Radiative Gas Dynamics - Barbara Ryden

ASTRONOMY 825: Radiative Gas Dynamics
Winter Quarter 2011 · Tu Th 1:30 - 2:48 pm · McPherson 4054
Instructor: Professor Barbara Ryden

Office: 4035 McPherson Lab
Telephone: 292-4562
Email: [email protected]
Text: Radiative Gas Dynamics, by Barbara Ryden – a set of lecture notes

that I will be handing out in class.
Class website: www.astronomy.ohio-state.edu/∼ryden/ast825/

The website will contain PDF lecture notes, the course syllabus, and the
assigned problem sets.
Grading policy: Problem sets will be assigned on alternate Thursdays, starting

on January 13, and will be due exactly one week later. The problem sets will
count for 2/3 of your final grade. The final exam will be a takehome exam,
and will be handed out during finals week, at a time to be decided. The final
exam will count for 1/3 of your final grade.
Academic Misconduct: It is the responsibility of the Committee on Academic

Misconduct to investigate or establish procedures for the investigation of all
reported cases of student academic misconduct wherever committed; illus-
trated by, but not limited to, cases of plagiarism and cheating on exami-
nations. Instructors shall report all instances of alleged misconduct to the
committee (Faculty Rule 3335-5-487). For additional information, see the
Code of Student Conduct (studentaffairs.osu.edu/resources).
Disability Services: Students with disabilities that have been certified by the
Office for Disability Services will be accommodated; please inform the pro-
fessor as soon as possible of your needs. The Office for Disability Studies is
located in 150 Pomerene Hall, 1760 Neil Avenue, telephone 292-3307, TDD
292-0901, www.ods.ohio-state.edu.
Tentative Course Outline: Astronomy 825

• Chapter 1: Fundamentals (Jan 4, 6)
• Chapter 2: Viscosity & Heat Conduction (Jan 11)
• Chapter 3: Sound & Shocks (Jan 13, 18)
• Chapter 4: Radiative & MHD Shocks (Jan 20)
• Chapter 5: Spherical Blastwaves & Supernova Remnants (Jan 25)
• Chapter 6: Ionization Fronts & HII Regions (Jan 27, Feb 1)
• Chapter 7: Basic Turbulence (Feb 3)
• Chapter 8: Spherical Accretion (Feb 8, 10)
• Chapter 9: Accretion Disks for Beginners (Feb 15, 17)
• Chapter 10: Advanced Accretion Disks (Feb 22)
• Chapter 11: Solar Wind (Feb 24, Mar 1)
• Chapter 12: Winds from Hot & Cool Stars (Mar 3, 8)
• Chapter 13: Astrophysical Jets (Mar 10)
Due dates for problem sets = January 20, February 3, February 17, March 3
Astronomy 825:
Radiative Gas Dynamics
Winter 2011
Barbara Ryden
January 3, 2011
ii
Contents
Notes on the Notes v
1 Fundamentals of Gas Dynamics 1
2 Viscosity & Heat Conduction 13
3 Sound & Shocks 21
4 Radiative & MHD Shocks 35
5 Spherical Blastwaves & SNRs 45
6 Ionization Fronts 55
7 Basic Turbulence 67
8 Spherical Accretion 79
9 Accretion Disks for Beginners 89
10 Advanced Accretion Disks 101
11 The Solar Wind 111
12 Winds from Hot & Cool Stars 123
13 Astrophysical Jets 133
iii
iv CONTENTS
Notes on the Notes
These lecture notes are the result of my having taught Astronomy 825 several
times, starting in 1993, when my knowledge of radiative gas dynamics was
nearly non-existent. The most valuable resource I found while cramming to
teach the course was the two-volume Physics of Astrophysics set by Frank
Shu. Volume I of Shu’s work is titled Radiation and Volume II is titled Gas
Dynamics, and between them they cover more than you might want to know
about radiative gas dynamics. More recent works that I have consulted while
updating these notes are An Introduction to Astrophysical Fluid Dynamics
by Michael Thompson and Principles of Astrophysical Fluid Dynamics by C.
J. Clarke and R. F. Carswell.
In a way, these (not-a-textbook) notes are a companion to Rick Pogge’s
(not-a-textbook) notes for Astronomy 871, Physics of the Interstellar Medium.
I have attempted to make notation and units consistent between the two sets
of notes. Thus, I adopt cgs units, supplemented by angstroms (Å ), electron
volts (eV), and parsecs (pc).
v
vi NOTES ON THE NOTES
Figure 1: Turbulent fluids as drawn by Leonardo da Vinci.

Chapter 1
Fundamentals of Gas Dynamics
Gas dynamics is the study of continuous compressible fluids in motion.

Real gases are not perfectly continuous, since they are made of individual
particles: atoms, molecules, ions, and/or electrons. However, a gas can be
adequately approximated as a continuous fluid when
λ≪L (1.1)
where λ is the mean free path of the gas particles, and L is the characteristic
size of the system. The mean free path in a gas of neutral particles is
1
λ= (1.2)
nσ
where n is the number density of particles, and σ is the cross section for col-
lisions. A typical cross section for atoms or small molecules is σ ∼ 10−15 cm2 ,
or about 1 gigabarn. In air at room temperature, n ∼ 1019 cm−3 , and
hence λ ∼ 10−4 cm ∼ 1 µm. Thus, a volume of air that is larger than
several microns on a side can be treated as a continuous fluid. In the in-
terstellar medium (ISM), particle number densities are much lower than in
the Earth’s atmosphere. In a molecular cloud, n ∼ 1000 cm−3 , and hence
λ ∼ 1012 cm ∼ 0.07 AU. In the warm neutral medium, n ∼ 0.5 cm−3 , and
hence λ ∼ 2 × 1015 cm ∼ 100 AU ∼ 6 × 10−4 pc.
The situation is more complicated in a plasma. Consider a gas of fully
ionized hydrogen. The effective radius of interaction re for a free electron
can be found by setting the magnitude of its potential energy at a distance
re from an electron or proton equal to its thermal kinetic energy: is found
1
2 CHAPTER 1. FUNDAMENTALS OF GAS DYNAMICS
by setting
e2
∼ me ve2 . (1.3)
re
Since me ve2 ∼ kT , where T is the kinetic temperature of the free electrons,
we can write
e2 e2
re ∼ ∼ . (1.4)
me ve2 kT
The cross section is thus
πe4
σ ∼ πre2 ∼ , (1.5)
k2T 2
and the mean free path for an electron is
k2T 2
λ∼ . (1.6)
πe4 n
(Note: a more accurate calculation would contain the Coulomb logarithm
ln Λ, but this is good enough for an order-of-magnitude estimate.) In the hot
ionized interstellar medium, T ∼ 106 K and n ∼ 3 × 10−3 cm−3 . The mean
free path is thus λ ∼ 4 × 1019 cm ∼ 10 pc.
It’s useful at this point to provide a sketch of the different components of
the ISM. Molecular Clouds consist mainly of molecular gas (H2 , CO, etc.)
along with stubbornly solitary He atoms. Typical temperatures and number
densities are
T ∼ 15 K and n > ∼ 1000 cm .
−3
(1.7)
The Cold Neutral Medium consists mainly of neutral atomic gas (H I,
He I, etc.) Typical temperatures and number densities are
T ∼ 90 K and n ∼ 50 cm−3 . (1.8)
The Warm Neutral Medium consists mainly of neutral atomic gas, but
at a higher temperature and lower density than the Cold Neutral Medium.
Typical temperatures and number densities are
T ∼ 8000 K and n ∼ 0.5 cm−3 . (1.9)
The Warm Ionized Medium consists of partially ionized gas (H I, He I,

H II, He II, He III, etc.) Typical temperatures and number densities are
T ∼ 104 K and n ∼ 0.1 cm−3 . (1.10)

3
The Hot Ionized Medium consists of ionized gas at very high tempera-
tures. Typical temperatures and number densities are
T >
∼ 10 K and n <
6 −3
∼ 0.003 cm . (1.11)
Although the different components of the ISM are not in perfect pressure
equilibrium, it is notable that the pressures implied by the values of n and T
given above are all within an order of magnitude of each other, with a typical
value
P = nkT ∼ 3 × 10−13 dyne cm−2 ∼ 3 × 10−19 atm . (1.12)
The “atmosphere” (atm) is approximately equal to the pressure of the Earth’s
atmosphere at sea level. The ISM is obviously very low in pressure compared
to the air around us; thus, our intuitions about gas dynamics, which are
largely based on the behavior of air, may not always apply to the ISM.
The intergalactic medium (IGM) is typically at densities much lower than
even the Hot Ionized Medium of the ISM. The Warm/Hot Intergalactic
Medium (WHIM) has typical temperatures and number densities
T ∼ 106 K and n ∼ 5 × 10−6 cm−3 . (1.13)
The intracluster medium (ICM) in dense clusters of galaxies can have
T ∼ 108 K and n ∼ 10−3 cm−3 . (1.14)
Although the intracluster medium is thousands of times higher in pressure

than the WHIM, it is kept in hydrostatic equilibrium by the gravitational
pull of the dark matter in clusters.
Much of the ISM is flowing out of, or accreting onto, compact objects.
Examples of outflow are blast waves, winds, and jets. Examples of accre-
tion are cooling flows and accretion disks. Interesting things also happen
where one phase of the interstellar medium meets another. A shock front
is a boundary between low density, supersonic flow and high density, sub-
sonic flow. An ionization front is a boundary between ionized matter and
neutral matter.
In many astrophysical contexts, radiative effects influence the flow of gas.
Molecular clouds can absorb and emit radiation by rotational and vibra-
tional transitions. Sufficiently energetic radiation can photodissociate the
molecules; the dissociation energy of H2 is 4.48 eV, corresponding to a wave-
length of 2770Å. Neutral atomic gas can absorb and emit radiation through
bound-bound electronic transitions. Sufficiently energetic radiation can pho-

toionize the atoms; the ionization energy of H is 13.6 eV, corresponding to a
wavelength of 912Å. Ionized gas can emit radiation through bremsstrahlung
(free-free transitions) and radiative recombination. The electrons and ions
can also interact with photons via Compton scattering. In the presence of a
magnetic field, electrons can lose energy by synchrotron radiation.
Radiative transfer in the interstellar medium isn’t always pretty. To make
life simpler, let’s start by considering the dynamics of gas that is not radiat-
ing. A good place to begin is with the Boltzmann Equation.1 At a given
time t, any particle of the gas has a position ⃗x and a velocity ⃗v . So, if we live
in a Newtonian deterministic universe, we can compute the development of
a system if we know all the positions and velocities at a given time and if we
know the external forces (gravitational, electrostatic, magnetic, etc.) work-
ing on the particles. This fully deterministic approach, however, becomes a
major pain in the neck as the number of gas particles becomes large. If you
are dealing with many, many gas particles, a probabilistic approach becomes
much more practical and much more useful.
Let f (⃗x, ⃗v , t)d3 xd3 v be the probability of finding a gas particle within the
six-dimensional phase space volume d3 xd3 v at position ⃗x with a velocity ⃗v
at time t. The six-dimensional space mapped out by ⃗x and ⃗v is known as
phase space. The three-dimensional space mapped by ⃗x is known as posi-
tion space; the three-dimensional space mapped by ⃗v is known as velocity
space. The function f (⃗x, ⃗v , t) is known as the distribution function.
The distribution function is normalized so that
! !
f (⃗x, ⃗v , t)d3 xd3 v = N , (1.15)
where N is the total number of particles in the system. If we choose, we can

also define a mass distribution function
fm (⃗x, ⃗v , t) = mf (⃗x, ⃗v , t) , (1.16)
where m is the mass of an individual particle. If n types of particle are

present, each with a different mass, the total mass distribution function is
n
"
fm = mk fk (⃗x, ⃗v , t) , (1.17)
k=1
1
The Boltzmann Equation also plays a key role in the dynamics of stellar systems.
5
where mk is the mass of the k th type of particle. If particles are neither

created nor destroyed,
3
# $
∂f " ∂f ∂f df %%
+ ẋi + v̇i = % . (1.18)
∂t i=1 ∂xi ∂vi dt c
The factor on the right hand side of the above equation is the rate at which
particles are bumped into a phase space volume by collisions with other par-
ticles. The word “collisions”, when used in this context, can be misleading,
since it implies that the individual particles behave like billiard balls, inter-
acting only when they touch each other. In an ionized gas, the individual
charged particles interact through long-range electromagnetic forces. A “col-
lision”, in that case, is a close encounter between charged particles during
which the electromagnetic force changes their direction of motion by ∼ 90◦
or more. Similarly, in stellar dynamics, a “collision” between stars is a close
encounter during which gravity changes their direction of motion by ∼ 90◦
or more.
We may rewrite equation (1.18) by making use of the fact that ⃗x˙ = ⃗v and
that ⃗v˙ = ⃗g , where ⃗g is the acceleration of a particle at position ⃗x with velocity
⃗v . The gravitational force acting on the particle results in an acceleration
⃗ ,
⃗g = −∇Φ (1.19)
where the potential Φ(⃗x) has contributions both from the self-gravity of the
system of particles and from any external gravitational fields that may be
present.2 The electromagnetic force acting on the particle results in an ac-
celeration # $
q ⃗ ⃗v ⃗ ,
⃗g = E+ ×B (1.20)
m c
⃗ is the electric field at
where q is the charge of the particle, m is its mass, E
⃗ is the magnetic field there.
the particle’s location, and B
The phase space continuity equation can now be written in the form
3
# $
∂f " ∂f ∂f df %%
+ vi + gi = % . (1.21)
∂t i=1 ∂xi ∂vi dt c
2
These notes are Newtonian. Unless I make an explicit statement to the contrary, all
motions are assumed to be non-relativistic, and gravity is assumed to be an inverse-square-
law force.
This equation is known as the Boltzmann equation.

In equilibrium, particles are bumped into an element of phase space at the
same rate that they are bumped out, and the collisional term on the right-
hand side is equal to zero. If the collisional term is ignored, the resulting
equation is called the collisionless Boltzmann equation. The collisionless
Boltzmann equation is the basis of stellar dynamics and of hydrodynamics.
However, we are often unable to determine the full velocity distribution at
every point in position space. What we usually want to deal with is the
number density or mass density in position space, plus the mean (or bulk)
velocity of the gas, plus its velocity dispersion.
Now, let’s take the moments of the collisionless Boltzmann equation.
Let’s start with fm (the mass density in phase space) rather than f (the
number density in phase space), because it is useful to think about the con-
servation of mass (m), the conservation of momentum (m × ⃗v ), and the
conservation of kinetic energy (m × v 2 /2). The mass density in position
space is found by integrating over all velocities:
!
ρ(⃗x, t) = fm (⃗x, ⃗v , t)d3 v . (1.22)
For any measurable quantity Q, the mass-weighted average value at a position

⃗x at time t is
1!
⟨Q⟩ = Qfm d3 v . (1.23)
ρ
Integrate the collisionless Boltzmann equation over all velocities:
∂ ! " ∂ ! " ! ∂fm

fm d3 v + vi fm d3 v = − gi d3 v . (1.24)
∂t i ∂x i i ∂v i
The partial differentials have been taken outside the integrals since ⃗x, ⃗v , and
t are all independent variables. The right hand side of the above equation
vanishes by the divergence theorem, as long as fm → 0 as v → ∞.
Then
∂ρ " ∂
+ (ρ⟨vi ⟩) = 0 . (1.25)
∂t i ∂xi
Now introduce a new symbol:
⃗u ≡ ⟨⃗v ⟩ . (1.26)
7
The velocity ⃗u is the bulk velocity at a given point in space. The velocity
⃗v of a particular particle may then be broken into two components:
⃗v = ⃗u + w
⃗ , (1.27)
where w⃗ is the random velocity.

In vector form,
∂ρ ⃗
+ ∇ · (ρ⃗u) = 0 . (1.28)
∂t
This is the continuity equation. It states that mass is conserved. More-
over, the flow of matter is continuous; mass does not disappear at one point
and simultaneously appear at another point some distance away.
The continuity equation is a single equation in four unknowns (ρ and the
three components of ⃗v ). To obtain more equations, let’s find the next higher
moment of the collisionless Boltzmann equation; multiply by vj and integrate
over all velocities.
∂ ! " ∂ ! " ! ∂f 3
vj fm d3 v + vj vi fm d3 v = − gi vj dv. (1.29)
∂t i ∂xi i ∂vi
Rewriting the integrand on the right hand side, we have
& '
∂ " ∂ " ! ∂
(ρ⟨vj ⟩) + (ρ⟨vj vi ⟩) = − gi (vj f ) − δij f d3 v . (1.30)
Now we use the decomposition ⃗v = ⃗u + w ⃗ on the left, and the divergence

theorem on the right (assuming that fm v → 0 as v → ∞). This tells us that
∂ " ∂
(ρuj ) + (ρui uj + ρ⟨wi wj ⟩) = ρgj . (1.31)
∂t i ∂xi
This is the equation of momentum conservation. It is convenient to write

the pressure tensor ρ⟨wi wj ⟩ in the form
ρ⟨wi wj ⟩ = P δij − πij , (1.32)
where P is the pressure,

1
⃗ 2⟩ ,
P ≡ ρ⟨|w| (1.33)
3
and πij is the viscous stress tensor
πij ≡ P δij − ρ⟨wi wj ⟩ . (1.34)

It is useful to make this division because in some cases the viscous stress ten-
sor can be ignored. In other cases, the viscous stress tensor can be computed
in terms of the shear of the bulk velocity.
The equation for momentum conservation using the formalism that we
have adopted:
∂ " ∂
(ρuj ) + (ρui uj + P δij − πij ) = ρgj . (1.35)
∂t i ∂xi
The equation can be written in tensor form:

∂ ↔
⃗ · (ρ⃗u⃗u + P I − ↔
(ρ⃗u) + ∇ π) = ρ⃗g . (1.36)
∂t
↔
In the above equation, I is the unit matrix:
⎛ ⎞
↔
1 0 0
I= ⎜
⎝0 1 0⎠ .
⎟
(1.37)
0 0 1
The tensor product of the velocities is

⎛ ⎞
u1 u1 u 1 u2 u1 u 3
⃗u⃗u = ⎜
⎝ u2 u1 u 2 u2 u2 u 3 ⎟
⎠ . (1.38)
u3 u1 u 3 u2 u 3 u3
Finally, the viscous stress tensor is the symmetric traceless tensor

⎛ ⎞
π11 π12 π13
↔
π= ⎝ π12
⎜
π22 π23 ⎠ .
⎟
(1.39)
π13 π23 −π11 − π22
The tensor form of the momentum conservation equation tells us that the
time derivative of a conserved quantity plus the divergence of a flux is equal
to a source term; this is the standard form of a conservation equation.
Let’s go one step further and look at the conservation of kinetic energy.
Multiply the collisionless Boltzmann equation by v 2 , and integrate over all
velocities.
∂ ! 2 3
" ∂ !
2 3
" ! ∂fm 3
v fm d v + vi v fm d v = − gi v 2 dv. (1.40)
9
Rewriting the integrand on the right hand side,

& '
∂ " ! ∂
(ρ⟨v 2 ⟩) + (ρ⟨vi v 2 ⟩) = − (v 2 fm ) − 2vi fm d3 v .
"
gi (1.41)
∂t i i ∂vi
Breaking the velocity into its ordered and random components (⃗v = ⃗u + w)
⃗
we find
∂
(ρu2 + ρ⟨w2 ⟩) + (1.42)
∂t
" ∂
(ρ[ui u2 + ui ⟨w2 ⟩ + 2 uj ⟨wi wj ⟩ + ⟨wi w2 ⟩]) = 2
" "
gi ρui .
i ∂xi j i
Some more definitions: the specific internal energy of a monatomic gas is

1
ε ≡ ⟨w2 ⟩ ; (1.43)
2
this is just the energy per unit mass contributed by the random internal
motions. Note that for a monatomic gas
3P
ε= . (1.44)
2ρ
Next, the conductive heat flux is
1
F⃗ ≡ ρ⟨ww
⃗ 2⟩ . (1.45)
2
If the distribution of random velocities w
⃗ is symmetric about zero, then the
⃗
conductive heat flux F will vanish. However, if w⃗ has a skewed distribution,
then the “hot” particles (those with large random velocities) will have a net
drift relative to the “cold” particles.
Using the newly defined quantities, the equation of energy conservation
is
∂ 1 2 ⃗ · ( 1 ρu2 + P + ρε)⃗u− ↔
. /
( ρu + ρε) + ∇ π ·⃗u + F⃗ = ρ⃗u · ⃗g . (1.46)
∂t 2 2
By combining the mass conservation equation with the momentum conser-
vation equation, we can write down an equation for the conservation of the
bulk kinetic energy ( ρu2 /2):
∂ 1 2 ⃗ · ( 1 ρu2⃗u) = ρ⃗u · ⃗g − ⃗u · ∇P ⃗ ↔
⃗ + ⃗u · (∇·
( ρu ) + ∇ π) . (1.47)
∂t 2 2
Figure 1.1: Jim and Huckleberry Finn, drifting along, are Lagrangian ob-
servers; someone on the bank would be an Eulerian observer of the river.
By subtracting this equation from the equation for the conservation of the
total energy, we find the internal energy equation:
∂ ⃗ · (ρε⃗u) = −P ∇
⃗ · ⃗u − ∇
⃗ · F⃗ + Ψ ,
(ρε) + ∇ (1.48)
∂t
where
" ∂ui
Ψ≡ πij . (1.49)
i,j ∂xj
The function Ψ is the rate of viscous dissipation; viscosity converts bulk
kinetic energy into internal energy.
The partial time derivative of a function Q, which we have written as
∂Q/∂t, is the rate of change as viewed by an observer at a fixed coordinate
position, who is watching the gas flow by. Such an observer is called an
Eulerian observer. An alternative point of view is that of an observer who
is moving along with the bulk flow of the gas. Such an observer is called a
Lagrangian observer.
The Lagrangian time derivative,
DQ ∂Q ⃗ ,
= + ⃗u · ∇Q (1.50)
Dt ∂t
11
is the rate of change of Q as seen by a Lagrangian observer, who is moving

along with a designated parcel of gas. Note that the Lagrangian time deriva-
⃗ Q
tive of a vector is (⃗u · ∇) ⃗ = (⃗u · ∇Q
⃗ x )ı̂ + (⃗u · ∇Q
⃗ y )ȷ̂ + (⃗u · ∇Q
⃗ z )k̂), when
written out in Cartesian coordinates. For a spherically symmetric system,
where all quantities are functions only of t and r (the radial coordinate), it
is useful to write the Lagrangian time derivatives in spherical coordinates:
DQ(r, t) ∂Q ∂Q
= + ur . (1.51)
Dt ∂t ∂r
and, for vectors,
⃗ t)
DQ(r, ∂Q⃗ ⃗
∂Q
= + ur . (1.52)
Dt ∂t ∂r
In a Lagrangian form, the continuity equation is
Dρ ⃗ · ⃗u .
= −ρ∇ (1.53)
Dt
The momentum equation is
D⃗u 1⃗ 1⃗ ↔
= − ∇P + ∇· π +⃗g . (1.54)
Dt ρ ρ
The acceleration is provided by the pressure gradient, the viscous drag, and
by the gravitational and electromagnetic forces that we have included in ⃗g .
The internal energy equation is
Dε P⃗ 1⃗ ⃗ 1
=− ∇ · ⃗u − ∇ ·F + Ψ . (1.55)
Dt ρ ρ ρ
The change in internal energy is given by a P dV work term, a heat conduction
term, and a viscous heating term.
Equations (1.53), (1.54), and (1.55) are valid for nonradiative gas dy-
namics. Since this is a course on radiative gas dynamics, we must take into
account the effect of radiation on the momentum and energy of the gas. To
the right hand side of the internal energy equation, we must add the term
(Γ − Λ)/ρ, where Γ is the volumetric heating rate (or “gain”), and Λ is the
volumetric cooling rate (or “loss”).3
3
In certain cases, such as systems near the Eddington Limit, radiative pressure has
an important effect on the motion of the gas, and must be included in the momentum
conservation equation.
The equations for conservation of mass, momentum, and energy provide

five equations in all. How many unknowns are there? There’s the density
ρ, the pressure P , the bulk velocity ⃗u (3 unknowns), the conductive heat
↔
flux F⃗ (3 unknowns), and the viscous stress tensor π (5 independent com-
ponents). The internal energy, for a monatomic gas, is given by the relation
↔
ε = 3P/(2ρ), the rate of viscous dissipation can be computed from π and ⃗u,
the acceleration due to self-gravity is given by Poisson’s equation
⃗ · ⃗g = −4πGρ ,
∇ (1.56)
and the acceleration from any external sources is assumed to be known.

So, we have 5 equations and 13 unknowns. The clever way to break out
↔
of the hierarchy of equations is to express the viscous stress tensor π and
the conductive heat flux F⃗ in terms of ρ, P , and ⃗u. This will leave us with
5 equations and 5 unknowns. Once we specify our initial conditions and
boundary conditions, we can proceed to solve the equations.
Chapter 2
Viscosity, Heat Conduction,

and Other Complications
↔
The task at hand is to find expressions for the viscous stress tensor π and the
conductive heat flux F⃗ in terms of the density, pressure, and bulk velocity of
a gas. We will start with an ideal gas.
The distribution of random velocities for an ideal gas (otherwise known
as a “perfect gas”) is the Maxwellian distribution
mw2 3
13/2 # $
m
0
3
f (w)d
⃗ w= exp − dw, (2.1)
2πkT 2kT
where k is the Boltzmann constant, m is the particle mass, and T is the ki-
netic temperature. The mean square random velocity is then ⟨w2 ⟩ = 3kT /m.
Using equation (1.33), we find that the pressure is given by the familiar ideal
gas law
ρ
P = kT . (2.2)
m
For the Maxwellian distribution, ⟨wi wj ⟩ = 0 when i ̸= j; hence, all elements
↔
of π vanish. Furthermore, ⟨w2 wi ⟩ = 0, so the conductive heat flux also
vanishes. Ideal gases are delightfully simple: no viscosity, no heat conduction.
The conservation laws for an ideal gas take the simplified form
Dρ ⃗ · ⃗u
= −ρ∇ (2.3)
Dt
D⃗u 1⃗
= − ∇P + ⃗g (2.4)
Dt ρ
13
14 CHAPTER 2. VISCOSITY & HEAT CONDUCTION
Dε P⃗
= − ∇ · ⃗u . (2.5)
Dt ρ
These three equations are known collectively as the Euler equations. In
conjunction with the ideal gas law
3P 3kT
ε= = , (2.6)
2ρ 2m
they describe the motions of an inviscid, ideal gas of point masses.1 Com-
bining the continuity and energy conservation equations, we see that for an
ideal gas,
Dε P Dρ DV
= 2 = −P , (2.7)
Dt ρ Dt Dt
where V ≡ 1/ρ is the specific volume of the gas. The change in internal
energy of a ideal gas is equal to the P dV work that is done on the gas.
The first law of thermodynamics states that
T ds = dε + P dV , (2.8)
where T and P are the temperature and pressure of the gas, s is the specific
entropy, ε is the specific internal energy, and V is the specific volume. Since,
for an ideal gas,
Dε DV
= −P , (2.9)
Dt Dt
it is necessary that
Ds
T =0. (2.10)
Dt
An ideal gas, in the absence of heat sources and sinks, undergoes only adi-
abatic processes. (That is, its entropy remains constant.)
If a gas is compelled to have a constant volume (by being enclosed in a
rigid box, for instance), the first law of thermodynamics reduces to dq = dε,
where dq is the heat added. The specific heat at constant volume is then
# $ # $
∂q ∂ε
cV ≡ = . (2.11)
∂T V
∂T V
However, for an ideal gas, ε(V, T ) = ε(T ). Thus, we may write, quite gener-
ally,
dε = cV dT (2.12)
1
Inviscid = having no viscosity.
15
and
dq = cV dT + P dV . (2.13)
Now consider gas that is kept at a constant pressure. The ideal gas law tells
us that
k
P dV = dT (2.14)
m
when pressure is kept constant, and hence that
# $
∂q k
cP ≡ = cV + . (2.15)
∂T P
m
Note that cP > cV ; when pressure is held constant, some of the added heat
goes into P dV work instead of into internal energy.
The adiabatic index of a gas is defined as γ ≡ cP /cV . For an ideal gas
of point particles, ε = (3kT )/(2m), cV = (3k)/(2m), cP = (5k)/(2m) and
γ = 5/3. The adiabatic index for a gas of diatomic molecules is γ = 7/5.
The adiabatic index is a function of the number of degrees of freedom of
the particles; diatomic molecules have rotational degrees of freedom that
are not present for point masses. The internal energy ε, if the gas particles
are not spherical atoms, also includes rotational energy in addition to the
translational energy. In general, the internal energy is
1 kT 1 P
ε= = . (2.16)
γ−1 m γ−1 ρ
Now consider an adiabatic process, for which the first law of thermodynamics
may be written
dε + P dV = 0 (2.17)
P
cV dT − 2 dρ = 0 . (2.18)
ρ
For an ideal gas, # $
mP dP dρ
dT = − . (2.19)
kρ P ρ
Combining equations (2.18) and (2.19),
& # $ '
mP dP cV + k/m dρ
cV − =0. (2.20)
kρ P cV ρ
Thus, dP/P = γ(dρ/ρ), and
P (ρ) = P0 (ρ/ρ0 )γ . (2.21)
A gas that has an equation of state in this form (P ∝ ργ ) is known as a

polytrope.
An ideal gas has no viscosity. However, real gases aren’t perfectly ideal,
and in many astrophysical applications (such as accretion disks) viscosity is
important. To handle viscosity in a relatively simple manner, we need to
↔
express the tensor π in terms of the bulk velocity ⃗u. First, viscous frictional
forces will occur when two fluid elements move relative to each other. Hence,
↔
π must depend on the spatial derivatives of the velocity, ∂ui /∂xj . Second,
viscous forces must disappear when the fluid is at rest or is in uniform transla-
tional or rotational motion. Third, for small velocity gradients, the elements
↔
of π will be linearly proportional to the velocity gradient. (Fluids for which
πij ∝ ∂ui /∂xj are known as linear fluids, or Newtonian fluids.)
The most general viscous stress tensor that satisfies these three require-
ments is
πij = µDij + β ∇⃗ · ⃗uδij , (2.22)
where µ is the coefficient of shear viscosity, β is the coefficient of bulk
viscosity, and
∂ui ∂uj 2⃗
Dij ≡ + − ∇ · ⃗uδij . (2.23)
∂xj ∂xi 3
The deformation tensor Dij vanishes in the case of uniform expansion or
contraction. Thus, the shear viscosity term represents pure shear, with no
change in volume, while the bulk viscosity term represents pure expansion
or contraction.
In the cgs system, the coefficients of viscosity µ and β are measured
in ‘poises’, where 1 poise equals 1 g/cm/sec.2 At room temperature, the
coefficient of shear viscosity for air is µ ∼ 2 × 10−4 poise; for water, µ ∼
10−2 poise.
The coefficient of shear viscosity can be approximately calculated in a
fairly simple manner. A gas has a number density n of particles, each with
a mass m. The gas has a temperature T , and hence a thermal velocity
vt = (kT /m)1/2 . In addition to the random thermal velocity of the particles,
2
The ‘poise’ is named after the French physiologist Jean Poiseuille.
17
there is also a bulk velocity ux in the x direction, where ux is a function of

y. (In other words, there exists a shear, ∂ux /∂y ̸= 0.)
Now consider the plane y = y0 . Because of the random velocities of the
particles, there is a flux of particles moving downward through the plane, with
a magnitude per unit area of ∼ nvt /2. There is a flux of equal magnitude
upward through the plane. Thus, there is no net flux of mass density through
the plane. However, if ∂ux /∂y > 0, the downward flux of particles will have a
greater momentum in the x direction than will the upward flux of particles.
The net result: a transfer of momentum opposite to the direction of the
velocity gradient ∂ux /∂y. The particles that cross the plane from above will
have had their last collision at a distance ∼ λ above the plane. Thus, the
downward flux of angular momentum will be
& '
nvt ∂ux
Pyx ∼ m ux (y0 ) + λ . (2.24)
2 ∂y
Similarly, the upward flux of angular momentum will be
& '
nvt ∂ux
Pyx ∼ m ux (y0 ) − λ . (2.25)
2 ∂y
The net momentum flux is then
∂ux
Pyx ∼ −nvt mλ , (2.26)
∂y
which tells us that the coefficient of shear viscosity is
µ ∼ nvt mλ ∼ (mkT )1/2 /σ . (2.27)
where σ is the cross sectional area of the gas particles. To lowest order, the
coefficient of bulk viscosity, β, is equal to zero. Note that µ is independent
of the density of the gas. Note also that the viscosity of a gas increases
with temperature, in contrast to the behavior of liquids, in which viscosity
generally decreases with temperature.3 For neutral atomic hydrogen, the
coefficient of shear viscosity (including relevant factors of π) is
11/2
T
0
−3
µ = 6 × 10 poise . (2.28)
104 K
3
The common English simile, “as slow as molasses in January” makes reference to the
higher viscosity of molasses at lower temperatures.
Another parameter that is frequently used to describe viscosity is the kine-

matic viscosity,
ν ≡ µ/ρ . (2.29)
From equation (2.27), we expect that the kinematic viscosity of a gas should
be of order
ν ∼ µ/(nm) ∼ vt λ , (2.30)
where vt is the thermal speed, and λ is the mean free path of gas particles.
The kinematic viscosity of the air around you is ν ∼ 0.15 cm2 s−1 . For a gas
of neutral atomic hydrogen,
11/2 0
T n
0 1
21 2 −1
ν ∼ 4 × 10 cm s −1 . (2.31)
104 K 1 cm−3
A dimensionless number much used by fluid dynamicists is the Reynolds
number,
ρuL uL u L
0 10 1
Re ≡ = ∼ , (2.32)
µ ν vt λ
where L is the typical length scale of the system we’re looking at. The
Reynolds number is the ratio of the inertial forces (∼ ρu2 /L) to the vis-
cous forces (∼ µu/L2 ). Thus, when Re ≫ 1, the viscous forces are negligi-
ble. As the scale of interest L becomes smaller and smaller, there is some
length scale on which viscosity becomes important. Viscous flow tends to be
laminar, while less-viscous flow tends to be turbulent. However, the transi-
tion between laminar and turbulent flow is not determined uniquely by the
Reynolds number; the geometry of the system is also important. For fluids
moving through a straight pipe, to take a well-studied example, flow with
Re <∼ 3000 is laminar and flow with Re > ∼ 3000 is turbulent. For a dimpled
golf ball, the critical Reynolds number for the transition from laminar to
turbulent flow is Recr ∼ 30,000; for a smooth golf ball, it’s Recr ∼ 300,000.
What about the conductive heat flux? It is found empirically that heat
flows from hot regions to cold regions, with the flux proportional to the
temperature gradient. Mathematically, this is expressed by Fourier’s Law:
F⃗ = −K ∇T
⃗ , (2.33)
where K is the coefficient of thermal conductivity. For a neutral gas,
# $1/2
5 k kT
K = cV µ ∼ . (2.34)
2 σ m
19
Thus, for a gas of neutral atomic hydrogen,

11/2
T
0
6 −1 −1 −1
K = 2 × 10 erg cm s K . (2.35)
104 K
Under many astrophysical circumstances, the conductive heat flux is small
compared to other energy fluxes. For instance, in stellar interiors, the ra-
diative heat flux and the convective heat flux battle for supremacy, with
the conductive heat flux trailing badly in third place. However, in the Sun’s
transparent corona, to mention an example we’ll be discussing in Chapter 11,
conduction is the main transport mechanism for heat.
Summary of Results So Far

The conservation equations are
Dρ ⃗ · ⃗u
= −ρ∇ (2.36)
Dt
D⃗u 1⃗ 1⃗ ↔
= − ∇P + ∇ · π + ⃗g (2.37)
Dt ρ ρ
Dε P⃗ 1⃗ ⃗ 1 1
= − ∇ · ⃗u − ∇ · F + Ψ + (Γ − Λ) , (2.38)
Dt ρ ρ ρ ρ
in conjunction with the equation of state for an ideal monatomic gas
3P 3kT
ε= = . (2.39)
2ρ 2m
In the simplest approximation, we ignore the viscosity and heat conduction
↔
by setting π, F⃗ , and Ψ equal to zero. When this approximation is made, the
conservation equations are known as the Euler equations.
When viscosity and heat conduction are not negligible, we make the ap-
proximations
⃗ · ⃗uδij
πij = µDij + β ∇ (2.40)
and
F⃗ = −K ∇T
⃗ . (2.41)
In this approximation, the viscous force per unit volume is
⃗ ·↔
∇ ⃗ ∇
π = µ∇2⃗u + (β + µ/3)∇( ⃗ · ⃗u) . (2.42)
The rate of viscous energy dissipation is equal to

µ ↔ 2 ⃗ · ⃗u)2 ,
Ψ= | D | + β(∇ (2.43)
2
where the square of the scalar norm of the deformation tensor is
↔
| D |2 =
"
Dij Dij . (2.44)
i,j
Chapter 3
Introduction to Sound &

Shocks
Suppose we are in an inviscid, non-heat-conducting, nonradiative medium.

We will first consider a case with plane parallel symmetry (all the properties
of the gas depend only on x and t). The mass continuity equation is then
∂ρ ∂ρ ∂u
+u +ρ =0. (3.1)
∂t ∂x ∂x
The momentum equation is
∂u ∂u 1 ∂P
+u =− +g . (3.2)
∂t ∂x ρ ∂x
If the medium is self-gravitating, the acceleration g is given by Poisson’s
equation:
∂g
= −4πGρ . (3.3)
∂x
A uniform static medium, with ρ = ρ0 , u = 0, and P = P0 , will satisfy the
equations of continuity and of motion if we perpetuate the Jeans swindle.
In a homogeneous, isotropic universe, the gravitational acceleration must be
⃗g0 = 0, by symmetry. However, Poisson’s equation will yield g0 = 0 only
if ρ0 = 0. From the Newtonian point of view, in other words, an infinite,
static, matter-filled universe cannot exist. The Jeans swindle deals with this
difficulty by ignoring it.1 Let us, like Jeans, assume that g0 = 0 for the
uniform medium.
1
If you strongly object to being swindled, try reading ”Mathematical Vindications of
the Jeans Swindle”, by M. Kiessling (astro-ph/9910247).
21
22 CHAPTER 3. SOUND & SHOCKS
Now, let us introduce small perturbations to the system, so that
ρ = ρ0 + ρ1 (x, t) (3.4)
u = u1 (x, t) (3.5)
P = P0 + P1 (x, t) (3.6)
g = g1 (x, t) . (3.7)
where |ρ1 |/ρ0 ≪ 1 and |P1 |/P0 ≪ 1. The linearized perturbation equations
are then
∂ρ1 ∂u1
+ ρ0 = 0 (3.8)
∂t ∂x
∂u1 dP ∂ρ1
ρ0 + | = g1 ρ 0 (3.9)
∂t dρ 0 ∂x
∂g1
= −4πGρ1 . (3.10)
∂x
In writing the equation of motion, I have made the implicit assumption that
P = P (ρ). Taking the time derivative of equation (3.8), and subtracting the
spatial derivative of equation (3.9), we find
∂ 2 ρ1 dP ∂ 2 ρ1
− | = 4πGρ0 ρ1 . (3.11)
∂t2 dρ 0 ∂x2
If the self-gravitation term on the right hand side is small enough to be

ignored, this is just a wave equation with a propagation speed
# $1/2
dP
a= . (3.12)
dρ
The density and pressure perturbations that propagate through the medium
with velocity ±a are sound waves. For a polytrope with adiabatic index γ,
the sound speed at density ρ0 and pressure P0 is
# $1/2 # $1/2
γP0 γk
a0 = = T . (3.13)
ρ0 m
For a neutral atomic gas, the sound speed is a = 0.12 km s−1 µ−1/2
a (T /1 K)1/2 ,
where µa is the mean atomic mass in units of the proton mass.
23
When is the self-gravity of the medium negligible? Consider a sound wave

of the form ρ1 (x, t) ∝ exp[i(ωt−kx)]. In that case, equation (3.11), including
the self-gravity term on the right hand side, reduces to the dispersion relation
ω 2 = k 2 a20 − 4πGρ0 . (3.14)
Thus, ω is real for wavenumbers k > kJ , where
2
kJ ≡ 4πGρ0 /a0 . (3.15)
For k < kJ , the frequency ω is imaginary, and the perturbations grow expo-
nentially due to their self-gravity.
The Jeans length for a neutral atomic gas is
11/2 0 1−1/2
2π T n
0
λJ ≡ ∼ 20 pc µ−1
a (3.16)
kJ 1K 1 cm−3
and the Jeans mass is
13/2 0 1−1/2
π T n
0
mJ ≡ λ3J ρ0 ∼ 100 M⊙ µ−2
a . (3.17)
6 1K 1 cm−3
Sound waves with a wavelength longer than λJ will collapse gravitation-
ally. On the other end of the size spectrum, sound waves with a wavelength
shorter than the mean free path λ = 1/(nσ) cannot be created. Even in
relatively dense molecular clouds, λ ∼ 0.2 AU; a sound with this wavelength
will have a frequency ∼ 10−8 Hz. From now on, we will deal with sound
whose wavelength lies between the mean free path λ and the Jeans length
λJ , so that we can ignore the effects of discreteness and self-gravitation. But
what about the effects of viscosity and heat conduction?
The linearized 1-d continuity equation, as always, is
∂ρ1 ∂u1
= −ρ0 . (3.18)
∂t ∂x
The momentum equation, including viscosity, is
∂u1 ∂P1 ∂ 2 u1
ρ0 =− + µ′ 2 , (3.19)
∂t ∂x ∂x
where the effective coefficient of viscosity in the one-dimensional case is µ′ ≡
4µ/3 + β. The energy equation, including heat conduction, is
∂ε1 ∂u1 ∂ 2 T1
ρ0 = −P0 +K . (3.20)
∂t ∂x ∂x2
The temperature perturbation, in terms of the density and pressure, is

m P0
T1 = [P1 − ρ1 ] . (3.21)
kρ0 ρ0
The perturbation to the specific internal energy is
1 P0
ε1 = [P1 − ρ1 ] . (3.22)
(γ − 1)ρ0 ρ0
Using these relations in conjunction with the continuity equation, the energy
equation takes the form
∂ ∂2
(P1 − a20 ρ1 ) = γχ 2 (P1 − a20 ρ1 /γ) , (3.23)
∂t ∂x
where χ = K/(ρ0 cP ).
If the perturbations are sine waves, of the form
ρ1 = R exp[i(ωt − kx)] (3.24)

P1 = P exp[i(ωt − kx)] (3.25)
u1 = U exp[i(ωt − kx)] , (3.26)
then the conservation equations yield the relations
iωR − iρ0 kU = 0 (3.27)

−ikP + [iωρ0 + µ′ k 2 ]U = 0 (3.28)
(iω + γk 2 χ)P − (iω + k 2 χ)a20 R = 0 . (3.29)
This set of equations yields the dispersion relation
1 − ik 2 χ/ω 2 2
ω2 = a k + iµ′ k 2 ω/ρ0 . (3.30)
1 − iγk 2 χ/ω 0
In the absence of viscosity and heat conduction, the dispersion relation is

ω 2 = a20 k 2 . Suppose, however, that we add a small µ and K, so that the
wavenumber is now k = ω/a0 + k1 , with |k1 | ≪ ω/a0 . The perturbation to
the wavenumber is then
ω2
k1 = −i [µ′ + (γ − 1)ρ0 χ] . (3.31)
2a30 ρ0
25
The density perturbation is now
ρ1 = R exp[i(ωt − ωx/a0 )] exp(−x/L1 ) , (3.32)
with the attenuation length

2a30 ρ0 ′
L1 = 2
[µ + (γ − 1)ρ0 χ]−1 . (3.33)
ω
The viscosity and heat conduction damp the sound waves by converting the
sound energy into random kinetic energy.
If we use the sound speed a2 ∼ kT /m and viscosity µ ∼ (mkT )1/2 /σ for
an atomic gas, we find that the attenuation length is
a20 Λ2
L1 ∼ nσ ∼ , (3.34)
ω2 λ
where Λ is the wavelength of the propagating sound and λ is the mean free
path of the gas. For sound to propagate, we require Λ ≫ λ, and hence
L1 ≫ Λ; the sound is not attenuated significantly until it has traveled for
many wavelengths. In air at room temperature, L1 ∼ 600 km(ω/1000 Hz)−2 .
In the interstellar medium (ISM), L1 ∼ 700 AU(ω/10−10 Hz)−2 .
So far, we have been assuming that the sound waves are of infinitesimal
magnitude, with |ρ1 |/ρ0 ≪ 1, |P1 |/P0 ≪ 1, and u1 ≪ a0 . As a consequence,
we have assumed that the sound speed in the medium has the uniform value
a = a0 . However, the sound speed is a function of density; for a polytrope,
a ∝ ρ(γ−1)/2 . As a consequence, crests of sound waves will travel more rapidly
than troughs of sound waves, as illustrated in Figure 3.1. Although a wave
on the surface of water can be triple-valued, as shown at time 3 in Figure 3.1,
creating a “breaker”, this is forbidden for a sound wave. Sound waves, there-
fore, will steepen until a shock forms. A shock front is a surface that marks
a sudden jump in the density, pressure, and velocity of a gas. The shock
front is supersonic – that is, it propagates at a velocity faster than the sound
speed in the unshocked medium.
Shocks are ubiquitous in the ISM. Whenever the bulk velocity u is larger
than the sound speed, you are likely to form shocks. For instance, there are
shocks associated with:
• cloud - cloud collisions (u ∼ 10 km s−1 ),
• galaxy - galaxy collisions (u ∼ 300 km s−1 ),

Figure 3.1: The steepening of a sound wave as a crest overtakes a trough.

27
• stellar winds encountering the ambient ISM (u ∼ 3000 km s−1 ),

• jets from radio galaxies encountering the ambient intergalactic medium
(u ∼ 5000 km s−1 ),
• supernova ejecta encountering the ambient ISM (u ∼ 2 × 104 km s−1 ),
and
• gas accreting onto neutron stars (u ∼ 105 km s−1 ).
Shocks play an important role role in determining the structure of the ISM.
For instance, they heat the ISM. Moreover, in the spiral arms of galaxies,
shocks compress the gas, which can trigger star formation.
Let us see how shocks behave once they are created through the steepen-
ing of a sound wave. Consider, to begin with, a simple plane parallel shock,
as shown in Figure 3.2. The math is easiest if we place ourselves in a frame of
reference that is comoving with the shock front. Let u1 be the bulk velocity
of the unshocked gas (upstream of the shock) in this frame of reference, and
let ρ1 , P1 , and a1 be the density, pressure, and sound speed of the unshocked
gas, which is assumed to be uniform. The bulk velocity, density, pressure,
and sound speed immediately downstream from the shock are u2 , ρ2 , P2 , and
a2 . The transition layer between the unshocked gas and the postshock gas
is very thin. For a neutral atomic gas, ∆x ∼ λ, the mean free path length.
For most purposes, it is adequate to regard the shock transition layer as
an infinitesimally thin surface. For a steady-state shock, the conservation
equations have the form
d
(ρu) = 0 (3.35)
dx
d
(ρu2 + P ) = 0 (3.36)
dx
d
(ρ[u2 /2 + ε] + P ) = 0 . (3.37)
dx
The gas properties immediately before and after being shocked are conse-
quently linked by the Rankine-Hugoniot jump conditions:
ρ 1 u 1 = ρ 2 u2 (3.38)
ρ1 u21+ P1 = ρ2 u22 + P2 (3.39)
1 2 1 2
u1 + ε1 + P1 /ρ1 = u + ε2 + P2 /ρ2 . (3.40)
2 2 2
Figure 3.2: The geometry of a plane parallel shock; we are in the shock’s
frame of reference, with pre-shocked gas flowing in from the left.
29
For a polytropic gas, the last of the Rankine-Hugoniot conditions may be

rewritten in the form
1 2 γ P1 1 γ P2
u + = u22 + . (3.41)
2 1 γ − 1 ρ1 2 γ − 1 ρ2
I have made the implicit assumption that γ is the same for the preshock and
postshock gas.
The Rankine-Hugoniot conditions are just the conservation equations in
a new guise. A dimensionless number that is often cited in the context of
shocks is the Mach number,
$1/2
ρ1 u21
#
M1 ≡ u1 /a1 = . (3.42)
γP1
The Mach number is the ratio of the velocity of the shock (relative to the
unshocked medium) to the sound speed in the unshocked medium. Using
the Rankine-Hugoniot conditions, we may solve for the density, pressure,
and temperature jumps in terms of M1 .
ρ2 (γ + 1)M12 u1
= = (3.43)
ρ1 (γ − 1)M12 + 2 u2
2
P2 2γM1 − (γ − 1)
= (3.44)
P1 γ+1
T2 [(γ − 1)M12 + 2][2γM12 − (γ − 1)]
= . (3.45)
T1 (γ + 1)2 M12
A strong shock is defined as one that is highly supersonic, with M1 ≫ 1. For

a strong shock,
ρ2 γ+1
≈ (3.46)
ρ1 γ−1
2
P2 ≈ ρ1 u21 (3.47)
γ+1
2(γ − 1) m 2
T2 ≈ u . (3.48)
(γ + 1)2 k 1
Thus, no matter how strong the shock is, the ratio ρ2 /ρ1 has a finite value;
for a monatomic gas, with γ = 5/3, the ratio is ρ2 /ρ1 = 4. However, a strong
shock is very efficient at converting the bulk kinetic energy of the upstream
gas (∼ ρ1 u21 ) into thermal energy.
A weak shock is defined as one that is barely supersonic, with M1 − 1 =
ϵ ≪ 1. For a weak shock,
ρ2 4
≈ 1+ ϵ (3.49)
ρ1 γ+1
P2 4γ
≈ 1+ ϵ (3.50)
P1 γ+1
T2 4(γ − 1)
≈ 1+ ϵ. (3.51)
T1 γ+1
Generally speaking, a shock converts supersonic gas (M1 > 1) into subsonic
gas (in the shock’s frame of reference). It increases density, decreases bulk
velocity (relative to the shock front), increases pressure, and increases tem-
perature.
The conversion of bulk kinetic energy to random thermal energy occurs
by dissipation within the shock layer itself. Within the shock layer, a jump
in velocity of ∆u ∼ a1 ∼ (kT /m)1/2 occurs over a length ∆x ∼ λ ∼ 1/(nσ).
The xx component of the viscous stress tensor within the shock is
√ √
4 ∆u mkT kT
πxx = ( π + β) ∼ √ (nσ) ∼ nkT . (3.52)
3 ∆x σ m
The increase in the specific entropy as the gas crosses the shock front is
s2 − s1 = cP ln(T2 /T1 ) − (k/m) ln(P2 /P1 ). For an extremely strong shock
(M1 → ∞), the entropy increase is s2 − s1 ∝ ln M1 .
So far, we’ve been dealing with normal shocks; that is, shocks in which
the the velocity vector ⃗u1 of the upstream, unshocked gas is perpendicular
to the shock front. However, some shocks are better described as oblique
shocks, in which ⃗u1 approaches the shock front at an angle other than 90
degrees. For instance, the shock waves associated with spiral arms of galaxies
can be approximated as oblique shocks. An oblique shock is illustrated in
Figure 3.3. Again, let us put ourselves into a frame of reference that is
moving along with a plane parallel shock front. This time, however, the
unshocked gas is flowing into the shock at an angle φ with respect to the
plane of the shock front. Thus, the bulk velocity u1 can be broken down into
a perpendicular component
u⊥1 = u1 sin φ (3.53)
31
Figure 3.3: The geometry of an oblique shock; we are in the shock’s frame
of reference, with pre-shocked gas flowing in from the left.
and a parallel component

u∥1 = u1 cos φ , (3.54)
which are perpendicular and parallel, respectively, to the shock front. The
postshock velocity ⃗u2 is rotated through an angle ψ away from ⃗u1 and towards
the shock front. The perpendicular and parallel components of ⃗u2 are
u⊥2 = u2 sin(φ − ψ) (3.55)
and
u∥2 = u2 cos(φ − ψ) . (3.56)
The parallel component of the velocity is unchanged by passage through the
shock. Thus,
u1 cos φ = u2 cos(φ − ψ) . (3.57)
The Rankine-Hugoniot jump conditions for the perpendicular component are
ρ1 u⊥1 = ρ2 u⊥2 (3.58)
ρ1 u2⊥1 = ρ2 u2⊥2 (3.59)
1 2 P1 1 2 P2
u⊥1 + ε1 + = u⊥2 + ε2 + . (3.60)
2 ρ1 2 ρ2
In an oblique shock, u⊥ obeys the same relations as u does in a normal shock.
From our previous encounter with the Rankine-Hugoniot jump conditions,
we know that the ratio of the perpendicular velocities is
u⊥2 2 + (γ − 1)M12 sin2 φ
= , (3.61)
u⊥1 (γ + 1)M12 sin2 φ
where M1 = u1 /a1 is the Mach number of the upstream, unshocked flow.
Since the parallel component of ⃗u is conserved and the perpendicular
component is decreased, the velocity vector is refracted away from the normal
to the shock front. Computing the actual value of the angle ψ through which
it is refracted cumbersome in the general case. Combining the equations for
the perpendicular and parallel components of the velocity, we find
1 − cos 2φ − 2/M12 sin 2φ
tan ψ = . (3.62)
γ − cos 2φ + 2/M12 1 − cos 2φ
In the case of a strong shock, the relation between φ and ψ takes the simpler
form
2 tan φ
tan ψ = . (3.63)
(γ + 1) + (γ − 1) tan2 φ
33
Figure 3.4: Shadowgraph of the detached bow shock in front of a bullet

traveling at M1 = 1.5; also note the turbulent wake. (Andrew Davidhazy:
RIT).
The maximum value ψm , in this case, is given by the relation

# $1/2
1
tan ψm = 2
, (3.64)
γ −1
which occurs when tan2 φ = (γ + 1)/(γ − 1). For a monatomic gas (γ = 5/3)
the maximum angle of refraction is ψm = 36.9◦ , which occurs when φ = 63.4◦ .
Even an arbitrarily strong shock can’t divert the flow through an angle of 90◦ .
If a blunt object, therefore, moves through a gas at supersonic speeds, then
a detached bow shock must form ahead of the object, as shown in Figure 3.4.
Chapter 4
Radiative &
Magnetohydrodynamic Shocks
I have been dealing, so far, with non-radiative shocks. Since, as we have seen,
a shock raises the density and temperature of the gas, it is quite likely, under
astrophysical conditions, that the post-shock gas will be highly radiative.
Consider, again, plane parallel shocks in which the gas flow is perpen-
dicular to the shock front, as shown in Figure 4.1. A gas with density ρ1
and pressure P1 flows into the shock front with a velocity u1 relative to the
front. The pre-shock gas is assumed to be in thermal equilibrium, and is not
radiating energy. The gas passes through the thin shock layer; immediately
downstream, it has density ρ2 , pressure P2 , and bulk velocity u2 given by
the Rankine-Hugoniot jump conditions. So far, everything is the same as in
the nonradiative shock. Now, however, we assume that there is an optically
thin radiative relaxation layer downstream of the shock front, in which
the cooling function L(ρ, T ) is greater than zero.1
Within the radiative relaxation layer, the conservation equations are
ρu = constant (4.1)
2
ρu + P = constant (4.2)
dε du
ρu = −P − ρL . (4.3)
dx dx
For an ideal gas, the energy equation reduces to
u dP γ du
+ P = −ρL (4.4)
γ − 1 dx γ − 1 dx
1
The cooling function L has units of energy per unit mass per unit time.
35
36 CHAPTER 4. RADIATIVE & MHD SHOCKS
Figure 4.1: The geometry of a radiative plane parallel shock; we are in the
shock’s frame of reference, with pre-shocked gas flowing in from the left.
37
or, using the continuity and momentum equations,

a2 − u2 du
= −L(ρ, T ) . (4.5)
γ − 1 dx
The cooling function for a very hot ionized gas, as an example, is domi-
nated by bremsstrahlung, for which L ∝ ρT 1/2 . At lower temperatures, the
cooling may be dominated by radiative recombination, or line radiation, or
some other process. Computing the exact values of ρ, P , T , and u in the
radiative relaxation layer is usually done numerically, with tabulated values
of L (or a functional fit). Without resorting to a computer, we can still derive
the qualitative behavior of the gas in the radiative relaxation layer.
As initial conditions, take the values of ρ2 , P2 , T2 and u2 immediately
downstream of the shock front. The four equations
ρu = ρ2 u2 (4.6)
ρu + P (x) = ρ2 u22 + P2
2
(4.7)
1 du
(γP/ρ − u2 ) = −L(ρ, T ) (4.8)
γ−1 dx
mP
T = (4.9)
k ρ
tell us the physical conditions within the radiative relaxation layer. Since
the post-shock gas is subsonic (u2 < γP/ρ) and since L > 0, equation (4.8)
indicates that the flow velocity u decreases as you move through the radiative
relaxation layer away from the shock front. Since u decreases, equation (4.6)
tells us that ρ increases. From equations (4.6) and (4.7), the pressure is
P = P2 + ρ2 u22 (1 − u/u2 ) . (4.10)
For a strong shock P2 > ρ2 u22 , and the pressure increases only slightly as the
gas flows through the radiative relaxation layer. The temperature, in terms
of u, is
ρ2 u22
! "# $
u u
! "
T = T2 1+ (1 − u/u2 ) ≈ T2 . (4.11)
u2 P2 u2
For a strong shock, T ∝ u and the temperature decreases in the radiative
relaxation layer. The gas will approach a steady state equilibrium in which
L = 0, and the density, pressure, temperature, and bulk velocity have the
constant values ρ3 , P3 , T3 , and u3 .
Frequently, the final temperature T3 is the same as the initial value T1 .

A radiative shock for which this equality holds true is called an isothermal
shock. The jump conditions for an isothermal shock are
ρ 3 u 3 = ρ 1 u1 (4.12)
ρ3 u23 + P3 = ρ1 u21 + P1 (4.13)
T3 = T1 . (4.14)
Making use of the equation of state P = ρkT /m, and using the “isothermal
sound speed” aT ≡ (kT1 /m)1/2 , we find that the increase in density is
"2
ρ3 u1
!
= = MT2 , (4.15)
ρ1 aT
where MT is the upstream isothermal Mach number. The bulk velocity is

decreased by a factor u3 /u1 = 1/MT2 . The properties of a MT = 2 isothermal
shock are displayed in Figure 4.2. Radiative isothermal shocks can achieve
arbitrarily high compression as the Mach number approaches infinity. Su-
pernova remnants in the early stages of their expansion are surrounded by
nonradiative shocks. However, as the spherical shocks expand and slow down,
the postshock temperatures drop below ∼ 104 K. At this point, line emission
kicks in, and the cooling becomes much more efficient. Thus, highly evolved
supernova remnants (such as the Cygnus Loop) have structures with very
high density contrasts.
The effectiveness of cooling is indicated by the cooling time tT ∼ ε/L. At
high temperatures, T ∼ 107 K, the cooling time is tT ∼ 6 Myr (1 cm−3 /nH ).
At a temperature of T ∼ 8000 K, cooling is much more efficient, and tT ∼
0.02 Myr (1 cm−3 /nH ). At lower temperatures, in the range 50 K < ∼ T < ∼
−3
600 K, the cooling time is tT ∼ 0.2 Myr (1 cm /nH ).
It is now time to consider magnetohydrodynamics (or MHD). MHD
is the study of the motions of a conducting fluid in the presence of mag-
netic fields. Magnetic field strengths encountered in astrophysics range from
10−6 G in the hot ISM to 1012 G on the surface of a neutron star. Magnetic
fields have a significant effect on the dynamics of astrophysical gases.
A good place to start our discussion of MHD is by writing down Maxwell’s
⃗ and the magnetic flux B.
equations for the electric field E ⃗
⃗ ·E
∇ ⃗ = 4πρe (4.16)
39
Figure 4.2: The density, bulk velocity, temperature, and pressure of material
passing through a Mach 2 isothermal shock.
⃗
⃗ ×E
∇ ⃗ = − 1 ∂B (4.17)
c ∂t
⃗ ·B
∇ ⃗ = 0 (4.18)
⃗
∇ ⃗ = 4π ⃗je + 1 ∂ E .
⃗ ×B (4.19)
c c ∂t
In the above equations, ρe is the charge density and ⃗je is the current density.
In nonrelativistic systems, the term c−1 ∂ E/∂t⃗ in equation (4.19) may be
ignored.
In an ionized gas, the current, electric field, and magnetic flux are related
through the expression
E⃗ = 1 ⃗je − 1 ⃗u × B⃗ , (4.20)
σ c
where σ is the electrical conductivity. Using equation (4.19) to eliminate je ,
⃗ = c ⃗ ⃗ − 1 ⃗u × B
⃗ .
E ∇×B (4.21)
4πσ c
Substituting this result into equation (4.17), we find the basic equation for
⃗
the evolution of B:
⃗
∂B ⃗ × (B
+∇ ⃗ × ⃗u) = −∇
⃗ × (η ∇
⃗ × B)
⃗ . (4.22)
∂t
The quantity η is the electrical resistivity,
c2
η≡ . (4.23)
4πσ
If η is constant throughout the gas, equation (4.23) may be rewritten in the
form
⃗
∂B
+∇⃗ × (B ⃗ × ⃗u) = η∇2 B
⃗ , (4.24)
∂t
where I have made use of the fact that ∇ ⃗ ·B ⃗ = 0. If ⃗u = 0, then the
⃗ with a diffusion
above equation represents the diffusion of the vector field B,
constant η.
The electric and magnetic fields exert force on the charged particles that
make up an ionized gas. For a completely ionized gas, with ions of charge
Ze, the mean force per unit volume is
⃗ + 1 ⃗ui × B)
f⃗L = Zeni (E ⃗ + 1 ⃗ue × B)
⃗ − ene (E ⃗ . (4.25)
c c
41
The ions have a number density ni and a bulk velocity ⃗ui ; the electrons
have a number density ne and a bulk velocity ⃗ue . Since maintaining charge
neutrality requires that Zeni − ene = 0, and the current density is given by
the relation ⃗je = Zeni⃗ui − ene⃗ue , the Lorenz force per unit volume is
1
f⃗L = ⃗je × B
⃗ (4.26)
c
1 ⃗ ⃗ ×B
⃗ .
= (∇ × B) (4.27)
4π
Using the appropriate identity from vector algebra,
1 ⃗ ⃗ ⃗ ⃗ B2
% &
f⃗L = (B · ∇)B − ∇ . (4.28)
4π 8π
The first term on the right hand side represents a ‘magnetic tension’; it
exists only if the magnetic field lines are curved, and acts in such a way as
to straighten them out. The second term on the right hand side represents
the gradient of the ‘magnetic pressure’, Pm = B 2 /(8π).
The magnetic field also adds a heating term to the internal energy con-
servation equation. An electric current je passing through a medium of re-
sistivity η converts electromagnetic energy to heat energy at the rate
4πη 2 η ⃗ ⃗ 2.
P = 2
je = |∇ × B| (4.29)
c 4π
For the sake of monumental completeness, let’s write down the basic equa-
tions of MHD in all their glory.
Dρ ⃗ · ⃗u
= −ρ∇ (4.30)
Dt
D⃗u ⃗ + ∇· ⃗ ↔ ⃗ + f⃗rad + 1 (∇⃗ × B)
⃗ ×B ⃗
ρ = −∇P π −ρ∇Φ (4.31)
Dt 4π
Dε ⃗ · F⃗ + ψ + Γ − Λ + η |∇
⃗ · ⃗u − ∇ ⃗ × B|
⃗ 2
ρ = −P ∇ (4.32)
Dt 4π
∇2 Φ = 4πGρ (4.33)
⃗
∂B ⃗ × (B
+ ∇ ⃗ × ⃗u) = −∇
⃗ × (η ∇
⃗ × B)
⃗ (4.34)
∂t
⃗ ·B
∇ ⃗ = 0 (4.35)
1 P
ε = . (4.36)
γ−1 ρ
Now that we’ve taken the trouble to write down these complicated equations,
we will ruthlessly simplify them. First we will ignore viscosity and heat
↔
conduction, setting π= 0, ψ = 0, and F⃗ = 0. We will ignore gravity, and set
⃗ = 0. We will ignore radiative effects, and set f⃗rad = 0 and Γ − Λ = 0.
∇Φ
We will ignore diffusion of the magnetic field, and set η = 0.
All of these simplifications will make it possible for us to consider a steady-
state planar shock in the presence of a magnetic field. If a planar shock front
exists, then the velocity vector ⃗u and the magnetic field vector B ⃗ can both be
broken down into components perpendicular to the shock front (designated
by the subscript ⊥) and parallel to the shock front (designated by the sub-
script ∥). Quantities measured just upstream of the shock will be designated
by the subscript ‘1’, as usual, and quantities measured just downstream of
the shock will be designated by the subscript ‘2’. The jump conditions across
the shock are:
ρ1 u⊥1 = ρ2 u⊥2 (4.37)
2 2
2
B∥1 B∥2
ρ1 u⊥1 + P1 + = ρ2 u2⊥2 + P2 + (4.38)
8π 8π
B⊥1 B∥1 B⊥2 B∥2
ρ1 u⊥1 u∥1 − = ρ2 u⊥2 u∥2 − (4.39)
4π 4π
γ P1 u21 B∥1
ρ1 u⊥1 ( + ) − (B⊥1 u∥1 − B∥1 u⊥1 ) = (4.40)
γ − 1 ρ1 2 4π
γ P2 u22 B∥2
ρ2 u⊥2 ( + )− (B⊥2 u∥2 − B∥2 u⊥2 )
γ − 1 ρ2 2 4π
B⊥1 u∥1 − B∥1 u⊥1 = B⊥2 u∥2 − B∥2 u⊥2 (4.41)
B⊥1 = B⊥2 (4.42)
Note that the component of the velocity parallel to the shock is no longer
conserved. The discontinuity in u∥ occurs because there is a current within
the shock front of strength (c/4π)(B∥2 − B∥1 ), which increases u∥ by a factor
B⊥
u∥2 − u∥1 = (B∥2 − B∥1 ) . (4.43)
4πρu⊥
The most mathematically tractable MHD shock is a normal shock (u∥ = 0,
u⊥ = u) in which the magnetic field is parallel to the shock front (B⊥ = 0,
B∥ = B). In this case, the jump relations simplify to the form
ρ 1 u 1 = ρ 2 u2 (4.44)
43
B12 B2
= ρ2 u22 + P2 + 2
ρ1 u21 + P1 + (4.45)
8π 8π
γ P1 1 2 B1 γ P2 1 2 B2
+ u1 + = + u2 + 2 (4.46)
γ − 1 ρ1 2 4πρ1 γ − 1 ρ2 2 4πρ2
B1 u1 = B2 u2 (4.47)
Using these four equations, we can solve for the density jump ρ2 /ρ1 in terms
of the upstream Mach number M1 and the ratio of the magnetic pressure to
the thermal pressure,
B12
α1 ≡ . (4.48)
8πP1
After discarding the trivial solution ρ2 = ρ1 , the four jump relations simplify
to the quadratic relation
% &2 % &
ρ2 ρ2
2(2 − γ)α1 + γ[(γ − 1)M12 + 2(α1 + 1)] − γ(γ + 1)M12 = 0 . (4.49)
ρ1 ρ1
When the magnetic pressure is relatively insignificant (α1 ≪ M12 ) the change
in density is approximately
γ + M12
# $
ρ2 4
= y0 1 − α1 , (4.50)
ρ1 γ [2 + (γ − 1)M12 ]2
where y0 is the density ratio in the absence of a magnetic field. The presence
of a magnetic field thus tends to decrease the density jump from what it
would be in the absence of magnetism. In fact, if we again examine the
quadratic equation for ρ2 /ρ1 , we see that ρ2 > ρ1 when
2
M12 > Mcr2 ≡ 1 + α1 . (4.51)
γ
The existence of a magnetic field permits you to have supersonic motions
(M > 1) without the formation of shocks. Magnetosonic waves (fluctua-
tions in the combined magnetic and gas pressure) can travel ahead of the
shock front. Fluctuations in the gas pressure have a characteristic speed
a ∝ (P/ρ)1/2 . Similarly, fluctuations in the magnetic pressure have a char-
acteristic speed, called the Alfven velocity, uA ≡ B/(4πρ)1/2 ∝ (Pm /ρ)1/2 .
Magnetosonic waves that travel in a direction perpendicular to B ⃗ have a
2 2 1/2 1/2
phase velocity u = (a + uA ) = a[1 + α(2/γ)] .
Chapter 5
Spherical Blastwaves &

Supernova Remnants
One way to create a shock is to inject a very large amount of energy into
a very small volume. This happens in the interstellar medium whenever a
supernova goes off. Suppose an explosion instantaneously injects an amount
of energy E into an ambient medium of uniform density ρ1 . The initial energy
release is considered to take place within an infinitesimally small volume.
Afterward, however, a spherical shock front will expand into the ambient
medium. Early in the course of expansion, the pressure within the shock,
P2 ∼ ρ1 u2sh , is much larger than the ambient pressure P1 and any radiated
energy is much smaller than the explosion energy E. This regime, during
which the energy E remains constant, is known as the blastwave regime. In
a blastwave, the expansion velocity ush (r, t), density ρ(r, t), pressure P (r, t),
and other properties, are determined solely by the two initial parameters of
the system, E and ρ1 .
The energy E has the dimensionality M L2 T −2 ; the density ρ1 has the
dimensionality M L−3 . These two parameters cannot be combined to form
a characteristic length scale or time scale for the problem. The solution for
the expanding shock front must then be a scale-free or self-similar solution.
The self-similar solution is a function of the dimensionless variable ξ, where
ξ ≡ rtl ρm n
1 E ; (5.1)
the exponents l, m, and n are determined by the requirement that ξ be

dimensionless. Since ξ has the dimensionality L1−3m+2n T l−2n M m+n , we see
45
46 CHAPTER 5. SPHERICAL BLASTWAVES & SNRS
that the required solution has the exponents l = −2/5, m = 1/5, and n =
−1/5, or
ρ1 1/5
! "
ξ=r . (5.2)
Et2
When expressed in the dimensionless units, the properties of the expanding
shock front will depend only on ξ. For instance, the radius of the spherical
shock is $1/5
Et2
#
rsh = ξ0 , (5.3)
ρ1
where ξ0 is a factor of order unity (for γ = 5/3, it turns out that ξ0 = 1.17).
The rate of expansion of the shock is
# $1/5
2 E
ush = ξ0 . (5.4)
5 ρ1 t3
Thus,
# $1/2
2 5/2 E −3/2
ush = ξ0 rsh . (5.5)
5 ρ1
The expanding shock wave slows as it expands. Using typical values for
supernova explosions,
# $1/5 # $−1/5 # $2/5
E ρ1 t
rsh = 2.3 pc 51
(5.6)
10 erg 10 g cm−3
−24 100 yr
# $1/5 # $−1/5 # $−3/5
E ρ1 t
ush = 9000 km s−1 51
(5.7)
.
10 erg 10 g cm−3
−24 100 yr
For instance, SN1054 (which gave birth to the Crab Nebula) should have a
radius rsh ∼ 5 pc and expansion velocity ush ∼ 2000 km s−1 , if it is still in the
blastwave phase of expansion. In fact, the Crab Nebula is observed to have
a radius of 3 pc and an expansion velocity of 900 km s−1 . In fact, the Crab is
not exactly spherical and its expansion is not exactly energy-conserving, so
the discrepancy is not surprising.
Obviously, the self-similar blastwave solution is not good for all times.
As t → 0, for instance, the solution predicts infinitely rapid expansion. The
self-similar blastwave solution breaks down when the predicted expansion
velocity is larger than the velocity with which the supernova initially ejects
matter into the ambient medium. For a typical supernova, the initial ejection
47
velocity is uej ∼ 104 km s−1 ; thus, the blastwave solution is invalid when the
supernova remnant is younger than ∼ 70 yr, for our ‘standard’ supernova
parameters.
The self-similar blastwave solution also breaks down when the expanding
shock front is very old. The pressure immediately inside the shock is, for an
extremely strong shock,
2
P2 = ρ1 u2sh (5.8)
γ+1
8 1 5 −3
= ξ Er . (5.9)
25 γ + 1 0 sh
Thus, the pressure within the blastwave falls off rapidly with radius. If we
insert numerical values, we find that
# $1/3 # $−1/3
E P2
rsh ≈ 300 pc 51
. (5.10)
10 erg 4 × 10 dyne cm−2
−13
Thus, when a typical supernova remnant reaches a radius of 300 parsecs, the
internal pressure has dropped until it is comparable to the ambient pressure of
the interstellar medium. This violates our initial assumption that P2 ≫ P1 .
For our typical supernova remnant, the assumption of negligible pressure
breaks down badly at radii of rsh ∼ 300 pc, corresponding to an age of t ∼
30 Myr.
In finding the blastwave solution, however, we made another assumption;
namely, that the energy lost in radiation was small compared to the explosion
energy E. At what point does this assumption break down? The temperature
within the shock will be, for ionized matter of cosmic abundances,
2(γ − 1) m 2
T2 = u (5.11)
(γ + 1)2 k sh
# $# $2/5 # $−2/5 # $−6/5
5 m E ρ1 t
≈ 2 × 10 K (5.12).
10−24 g 51
10 erg 10 g cm−3
−24 105 yr
At an age of 105 yr, the radius of the shock will be rsh ∼ 30 pc. The tem-
perature immediately within the shock will be T2 ∼ 2 × 105 K; the cooling
function at this temperature has the value ρL ∼ 10−21 erg cm−3 s−1 . The
total energy radiated during the course of 105 years will be, very roughly,
4π 3
Erad ∼ ρL( rsh )t ∼ 1051 erg . (5.13)
3
Thus, we expect the energy-conserving blastwave solution to break down

because of radiative losses before it breaks down because of pressure equality.
Once radiative losses become significant, a dense shell forms behind the
radiative shock. The expanding supernova remnant then passes from its
blastwave phase to its snowplow phase. Figure 5.1 illustrates the transition
from the blastwave phase to the snowplow phase. During the ‘snowplow’
phase, the matter of the ambient interstellar medium is swept up by the ex-
panding dense shell, just as snow is swept up by a coasting snowplow. During
the blastwave phase, energy was conserved; during the snowplow phase, mo-
mentum is conserved. Let µ be the total momentum of the expanding shell,
so that a region of solid angle δΩ has momentum µδΩ/(4π). Since the mo-
mentum has dimensionality M LT −1 , the radius in the self-similar snowplow
phase has the dependence
# $1/4
µt
rsh = β0 , (5.14)
ρ1
where β0 is another factor of order unity. The expansion velocity during the
snowplow phase is
# $1/4
1 µ
ush = β0 t−3/4 (5.15)
4 ρ1
# $
1 −2 µ −3
= β0 rsh . (5.16)
4 ρ1
During the snowplow phase, the momentum of the expanding supernova
shell will be of order µ ∼ Msh ush ∼ 2 × 1043 g cm s−1 . The snowplow phase
eventually ends when the expansion speed ush drops to the sound speed a1 =
(γP1 /ρ1 )1/2 of the ambient medium. For our typical supernova remnants, this
happens at a time t ∼ 2 Myr, when the radius of the dense shell is rsh ∼ 60 pc.
At this point, the self-similarity of the solution breaks down, since there is
now a characteristic length scale rc = (E/P1 )1/3 in the problem. During this
last phase of expansion, the shock front degenerates into an acoustic wave,
with a velocity of expansion ush = a1 .
So far, we have found the radius of the shock front as a function of time
for the blastwave and snowplow solutions. However, we do not yet know the
density, velocity, and pressure as a function of radius within the spherical
shock front. A solution for ρ, u, and P within a self-similar blastwave was
first found by the Russian physicist Leonid Sedov in the 1940’s.
49
Figure 5.1: An expanding supernova remnant makes the transition from the
blastwave phase (upper image) to the snowplow phase (lower image). [Shu,
The Physics of Astrophysics, Volume II, Figure 17.4]
The radius of the shock, for the adiabatic blastwave, is rsh = ξo (Et2 /ρ1 )1/5 ,
and the velocity of expansion is ush = (2/5)rsh /t. If the shock is very strong,
the density, bulk velocity and pressure immediately inside the shock front
are found from the Rankine-Hugoniot conditions:
γ+1
ρ2 = ρ1 (5.17)
γ−1
2
u2 = ush (5.18)
γ+1
2
P2 = ρ1 u2sh . (5.19)
γ+1
But now, we ask, what are ρ, u, and P for 0 < r < rsh ?
We can find these quantities by solving the equation of mass conservation,
∂ρ 1 ∂
+ 2 (r2 ρu) = 0, (5.20)
∂t r ∂r
the equation of momentum conservation,
∂u ∂u ∂P
+u =− , (5.21)
∂t ∂r ∂r
and the equation of energy conservation,
∂ 1 1 ∂ 1
(ρε + ρu2 ) + 2 [r2 ρu(ε + P/ρ + u2 )] = 0 , (5.22)
∂t 2 r ∂r 2
where ε = P/[(γ − 1)ρ].
Since the solutions for the blastwave is self-similar, the solutions must
take the form
% &
γ+1
ρ(r, t) = ρ1 α(ξ) (5.23)
γ−1
% &
4 r
u(r, t) = v(ξ) (5.24)
5(γ + 1) t
r2
% &
8
P (r, t) = ρ1 2 p(ξ) . (5.25)
25(γ + 1) t
The factors of ρ1 , r, and t provide the proper dimensionality for the solutions.
The numerical factors in square brackets are inserted so that the dimension-
less density, velocity, and pressure have the normalization α(ξ0 ) = v(ξ0 ) =
51
Figure 5.2: The Sedov solution for a spherical blastwave with γ = 5/3, in
units of the immediate post-shock values. [Shu, The Physics of Astrophysics,
Volume II, Figure 17.3]
p(ξ0 ) = 1. The equations of mass, momentum, and energy conservation can

now be converted into equations involving the functions α(ξ), v(ξ), and p(ξ).
After some numerical manipulation, the mass conservation equation becomes
% &
dα 2 d
−ξ + 2αv + ξ (αv) = 0 . (5.26)
dξ γ+1 dξ
The momentum conservation equation becomes

% & % &
2 dv 4 dv 2(γ − 1) dp
−v − ξ + v 2 + vξ + 2p + ξ = 0 . (5.27)
5 dξ 5(γ + 1) dξ 5(γ + 1)α dξ
The energy conservation equation, finally, becomes

2
−2(p + αv 2 ) − ξ(p + αv 2 ) (5.28)
5 % &
4 d
+ 5v(γp + αv 2 ) + ξ (v[γp + αv 2 ]) = 0 .
5(γ + 1) dξ
Astonishingly, Sedov found an analytic solution to these equations. A plot

of the solutions is given Figure 5.2. As ξ → 0, the density α and the velocity
v go to zero, but the pressure p reaches the limiting value p = 0.306 (when
γ = 5/3). This implies that the temperature (T ∝ P/ρ) goes to infinity as
the radius goes to zero.
The time has come to compare our theoretical results with observations
of actual supernova remnants (SNRs). Supernova remnants can be observed
that correspond to the first three phases of theoretical expansion. The first
phase is the free expansion of the matter of the exploded star. During this
phase, matter moves outward with a constant velocity veject . The remnant
associated with SN1987A is in the free expansion phase. The remnant Cas
A is at the end of its free expansion phase; it is ∼ 300 years old. At visible
wavelengths, many ‘knots’ of emission, moving radially outward with a ve-
locity of ∼ 6000 km s−1 , can be seen. The knots are rich in oxygen, and are
interpreted as being clumps of matter that have been ejected from the center
of the star, where nucleosynthesis took place.
When the amount of interstellar gas swept up becomes comparable to
the initial mass of ejected matter, the SNR enters the blastwave phase. The
remnant of Tycho’s supernova is in the blastwave phase. The Crab nebula is
also in the blastwave phase; its age is ∼ 940 yr, and its expansion velocity is
∼ 900 km s−1 .
When the energy radiated just behind the shock front is comparable to the
initial energy of the explosion, the SNR enters the snowplow phase. During
this phase, a dense cool shell forms directly behind the shock front. In theory,
the momentum of this shell is conserved. Observationally, it is found that
the internal pressure of SNRs in the snowplow phase is large enough to give
a significant push to the shell. The Cygnus Loop (alias the Veil Nebula) is
in the snowplow phase; its age is ∼ 4 × 104 yr, and its expansion velocity
is ∼ 120 km s−1 . The dense shell is thermally unstable to the formation of
filaments; such filaments can be seen in optical photographs of the Cygnus
Loop.
Although the stages of SNR evolution can be fairly well described by our
simple theory, there are many deviations that should be mentioned. First of
all, the progenitors of SNRs are massive hot stars that have copious stellar
winds. Thus, at the time that the supernova goes off, it will be surrounded
by a low-density bubble that has been excavated by the stellar wind. As
long as the supernova ejecta are expanding into this low-density region, they
will sweep up very little mass, and the free expansion phase will be pro-
longed. Another complication is that the ambient interstellar medium is not
uniform. This inhomogeneity will blur the distinction between the blastwave
and snowplow phases; a portion of the shock front that is passing through a
dense cloud may be in the snowplow phase, while a neighboring portion that
is passing through the rarefied intercloud medium is still in the blastwave
53
phase.
It should also be noted that massive stars tend to be born in stellar
associations. They enter their supernova phase before they have a chance
to drift apart. Thus, the remnants of the neighboring supernovas will merge
to form a single superbubble, which may be hundreds of parsecs across.
Such superbubbles are a primary source of the hot ionized component of the
interstellar medium.
Chapter 6
Ionization Fronts, HII Regions,

& Planetary Nebulae
A shock front is the abrupt transition between two regions with different
densities, bulk velocities, and pressures. An ionization front is the abrupt
transition between a region of ionized gas and a region of neutral gas. The
boundary of an HII region is an approximation to a spherical ionization
front.
An HII region is a volume of photoionized gas surrounding a hot young
star (of spectral type O or B). A planetary nebula also consists of a volume
of ionized gas surrounding a central photoionizing star; the central star in a
planetary nebula, however, is an evolved star approaching its death throes.
Planetary nebulae tend to be denser and smaller than HII regions. The
physics of planetary nebulae is complicated by the fact that the photoioniza-
tion phase is preceded by the ejection of a large quantity of neutral gas with
a velocity of ∼ 20 km s−1 . Because the structure of an HII region is simpler,
I will use it as my basic example of a photoionization region.
The gas in an HII region is ionized by photons emitted by the central star.
The ionization energy of hydrogen is 13.6 eV, corresponding to a wavelength
of 912 angstroms. To photoionize significant amounts of hydrogen, the stellar
temperature must be T > ∼ 25,000 K. As the star continues to emit photons,
the HII region becomes larger. Eventually, enough energy is pumped into
the ionized material to raise its temperature and cause it to expand. Thus, a
generic HII region consists of a low-density, high-temperature, ionized region
expanding outward into the ambient high-density, low-temperature, neutral
medium.
55
56 CHAPTER 6. IONIZATION FRONTS
The original classic study of HII regions was performed by Bengt Strömgren
in 1939. He considered a simple model, which still managed to convey the
basic physics of HII regions. We will examine the ‘Strömgren sphere’, as his
model is called, and later examine real HII regions to see how they differ
from the ideal case. Strömgren assumed the presence of a uniform medium,
consisting of neutral atomic hydrogen, with number density n0 . Suddenly,
at time t = 0, a hot star turns on at the origin. The star’s total output
of ionizing photons (in photons per second) is Nu . The initial effect of the
star’s radiation will be solely to ionize the hydrogen. Thus, the star will
be surrounded by a sphere of electrons and protons, with number density
ne = np = n0 . The spherical volume of ionized gas will be separated from
the neutral medium by a thin transition layer, whose thickness is comparable
to the mean free path of an ionizing photon in the neutral medium. The mean
free path λi is given by the relation λi = 1/(n0 σi ), where the cross-section
for photoionization is σi ≈ 6 × 10−18 cm2 for hydrogen atoms.
Initially, the HII region expands outward at a rate given by the relation
dR
Nu = n0 4πR2 , (6.1)
dt
which integrates to
3Nu 1/3
! "
R(t) = t . (6.2)
4πn0
However, an ionized hydrogen atom does not remain ionized forever. The
electrons and protons collide and recombine to form neutral hydrogen at a
rate Nrec = αne ni . The recombination rate α(T ) has a value of α = 2.6 ×
10−13 cm3 s−1 at a temperature T = 104 K, which is a typical temperature
for an HII region. Once recombinations become important, the radius of the
HII region is given by the relation
dR 4π
Nu = 4πn0 R2 + αn20 R3 . (6.3)
dt 3
This has the solution
' (1/3
R(t) = Rs 1 − e−t/ts , (6.4)
where the characteristic time scale is ts = 1/(αn0 ) and the characteristic
length scale is the Strömgren radius,
# $1/3
3Nu
Rs ≡ . (6.5)
4παn20
57
Figure 6.1: The radius of a Strömgren sphere, in units of Rs , as a function

of time, in units of ts .
A plot of the Strömgren sphere’s radius as a function of time is given in

Figure 6.1. At times t ≫ ts , the radius of the HII region will approach the
Strömgren radius. An O5 star has Nu = 5 × 1049 s−1 , and a B0 star has
Nu = 4 × 1046 s−1 . If an O5 star is embedded in a cool neutral cloud with
n0 = 30 cm−3 , it will create an HII region with radius Rs ∼ 4 × 1019 cm ∼
12 pc. The total mass of ionized hydrogen is ∼ 7000 M⊙ . The time to create
the Strömgren sphere is ts ∼ 4000 yr.
The above analysis assumes that the hydrogen gas is static; this is a false
assumption. Although the Strömgren sphere is in ionization equilibrium, it is
not in dynamic equilibrium. The temperature of the neutral medium in which
the HII region will typically be ∼ 100 K. The measured temperatures of HII
regions are ∼ 7000 K. Moreover, the HII region has twice as many particles
per unit volume as the ambient medium, since its hydrogen is completely
ionized into protons and electrons. Thus, the pressure inside the HII region
is ∼ 140 times the pressure outside. The ionized region will expand outward.
Consider a plane parallel ionization front, across which ρ, P , and u, as
well as the degree of ionization, are discontinuous. Let Φi be the flux of

ionizing photons reaching the ionization front. Let ρ1 and P1 be the density
and pressure of the neutral medium, and let u1 be the bulk velocity of the
neutral matter relative to the ionization front. Let ρ2 and P2 be the density
and pressure of the ionized medium, and let u2 be the bulk velocity of the
ionized matter relative to the front. The continuity equation takes the form
ρ1 u1 = ρ2 u2 = mi Φi , (6.6)
where mi is the mean particle mass of the ionized gas. The momentum
conservation equation has the familiar form
P1 + ρ1 u21 = P2 + ρ2 u22 . (6.7)
The energy equation, however, takes a new and unfamiliar form. Our com-
bination ionization & shock front is radiative, but not isothermal. The tem-
perature T1 of the neutral medium is constant. The ionized medium is ra-
diatively cooled until it reaches a constant temperature T2 ̸= T1 . In a fully
ionized medium, the equilibrium temperature that is set by a balance be-
tween heating and cooling is generally T2 ∼ 104 K, considerably higher than
the temperature T1 ∼ 100 K that is typical of the neutral atomic interstellar
medium.
In a gas consisting of pure hydrogen, the energy requirements are
P1 kT1
= (6.8)
ρ1 mH
for the neutral gas, and
P2 2kT2
= (6.9)
ρ2 mH
for the ionized gas. We can define the isothermal sound speeds for the
neutral and ionized media as a1 ≡ (P1 /ρ1 )1/2 and a2 ≡ (P2 /ρ2 )1/2 . Making
use of the equations for the conservation of mass and momentum, the jump
in density across the front can be written in terms of the velocity u1 and the
sound speeds a1 and a2 . A bit of algebraic manipulation tells us
ρ2 1 2 2
)
= 2 [a1 + u1 ± f (u1 )] , (6.10)
ρ1 2a2
where f (u1 ) = (a21 + u21 )2 − 4a22 u21 . The function f (u1 ) can also be written in
the form
f (u1 ) = (u21 − u2R )(u21 − u2D ) (6.11)
59
where
)
uR ≡ a 2 + a22 − a21 (6.12)
)
uD ≡ a 2 − a22 − a21 . (6.13)
Physical reality demands that the ratio ρ2 /ρ1 be a real number. Thus, the
velocity of the ionization front (relative to the neutral gas) must have either
u1 ≥ uR or u1 ≤ uD . The rapidly propagating ionization fronts (u1 ≥ uR ) are
called R-type fronts; the slowly propagating ionization fronts (u1 ≤ uD )
are called D-type fronts. In this naming scheme, ‘R’ stands for ‘rarefied’
and ‘D’ stands for ‘dense’.1
In HII regions, it’s a fair approximation that a2 ≫ a1 , so uR ≈ 2a2 and
uD ≈ (a1 /a2 )2 a2 /2. An R-type front always has u1 > uR > a2 > a1 , and is
supersonic with respect to the neutral medium. A D-type front always has
u1 < uD < a1 < a2 , and is subsonic with respect to the neutral medium.
For a given propagation velocity u1 , there are two possible values of ρ2 /ρ1 ,
corresponding to taking the plus or the minus sign in the solution of the
quadratic equation. This is shown graphically in Figure 6.2, which is a plot
of the ratio ρ2 /ρ1 as a function of u1 . A front that has the larger density
contrast is called a strong front; a front that has the smaller density contrast
is a weak front. Thus, there are four types of ionization front possible; a
weak R-type front, a strong R-type front, a weak D-type front, and a strong
D-type front. It turns out that strong R-type fronts are unstable, and are not
seen in nature. Weak R-type fronts and both strong and weak D-type fronts
are known to exist, however, around HII regions and planetary nebulae.
In a weak R-type front, the incoming neutral gas is supersonic with re-
spect to the front (u1 > a1 ), and the outflowing ionized gas is also supersonic
with respect to the front (u2 > a2 ). In a strong R-type front, the incoming
neutral gas is supersonic, and the outflowing ionized gas is subsonic. In a
weak D-type front, the incoming neutral gas is subsonic, and the outflowing
ionized gas is also subsonic. In a strong D-type front, the incoming gas is
subsonic, and the outflowing ionized gas is supersonic.
Consider, for example, an expanding HII region. When the ionizing star
is first ‘turned on’, the spherical ionization front is very small; hence the flux
Φi is very large, resulting in a high velocity u1 . Thus, the front is initially
1
For a given flux Φi , as the gas becomes more rarefied (ρ1 → 0), the velocity u1 → ∞,
and an R-type front results. Similarly, as ρ1 → ∞, u1 → 0, and a D-type front results.
Figure 6.2: Physically permissible values of the density contrast ρ2 /ρ1 across
an ionization front with a2 = 10a1 .
61
a weak R-type front. In the limit that u1 ≫ a2 ≫ a1 , the ratio of densities

inside and outside the HII region is
ρ2 a2
≈ 1 + 22 weak R . (6.14)
ρ1 u1
A weak R-type front compresses the gas only slightly. This period, when
the interior and exterior densities are nearly the same, is when Strömgren’s
approximation holds good.
As the ionization front surrounding the HII region expands, the flux of
ionizing photons Φi steadily decreases. Eventually, the velocity u1 drops to
a value u1 = uR . At this point, the front is an ‘R-critical’ front; the density
ratio is ρ2 /ρ1 ≈ 2 and the velocity of the ionized gas is u2 ≈ a2 . Once
the ionization front slows still farther, the R-type front can no longer exist.
What happens next is illustrated in Figure 6.3. When the velocity u1 drops
below uR , the R-critical front splits into a shock front followed by a D-critical
front. The shock front increases the density of the gas (by a factor of 4 if
M1 ≫ 1 and γ = 5/3), decreases the velocity of the gas, and increases the
sound speed. Because of the decrease in the bulk velocity and the increase in
the sound speed, the flow of neutral gas is now subsonic, and is ready to pass
through a D-type front that ionizes it. As the HII region expands farther,
the leading shock front gradually weakens, and the trailing D-critical front
develops into a weak D-type front. Thus, a bit of neutral gas will be first
compressed by the shock and then blasted with UV photons and ionized.
The shock front expands outward and decreases in amplitude until it fi-
nally becomes a sound wave of infinitesimal amplitude expanding outward
with velocity u1 = a1 . If the photoionizing star is immortal, then the ion-
ization front will expand more and more slowly until HII region attains an
equilibrium state in which the pressure of the expanded ionized gas is equal
to the pressure of the neutral surrounding medium. The final equilibrium
radius, Rfinal , is given by the relation
# $1/3
3Nu
Rfinal = , (6.15)
4παn2final
where the density nfinal within the HII region is given by the requirement of
pressure equilibrium:
2nfinal T2 = n0 T1 . (6.16)
Figure 6.3: The transition from an R-critical front to a shock front followed
by a D-critical front.
63
Thus,
"2/3
2T2
!
Rfinal = Rs ∼ 27Rs . (6.17)
T1
In practice, however, HII regions never reach this equilibrium state. The
massive O and B stars that ionize them become supernovae before then.
Observers classify real HII regions into five categories that correlate well
with the age of the region. An ultracompact HII region contains a single,
very young, hot star. Ultracompact HII regions are embedded deep within
molecular clouds, and can be observed only in the radio and the infrared.
They have radii R < ∼ 0.1 pc and densities n >
4 −3
∼ 10 cm . An example of
an ultracompact HII region is the Becklin-Neugebauer object, located in a
molecular cloud near the Orion nebula.
As the HII region expands, it becomes a compact HII region. A compact
HII region contains a single, young, hot star; it has broken out of its molecular
cocoon, so it can be seen at optical wavelengths. Compact HII regions have
radii R ∼ 0.5 pc and densities n ∼ 5 × 103 cm−3 . An example of a compact
HII region is the Orion nebula.
A standard HII region contains a single hot star; it typically forms a
‘blister’ on the surface of a molecular cloud. On one side, the expanding
ionized gas of the HII region is confined by the dense molecular cloud; on the
other side, it pushes out into the intercloud medium. Standard HII regions
have R ∼ 3 pc and n ∼ 300 cm−3 . An example of a standard HII region is
the Omega nebula.
A large HII region contains several hot stars; it consists of several stan-
dard HII regions that have merged together. Large HII regions have R ∼
10 pc and n ∼ 50 cm−3 . An example of a large HII region is the North
America nebula.
Finally, a giant HII region contains as many as 104 hot stars; it consists
of numerous large HII regions plus a common ionized envelope. Giant HII
regions have R > ∼ 100 pc and n <
−3
∼ 10 cm . A famous example of a giant HII
region is 30 Doradus in the Large Magellanic Cloud.
Planetary nebulae contain one hot star apiece, but they show an increas-
ing radius and decreasing density similar to that of HII regions. The youngest
visible planetary nebulae have R ∼ 0.03 pc and n ∼ 104 cm−3 . The oldest
planetary nebulae have R ∼ 1 pc and n ∼ 10 cm−3 . The progenitor of a
planetary nebula is a star of initial mass 1 − 8 M⊙ that has evolved onto the
asymptotic giant branch (AGB). At the end of its AGB stage, such a star
Figure 6.4: A schematic luminosity – temperature diagram for the evolution

of stars destined to become planetary nebulae.
ejects its outer envelope at the relatively low velocity of 20 − 30 km s−1 . A

central core of mass 0.6 − 1.4 M⊙ is left behind. This central core, which
evolves into a white dwarf, is the photoionization source for the planetary
nebula. The evolution of the central stars of planetary nebulae is given in
Figure 6.4.
The central photoionizing core has a temperature of 25,000 K → 200,000 K,
and a luminosity 10 L⊙ → 10,000 L⊙ . The ejected matter that is photoion-
ized by the central core is initially cold and neutral, and consists partly of
molecules. (Thus, the ionization front in a planetary nebula is preceded by
a dissociation front.) Initially, the ionized matter surrounding the central
core consists entirely of the matter that has been ejected from the giant pro-
genitor. However, as the ionization front expands, it reaches the boundary of
the ejected matter, and reaches the lower-density interstellar matter beyond.
The ionization front races outward into the low-density surrounding matter.
Eventually, however, the temperature of the central white dwarf drops below
25,000 K and fails to maintain the ionization of the surrounding matter. The
planetary nebula gradually recombines, and dims into invisibility.
Chapter 7
Basic Turbulence
The universe is a highly turbulent place, and we must understand turbulence

if we want to understand a lot of what’s going on. Interstellar turbulence
causes the “twinkling” of radio sources, just as turbulence in the earth’s
atmosphere causes the twinkling of stars. Turbulence in stellar atmospheres
strongly modifies the structure of the star; in fact, the turbulent heat flux
can be larger than the conductive heat flux F⃗ and the turbulent stress tensor
↔
can be larger than the viscous stress tensor π. Turbulent motions are seen
in HII regions, in molecular clouds, and in jets.
The first difficulty in dealing with turbulence is providing an adequate
definition. According to Webster’s New Collegiate Dictionary, turbulence is
“departure in a fluid from a smooth flow”. Turbulence is characterized by
the presence of irregular eddying motions – that is, motions in which the
vorticity ω⃗ ≡∇ ⃗ × ⃗u is non-zero. An image of turbulent flow in a terrestrial
laboratory is given in Figure 7.1. Usually, a turbulent flow has a spectrum
of eddy sizes.1
In a turbulent flow, the fluid velocity ⃗u(⃗x, t) at a fixed point ⃗x varies
with time in a nearly random manner, as shown in the data of Figure 7.2.
From an alternative viewpoint, the velocity at a fixed time t varies with
position in a nearly random manner. Turbulence is a chaotic process; a
small change in the initial conditions ⃗u(⃗x, t0 ) results in a large change in the
conditions at a later time t. Since we cannot examine the development of the
turbulent velocity field in a deterministic manner, we are reduced to studying
the statistical properties of the turbulence.
1
As it says in the poem by Lewis F. Richardson, ‘Big whirls have little whirls / Which
feed on their velocity, / And little whirls have lesser whirls / And so on to viscosity.’
65
66 CHAPTER 7. BASIC TURBULENCE
Figure 7.1: Laser-induced fluorescence image of an incompressible turbulence

boundary layer. [University of Iowa Fluids Laboratory]
Figure 7.2: Turbulent velocities within the Earth’s atmosphere, measured

18 m above the ground. Time and velocity are in arbitrary units.
67
A flow that is not turbulent, in which the velocity varies smoothly and
predictably in space and time, is known as a laminar flow. Many flows
can be broken down into a laminar and a turbulent component. Note that
the random turbulent component of a fluid’s motion is distinct from the
random thermal component. The thermal component consists of the random
Brownian motion of the individual gas particles; each particle moves in a
straight line for a distance ∼ λ before an encounter with another particle
sends it off in a different direction. The turbulent component consists of the
eddying motion of macroscopic fluid elements; the eddies have a range of
sizes, but all eddies are very much larger than λ.
By distinguishing between the laminar component and the turbulent com-
ponent, we are separating the bulk velocity ⃗u of a gas into two parts:
⃗ (⃗x) + ⃗u′ (⃗x) .
⃗u(⃗x) = U (7.1)
where the laminar component is U ⃗ = ⟨⃗u⟩ and the turbulent component is

′
⃗u = ⃗u − ⟨⃗u⟩. When I use the symbol ⟨⃗u⟩, I am referring to the average
value of ⃗u – but what does it mean, in this context, to take the average of
the velocity? In practice, what we must do is calculate the spatial average
within a volume V ,
1 !
⟨⃗u⟩ = ⃗ud3 x , (7.2)
V V
or the time average during an interval τ ,
1 ! t+τ
⟨⃗u⟩ = ⃗udt . (7.3)
τ t
The Taylor hypothesis states that for fully developed turbulence, the spa-
tial average and the time average are equivalent.
Now, after removing the laminar velocity U ⃗ , we are left with the turbu-
′
lent velocity ⃗u . We will make things easier for ourselves by placing some
restrictions on the form of the turbulent velocity field. First, we will assume
that the turbulence is homogeneous and isotropic; that is, the statistical
properties of ⃗u′ are independent of position and direction. In addition, we
assume that the turbulence is incompressible; that is, the density of the
turbulent fluid is the same everywhere. This assumption is an extraordinarily
poor one for the interstellar medium, in which the density varies wildly from
place to place. However, the case in which the fluid is incompressible has the
great advantage of simplicity; if we assume incompressibility, we can actually
find a few analytic results! These results will give us some insight into the
physics of the problem, which we can then apply to the more realistic case
of a self-gravitating, radiating, compressible fluid.
In an incompressible fluid, the continuity equation for the turbulent flow
simplifies to the form
∇⃗ · ⃗u′ = 0. (7.4)
The equation for momentum conservation is
∂⃗u′ ⃗ u′ = 1 ∇P
⃗ + ν∇2⃗u′ .
+ (⃗u′ · ∇)⃗ (7.5)
∂t ρ
I am assuming that the fluid exhibits Newtonian viscosity, and that the
kinematic viscosity, ν ≡ µ/ρ, is constant. The kinematic viscosity of air is
ν ≈ 0.15 cm2 s−1 at sea level and at room temperature. For neutral atomic
hydrogen, the kinematic viscosity is
#1/2 " #−1
T n
"
ν ∼ 4 × 1021 cm2 s−1 . (7.6)
104 K 1 cm−3
The kinematic velocity is approximately equal to the sound speed in a fluid
times the mean free path λ.
We can define a velocity correlation tensor for the turbulent velocity
in the following manner:
Rij (r) ≡ ⟨u′i (⃗x)u′j (⃗x + ⃗r)⟩ . (7.7)
Since the turbulent velocity field is homogeneous and isotropic (by assump-
tion), the correlation tensor is a function only of the distance r between the
two points and not on their location ⃗x within the velocity field. In an ideal
universe, we could measure the velocity vector everywhere within the tur-
bulent flow, subtract away the laminar component, and then compute the
correlation tensor Rij exactly. Unfortunately, this is difficult to do even in-
side a laboratory. Astronomers are even more badly hampered; usually the
best they can get is a line-of-sight velocity profile, which adds together the
radial velocities of all the matter along a given line of sight.
The trace of the velocity correlation tensor is
$
R(r) ≡ Rii (r) . (7.8)
i
69
In the limit that r → 0, R(r) → ⟨|⃗u′ |2 ⟩. The trace can be used to define a
correlation length
1 !∞
Λt ≡ R(r)dr . (7.9)
R(0) 0
The correlation length, roughly speaking, is the size of the biggest eddies in
the turbulent flow.
Since turbulence exists with a range of eddy sizes, it is frequently conve-
nient to take the Fourier transform of the velocity field in order to consider
the Fourier components of different wavenumber. The components of the
velocity in Fourier space will be
1 ! ′ ⃗
⃗u⃗k = 3
⃗u (⃗r)e−ik·⃗r d3 r . (7.10)
(2π)
The Fourier transform of the velocity correlation tensor is the energy spec-
trum tensor,
⃗ 1 ! ⃗
Φij (k) = 3
Rij (⃗r)e−ik·⃗r d3 r . (7.11)
(2π)
The spectrum Φij (⃗k) tells how much kinetic energy is contained in eddies with
wavenumber k. The tensors Rij and Φij both contain the same information
about the field; which tensor you use depends merely on whether it is more
convenient to work in real space or Fourier space.
In a homogeneous isotropic turbulent flow, it is possible, and also useful,
to define an energy spectrum function E(k) such that
E(k) ≡ 2πk 2 Φii (⃗k) .

$
(7.12)
i
The total turbulent kinetic energy per unit mass is then

1 ′2 ! ∞
⟨|⃗u | ⟩ = E(k)dk . (7.13)
2 0
In a Newtonian fluid, the rate at which the turbulent kinetic energy is dissi-
pated by viscosity is ! ∞
ϵd = 2ν k 2 E(k)dk . (7.14)
0
Thus, the dissipation occurs at high wavenumbers – it is the smallest eddies

present that dissipate the turbulent energy through viscosity.
For a turbulent flow to remain in a steady state, turbulent energy must be

added at the largest scales at the same rate ϵd at which it is being dissipated
at the smallest scales. If additional energy is not added, the turbulence
will gradually decay. In the earth’s atmosphere, for instance, the turbulence
is maintained by the addition of solar energy, which adds energy to the
atmosphere at the time-averaged rate ϵd ∼ 10 erg g−1 s−1 .
In 1941, Kolmogorov predicted that the velocity field in incompressible
turbulent flows would be self-similar over a range of velocity scales.2 The
self-similarity, he stated, results from a dissipationless cascade of energy from
large scales (the big whirls) to small scales (the little whirls). The average
velocity difference between two points, in the Kolmogorov theory, is a func-
tion only of the distance r between the points and the energy dissipation rate
per unit mass ϵd .
The fully developed turbulent medium is characterized by only two quan-
tities, the mean rate of energy dissipation, ϵd , and the kinematic viscosity ν.
The dimensionality of ϵd is energy/time/mass, or L2 T −3 , and the dimension-
ality of ν is L2 T −1 . We can combine these two quantities to find the length
scale % &1/4
ν3
lK = , (7.15)
ϵd
known as the Kolmogorov, or dissipational, length scale. The Kolmogorov
length scale is the size of the smallest eddies in the fluid. Eddies smaller
than the Kolmogorov scale rapidly dissipate their kinetic energy by viscous
heating and disappear. The eddies of size lK rotate with a velocity
uK = (νϵd )1/4 , (7.16)
and dissipate their energy in a time roughly equal to
#1/2
ν
"
τK = . (7.17)
ϵd
In the Earth’s atmosphere, ϵd ∼ 10 cm2 s−3 and ν ∼ 0.1 cm2 s−1 . Thus,
the smallest eddies in the Earth’s atmosphere have lK ∼ 0.1 cm, which is
significantly longer than the mean free path λ ∼ 10−4 cm. The smallest
eddies swirl around with a velocity uK ∼ 1 cm s−1 , dying away with a decay
time of τK ∼ 0.1 s.
2
His results were published in the Proceedings of the Soviet Academy of Sciences, and
were initially neglected; Russians had other things to worry them during the year 1941.
71
Kolmogorov, in making his assumption of self-similar turbulence, saw

that the power spectrum of the turbulence should have the form
E(k, t) = u2K lK E∗ (lK k) , (7.18)
where E∗ is a dimensionless function of the dimensionless wavenumber lK k.

1/4
The factor u2K lK = ν 5/4 ϵd in front gives the energy spectrum its proper
dimensionality. When the wavenumber k of a turbulent element lies in the
range Λ−1 −1
t ≪ k ≪ lK , then it is in the inertial subrange, in which negligible
dissipation occurs, and the dominant energy process is the transfer of kinetic
energy from large eddies to smaller eddies by inertial forces. In the inertial
subrange, the power spectrum must be scale free, so that
E∗ (lK k) = α(lK k)n (7.19)
is a pure power law (with α being our old friend, the “factor of order unity”).
Since viscous forces are negligible on this scale, the power spectrum E(k)
must also be independent of the value of the kinematic viscosity ν. Since
1/2 −1/4
u2K = ν 1/2 ϵd and lK = ν 3/4 ϵd , we find
1/2 −1/4 −n/4 n
E(k) = ν 1/2 ϵd ν 3/4 ϵd αν 3n/4 ϵd k . (7.20)
For the dependence of E on ν to vanish, we require n = −5/3, and

2/3
E(k) = αϵd k −5/3 . (7.21)
This spectrum, of the form E ∝ k −5/3 , is referred to as the Kolmogorov

spectrum, and is a pretty good fit for many turbulent flows on scales be-
tween the correlation length Λt and the Kolmogorov length lK . A schematic
plot of the Kolmogorov spectrum is shown in Figure 7.3.
If u(l) is the velocity of eddies with size l, dimensional analysis tells us
that
2/3
E(k) ∼ lu(l)2 ∼ ϵd l5/3 (7.22)
and hence that
1/3
u(l) ∼ ϵd l1/3 = uK (l/lK )1/3 . (7.23)
On scales larger than lK the Reynolds number scales as Re ∝ l4/3 , so on large
scales dissipation is negligible.
Figure 7.3: The Kolmogorov spectrum for incompressible turbulence.
In the Earth’s atmosphere, the inertial range stretches from lK ∼ 0.1 cm

to the correlation length Λt ∼ 1 km. Within that range, the typical turbulent
velocity is
% &1/3
−1 l
u(l) ≈ 10 cm s . (7.24)
1m
The smallest eddies, with l ∼ lK ∼ 0.1 cm, take 0.1 seconds to whirl around.
The largest eddies, with l ∼ Λt ∼ 1 km, take 1000 seconds to whirl around.
It is these turbulent eddies that cause ‘twinkling’ and ‘seeing’.
Does the Kolmogorov spectrum apply (even approximately) to the inter-
stellar medium, which is most distinctly not an incompressible fluid? For in-
stance, it has been proposed that turbulence occurs within molecular clouds.
The temperature of molecular clouds is usually about 10 K. If the emission
lines in the spectra of molecular clouds are broadened only by the thermal
motions, they should have a width of ∆v ∼ 0.3 km s−1 . In fact, the measured
widths are greater than this value. For instance, in 1981, Richard Larson
found that the relation
% &0.38
−1 l
∆v ≈ 1.1 km s (7.25)
1 pc
gave a good fit to the measured velocity dispersions of molecular clouds.
Larson’s results are plotted in Figure 7.4. This observationally determined
relation is close to the ∆v ∝ l1/3 relation expected for a Kolmogorov spectrum
of turbulence.
73
Figure 7.4: The internal velocity dispersion σ plotted versus the maximum
linear size L of molecular clouds. (The different letter symbols represent
different molecular cloud complexes.) [Larson, 1981, MNRAS, 194, 809]
If the relation were actually ∆v ∼ 1 km s−1 (l/1 pc)1/3 , and the molecular
gas were incompressible, that would imply ϵd = (∆v)3 /l ∼ 3×10−4 cm2 s−3 ∼
2 × 10−4 L⊙ / M⊙ . The kinematic viscosity of a dense molecular cloud is
ν ∼ 1017 cm2 s−1 , implying a Kolmogorov length lK ∼ 4 × 1013 cm ∼ 3 AU.
The velocity of the smallest eddies would be uK ∼ 2 × 103 cm s−1 and they
would whirl around on a timescale τK ∼ 2 × 1010 s ∼ 600 yr.
However, treating molecular clouds as if they were incompressible fluids
with a Kolmogorov spectrum of turbulence is incredibly naı̈ve. Most studies
of molecular clouds give the relation
% &0.5
l
∆v ≈ 1 km s−1 (7.26)
1 pc
for the line width on a scale l; the results from one study are shown in
Figure 7.5. In addition, the average density of a molecular cloud scales as
% &−1.2
−3 l
n ≈ 4000 cm . (7.27)
1 pc
Thus, the assumption of a constant density, incompressible fluid is a poor
one. In addition, the turbulent velocities are larger than the sound speed
Figure 7.5: Velocity dispersion as a function of linear size for a sample of 273
molecular clouds within our galaxy. [Solomon et al., 1987, ApJ, 319, 730]
75
a ∼ 0.3 km s−1 ; therefore, it is highly likely that shocks will form. Further-
more, energy will be added to the molecular clouds not only on the correla-
tion length Λt , but also on smaller scales (as the result of supernovae, stellar
winds, and other processes). Also, in a realistic view of interstellar turbu-
lence, the effects of gravity and magnetism must be taken into account. In
view of these complications, people who study turbulence in the ISM must
resort to numerical simulations. The simulation shown in Figure 7.6, for
instance, shows the large density contrasts typical of realistic simulations of
compressible turbulence.
Figure 7.6: Gas density from a 10243 adaptive mesh simulation of supersonic
compressible turbulence (darker regions are denser). [Kritsuk et al., 2006,
ApJ, 638, L25]
Chapter 8
Spherical Accretion
Accretion may be defined as the gravitational attraction of material onto a

compact object. The compact object may be a black hole with a Schwarzschild
radius R∗ = 2GM∗ /c2 ∼ 3 km(M∗ / M⊙ ). Another possible compact object
is a neutron star, with mass M∗ ∼ 2 M⊙ and radius R∗ ∼ 10 km. Yet an-
other possible compact object is a white dwarf, with mass M∗ ∼ 1 M⊙ and
radius R∗ ∼ 104 km. The accretion of gas onto a compact object can be a
very efficient way of converting gravitational potential energy into radiation.
Consider a compact object of mass M∗ and radius R∗ . If a blob of hydrogen
with mass m is allowed to drop onto the compact object, starting at infinity,
the amount of energy released is
GM∗ m RSch 2
Eacc = = mc , (8.1)
R∗ 2R∗
where RSch is the Schwarzschild radius of the mass M∗ . By comparison,
the conversion of the hydrogen into helium would yield an energy Enuc =
0.007mc2 . Thus, accretion accretion onto a compact object with R∗ < ∼
70RSch , such as a black hole or neutron star, is a more efficient mechanism
than nuclear fusion of H to He.
There is a limit to the rate Ṁ at which a compact object can accrete
matter. Suppose the infalling matter consists of ionized hydrogen. The
luminosity L of the central compact object exerts a radiation force on the free
electrons by Thomson scattering. The outward radial force on an electron
at a radius r is fout = σT L/(4πr2 c), where σT = 6.7 × 10−25 cm2 is the
Thomson cross section of the electron. As each electron moves outward, it
drags a proton along with it in order to conserve charge neutrality. The
79
80 CHAPTER 8. SPHERICAL ACCRETION
inward gravitational force on the electron - proton pair is fin = GM∗ mp /r2 .
There exists a limiting luminosity, called the Eddington luminosity, at
which the two forces cancel:
! "
4πGM∗ mp c M∗
LEdd = = 3.4 × 104 L⊙ . (8.2)
σT M⊙
When the luminosity of the central compact object is greater than this value,
the surrounding hydrogen gas will be blown away by the radiation pressure.
If the Eddington luminosity is emitted as black-body radiation, the temper-
ature will be ! "1/4 ! "1/4
LEdd GM∗ mp c
Tbb = = . (8.3)
4πR∗2 σSB R∗2 σSB
where σSB is the Stefan-Boltzmann constant, and R∗ is the radius of the
surface from which the radiation is emitted (for a black hole, of course,
this surface will be outside the Schwarzschild radius). For a black hole ac-
creting at the Eddington limit, the temperature of the radiation will be
Tbb ∼ 4 × 107 K(M∗ / M⊙ )−1/4 , if the radiation comes from immediately out-
side the Schwarzschild radius. The spectrum of the emitted photons will
then peak at a photon energy E ∼ 20 keV(M∗ / M⊙ )−1/4 .
The existence of the Eddington luminosity implies the existence of a max-
imum accretion rate, ṀEdd , for an accreting compact object. If the accre-
tion energy Eacc is converted entirely into radiation, then the luminosity is
Lacc = GM∗ Ṁ /R∗ , and the maximum possible accretion rate is
4πmp cR∗ R∗ R∗
ṀEdd = = 9×1016 g sec−1 ( ) = 1×10−3 M⊙ yr−1 ( ) . (8.4)
σT 1 km R⊙
In reality, the conversion is not 100% efficient, the accretion is not perfectly
spherically symmetrical, and the radiation is not perfectly spherically sym-
metrical; thus, matter can be accreted at rates somewhat greater than ṀEdd .
Quasars and active galactic nuclei can have luminosities as great as 1014 L⊙ .
If they are powered by accretion onto a central compact object, its mass must
be M∗ > ∼ 3 × 10 M⊙ , with a Schwarzschild radius RSch >
9
∼ 8 light-hours. The
X-ray variability of bright AGNs and quasars is on the scale of hours, suggest-
ing that the central accreting object must be a black hole near its Eddington
luminosity; otherwise its radius and variability timescale would be too large.
If the central object is a black hole, the rate at which it must accrete matter in
order to radiate with L = 1014 L⊙ is Ṁ = L/(c2 η) ∼ (η −1 )7 M⊙ yr−1 , where
81
η is the efficiency with which the accretion energy is converted to radiation.

The most luminous X-ray binaries within our own galaxy have luminosities
of about 3×104 L⊙ , implying a mass of 1 M⊙ if they are powered by accretion
at the Eddington limit.
Now, let us consider the dynamics of spherically symmetric accretion. A
compact object of mass M is embedded in the ISM. Far from the accreting
compact object, the medium has a uniform density ρ∞ and a uniform pressure
P∞ ; the sound speed far from the accreting object thus has the value a∞ =
(γP∞ /ρ∞ )1/2 . If the infall of matter has reached a steady state, then the
equation of mass conservation reduces to the form Ṁ = −4πr2 ρu, where Ṁ
is the rate of accretion onto the compact object. Ignoring the self-gravity of
the accreting gas, the equation of momentum conservation is
du a2 dρ GM
u + + 2 =0. (8.5)
dr ρ dr r
From the equation of mass conservation, we know that
1 dρ 2 1 du
=− − . (8.6)
ρ dr r u dr
Substituting this result into the momentum conservation equation, we find
the result
a2 d 2 2a2 r
! " # $
1 GM
1− 2 (u ) = − 2 1 − . (8.7)
2 u dr r GM
This equation is called the “Bondi equation”, after Hermann Bondi, who pi-
oneered the study of spherical accretion. It represents spherically symmetric,
steady state accretion of a non-self-gravitating gas.1
Now consider the term in square brackets on the right hand side of the
Bondi equation, above. As r → ∞, the sound speed a approaches the finite
value a∞ . Thus, the term in square brackets is negative at large radii. As
r → 0, however, the term in square brackets tends to increase. At some
radius rs , it will equal zero, unless some form of heating raises the sound
speed to values a(r)2 > GM/(2r). Let us assume that there is a radius rs
at which 2a(rs )2 rs = GM , and the right hand side of the Bondi equation
vanishes. At the radius rs , the left hand side of the equation must also
vanish. This means that either
u(rs )2 = a(rs )2 (8.8)
1
Bondi, 1952, MNRAS, 112, 195
Figure 8.1: Solutions to the Bondi equation, divided into six families.
or
d(u2 ) %%
=0. (8.9)
dr r=rs
%
There are six families of solutions to the Bondi equation, characterized by

their behavior at the radius rs and in the limits r → ∞ and r → 0. These
families are illustrated in Figure 8.1.
Type 1: u(rs )2 = a(rs )2 , u2 → 0 as r → ∞ . (8.10)
Type 2: u(rs )2 = a(rs )2 , u2 → 0 as r → 0 . (8.11)

The type 1 and type 2 solutions are known as the transonic solutions; in
these solutions the radius rs is known as the sonic point.
d(u2 ) %%
Type 3: =0, u2 < a2 . (8.12)
dr r=rs
%
d(u2 ) %%
Type 4: =0, u2 > a2 . (8.13)
dr r=rs
%
83
The type 3 and type 4 families of solution represent flow that is everywhere
subsonic (type 3) or everywhere supersonic (type 4).
Type 5: u(rs )2 = a(rs )2 , r > rs . (8.14)
Type 6: u(rs )2 = a(rs )2 , r < rs . (8.15)

The type 5 and type 6 families of solutions represent solutions that are math-
ematically permissible but physically impossible. They are double-valued,
giving two values of u2 at a given value of r. Since the bulk velocity u must
have a unique value at every point, we exclude families (5) and (6) as possible
solutions for spherical accretion. Moreover, for the case of spherical accre-
tion, we want solutions for which u2 → 0 as r → ∞. We thus exclude family
(4) and solution (2) from consideration, since they have supersonic infall at
large radius. This leaves us with two possible solutions. The solutions of type
(3) represent an accretion flow that is subsonic at all radii; they consist of a
gentle infall that gradually settles into hydrostatic equilibrium at small radii.
In most cases, accretion is not gentle; the accreted matter comes zooming in
at supersonic speeds at small radii. Therefore, we will focus our attention
on solution (1), which represents an accretion flow that is subsonic at r > rs
and supersonic at r < rs .
If the accretion is transonic, of type (1), then we can uniquely determine
the accretion rate Ṁt in terms of the mass M of the accreting object and
the density ρ∞ and the sound speed a∞ at infinity. Let us suppose that the
infall is adiabatic, with P ∝ ργ . The equation of momentum conservation
can then be integrated to yield the result
u(r)2 a(r)2 GM a2
+ − = ∞ . (8.16)
2 γ−1 r γ−1
This quantity is known as the Bernoulli integral. Evaluating the Bernoulli

integral at the sonic point rs tells us that
a2
! "
1 1
2
a(rs ) + −2 = ∞ , (8.17)
2 γ−1 γ−1
or ! "1/2
2
a(rs ) = a∞ . (8.18)
5 − 3γ
This implies that

5 − 3γ GM
rs = (8.19)
4 a2∞
and ! "1/(γ−1)
2
ρ(rs ) = ρ∞ . (8.20)
5 − 3γ
When the adiabatic index is γ = 5/3, the accreting gas doesn’t reach the
sonic point until rs = 0.
Using the relation Ṁ = 4πrs2 ρ(rs )a(rs ), we find that the transonic accre-
tion rate for the type (1) solution is
G 2 M 2 ρ∞
Ṁt = 4πqs , (8.21)
a3∞
where ! "(5−3γ)/(2γ−2)
1 2
qs (γ) = . (8.22)
4 5 − 3γ
The numerical value of qs ranges from qs = 1/4 at γ = 5/3 to qs = e3/2 /4 ≈
1.12 when γ = 1. If the accreting matter is ionized hydrogen, the transonic
accretion rate has the value
! "2 ! "& '−3/2
10 −1 M ρ∞ T∞
Ṁt = 1.2 × 10 g sec . (8.23)
M⊙ 10 g cm−3
−24 104 K
This amounts to about 10−16 M⊙ yr−1 for a 1 M⊙ accreting mass. The Ed-
dington accretion rate for a black hole is ṀEdd ∼ 3 × 1017 g sec−1 (M/ M⊙ ), so
an isolated black hole in the ISM is in little danger of exceeding its Eddington
accretion rate.
The relation between the bulk velocity u(r) and the sound speed a(r) can
be computed from the equation
! "2/(γ−1)
Ṁ Ṁ a∞
−u = = . (8.24)
4πr2 ρ(r) 4πr2 ρ∞ a(r)
Thus "−2/(γ−1)
qs G2 M 2
!
a(r)
u=− 2 3 , (8.25)
r a∞ a∞
85
or "−2/(γ−1)
'−2 !
u qs r a(r)
&
=− , (8.26)
a∞ 4 ra a∞
with ra ≡ 2GM/a2∞ . The “accretion radius” ra is the radius at which the
density ρ and sound speed a start to significantly increase above their ambient
values of ρ∞ and a∞ . The relation between the sonic radius and the accretion
radius is rs = [(5−3γ)/8]ra . At large radii (r ≫ ra ), the infall velocity, sound
speed, and density of the transonic flow are
q s a∞r −2 1 ra
& ' ( )
u ≈ − 1− (8.27)
4 ra 2r
γ − 1 ra
( )
a ≈ a∞ 1 + (8.28)
4 r
1 ra
( )
ρ ≈ ρ∞ 1 + . (8.29)
2r
For a gas with γ = 5/3, the infall velocity, sound speed, and density at small
radii (r ≪ ra ) are
'−1/2
a∞ r
&
u ≈ −a ≈ − (8.30)
2 ra
ρ∞ r −3/2
& '
ρ ≈ . (8.31)
8 ra
If 1 ≤ γ < 5/3, the infall at r ≪ rs is supersonic, and the infalling gas is in
free fall. From the Bernoulli integral, we find that u2 /2 ≈ GM/r, or
'−1/2
r
&
u ≈ −a∞ . (8.32)
ra
The density is then
qs ρ∞ r −3/2
& '
ρ≈ . (8.33)
4 ra
Spherical accretion of gas thus has a characteristic density profile, with ρ ∝
r−3/2 at small radii and ρ = constant at large radii.
The gas will come pouring in with velocity u ∝ r−1/2 at small radii, while
the temperature of the gas increases to the value
'−1
T∞ r
&
T ≈ (8.34)
4 ra
when γ = 5/3 and

'γ−1 & '−3(γ−1)/2
qs r
&
T ≈ T∞ (8.35)
4 ra
when 1 ≤ γ < 5/3. The velocity will not increase indefinitely; eventually the
gas will slam into the surface of the compact object – or disappear within
the event horizon if the compact object is a black hole. The velocity of the
gas just before it hits the compact object will be u ∼ −(2GM/R) ∼ −Vesc .
Spherical accretion is only a good approximation for isolated compact
objects. If the accreting mass is fed by a binary companion, or is located in
the center of an active galactic nucleus, an accretion disk will generally form.
The case to which spherical accretion applies is the case of an isolated black
hole or neutron star in the middle of the ISM. It is very unlikely, however,
that a stellar remnant will be at rest with respect to the ambient ISM. Direct
measurements of the radial velocities of pulsars indicate that pulsars in the
disk of our galaxy have a velocity dispersion of ∼ 100 km s−1 . This is much
larger than the sound speed a ∼ 12 km s−1 in the warm neutral medium.
Thus, a typical pulsar will be moving at supersonic speeds with respect to
its ambient medium.
If the accreting body has a velocity V with respect to the ambient medium,
the transonic accretion rate has the form
G 2 M 2 ρ∞
Ṁt = 4π q̃ , (8.36)
(a2∞ + V 2 )3/2
where q̃ is a factor of order unity. When V > a∞ , a bow shock forms in front
of the accreting object, as shown in Figure 8.2, raising the temperature of the
gas and decreasing its bulk velocity relative to the central accreting mass. At
radii r ≪ ra ∼ 2GM/(V 2 + a2∞ ), the flow of the gas is approximately radial,
and takes the form of the spherically symmetric Bondi solution.
A compact object that is moving at highly supersonic speeds will accrete
mass at a rate Ṁ ∝ V −3 . However, there is a lower limit to the accretion
rate that is dictated by the fact that the compact object has a physical cross
section equal to πR2 . Thus, in the limit that the accreting object is moving at
a velocity V that is much greater than the escape velocity Vesc = (2GM/R)1/2
from its surface, the accretion rate is
Ṁt ≈ πR2 ρ∞ V . (8.37)

87
Figure 8.2: The isovelocity contours (above) and isodensity contours (below)
for accretion onto a compact object moving with speed V = 2.4a∞ with
respect to the ambient gas.
Figure 8.3: The mass accretion rate for a compact object as a function of
the speed with which is it moving relative to the surrounding gas.
To minimize the rate of accretion for a compact object, you should send it off
at a velocity equal to its escape velocity. A schematic plot of the accretion
rate as a function of the compact object’s velocity is given in Figure 8.3.
Chapter 9
Accretion Disks for Beginners
When the gas being accreted has high angular momentum, it generally forms
an accretion disk. If the gas conserves angular momentum but is free to
radiate energy, it will lose energy until it is on a circular orbit of radius
Rc = j 2 /(GM ), where j is the specific angular momentum of the gas, and
M is the mass of the accreting compact object. The gas will only be able to
move inward from this radius if it disposes of part of its angular momentum.
In an accretion disk, angular momentum is transferred by viscous torques
from the inner regions of the disk to the outer regions.
The importance of accretion disks was first realized in the study of binary
stellar systems. Suppose that a compact object of mass Mc and a ‘normal’
star of mass Ms are separated by a distance a. The normal star (a main-
sequence star, a giant, or a supergiant) is the source of the accreted matter,
and the compact object is the body on which the matter accretes. If the
two bodies are on circular orbits about their center of mass, their angular
velocity will be
! "1/2
⃗ = G(Mc + Ms )
Ω ê , (9.1)
a3
where ê is the unit vector normal to the orbital plane.

In a frame of reference that is corotating with the two stars, the equation
of momentum conservation has the form
∂⃗u ⃗ u = − 1 ∇P
⃗ − 2Ω
⃗ × ⃗u − ∇Φ
⃗ R,
+ (⃗u · ∇)⃗ (9.2)
∂t ρ
89
90 CHAPTER 9. ACCRETION DISKS FOR BEGINNERS
Figure 9.1: Sections in the orbital plane of equipotential surfaces, for a binary
with q = 0.2. The five Lagrange points, L1 through L5 , are shown.
where ΦR is the Roche potential,

GMc GMs 1 ⃗
ΦR (⃗r) = − − − |Ω × ⃗r|2 . (9.3)
|⃗r − ⃗r1 | |⃗r − ⃗r2 | 2
The vectors ⃗r1 and ⃗r2 are the positions of the two stars in the rotating frame
of reference; all distances are measured relative to the center of mass of the
system.
The shapes of the equipotential surfaces (where ΦR = constant) are dic-
tated by the mass ratio q ≡ Ms /Mc . Figure 9.1 shows the equipotential
surfaces for a binary system with Ms /Mc = 0.2. At very large distances,
r ≫ a, the equipotential surfaces are nearly spherical, and are centered on
the center of mass of the binary system. At very small distances from the
91
two stars, the equipotential surfaces form a pair of nearly spherical surfaces,
centered on the two stars. It is at intermediate distances that the equipoten-
tial surfaces take on interesting shapes. Of particular interest in the context
of accretion is the ‘critical surface’, which has a dumbbell shape, and consists
of two Roche lobes, one for each star, which are connected at the inner
Lagrangian point L1 . The L1 point is a saddle-point in the potential; the
gravitational acceleration at the L1 point is equal to zero. If the normal star,
in the process of stellar evolution, expands to fill its Roche lobe, then the gas
at the L1 point will be pushed, by its internal gas pressure, from the Roche
lobe of the normal star (Ms ) to the Roche lobe of the compact star (Mc ).
The mass transfer between the two components will change the period of the
binary system. If the total mass and the orbital angular momentum of the
system are constant, then the period of the system is P ∝ Mc−3 Ms−3 . The
period is minimized, given the constraint that Mc + Ms is constant, when
Mc = Ms . Kepler’s law tells us that a ∝ Mc−2 Ms−2 , which is also minimized
when Mc = Ms . When the normal ‘donor’ star is more massive than the
compact accretor (Ms > Mc ), the loss of mass will decrease the separation
a and will decrease the size of the Roche lobe of the normal star; thus, the
mass loss will be naturally self-perpetuating. When Ms < Mc , however, the
separation a will increase as the normal star loses mass, and thus the mass
loss will tend to be cut off (unless evolutionary swelling continues to increase
the size of the mass-losing star, or loss of angular momentum from the binary
system shrinks the separation between the stars.)
The distance of the L1 point from the center of the compact star is well
approximated by the formula
b1
= 0.500 + 0.227 log q , (9.4)
a
so that even when the mass shed by the normal star crosses over the L1
saddle point, it has a considerable distance to travel until it can collide with
the compact star. Moreover, the gas that is passing through the L1 point has
considerable angular momentum. In a nonrotating frame of reference fixed
to the compact star, the L1 point is orbiting with a velocity v ∼ b1 Ω. The
accreting matter that is being squirted through the L1 nozzle therefore has a
bulk velocity u⊥ ∼ v ∼ 2πb1 /P perpendicular to the line between the stars.
Since the accreting matter is being squirted by pressure forces, the velocity of
the gas parallel to the line between the stars will be u∥ ∼ as , where as is the
sound speed in the outermost envelope of the mass-losing star. The normal
temperatures of stellar envelopes are T < 5

∼ 10 K, so typically you will have
u∥ < −1
∼ 30 km s . If the masses of the two stars are roughly equal, however,
the use of Kepler’s law tells us
# $1/3 # $−1/3
−1 Mc P
u⊥ ∼ 100 km s . (9.5)
M⊙ 1 day
In general, then, u⊥ will be greater than u∥ . The specific angular momentum

of the accreting gas will be j = b21 (2π/P ) = b21 (GMc )1/2 (1 + q)1/2 a−3/2 . If
it conserves angular momentum during infall, it will end up on a circular
orbit of radius Rc = (1 + q)(b1 /a)4 a. For q >
∼ 0.5, the circular radius will be
roughly Rc ∼ 1 R⊙ (P/1 day)2/3 . Even for orbital periods as short as hours,
this radius is larger than the radius of a white dwarf (R ∼ 0.01 R⊙ ), to say
nothing of a neutron star (R ∼ 2 × 10−5 R⊙ ), or a solar mass black hole
(R ∼ 4 × 10−6 R⊙ ).
Let us start our study of the dynamics of disks with the thin disk ap-
proximation; we will assume that the height H of the disk in the z direction
is much smaller than the extent of the disk in the R direction. We will also
assume that the disk is axisymmetric. If the mass of the disk is negligible
compared to the mass of the central compact object, the angular velocity
will have the Keplerian form
Ω(R) = (GM/R3 )1/2 , (9.6)
so that the circular velocity will be uφ (R) = RΩ(R) ∝ R−1/2 . In addition to

the circular velocity uφ , the matter of the disk will have a small radial drift
velocity uR (R, t) that carries the gas inward or outward.
Consider the two thin annuli of matter on either side of the surface R =
constant. If the kinematic viscosity of the gas is ν, the viscous torque exerted
on the inner annulus by the outer annulus is
dΩ
T (R) = 2πνΣR3 , (9.7)
dR
where Σ(R, t) is the surface density of the disk. For a disk in which Ω
decreases with radius, T (R) is negative.
The mass conservation equation for a thin disk takes the form
∂Σ ∂
R + (RΣuR ) = 0 . (9.8)
∂t ∂R
93
The conservation equation for angular momentum takes the form

∂ ∂ 1 ∂T
R (ΣR2 Ω) + (ΣuR R3 Ω) = . (9.9)
∂t ∂R 2π ∂R
These two equations may be combined into a single equation
! "
∂Σ 1 ∂ 1 ∂T
R =− . (9.10)
∂t 2π ∂R (R Ω)′ ∂R
2
(The prime denotes differentiation with respect to R). If we use the Keplerian
angular velocity Ω = (GM/R3 )1/2 and the Newtonian viscous torque T =
2πνΣR3 Ω′ , the time evolution of the surface density is
! "
∂Σ 3 ∂ ∂
= R1/2 (νΣR1/2 ) . (9.11)
∂t R ∂R ∂R
This is the basic equation for the evolution of a Keplerian accretion disk.
Once we know the surface density as a function of time, the radial drift
velocity follows from mass conservation:
3 ∂
uR (R, t) = − (νΣR1/2 ) . (9.12)
ΣR1/2 ∂R
Generally, the kinematic viscosity ν will be a function of R and t. How-
ever, the qualitative behavior of a viscous disk can be found by looking at
the simplest case, in which ν = constant. Suppose we start with an accretion
disk that is an infinitesimally thin ring, with mass m and radius R0 :
mδ(R − R0 )
Σ(R, t = 0) = . (9.13)
2πR0
In terms of the dimensionless radius variable x ≡ R/R0 and the dimensionless
time variable τ ≡ t(12ν/R02 ), the solution for the surface density is
1 + x2
! "
m −1 −1/4 2x
% &
Σ(x, τ ) = 2
τ x exp − I1/4 , (9.14)
πR0 τ τ
where I1/4 is a modified Bessel function. This solution is plotted in Figure 9.2
at four increasing values of τ . At early times (τ ≪ 1), the surface density is
nearly Gaussian around R = R0 , with a width σ ≈ (6νt)1/2 . At later times,
most of the mass has lost angular momentum and has moved inward, but
Figure 9.2: The viscous evolution of an initially thin ring of mass m and
initial radius R0 . [Pringle, 1981, ARAA, 19, 137]
there is a tail of matter that moves outward to larger and larger radii, and
which carries off most of the angular momentum.
A disk that starts as a thin ring will spread with time. In many cases of
interest, the accretion disk is fed by a steady stream of matter from outside,
and the disk settles into a steady state in which the rate at which gas is lost
from the inner edge is equal to the rate at which gas is added at the outer
edge. In a steady-state disk of this sort, the radial momentum equation is
∂uR u2φ 1 ∂P GM
uR − + + 2 =0, (9.15)
∂R R ρ ∂R R
assuming Keplerian rotation. The inflow velocity uR is of order ν/R, and is
much smaller than the rotation velocity uφ . If we define the Mach number
of the rotation as Mφ ≡ uφ /a,
&1/2
GM
%
uφ = [1 + O(Mφ−2 )] . (9.16)
R
Perpendicular to the disk, the gas is in hydrostatic equilibrium, with
1 ∂P GM z
=− 3 (9.17)
ρ ∂z R
95
or
a2 ∂ρ u2
≈ − φ2 z (9.18)
ρ ∂z R
for a thin disk. Equation (9.18) can be rewritten in the form
1 ∂ρ Mφ2
≈− 2z . (9.19)
ρ ∂z R
This has a solution
z2
! "
ρ ≈ ρ0 exp − 2 , (9.20)
2H
where the scale height is H ≈ R/Mφ . Thus, the thin disk approximation
(H ≪ R) is equivalent to saying that the disk rotation is supersonic (Mφ ≫
1).
In a steady-state disk, the inward mass flux is
Ṁ = −2πRΣuR . (9.21)
Moreover, when ∂Σ/∂t = 0, the equation for the conservation of angular
momentum can be integrated to yield
∂Ω C
−νΣ = −ΣuR Ω − , (9.22)
∂R 2πR3
where C is a constant of integration. For a disk in Keplerian rotation around
a star of radius R∗ , the integration constant is C = Ṁ (GM R∗ )1/2 . Thus,
! &1/2 "
Ṁ R∗
%
νΣ = 1− . (9.23)
3π R
When R ≫ R∗ , the surface density of a Keplerian viscous disk will be Σ ≈
Ṁ /(3πν), and the radial velocity will be uR ≈ −3ν/(2R).
Let D(R) be the rate per unit time per unit area at which the kinetic
energy of rotation is dissipated into heat by viscosity. The value of D is
# $2
1 ∂Ω
D(R) = νΣ R . (9.24)
2 ∂R
For a Keplerian disk, the viscous dissipation rate is
! &1/2 "
3GM Ṁ R∗
%
D(R) = 3
1− . (9.25)
4πR R
The formula for the dissipation given above is independent of the value of
ν; it merely includes the assumption that the transfer of mass through the
accretion disk and the dissipation of energy within the disk are both regulated
by the viscosity.
The total disk luminosity is
' ∞ 1 GM Ṁ
Ldisk = 2π D(R)RdR = . (9.26)
R∗ 2 R∗
The disk luminosity is half of the total accretion luminosity Lacc = GM Ṁ /R∗ .
Only half of the luminosity is emitted during the gradual passage of the gas
inward through the accretion disk; the other half must be emitted when the
gas makes the transition from the inner edge of the accretion disk to the
surface of the compact object.
What is the source of the kinematic viscosity ν that causes the inward
drift of the gas in the accretion disk? The first guess would be that ν comes
from standard molecular viscosity, the result of thermal collisions between
individual gas particles in a hot medium. In the case of standard viscosity,
the kinematic viscosity is ν ∼ aT λ, where where aT = (kT /m)1/2 is a typical
thermal velocity, and λ is the mean free path length.
Some more-or-less typical values for accretion disks, at a radius R ∼
1010 cm, are T ∼ 104 K and n ∼ 1016 cm−3 . The mean free path (in an
ionized gas) is
k2T 2
λ = 4 ∼ 10−3 cm , (9.27)
πe n
the thermal velocity in the accretion disk is aT ∼ 106 cm s−1 , and the kine-
matic viscosity is ν ∼ 103 cm2 sec−1 . This value of the kinematic viscosity
yields a viscous accretion time scale of tacc = R2 /ν ∼ 1017 sec ∼ 3 × 109 yr,
and the mass will be flowing in at the excruciatingly slow rate of uR =
−(3ν)/(2R) ∼ −5 cm yr−1 .
To explain the observed accretion rates in X-ray binaries and protostars, a
much larger value of the kinematic viscosity is required. The Reynolds num-
ber for the standard molecular viscosity is Re = uφ R/ν = (GM R)1/2 /ν ∼
1015 at R = 1010 cm from a 1 M⊙ body, with ν = 103 cm2 sec−1 . At such a high
Reynolds number, some physicists have argued, we might expect turbulence
to set in. The random eddies of the turbulence would cause viscosity, just as
random thermal motions cause viscosity on the molecular scale. The kine-
matic viscosity due to turbulence should be, from dimensional arguments,
97
νt ∼ Λt ut , where Λt is the size of the largest eddies, and ut is the r.m.s.

turbulent velocity. In a thin disk, the largest eddies can be no larger than
the disk thickness, so λt <
∼ H. The r.m.s. turbulent velocity is unlikely to
be much larger than the sound speed (if it were, shocks would form and the
kinetic energy of turbulent motion would become thermalized); thus, we ex-
pect ut <
∼ a. These considerations led Shakura and Sunyaev (1973) to invent
the alpha disk, in which the kinematic viscosity at a given radius is
ν(R, t) = αa(R, t)H(R, t) . (9.28)
The parameter α is expected, from the arguments given above, to have the
numerical value α < ∼ 1.
The alpha disk is the most commonly used model for accretion disks.
However, by introducing α, we have merely parameterized our ignorance.
There is not even agreement as to the actual source of the viscosity. Some
astronomers, after pointing out that there are no observations that prove that
disks are turbulent, proposed that the viscosity is caused by magnetic stresses
due to a tangled magnetic field within the accretion disk. Such a disordered
field would give rise to an alpha disk with α ∼ vA2 /a2 , where vA is the Alfven
velocity. Other researchers mutter about convective instabilities as a means of
transporting angular momentum; others invoke spiral density waves; others
propose more exotic mechanisms for transporting angular momentum. In
other words, no one knows for sure what causes the viscosity in accretion
disks; we just know that it exists, and that the most plausible mechanisms
result in viscosities that can be written in the form ν = αaH, where α < ∼ 1.
Chapter 10
Advanced Accretion Disks
The total luminosity of a disk with the viscous dissipation rate D(R) is
! ∞ 1 GM∗ Ṁ
Ldisk = 2π D(R)RdR = . (10.1)
R∗ 2 R∗
The disk luminosity is half of the total accretion luminosity Lacc = GM∗ Ṁ /R∗ ;
the other half of the luminosity is emitted when the gas makes the transition
from the inner edge of the accretion disk to the surface of the compact object.
The physical characteristics of the boundary layer between the Keplerian
disk and the compact object are poorly known. Let’s start by considering
the better-known luminosity that comes from the disk itself.
If the disk is optically thick, and the luminosity Ldisk is in the form of
black body radiation, then the temperature Tbb (R) at a given radius is set
by the relation
1
σSB Tbb4 = D(R) . (10.2)
2
The factor of 1/2 enters because we are considering the radiation from only
one side of the disk. Using D(R) for a viscous accretion disk,
#−3/4
R
"
Tbb (R) = T∗ [1 − (R∗ /R)1/2 ]1/4 , (10.3)
R∗
where $ %1/4
3GM∗ Ṁ
T∗ = . (10.4)
8πR∗3 σSB
The temperature of the disk is plotted in Figure 10.1. The hottest part of
99
100 CHAPTER 10. ADVANCED ACCRETION DISKS
Figure 10.1: The temperature of an optically thick accretion disk.

101
the disk is at R = 1.36R∗ , where Tbb = 0.488T∗ . The characteristic disk

temperature T∗ , for typical X-ray binaries, is
$ %1/4 $ %1/4 " #−3/4
7 Ṁ M∗ R∗
T∗ = 1 × 10 K . (10.5)
1017 g sec−1 1 M⊙ 10 km
The spectrum of the disk is the black-body spectrum integrated over all radii:
! Ro ν3
Sν ∝ RdR , (10.6)
R∗ exp[hν/kTbb (R)] − 1
where Ro is the outer radius of the accretion disk. At high frequencies,

ν ≫ kT∗ /h, the spectrum falls away exponentially. At lower frequencies,
ν ≪ kT∗ /h, the temperature has the radial dependence T ∝ R−3/4 , and the
spectrum has the form
1/3
! hν/kTo x5/3 dx
Sν = ν . (10.7)
0 ex − 1
When kTo /h ≪ ν ≪ kT∗ /h, the upper limit of the integral may be taken as
infinity, yielding the spectrum Sν ∝ ν 1/3 at intermediate frequencies. At very
low frequencies, ν ≪ kTo /h, the main contribution to the spectrum is the
Rayleigh-Jeans tail of the cool, outer edge of the accretion disk; the resulting
frequency dependence is Sν ∝ ν 2 . The integrated spectrum Sν for a disk
with T∗ = 100To is plotted in Figure 10.2.
One check to see whether the ‘physically thin, optically thick’ model is
correct is to check for the characteristic ν 1/3 spectrum. Unfortunately, in
binary systems, the luminosity of the normal star that is providing the ac-
creting matter may overpower the luminosity of the disk. The best system
for seeing the entire spectrum of the accretion disk is a cataclysmic vari-
able in which the compact star is a white dwarf and the normal star is a
low mass (0.1 → 1 M⊙ ) main sequence star. In the quiescent state, between
cataclysmic outbursts, most of the luminosity comes from the disk. The cata-
clysmic variable VW Hydri (a dwarf nova) has a very dim normal companion,
and a large disk. The observed spectrum of VW Hydri, in its post-outburst
state, is shown in Figure 10.3. It is a fairly good example of a ν 1/3 law.
It is enlightening to do slightly more elaborate modeling of alpha disks,
to find its properties as a function of the radius R, the mass M∗ of the central
object, the accretion rate Ṁ , and the assumed value of α. Let ρ, a, P , and
Figure 10.2: The spectrum Sν of an optically thick accretion disk. The units
of ν and Sν are arbitrary. The temperature at the outer edge of the disk is
Tout = To = 0.01T∗ .
103
Figure 10.3: The dots are the observed spectrum Fλ of VW Hydri after
an outburst. Line ‘a’ is a stellar atmosphere fit; line ‘b’ is a line of slope
Fλ ∝ λ−7/3 , corresponding to Sλ ∝ ν 1/3 .
T be the density, sound speed, pressure, and temperature in the z = 0 plane.

The optical depth τ = ρHκR of the disk (where κR is the Rosseland mean
opacity) is assumed to be much greater than one. We can solve for the eight
unknowns ρ, Σ, H, a, P , T , τ , and ν as a function of M∗ , Ṁ , R, and α.
Equations for the alpha disk:
Σ
ρ= (10.8)
H
%1/2
R3
$
H=a (10.9)
GM∗
P
a2 = (10.10)
ρ
ρkT
P = (10.11)
m
#1/2 '
4σSB T 4
&
3GM∗ Ṁ R∗
"
= 3
1− (10.12)
3τ 8πR R
τ = ΣκR (ρ, T ) (10.13)
& #1/2 '
Ṁ R∗
"
νΣ = 1− (10.14)
3π R
ν = αaH (10.15)
When the dominant source of opacity in the disk is free-free absorption, the
Rosseland mean opacity κR (ρ, T ) is well approximated by Kramers’ law:
$ %" #−7/2
2 −1 ρ T
κR = 6.6 cm g . (10.16)
10 g cm−3
−8 104 K
At higher temperatures and lower densities, the main source of opacity is
Thomson scattering of photons by free electrons, with κR = 0.40 cm2 g−1 .
The above set of equations has an algebraic solution. Let f ≡ 1 −
(R∗ /R)1/2 , R10 ≡ R/1010 cm, Mc ≡ M∗ /1 M⊙ , and Ṁ16 ≡ Ṁ /1016 g s−1 .
If the dominant source of opacity is free-free absorption, then
7/10 −3/4
Σ = 5 g cm−2 α−4/5 Ṁ16 Mc1/4 R10 f 14/5 (10.17)
3/20 9/8
H = 2 × 108 cm α−1/10 Ṁ16 Mc−3/8 R10 f 3/5 (10.18)
11/20 −15/8 11/5
ρ = 3 × 10−8 g cm−3 α−7/10 Ṁ16 Mc5/8 R10 f (10.19)
105
3/10 −3/4
T = 1 × 104 K α−1/5 Ṁ16 Mc1/4 R10 f 6/5 (10.20)
1/5
τ = 30 α−4/5 Ṁ16 f 4/5 (10.21)
3/10 3/4
ν = 2 × 1014 cm2 sec−1 α4/5 Ṁ16 Mc−1/4 R10 f 6/5 (10.22)
3/10 −1/4
uR = 3 × 104 cm s−1 α4/5 Ṁ16 Mc−1/4 R10 f −14/5 . (10.23)
Fortunately, none of these results have an exorbitantly strong dependence on
the unknown value of α.
We can check whether our assumption of Kramers’ law is valid. The
opacity within the disk will be
−1/2 3/4
κR (Kramers) = 6.3 cm2 g−1 Ṁ16 Mc−1/4 R10 f −2 . (10.24)
The bremsstrahlung opacity dominates over the electron scattering opacity

at radii
2/3
R > 2.5 × 108 cm Ṁ16 Mc1/3 f 8/3 . (10.25)
The crossover radius is smaller than the radius of a white dwarf for values
of Ṁ typical of cataclysmic binaries. Thus, for white dwarfs, Kramers’ law
usually holds. In X-ray binaries with a central neutron star, the inner parts
of the accretion disk will have an opacity dominated by electron scattering.
The boundary layer emits half the accretion luminosity. In the absence
of a magnetic field, the accretion disk can extend all the way to the surface
of the compact object, and the boundary layer consists of a very thin region
just above the surface of the compact object. Consider the boundary layer
in the accretion disk around a non-magnetic compact object of mass M∗ and
radius R∗ . The boundary layer consists of the region R∗ < R < R∗ +b, where
the angular velocity decreases from its Keplerian value
$ %1/2
GM∗
Ω(R∗ + b) = (10.26)
(R∗ + b)3
to the angular velocity of the star’s surface

$ %1/2
GM∗
Ω∗ < . (10.27)
R∗3
The angular velocity in the vicinity of the boundary region is plotted in

Figure 10.4. The radial extent b of the boundary layer is less than the disk
Figure 10.4: The distribution of angular velocity Ω in the boundary layer,

compared to the Keplerian value.
thickness H just outside the boundary layer, which in turn is less than the
radius R∗ of the compact object. To show that b < H < R∗ , start with the
equation for conservation of radial momentum:
∂uR u2φ 1 ∂P GM∗

uR − + + =0. (10.28)
∂R R ρ ∂R R2
In the Keplerian disk, the gravitational term (GM∗ /R2 ) is balanced by the
centrifugal term (u2φ /R). Within the boundary layer, the gravitational term
is balanced by the pressure gradient term. The magnitude of the pressure
gradient is
1 ∂P a2
∼ , (10.29)
ρ ∂R b
where a is the sound speed at the outer edge of the boundary layer and b
is the thickness of the boundary layer. From the balance of forces, a2 /b ∼
GM∗ /(R∗ + b)2 , we see that
R∗ + b
b∼ , (10.30)
Mφ2
107
Figure 10.5: A sketch of the boundary layer of an optically thick disk.
where Mφ is the rotational Mach number at the outer edge of the boundary
layer. In a thin disk, Mφ ≫ 1, and H ∼ R/Mφ , so b ∼ H/Mφ ∼ R∗ /Mφ2 .
Figure 10.5 shows the resulting geometry, with b ≪ H ≪ R∗ . The black-
body temperature within the boundary layer is
1/4
TBL ≈ Mφ T∗ . (10.31)
Magnetic fields can affect the accretion of matter. The magnetic field of
compact stars is a dipole field, with dipole moment µ = B∗ R∗3 , where R∗ is
the radius of the star and B∗ is the magnetic field at the surface of the star.
A neutron star typically will have R∗ ≈ 106 cm and B∗ ≈ 1012 G, yielding
µ = 1030 G cm3 . A white dwarf with R∗ ≈ 5 × 108 cm and B∗ ≈ 104 G will
have the same magnetic moment, µ ≈ 1030 G cm3 .
At distances r ≫ R∗ , the amplitude of the dipole magnetic field is
µ(1 + 3 cos2 θ)1/2

B≈ , (10.32)
r3
where θ is angle measured from the magnetic pole. The magnetic energy
density in the equatorial plane is
B2 1 µ2
#−6
r
"
Emag = = 6
≈ 8 × 1022 erg cm−3 µ230 R6−6 . (10.33)
4π 4π r R∗
The quantity µ30 is the magnetic moment of the star, measured in units of
1030 G cm3 ; the quantity R6 is the radius R∗ of the star, measured in units
of 106 cm.
Consider a neutron star or white dwarf isolated in the middle of a uniform
gaseous medium. The accreting matter will fall in radially at large distances,
where the magnetic pressure is small. The infall will be significantly deflected
from a radial flow at a radius rA (the Alfven radius), where the magnetic
energy density Emag is equal to the kinetic energy density Ekin of the gas.
From the continuity equation,
Ṁ
ρu = − . (10.34)
4πr2
For an ionized gas (γ = 5/3) inside the accretion radius, the infall velocity is
#1/2
GM∗
"
u=− . (10.35)
2r
The kinetic energy density is thus
√
1 2 Ṁ GM∗ −5/2
Ekin = ρu = √ r . (10.36)
2 8π 2
The Alfven radius is %1/7
µ4
$
rA ≈ . (10.37)
Ṁ 2 GM∗
Using the appropriate numerical values for a neutron star or white dwarf,
4/7 −2/7
rA = 7 × 108 cm µ30 Ṁ16 Mc−1/7 . (10.38)
Accreting neutron stars will generally have rA ≫ R∗ . A typical white dwarf

will have rA ∼ R∗ .
There exists, however, a class of magnetic white dwarfs, which have
magnetic dipole moments as large as µ ∼ 1034 G cm3 , and Alfven radii
rA ∼ 1011 cm. One class of cataclysmic variables, known as AM Herculis
systems, consists of magnetic white dwarfs with low mass main sequence
companions, on an orbit with P < ∼ 3 hr. In these systems, the Alfven radius
is larger than the orbital radius. The mass that is lost by the main sequence
star is compelled to flow along the magnetic field lines.
109
Consider a binary system in which the magnetic field of the compact star
is small enough to allow an accretion disk to form. The accretion disk will
be disrupted at a radius RA , which is the radius at which the torque exerted
by the magnetic field on the disk is equal to the viscous torque. Computing
the magnetic torque is, unfortunately, difficult. The torque exerted depends
on the azimuthal component Bφ of the magnetic field, which in turn depends
on the extent to which the magnetic dipole configuration is distorted by the
interaction of the magnetic field with the disk. Calculations usually show
that RA ≈ rA /2.
Because the magnetic field lines are pinned to the compact object, they
have an angular velocity equal to Ω∗ , the angular velocity with which the
compact object rotates. At radii R ≥ RA , the accreting gas rotates with
an angular velocity Ω(R) = (GM∗ /R3 )1/2 . At radii R ≤ RA , the gas flows
along the magnetic field lines, and hence rotates with an angular velocity
Ω(R) = Ω∗ . For steady accretion to occur, we must have a situation in
3 1/2
which Ω∗ ≤ (GM∗ /RA ) . Numerically, this requires that
−6/7 3/7
Ω∗ < 2 sec−1 µ30 Ṁ16 Mc5/7 . (10.39)
If the magnetized compact object rotates more rapidly than this, gas will be
unable to accrete.
Matter will leave the accretion disk at a radius RA and flow along the
magnetic field lines to the magnetic poles of the neutron star, as shown in
Figure 10.6. The rotation axis of the the neutron star is aligned with the
rotation axis of the disk. However, the magnetic axis of the neutron star is
generally at an angle to the rotation axis of the disk. Let α be the angle
between the magnetic axis of the star and the plane of the disk, as shown in
Figure 10.6. In polar coordinates (r, θ) aligned with the magnetic axis, the
magnetic field lines are described by the equation r(θ) = C sin2 θ. The field
line that passes through the disk at the radius r = RA and angle θ = α must
then have the equation
sin2 θ
r(θ) = RA 2 . (10.40)
sin α
The gas that is ripped away from the disk will follow along this path until
it reaches the surface of the neutron star at a radius r = R∗ . The accretion
thus takes place on a ring of angular radius θc around the magnetic pole,
where θc is given by the relation
R∗
sin2 θc = sin2 α . (10.41)
RA
Figure 10.6: Accretion from a gaseous disk onto a magnetized neutron star.
111
Diffusion effects tend to smear the area of accretion into a circular cap cen-
tered on the magnetic pole. The area of the two accreting polar caps takes
up a fraction
πR2 sin2 θc R∗ sin2 α
f ∼2 ∗ 2 ∼ (10.42)
4πR∗ 2RA
of the total surface area of the neutron star. The fraction f is small; f ∼ 10−3
for a neutron star with α ∼ π/2. All of the ‘boundary layer’ accretion lumi-
nosity, equal in value to L = (GM∗ Ṁ )/(2R∗ ), will be emitted from two small
regions. The rotation of the neutron star will cause a periodic modulation in
the observed flux from the accreting polecaps. Pulsation periods are observed
in X-ray binaries, with typical periods in the range 1 sec < ∼P <
3
∼ 10 sec. In
many X-ray binaries, the pulsation period P is observed to be decreasing
steadily, on time scales of ∼ 104 yr. This spin-up is the result of the torque
exerted by the accretion disk on the magnetic field of the neutron star. For
a magnetized neutron star accreting matter from a disk, the Keplerian an-
3 1/2
gular velocity at the Alfven radius, Ω = (GM∗ /RA ) , is larger than Ω∗ ,
the angular velocity of the neutron star and its attached magnetic field. The
star + magnetic field combination is thus accreting angular momentum from
2
the disk at the rate Ṁ RA Ω(RA ). If the moment of inertia of the star is
2
I ∼ M∗ R∗ , then the spin-up rate is given by the relation
2
I Ω̇∗ = Ṁ RA Ω(RA ) = Ṁ (GM∗ RA )1/2 . (10.43)
The spin-up rate for the neutron-star is then
Ṗ 6/7 2/7
− ≈ 10−5 yr−1 I45
−1
Ṁ16 µ30 Mc3/7 P1 , (10.44)
P
where I45 is the moment of inertia in units of 1045 g cm2 .
Chapter 11
The Solar Wind
A stellar wind consists of particles emitted from the stellar atmosphere with
a sufficiently large velocity to escape the star’s gravitational attraction. The
escape velocity at the surface of a star with mass M∗ and radius R∗ is
"1/2
2GM∗
!
vesc = = 620 km s−1 (M∗ / M⊙ )1/2 (R∗ / R⊙ )−1/2 . (11.1)
R∗
Since a star will tend to accrete mass, due to its gravitational attraction, some
non-gravitational force is needed to counteract the inward pull of gravity, and
accelerate the outermost layers of the stellar atmosphere away from the star.
This wind-driving force can be, depending on the type of wind, a gradient
in the gas pressure, a gradient in the radiation pressure, or a gradient in the
magnetic pressure.
Observed stellar winds may be placed into one of four main categories.
1. Winds from main sequence stars. Stars similar to the sun have a low
rate of mass loss: Ṁ ∼ 10−14 M⊙ yr−1 . (1 M⊙ yr−1 = 6.3×1025 g sec−1 .)
The asymptotic velocity u∞ of the wind is comparable to the escape
velocity vesc . Winds from solar-type stars are thought to be driven
by gas pressure gradients in the corona. The prototypical star in this
category is the sun.
2. Winds from hot, luminous stars. Stars with Teff >∼ 15,000 K and L > ∼
3000 L⊙ have a high rate of mass loss: Ṁ ∼ 10−7 → 10−5 M⊙ yr−1 .
Their winds are very fast, with u∞ > vesc . Winds from hot stars
are thought to be driven by radiation pressure. Wolf-Rayet stars fall
111
112 CHAPTER 11. THE SOLAR WIND
into this category. WR stars are hot, luminous stars with extended
envelopes; their mass loss rates (Ṁ ∼ 10−5 M⊙ yr−1 ) and wind terminal
velocities (u∞ ∼ 2500 km s−1 ) are among the largest observed for any
type of star.
3. Winds from cool, luminous stars. Stars with Teff < ∼ 6000 K and L > ∼
100 L⊙ have a high rate of mass loss: Ṁ ∼ 10 → 10−5 M⊙ yr−1 . The
−8
wind velocity, however, is very low, with u∞ < vesc . The mechanism
that drives the wind from cool stars is uncertain; a leading candidate
is radiation pressure on dust grains, aided by stellar pulsations in the
outer atmosphere. Stars in this category are K and M giants and
supergiants.
4. Winds from extremely young stars. T Tauri stars have mass loss rates
of Ṁ ∼ 10−9 → 10−7 M⊙ yr−1 . The wind velocity is u∞ ∼ 200 km s−1 .
T Tauri stars have circumstellar disks that might be accretion disks.
The strength of the wind is correlated with the luminosity of the disk;
this suggests that the outgoing wind might be powered by the accretion
disk.
Figure 11.1 shows the location on a Hertzsprung-Russell diagram of stars

with high rates of mass loss. Hot, luminous stars with winds (including Wolf-
Rayet stars) are in the upper left; cool, luminous stars with winds are in the
upper right. The approximate location of T Tauri stars on the H-R diagram
is also shown.
The best-studied stellar wind is the solar wind, which is the weakest
of all measured stellar winds. The solar wind consists mainly of ionized
hydrogen and fully ionized helium, with heavier elements present in solar
abundances. At 1 AU from the sun, the solar wind is supersonic (u > a)
and super-Alfvenic (u > vA ). As shown in Figure 11.2, the solar wind is
“gusty”, with significant variations in velocity and density. As a lowest-order
approximation, we can recognize two distinct types of solar wind: a high-
speed wind and a low-speed wind. In the high-speed wind, the mean proton
velocity is up = 700 km s−1 and the mean proton density is np = 3.4 cm−3 .
The gas pressure is P = 1.9 × 10−10 dyne cm−2 and the magnetic pressure is
B 2 /(8π) = 1.7 × 10−10 dyne cm−2 ; both of these pressures are much smaller
than the kinetic energy density ρu2 /2 = 1.4×10−8 erg cm−3 . In the low-speed
wind, the mean proton velocity is up = 330 km s−1 and the mean proton
density is np = 10.3 cm−3 . The gas pressure is P = 2.6 × 10−10 dyne cm−2
113
Figure 11.1: A Hertzsprung-Russell diagram (luminosity versus effective tem-

perature) showing the location of stars with high rates of mass loss from
stellar winds.
Figure 11.2: The velocity and density of the solar wind at R = 1 AU from
the Sun, as measured by the Mariner 2 spacecraft.
and the magnetic pressure is B 2 /(8π) = 1.7 × 10−10 ; much smaller than the
kinetic energy density ρu2 /2 = 9.4 × 10−9 erg cm−3 .
At any given time, part of the sun’s corona is emitting a low-speed wind,
and part is emitting a high-speed wind. (The high-speed winds appear to
come from “coronal holes” – regions of low density and low temperature
where the magnetic field lines are not closed.) Both the high and low speed
winds produce a proton flux of ∼ 3 × 108 protons cm−2 sec−1 at 1 AU. This
leads to a total mass loss rate of 2 × 10−14 M⊙ yr−1 .
The study of stellar atmospheres usually starts with the statement, “We
assume that the atmosphere is in hydrostatic equilibrium.” Well, that state-
ment is fraudulent – a stellar atmosphere cannot be in hydrostatic equilib-
rium. There is always going to be a loss of high velocity particles. To show
that the sun’s corona is not in hydrostatic equilibrium, we first assume that
it is in equilibrium, and then show that the assumption leads to an absurd
conclusion. The corona is the outermost part of the sun’s atmosphere; it
extends from a distance of 2000 km above the sun’s visible surface out into
interplanetary space. At the base of the corona, r0 = 7×1010 cm, the temper-
ature is T0 = 2 × 106 K and the number density of protons is n0 = 108 cm−3 .
The resulting gas pressure is P0 = 2n0 kT0 = 0.06 dyne cm−2 . At the high
temperatures present in the corona, the thermal structure is determined by
heat conduction. The coefficient of thermal conductivity for an ionized gas
is equal to
"5/2
T
!
9 −3 −1
K(T ) = 3 × 10 g cm sec K . (11.2)
2 × 106 K
In the corona, where there are no local heat sources, the fact that heat flow
is in a steady state tells us that
# $
∇ ⃗ ) = 1 d r2 K dT
⃗ · (K ∇T =0. (11.3)
r2 dr dr
In conjunction with the relation K ∝ T 5/2 , this yields

"−2/7
r
!
T (r) = T0 . (11.4)
r0
In general, if K ∝ T n , T ∝ r−1/(n+1) ; so if the corona were neutral, with

n = 1/2, the temperature would fall off at the more rapid rate T ∝ r−2/3 .
115
If, as we are assuming, the corona is in hydrostatic equilibrium, we have

the familiar equation
1 dP GM
=− 2 . (11.5)
ρ dr r
If the corona consists of ionized hydrogen, ρ ≈ mP n and P = 2nkT . Using
the substitution T = T0 (r/r0 )−2/7 ,
% ! "−2/7 &
d r GM mp n
n =− . (11.6)
dr r0 2kT0 r2
This equation has the solution
7 r0 −5/7
' (
2/7
n(r) = n0 x exp (x − 1) , (11.7)
5h
where x ≡ r/r0 and the scale height h is given by the relation
$2 # $−1
2kT0 r02
"#
T0 r0 M
!
h≡ = 1.2 × 1010 cm . (11.8)
GM mp 2 × 106 R⊙ M⊙
Near the base of the corona, where r − r0 ≪ r0 , the density falls off expo-
nentially, with
n(r) ≈ n0 e−(r−r0 )/h . (11.9)
At large radii, the density increases at the rate n ∝ r2/7 . In fact, at a distance
of 1 AU, where r = 210r0 , the hydrostatic corona must have a number density
n = 1.4 × 10−3 n0 ∼ 105 cm−3 . But this, of course, is much larger than the
number density of protons that are actually observed at the earth’s orbit.
The gas pressure as a function of radius is
7 r0 −5/7
' (
P (r) = P0 exp (x − 1) . (11.10)
5h
To maintain the corona in hydrostatic equilibrium, there must be a finite
pressure at large radii, with the value
7 r0
' (
P (∞) = P0 exp − ≈ 2 × 10−5 dyne cm−2 . (11.11)
5h
The pressure required to keep the solar corona in hydrostatic equilibrium is
much greater than the pressure P ≈ 3 × 10−13 dyne cm−2 that is found in the
interstellar medium.
Since the corona can’t be stationary, let’s approximate it as having a

steady-state outflow. The continuity equation tells us that Ṁ = 4πr2 ρu.
This is similar to the continuity equation for spherical accretion; the only
difference is in the sign convention. For accretion, we adopted Ṁ > 0 and
u < 0; for winds, we adopt Ṁ > 0 and u > 0. The equation for momentum
conservation in an unmagnetized, spherical, steady-state wind is
du a2 dρ GM
u + + 2 =0. (11.12)
dr ρ dr r
The self-gravity of the wind has been ignored. Using the continuity equation
to eliminate ρ, we find, once again, the Bondi equation:
a2 2a2 r
# $ # $
1 d 2 GM
1− 2 (u ) = − 2 1− . (11.13)
2 u dr r GM
Care must be taken in choosing, from among the possible solutions of the
Bondi equation, the solution that matches the boundary conditions of the
flow. For accretion, we wanted solutions with u2 → 0 as r → ∞. For winds,
we want solutions with u2 → 0 as r → 0. In the nomenclature of Chapter 8, a
stellar wind must have a subsonic solution of type 3, or a transonic solution
of type 2. The subsonic solutions, in this context, are known as “stellar
breezes”; the gas in these flows never becomes supersonic or escapes from
the gravitational well of the star. The transonic solution is known as the
“Parker wind”. Since the solar wind is known to be supersonic at the earth’s
orbit, the “Parker wind” is the appropriate solution for the solar wind.
The temperature is nearly constant throughout the inner regions of the
corona. This fact prompted Parker to look at isothermal winds. An isother-
mal wind of ionized hydrogen has the constant sound speed
# $1/2 "1/2
2kT0 T0
!
−1
a0 = = 180 km s . (11.14)
mp 2 × 106 K
In the transonic solution, the wind velocity will be u = a0 at the sonic radius
# $! "−1
GM M T0
rs = 2
= 2.0 × 1011 cm . (11.15)
2a0 M⊙ 2 × 106
The Bondi equation for an isothermal wind is
a20 u2
# $ # $
d rs r
! "
1− 2 = −4 2 1 − . (11.16)
u dr a20 r rs
117
Figure 11.3: The wind velocity of an isothermal Parker wind, for different
values of the temperature T0 .
This equation may be integrated, using the boundary restriction u = a0 at

r = rs , to yield the Bernoulli integral for an isothermal wind:
u2 u2
# $
r rs
− ln = 4 ln +4 −3 . (11.17)
a20 a20 rs r
This equation gives the wind velocity u as a function of r; a plot of u(r)

for different temperatures is shown in Figure 11.3. At large radii, the wind
velocity is approximately u ≈ 2a0 [ln(r/rs )]1/2 . At small radii, the velocity is
u ≈ a0 e3/2 exp(−2rs /r). For the sun, the sonic radius rs = 2×1011 cm is only
three times the radius of the base of the corona. At 1 AU, the radius is r =
75rs , and the solar wind velocity predicted by the isothermal Parker model
is u ≈ 4.1a0 ≈ 740 km s−1 (T0 /2 × 106 )1/2 . This velocity is approximately
equal to the velocity actually measured for the solar wind (at least its high-
speed component). The agreement, however, is partly due to chance. The
solar wind is not, in fact, perfectly isothermal. At 1 AU, the temperature has
dropped to 1.5 × 105 K.
The magnetic field associated with the solar wind at the earth’s orbit is
⃗ ·B
B ≈ 5 × 10−5 G. The fact that ∇ ⃗ = 0 implies that
Br = B0 (r/r0 )−2 , (11.18)
where B0 = Br (r0 ). Since the base of the solar corona is at a radius r0 =

4.6 × 10−3 AU, the radial component of the magnetic field at the base of the
corona must be of magnitude B0 ∼ 1 G. Since the corona and solar wind
are fully ionized, the magnetic field lines are pinned to the solar wind as
it streams outward. The ends of the magnetic field lines are anchored to
the sun, as it rotates with an angular velocity Ω∗ = 3 × 10−6 sec−1 . If the
magnetic pressures are negligibly small, and the ionized gas streams radially
outward, then the magnetic field lines in the equatorial plane will be drawn
outward in spirals. If the wind speed is constant, the magnetic field lines are
Archimedean spirals, with the shape
ur
r(φ) − r0 = (φ − φ0 ) . (11.19)
Ω∗
The above analysis assumes that the magnetic force terms in the equations
of motion are negligibly small, and that the magnetic field lines are therefore
passively carried about by the gas flow. The magnetic force terms in the
radial direction are, in fact, very small compared to the gravitational term,
so the magnetic field has only a small effect on the radial outflow. However,
the magnetic force terms in the azimuthal direction are crucially important
in determining the rotational component of the wind’s velocity.
Start by assuming a steady-state axisymmetric model. In the equato-
rial plane of the sun, we assume that the magnetic field has no component
perpendicular to the plane. The magnetic field in the plane is
⃗ = Br (r)êr + Bφ (r)êφ
B (11.20)
and the wind velocity is
⃗u = ur (r)êr + uφ (r)êφ . (11.21)
⃗ = −⃗u × B/c,
The electric field in the ionized wind is E ⃗ and the steady-state
Maxwell’s equation ∇ ⃗ ×E ⃗ = 0 reduces to the equation
1d
[r(ur Bφ − uφ Br )] = 0 . (11.22)
r dr
119
Integrating, we find
r(ur Bφ − uφ Br ) = constant = −Ω∗ r2 Br , (11.23)
where Ω∗ is the rotation velocity of the sun.

Since the model is axisymmetric, the equation for the conservation of
angular momentum in the equatorial plane is
d Br r2 d
(ruφ ) = (rBφ ) . (11.24)
dr Ṁ dr
Since Br r2 = constant, we may instantly perform the integration to find that
r2 Br
# $
r uφ − Bφ = L = constant . (11.25)
Ṁ
The constant L is the specific angular momentum carried away by the solar
wind. The first term in the above equation is the specific angular momentum
carried by the motions of the gas; the second term represents the angular
momentum carried by the magnetic field.
It is convenient to define the radial Alfvenic Mach number, MA ≡ ur /vA ,
where vA is the Alfven velocity,
$1/2
Br2
#
vA = . (11.26)
4πρ
At the base of the corona, vA ≈ 300 km s−1 ; at 1 AU, the Alfven velocity has
fallen to vA ≈ 40 km s−1 . The azimuthal velocity of the solar wind, in terms
of Ω∗ , MA , and L, is
LMA2 r−2 Ω−1

∗ −1
uφ = Ω∗ r 2
. (11.27)
MA − 1
At the base of the corona, where the radial velocity ur is still small, the
Alfvenic Mach number is much smaller than one. At 1 AU, the Alfvenic
Mach number is MA ≈ 10. Thus, somewhere between the sun and the earth,
there exists an Alfvenic critical radius at which MA = 1. Let the radius at
this point be rA and the radial velocity be uA . Since the denominator in
the equation for uφ vanishes at rA , the numerator must vanish, too. Thus,
2
L = Ω∗ rA . The angular momentum per unit mass within the solar wind can
be computed as if there were solid body rotation, at an angular velocity Ω∗ ,

out as far as the Alfven radius rA .
Because the angular momentum of the corona is being lost to the solar
wind, the sun is slowly being spun down. The rate at which the sun is losing
angular momentum is
J˙ = −LṀ = −Ω∗ rA
2
Ṁ . (11.28)
For typical models of the magnetized solar wind, the Alfven radius is rA =
1.7 × 1012 cm = 24 R⊙ = 0.11 AU. With the sun’s angular velocity of Ω∗ =
3 × 10−6 sec−1 and mass loss rate of Ṁ = 2 × 10−14 M⊙ yr−1 , this implies
J˙ = −1 × 1031 g cm2 sec−2 . (11.29)
Since the sun’s total angular momentum is J = 1.6 × 1048 g cm2 sec−1 , the
time scale over which the sun will be spun down is
tJ = −J/J˙ ≈ 2 × 1017 sec ≈ 5 × 109 yr . (11.30)
This is a time comparable to the age of the sun, in contrast to the mass loss
time scale tM = M⊙ /Ṁ ≈ 5 × 1013 yr. Although the solar wind will not
significantly affect the total mass of the sun, it will affect the total angular
momentum of the sun.
The angular velocity of the wind may be rewritten as
Ω∗ r uA − ur
uφ = (11.31)
uA 1 − MA2
and the azimuthal component of the magnetic field is

2
Ω∗ r rA − r2
Bφ = −Br 2
. (11.32)
uA r A (1 − MA2 )
If we knew the velocity ur as a function of r, we could solve these equations to

find uφ (r) and Bφ (r). In practice, since the presence of rotation and magnetic
fields have little effect on the radial motions, we can use the standard Parker
wind solution to find ur (r). Well within the Alfven radius, the azimuthal
velocity has the value uφ = Ω∗ r. Far outside the Alfven radius, ur is nearly
constant, and the Alfvenic Mach number increases as MA ∝ r. Thus, uφ
decreases at the rate uφ ∝ r−1 . The maximum azimuthal velocity uφ occurs in
121
Figure 11.4: The angular momentum content of the gas in the solar wind
(solid line) compared to the angular momentum content of the magnetic field
(dashed line).
the vicinity of the Alfven radius. The azimuthal component of the magnetic
field decreases at the rate Bφ ∝ r−1 . Since Br ∝ r−2 , at large radii, the
magnetic field will be mainly azimuthal; at small radii, the field will be
mainly radial.
At a radius r, the gas in the solar wind has a specific angular momentum
Ω∗ r2 uA − ur
Lgas = uφ r = , (11.33)
uA 1 − MA2
plotted as the solid line in Figure 11.4. The magnetic field carries the specific
angular momentum
2
rA − r2
Lmag = L − Lgas = Ω∗ , (11.34)
1 − MA2
which is shown as the dashed line in Figure 11.4. The ratio of angular
momentum in the gas to angular momentum in the magnetic field is, at
small radii, # $2
! "2
r −3 r
Lgas /Lmag ≈ ∼ 2 × 10 . (11.35)
rA R⊙
Figure 11.5: The termination shock, heliopause, and bow shock surrounding
the solar sytem.
At large radii, where r ≫ rA ,

ur ur
! "
Lgas /Lmag ≈ −1∼2 −1 . (11.36)
uA 700 km s−1
Inside the Alfven radius, the magnetic field carries most of the angular mo-
mentum of the solar wind; outside the Alfven radius, it shares the job with
the gas.
The solar wind is shocked and decelerated when its kinetic energy density
2
ρur /2 is comparable to the ambient pressure Pi of the interstellar medium.
This happens at a radius
# $1/2 ! "1/2
Ṁ ur
rs ≈ 140 AU , (11.37)
10 M⊙ yr−1
−14 700 km s−1
if the pressure of the ISM is taken to be Pi ≈ 3×10−13 dyne cm−2 . The space-
craft Voyager 1 went through this termination shock in 2004 December,
when it was 94 AU from the Sun (Figure 11.5). Voyager 2 went through the
termination shock in 2006 May, when it was only 76 AU from the Sun; this
gives a idea of the non-sphericity of the termination shock.
Chapter 12
Winds from Hot & Cool Stars
Hot luminous stars, such as O and B supergiants, are known to have stellar
winds. In the UV, they display P Cygni line profiles.1 P Cygni profiles show
absorption at short wavelengths and emission at longer wavelengths, as seen
in the lower panel of Figure 12.1. This asymmetric absorption/emission
profile is the characteristic signature of an expanding stellar atmosphere.
The absorption comes from material between us and the photosphere of the
star (region 5 in the upper panel of Figure 12.1). The emission comes from
material in regions 1, 2, 3, and 4. The width of the P Cygni line tells us
the terminal velocity u∞ of the expanding gas. The depth of the absorption
component tells us the column density of absorbing matter along the line of
sight to the star. From this, we can use a model of the expanding atmosphere
to deduce the total mass of the gas surrounding the star. The P Cygni lines
of OB supergiants show expansion velocities of ∼ 2000 km s−1 , and mass
loss rates of ∼ 10−6 M⊙ yr−1 . The hottest stars can have maximum wind
velocities of as much as 4000 km s−1 .
Mass loss from hot luminous stars cannot be explained by gas pressure
gradients like those that drive the solar wind. Among the P Cygni lines that
are observed in OB supergiants are those of CIV and SiIV. These ions are
present at a temperature of ∼ 3 × 105 K; if the gas were any hotter, they
would be collisionally ionized to higher ionization states. In a hot coronal
wind driven by gas pressure, the maximum observed velocity can be only a
few times the sound speed, which is a ∼ 70 km s−1 at T ∼ 3 × 105 K. Thus,
1
P Cygni itself is a variable blue hypergiant (B1 Ia); it is visible to the naked eye,
despite being some 2 kpc from the Earth. It’s an example of the type of hot, luminous
star that produces a fast stellar wind with a high mass-loss rate.
123
124 CHAPTER 12. WINDS FROM HOT & COOL STARS
Figure 12.1: Formation of a P Cygni line profile.

125
we need some additional force to explain the very high wind velocities seen
in OB supergiants.
Winds in hot luminous stars are driven by radiation pressure. The stellar
winds emerging from OB supergiants have numerous resonance lines in the
UV, which coincidentally is where the continuum radiation of an OB star has
its maximum. The winds from OB stars are thus referred to as line-driven
winds, since the opacity of the accelerated material is provided by absorption
lines. A rough estimate of the mass-loss rate in a radiatively driven wind can
be computed by assuming that each photon emitted by the star transfers
its momentum of hν/c to a gas particle in the wind. The rate at which the
star loses momentum is L/c, where L is the radiative luminosity. The rate
at which the wind carries away the momentum is Ṁ u∞ , where u∞ is the
asymptotic wind velocity. By setting the two rates equal to each other, we
estimate a mass loss of
L
Ṁest = (12.1)
u∞ c
and a corresponding kinetic energy luminosity of
1 u∞
Ṁest u2∞ = L. (12.2)
2 2c
For a luminous O or B star, this estimated mass loss rate is Ṁest ∼ 10−6 →
10−5 M⊙ yr−1 . This is in the right range to account for the observed mass loss
from OB stars. The kinetic energy luminosity is less than 1% of the radiation
luminosity L. If each photon from the star undergoes multiple scatterings,
then the mass loss rate can be several times Ṁest .
Even if the circumstellar gas had no resonance lines, there would still be
radiation pressure on the ionized gas as a result of Thomson scattering. A
star of luminosity L will exert an outward radial force of magnitude f =
(σT L)/(4πr2 c) on every electron-proton pair (where σT = 6.7 × 10−25 cm2 is
the Thomson cross section). The ratio of the outward acceleration due to
Thomson scattering, to the inward acceleration due to gravitation, is
σT L 3.2 × 10−5
Γ= = , (12.3)
4πGM mp c M/L
where the mass-to-light ratio M/L is in solar units. The dimensionless

number Γ is the ratio of a star’s luminosity L to its Eddington luminos-
ity LEdd = 4πGM mp c/σT . Thomson scattering alone will not be enough to
drive the stellar winds. For a very luminous O star, with L ≈ 106 L⊙ and
M ≈ 60 M⊙ , you find Γ ≈ 1/2; for less luminous stars, Γ will be even smaller.
A single line will provide an acceleration
κL ! ∞
gL = πFν (ν)φ(ν)dν , (12.4)
c ν=0
where κL is the line opacity (in units of cm2 g−1 ), πFν is the flux from the
central star at frequency ν (in units of erg sec−1 cm−2 Hz−1 ) and φ(ν) is the
line profile function.
A complication is added by the fact that the gas of the stellar wind
is being steadily accelerated away from the star. Thus, the radiation that
each bit of gas sees from the central star will be more and more redshifted.
Suppose that the gas has a strong line at a frequency ν0 , as measured in the
rest frame. A photon emitted by the star with a frequency ν > ν0 will be
scattered by the shell of gas that has a velocity
ν − ν0
u(r) = c . (12.5)
ν0
The stellar wind will be able to scatter photons with frequencies between
ν0 and ν0 + δν, where δν = ν0 (u∞ /c). Photons with a frequency ν0 will be
scattered near the surface of the star; photons with a frequency ν0 + δν will
be scattered far from the star, where the expansion velocity has reached its
asymptotic value of u∞ .
That maximum mass loss rate that can be produced by a single strong
line is found by setting the mass momentum flux equal to the radiation
momentum flux in the frequency range ν0 → ν0 + δν:
Lν δν L ν ν0
Ṁ u∞ = = 2 u∞ . (12.6)
c c
The quantity Lν δν is the stellar luminosity in the frequency range ν → ν +δν.
If ν0 is near the peak of the star’s energy distribution, then Ṁ ∼ L/c2 . This
rate is smaller by a factor u∞ /c than the mass loss rate Ṁest that occurs
if all photons are scattered or absorbed, rather than those in the restricted
frequency range ν0 → ν0 (1 + u∞ /c). Even for the fastest stellar winds, the
ratio u∞ /c is only 0.01. If a line-driven wind is to effectively convert the
momentum of photons into the momentum of gas, it must have many strong
lines in the frequency range where the star emits most of its radiation.
127
Suppose that a radiative flux Fν is incident on the inner side of a thin gas
shell. If the gas scatters photons with a frequency ν0 , then the observed res-
onance line will have a Doppler width ∆νD = ν0 ut /c, where ut = (kT /m)1/2
is the thermal velocity within the shell. The radiative acceleration of the
shell is
πFν 1 − e−τL
gL = κL ∆νD . (12.7)
c τL
In the above equation, τL is the effective optical depth of the shell, and is
given approximately by the equation
" #−1
du
τL = κL ρ(r)ut (r) . (12.8)
dr
For strong lines, where τL ≫ 1, the acceleration is
πFν ν0 du
gL ≈ , (12.9)
c ρc dr
while for weak lines, where τL ≪ 1,
πFν ν0 κL ut
gL ≈ . (12.10)
c c
In real stellar winds, the total line acceleration,
$
gL = gL,i , (12.11)
i
will be a sum over weak and strong lines. It is customary (following Cas-
tor, Abbott, & Klein 1975, ApJ, 195, 157), to parameterize the total line
acceleration in a spherically symmetric wind as
" #α
GM 1 du
% &
gL = 2
Γ k , (12.12)
r κρut dr
where the force constant k and the slope α are found by an empirical fit to the
observed line strengths in stellar atmospheres. (The parameter κ = σT /mp =
0.40 cm2 g−1 is the Thomson scattering opacity.) If all the observed lines are
strong, α = 1; if all the lines are weak, α = 0. An O4 star has k ≈ 1/30 and
α ≈ 0.7.
With this parametric form for the line acceleration, the isothermal Bondi
equation takes the modified form
1 d(u2 ) a2 2a2 r
' (
GM
(1 − 02 ) = − 2 1 − 0 − Γ − Γkt−α , (12.13)
2 dr u r GM
where #−1
r2 d(u2 )
" #"
κut Ṁ
t= . (12.14)
4π 2 dr
In the absence of line opacity, the solution to this equation is a Parker wind
with a sonic radius at
GM (1 − Γ)
rs = . (12.15)
2a20
Once the line opacity is added, the sonic radius is no longer a critical point.
There is, however, a unique solution which has a smooth transition from
subsonic flow near the star to supersonic flow far from the star. The terminal
velocity for this model has the value
&1/2
α
%
u∞ = vesc , (12.16)
1−α
where ' (1/2
2GM (1 − Γ)
vesc = . (12.17)
R∗
The observed values of u∞ /vesc for a sample of hot, luminous stars are plotted
in Figure 12.2. The O stars are observed to have u∞ /vesc ≈ 3, suggesting a
value of α ≈ 0.9 for these stars. Late B stars, by contrast, have u∞ /vesc ≈ 1,
yielding α ≈ 0.5. The mass loss rate for a line-driven wind is
&(1−α)/α
L 1/α Γ
%
Ṁ = k α(1 − α)(1−α)/α . (12.18)
ut c 1−Γ
This leads to a luminosity dependence Ṁ ∝ L1/α . If all the lines are optically
thick, then Ṁ ∝ L; if all the lines are optically thin, then there will be a
very steep dependence of Ṁ on L.
An O5 star has a mass M = 60 M⊙ , a luminosity L = 106 L⊙ , and a radius
R∗ = 14 R⊙ . The effective temperature is then 49,000 K. Castor, Abbott,
and Klein used their parametric model, with k = 1/30 and α = 0.7, to deduce
129
Figure 12.2: The ratio of terminal wind velocity to escape speed as a function
of spectral type for stars from O4 to A2.
a terminal velocity u∞ = 1500 km s−1 for such a star, and a mass loss rate
of Ṁ = 7 × 10−6 M⊙ yr−1 . This is in good agreement with observations of O
stars. The time scale for mass loss in such a star is tM = M/Ṁ = 1 × 107 yr,
which is only three times the main sequence lifetime of an O5 star.
The terminal velocities deduced for cool luminous stars range from u∞ ∼
10 km s−1 for M supergiants to u∞ ∼ 75 km s−1 for K giants. These velocities
are smaller than the corresponding escape velocities vesc from the stellar
surface. An approximate fit, found empirically, is
2
vesc
u∞ ∼ . (12.19)
1000 km s−1
The deduced mass loss rate for cool luminous stars lies in the approximate
range 10−8 M⊙ yr−1 → 10−5 M⊙ yr−1 .
Winds in cool K and M stars cannot be driven by pressure gradients, since
cool stars with winds lack hot, extended coronas. Thomson scattering is not
enough to drive the winds of cool stars. An M supergiant, with L = 3×104 L⊙
and M = 20 M⊙ , will have Γ = L/LEdd = 3.2 × 10−5 (M/L)−1 = 0.05. A K
giant, with L = 130 L⊙ and M = 4 M⊙ , will have Γ = 0.001. The radiative
acceleration from lines is also small; the continuum of cool stars peaks in
infrared or red, where there are few resonance lines in the atmosphere. How-
ever, in cool giants and supergiants, there is an additional source of opacity.
Cool giants and supergiants are observed to emit an excess of radiation in
the infrared. This infrared excess can be attributed to the formation of
dust grains in the cool, extended stellar envelope; dust condensation requires
temperatures lower than T ∼ 1000 K.
The infrared opacity of the atmosphere increases very rapidly at the radius
where the dust condenses out. The grains, once they form, are accelerated
outward as the result of the momentum they gain by absorbing photons.
Some of the momentum of the grains is transferred to the gas by collisions.
As the grains are driven outward, they thus drag the gas along with them.
When the dust grains are accelerated to high velocities, however, they begin
to be destroyed by ‘sputtering’. The acceleration ceases, and the terminal
wind velocity stays at the relatively low value of ∼ 20 − 50 km s−1 . If dust
particles were indestructible, the wind could be accelerated to velocities of
∼ 100 km s−1 or more. The radiative acceleration on the dust grains is greater
than the gravitational acceleration when
4πGM c M
% &
κd > = 1.3 × 104 cm2 g−1 , (12.20)
L L
131
where κd is the frequency-averaged opacity of the dust, and the mass-to-light

ratio M/L is in solar units. An M supergiant, with M/L = 7 × 10−4 M⊙ / L⊙ ,
will require a dust opacity of κd > 9 cm2 g−1 in order to have a dust-driven
wind; a K giant, with M/L = 0.03 M⊙ / L⊙ , will require κd > 400 cm2 g−1 .
Radiative pressure on dust cannot be the sole mechanism driving winds
from cool luminous stars, since dust can only exist in a relatively cool envi-
ronment. Infrared observations of Betelgeuse (an M supergiant) show that
less than 20% of the emission from dust comes from within a distance 12R∗ of
the star. For K stars, with their higher effective temperatures, the problem
of forming dust close to the star’s surface is even greater. If dust forms at 10
times the stellar radius, then some other mechanism must lift gas 90% of the
way out of the star’s gravitational well before it even encounters the dust.
Winds from cool, luminous stars might be driven by Alfvén waves prop-
agating outward in the atmosphere of a magnetized star. Large amplitude
Alfvén waves (δB ∼ B) are observed in the sun’s atmosphere, so it is not im-
plausible that such waves may propagate through other stellar atmospheres
as well. The rate at which Alfvén waves deposit momentum in the star’s at-
mosphere depends on whether the waves are adiabatic (undamped) or dissi-
pational (damped). Hartman & MacGregor, 1980 (ApJ, 242, 260) examined
models of Alfvén-wave driven winds. For a supergiant with M = 16 M⊙ ,
R = 400 R⊙ and a surface magnetic field of B = 10 G, the predicted terminal
velocity is u∞ = 400 km s−1 if the Alfvén waves are undamped. If the Alfvén
waves are heavily damped, then all of the wave energy goes into heating the
base of the atmosphere. If, however, the damping length of the Alfvén waves
is comparable to the radius of the star, then the observations can be matched.
For a 16 M⊙ , 400 R⊙ , 10G supergiant, with an Alfvén damping length of
L = R∗ = 400 R⊙ , the terminal velocity of the wind is u∞ = 50 km s−1 , and
the mass loss rate is Ṁ = 5 × 10−7 M⊙ yr−1 , in agreement with observed
values.
Another possible mechanism for driving winds from cool, luminous stars
is pulsationally driven, radially expanding shock waves in the stellar atmo-
sphere. Many M giants and supergiants are observed to be variable, with long
periods. For instance, Mira (an M giant) has a period of 330 days, during
which its luminosity varies by 8 magnitudes. The variability in luminosity
is due to the radial pulsation of the star. The pulsating star can be thought
of as a spherically symmetric piston, driving periodic shock waves into the
stellar atmosphere. Every year or so, before the gas has a chance to settle
down into hydrostatic equilibrium again, it is hit by another shock, and yet
Figure 12.3: The density (upper panel) and velocity (lower panel) for shock-
driven winds. On the left, the wind is assumed to be adiabatic; on the right,
the wind is assumed to be isothermal. [Wood 1979, ApJ, 227, 220]
more momentum and energy is added to the gas. Eventually, the outer layers
of the star are unbound, and flow outward in a wind. The properties of a
shock-driven wind, as illustrated in Figure 12.3, depend on whether the gas
is adiabatic or isothermal. A model star with M = M⊙ , L = 104 L⊙ and pul-
sational period P = 373days will have a mass loss rate of Ṁ = 0.02 M⊙ yr−1
if the gas is adiabatic, and a mass loss rate of Ṁ = 10−12 M⊙ yr−1 if the gas is
isothermal. The main difference in the two types of flow is that the adiabatic
case has much higher densities at large radii. A combination of isothermal
shock-driven flow (effective at small radii) and dust-driven flow (effective at
large radii) gives a mass loss rate of Ṁ = 3 × 10−7 and a terminal velocity
u∞ = 7 km s−1 , which are of the right magnitude for mass loss from a Mira
variable.
Chapter 13
Astrophysical Jets
In an astrophysical context, jets may be defined as well-collimated outflows

from a compact object. Jets are commonly bipolar, with two back-to-back
jets streaming away from the central compact object. There are two types
of astrophysical jets on which attention has been focused.
1. Young stellar jets are seen near newly forming stars. The length of a
young stellar jet is l = 0.01 → 1 pc, and the velocity of gas within the
jet is v = 100 → 400 km s−1 ; the central compact object is a newborn
star of mass M ∼ 1 M⊙ . A young stellar jet carries off mass at a rate
Ṁ = 10−9 → 10−6 M⊙ yr−1 .
2. Extragalactic jets are seen near the nuclei of radio galaxies and
quasars. The length of an extragalactic jet is l = 0.01 → 1 Mpc, and
the velocity of gas within the jet may be as high as v ∼ c; the central
compact object is (probably) a black hole of mass M = 108 → 109 M⊙ .
Despite the very different length scales and velocities for the two types
of jets, the basic physics involved is the same. Long, highly collimated flows
originate in a compact object, and appear to be perpendicular to an ac-
cretion disk. In addition to a force that counteracts the gravitational force
of the central compact object, the formation of a jet requires a ‘nozzle’ to
shape the gas flow into a narrow jet. Theorists have divided the study of
astrophysical jets into two parts. (1) What forces act to accelerate and col-
limate the beam? (2) How does the jet propagate through space once it has
been accelerated and collimated? The first question is the more difficult one.
Jets are presumably accelerated by one of the mechanisms that accelerate
133
134 CHAPTER 13. ASTROPHYSICAL JETS
winds: jets might be pressure-driven, radiation-driven, Alfven-wave-driven,

or shock-driven. No one is sure how jets are collimated; by the time they
are visible to observers, they are already tightly collimated. Because of our
vast ignorance of how jets are born, theoretical studies have focused on the
question of how jets propagate through space once they are born.
Young stellar jets have more observational data than extragalactic jets
to constrain any models that we may make. For instance, spectroscopy of
stellar jets gives information about the bulk velocity, velocity dispersion,
electron density, and line excitation as a function of position along the jet.
Extragalactic jets, however, do not emit line radiation, so we cannot use
Doppler shifts and Doppler widths to measure the bulk velocity and velocity
dispersion within the jet. We will focus our attention on the better-known
young stellar jets.
Stars form in molecular clouds, where the temperature is T ≈ 10 K, and
the sound speed is a ≈ 0.3 km s−1 . The molecular clouds contain dense cores
that are supported by magnetic pressure. As the magnetic field slowly dif-
fuses outward, the core collapses to form a rotating protostar, which accretes
matter both from an accretion disk and from a roughly spherical infall of gas
and dust. When the protostar becomes sufficiently massive, deuterium burn-
ing ignites, and a stellar wind starts to push its way outward. At first, the
stellar wind is contained by the infalling matter, which has been compared
to the lid of a pressure cooker. Eventually, the ram pressure of the stellar
wind breaks through the ‘lid’ at its weakest points, the rotational poles of
the protostar (where the column depth of accreting matter is smallest). The
next stage is a young stellar object, with jets flowing outward along the two
rotational poles, and an accretion disk lying in the equator. The opening
angle of the jet widens with time, until the young stellar object becomes an
optically visible T Tauri star.
The jet associated with a young stellar object remains tightly collimated
for some length, then expands outward in a ‘stellar wind cavity’. The cavity
ends at a working surface, where the jet flow is decelerated by a shock. The
basic morphology of a young stellar jet is shown in Figure 13.1. The shock
heating at the working surface creates a Herbig-Haro object, which is a
low-excitation emission-line nebula. The working surface of the jet moves
outward at a velocity uw with respect to the central star. The velocity uw
depends on the bulk velocity uj of matter within the jet, and on the ratio
η ≡ ρj /ρ0 , where ρj is the gas density within the jet and ρ0 is the density
of the ambient medium. In a frame of reference fixed to the working surface
135
Figure 13.1: The components of a typical jet from a young stellar object.
of the jet, the ambient gas is moving with a velocity −uw and the jet gas
is moving with a velocity uj − uw . Since the working surface is stationary
in this frame, the ram pressure of the gas on the two sides of the working
surface must be equal. That is,
ρj (uj − uw )2 = ρ0 u2w . (13.1)
This yields a quadratic equation for uw ,
(η − 1)u2w − (2ηuj )uw + ηu2j = 0 , (13.2)
of which the positive root is
η 1/2
uw = uj . (13.3)
1 + η 1/2
For heavy jets, with η ≫ 1, the working surface moves outward with a
velocity uw ≈ uj . For light jets, with η ≪ 1, the working surface moves
outward with the low velocity uw ≈ uj η 1/2 . Observations of the velocities of
young stellar jets, and of the Herbig-Haro objects at their working surfaces,
indicate that η ∼ 1 − 2. The density of hydrogen atoms in the stellar jet and
in the ambient molecular cloud are n ∼ 50 cm−3 .
The age of a young stellar jet, as deduced from its expansion velocity, is
! "# $−1
3 R uw
tj ≈ 10 yr , (13.4)
0.1 pc 100 km s−1
where R is the length of the jet. The mass of the jet is
! "# $−1 ! "
R uw Ṁ
Mj ≈ 10−4 M⊙ , (13.5)
0.1 pc 100 km s−1 10 M⊙ yr−1
−7
and its kinetic energy is

! "# $! "
43 R uw Ṁ
Ej ≈ 10 erg , (13.6)
0.1 pc 100 km s−1 10−7 M⊙ yr−1
as long as η >
∼ 1 and uw ≈ uj . During the jet’s lifetime, the energy carried
away from the star by photons is
! "# $−1 ! "
44 R uw L
Ltj ≈ 10 erg , (13.7)
0.1 pc 100 km s−1 1 L⊙
so the energy carried away by the jet can be comparable to the energy carried
away by radiation.
A bit of gas that is ejected into the jet moves away from the star at a
velocity uj until it reaches the working surface. The gas is then shocked and
heated, and enters a cocoon of hot gas that surrounds the jet. Conservation
of mass implies that the mass of the cocoon increases at the rate Ṁcoc =
ρj (uj − uw )Aj , where Aj is the area of the working surface. The cocoon
increases its mass at the rate
ρj uj Aj
Ṁcoc = , (13.8)
1 + η 1/2
while the jet itself increases its mass at the rate
ρj uj Aj
Ṁj = ρj uw Aj = η 1/2 . (13.9)
1 + η 1/2
If the rates of mass gain are constant, the ratio of the cocoon mass to the jet
mass is
Mcoc /Mj = η −1/2 . (13.10)
137
In addition to the dependence on the density ratio η, the structure of the jet
also depends on its Mach number Nj = uj /aj , where aj = (γPj /ρj )1/2 s. Jets
with low Mach numbers have cocoons that extend only a short distance from
the working surface; jets with high Mach numbers have cocoons that extend
for the entire length of the jet. Figure 13.2 shows a series of jet simulations;
all the jets have Mach number mj = 6, but their value of η ranges from
η = 10 (no cocoon worth mentioning) to η = 0.01 (a cocoon much larger
than the central jet). Figure 13.3 shows a series of jet simulations in which
all the jets have η = 0.1, but their Mach number ranges from mj = 1.5 (short
cocoon) to mj = 12 (cocoon as long as the jet).
Since the jet has an internal pressure Pj , what keeps the jet from expand-
ing perpendicular to its long axis? In non-magnetic jets, the confinement is
provided by the external pressure Pex . If the pressure of the external medium
is constant, then the diameter d of the jet is roughly constant as a function of
the distance r along the jet. If, however, the external pressure Pex varies with
distance along the jet, the diameter of the jet will vary in order to maintain
pressure equilibrium at the sides of the jet. Suppose that the gas within the
jet is polytropic, with
Pj = P0 (ρj /ρ0 )γ . (13.11)
If the external pressure has a power-law dependence on r,
Pex = P0 (r/r0 )−n , (13.12)
then the requirement that Pj = Pex yields a density profile
ρj (r) = ρ0 (r/r0 )−n/γ . (13.13)
If the mass flux Ṁ and the bulk velocity uj are constant along the jet, then
the mass continuity relation
π
Ṁ = d(r)2 ρj (r)uj (13.14)
4
tells us that ! "1/2 # $n/(2γ)
4Ṁ r
d(r) = . (13.15)
πuj ρ0 r0
The opening angle θ of the jet is θ ≈ d/r as long as d ≪ r. When the jet is
narrow,
# $(n−2γ)/(2γ)
r
θ(r) = θ0 . (13.16)
r0
Figure 13.2: A sequence of simulated mj = 6 jets, with η = 10, 1, 0.1, and

0.01 (top to bottom).
139
Figure 13.3: A sequence of simulated η = 0.1 jets, with mj = 1.5, 3, 6, and

12 (top to bottom).
The Mach number mj of the jet is

! "−1/2 $n(γ−1)/(2γ)
γPj r
#
m j = uj = m0 . (13.17)
ρj r0
When γ > 1, the Mach number of the jet increases as the external pressure
decreases.
If the jet has a very high Mach number, or is very broad, it can no longer
be confined by the pressure of the ambient medium. Place yourself in a
frame of reference traveling along the axis of symmetry of the jet with the
gas velocity uj . In this frame of reference, the walls of the jet are moving
away from you with a velocity
! "
θ θ
uwall ≈ uj tan ≈ uj . (13.18)
2 2
Information about changes in the external pressure are carried toward you
from the wall of the jet with a velocity equal to the sound speed aj . In
order for the jet to remain confined by the pressure of the outside gas, it
must remain in causal contact with the gas outside the walls of the jet. This
requires that uwall < aj , or, in other words, that θmj < 2. In our model with
Pex ∝ r−n , the product of the opening angle θ and the Mach number mj is
$(n−2)/2
r
#
θmj = θ0 m0 . (13.19)
r0
Thus, when n > 2, there will be a critical radius rf at which θmj = 2; beyond
this point, the jet will no longer be confined by the pressure of the ambient
medium. It will be a free jet. Free jets have a constant opening angle
2
θf = . (13.20)
mj (rf )
The diameter of the free jet will increase at the rate d ∝ r, the density will
decrease at the rate ρj ∝ r−2 , and the Mach number will increase at the rate
mj ∝ rγ−1 . Some extragalactic jets have a constant opening angle, and are
conjectured to be free jets. One such jet is seen in NGC 6251 (Figure 13.4).
Radiative cooling can have an important effect on the structure of stellar
jets. Cooling creates a dense, cool shell at the head of the jet. The density
within the shell, simulations indicate, can be as much as 100 times the density
141
Figure 13.4: The jet of the radio galaxy NGC 6251, seen at radio frequencies
at over a wide range of angular resolutions.
Figure 13.5: A numerical simulation of the growth of jet instabilities. Note

that periodic wiggles grow into turbulent eddies. [Norman & Hardee (1988)
APJ, 334, 80]
of the ambient medium. This radiating, dense shell can be identified with
Herbig-Haro objects seen at the ends of young stellar jets. Cooling also
decreases the size and the pressure of the cocoon.
Jets are not perfectly smooth and straight; they show a tendency to
break up into knots and to deviate into ‘wiggles’. These knots and wiggles
can occur because the jet is subject to instabilities that are analogous to the
classic Kelvin-Helmholtz instability. Figure 13.5 shows the development of
jet wiggles in a numerical simulation. Numerical studies indicate that there
exist growing modes of instability for jets with any Mach number. Some
of these modes are axisymmetric, pinching modes (these would tend to give
rise to observed knots) and some modes are nonaxisymmetric helical modes
(these would tend to give rise to observed wiggles). The fastest growing
modes are those with wavelengths comparable to the diameter d of the jet
(this is why knots are roughly circular and wiggles have radii of curvature
roughly equal to the jet width).

Radiative Gas Dynamics - Barbara Ryden

Uploaded by

Copyright:

Available Formats

Radiative Gas Dynamics - Barbara Ryden

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Radiative Gas Dynamics - Barbara Ryden

Uploaded by

Copyright:

Available Formats

ASTRONOMY 825: Radiative Gas Dynamics

Winter Quarter 2011 · Tu Th 1:30 - 2:48 pm · McPherson 4054

Instructor: Professor Barbara Ryden

Text: Radiative Gas Dynamics, by Barbara Ryden – a set of lecture notes

Class website: www.astronomy.ohio-state.edu/∼ryden/ast825/

Grading policy: Problem sets will be assigned on alternate Thursdays, starting

Academic Misconduct: It is the responsibility of the Committee on Academic

Tentative Course Outline: Astronomy 825

• Chapter 2: Viscosity & Heat Conduction (Jan 11)

• Chapter 3: Sound & Shocks (Jan 13, 18)

• Chapter 4: Radiative & MHD Shocks (Jan 20)

• Chapter 5: Spherical Blastwaves & Supernova Remnants (Jan 25)

• Chapter 6: Ionization Fronts & HII Regions (Jan 27, Feb 1)

• Chapter 7: Basic Turbulence (Feb 3)

• Chapter 8: Spherical Accretion (Feb 8, 10)

• Chapter 9: Accretion Disks for Beginners (Feb 15, 17)

• Chapter 10: Advanced Accretion Disks (Feb 22)

• Chapter 11: Solar Wind (Feb 24, Mar 1)

• Chapter 12: Winds from Hot & Cool Stars (Mar 3, 8)

• Chapter 13: Astrophysical Jets (Mar 10)

Notes on the Notes v

1 Fundamentals of Gas Dynamics 1

2 Viscosity & Heat Conduction 13

3 Sound & Shocks 21

4 Radiative & MHD Shocks 35

5 Spherical Blastwaves & SNRs 45

9 Accretion Disks for Beginners 89

10 Advanced Accretion Disks 101

11 The Solar Wind 111

12 Winds from Hot & Cool Stars 123

13 Astrophysical Jets 133

Figure 1: Turbulent fluids as drawn by Leonardo da Vinci.

Fundamentals of Gas Dynamics

Gas dynamics is the study of continuous compressible fluids in motion.

T ∼ 90 K and n ∼ 50 cm−3 . (1.8)

T ∼ 8000 K and n ∼ 0.5 cm−3 . (1.9)

The Warm Ionized Medium consists of partially ionized gas (H I, He I,

T ∼ 104 K and n ∼ 0.1 cm−3 . (1.10)

T ∼ 106 K and n ∼ 5 × 10−6 cm−3 . (1.13)

The intracluster medium (ICM) in dense clusters of galaxies can have

T ∼ 108 K and n ∼ 10−3 cm−3 . (1.14)

Although the intracluster medium is thousands of times higher in pressure

bound-bound electronic transitions. Suﬃciently energetic radiation can pho-

where N is the total number of particles in the system. If we choose, we can

fm (⃗x, ⃗v , t) = mf (⃗x, ⃗v , t) , (1.16)

where m is the mass of an individual particle. If n types of particle are

where mk is the mass of the k th type of particle. If particles are neither

This equation is known as the Boltzmann equation.

For any measurable quantity Q, the mass-weighted average value at a position

∂ ! " ∂ ! " ! ∂fm

Now introduce a new symbol:

where w⃗ is the random velocity.

Now we use the decomposition ⃗v = ⃗u + w ⃗ on the left, and the divergence

This is the equation of momentum conservation. It is convenient to write

ρ⟨wi wj ⟩ = P δij − πij , (1.32)

where P is the pressure,

πij ≡ P δij − ρ⟨wi wj ⟩ . (1.34)

The equation can be written in tensor form:

The tensor product of the velocities is