈ Ω.

The instability has an interesting property — while the magnetic field is

essential for its existence, the maximum growth rate is independent of the
magnetic field. Formally, within ideal hydrodynamics, the instability exists as
B0 → 0 but not at B0 = 0. Of course, the wavelength of the mode with the
maximum growth rate kmax → ∞ as B0 → 0 and so in practice finite viscosity
or finite conductivity effects will kill the MRI for sufficiently small B0 .
The MRI is central to the modern theory of accretion disks. MRI drives the
turbulence that, as we have described previously, is essential for the transport
of angular momentum in an accretion disk.

114
 CHAPTER K

A PPENDIX : T YPOS AND

C ORRECTIONS

For convenience, here we collect together typos that were present in the set of
notes posted at the start of Lent-2021 term and which have been corrected in
the current set of notes.

• Section I.3 : Vorticity in Viscous Flows (page 92) : There is a missing u

in the second line following the statement “To tidy up the LHS...”. The
expression should read

⇒ ∇ × (u · ∇u) = −∇ × (u × w). (CORRECT),

and not

⇒ ∇ × (u · ∇) = −∇ × (u × w). (INCORRECT),

• Section I.6 : Accretion Disks (page 99) : There was a typo in the
derivation and final expression for the radial velocity, uR . The final
expression should read
 
∂ dΩ
∂R νΣR3 dR
uR = (CORRECT),
∂
RΣ ∂R (R2 Ω)

and not  
∂ dΩ
∂R νΣR3 dR
uR = (INCORRECT),
∂
RΣ ∂R (RΩ2 )
This same typo was repeated in the line above (i.e. the line immediately
prior to the statement of the final result):
∂ ∂ dΩ
 
RΣuR (R2 Ω) = νΣR3 (CORRECT)
∂R ∂R dR

115
K. A PPENDIX : T YPOS AND C ORRECTIONS

and not
∂ ∂ dΩ
 
RΣuR (RΩ2 ) = νΣR3 (INCORRECT)
∂R ∂R dR
The same typo was repeated when the expression for uR was restated
on page 100.

116
 Bibliography

PAFD [1] C.J. Clarke & R.F. Carswell, Principles of Astrophysical Fluid Dynamics
(CUP 2007)

117