You are on page 1of 216

Theoretical Astrophysics

Matthias Bartelmann
Institut für Theoretische Astrophysik
Universität Heidelberg
2
Inhaltsverzeichnis

1 Macroscopic Radiation Quantities, Emission and Absorpti-


on 1
1.1 Specific Intensity . . . . . . . . . . . . . . . . . . . . 1
1.2 Relativistic Invariant . . . . . . . . . . . . . . . . . . 2
1.2.1 Lorentz Transformation of Iν . . . . . . . . . . 2
1.2.2 Example: The CMB Dipole . . . . . . . . . . 4
1.3 Einstein coefficients and the Planck spectrum . . . . . 5
1.3.1 Transition Balance . . . . . . . . . . . . . . . 5
1.3.2 Example: The CMB Spectrum . . . . . . . . . 6
1.4 Absorption and Emission . . . . . . . . . . . . . . . . 7
1.5 Radiation Transport in a Simple Case . . . . . . . . . 8
1.6 Emission and Absorption in the Continuum Case . . . 10

2 Scattering 13
2.1 Maxwell’s Equations and Units . . . . . . . . . . . . . 13
2.2 Radiation of a Moving Charge . . . . . . . . . . . . . 14
2.3 Scattering off Free Electrons . . . . . . . . . . . . . . 15
2.3.1 Polarised Thomson Cross Section . . . . . . . 15
2.3.2 Unpolarised Thomson Cross Section . . . . . . 16
2.4 Scattering off Bound Charges . . . . . . . . . . . . . . 17
2.5 Radiation Drag . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 Time-Averaged Damping Force . . . . . . . . 19
2.5.2 Energy Transfer to a Radiation Field . . . . . . 20
2.6 Compton Scattering . . . . . . . . . . . . . . . . . . . 21

3
4 INHALTSVERZEICHNIS

2.6.1 Energy-Momentum Conservation . . . . . . . 21


2.6.2 Energy Balance . . . . . . . . . . . . . . . . . 22
2.7 The Kompaneets Equation . . . . . . . . . . . . . . . 24

3 Radiation Transport and Bremsstrahlung 27


3.1 Radiation Transport Equations . . . . . . . . . . . . . 27
3.2 Local Thermodynamical Equilibrium . . . . . . . . . . 29
3.3 Scattering . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Bremsstrahlung . . . . . . . . . . . . . . . . . . . . . 34
3.4.1 Spectrum of a Moving Charge . . . . . . . . . 34
3.4.2 Hyperbolic Orbits . . . . . . . . . . . . . . . . 35
3.4.3 Integration over the Electron Distribution . . . 37

4 Synchrotron Radiation, Ionisation and Recombination 41


4.1 Synchrotron Radiation . . . . . . . . . . . . . . . . . 41
4.1.1 Electron Gyrating in a Magnetic Field . . . . . 41
4.1.2 Beaming and Retardation . . . . . . . . . . . . 42
4.1.3 Synchrotron Spectrum . . . . . . . . . . . . . 44
4.2 Photo-Ionisation . . . . . . . . . . . . . . . . . . . . . 45
4.2.1 Transition Amplitude . . . . . . . . . . . . . . 45
4.2.2 Transition Probability . . . . . . . . . . . . . 46
4.2.3 Transition Matrix Element . . . . . . . . . . . 48
4.2.4 Cross Section . . . . . . . . . . . . . . . . . . 50

5 Spectra 53
5.1 Natural Width of Spectral Lines . . . . . . . . . . . . 53
5.2 Cross Sections and Oscillator Strengths . . . . . . . . 53
5.2.1 Transition Probabilities . . . . . . . . . . . . . 54
5.3 Collisional Broadening of Spectral Lines . . . . . . . . 56
5.4 Velocity Broadening of Spectral Lines . . . . . . . . . 57
5.5 The Voigt Profile . . . . . . . . . . . . . . . . . . . . 58
INHALTSVERZEICHNIS 5

5.6 Equivalent Widths and Curves-of-Growth . . . . . . . 59

6 Energy-Momentum Tensor and Equations of Motion 63


6.1 Boltzmann Equation and Energy-Momentum Tensor . 63
6.1.1 Boltzmann Equation . . . . . . . . . . . . . . 63
6.1.2 Moments; Continuity Equation . . . . . . . . . 64
6.1.3 Energy-Momentum Tensor . . . . . . . . . . . 66
6.2 The Tensor Virial Theorem . . . . . . . . . . . . . . . 69
6.2.1 A Corollary . . . . . . . . . . . . . . . . . . . 69
6.2.2 Second Moment of the Mass Distribution . . . 70

7 Ideal and Viscous Fluids 73


7.1 Ideal Fluids . . . . . . . . . . . . . . . . . . . . . . . 73
7.1.1 Energy-Momentum Tensor . . . . . . . . . . . 73
7.1.2 Equations of Motion . . . . . . . . . . . . . . 75
7.1.3 Entropy . . . . . . . . . . . . . . . . . . . . . 77
7.2 Viscous Fluids . . . . . . . . . . . . . . . . . . . . . . 78
7.2.1 Stress-Energy Tensor; Viscosity and Heat Con-
ductivity . . . . . . . . . . . . . . . . . . . . 78
7.2.2 Estimates for Heat Conductivity and Viscosity 80
7.2.3 Equations of Motion for Viscous Fluids . . . . 82
7.2.4 Entropy . . . . . . . . . . . . . . . . . . . . . 83
7.3 Generalisations . . . . . . . . . . . . . . . . . . . . . 84
7.3.1 Additional External Forces; Gravity . . . . . . 84
7.3.2 Example: Cloud in Pressure Equilibrium . . . 85
7.3.3 Example: Self-Gravitating Gas Sphere . . . . . 86

8 Flows of Ideal and Viscous Fluids 89


8.1 Flows of Ideal Fluids . . . . . . . . . . . . . . . . . . 89
8.1.1 Vorticity and Kelvin’s Circulation Theorem . . 89
8.1.2 Bernoulli’s Constant . . . . . . . . . . . . . . 91
8.1.3 Hydrostatic Equlibrium . . . . . . . . . . . . . 92
6 INHALTSVERZEICHNIS

8.1.4 Curl-Free and Incompressible Flows . . . . . . 93


8.2 Flows of Viscous Fluids . . . . . . . . . . . . . . . . . 94
8.2.1 Vorticity; Incompressible Flows . . . . . . . . 94
8.2.2 The Reynolds Number . . . . . . . . . . . . . 95
8.3 Sound Waves in Ideal Fluids . . . . . . . . . . . . . . 96
8.3.1 Linear Perturbations . . . . . . . . . . . . . . 96
8.3.2 Sound Speed . . . . . . . . . . . . . . . . . . 98
8.4 Supersonic Flows . . . . . . . . . . . . . . . . . . . . 99
8.4.1 Mach’s Cone; the Laval Nozzle . . . . . . . . 99
8.4.2 Spherical Accretion . . . . . . . . . . . . . . . 100

9 Shock Waves and the Sedov Solution 105


9.1 Steepening of Sound Waves . . . . . . . . . . . . . . . 105
9.1.1 Formation of a Discontinuity . . . . . . . . . . 105
9.1.2 Specific Example . . . . . . . . . . . . . . . . 107
9.2 Shock Waves . . . . . . . . . . . . . . . . . . . . . . 110
9.2.1 The Shock Jump Conditions . . . . . . . . . . 110
9.2.2 Propagation of a One-Dimensional Shock Front 111
9.2.3 The Width of a Shock . . . . . . . . . . . . . 114
9.3 The Sedov Solution . . . . . . . . . . . . . . . . . . . 115
9.3.1 Dimensional Analysis . . . . . . . . . . . . . 115
9.3.2 Similarity Solution . . . . . . . . . . . . . . . 117

10 Instabilities, Convection, Heat Conduction, Turbulence 121


10.1 Rayleigh-Taylor Instability . . . . . . . . . . . . . . . 121
10.2 Kelvin-Helmholtz Instability . . . . . . . . . . . . . . 124
10.3 Thermal Instability . . . . . . . . . . . . . . . . . . . 126
10.4 Heat Conduction and Convection . . . . . . . . . . . . 130
10.4.1 Heat conduction . . . . . . . . . . . . . . . . 130
10.4.2 Convection . . . . . . . . . . . . . . . . . . . 132
10.5 Turbulence . . . . . . . . . . . . . . . . . . . . . . . 133
INHALTSVERZEICHNIS 7

11 Collision-Less Plasmas 135


11.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . 135
11.1.1 Shielding; the Debye length . . . . . . . . . . 135
11.1.2 The plasma frequency . . . . . . . . . . . . . 136
11.2 The Dielectric Tensor . . . . . . . . . . . . . . . . . . 137
11.2.1 Polarisation and dielectric displacement . . . . 137
11.2.2 Structure of the dielectric tensor . . . . . . . . 138
11.3 Dispersion Relations . . . . . . . . . . . . . . . . . . 139
11.3.1 General form of the dispersion relations . . . . 139
11.3.2 Transversal and longitudinal waves . . . . . . 140
11.4 Longitudinal Waves . . . . . . . . . . . . . . . . . . . 141
11.4.1 The longitudinal dielectricity . . . . . . . . . . 141
11.4.2 Landau Damping . . . . . . . . . . . . . . . . 143
11.5 Waves in a Thermal Plasma . . . . . . . . . . . . . . . 144
11.5.1 Longitudinal and transversal dielectricities . . 144
11.5.2 Dispersion Measure and Damping . . . . . . . 147

12 Magneto-Hydrodynamics 149
12.1 The Magneto-Hydrodynamic Equations . . . . . . . . 149
12.1.1 Assumptions . . . . . . . . . . . . . . . . . . 149
12.1.2 The induction equation . . . . . . . . . . . . . 150
12.1.3 Euler’s equation . . . . . . . . . . . . . . . . 151
12.1.4 Energy and entropy . . . . . . . . . . . . . . . 153
12.1.5 Magnetic advection and diffusion . . . . . . . 154
12.2 Generation of Magnetic Fields . . . . . . . . . . . . . 155
12.3 Ambipolar Diffusion . . . . . . . . . . . . . . . . . . 157
12.3.1 Scattering cross section . . . . . . . . . . . . . 157
12.3.2 Friction force; diffusion coefficient . . . . . . . 159

13 Waves in Magnetised Plasmas 161


13.1 Waves in magnetised cold plasmas . . . . . . . . . . . 161
8 INHALTSVERZEICHNIS

13.1.1 The dielectric tensor . . . . . . . . . . . . . . 161


13.1.2 Contribution by ions . . . . . . . . . . . . . . 163
13.1.3 General dispersion relation . . . . . . . . . . . 164
13.1.4 Wave propagation parallel to the magnetic field 165
13.1.5 Faraday rotation . . . . . . . . . . . . . . . . 166
13.1.6 Wave propagation perpendicular to the magnetic
field . . . . . . . . . . . . . . . . . . . . . . . 167
13.2 Hydro-Magnetic Waves . . . . . . . . . . . . . . . . . 168
13.2.1 Linearised perturbation equations . . . . . . . 168
13.2.2 Alfvén waves . . . . . . . . . . . . . . . . . . 170
13.2.3 Slow and fast hydro-magnetic waves . . . . . . 171

14 Jeans Equations and Jeans Theorem 173


14.1 Collision-less motion in a gravitational field . . . . . . 173
14.1.1 Motion in a gravitational field . . . . . . . . . 173
14.1.2 The relaxation time scale . . . . . . . . . . . . 174
14.2 The Jeans Equations . . . . . . . . . . . . . . . . . . 177
14.2.1 Moments of Boltzmann’s equation . . . . . . . 177
14.2.2 Jeans equations in cylindrical and spherical coor-
dinates . . . . . . . . . . . . . . . . . . . . . 179
14.2.3 Application: the mass of a galaxy . . . . . . . 180
14.3 The Virial Equations . . . . . . . . . . . . . . . . . . 181
14.3.1 The tensor of potential energy . . . . . . . . . 181
14.3.2 The tensor virial theorem . . . . . . . . . . . . 182
14.4 The Jeans Theorem . . . . . . . . . . . . . . . . . . . 184

15 Equilibrium, Stability and Disks 187


15.1 The Isothermal Sphere . . . . . . . . . . . . . . . . . 187
15.1.1 Phase-space distribution function . . . . . . . 187
15.1.2 Isothermality . . . . . . . . . . . . . . . . . . 188
15.1.3 Singular and non-singular solutions . . . . . . 189
15.2 Equilibrium and Relaxation . . . . . . . . . . . . . . . 190
INHALTSVERZEICHNIS 9

15.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . 191


15.3.1 Linear analysis and the Jeans swindle . . . . . 191
15.3.2 Jeans length and Jeans mass . . . . . . . . . . 193
15.4 The rigidly rotating disk . . . . . . . . . . . . . . . . 194
15.4.1 Equations for the two-dimensional system . . . 194
15.4.2 Analysis of perturbations . . . . . . . . . . . . 195
15.4.3 Toomre’s criterion . . . . . . . . . . . . . . . 196

16 Dynamical Friction, Fokker-Planck Approximation 199


16.1 Dynamical Friction . . . . . . . . . . . . . . . . . . . 199
16.1.1 Deflection of point masses . . . . . . . . . . . 199
16.1.2 Velocity changes . . . . . . . . . . . . . . . . 201
16.1.3 Chandrasekhar’s formula . . . . . . . . . . . . 202
16.2 Fokker-Planck Approximation . . . . . . . . . . . . . 204
16.2.1 The master equation . . . . . . . . . . . . . . 204
16.2.2 The Fokker-Planck equation . . . . . . . . . . 205
10 INHALTSVERZEICHNIS
Kapitel 1

Macroscopic Radiation
Quantities, Emission and
Absorption
further reading: Shu, “The Phy-
sics of Astrophysics, Vol I: Radia-
1.1 Specific Intensity tion”, chapter 1; Rybicki, Light-
man, “Radiative Processes in
• to begin with, radiation is considered as a stream of particles; Astrophysics”, chapter 1; Pad-
energy, momentum and so on of this stream will be investigated as manabhan, “Theoretical Astro-
well as changes of its properties; physics, Vol. I: Astrophysical
Processes”, sections 6.1–6.3
• a screen of area dA is set up; which energy is streaming per time
interval dt enclosing the angle θ with the direction normal to the
screen into the solid angle element dΩ and within the frequency
interval dν?

• we begin with the occupation number: let nα~p be the number the occupation number is the
density of photons with momentum ~p and the polarisation state α number density of occupied
(α = 1, 2); states per phase space element

• the energy per photon is E = hν = cp (because the photon has


zero rest mass); thus


p = |~p| = ; (1.1)
c

• the volume element in momentum space is d3 p; the number of


independent phase-space cells is Heisenberg’s uncertainty relation
implies that points in phase space
d3 p p2 dpdΩ ν2 dνdΩ cannot be observed; rather, obser-
= = , (1.2)
(2π~)3 h3 c3 vable cells in phase space have a
finite volume
where momentum has been expressed by frequency ν in the last
step;

1
2KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

• in terms of these quantities, the following amount of energy is


flowing through the screen: (number of phase space cells) times
(photon occupation number) times (energy per photon) times (vo-
lume filled by the photons); thus
2
ν2 dνdΩ X
dE = nα~p hν dA cos θ dt (1.3)
c3 α=1

• the energy flowing through the screen per unit time, frequency and
solid angle is
2
dE X hν3
= nα~p 2 cos θ ≡ Iν cos θ , (1.4)
dtdνdAdΩ α=1 c

where Iν is called specific intensity of the radiation;

• for unpolarised light, we obviously have


2hν3
Iν = nα~p ; (1.5)
c2

1.2 Relativistic Invariant

1.2.1 Lorentz Transformation of Iν


• switching from one reference frame to another, the transformation
properties of the physical quantities is important to be known; we
shall now show by Lorentz transformation that the quantity

(1.6)
ν3
is relativistically invariant;

• let us assume two observers O and O0 , which are moving relative-


ly to each other with velocity v in x3 direction; O0 is collecting
photons on a screen dA0 in the x10 -x20 plane which move under the
angle θ0 with respect to the area normal into the solid angle dΩ0 ;
he finds
p02 dp0 dΩ0 0
dN 0 = 2 3
n p0 dA0 c cos θ0 dt0 (1.7)
(2π~)
photons on his screen;

• likewise, observer O expects the same screen to collect the photon


number
p2 dpdΩ
dN = 2 n p dA(c cos θ − v) dt0 (1.8)
(2π~)3
and of course the two numbers must be equal, dN 0 = dN;
1.2. RELATIVISTIC INVARIANT 3

• the Lorentz transformation relating O and O0 is


 γ 0 0 βγ 
 
 0 1 0 0 
Λ =   (1.9)
 0 0 1 0 
βγ 0 0 γ

with β ≡ v/c and γ ≡ (1 − β2 )−1/2 ;


• for the screen at rest in O0 , dx30 = 0, thus

dx0 = cdt = γ dx00 = γ cdt0 , (1.10)

so that dt = γ dt0 , which is the usual relativistic time dilation;


• energy and momentum are combined in the four-vector
E   E
pµ =

, ~p ≡ p0 , ~p ; p0 = |~p| because |~p| = , (1.11)
c c
and we obtain

p0 = γ(p00 + βp03 ) ; p3 = γ(βp00 + p03 ) , (1.12)

and the other components are p1 = p01 , p2 = p02 ;


• since, from simple geometry, p03 = cos θ0 |~p0 | = p00 cos θ0 and
p3 = p0 cos θ, we then find
p3 βp00 + p03 β + cos θ0
cos θ = = = (1.13)
p0 p00 + βp03 1 + β cos θ0
for the Lorentz transformation of the angle θ;
• this implies for the solid-angle element
β + cos θ0 d 2 Ω0
" #
d Ω = d(cos θ)dφ = dφ d
2 0
= ;
1 + β cos θ0 γ2 (1 + β cos θ0 )2
(1.14)
• summarising, we find for the number of photons in the system O:
2  d 2 Ω0
dN = γ(1 + β cos θ 0 3 02
) p dp 0
h3 | {z } γ2 (1 + β cos θ0 )2
=p dp
2 | {z }
=d2 Ω
β + cos θ0
" ! #
× n p |{z}0
dA c − v γdt0 ; (1.15)
1 + β cos θ0 |{z}
=dA | {z } =dt
=c cos θ−v

equating this to dN 0 from (1.7) yields


β + cos θ0
" !#
n p0 cos θ = γ (1 + β cos θ )
0 0 2 0
− β np
1 + β cos θ0
= γ2 (1 − β2 ) cos θ0 n p = n p cos θ0 ; (1.16)
4KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

• thus, the occupation number is obviously a relativistic invariant,


n p = n0p0 , (1.17)
and Iν ∝ ν n p implies the claimed invariance (1.6),
3

Iν Iν00
= ; (1.18)
ν3 ν03
the Lorentz transformation of the solid angle (1.14) will be used
later in the discussion of synchrotron radiation

1.2.2 Example: The CMB Dipole


• this relativistic invariance of Iν /ν3 allows the dipole of the cosmic
microwave background to be computed: a photon flying at an angle
θ relative to the x axis of the observer will be redshifted by an
amount which directly follows from Lorentz transformation;
• using pµ = (E/c, ~p) and p1 = |~p| cos θ, the Lorentz transformation
yields
 γ βγ 0 0   E/c 
  
 βγ γ 0 0   p1 
p0µ =    
 0 0 1 0   p2 
0 0 0 1 p3
 γE/c + βγ|~p| cos θ 
 
 βγE/c + γ|~p| cos θ 
=  2
 , (1.19)
 p 

p3

i.e. the energy in the primed system is


 E E 
E = c γ + βγ cos θ = γ(1 + β cos θ)E ;
0
(1.20)
c c
the frequency is thus increased to
ν0 = γ(1 + β cos θ)ν ; (1.21)
• with the occupation number n p being a relativistic invariant,
1 1
n p = hν/kT = hν0 /kT 0 = n0p0 , (1.22)
e −1 e −1
the temperature T must change exactly as the frequency ν, thus
T 0 = T γ(1 + β cos θ) ; (1.23)
• for non-relativistic velocities v  c, γ ≈ 1, and thus
 v 
T 0 ≈ T 1 + cos θ ; (1.24)
c
the motion of the Earth relative to the microwave background
thus causes a dipolar pattern in its measured temperature; with
v ∼ 10−3 c and T ∼ 3 K, the amplitude of the dipole is of order a
few milli-Kelvins;
1.3. EINSTEIN COEFFICIENTS AND THE PLANCK SPECTRUM5

1.3 Einstein coefficients and the Planck spec-


trum

1.3.1 Transition Balance


• we consider mean transition rates in an emission- and absorption
process between two energy levels E1 and E2 ; the rates of absorp-
tion and of stimulated emission will be proportional to the specific stimulated emission is a conse-
intensity, absorption rate ∝ Iν B12 and stimulated emission rate ∝ quence of the Bose character of
Iν B21 , while the rate of spontaneous emission will not depend on Iν , photons: if a quantum state is oc-
spontaneous emission rate ∝ A21 ; A and B are called the Einstein cupied by photons, an increase in
coefficients; the occupation number is more
likely
• now, let N1 and N2 be the mean number of states with the energies
E1 and E2 ; equilibrium between transitions will require as many
transitions from E1 to E2 as there are from E2 to E1 , thus

N1 Iν B12 = N2 [A21 + Iν B21 ] , (1.25)

which can be satisfied if the specific intensity is


N2 A21 A21 A21
Iν = = N1
= N , (1.26)
N1 B12 − N2 B21 B
N2 12
− B21 B21 N1 − 1
2

where we have used that B12 = B21 (E1 and E2 are eigenstates of
the Hamilton operator);
• according to the definition of A21 and B21 , we must have [cf. Eq. (1.5)]
hν3
A21 = 2 B21 ; (1.27)
c2
• if there is thermal equilibrium between the states E1 and E2 , we
have the Boltzmann factor between N1 and N2 ,
!
N2 hν
= exp − , (1.28)
N1 kT
where E2 = E1 + hν;
• under this condition, (1.28) implies
2hν3 1
Iν = 2 hν/kT ≡ Bν , (1.29)
c e −1
which is the Planck spectrum;
• limiting cases of the Planck spectrum for high and low frequencies
are
2hν3 −hν/kT kT
Bν ≈ e for ν (Wien’s law) (1.30)
c2 h
6KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

and
2ν2 kT
Bν ≈ kT for ν  (Rayleigh-Jeans law) (1.31)
c2 h
• the spectral energy density is
dE dE Iν
dUν = 3
= = d2 Ω , (1.32)
dνdx dνdA(cdt) c
thus Z
cUν = Iν d2 Ω , (1.33)

which equals 4πIν for isotropic radiation;


• a unit for the spectral energy density which is frequently used in
note that 1 Jy is not the unit of astronomy is the Jansky, defined by
specific intensity, which would W erg
be Jy/sr 1 Jy = 10−26 2
= 10−23 ; (1.34)
m Hz cm2 s Hz

1.3.2 Example: The CMB Spectrum


• for example, the spectral energy density of the CMB is given by
4π 4π 2hν3 1
Uν = Bν = 2 hν/kT
c c c e −1
erg
= 2.4 × 10 −25
= 23.9 mJy (1.35)
cm2 s Hz
at a frequency of ν = 30 GHz;
• the maximum of the Planck spectrum is located at

x≡ ≈ 2.82 ; (1.36)
kT
for the CMB, this corresponds to a frequency of
ν ≈ 1.60 × 1011 Hz = 160 GHz ; (1.37)

• inserting the Planck spectrum for Iν in (1.28) and integrating over


all frequencies yields
π2 (kT )4
Z ∞
U= Uν dν = (1.38)
0 15 (~c)3
for the energy density of a Planckian radiation field;
• the number density of the photons is clearly
Z ∞ !3
Uν 2ζ(3) kT
n= dν = , (1.39)
0 hν π2 ~c
where the Riemann ζ function takes the numerical value ζ(3) ≈
1.202;
1.4. ABSORPTION AND EMISSION 7

• for the cosmic microwave background, T = 2.7 K, and thus


erg
n ≈ 400 cm−3 , U ≈ 4.0 × 10−13 ; (1.40)
cm3

• the Rayleigh-Jeans law is often used to define a radiation tempera-


ture T rad by requiring

2ν2 !
2
kT rad = Iν ; (1.41)
c
obviously, this agrees well with the thermodynamic temperature if
hν/kT  2.82 and Iν = Bν , but the deviation becomes considera-
ble for higher frequencies;

1.4 Absorption and Emission


• the absorption coefficient αν is defined in terms of the energy
absorbed per unit volume, time and frequency from the solid angle
d2 Ω, !
dE
αν Iν = 3 ; (1.42)
d xdtdνd2 Ω abs

• since the stimulated emission is also proportional to Iν , an analo-


gous definition applies for the “induced” emission,
!
dE
αν Iν = 3
ind
; (1.43)
d xdtdνd2 Ω ind

• for the spontaneous emission, we define the emissivity


!
dE
jν = 3 , (1.44)
d xdtdνd2 Ω spn

i.e. the spontaneous energy emission per unit volume, time and
frequency into the solid-angle element d2 Ω;

• the effective net absorption is

αnet
ν = αν − αν ;
ind
(1.45)

• since the unit of Iν is


energy
, (1.46)
time × area × frequency × solid angle

αν must obviously have the dimension (length)−1 ; the “mean free


path” for a photon of frequency ν is thus approximately α−1
ν ;
8KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

• let σν be the cross section of an atom, molecule or other particle


for the absorption of light of frequency ν, then

αν = nσν ≡ ρκ , (1.47)

where n is the number density of absorbing systems and ρ is their


mass density; κ is called “opacity”, whose physical meaning is the
absorption cross section per unit mass,

cm2
[κ] = ; (1.48)
g

• if matter is in equilibrium with a radiation field, the emitted and


absorbed amounts of energy must equal, hence

jν + αind
ν Iν = αν Iν ⇒ jν = αnet
ν Iν ; (1.49)

using (1.28) and (1.29), we then find


!−1
jν hν3 N1
Iν = net = 2 2 −1 ; (1.50)
αν c N2

i.e. if the occupation numbers are known, the emission and absorp-
tion properties in equilibrium can be calculated, and vice versa;

• in particular, in thermal equilibrium with matter, we have


Iν = Bν ⇒ αnet
ν = ; (1.51)

1.5 Radiation Transport in a Simple Case


• we consider an emitting and absorbing medium which does not
scatter for now and is being irradiated by a light bundle; let the
medium be characterised by an emissivity jν and a net absorption
coefficient αnet
ν ;

• per unit of the traversed distance, the intensity of the light bundle
changes according to

dIν = jν dl − α },
Iν dl (1.52)
|{z} |ν{z
emission absorption

from which we obtain the equation of radiation transport in its


simplest case,
dIν
= jν − αν Iν . (1.53)
dl
1.5. RADIATION TRANSPORT IN A SIMPLE CASE 9

• the homogeneous equation (1.53) is easily solved:


dIν
= −αν Iν ⇒ d ln Iν = −αν dl , (1.54)
dl
thus Z !
Iν = C1 exp − αν dl ; (1.55)

• for solving the inhomogeneous equation (1.53), we assume C1 =


C1 (l) and find
" Z !#
dIν d
= C1 (l) exp − αν dl (1.56)
dl dl
Z !
= C1 (l) − C1 (l)αν exp − αν dl
 0 

Z !
!
= jν − αν Iν = jν − ανC1 (l) exp − αν dl ;

this implies Z !
C10 (l) exp − αν dl = jν , (1.57)

which has the solution


Z " Z !#
C1 (l) = dl jν exp αν dl + C2 ; (1.58)

• if αν is a constant along the light path, the integral is simply


Z
αν dl = αν l , (1.59)

and then we have


jν αν l jν
C1 (l) = e , Iν (l) = − C2 e−αν l ; (1.60)
αν αν

• for example, if the intensity satisfies the boundary condition Iν = 0


at l = 0, the intensity as a function of path length becomes
jν  
Iν (l) = 1 − e−αν l ; (1.61)
αν

• interesting limiting cases: let L be the entire path length through


the medium; if

αν L  1 : Iν (L) = (αν L) = jν L ,
αν

αν L  1 : Iν (L) = (1.62)
αν
the former is the “optically thin”, the latter the “optically thick”
case; this amounts to comparing the mean free path α−1ν to the total
path length L;
10KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

• if the radiation is in thermal equilibrium with the irradiated mate-


rial, we must have
jν  
Iν = Bν = 1 − e−αν L , (1.63)
αν
which implies that the source emits at most the intensity of the
Planck spectrum
• we consider optically thin, thermal emission of radio waves; op-
tically thin implies αν L  1 and Iν = jν L, thermal equilibrium
requires Iν = Bν , and in the radio regime we have
hν 2ν2
1, Bν ≈ kT ; (1.64)
kT c2
combining these conditions, we find
2ν2 2ν2
Iν ≈ jν L = αν Bν L = αν kT L = kT b , (1.65)
c2 c2
where T b is the observed temperature, which is obviously related
to the emission temperature T by

T b ≈ αν LT ; (1.66)

this absurd conclusion shows indicates that the two assumptions,


thermal equilibrium and optically-thin radiation, are in conflict
with each other;

1.6 Emission and Absorption in the Continu-


um Case
• in the discrete case, the energy balance for the emitted energy was

N2 A21
|{z} × hν12
|{z} = δE (1.67)
transition number energy per transition

• the emissivity (per unit solid angle) is


N2 A21 hν12 N2 A21 hν
jν = → δD (ν − ν12 ) , (1.68)
4π 4π
with the Dirac delta function modeling a sharp line transition;
• correspondingly, we generalise this expression by a “line profile
function” φ(ν),
N2 A21 hν
jν = φ(ν) , (1.69)

where φ(ν) quantifies the transition probability as a function of
frequency;
1.6. EMISSION AND ABSORPTION IN THE CONTINUUM CASE11

• by an analogous procedure for the absorption coefficient, we find


N1 B12
αν = hν φ(ν) ; (1.70)

• we now consider an electron of energy E which emits the energy
d
≡ P(ν, E) (1.71)
dνdt
per unit time and unit frequency; let further f (~p) be the momentum
distribution of the electrons, then the number of electrons with
energies between E and E + dE is
d3 p dp
n(E)dE = f (~p) dE = 4πp2 f (~p) dE , (1.72)
dE dE
if we assume the distribution to be isotropic in momentum space;
since each electron emits the energy
d = P(ν, E) dνdt , (1.73)
we obtain for the emissivity
Z ∞ Z ∞
dp
4π jν = n(E)P(ν, E)dE = 4π p2 f (p) P(ν, E)dE
0 0 dE
(1.74)
• by definition, we have for a continuous transition
Z E2
P(ν, E2 ) = hν A21 φ(ν)dE1 , (1.75)
0

i.e. electrons with the energy E2 can emit in transitions to all


possible states with E1 < E2 ; thus
2hν3 E2
Z
P(ν, E2 ) = hν 2 B21 φ(ν)dE1 ; (1.76)
c 0

• likewise, the net absorption coefficient is


 
Z Z  

αν = dE1 dE2  n(E1 )B12 − n(E2 )B21  φ(ν) ;
 
4π  | {z } | {z } 
absorption stim. emission
(1.77)
• the second term in this expression can be written
Z Z

dE1 dE2 n(E2 )B21 φ(ν)

Z Z E2

= dE2 n(E2 ) dE1 B21 φ(ν)
4π 0
c2
Z
= dE2 n(E2 )P(ν, E2 ) , (1.78)
8πhν3
12KAPITEL 1. MACROSCOPIC RADIATION QUANTITIES, EMISSION AND A

while the first term reads


Z Z

dE1 dE2 n(E1 )B12 φ(ν)
4π Z Z

= dE2 n(E2 − hν) dE1 B12 φ(ν)

c2
Z
= dE2 n(E2 − hν)P(ν, E2 ) ; (1.79)
8πhν3

• we thus obtain for the absorption coefficient

c2
Z
αν = dE [n(E − hν) − n(E)] P(ν, E) ; (1.80)
8πhν3

• in thermal equilibrium and far from the Fermi edge, the electron
number density is
 E
n(E) ∝ exp − , (1.81)
kT
thus
" ! #

n(E − hν) − n(E) = n(E) exp −1 , (1.82)
kT

from which we obtain


c2  hν/kT Z
αν = e −1 dE n(E) P(ν, E)
8πhν3
c2  hν/kT  jν
= 3
e − 1 jν = , (1.83)
2hν Bν
just as in the discrete case;
Kapitel 2

Scattering
further reading: Rybicki, Light-
man, “Radiative Processes in
2.1 Maxwell’s Equations and Units Astrophysics”, chapter 7; Pad-
manabhan, “Theoretical Astro-
• we use cgs units, i.e. the dielectric constant and the magnetic physics, Vol. I: Astrophysical
permeability of the vacuum are both unity, 0 = 1 = µ0 ; Maxwell’s Processes”, sections 6.4–6.7
equations in vacuum then read
~ · E~ = 4πρ , ∇
∇ ~ ·B~=0,
~ ~
~ × E~ = − 1 ∂ B , ∇
∇ ~ = 4π ~j + 1 ∂E ,
~ ×B (2.1)
c ∂t c c ∂t
where ρ is the charge density and ~j is the current density;

• the Lorentz force per unit charge is


~v ~
f~L = E~ + × B ; (2.2)
c

• the energy density of the electromagnetic field is


1  ~ 2 ~ 2
U= E +B ; (2.3)

• consequently, the field components E~ and B


~ have dimension
!1/2 
 erg 1/2 g cm2 g 1/2
= 2 3 = (2.4)
cm3 s cm cm s2

• forces have the dimension


g cm
≡ dyn ; (2.5)
s2
thus, the Lorentz force
g cm  g 1/2
[F~L ] = [q][E]
~ ⇒ = [q] , (2.6)
s2 cm s2

13
14 KAPITEL 2. SCATTERING

implies that the unit of charge must be


g1/2 cm3/2
[q] = ; (2.7)
s
• in these units, the elementary charge is
cm3/2 g1/2
e = 4.8033 × 10−10 ; (2.8)
s
• the Poynting vector, i.e. the vector of the energy current density of
the electromagnetic field, is
c  ~ ~
S~ = E×B , (2.9)

with dimension
cm erg
[S~ ] = (2.10)
s cm2 s
which is obvious because the unit of E~ 2 is
erg
[E~ 2 ] = ; (2.11)
cm3

2.2 Radiation of a Moving Charge


• far from its source, the electric field of an accelerated charge is, in
see, for example, Jackson, Classi- the non-relativistic limit |~β|  1
cal Electrodynamics, eq. (14.18) q  
˙

E~ = ~e × ~e × ~β , (2.12)
cR
where ~e is the unit vector pointing from the radiating charge to the
observer, and R is the distance;
~ = ~e × E~ and E~ = B
• since B ~ × ~e in vacuum, the B ~ field is
~ = − q ~e × ~β˙ ,
 
B (2.13)
cR
and the Poynting vector is
c h ~ ~ = c B ~ = c B
S~ = ~ 2~e − ( B
~ · ~e) B ~ 2~e
 i h i
B × ~e × B (2.14)
4π 4π 4π
~ = 0;
because ~e · B
• per unit time, the energy
dE ~ ~
= S · dA (2.15)
dt
~ since dA
is radiated through the area element dA; ~ is related to the
~
solid-angle element dΩ as dA ~ = R ~edΩ, we find
2

dE c ~2 2
= B R dΩ ; (2.16)
dt 4π
thus, the energy radiated per unit time into the solid-angle element
dΩ is
dE c ~ 2
= R B ; (2.17)
dtdΩ 4π
2.3. SCATTERING OFF FREE ELECTRONS 15

2.3 Scattering off Free Electrons

2.3.1 Polarised Thomson Cross Section


• a point charge q is accelerated by an incoming electromagnetic
wave with the electric field component E~ 0 ; the equation of motion
for the charge is

m~x¨ = F~L = q E~ 0 + ~β × B
~ 0 ≈ qE~ 0 + O(β) ,
 
(2.18)

i.e. the last approximation employs the non-relativistic limit of the


Lorentz force; thus, the acceleration is
˙ q
~x¨ = c~β = E~ 0 ; (2.19)
m

• using the dipole moment d~ ≡ q~x, we can write eq. (2.13) for the
magnetic field in the form


~ = − ~e × d ;
B (2.20)
c2 R
• according to (2.19), the second time derivative of the dipole mo-
ment is
2
~¨ q ~0
d= E , (2.21)
m
which, when combined with (2.20) and (2.17), implies

dE c 1 !2
¨~
= ~e × d
dtdΩ 4π c2
q4 ~ 0 = q
2 4
= ~e × E E~ 02 sin2 α , (2.22)
4πc3 m2 4πc3 m2

where α was introduced as the angle between the incoming electric


field E~ 0 and the direction of the outgoing radiation, ~e;
• the incoming energy current density is
c ~ 02
S0 = E ; (2.23)

thus the differential scattering cross section is
!2
dσ 1 dE q2
= = sin2 α ; (2.24)
dΩ S 0 dtdΩ mc2

• suppose the elementary charge −e is homogeneously distributed


on the surface of a sphere with radius re ; then, its absolute potential
energy is
e2
∼ ; (2.25)
re
16 KAPITEL 2. SCATTERING

equating this to an electron’s rest-mass energy me c2 , we can solve


for re ,

e2 ! e2
= me c2 ⇒ re = ≈ 2.8 × 10−13 cm ; (2.26)
re me c2

this is the so-called “classical electron radius”;

• generally, the radius


q2
r0 ≡ (2.27)
mc2
is associated with a particle of charge q and rest-mass m; using
this radius, the differential scattering cross section reads


= r02 sin2 α ; (2.28)
dΩ

• the total cross section is


Z Z π
8π 2
σ = r0 sin αdΩ = 2πr0
2 2 2
sin3 αdα = r ; (2.29)
0 3 0

for electrons, we obtain the Thomson cross section,


!2
8π 2 8π e2
σT = r = ≈ 6.6 × 10−25 cm2 ; (2.30)
3 e 3 me c2

2.3.2 Unpolarised Thomson Cross Section

• this scattering cross section is valid for one particular polarisa-


tion direction; we now average over all incoming polarisation
directions; for doing so, we introduce the angle ϕ in the plane per-
pendicular to the incoming direction ~n0 ; the polarisation direction
is then
 cos ϕ 
 
~e0 =  sin ϕ  , (2.31)
 
0
 

if ~e0 is parallel to the x3 axis; the outgoing direction of the scattered


radiation is
 sin θ 
 
~e =  0  ; (2.32)
 
cos θ
 

• using this, one obtains the differential scattering cross section

dσ h i
= r02 sin2 α = r02 (1 − cos2 α) = r02 1 − (~e · ~e0 )2 ; (2.33)
dΩ
2.4. SCATTERING OFF BOUND CHARGES 17

using ~e · ~e0 = sin θ cos ϕ, averaging over ϕ yields

r02
* + Z 2π 
dσ 
= dϕ 1 − sin2 θ cos2 ϕ
dΩ 2π 0
sin2 θ 2π
" Z #
= r0 1 −
2
dϕ cos ϕ
2
2π 0
r2
= 0 (1 + cos2 θ) ; (2.34)
2
this is the unpolarised Thomson cross section;

2.4 Scattering off Bound Charges


• an accelerated charge radiates energy and thus damps the incoming,
accelerating wave; the non-relativistic Larmor formula asserts that cf. Jackson, Classical Electrody-
a non-relativistic, accelerated charge q emits the power namics, eq. (14.22)

2q2 2
P= ~v˙ ; (2.35)
3c3

• this is interpreted as damping with a force F~D ,

2q2 2
− F~D · ~v = P ⇒ ~v · F~D = − 3 ~v˙ ; (2.36)
3c

• the temporal average over a time interval T is

1 T 2q2 ˙ 2
* + Z
dE
= dt 3 ~v
dt T 0 3c
Z T
1 2q2 ˙ T
" #
= ~v~v 0 − dt~v~v¨ ; (2.37)
T 3c3 0

• the first term vanishes for bound charges and large T , thus
E 2q2 D... E
~
D
FD · ~v = 3 ~x · ~v ; (2.38)
3c
we thus identify the expression

2q2 ...
F~D = 3 ~x (2.39)
3c
with the time-averaged damping force;

• for bound orbits with an angular frequency of ω0 , we have


...
~x¨ = −ω20 ~x ⇒ ~x = −ω20 ~x˙ ; (2.40)
18 KAPITEL 2. SCATTERING

thus, the equation of motion reads


q
~x¨ + ω20 ~x = E~ 0 e−iωt − γ~x˙ (2.41)
m
with the damping term

2q2 2
γ= ω ; (2.42)
3c3 0
the first term on the right-hand side of (2.41) is the external exci-
tation, the second is the damping; this equation models a driven,
damped harmonic oscillator, whose solution is known to read

q E~ 0 e−iωt
~x = ; (2.43)
m ω20 − ω2 − iωγ

• we put this back into Larmor’s equation (2.35) and obtain

dE 2q2 ¨ 2 2q2 ¨ ¨∗
=P = ~x = 3 ~x · ~x
dt 3c3 3c
4
2q ~ 2 ω4
= E ; (2.44)
3m2 c3 (ω2 − ω20 )2 + ω2 γ2
0

• the incoming energy current is |S~ | = cE~ 02 /(4π), and thus the scatte-
ring cross section becomes

1 dE 8π 2 ω4
σ= = r0 2 (2.45)
|S~ | dt 3 (ω − ω20 )2 + ω2 γ2

with the typical resonance behaviour at ω = ω0 ;

• interesting limiting cases are:

8π 2
ω  ω0 : σ ≈ r = σT ;
3 0
(binding forces are then irrelevant)
!4
ω
ω  ω0 : σ ≈ σT ;
ω0
(Rayleigh scattering)
ω20
ω ≈ ω0 : σ ≈ σT
4(ω − ω0 )2 + γ2
2 2
γ/(2π)
" #
2π q
= , (2.46)
mc (ω − ω0 )2 + (γ/2)2

where the term in square brackets defines the so-called Lorentz


profile;
2.5. RADIATION DRAG 19

2.5 Radiation Drag

2.5.1 Time-Averaged Damping Force


• in the case of Thomson scattering, the scattering charge damps the
motion which is caused by the incoming electric field according
to the damping force (2.39)
2q2 ...
FD = ~x ; (2.47)
3c3
• an incoming electromagnetic wave exerts the Lorentz force

FL = q(E~ − ~β × B)
~ = m~x¨ (2.48)

on the scattering charge; the last equality in (2.48) assumes that


F~D  F~L , i.e. the back reaction of the radiation by the charge was
neglected;
• from (2.48), we find
... q  ~˙ ~˙ ~ ~ ~˙ 
~x = E+β×B+β×B
m
q  ~˙ ~ ~˙ ~x¨ ~ 
 
= E + β × B + × B
m c
q ~˙ ~ ~˙
 q  ~ ~ ~ ~
= E+β×B+ E+β×B ×B ; (2.49)
m mc
in the non-relativistic limit, we can drop the terms proportional to
~β and find
... q  ˙ q ~ ~ 
~x = ~
E+ (E × B) , (2.50)
m mc
and thus
2q3  ~˙ q ~ ~ 
F~D = E + (E × B) ; (2.51)
3mc3 mc
• averaging over time yields
!2
2q3 ~˙ + 2 q
2
hF~D i = h Ei hE~ × Bi
~ ; (2.52)
3mc3 |{z} 3 mc2
=0

using

hE~ × Bi
~ = hE~ + (~e × E)i
~ = E~ 2~e − (E~ · ~e) E~
|{z}
=0
= 4πU~e , (2.53)

we finally find
8π 2
hF~D i =r U~e = σT U~e (2.54)
3 q
for the time-averaged damping force;
20 KAPITEL 2. SCATTERING

2.5.2 Energy Transfer to a Radiation Field


• we now consider a charge moving with a relative velocity ~v through
a radiation field which is isotropic in its rest frame; in the rest frame
of the radiation, we have
~ = 0 = h Bi
hEi ~ , hE~ 2 i = 4πU = h B
~ 2i ; (2.55)

• in the rest frame of the charge, the Lorentz force is

F~L0 = q(E~ 0 + ~β0 × B


~ 0 ) = qE~ 0 , (2.56)

because ~β0 = 0 in the charge’s rest frame; in addition, we have

E~ 0 = E~ ⊥0 + E~ k0 = γ E~ ⊥ + ~β × B
~ + E~ k ,
 
(2.57)

where γ is the usual Lorentz factor; for Larmor’s equation, we


further need
dE 2q2 ¨0 2 F~ 0
= 3 ~x , ~x¨0 = L ; (2.58)
dt 3c m
• in a first step, we compute

q2 h i2 
γ(E~ ⊥ + ~β × B)
~ + E~ k
D E
~x¨0 = (2.59)
m2
q2 h i2 
= γ(E~ − E~ k + ~β × B)
~ + E~ k
m2
q2 h i2 
= γ(E~ + ~β × B)
~ + (1 − γ)E~ k
m2
q2
γ2 hE~ 2 i + γ2~β2 h B
~ 2 ihsin2 θi + (1 − γ2 )hE~ k2 i ,
h i
=
m2
where we have used that

hE~ k · (~β × B)i


~ =0 (2.60)

because the direction of ~β is random with respect to the direction


of E~ × B;
~ thus, we obtain

q2 2β2 1 − γ2
 2  !
~x¨0 = 4πγ U 2 1 +
2
+
m 3 3γ2
q2 β2
!
= 4πγ U 2 1 +
2
; (2.61)
m 3

• with that, we find the result


2q4 β2 β2
! !
dE
= 4πγ U 2 3 1 +
2
= σT Ucγ 1 +
2
(2.62)
dt 3m c 3 3
for the radiation which is on average radiated by the charge;
2.6. COMPTON SCATTERING 21

• according to the radiation damping, the energy which is on average


absorbed by the charge is
!
dE
= σT Uc , (2.63)
dt abs
and thus the total energy change of the radiation field per unit time
is
β2
" ! #
dE 4
= σT Uc γ 1 +
2
− 1 = σT Ucγ2 β2 ; (2.64)
dt 3 3
this amount of energy is added to the radiation field per unit time
by a single charge;
• the number of collisions between the charge and photons per unit
time is
dNc U
= σT c ; (2.65)
dt hν
combining this with (2.64), we find the energy gain of the radiation
field by scattering of the charge per scattering process,
D E dEc dNc !−1 4 4
∆Eγ = = hνγ2 β2 = γ2 β2 Eγ ; (2.66)
dt dt 3 3

2.6 Compton Scattering

2.6.1 Energy-Momentum Conservation


• we now consider electromagnetic radiation as being composed
of photons; if an ensemble of charges is embedded into a radia-
tion field, energy is transfered by scattering from the photons to
the charges and back; if the radiation temperature is higher than
the temperature of the charge ensemble, energy flows from the
radiation to the charges; this process is called Compton scattering;
in astrophysics, the inverse Compton scattering is typically more
important, during which energy is transfered from the charges to
the radiation field;
• an incoming photon with momentum hν~e/c hits an electron with
momentum ~p; after scattering, the photon and the electron have
momenta hν0~e0 /c and ~p0 ;
• conservation of momentum and energy imply
hν~e + c~p = hν0~e0 + c~p0 , hν + E = hν0 + E 0 , (2.67)
where
E 2 = c2 p2 + m2 c4 (2.68)
according to the relativistic energy-momentum relation;
22 KAPITEL 2. SCATTERING

• solving the energy equation for E 02 and inserting (2.68) yields

c2 ~p02 = c2 ~p2 + h2 (ν − ν0 )2 + 2Eh(ν − ν0 ) , (2.69)

while the momentum equation implies

c2 ~p02 = c2 ~p2 + h2 (ν~e − ν0~e0 )2 + 2h(ν~e − ν0~e0 )c~p ; (2.70)

• subtracting (2.69) from (2.70) and cancelling suitable terms gives

hνν0 (1 − cos θ) = E(ν − ν0 ) − c~p(ν~e − ν0~e0 ) , (2.71)

where θ is the angle between ~e and ~e0 ;

• if the electron is originally at rest, ~p = 0 and E = mc2 , and (2.71)


simplifies to
h
ν − ν0 = νν0 (1 − cos θ) , (2.72)
mc2
and in the limit of very low photon energy, hν  mc2 , we find for
the relative energy change of the Compton-scattered photon

∆Eγ ν0 − ν Eγ
= = − 2 (1 − cos θ) , (2.73)
Eγ ν mc

and if hν & mc2 , quantum electrodynamics must be used anyway;

• averaging (2.73) over all scattering angles θ, we find the mean


energy loss per Compton scattering,

Eγ2 1 Eγ2
Z
hEγ i = − (1 − cos θ)d(cos θ) = − ; (2.74)
mc2 −1 mc2

2.6.2 Energy Balance


• the total energy transfer to the radiation field due to the motion of
a single charge is given by the difference between the energy gain
(2.66) per scattering and the energy loss per Compton scattering
(2.74), !
D E 4 2 2 Eγ
∆Eγ = γ β − 2 Eγ ; (2.75)
3 mc

• for photons with Eγ  mc2 in the relativistic limit, β ≈ 1, and


D E 4
∆Eγ ≈ γ2 Eγ , (2.76)
3
which can become a very large number; in that way, for example,
CMB photons can be converted to X-ray photons;
2.6. COMPTON SCATTERING 23

• in the thermal limit of (2.75), we can approximate v  c, thus


γ ≈ 1, and mv2 = 3kT e ; then

4v2
!
D E Eγ   Eγ
∆Eγ ≈ − E γ = 4kT e − E γ ; (2.77)
3c2 mc2 mc2

thus, the photons gain energy (on average), if

4kT e > Eγ (2.78)

(inverse Compton scattering), and lose energy otherwise (Compton


scattering)

• Compton scattering causes fast charges to lose energy; typical time


scales are, according to (2.64)

E γmc2 3 mc2
tc ≡ = 4 = ; (2.79)
dE/dt σ Ucγ2 β2 4 γβ2 σT U
3 T

for non-relativistic, thermal electrons, E = 3kT e /2 and γ ≈ 1, and


3
kT e c2 9 mc
tc = 2
= ; (2.80)
4
σ Ucv2
3 T
8 σT U

• after Ns scatterings, the total energy transfer from thermal electrons


to the photons is
!Ns
E0
!
4kT e 4kT e Ns
= 1+ ≈ exp ≡ e4y , (2.81)
E mc2 mc2

where the Compton parameter

4kT e Ns
y≡ (2.82)
mc2

was introduced;

• if the electron number density is ne , the number of scatterings per


path length dl is
Z
dNs = ne σT dl ⇒ Ns = σT ne dl , (2.83)

and thus the Compton-y parameter becomes


Z
kT e
y= σT ne dl ; (2.84)
mc2
24 KAPITEL 2. SCATTERING

2.7 The Kompaneets Equation


• we need an additional equation which specifies how the photon
spectrum is changed due to the scatterings; for deriving it, we
assume that a homogeneous, thermal distribution of electrons is
located in a homogeneous sea of radiation, such as, for example, a
galaxy cluster in the microwave background; the collisions with
the electrons change the photon energy, but not their number, and
thus their spectrum cannot remain a Planck spectrum;

• let n(ν) be the occupation number of photon states with frequency


ν; then, the Boltzmann equation requires

∂n(ν)
Z Z !

= 3
d p dΩ c (2.85)
∂t dΩ
× n(ν0 ) [1 + n(ν)] N(E 0 ) − n(ν) 1 + n(ν0 ) N(E) ;
  

this equation has the following meaning: the occupation number


at the frequency ν changes due to scattering from ν to ν0 , and from
ν0 to ν; the term
n(ν) 1 + n(ν0 ) N(E)
 
(2.86)
quantifies how many photons there are at frequency ν, corrected
by the factor for stimulated emission from ν to ν0 , and multiplies
with the number of collision partners N(E) at energy E; in other
words, it quantifies the number of collisions away from frequency
ν; analogously, the term

n(ν0 ) [1 + n(ν)] N(E 0 ) (2.87)

quantifies the opposite scattering, i.e. scattering processes increa-


sing the occupation number at frequency ν; of course, the energy
difference between photon frequencies ν and ν0 must be balanced
by the difference between the energies E and E 0 ; the integral over
d3 p integrates over the electron distribution, and the factor

dΩ (2.88)
dΩ
specifies the probability for scattering photons from frequency ν
to frequency ν0 or backward;

• we assume thermal photon and electron distributions, and restrict


ourselves to the limit of Thomson scattering, which applies if

hν  mc2 ; (2.89)

moreover, we assume small changes in the photon frequency, hence

δν ≡ ν0 − ν  ν ; (2.90)
2.7. THE KOMPANEETS EQUATION 25

moreover, the electron energy distribution is


!
E
N(E) ∝ exp − , (2.91)
kT e
and energy conservation requires
E 0 = hν − hδν ; (2.92)

• now, both n(ν) and N(E) can be expanded in Taylor series up to


second order,
∂n 1 ∂2 n 2
n(ν0 ) = n(ν) + δν + δν + O(δν3 ) , (2.93)
∂ν 2 ∂ν2
∂N 1 ∂2 N 2 2
N(E 0 ) = N(E) − hδν + h δν + O(δν3 ) ,
∂E 2 ∂E 2

where (2.91) allows us to use


∂N N(E) ∂2 N N(E)
=− , = ; (2.94)
∂E kT e ∂E 2 (kT e )2
• for simplification, we now define the dimension-less photon energy,
scaled by the thermal electron energy

x≡ (2.95)
kT e
and find
∂n 1 ∂2 n 2
n(x0 ) ≈ n(x) + δx + δx ,
" ∂x 2 ∂x#2
δx2
N(E 0 ) ≈ N(E) 1 + δx + ; (2.96)
2
• with these approximations, we return to the original equation (2.86)
for n(ν) and obtain
∂n ∂n
" #
= + n(n + 1) I1
∂t ∂x
1 ∂2 n ∂n
" #
+ + 2(1 + n) + n(n + 1) I2 , (2.97)
2 ∂x2 ∂x
with the abbreviations
Z Z
3 dσ
Ii ≡ d p dΩ cδxi N(E) (2.98)
dΩ
• the energy change of a photon scattering off a moving electron
follows from (2.71), adopting the non-relativistic limit
p2
E = mc2 + (2.99)
2m
and using (2.89) and (2.90); this yields
hν x
hδν = − (~e − ~e0 ) · ~p ⇒ δx = − (~e − ~e0 ) · ~p ; (2.100)
mc mc
26 KAPITEL 2. SCATTERING

• using this result, the integrals Ii can be carried out straightforward-


ly; with the unpolarised Thomson cross section (2.34), we first
find
kT e x2
I2 = 2σT ne c ; (2.101)
mc2
• for evaluating I1 , we note that I1 is the mean rate of relative energy
transfer, quantified by δx from the electrons to the photons, and
therefore the mean energy transfer rate, divided by kT e ; from
(2.77), we know that this is
D E x(kT e )2
∆Eγ = (4 − x) (2.102)
me c2
per scattering, and multiplying with the collision rate ne σT c gives
kT e
I1 = ne σT c x(4 − x) ; (2.103)
me c2

• with these two expression for Ii , we find the time derivative of n to


be
me c2 1 ∂n 1 ∂ 4 ∂n
" !#
= 2 x +n+n 2
; (2.104)
kT e ne σT c ∂t x ∂x ∂x

• we finally transform the time t to the Compton parameter, using


kT e
dy = ne σT c dt (2.105)
me c2
to find the Kompaneets equation

∂n 1 ∂ 4 ∂n
" !#
= x +n+n 2
; (2.106)
∂y x2 ∂x ∂x

• the hot gas in galaxy clusters is much hotter than the cosmic
background radiation; then, we can approximate the right-hand
side of (2.106) to lowest order in x,

∂n ∂2 n ∂n
≈ x2 2 + 4x ; (2.107)
∂y ∂x ∂x

• inserting here the occupation number in thermal equilibrium, n ≈


(e x − 1)−1 , we find

δn x2 e x (1 + e x ) 4xe x
!
= δy − x (2.108)
n (e x − 1)2 e −1
for the relative change of the occupation number, where x is now
hν/kT and no longer hν/kT e !
Kapitel 3

Radiation Transport and


Bremsstrahlung
further reading: Shu, “The Phy-
sics of Astrophysics, Vol I: Ra-
3.1 Radiation Transport Equations diation”, chapters 2, 3, and 15;
Rybicki, Lightman, “Radiative
• we start with the collision-less Boltzmann equation for describing Processes in Astrophysics”, chap-
the temporal change of the photon distribution function in phase ter 5; Padmanabhan, “Theoretical
space, Astrophysics, Vol. I: Astrophysi-
∂n ~ ∂n ˙ ∂n cal Processes”, sections 6.8–6.9
+∇· + ~p · =0, (3.1)
∂t ∂~x ∂~p
which is valid in absence of collisions;

• for photons, we have ~v = c~e, where ~e is the unit vector in the


direction of light propagation; moreover, ~p˙ = 0 in absence of
systematic external forces (such as gravitational lensing); since
the intensity Iν is proportional to n, the Boltzmann equation for
photons can also be written as
1 ∂Iν ∂Iν
+ ~e · =0; (3.2)
c ∂t ∂~x

• we now define the following quantities:


Z Z
~ 1
Fν ≡ dΩ ~e · Iν , Pν,i j ≡ dΩ ei e j Iν (3.3)
c
and recall the spectral energy density
Z
1
Uν = dΩ Iν ; (3.4)
c

• integrating the Boltzmann equation (3.2) first over dΩ, we obtain


the equation
∂Uν ~ ~
+ ∇ · Fν = 0 ; (3.5)
∂t

27
28KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

which has the form of a continuity equation and identifies F~ν as


the spectral radiation current density (spectral because it retains
the dependence on frequency ν); this equation expresses energy
conservation in the radiation field;
• if we multiply (3.2) with ei first before integrating over dΩ, we
find
∂Iν ∂Iν
Z Z
1
dΩ ei + dΩ ei e j =0, (3.6)
c ∂t ∂x j
and hence
1 ∂Fν,i ∂Pν,i j
+c =0; (3.7)
c ∂t ∂x j
• this equation describes the change of the momentum current den-
sity, because U 
ν
(c~e) (3.8)
c
is the momentum density of the radiation field, and thus

1 ∂F~ 1 ∂
Z
= 2 dΩ Iν~e (3.9)
c ∂t c ∂t
is c times the temporal change of the momentum current density;
Eq. (3.7) expresses momentum conservation;
• in presence of emission, stimulated emission and absorption, we
know from the first chapter that the energy equation must be
augmented by source and sink terms on its right-hand side; we had
dIν 1 dIν
= jν − αν Iν = ; (3.10)
dl c dt
integrating over dΩ, and assuming that jν and αν are isotropic, we
find
dUν
= 4π jν − αν Uν c = 4π jν − ρκν cUν ; (3.11)
dt
we now re-define the emissivity,
4π jν → ρ j0ν ≡ ρ jν , (3.12)
i.e. we refer it to the mass density, and write
dUν
= ρ( jν − κν cUν ) ; (3.13)
dt
• likewise, the momentum-conservation equation
1 dIν
= jν − αν Iν (3.14)
c dt
becomes after multiplication with ~e and integration over dΩ
Z Z Z
1d
dΩ ~e · Iν = dΩ jν~e − dΩ αν Iν~e , (3.15)
c dt
3.2. LOCAL THERMODYNAMICAL EQUILIBRIUM 29

and thus
1 dF~
= −αν F~ν = −ρκν F~ν , (3.16)
c dt
where we have assumed again that jν and κν are isotropic
• including the emission and absorption terms, the transport equati-
ons are modified to read
∂Uν ~ ~
+ ∇ · Fν = ρ( jν − κν cUν )
∂t
1 ∂Fν,i ∂Pν,i j
+c = −ρκν Fν,i ; (3.17)
c ∂t ∂x j
these equations do not contain scattering terms yet!
• since the change in the momentum current density corresponds to
a force density, and this force is caused by the interaction between
radiation and matter, an oppositely directed and equally strong
force must act on the matter as radiation pressure force; thus
ρ
Z ∞
f~rad = κν F~ν dν (3.18)
c 0
is the density of the radiation pressure force;

3.2 Local Thermodynamical Equilibrium


• the moment equations for Uν and F~ν are by no means easier to
handle than the Boltzmann equation whose moments they are;
we obviously need an additional approximation, or condition, in
order to “close” the moment equations; the “closure” means that
they can then be solved without progressing indefinitely to higher
orders of moments;
• often, the mean free path of the photons is much smaller than
the dimensions of the system under consideration; then, we can
assume that thermodynamical equilibrium is locally established
between the radiation field and the matter; under this condition,
Iν ≈ Bν (T ) , (3.19)
i.e. the specific intensity of the radiation field is the Planckian
intensity of a black body, and
Z
1 4π
Uν = dΩ Iν = Bν (T ) ; (3.20)
c c
• under such circumstances, there is obviously no radiation flux any
more because the radiation field is isotropic; in order to estimate
the flux nonetheless, we study the orders of magnitude of the
different terms in the moment equations;
30KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

• time derivatives can typically be neglected because temporal chan-


ges of the quantities Uν , F~ν and Pν,i j occur on an evolutionary time
scale, while the other terms change according to the streaming of
the photons, thus approximately on time scales of order (mean free
path)/c;
• if we first ignore ∂F~ν /∂t, we obtain
c ∂Pν,i j
Fν,i ≈ − (3.21)
ρκν ∂x j
in the approximation of Local Thermodynamical Equilibrium
(LTE), we further have

Pν,i j ≈ Pν δi j = δi j (3.22)
3
because of the (local) isotropy of the radiation field, and thus
c ∂Uν c Uν
Fν,i ≈ − ≈− , (3.23)
ρκν ∂xi ρκν R
where R is a typical dimension of the system; the mean free path
λν is determined by

λν nσν = λν ρκν ≈ 1 , (3.24)

and (3.23) can thus be approximated by


λ 
ν
|Fν,i | ≈ cUν , (3.25)
R
which is smaller by a factor λν /R compared to the transparent case
(in which κ → 0 and λν → R;
• using this estimate for F~ν , we return to the Eq. (3.17) for the partial
time derivative of Uν ; as before, we ignore the time derivative, such
that the only term remaining on the left-hand side is

~ · F~ν ≈ cUν λν ;
∇ (3.26)
R R
the second term on the right-hand side is
!2
c R ~ ~ ~ · F~ν ;
ρκν cUν ≈ Uν ≈ ∇ · Fν  ∇ (3.27)
λν λν
thus, because of the assumption of local thermodynamical equili-
brium, the divergence of F~ν is negligibly small; consequently, we
must require
ρκν 4π
ρ jν ≈ ρκν cUν ⇒ Uν ≈ = Bν (T ) , (3.28)
cαν c
as anticipated;
3.2. LOCAL THERMODYNAMICAL EQUILIBRIUM 31

• accordingly, if λν  R and tevol  λν /c, the solutions of the


moment equations are
c ∂Pν,i j 4π
Fν,i ≈ − , Uν ≈ Bν (T ) ; (3.29)
ρκν ∂x j c

• because of (local) isotropy, we had


Uν 4π
Pν,i j ≈ δi j ≈ Bν (T ) δi j , (3.30)
3 3c
and thus
4π ∂Bν ∂T
!
Fν,i ≈ − , (3.31)
3ρκν ∂T ∂xi
i.e. the flux will become proportional to the temperature, which is
characteristic for diffusion processes;
• for convenience, we now introduce the Rosseland mean opacity,
R ∞  ∂B (T ) 
0
dν κν−1 ∂Tν

κR ≡ R ∞  ∂B (T )  ;
−1
(3.32)
ν
0
dν ∂T

here, we can use the fact that


∂Bν (T ) ∂ ∂ caT 4
Z ∞ ! Z ∞ !
dν = dν Bν (T ) = , (3.33)
0 ∂T ∂T 0 ∂T 4π
where
π2 k 4 erg
a≡ 3
= 7.6 × 10−15 4 3 (3.34)
15 (~c) K cm
is the so-called Stefan-Boltzmann constant;
• using this, we obtain the expression
4π ∂T dν ∂Bν
Z ∞ Z ∞
~
F = ~
dνFν = −
0 3ρ ∂~x 0 κν ∂T
4π ~ d caT 4
!
c ~  4
= − ∇T =− ∇ aT (3.35)
3ρκR dT 4π 3ρκR
for the radiative energy flux;
• the energy which is streaming away interacts with the absorbing
matter and thus exerts a force on it, which is determined by the
right-hand side of the momentum-conservation equation, as des-
cribed above:
ρ ∞ ρ ∂Bν ∂T
Z Z ∞
~ 4π
frad = dν κν F~ν = − dν
c 0 c 3ρ 0 ∂T ∂~x
1~ ~ ,
= − ∇(aT 4
) = −∇P (3.36)
3
which equals just the negative pressure gradient;
32KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

• a remark on units: the unit of Uν is


erg
[Uν ] = , (3.37)
cm3 Hz
the unit of κν is
cm2
[κν ] = , (3.38)
g
and thus the unit of F~ν is
erg
[F~ν ] = [c][Uν ] = , (3.39)
cm2 t Hz

and the unit of f~rad is

g s cm2 erg erg


[ f~rad ] = Hz = (3.40)
cm3 cm g cm2 s Hz cm4
g cm2 1 g cm 1 dyn force
= = = = ,
s4 cm4 s2 cm3 cm3 volume
as it should be;

3.3 Scattering
• so far, we have only considered emission and absorption, but
neglected scattering; scattering changes the distribution function
of the photons by exchanging photons with different momenta; if
we assume for simplicity that the scattering process changes the
photon’s momentum, but not its energy, we can write the scattering
cross section in the form
dσ(~e → ~e0 )
= σφ(~e, ~e0 ) , (3.41)
dΩ
where ~e and ~e0 are unit vectors in the propagation directions of
the incoming and the outgoing photon; the function φ(~e, ~e0 ) is
normalised, symmetric in its arguments and dimension-less and
describes the directional distribution of the scattered photons;

• scattering increases the distribution function n(~e) according to


" # Z
dn(~e)
= dΩ0 Ne cσφ(~e, ~e0 ) n(~e0 ) [1 + n(~e)] ,
 
dt + | {z }
# of scatterings ~e → ~e0
(3.42)
where the factor [1 + n(~e)] is included for describing stimulated
emission of photons with momentum direction ~e;
3.3. SCATTERING 33

• analogously, losses due to scattering are given by


" # Z
dn(~e)
= dΩ0 Ne cσφ(~e, ~e0 ) n(~e) [1 + n(~e0 )] ,
 
(3.43)
dt −

and thus the total change of n(~e) due to scatterings becomes


Z
dn(~e)
= dΩ0 Ne cσφ(~e, ~e0 )
 
dt
× n(~e0 ) [1 + n(~e)] − n(~e) [1 + n(~e0 )]

Z
= dΩ0 Ne cσφ(~e, ~e0 ) n(~e0 ) − n(~e) , (3.44)
  

in which the terms from stimulated emission cancel exactly;

• since the integral over the solid angle only concerns the direction
of ~e0 , we obtain from (3.44)
Z
n(~e)
= −Ne cσn(~e) + Ne cσ dΩ0 φ(~e, ~e0 )n(~e0 ) , (3.45)
dt
and thus
Z
1 dn(~e)
= −ρκνsca n(~e) + ρκνsca dΩ0 φ(~e, ~e0 )n(~e0 ) , (3.46)
c dt
where we have introduced the scattering opacity through κνsca =
Ne σ;

• since the intensity at fixed frequency is proportional to the occupa-


tion number, the same equation (3.46) also holds for Iν ; therefore,
the transport equation for the specific intensity is changed in pre-
sence of scattering to
1 dIν 1 ∂Iν ∂Iν
= + ~e · (3.47)
c dt c ∂t ∂~x
ρ jν
" Z #
= − ρκν Iν − ρκν Iν − dΩ φ(~e, ~e )Iν (~e ) ;
abs sca 0 0 0

• again, we now take the moments of the transport equation in order


to see how the moment equations are changed by scattering; the
first moment is obtained by integrating (3.48) over dΩ,
∂Uν ~ ~
+ ∇ · Fν = ρ jν − ρκνabs cUν − ρκνsca cUν + ρκνsca cUν (3.48)
∂t
due to the normalisation of φ(~e, ~e0 ); therefore, the scattering terms
cancel, and the equation for Uν remains unchanged;

• the next moment equation simplifies if we further assume that


Z Z
dΩφ(~e, ~e )~e = 0 =
0
dΩ0 φ(~e, ~e0 )~e0 , (3.49)
34KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

which holds for many scattering processes (e.g. Thomson scatte-


ring); then, the second moment equation reads
1 ∂Fν,i ∂Pν,i j
+c = −ρκνabs Fν,i − ρκνsca Fν,i
c ∂t ∂x j
= −ρ(κνabs + κνsca )Fν,i , (3.50)
i.e. the scattering opacity is simply added to the absorption opacity
here; with a suitable modification of the Rosseland mean opacity,
the diffusion approximation remains valid which we have obtained
above;

3.4 Bremsstrahlung

3.4.1 Spectrum of a Moving Charge


• a radiation process which is very important in astrophysics is due
to electrons which are scattered off ions and radiate due to the
acceleration they experience; in order to describe it, we start again
from Larmor’s equation, which says
dE 2e2 2
= 3 ~x¨ , (3.51)
dt 3c
where e is the charge of the accelerated particle (the electron, in
most cases), ~x¨ is its acceleration, and dE/dt is the power radiated
away;
• as a function of frequency, this equation can be written as follows:
2e2 ¨ 2 2e2 ∞ ¨ 2
Z
dE = 3 ~x ⇒ E= 3 dt ~x ; (3.52)
3c 3c −∞
if we Fourier-transform the particle’s trajectory,
Z ∞ Z ∞
dω ˆ
~xˆ(ω) = dt ~x(t)e , ~x(t) =
iωt
~x(ω)e−iωt , (3.53)
−∞ −∞ 2π

we first have
Z ∞
dω h 2 ˆ i −iωt
~x¨ = −ω ~x(ω) e ⇒ ~xˆ¨ = −ω2 ~xˆ(ω) , (3.54)
−∞ 2π

and we can use Parseval’s equation,


Z ∞ Z ∞
dω 2
dt f (t) =
2
fˆ(ω) ; (3.55)
−∞ −∞ 2π

combining these results yields


2e2 ∞ dω ˆ¨ 2 4e2 ∞ dω 4 ˆ 2
Z Z
E= 3 ~x(ω) = 3 ω ~x(ω) , (3.56)
3c −∞ 2π 3c 0 2π
3.4. BREMSSTRAHLUNG 35

and, by differentiation,
dE 2e2 4 ˆ 2
= ω ~x(ω) ; (3.57)
dω 3πc3
this is a general expression valid for all radiation processes; in
order to make progress, we need the Fourier transform of the
specific particle trajectory;

3.4.2 Hyperbolic Orbits


• classically, the electron follows a hyperbola around the ion in the
orbital plane perpendicular to the (conserved) angular momentum;
the focal point of the hyperbola is the centre of mass, which we
assume to coincide with the centre of the scattering ion, i.e. we
neglect the mass of the electron; in polar coordinates, the trajectory
is given by
p
r(ϕ) = , (3.58)
1 +  cos ϕ
with the parameters
!1/2
Lz2 2Lz2 E
p= = a( − 1) and  ≡ 1 + 2
2
, (3.59)
αm αm
where Lz = bmv∞ is the angular momentum in z direction, v∞ is
the initial velocity at infinity, E = mv2∞ /2 is the kinetic energy at
infinity and thus the total energy, and α quantifies the coupling
strength; for electrons orbiting nuclei at rest with charge Ze,
α = Ze2 (3.60)

• as for solving the Kepler problem, we introduce a parameter ψ (the


eccentric anomaly), of which we require that it satisfy
r = a( cosh ψ − 1) ; (3.61)
we find the relation between ϕ and ψ by inserting (3.61) into (3.58),
using (3.59)
a( 2 − 1)  − cosh ψ
= a( cosh ψ − 1) ⇒ cos ϕ = ,
1 +  cos ϕ  cosh ψ − 1
(3.62)
if we want ψ = 0 where ϕ = 0;
• energy conservation implies that the time t when the electron
reaches the distance r from the scattering ion is
Z r Z r
dx xdx
t=    1/2
=   1/2 ,
α Lz2 Lz2
r0 2
m
E + x − 2mx2 r0 2
m
Ex + αx − 2m
2

(3.63)
36KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

where
Lz2 = mαa( 2 − 1) (3.64)
was used from (3.59); furthermore, we have
p Lz2 α2 m α α
a= = = ⇒ E= ; (3.65)
 2 − 1 αm 2Lz2 E 2E 2a
combining, we first find
r Z r
m xdx
t= , (3.66)
2α r0 x2 + x − a ( 2 − 1)i1/2
h
2a 2

which can be transformed with (3.61) to obtain


ma3 ψ
r Z r
ma3
t= ( cosh ψ − 1)dψ = ( sinh ψ − ψ) ; (3.67)
α 0 α
• the coordinates x and y in the orbital plane satisfy
 − cosh ψ
x = r cos ϕ = a( cosh ψ − 1) = a( − cosh ψ)
 cosh ψ − 1

y = r sin ϕ = a  2 − 1 sinh ψ , (3.68)
where we have used (3.61) and (3.62); with these expressions, we
return to the Fourier transform of x and y
• since
ẋˆ = −iω x̂(ω) , (3.69)
we have
Z ∞
i ˆ i
x̂(ω) = ẋ(ω) = dt ẋ(t)eiωt
ω ω −∞
Z ∞
i dx dψ iωt(ψ)
= dt e
ω −∞ dψ dt
ia ∞
Z
= − dψ sinh ψeiωt(ψ) ; (3.70)
ω −∞
• using (3.67), we can write
 r 
ma 3
eiωt(ψ) = exp iω ( sinh ψ − ψ) ;

 
(3.71)
α
from (3.65), we find
α
a= , (3.72)
mv2∞
moreover,
r s
ma3 mα2 αω
ω =ω = ≡µ, (3.73)
α m3 v6∞ mv3∞
and thus
eiωt(ψ) = eiµ( sinh ψ−ψ) (3.74)
3.4. BREMSSTRAHLUNG 37

• putting this result into (3.70), we write

ia ∞
Z
x̂(ω) = − dψ sinh ψeiµ( sinh ψ−ψ)
ω −∞

ia  2 − 1 ∞
Z
ŷ(ω) = dψ cosh ψeiµ( sinh ψ−ψ) ; (3.75)
ω −∞

these expressions can be analytically integrated and lead to first-


order Hankel functions, which will not be discussed in detail here;

• forming
|~xˆ(ω)|2 = x̂(ω) x̂∗ (ω) + ŷ(ω)ŷ∗ (ω) (3.76)
and inserting the result into (3.57) yields the desired bremsstrah-
lung spectrum;

3.4.3 Integration over the Electron Distribution


• having obtained the spectrum dE/dω for a single charge, we now
have to integrate over a distribution of charges; we do this by inte-
grating over all impact parameters b from 0 to ∞ after multiplying
dE/dω with
ni ne v · 2πbdb , (3.77)
which is the number of scatterings per unit volume and unit time
between ions and electrons with the number densities ni and ne ,
respectively; for a fully ionised pure hydrogen gas, ni = ne ≡ n;

• preparing the integration, we note from (3.59) and (3.72) that

b2 m2 v4∞ b2
2 = 1 + = 1 + , (3.78)
α2 a2
such that the integration over b can be transformed into an integra-
tion over ,
bdb bdb
d =  = 2 ⇒ bdb = a2 d , (3.79)
a2 a
where  ∈ [1, ∞) while b ∈ [0, ∞);

• inserting suitable approximations for the first-order Hankel func-


tions, we find after carrying out the b integration and inserting
(3.60)
    
2 mv3 mv3
dE  ln γ Ze2 ω ω  Ze2 
16Z 2 e6 n2 
,

≈ (3.80)
 √π
3
dVdtdω 3m2 c3 v  ω  mv

3 Ze2

where γ ≈ 1.78 is Euler’s constant;


38KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG

• we write this result as

dE 16πZ 2 n2 e6 1
= √ gff (v, ω) , (3.81)
dVdtdω 3 3m2 c3 v
introducing the so-called gaunt factor gff , which usually depends
at most weakly on v;

• in a dilute thermal plasma, the electrons have a Maxwellian veloci-


ty distribution, but for emitting a photon of energy ~ω, an electron
needs at least an energy of
r
m 2 2~ω
vmin = ~ω ⇒ vmin = ; (3.82)
2 m
the thermal average of the inverse velocity is then
* +
1  m 3/2 Z ∞ 1
4πv2 dv e−mv /2kT
2
=
v 2πkT vmin v
r
2m −~ω/kT
= e ; (3.83)
πkT

• replacing 1/v in (3.81) by the average (3.83) finally yields the


specific emissivity of a thermal plasma due to bremsstrahlung,
r
dE 16πZ 2 n2 e6 2m −~ω/kT
≡ j(ω) = √ e gff (v, ω) ; (3.84)
dVdtdω 3 3m2 c3 πkT

• the volume emssivity is the integral of j(ω) over frequency ω,


r
dE 16Z 2 n2 e6 2mkT √
j= = g ff (v, ω) ∝ n2
T; (3.85)
dVdt 3~m2 c3 π

• numerically, the volume emissivity in cgs units is


 !1/2

 −20 kT
6.69 × 10



erg



 n 2 

 !1/2
j = Z 2 gff

 −24 kT (3.86)
cm−3 
 2.68 × 10



 keV


  T 1/2
−28
 7.86 × 10



K

• as an example, we consider the X-ray emission of a massive galaxy


cluster with kT = 10 keV; typical clusters reach electron densities
of n ≈ 10−3 cm−3 in their cores; let us assume for simplicity that
the X-ray emitting gas with that electron density fills a volume of
1 Mpc3 ;
3.4. BREMSSTRAHLUNG 39

• assuming fully ionised pure hydrogen, we put Z = 1 and gff = 1


for simplicity; then, (3.86) yields

LX ≈ V j ≈ (3.1 × 1024 )3 · 10−6 · 2.68 × 10−24 · 10
≈ 2.5 × 1044 erg s−1 , (3.87)

which makes galaxy clusters the most luminous X-ray sources on


the sky;

• with an average energy of ∼ 10 keV = 1.6e − 8 erg per photon, this


luminosity corresponds to

NX ≈ 1.6 × 1052 s−1 (3.88)

photons emitted by the cluster per second; if the cluster is at a


distance of, say, 100 Mpc ≈ 3.1 × 1026 cm, these photons are
distributed over an area of ≈ 1.2 × 1054 cm2 , such that an X-ray
detector with a typical collecting area of a few hundred cm2 sees

1.6 × 1052 −1
≈ 100 s ≈ 1 s−1 (3.89)
1.2 × 1054
i.e. this enormous X-ray luminosity produces a flux of approxima-
tely one photon per second in a typical X-ray detector;
40KAPITEL 3. RADIATION TRANSPORT AND BREMSSTRAHLUNG
Kapitel 4

Synchrotron Radiation,
Ionisation and Recombination
further reading: Shu, “The Phy-
sics of Astrophysics, Vol I: Ra-
4.1 Synchrotron Radiation diation”, chapters 18, 19, 21–
23; Rybicki, Lightman, “Radia-
tive Processes in Astrophysics”,
4.1.1 Electron Gyrating in a Magnetic Field chapters 6 and 10; Padmanabhan,
“Theoretical Astrophysics, Vol. I:
• a further very important radiation process is the emission of ra-
~ in such a field, Astrophysical Processes”, secti-
diation by electrons moving in a magnetic field B;
ons 6.10–6.12
electrons spiral around field lines, with their angular frequency
given by
ceB eB
ωB = = , (4.1)
E γmc
where E is the electron energy, and γ is the usual Lorentz factor;
numerically, we have
 B 
ωB ≈ 17.6γ−1 MHz , (4.2)
1G
i.e. synchrotron radiation is typically emitted at radio frequencies;
• the radius of the projection of the spiral orbit perpendicular to the
magnetic field is
v γmcv
rB = = , (4.3)
ωB eB
and the complete motion is the circular motion around B, ~ superpo-
~
sed by a drift along B;
• we employ Larmor’s equation
dE 2e2 ¨ 2
= 3 ~x (4.4)
dt 3c
again to calculate the emission; assuming E~ = 0, the Lorentz force
is
γe γe
F~L = (~v × B)~ ⇒ ~x¨ = ~ ;
(~v × B) (4.5)
c mc

41
42KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

~ then |~v × B|
• let now α be the angle between ~v and B, ~ = vB sin α and
 e 2
¨
|~x| =
2
(γvB sin α)2 , (4.6)
mc
and, from Larmor’s equation (4.4),
dE 2e4 γ2 v2 B2 sin2 α
= (4.7)
dt 3m!2 c5
2 e2 B2 2 2 2
= cγ β
2 2 2
B sin2
α = 2cσ T γ β sin α ,
3 mc2 8π
where we have identified the Thomson scattering cross section;
• averaging over all pitch angles α, finding
1 π 3
Z
2
hsin αi =
2
sin αdα = , (4.8)
2 0 3
we obtain the radiation energy emitted per unit time by an isotropic
electron distribution in a magnetic field,
dE 4 B2 4
= cσT γ2 β2 = UB cσT γ2 β2 , (4.9)
dt 3 8π 3
where we have inserted the energy density UB = B2 /(8π) of the
magnetic field;

4.1.2 Beaming and Retardation


• we had seen in Sect. 2 of Chap. 1 that solid angles are deformed
by Lorentz transformations according to
dΩ
dΩ0 = ; (4.10)
γ2 (1 − β cos θ)2
relativistically moving charges thus focus their emission into their
forward direction; for estimating the opening angle ∆θ of the
resulting cone, we require that
1 1
≈ , (4.11)
1 − β cos ∆θ 2(1 − β)
i.e. we want to estimate the half-width of the cone; approximating
the cosine in (4.11) to second order,
∆θ2
1−β+ ≈ 2(1 − β) , (4.12)
2
thus
∆θ2 ≈ 2(1 − β) ≈ (1 + β)(1 − β) = γ−2 (4.13)
where we have approximated 2 ≈ (1 + β) for relativistic particles,
and
1
∆θ ≈ (4.14)
γ
4.1. SYNCHROTRON RADIATION 43

• due to the narrow emission cone with the opening angle ∆θ ≈ γ−1 ,
an observer sees an electron spiralling in a magnetic field only
during the short moment while the cone is moving past him;
• moreover, this means that the radiated electric field can depend on
θ only through the combination ∆θγ, because a change in γ causes
an immediate change in ∆θ;
• finally, the arrival time of the radiation at the observer depends
also on the angle θ
• we now consider two radiation signals which leave the electron at
times t1 and t2 = t1 + t; if the velocity of the electron is v and its
orbital radius is rB , we have
rB θ
rB θ = vt ⇒ t = ; (4.15)
v
the signal leaving at t2 needs less time to get to the observer,
namely by the amount
rB θ
; (4.16)
c
this is of course the usual retardation due to the finite light speed;
the observing time for the second signal is thus
rB θ rB θ  v
tobs = t − = 1− ; (4.17)
c v c
• now we obtain for the angle, where the electron was when it
emitted the radiation which arrives at the observer at time tobs
vtobs  v −1
θobs = 1− ; (4.18)
rB c
approximating the squared Lorentz factor by
!−1 
v2 v −1  v −1 1  v −1
γ = 1− 2
2
= 1− 1+ ≈ 1− (4.19)
c c c 2 c
for v ≈ c, we finally find
2vtobs 2
θobs = γ , (4.20)
rB
from which we obtain
2vtobs 3   4
γθobs ≈ γ = tobs 2ωB γ3 sin α ≡ ωc tobs , (4.21)
rB 3
because v = rB ωB sin α; moreover, the last equation defines the
angular frequency
3 3  eB 
ωc ≡ ωB γ3 sin α = sin α γ2
2 !2 mc 2
B E
≈ 100 MHz , (4.22)
µG GeV
where the factor 3/2 was introduced for later convenience;
44KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

4.1.3 Synchrotron Spectrum


• since the electric field can depend on ∆θ only through the factor
γ∆θ and γ∆θ depends only through the factor ωc tobs on the ob-
serving time, a Fourier transform must find that the spectrum of
synchrotron radiation can depend on frequency ω only through the
ratio
ω
; (4.23)
ωc
• qualitatively, synchrotron radiation is determined by the following
properties:
– the basic frequency for γ ≈ 1 is the cyclotron frequency
eB
ωcyc = ; (4.24)
mc
– for higher velocities, overtones of the cyclotron frequency
will be added whose amplitudes will decrease by powers of
v/c;
– since ωB ∝ γ−1 , the electron’s angular orbital frequency will
decrease as v/c increases;
– finally, the radiation is limited to a cone which allows the
observer to see the radiation only during time intervals
∆t ∝ γ−3 ω−1
B ; (4.25)
• therefore, we expect a spectrum which consists of a sequence of
sharp maxima at ωB and its overtones and cuts off at
1
ωc ∝ ∝ ωB γ3 ; (4.26)
∆t
• while γ is increasing, the contribution due to the overtones will
increase, and the maxima will be broadened due to the electron
distribution in the Lorentz factor γ and the pitch angle α; thus, a
continuum is formed, which ends near ωc ;
• most of the energy will be emitted near ωc ; since
ωc ∝ ωB γ3 ∝ Bγ2 ∝ BE 2 (4.27)
the electron energy E is proportional to
ωc
r
E∝ , (4.28)
B
more precisely
E ≈ 4νc1/2 B−1/2 erg , (4.29)
where νc = ωc /(2π) is and all quantities are expressed in cgs units;
• synchrotron radiation is strongly linearly polarised because is
results from the orbital motion of electrons perpendicular to the
magnetic field; a circular polarisation parallel to the magnetic field
is effectively averaged out by the pitch-angle distribution;
4.2. PHOTO-IONISATION 45

4.2 Photo-Ionisation

4.2.1 Transition Amplitude


• many astrophysical radiation processes are accompanied by io-
nisation and recombination; in ionising processes, a sufficiently
energetic photon removes an electron from a bound state and pla-
ces it into an unbound state, and the reverse process happens in
recombinations;

• as an example, we consider the cross section for ionising a hy-


drogen atom from its ground state; for doing so, we first need to
compute the transition rate between bound and unbound states
which is caused by a “perturbing” photon;

• according to the quantum-mechanical perturbation theory, the


transition probability within a time interval ∆t is given by

P(∆t) = |a2 (∆t)| , (4.30)

where a is the probability amplitude


Z ∆t
dt 0
a(∆t) = hE |H|Ei , (4.31)
0 ~
and |Ei and |E 0 i are the initial and final states of the electron,
respectively, which are the bound and unbound states here; H is
the Hamiltonian of the “perturbing” interaction; the states |Ei and
|E 0 i are written as eigenstates of the Hamiltonian;

• the canonical momentum of a particle in an electromagnetic field


can be written as
~ = ~p − e A
P ~, (4.32)
c
where A~ is the vector potential; accordingly, the total Hamiltonian
is
~2
P 1  e ~ 2 ~p2 e ~ ~2
 e 2 A
H̄ = = ~p − A = − A · ~p + ; (4.33)
2m 2m c 2m mc c 2m
the Hamiltonian of the interaction between the electron and the
field is thus
e ~
H =− A · ~p ; (4.34)
mc
• we have assume here that A ~ satisfies the transversal (or Coulomb)
gauge, in which the scalar potential Φ = 0 and ∇ ~ ·A~ = 0; then, A
~
satisfies the wave equation

1 ∂2 ~ 2 ~
!
−∇ A=0; (4.35)
c2 ∂t2
46KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

~ x, t) is decomposed into plane waves,


• if A(~

d3 k ~
Z
~
~ x, t) =
A(~ Ak (t)e−ik·~x , (4.36)
(2π)3

the wave equation implies for the Fourier amplitudes

~¨ k (t) + c2 k2 A
A ~k (t) = 0 (4.37)

because the individual amplitudes A ~k (t) are independent of each


~k (t) satisfies an oscillator equation and can be
other; therefore, A
written in the form
~k (t) = A~e eiωt ;
A (4.38)
the dispersion relation ω = kc holds here, ~e is the unit vector in
polarisation direction, and Ak is the amplitude of the given mode
of the vector potential;

• the interaction Hamiltonian between field and electron is thus


eA i(ωt−~k·~x)
H =− e ~e · ~p ; (4.39)
mc

• the transition matrix element hE 0 |H|Ei can now be calculated as


follows: first, the state are represented by the wave functions
0
|Ei = φ(~x)eiEt/~ , |E 0 i = φ0 (~x)eiE t/~ , (4.40)

in which φ(~x) and φ0 (~x) are the spatial amplitudes, and the phase
factors describe the time evolution; using these, the probability
amplitude (4.31) turns into
∆t i~k·~x
Z Z 
A e e
~e · ~p φ(~x) ei(E /~−E/~−ω)t ;
 0
a(∆t) = dt  d3 x φ0∗ (~x)

~c 0 m
(4.41)

4.2.2 Transition Probability

• we abbreviate the notation by introducing the matrix element


between the initial and the final state,
ik·~x~ Z
~ fi ≡ e e
M
e
d3 x φ0∗ (~x) ~p φ(~x) , (4.42)
m m

in terms of which we can write the probability amplitude (4.41) as


Z ∆t
A ~ fi ) dt ei(E /~−E/~−ω)t ;
0
a(∆t) = (~e · M (4.43)
~c 0
4.2. PHOTO-IONISATION 47

• the time integral can be evaluated as follows:


Z ∆t Z ∆t
E0 E
dt e =
it
dt (cos t+i sin t) ,  ≡ − −ω ; (4.44)
0 0 ~ ~
taking the absolute square of this expression, which we shall later
need, we find
Z ∆t !2
sin(∆t/2)
dt e =
it

; (4.45)
/2

0

in the limit of ∆t → ∞ this expression approaches a δ function,

sin2 (∆t/2)
lim → 2π∆tδD () ; (4.46)
∆t→∞ (/2)2

• combining these results, the transition probability (4.31) turns out


to be
2πA2 ∆t
~
2
P(∆t) = δ D () ~
e · M fi ,
(4.47)
~2 c2

and thus the transition rate is
P(∆t) 2πA2 E0 E
!
~ 2
R= = 2 2 δD − − ω ~e · Mfi ; (4.48)
∆t ~c ~ ~

we further introduce the abbreviation


E0 − E
ωfi ≡ , (4.49)
~
i.e. we assign an angular frequency to the energy difference E 0 − E
between the final and initial states;

• what remains is the determination of the field amplitude Ak and


~ fi ; we first consider the amplitude Ak ;
the matrix element M

• the electric field belonging to the vector potential

~k (t) = Ak ~e ei(ωt−~k·~x)
A (4.50)

is
~k (t)
1 ∂A iωAk~e i(ωt−~k·~x)
E~ k = − =− e , (4.51)
c ∂t c
and the energy density in an electromagnetic field in vacuum can
be written as
E~ 2 + B
~ 2 E~ 2
U= = ; (4.52)
8π 4π
• similarly, the energy density in photons of angular frequency ω is
1 N~ω
, (4.53)
2 V
48KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

where the factor 1/2 appears because the two independent pola-
risation directions need to be distinguished; we set N = 1 so that
we shall only have to multiply with the number of photons in the
volume V later; thus, we find by comparing (4.51), (4.52) and
(4.53)
N~ω A2k ω2
r
2π~c2
= ⇒ Ak = (4.54)
2V 4πc2 Vω
for a single photon (N = 1); the unit of Ak is
!1/2
erg s cm2 s  erg 1/2 g1/2 cm1/2
[Ak ] = = = , (4.55)
cm3 s2 cm s

from which we find for the unit of the electric field


ω g1/2
~
[Ek ] = [Ak ] = [Ak ] cm−1 = , (4.56)
c cm1/2 s
as it must be;

• for the transition rate (4.48), we now have

4π2 ~ 2
R= ~e · Mfi δD (ωfi − ω) ; (4.57)
~Vω

• according to our choice of the amplitude Ak , this is the transi-


tion rate caused by a single photon; we must now multiply the
expression (4.57) with the number of available photons, which is

Vd3 k
n(ω) , (4.58)
(2π)3
in which n(ω) is the occupation number of photons with frequency
ω and polarisation state ~e, i.e. the photon number density per
phase-space cell;

• since ω = ck, the volume element in k-space is

ω2 dω
d3 k = k2 dkdΩ = dΩ , (4.59)
c3
and we obtain the number of ionisation transitions per unit time,
unit frequency and unit solid angle as
dP ωn(ω) 2
~ fi δD (ωfi − ω) ;
~e · M
= (4.60)
dtdωdΩ 2π~c3

4.2.3 Transition Matrix Element


• finally, we have to calculate the matrix element M ~ fi as defined in
(4.42); if the wave length of the incoming light, λ = 2π/k, is much
4.2. PHOTO-IONISATION 49

larger than the extent of the wave function of the bound electron,
we can approximate
~
eik·~x ≈ 1 (4.61)
and are left with
Z
~ fi ≈ e
M d3 x φ0∗ (~x) ~p φ(~x) ; (4.62)
m
this approximation is called “dipole approximation” for the follo-
wing reason: the momentum operator ~p can be expressed by the
commutator
im im
~p = [H, ~x] = (H ~x − ~xH) ; (4.63)
~ ~
~ fi (4.62) can be transformed to
therefore, the matrix element M
Z
~ e im 0
Mfi ≈ (E − E) d3 x φ0∗ (~x) ~x φ(~x) = iωfi e~x , (4.64)
m ~
i.e. it turns into a dipole matrix element;

• during the transition from the bound to the free state, the Hamil-
tonian changes, and thus it is preferable in this context to use the
momentum operator instead; we now insert the wave function of
the ground state in the hydrogen atom as the wave function for the
initial state,
e−r/a0
φ(~x) = q , (4.65)
πa30
where a0 is Bohr’s radius

~2
a0 ≡ = 4.7 × 10−8 cm , (4.66)
me2
while the free electron is described by the plane wave
~
eike ·~x
φ0 (~x) = √ , (4.67)
V

where ~ke is the wave vector of the free electron, which is related to
the momentum by ~pe = ~~ke ;

• we first confirm that the dipole approximation can be applied here;


for short-wave light, λ ≈ 1000 Å = 10−5 cm, which is almost three
orders of magnitude larger than Bohr’s radius a0 , λ  a0 ;

• now, the transition probability between the initial and the final
state equals the reverse transition rate,

2
2
final|~e · ~p|initial = initial|~e · ~p|final , (4.68)
50KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

~ we find
and inserting the momentum operator ~p = i~∇,

~~e · ~ke
Z
~
initial|~e · ~p|final = q d3 x e−x/a0 e−ike ·~x ;


(4.69)
πVa30

• denoting the angle between ~ke and ~x by θ, we can write


Z Z ∞ Z π
−x/a0 −i~ke ·~x
3
d xe e = 2π 2
x dx sin θdθ e−x/a0 e−ike x cos θ
0 0
Z ∞ Z 1
= 2π x2 dx d(cos θ) e−x/a0 e−ike x cos θ
−1
Z0 ∞
sin ke x
= 4π x2 dxe−x/a0
0 ke x
8πa30
= , (4.70)
(1 + ke2 a20 )2

which turns the square of the transition matrix element into

2 64π~2 e2 a3 |~e · ~k |2
~ fi =
~e · M 0 e
, (4.71)
(1 + ~ke2 a20 )4
Vm 2

and the number of transitions per unit time, unit frequency and
unit solid angle (4.60) becomes

dP ωn(ω) ~ 2
= ~e · Mfi δD (ωfi − ω) (4.72)
dtdωdΩ 2π~c3
32~e2 ωn(ω)a30 |~e · ~ke |2
= δD (ωfi − ω) ;
Vm2 c3 (1 + ~ke2 a2 )4 0

4.2.4 Cross Section

• the cross section for photo-ionisation is determined by the relation

dP dP Vd3 pe dP Vd3 ke
dNe = =
dt dt (2π~)3 dt (2π)3
! d3 pγ
= σ(ω) cn(ω) ; (4.73)
(2π~)3

this means that the flux of incoming photons, cn(ω) times the
number density of states in phase space, multiplied with the cross
section, is the number of transitions per unit time; this must equal
the transition rate times the number of available final states for the
electron;
4.2. PHOTO-IONISATION 51

• in order to obtain the cross section, we first need to integrate the


expression

dP Vd3 ke 4~e2 ωn(ω)a30 |~e · ~ke |2 d3 ke


= δD (ωfi − ω) (4.74)
dtdωdΩ (2π)3 π3 m2 c3 (1 + ~ke2 a20 )4

over all solid angles; we had assumed that the electron is photo-
ionised into a free final state, and thus we must have ke2  a−2
0 ;
using that, we can approximate

(1 + ke2 a20 )4 ≈ a80 ke8 , (4.75)

and we introduce a coordinate system in which the photon wave


vector ~k and the polarisation vector ~e are parallel to the z- and
x-axes, respectively; then, ~ke is described by the two angles (θ, ϕ),

 sin θ cos ϕ 


 
~ke = ke  sin θ sin ϕ  , (4.76)
cos θ
 

and
~e · ~ke = ke sin θ cos ϕ ; (4.77)

• integrating over the solid angle then yields


Z 2π Z π Z π

dϕ dθ sin θ cos φ = π
3 2
dθ sin3 θ = , (4.78)
0 0 0 3
which leads us to write the number of transitions per unit time and
unit photon frequency

dP Vd3 ke 16~e2 ωn(ω) dke


= δD (ωfi − ω) ; (4.79)
dtdω (2π)3 3π2 m2 c3 a50 ke4

• finally, if the electron is to become unbound,

pe ~2 ke2
~ω = = , (4.80)
2m 2m
thus !1/2
2mω mdω
ke = and dke = , (4.81)
~ ~ke
and !5/2
dke ~ mdω
4
= ; (4.82)
ke 2mω ~

• using
d3 pγ ω2 dω
= , (4.83)
(2π~)3 (2πc)3
52KAPITEL 4. SYNCHROTRON RADIATION, IONISATION AND RECOMBIN

we finally obtain from (4.73)


!5/2
dP Vd3 ke 16e2 ωn(ω) ~
= dω
dtdω (2π)3 3π2 mc3 a50 2mω
! ω2 dω
= σ(ω) cn(ω) (4.84)
(2πc)3
and thus, for the cross section,
!5/2
128πe2 ~
σ= ; (4.85)
3mca0 ω
5 7/2 2m

identifying the (dimension-less) fine-structure constant,

e2
α≡ , (4.86)
~c
and Bohr’s radius (4.66), we finally write this in the intuitive form

(2α)9/2  ω0 7/2
σ(ω) = πa20 , (4.87)
3 ω
with ω0 ≡ c/a0 ;
Kapitel 5

Spectra
further reading: Shu, “The Phy-
sics of Astrophysics, Vol I: Ra-
5.1 Natural Width of Spectral Lines diation”, chapters 22–23; Ry-
bicki, Lightman, “Radiative Pro-
cesses in Astrophysics”, chapters
5.2 Cross Sections and Oscillator Strengths 10; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophysi-
• the analysis of photo-ionisation is equally applicable to systems in cal Processes”, sections 7.1–7.3
which transitions occur between two bound levels; as before, we
have from (4.60) and (4.64)
dP 4e2 ω3fi n(ωfi )
= 3
|~xfi |2 (5.1)
dt ~c
for the dipole transition probability, where |~xfi | is the matrix ele-
ment of the position operator ~x between the two bound states;
• since the photon flux per unit frequency can be expressed by the
occupation number n(ω) times the number of states in phase space,
d3 k n(ω)c 4π k2 dk n(ω)ω2
n(ω)c = = , (5.2)
(2π)3 dω (2π)3 dω 2π2 c2
we can identify the cross section
2π2 c2 4e2 ω3fi n(ωfi )
! !
σ(ω) = |~xfi | δD (ω − ωfi )
2
n(ωfi )ω2fi ~c3
4πe2 ωfi
= |~xfi |2 2πδD (ω − ωfi )
~c
πe2
≡ ffi φ(ω) , (5.3)
mc
in which
φ(ω) = 2πδD (ω − ωfi ) (5.4)
is the line profile function, and ffi is called the “oscillator strength”;
the factor
e2
(5.5)
mc

53
54 KAPITEL 5. SPECTRA

has the dimension length2 time−1 , the profile function has the di-
mension time, i.e. the cross section (5.3) does have the dimension
of an area, as it needs be; the profile function is normalised such
that its integral over frequency is unity,
Z ∞ Z ∞
2π δD (ω − ωfi )dν = δD (ω − ωfi )dω = 1 ; (5.6)
0 0

accordingly, the oscillator strength is


4πe2 ωfi mc 4mωfi
ffi = |~xfi |2 2 = |~xfi |2 ; (5.7)
3~c πe 3~
it is dimension-less because its unit is
g s−1 cm2 g cm2 s2
= 2 =1; (5.8)
erg s s g cm2
for electric dipole transitions, f ≈ 1 typically;

5.2.1 Transition Probabilities


• transitions between bound states occur spontaneously only with a
certain transition probability, i.e. there is an uncertainty in energy
corresponding to their uncertainty in time; if Γ−1 is the mean life
time of the initial state, the uncertainty in energy is approximately
1
∆E ≈~ ⇒ ∆E ≈ Γ~ ; (5.9)
Γ

• therefore, spectral lines are not infinitely sharp but somewhat


widened; the profile function φ(ω) will thus not typically be a δD
function; we shall now calculate the shape of this function if the
mean life time Γ−1 is finite (rather than infinite);

• doing so, we consider a radiative transition between two bound


states 1 and 2 with energy eigenvalues E1 and E2 ; generally, the
Schrödinger equation requires for the amplitude an of the energy
eigenstate n
∂an X
i~ = hn|H|mi am ei(En −Em )t/~ , (5.10)
∂t m

where H is again the interaction Hamiltonian between the electron


and the electromagnetic field;

• modeling the transition between the states |2i and |1i, we can set
a2 = 1 and am = 0 for m , 2; due to the finite life time of the state
|2i, we can use the ansatz

a2 (t) = e−Γt/2 (5.11)


5.2. CROSS SECTIONS AND OSCILLATOR STRENGTHS 55

and obtain
∂a1
i~ = h1|H|2i a2 (t) ei[(E1 −E2 )/~−ω]t
∂t
E1 − E2 + ~ω Γt
( " # )
= h1|H|2i exp i t −
~ 2
Γt
" #
= h1|H|2i exp i(ω − ω12 )t − (5.12)
2

where
E1 − E2
ω12 ≡ (5.13)
~
is the angular frequency corresponding to the transition energy;
upon integrating (5.12) with the boundary condition a1 (t = 0) = 0,
we find
ih1|H|2i 1 − exp [i(ω − ω12 )t − Γt/2]
a1 (t) = ; (5.14)
~ ω − ω12 + iΓ/2

• in the limit of very long times, t  Γ−1 , thus Γt  1, part of the


exponential factor can be approximated as

e−Γt/2 → 0 , (5.15)

and the transition rate becomes


2
|h1|H|2i|2

1
2
|a1 (t)| =
~2 ω − ω12 + iΓ/2

|h1|H|2i|2 1
= ; (5.16)
~2 (ω − ω12 )2 + Γ2 /4

• this changes the profile function to read

Γ
φ(ω) = , (5.17)
(ω − ω12 )2 + (Γ/2)2
which is again defined such that its integral over frequency ν (rather
than angular frequency ω) is unity;

• we now need to determine Γ; we had seen above that the transition


rate is given by (5.1), which can be written as

dP 4mωfi n(ωfi ) e2 ω2fi e2 ω212 n(ω12 )


= |~xfi |2 = f12 , (5.18)
dt 3~ mc3 mc3
which corresponds to the decay rate; since f12 ≈ 1 for the transiti-
ons which are of interest here, a good estimate for Γ is

e2 ω212
Γ≈ ; (5.19)
mc3
56 KAPITEL 5. SPECTRA

• the cross section for the line transition is


πe2 πe2 Γ
σ12 (ω) = f12 φ(ω) = f12 ; (5.20)
mc mc (ω − ω12 )2 + (Γ/2)2
in the centre of the line, i.e. at ω = ω12 , we have
πe2 4Γ 4πe2
σ12 = f12 2 = f12 ; (5.21)
mc Γ mcΓ
using Γ from (5.18), we find
!2
4πe2 mc3 c
σ12 = = 4π ∝ λ212 , (5.22)
mc e2 ω212 ω12
i.e. the cross section in the centre of the line is proportional to the
square of the absorbed wave length;

5.3 Collisional Broadening of Spectral Lines


• collisions between atoms can change the occupation numbers and
the life times of states and modify the line profile of emission or
absorption in this way; the effect of collisions can be described by
random changes of the phase of a1 ,
∂a1
i~ = h1|H|2i exp [i(ω − ω12 )t − Γt/2] eiφ(t) , (5.23)
∂t
where φ(t) is a random function such that

D E 
 1 if there was no collision
= ,
iφ(t)

e until time t (5.24)



 0 else

which means that the average phase factor (5.24) expresses the
probability that the individual system under consideration experi-
enced no collision until time t; in this way, (5.23) formalises the
expectation that (sufficiently energetic) collisions can change the
phase of a1 completely;
• extending this consideration from a single system to an ensemble
of systems and averaging over them, the ensemble average will
turn into an exponential if we assume that the number of collisions
in the ensemble until time t follows a Poisson distribution,
D E
eiφ(t) → e−Γc t/2 , (5.25)

in which Γ−1
c is the mean time between collisions; using this, we
find the change in a1 after time t to be proportional to the integral
of (5.23),
Γt
Z " #
∆a1 ∝ dt exp i(ω − ω12 )t − + iφ(t) ; (5.26)
2
5.4. VELOCITY BROADENING OF SPECTRAL LINES 57

the averages over time and over all systems in the ensemble then
yield, using (5.25),

Γt D iφ(t) E
Z " #
h∆a1 i ∝ dt exp i(ω − ω12 )t − e
2
Γt Γc t
Z " #
= dt exp i(ω − ω12 )t − − ; (5.27)
2 2

• obviously, therefore, the sum of the mean decay and collision rates
Γ and Γc now takes the role that Γ had before, i.e. the collisions
shorten the mean life time to
1 1 1 1
→ = (5.28)
Γ Γ + Γc Γ 1 + Γc /Γ
and thus broadens the line profile;

5.4 Velocity Broadening of Spectral Lines


• a further broadening mechanism is caused by the Doppler effect;
if the emitting atoms (or molecules) move along the line-of-sight,
we observe the frequency
 vk 
ν = ν0 1 + (5.29)
c
instead of the frequency ν0 , if vk is the velocity component along
the line-of-sight;

• it is often appropriate the assume a Gaussian velocity distribution;


the observed line profile is then given by

vk  (vk − v̄)2
Z ∞ " #
dvk  
δD ν − ν0 1 + exp − , (5.30)
2σ2v
p
−∞ 2πσ2v c

where v̄ is the mean velocity of the emitting system, and σv is the


velocity dispersion of its particles; using the identity

1
δD (ax) = δD (x) , (5.31)
a
the Gaussian line profile
!2 
 1 ν − ν0

c 1
c − v̄ 

exp − 2
ν0 2πσ2v ν0
p
2σv
 c2 ν − ν̄ 2 
 !
c 1
= exp − 2 (5.32)
ν0 2πσ2v 2σv ν0
p 
58 KAPITEL 5. SPECTRA

follows, where we have defined


 v̄ 
ν̄ ≡ ν0 1 + , (5.33)
c
i.e. ν̄ is the central line frequency, shifted by the Doppler effect
due to the mean motion of the emitting or absorbing medium;

• according to the definition (5.3) of the scattering cross section, we


have for the Doppler-broadened line

πe2  c2 ν − ν̄ 2 


 !
c 1
σ= f12 exp − 2  ; (5.34)
ν0 2πσ2v 2σv ν0
p
mc

• if the motion of the particles in the medium is thermal, we obtain


for the velocity dispersion
m 2 kT kT
σ = ⇒ σ2v = , (5.35)
2 v 2 m
which gives the cross section
r
πe2  mc2 ν − ν̄ 2 
 !
1 mc2
σ= f12 exp −  ; (5.36)
mc ν0 2πkT 2kT ν0
the cross section in the centre of the line, i.e. at ν = ν̄, is thus
r
πe2 f12 mc2
σ= (5.37)
mc ν0 2πkT
and thus proportional to the inverse root of the temperature;

5.5 The Voigt Profile


• natural line width and collisional broadening result in the Lorentz
profile, while the (typically thermal) Doppler broadening results
in a Gaussian profile; if all effects need to be taken into account,
the resulting profile is a convolution of a Lorentz and a Gaussian
profile,

 v2k 


 
Γ
Z ∞
1
φ(ω) = exp − 2  dvk ,
−∞ (ω − ω12 ) + (Γ/2)
2 2
p
2πσ2v 2σv
(5.38)
where we have neglected for simplicity that the line may be shifted
as a whole due to the mean motion with velocity v̄;

• in (5.38), we need to replace ω12 by


 vk 
ω12 1 + (5.39)
c
5.6. EQUIVALENT WIDTHS AND CURVES-OF-GROWTH 59

to take the Doppler shift of the emission frequency into considera-


tion; this yields

Γ
Z ∞
φ(ω) =
−∞ (ω − ω12 − ω12 vk /c) + (Γ/2)
2 2

 v2k 
 
1
× p exp − 2  dvk ; (5.40)
2πσ2v 2σv

• we now set
√ ω − ω12 c Γ c
v0 ≡ 2σv , u≡ , a≡ (5.41)
ω12 v0 2ω12 v0

and
vk vk
q≡ √ = (5.42)
2σv v0
and obtain
∞ 2
e−q dq
Z
2ac
φ(ω) = √ , (5.43)
πv0 ω12 −∞ (u − q)2 + a2

which is the so-called Voigt profile;

• near its centre, this line profile has a Gaussian shape, while its
wings retain the Lorentzian shape;

5.6 Equivalent Widths and Curves-of-Growth


• two concepts have been introduced for describing the information
contained in observed spectral lines, namely the equivalent width
and the curve-of-growth;

• the equivalent width quantifies the area under a spectral line; if I0


is the specific intensity of the spectral continuum, the equivalent
with is defined as
Z
I0 − I(ν)
W≡ dν , (5.44)
I0

where I(ν) is the spectral (specific) intensity within the line; thus,
the equivalent width of an absorption line is a measure for the
intensity removed from the spectrum, or added to the spectrum by
an emission line;

• the optical depth within the line is

τ = N L σ(ν) , (5.45)
60 KAPITEL 5. SPECTRA

where N is the number of absorbers and L is the extent of the


absorbing medium; the specific intensity is then
I(ν) = I0 e−τ , (5.46)
and thus the equivalent width is
Z h i
W= dν 1 − e−τ(ν) ; (5.47)

• since the cross section is proportional to the profile function φ(ω),


(5.47) can equally be written as
Z h i
W= dν 1 − e−Cφ(ω) , (5.48)

with a constant (frequency-independent) C;


• for small optical depths, τ  1, the exponential function in (5.46)
or (5.47) can be expanded into a Taylor series; this results in
πe2
Z
W= dν N L σ(ν) = N L f12 (5.49)
mc
because the profile function had been normalised such that its inte-
gral over frequency ν yields unity; thus, for small optical depths,
we have
W∝N, (5.50)
i.e. the equivalent width is simply growing linearly with the num-
ber of absorbers;
• if τ  1, the function
1 − e−τ (5.51)
behaves like a step function across the absorption line whose step
width is determined by how τ approaches unity; let ∆ be this step
width in frequency space, then
W ≈ 2∆ ; (5.52)

• for Doppler-broadened lines, we have


r
πe2 f12 mc2 ∆
" !#
mc2
τ = N Lσ = N L exp − ; (5.53)
mc ν0 2πkT 2kT ν0
the condition τ ≈ 1 then requires
!−1 !−1/2
mc2 ∆2 ! 1 πe2 f12 mc2
" #
C1 1
exp − = ≡ (5.54)
2kT ν02 NL mcν0 2πkT C2 NL
with the abbreviations
r
2πkT πe2
C1 ≡ ν0 , C2 ≡ f12 ; (5.55)
mc2 mc
5.6. EQUIVALENT WIDTHS AND CURVES-OF-GROWTH 61

• from this and (5.54), we obtain


! !
2kT 2 NLC2 C1 1/2 NLC2
∆ =
2
ν ln ⇒ ∆ = √ ln ,
mc2 0 C1 π C1
√ (5.56)
i.e. W(N) grows approximately as ln N;

• for a Lorentz profile,


Γ
φ(ω) = , (5.57)
(ω − ω12 )2 + (Γ/2)2
we can approximate the cross section in the limit |ω − ω12 |  Γ by

πe2 Γ C2 Γ
σ= f12 = ; (5.58)
mc (ω − ω12 )2 (ω − ω12 )2
again, we conclude from
! p
τ = N Lσ = ⇒ NLC2 Γ = ∆2 ⇒ ∆= NLC2 Γ ,
√ (5.59)
i.e. the equivalent width grows in this case as N;

• summarising, the curve-of-growth W(N) behaves as





 N small N
 1/2

W(N) ∝   ln
√ N intermediate N (5.60)

 N
 large N

• for determining N, lines with different oscillator strengths f are


used because then the spectral lines fall into different sections of
the curve-of-growth W(N) for the same number N of absorbers;
this may prove difficult when
√ some lines fall into the flat section
of W(N) where W(N) ∝ ln N;
62 KAPITEL 5. SPECTRA
Kapitel 6

Energy-Momentum Tensor and


Equations of Motion
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
6.1 Boltzmann Equation and Energy-MomentumGas Dynamics”, chapters 1–
Tensor 3; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophy-
sical Processes”, sections 8.1–
6.1.1 Boltzmann Equation 8.4; Landau, Lifshitz, “Theoreti-
cal Physics, Vol VI: Hydrodyna-
• there are many possible approaches to hydrodynamics; one of the mics”, chapter I
most direct ones starts by considering the particles of a fluid as
they interact by collisions; the distribution of the particles in phase
space (~x, ~p) then becomes the fundamental physical quantity;

• let f (~x, ~p, t) be the distribution function in phase space, i.e. the
quantity
dN = f (~x, ~p, t) d3 xd3 p (6.1)
is the number of particles in the six-dimensional phase space
element d3 xd3 p;

• forces acting on the particles can be distinguished according to


whether they are “smooth” or “rough” on microscopic length
scales; forces which are “rough” on a microscopic scale are due to
the direct interactions between the particles and are summarised
as “collisions”; forces which are smooth on a microscopic scale
are described by a potential U,

~ ;
F~ = −∇U (6.2)

• without collisions, the distribution function would satisfy the


collision-less Boltzmann equation,

d f (~x, ~p, t)
=0, (6.3)
dt

63
64KAPITEL 6. ENERGY-MOMENTUM TENSOR AND EQUATIONS OF MOTI

in which the time derivative is to be taken along the particle tra-


jectories; this equation says that in absence of collisions, particle
trajectories are not lost in phase space, which is a consequence of
Liouville’s theorem for Hamiltonian systems;

• writing the time derivative in (6.3) explicitly, the collision-less


Boltzmann equation reads

∂f ˙∂f ∂f
+ ~x + ~p˙ =0 (6.4)
∂t ∂~x ∂~p

~
or, using (6.2) and ~p˙ = F,

∂f ˙ ~ ~ ∂f = 0
+ ~x · ∇ f − ∇U (6.5)
∂t ∂~p

• in presence of collisions, the right-hand side of the Boltzmann


equation is changed from zero to

0 → C[ f (~x, ~p, t)] (6.6)

where C[ f ] is a functional of the distribution function f which


describes how it is changed by collisions;

• collisions happen between particles with momenta ~p1 and ~p2 and
lead to momenta ~p01 and ~p02 ; the scattering cross section as a func-
tion of the solid angle be σ(Ω); if the distribution functions are
abbreviated as follows

f (~x, ~p1 , t) ≡ f1 , f (~x, ~p2 , t) ≡ f2


f (~x, ~p01 , t) ≡ f10 , f (~x, ~p02 , t) ≡ f20 , (6.7)

the collision term can be written as


Z
C[ f ] = dΩd3 p2 σ(Ω) |~v1 − ~v2 | f10 f20 − f1 f2 ;
 
(6.8)

this is a result of kinetic theory; we shall later see that this term
drops out when moments of the Boltzmann equation are formed;

6.1.2 Moments; Continuity Equation

• hydrodynamics builds upon the central assumption that the mean


free path λ of the particles making up a fluid is very much smaller
than the extent L of the system under consideration, λ  L; this
means that integrations over phase space cells typically average
over many collisions;
6.1. BOLTZMANN EQUATION AND ENERGY-MOMENTUM TENSOR65

• this suggests to solve Boltzmann’s equation by taking its moments


(as we did before when studying radiation transport); we assume
~ = 0, and form
for now that there are no external forces, F~ = −∇U
the first and second moments of the equation
∂f ˙ ~
+ ~x · ∇ f = C[ f ] ; (6.9)
∂t
• the moments are formed by multiplying Eq. (6.9) with 1 or the four-
momentum pµ and integrating over momentum space, weighting
with a factor E −1 (~p), in short, by applying the operators
Z 3 Z 3
d p d p µ
and p (6.10)
E(~p) E(~p)
to (6.9);
• because of momentum conservation during collisions, the collision
terms on the right-hand side of (6.9) drop out: no net momentum is
exchanged on average because under the basic underlying assump-
tion λ  L, collisions are very frequent and it can be assumed that
collisions “away from ~p” and “to ~p” are in equilibrium; note that
this is different from starting with a collision-less equation in the
first place; here, the collision terms are present, but have no net
effect because of the very short mean free path;
• the first moment yields
d p ∂f
Z 3 Z 3
d p ˙ ~
+ ~x · ∇ f = 0 ; (6.11)
E(~p) ∂t E(~p)

• using the definition


d3 p µ µ
Z
µ
J ≡c p f (x , ~p) , (6.12)
E(~p)
we find Z
J =
0
d3 p f (xµ , ~p) (6.13)

because p0 = E/c; quite obviously, J 0 is the particle density n(xµ )


at the position ~x and the time t = x0 /c;
• similarly, the spatial components of J µ are
Z 3
d p i µ
J =c
i
p f (x , ~p) , (6.14)
E(~p)

or, using the expressions E = γmc2 and ~p = γm~x˙ which are valid
for relativistic and non-relativistic particles alike,

~x˙
Z
~
J= d3 p f (xµ , ~p) ; (6.15)
c
66KAPITEL 6. ENERGY-MOMENTUM TENSOR AND EQUATIONS OF MOTI

since the average particle velocity is

d3 p ~x˙ f
R Z
1
~v = R = d3 p ~x˙ f , (6.16)
3
d pf n

the spatial vector J~ turns out to be the average particle current n~v
divided by the light speed c; therefore, the first moment equation
can be written in the form
∂J 0 ∂J i ∂n ~
+ i = + ∇ · (n~v) = 0 , (6.17)
x0 ∂x ∂t
upon multiplying this equation for the evolution of the number den-
sity n with the particle mass m, we obtain the continuity equation
for the mass,
∂ρ ~
+ ∇ · (ρ~v) = 0 , (6.18)
∂t
which can also be expressed by the vanishing four-divergence
∂J µ
=0 (6.19)
∂xµ
of the current J µ , which is the relativistic generalisation of the
continuity equation (6.18);

6.1.3 Energy-Momentum Tensor


• we now consider the second moments, which are obtained by
applying the second integral operator from (6.10) to the Boltzmann
equation (6.9); we first study the tensor
Z 3
d p µ ν
c 2
p p f ≡ T µν ; (6.20)
E(~p)

• in the non-relativistic limit, we have


E γmc2 ~x˙2 
 
p = =
0
= mc 1 + 2  ,

(6.21)
c c 2c
and thus the time-time component of the tensor T µν is

~x˙2 
Z  
00
= mc 2
d p 1 + 2  f
3 

T
2c
m~x˙2
Z
= ρc2 + d3 p f = ρc2 + n¯ , (6.22)
2
where ¯ is the mean kinetic energy of the fluid particles

d3 p m2 ~x˙2 f
R Z
1 m
¯ ≡ R = d3 p ~x˙2 f ; (6.23)
3
d pf n 2
6.1. BOLTZMANN EQUATION AND ENERGY-MOMENTUM TENSOR67

• the space-time part of the tensor T µν reads


Z Z
T 0i
= c d pp f =
3 i
d3 p γmc ẋi f

~x˙2 
!i
~q
Z  
d p 1 + 2  mc ẋ f = ρc~v +
3  i
, (6.24)


2c c

where ~q is the current of the kinetic energy


 ˙2 
 m~x  ˙
Z
~q ≡ 3 
d p   ~x f ; (6.25)
2

the first term on the right-hand side of (6.24), ρc~v, is the mass
current times the light speed;
• finally, the space-space components of T µν are
Z 3 Z 3
d p i j d p
T = c
ij 2
pp f =c 2
2
γm2 ẋi ẋ j f
E(~p) mc
Z
≈ d3 p m ẋi ẋ j f , (6.26)

which can be interpreted as the (three-dimensional) stress-energy


tensor;
• we now return to the (simplified) Boltzmann equation (6.9) and
rewrite it somewhat, using

~p ~x˙
E = γmc2 , ~p = γm~x˙ ⇒ = 2 (6.27)
E c
as well as p0 = E/c and x0 = ct; with that, we first obtain
∂f ~p ~
c + c2 ∇ f = C[ f ] , (6.28)
∂x 0 E
and from that,
p0 ∂ f ~p ~
c2
+ c2 ∇ f = C[ f ] ; (6.29)
E ∂x 0 E

• this can obviously be written in the form of a four-divergence,


∂ ∂ 2~
0
! !
2p p
fc + · fc = C[ f ] , (6.30)
∂x0 E ∂~x E

and an integration over d3 p pν yields


R

∂T 0ν
=0, (6.31)
∂xν
i.e. this moment of the Boltzmann equation is equivalent to the
vanishing four-divergence of T 0ν ;
68KAPITEL 6. ENERGY-MOMENTUM TENSOR AND EQUATIONS OF MOTI

• we can proceed in much the same way with the spatial components
T iν of T µν ; doing so, we first write
∂f ~ f = cpi ∂ f + pi ẋ j ∂ f = C[ f ] pi
cpi + pi ~x˙ · ∇ (6.32)
∂x 0 ∂x0 ∂x j
and insert
pj
ẋ j = c2 (6.33)
E
to find
E ∂f 2 i p ∂f
j
c2 pi + c p = C[ f ] pi , (6.34)
cE ∂x 0 E ∂x j

which equals

c2 0 i ∂ f c2 i j ∂ f
p p 0+ pp = C[ f ] pi ; (6.35)
E ∂x E ∂x j

after integration over d3 p, this yields

∂ i0 ∂ i j ∂T iν
T + T = =0; (6.36)
∂x0 ∂x j ∂xν

• thus, the moment equations of the Boltzmann equation can be ex-


pressed by the vanishing four-divergence of the energy-momentum
tensor T µν ,
∂T µν
=0; (6.37)
∂xν
this indicates how the hydrodynamical equations can be relativisti-
cally generalised; we now return to the non-relativistic expressions
for T µν ;

• the time-component of the divergence equation implies

1∂  2 ~q
!
~

ρc + n¯ + ∇ · ρc~v + =0; (6.38)
c ∂t c
using the continuity equation (6.18) to eliminate the partial time
derivative of the density ρ from (6.38), we find

~ · (ρ~v) + ∂ n¯ + c∇~ · (ρ~v) + 1 ∇


 
− c∇ ~ · ~q = 0 (6.39)
∂t c c
and thus
∂(n¯ ) ~
+ ∇ · ~q = 0 ; (6.40)
∂t
this is the continuity equation for the energy, i.e. the expression of
energy conservation;

• the spatial part of the divergence equation implies

1∂ qi ∂
!
cρv +
i
+ j Tij = 0 ; (6.41)
c ∂t c ∂x
6.2. THE TENSOR VIRIAL THEOREM 69

here, we can neglect the term ~q c−2 because the energy flow will
be much slower than the speed of light, and obtain

∂(ρvi ) ∂T i j
+ =0, (6.42)
∂t ∂x j
which is the equation of momentum conservation in the fluid
approximation;

• summarising, we obtain the equations of motion


 ˙2 
∂(n¯ ) ~  m~x  ˙
Z
+ ∇ · ~q = 0 , with ~q ≡ d3 p   ~x f
∂t 2
∂(ρvi ) ∂T i j
Z
+ =0 , with T ≡ij
d3 p m ẋi ẋ j f (6.43)
∂t ∂x j

• we now split the velocities ~x of the particles into the mean velocity
~v and a (usually thermal) velocity ~u about the mean,

~x˙ = ~v + ~u ; (6.44)

the kinetic energy then reads


Z Z Z
3 m 3 m 2 m
n¯ = ~v + ~u f = d p ~v f + d3 p ~u2 f ,
2
d p
2 2 2
(6.45)
because the (thermal) velocities ~u vanish on average, by definition;
we thus obtain
ρ nm D 2 E
n¯ = ~v2 + ~u , (6.46)
2 2
D E
where ~u2 is the mean-squared thermal velocity,

d3 p ~u2 f
R
D E
~u2 ≡ R ; (6.47)
d3 p f

the second term in (6.46) is the internal energy , the first is the
kinetic energy of the mean fluid motion; if the internal energy is
thermal,
3
 = nkT ; (6.48)
2

6.2 The Tensor Virial Theorem

6.2.1 A Corollary
• an important theorem for all systems satisfying the equations we
have derived is the tensor virial theorem, which holds independent
70KAPITEL 6. ENERGY-MOMENTUM TENSOR AND EQUATIONS OF MOTI

of the particular form of the energy-momentum tensor T µν ; in


order to prove it, we first demonstrate that the relation
Z Z
d dF
d x ρF =
3
d3 x ρ (6.49)
dt V V dt

holds for arbitrary functions F(~x, t) and integration volumina V;

• the proof proceeds as follows: first, the total time derivative of the
integral is

∂ ∂(ρF)
Z !Z Z
d ~
d x ρF =
3
+ ~v · ∇ d x ρF =
3
d3 x (6.50)
dt V ∂t V V ∂t

because the integration over d3 x removes the dependence on ~x and


makes the gradient vanish; then, the remaining integral over the
partial time derivative is

∂(ρF) ∂F
Z Z !
d x 3
= d x ρ 3 ~
− F ∇ · (ρ~v)
V ∂t V ∂t
∂F ~
Z !
= d x ρ
3 ~
− ∇ · (Fρ~v) + ρ~v · ∇F
V ∂t
∂F
Z !
= d x ρ
3 ~
+ ρ~v · ∇F
V ∂t
Z
dF
= d3 x ρ , (6.51)
V dt

where Gauss’ theorem was employed again to remove the diver-


gence term; this completes the proof;

6.2.2 Second Moment of the Mass Distribution

• we now consider the second spatial moment of the mass distributi-


on,
Z
ij
I ≡ d3 x ρxi x j (6.52)
V

and its second time derivative,

d2 I i j d(xi x j )
" Z # Z
d d d
= d3
x ρx i j
x = d3
x ρ , (6.53)
dt2 dt dt V dt V dt

where the theorem (6.49) was used with F = xi x j ; notice that the
volume V is fixed, so that the coordinates xi introduced in (6.52)
do not explicitly depend on time;
6.2. THE TENSOR VIRIAL THEOREM 71

• the integral on the right-hand side of (6.53) can be transformed to


read
d(xi x j )
Z Z
d xρ
3
= ~ i x j)
d3 x ρ~v · ∇(x
V dt
ZV
∂(xi x j )
= d3 x ρvk
∂xk
ZV  
= d3 x ρ vk δik x j + vk xi δkj
ZV  
= d3 x ρ vi x j + v j xi ; (6.54)
V

we now take the second total time derivative remaining from (6.53),
which can again be replaced by a partial time derivative as in (6.50)
before,
d2 I i j ∂(ρvi x j ) ∂(ρv j xi )
Z " #
= 3
d x + ; (6.55)
dt2 V ∂t ∂t

• now we establish a connection with the energy-momentum tensor


by noting that
∂(xi T jk ) i ∂T
jk
i ∂(ρv )
j
= T δ
jk i
+ x = T ji
− x (6.56)
∂xk k
∂xk ∂t
holds because of momentum conservation (6.42); we can use again
that the coordinates xi do not explicitly depend on time to write
this last result in the form
∂(ρxi v j ) ∂T jk
= T ji − xi k (6.57)
∂t ∂x

• inserting this into (6.55), we can bring the time derivative of I i j


into the following form:
d2 I i j ∂  il j
Z "  #
= d x T +T − l T x +T x ;
3 ij ji jl i
(6.58)
dt2 V ∂x
the second term in this last equation is a divergence, which may be
transformed by Gauss’ theorem into a vanishing boundary term;
the symmetry of T i j then implies
d2 I i j
Z
= 2 d3 x T i j , (6.59)
dt2 V

which is the tensor virial theorem;


• although the described prorcedure of taking moments is mathema-
tically meaningful and correct, it has not brought us closer to a
solution of Boltzmann’s equation: we do not know the form of the
stress-energy tensor yet; only when this is defined can we solve
the equations of motion;
72KAPITEL 6. ENERGY-MOMENTUM TENSOR AND EQUATIONS OF MOTI

• here, too, there are numerous possibilities for “closing” the mo-
ment equations; they are similar to the procedure which we have
applied to describe radiation transport in the local thermodynami-
cal equilibrium: the fluid approximation asserts that the mean free
path, λ, of the fluid particles is much smaller than the dimension
of the system; accordingly, we can define a small parameter
λ
ε≡ , ε1 (6.60)
L
and expand the distribution function in powers of ε, such as

f = f0 + ε f1 + ε2 f2 + . . . , (6.61)

where f0 is the distribution function in the limit ε → 0;


Kapitel 7

Ideal and Viscous Fluids


further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
7.1 Ideal Fluids Gas Dynamics”, chapters 3–
4; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophy-
7.1.1 Energy-Momentum Tensor sical Processes”, sections 8.5–
8.7; Landau, Lifshitz, “Theoreti-
• we first start with a distribution function f0 which describes an cal Physics, Vol VI: Hydrodyna-
infinitely extended medium in thermal equilibrium; then, f0 is the mics”, chapters I and II
Maxwellian distribution
m~u2
!
4n
f0 = √ exp − , (7.1)
π(2kT m)3/2 2kT

which contains the velocity ~u relative to the mean velocity of the


fluid, according to our previous notation

~u = ~x˙ − ~v ; (7.2)

• the stress-energy tensor is then


Z
T =
ij
d3 p m(vi + ui )(v j + u j ) f0 =
Z Z
= mv v i j
d p f0 + m d3 ui u j f0
3

Z Z
+ mv i
d p u f0 + mv
3 j j
d3 p ui f0 ; (7.3)

the two latter terms vanish because f0 is isotropic, the first term is
Z
mv v
i j
d3 p f0 = ρvi v j , (7.4)

and the second term is


Z
m d3 p ui u j f0 = 0 for i, j (7.5)

73
74 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

and
nm D 2 E
~u for i= j (7.6)
3
which is the gas pressure P,
ρ D 2E
P= ~u ; (7.7)
3
• this interpretation also follows from our earlier considerations; we
had
3 ρ
 = nkT = h~u2 i ⇒ ρh~u2 i = 3nkT = 3P (7.8)
2 2
in case of thermal motion;
• in this way, we obtain the complete stress-energy tensor,
T i j = ρvi v j + Pδi j ; (7.9)

• in addition, we require the flux ~q of the kinetic energy,


Z
m
~q = d3 p ~v + ~u 2 ~u + ~v f0 ;
 
(7.10)
2
we write it in components,
Z
m 2 
q =
i
d3 p ~v + ~u2 + 2~u · ~v (vi + ui ) f0
2
ρ 2 i ρ 2 i ρ ρ
= ~v v + ~v hu i + vi h~u2 i + hui~u2 i
2 2 2 2
+ ρvi v j hu j i + ρv j hui u j i , (7.11)
where we have used again the averages of arbitrary quantities Q,
Z
1
hQi ≡ d3 p (Q f ) ; (7.12)
n
due to the isotropy of f0 , the second, fourth and fifth term on the
right-hand side of (7.13) vanish, the third term equals vi , and for
the last term we use again
Z
m d3 p (ui u j f0 ) = Pδi j = ρhui u j i , (7.13)

thus
ρv j hui u j i = v j Pδi j = Pvi ; (7.14)
• then, the flux of kinetic energy becomes
~v2  P ~v2
! !
~q = + + ρ~v ≡ + w ρ~v , (7.15)
2 ρ ρ 2
where
+P
w≡ (7.16)
ρ
is the “heat function” (enthalpy) per mass; the enthalpy occurs
here instead of the energy because the the pressure work exerted
by the fluid needs to be taken into account;
7.1. IDEAL FLUIDS 75

7.1.2 Equations of Motion


• substituting now the expressions (7.9) for the stress-energy tensor
T i j and (7.15) for the flux of kinetic energy into the equations of
motion, we obtain the equations of motion for an “ideal” fluid:
first, the equation of continuity,

∂ρ ~ ∂ρ ~ · ~v = dρ + ρ∇
~ + ρ∇ ~ · ~v = 0 ; (7.17)
+ ∇ · (ρ~v) = + (~v · ∇)ρ
∂t ∂t dt
next, the equation of energy transport,

∂ ρ~v2 ~v
! " 2 ! #
+ +∇· ~ + w ρ~v = 0 , (7.18)
∂t 2 2

and finally the equation for the momentum transport,

∂(ρvi ) ∂  
+ j ρvi v j + Pδi j = 0 ; (7.19)
∂t ∂x

• we rewrite the last two equations in order to bring them into a


particularly manageable form; from energy conservation (7.18),
we have
~v2 ∂ρ ρ ∂~v2 ∂
0 = + +
2 ∂t 2 ∂t ∂t
~
v2
~ + ∇~
v2
~ · (ρ~v) + ∇(~
~ v) + ∇
~ · (P~v) ,
+ ρ~v · ∇ (7.20)
2 2
which we can simplify with the aid of the continuity equation,

ρ ∂~v2 ∂ ~ ~v + ∇(~
2
+ + ρ~v · ∇ ~ v) + ∇
~ · (P~v) = 0 ; (7.21)
2 ∂t ∂t 2
momentum conservation requires

∂ρ ∂vi
0 = vi +ρ
∂t ∂t
∂ρ ∂vi ∂v j ∂P
+ vi v j j + ρv j j + ρv j j + j δi j , (7.22)
∂x ∂x ∂x ∂x
which can be written in vector form as
∂ρ ∂~v
0 = ~v +ρ
∂t ∂t
~ ~ v + ρ~v(∇
+ ~v(~v · ∇)ρ + ρ(~v · ∇)~ ~ · ~v) + ∇P
~ ; (7.23)

for simplifying that, we first use the continuity equation again in


its form
∂ρ ~ + ρ∇~ · ~v = 0
+ ~v · ∇ρ (7.24)
∂t
76 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

to get rid of the first, third, and fifth terms on the right-hand side
of (7.23); we thus obtain
∂~v ~ v + ∇P
~ =0;
ρ + ρ(~v · ∇)~ (7.25)
∂t
moreover, we can use the identity

~
v2
!
~ v=∇
(~v · ∇)~ ~ ~ × ~v) × ~v
+ (∇ (7.26)
2
to find
∂~v ~ ~v
! ~
+∇ ~ × ~v) × ~v = − ∇P ;
+ (∇ (7.27)
∂t 2 ρ

• multiplying (7.25) with ~v yields

ρ ∂~v2 ρ ~ v2 + ~v · P
~=0;
+ (~v · ∇)~ (7.28)
2 ∂t 2

• inserting this into the equation of energy conservation, the first,


second, and next-to-last terms cancel to yield
∂ ~ ~ · ~v ;
+ ∇ · (~v) = −P∇ (7.29)
∂t
using here the energy per mass, ε ≡ /ρ, this equation reads
(ερ) ~ ∂ε ∂ρ ~ · (ρ~v) + ρ~v · ∇ε
~
+ ∇ · (ρε~v) = ρ + ε + ε∇
∂t ∂t ∂t !
∂ε ~ = −P∇ ~ · ~v ;
= ρ + ~v · ∇ε (7.30)
∂t

• we have now arrived at Euler’s equations,


∂ρ ~ ∂ ~ ~ · ~v ,
+ ∇ · (ρ~v) = 0 , + ∇ · (~v) = −P∇
∂t ∂t
∂~v ~
~ v = − ∇P
+ (~v · ∇)~ , (7.31)
∂t ρ
which described the conservation laws for mass, energy, and mo-
mentum in the approximation of an ideal fluid;

• had we allowed (conservative) external forces, with


~
F~ext = m~v˙ = −m∇Φ (7.32)

with the potential Φ of the force, the right-hand side of the momentum-
conservation equation had acquired an additional potential gradi-
ent,
∂~v ~
~ v = − ∇P − ∇Φ
+ (~v · ∇)~ ~ ; (7.33)
∂t ρ
7.1. IDEAL FLUIDS 77

7.1.3 Entropy
• the entropy of an ideal, monatomic gas is
!
3k P
s= ln γ (7.34)
2m ρ
per unit mass, where we have omitted an additive constant; γ = 5/3
is the adiabatic index; the total time derivative of this specific
entropy is
∂s ργ −γ ∂P −γ−1 ∂ρ
!
~
+ (~v · ∇)s = ρ − Pγρ (7.35)
∂t P ∂t ∂t
i ∂
! !
P
+ v i ln γ
∂x ρ
1 ∂P γ ∂ρ
= −
P ∂t ρ ∂t
γ
iρ −γ ∂P −γ−1 ∂ρ
!
+ v ρ − γρ P i
P ∂xi ∂x
1 ∂P γ ∂ρ vi ∂P γvi ∂ρ
= − + − ;
P ∂t ρ ∂t P ∂xi ρ ∂xi
• according to the continuity equation, we can simplify
γ ∂ρ γ ∂ρ ~
" # " #
~
+ (~v · ∇)ρ = ~ ~ v , (7.36)
+ ∇ · (ρ~v) − ρ∇ · ~v = −γ∇·~
ρ ∂t ρ ∂t
and we further had
2
P = nkT =  , (7.37)
3
and therefore
1 ∂P 1 ∂ 1 ∂P 1 ∂
= and = ; (7.38)
P ∂t  ∂t P ∂xi  ∂xi
• thus, the entropy equation (7.36) reads
∂s ~ 1 ∂ 1 ~ + γ∇ ~ · ~v
+ (~v · ∇)s = + (~v · ∇)
∂t  "∂t 
1 ∂
#
= ~ ~
+ (~v · ∇) + γ ∇ · ~v ; (7.39)
 ∂t
finally, we use
!
5 3 5 3
γ = · nkT = nkT = + 1 nkT =  + P , (7.40)
3 2 2 2
which allows us to conclude
∂s 1 ∂ ~
" #
~
+ (~v · ∇)s = ~
+ ∇ · (~v) + P∇ · ~v = 0 , (7.41)
∂t  ∂t
because the expression in square brackets vanishes due to energy
conservation;
78 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

• we thus obtain the conservation of entropy,

∂s ~ = ds = 0 ,
+ (~v · ∇)s (7.42)
∂t dt
which is intuitively expected in the absence of dissipation;

7.2 Viscous Fluids

7.2.1 Stress-Energy Tensor; Viscosity and Heat Con-


ductivity
• so far, we have neglected gradients in the temperature because we
have assumed a Maxwellian velocity distribution belonging to a fi-
xed temperature for the particles; likewise, the stress-energy tensor
of the ideal fluid does not contain velocity- or density gradients;
such terms will now be included;

• the form of the corresponding expressions in the stress-energy


tensor and in the energy current ~q can be computed by expanding
the phase-space distribution function to the next order beyond the
ideal-fluid term f0 ; we abbreviate here and justify the form of the
appearing terms by physical arguments;

• differential velocity terms can be expressed by the tensor

∂vi
vij ≡ , (7.43)
∂x j
whose trace is the divergence of ~v,

∂vi ~
trvij = = ∇ · ~v ; (7.44)
∂xi
we subtract this trace from vij in order to obtain a trace-free residu-
al,
1 ~
vij − δij ∇ · ~v ; (7.45)
3
this expression describes pure shear flows which deform the medi-
um, while the part proportional to ∇ ~ · ~v describes the compression
of the medium; we thus obtain the shear tensor
!
1 i~ ~ · ~v ,
σ j ≡ 2η v j − δ j ∇ · ~v − ζδij ∇
i i
(7.46)
3

in which η and ζ are constants which remain to be determined and


describe the strength of shear flows and compression, respectively;
7.2. VISCOUS FLUIDS 79

• the tensor vij from (7.43) can be split into a symmetric and an
antisymmetric part:

∂vi 1 ∂vi ∂v j 1 ∂vi ∂v j


! !
= + + − ; (7.47)
∂x j 2 ∂x j ∂xi 2 ∂x j ∂xi

• if the velocity field is caused by rigid rotation,

~v = ω
~ × ~x , vi =  ijk ω j xk , (7.48)

the antisymmetric part turns into

1  ∂(kli ωk xl ) ∂(kl ωk xl ) 


 j 
1 ∂vi ∂v j
!
− = −
2 ∂x j ∂xi 2  ∂x j ∂xi 
 

1 i 
= k j − kij ωk
2
= − ijk ωk , 0 , (7.49)

while the symmetric part vanishes;

• in order to prevent rigid rotation from causing dissipation, we use


only the symmetric part and write from now on

1 ∂vi ∂v j
!
vj ≡
i
+ ; (7.50)
2 ∂x j ∂xi

• we augment the energy-momentum tensor of the ideal fluid now


by this shear tensor and obtain

T ij = ρvi v j + Pδij − σij , (7.51)

in which the minus sign is conventional;

• accordingly, we can modify the energy current; first, a temperature


gradient will cause an energy current against the gradient which
will be quantified by a heat conductivity κ,

∂T
−κ ; (7.52)
∂xi
moreover, we need to add a contribution to the energy transport
which is due to the flow of the velocity gradient,

− v j σij ; (7.53)

and therefore the energy current then reads


ρ  ∂T
q =
i
~v + w vi − κ
2
− v j σij ; (7.54)
2 ∂xi
80 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

• the additional contributions are thus characterised by three coef-


ficients, i.e. the heat conductivity κ, and the two viscosity coeffi-
cients η and ζ;

• the form of these coefficients can be computed by inserting the


ansatz f = f0 + f1 for the phase-space distribution function into
the Boltzmann equation and iteratively searching for solutions,
evaluating the collision terms;

• κ must have the dimension (energy density × velocity)/(temperature


gradient), hence
erg cm cm erg
[κ] = 3
= ; (7.55)
cm s K cm s K

• similarly, one finds that the dimension of η is (energy densi-


ty)/(velocity gradient), or
erg s erg s
[η] = 3
= ; (7.56)
cm cm cm3

7.2.2 Estimates for Heat Conductivity and Viscosity


• we consider a gaseous system in thermal equilibrium with a tem-
perature T whose particles are moving randomly in all directions;
let ∆A be the area of a screen perpendicular to the y axis;

• per unit time,


n v ∆A
(7.57)
6
particles will fly through the screen, either from left to right or the
other way round; the factor of 1/6 is owed to the fact that typically
only 1/3 of the particles is flying along the y axis, and of those,
only 1/2 in either direction;

• the mean free path is λ = (nσ)−1 , if σ is the collisional cross


section of the particles; particles coming from the left transport
properties of the gas from y − λ to y, and particles coming from the
opposite direction transport properties from y + λ to y; interesting
effects occur if these properties have gradients;

• if the particle number density, n, changes along y, ∂n/∂y , 0, the


net number of flowing particles is
n(y + λ)v∆A n(y − λ)v∆A v∆A ∂n
− ≈ 2 λ, (7.58)
6 6 6 ∂y
where we have implicitly assumed that the mean free path λ is very
short compared to the typical length scale of the number-density
gradient;
7.2. VISCOUS FLUIDS 81

• the diffusion coefficient D, which relates the particle current per


unit area to the number-density gradient,
∆N ∂n
=D , (7.59)
∆A ∂y
thus is

D= ; (7.60)
3
• if the temperature changes along y, ∂T/∂y , 0, the particles
transport energy,
∆ nv∆A 
= (y + λ) − (y − λ)

(7.61)
∆A 6∆A
nvλ ∂ ∂T nv cv λ ∂T
!
= = ,
3 ∂T ∂y 3 ∂y
where cv is the heat capacity at constant volume; using
∆ ! ∂T
=κ , (7.62)
∆A ∂y
we find
nv cv λ v cv
κ= = ; (7.63)
3 3σ
since  is the energy per particle, we have
3k
cv = (7.64)
2
and thus
vk
κ= , (7.65)

which has the physical unit
cm erg 1 erg
[κ] = = , (7.66)
s K cm2 cm s K
as expected from (7.55);
• in complete analogy, the transport of momentum is
∆p x nv∆A ∂v x
= 2mλ , (7.67)
∆t 6 ∂y
i.e. momentum in the x direction is transported in this way along
the y axis;
• the change of momentum per unit time is a force; the force per
unit area is
∆p x nvmλ ∂v x ! ∂v x
Fx = = =η , (7.68)
∆t ∆A 3 ∂y ∂y
where η is the viscosity coefficient; from this, we obtain
nvmλ mv 2m
η= = = κ, (7.69)
3 3σ 3k
which clarifies where the form of the expressions for κ and η
originate from;
82 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

7.2.3 Equations of Motion for Viscous Fluids


• again, we can find the equations of motion by inserting ~q and T i j
into the general results (6.18) and (6.43), where
ρ 3 ρ
n¯ = ~v2 + nkT ≡ ~v2 +  ; (7.70)
2 2 2
obviously, the continuity equation remains valid without change;

• the force equation now reads, with the stress-energy tensor (7.51)
and the shear tensor (7.46)

∂(ρvi ) ∂(ρvi v j ) ∂P
+ + i
∂t ∂x j ∂x
~ · ~v)  ~

 + ζ ∂(∇ · ~v) ;
 ∂vi j 1 ∂(∇
= 2η  j − (7.71)
∂x 3 ∂xi ∂xi

• the right-rand side of this equation can be simplified to read

∂ ∂vi ∂v j 2η ∂ ∂v j ∂ ∂v j
!
η j + − + ζ
∂x ∂x j ∂xi 3 ∂xi ∂x j ∂xi ∂x j
η ~ · ~v)
 ∂(∇
~
= η∇ v +
2 i
+ζ ; (7.72)
3 ∂xi

• the left-hand side of (7.71) can be transformed by means of the


continuity equation,

∂~v ∂ρ ~ + ρ(~v · ∇)~


~ v + ρ~v · (∇
~ · ~v) =
ρ + ~v + ~v · (~v · ∇)ρ
∂t ∂t
∂~v ∂ρ
!
= ρ + ~v ~ ~ ~ v=
+ (~v · ∇)ρ + ρ∇ · ~v + ρ(~v · ∇)~
∂t ∂t
d~v
= ρ ; (7.73)
dt

• this leads us to the Navier-Stokes equation,

d~v η ~ · ~v)
 ∂(∇
~ ~
ρ = −∇P + η∇ v +
2 i
+ζ , (7.74)
dt 3 ∂xi
which simplifies to Euler’s equation if the viscosity parameters
vanish, η = 0 = ζ;

• the energy-conservation equation is

∂ ρ~v2 ∂ ~v2 ∂T
! " ! #
+ + i ρ +w v −κ
i
− v jσ = 0 ;
ij
(7.75)
∂t 2 ∂x 2 ∂xi

this expression can be simplified as follows:


7.2. VISCOUS FLUIDS 83

• the equation of momentum conservation can be written as

∂(ρvi ) ∂  
+ j ρvi v j + Pδi j − σi j = 0 ; (7.76)
∂t ∂x
multiplying this with vi and using the continuity equation enables
us to write
∂  ρ 2 ∂ 1 2 j ∂P ∂σi j
!
~v + j ρ~v v = −vi + vi j ; (7.77)
∂t 2 ∂x 2 ∂xi ∂x
subtracting this from the energy conservation equation yields

∂ ∂(ρwvi ) ∂ ∂T ~ − σi j ∂v j = 0 ; (7.78)
!
+ − i κ − (~v · ∇)P
∂t ∂x i ∂x ∂xi ∂xi

• using the definition of the enthalpy (7.16), we can transform

∂(ρwvi ) ∂[( + P)vi ] ~ + P) + ( + P)∇


~ · ~v , (7.79)
= = (~v · ∇)(
∂xi ∂xi
and the energy conservation equation can be cast into the form
∂ ~ ~ · ~v = ∇ ~ ) + σi j ∂vi ;
~ · (κ∇T
+ ∇ · (~v) + P∇ (7.80)
∂t ∂x j

7.2.4 Entropy
• we now introduce again the energy per unit mass, ε ≡ /ρ, to write

∂(ρε) ~ ∂ρ ~ ∂ε
" # !
+ ∇ · (ρε~v) = ε + ∇ · (ρ~v) + ρ ~
+ ~v · ∇ε
∂t ∂t ∂t

= ρ ; (7.81)
dt
this first implies the equation

ρ
dε ~ · ~v = ∇
+ P∇ ~ ) + σi j ∂vi ;
~ · (κ∇T (7.82)
dt ∂x j

• the term P∇~ · ~v can also be rewritten; because of the continuity


equation, we first obtain
dρ ∂ρ ~ = ∂ρ + ∇
~ · (ρ~v) − ρ∇
~ · ~v = −ρ∇
~ · ~v ,
= + ~v · ∇ρ (7.83)
dt ∂t ∂t
which allows us to write
dρ−1
!
~ · ~v = P − 1 dρ P dρ dV
P∇ =− = Pρ = Pρ , (7.84)
ρ dt ρ dt dt dt

where we have used the specific volume V ≡ ρ−1 ;


84 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

• thus, the left-hand side of the energy equation (7.82) can be cast
into the form
!
dε ~ dε dV ds
ρ + P∇ · ~v = ρ +P = ρT
dt dt dt dt
∂s
" #
= ρT ~
+ (~v · ∇)s , (7.85)
∂t
where s is again the specific entropy; therefore, we finally obtain
∂s ~ ) + σi j ∂vi ;
" #
ds ~ ~ · (κ∇T
ρT = ρT + (~v · ∇)s =∇ (7.86)
dt ∂t ∂x j
this describes how the entropy is changed due to heat conduction
and viscous dissipation; obviously, the entropy is conserved if
κ = 0 = σi j ;

7.3 Generalisations

7.3.1 Additional External Forces; Gravity


• the equations derievd so far can be generalised in obvious ways if
~ or a radiation pressure
external forces act such as gravity, −ρ∇Φ,
force f~rad ; such additional forces leave the continuity equation
unchanged; the force equation acquires the additional force density
terms
~ + f~rad ;
− ρ∇Φ (7.87)
• in the energy-conservation equation, two additional terms appear
which describe the work done by the external forces, thus
~ + ~v · f~rad ;
− ρ~v · ∇Φ (7.88)

• the stress-energy tensor of the gravitational field is


1 ∂Φ ∂Φ 1 i j ∂Φ ∂Φ
!
T =
ij
− δ (7.89)
4πG ∂xi ∂x j 2 ∂xk ∂xk
whose trace is
1 ∂Φ ∂Φ 3 ∂Φ ∂Φ 1 ∂Φ ∂Φ
!
T ii = − =− ; (7.90)
4πG ∂x ∂xi 2 ∂xk ∂x
i k 8πG ∂xi ∂xi
the spatial integral of the trace,
∂Φ ∂Φ
Z Z
1
d x Ti = −
3 i
d3 x i =
V 8πG V ∂x ∂xi
Z
1 h
~ · (Φ∇Φ)
~ − Φ∇
~ 2Φ
i
= − d3 x ∇
8πG V
Z
1
= d3 x Φρ , (7.91)
2 V
7.3. GENERALISATIONS 85

is the potential energy in the gravitational field; in the last equality


of (7.90), we have used the Poisson equation,
~ 2 Φ = 4πGρ ,
∇ (7.92)
and dropped the boundary term which results from the divergence
~ · (Φ∇Φ)
∇ ~ when we use Gauss’ theorem;

• then, the total stress-energy tensor of a self-gravitating fluid is


ij
T ges = T gas
ij
+ T grav
ij
= ρvi v j + Pδi j +
1 ∂Φ ∂Φ 1 i j ∂Φ ∂Φ
!
+ − δ (7.93)
4πG ∂xi ∂x j 2 ∂xk ∂xk
with the trace
~ 2
(∇Φ)
T ii ges = ρ~v2 + 3P − , (7.94)
8πG
whose volume integral is
Z Z !
1
d x T i ges =
3 i
d x ρ~v + 3P + Φρ
3 2
V V 2
Z
= 2T + U + 3 d3 x P , (7.95)
V

where T and U are the kinetic and potential energies, respectively;


• the tensor virial theorem (6.59) tells us
d2 I i j d2 Iii
Z Z
2
= 2 d xT
3 ij
⇒ 2
= 2 d3 x T ii ; (7.96)
dt V dt V

which means that the integral over the trace of T i j must vanish if
the system under consideration is static;

7.3.2 Example: Cloud in Pressure Equilibrium


• now, we briefly consider two astrophysical consequences of this
result; first, let a homogeneous, spherical cloud be given of mass
M and radius R which has the temperature T ; it be embedded into
the constant pressure P;
• its kinetic energy is
3M
kT , (7.97)
2m
and the potential energy is
GM 2 3
U = −α with α= (7.98)
R 10
for the homogeneous sphere; for other mass distributions, α will
be different but remains of order unity;
86 KAPITEL 7. IDEAL AND VISCOUS FLUIDS

• our earlier result now asserts

M GM 2
3kT −α + 3PV = 0 (7.99)
m R

for a static configuration; with V = 4πR3 /3, we find for the pressu-
re
1 αGM 2 3kT M
!
P= − ; (7.100)
4π R4 mR3
the external pressure must thus be reduced by the amount of the
gravitational force compared to the thermal pressure NkT/V of
the gas in the sphere;

• at the critical mass

3kT R4 3kT R
Mcr = · = , (7.101)
mR αG
3 mGα
the pressure P vanishes, such that the sphere is in equilibrium with
its self-gravity;

7.3.3 Example: Self-Gravitating Gas Sphere

• a further example concerns isolated systems in which the kinetic


energy of the gas is

3
T = (γ − 1)Uint , (7.102)
2
where Uint is the internal energy of the gas and γ the adiabatic
index; for such a static system, the tensor virial theorem requires

3(γ − 1)Uint + Ugrav = 0 , (7.103)

if we denote the gravitational potential energy by Ugrav

• the total energy is E = Uint + Ugrav , and thus (7.103) implies


 
3(γ − 1)E − 3(γ − 1) − 1 Ugrav
= 3(γ − 1)E − (3γ − 4)Ugrav = 0 (7.104)

and therefore
3γ − 4
E= Ugrav ; (7.105)
3(γ − 1)
since Ugrav < 0 and E < 0 for a bound system, we require

4
γ> ; (7.106)
3
7.3. GENERALISATIONS 87

• in order to see what happens in the limiting case γ → 4/3, we


write
1 d2 I
= 3(γ − 1)E − (3γ − 4)Ugrav ⇒ (7.107)
2 dt2
γ − 4/3 1 d2 I
E = Ugrav + ;
γ−1 6(γ − 1) dt2
for γ → 4/3, the first term on the right-hand side vanishes, and
because of E < 0, we must have

d2 I
<0, (7.108)
dt2
which typically implies a collapse;
88 KAPITEL 7. IDEAL AND VISCOUS FLUIDS
Kapitel 8

Flows of Ideal and Viscous


Fluids
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
8.1 Flows of Ideal Fluids Gas Dynamics”, chapters 6 and
14; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophy-
8.1.1 Vorticity and Kelvin’s Circulation Theorem sical Processes”, sections 8.6–
8.9; Landau, Lifshitz, “Theoreti-
• ideal fluids are such in which dissipative effects are unimportant cal Physics, Vol VI: Hydrodyna-
or absent, i.e. fluids for which we can put the viscosity coefficients mics”, chapters I, II and VIII
and the thermal conductivity to zero, ζ = η = κ = 0; in such fluids,
the entropy is conserved along flow lines,
ds ∂s ~ =0;
= + ~v · ∇s (8.1)
dt ∂t
• Euler’s equation for the force per unit mass is

∂~v ~
~ v = − ∇P − ∇Φ
+ (~v · ∇)~ ~ ; (8.2)
∂t ρ
here, we employ the identity
~ a · ~b) = (~a · ∇)
∇(~ ~ ~b + (~b · ∇)~
~ a + ~a × (∇
~ × ~b) + ~b × (∇
~ × ~a) (8.3)

and put ~a = ~v = ~b; then, the relation


~ v2 ) = 2(~v · ∇)~
∇(~ ~ v + 2~v × (∇
~ × ~v) (8.4)

follows, and thus

~ v = 1 ∇(~
(~v · ∇)~ ~ v2 ) − ~v × (∇
~ × ~v) , (8.5)
2
from which we obtain
∂~v ~
− ~v × (∇ ~ v2 ) − ∇P − ∇Φ
~ × ~v) = − 1 ∇(~ ~ ; (8.6)
∂t 2 ρ

89
90 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

• the curl of the velocity,

Ω ~ × ~v ,
~ ≡∇ (8.7)

is called the vorticity of the flow; if we take the curl of Euler’s


equation in its form (8.6), we find the equation for the vorticity

~ ~ 
 
∂Ω ~ × (~v × Ω)
~ −∇ ~ ×  ∇P 
= ∇
∂t  ρ 
~ ~
= ∇ ~ + ∇ρ × ∇P ;
~ × (~v × Ω) (8.8)
ρ2

• if the pressure P is a function only of ρ and not of other quantities


such as ~v, the gradients of P and ρ must align,

~ k ∇ρ
∇P ~ ⇒ ~ × ∇ρ
∇P ~ =0; (8.9)

for such barotropic fluids, the vorticity equation simplifies to

∂Ω ~ ~ ;
= ∇ × (~v × Ω) (8.10)
∂t

• we consider now the so-called circulation, which is the line integral


over the velocity along closed curves,
I
Γ ≡ ~v · d~l ; (8.11)
C

we are interested in the total change with time of the circulation


embedded into the flow, i.e. taking into consideration that the
contour C is deformed by the flow; we first use Stokes’ theorem
to write Z Z
~
Γ = (∇ × ~v) · dA =~ Ω~ · dA
~, (8.12)
A A

where A is the area enclosed by the contour C; dA~ is the directed


area element pointing along the local normal to the area A;

• the total time derivative of Γ now is

∂Ω ~ ~
~ · ∂A ,
Z Z

= · dA + Ω (8.13)
dt A ∂t A ∂t
and the change of the area due to the deformation of the contour is
I
~ · (~v × d~l) ,
Ω (8.14)
C

for (~vdt) × d~l gives the differential change of area per time interval
dt;
8.1. FLOWS OF IDEAL FLUIDS 91

• using (8.14) yields with (8.13)

∂Ω ~
Z Z
dΓ ~ × ~v) · d~l
= · dA + (Ω
dt A ∂t C
Z  ~ 
 ∂Ω ~ ~ ~=0,
=  ∂t + ∇ × (Ω × ~v) · dA

(8.15)
A

since the term in square brackets vanishes because of the vorticity


equation (8.10); we have further used that

Ω~ · (~v × d~l) = (Ω
~ × ~v) · d~l and
~ × (Ω
∇ ~ × ~v) = −∇ ~ × (~v × Ω)~ ; (8.16)

thus, the circulation Γ along contours comoving with the flow is


conserved in barotropic flows; this is Kelvin’s circulation theorem;

8.1.2 Bernoulli’s Constant


• if the flow is stationary, all partial derivatives with respect to time
vanish; in such cases, flow lines can be introduced which are the
integral curves of the velocity field; obviously,

dx dy dz
= dt = = (8.17)
vx vy vz

for the flow lines;

• in ideal fluids, the specific entropy is constant,

ds ∂s ~ =0 ~ =0
= + (~v · ∇)s thus (~v · ∇)s (8.18)
dt ∂t
for a stationary flow because ∂s/∂t = 0, i.e. the entropy remains
constant along flow lines; moreover, the enthalpy satisfes

dw = dε + Pd(ρ−1 ) + ρ−1 dP , (8.19)

because it is, per definition,

+P P
w= =ε+ , (8.20)
ρ ρ

and with
dε = T ds − PdV = T ds − Pd(ρ−1 ) (8.21)
we find
dw = T ds + ρ−1 dP = ρ−1 dP , (8.22)
since ds = 0 along flow lines;
92 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

• again for stationary flows, ∂~v/∂t = 0, Euler’s equation in its form


(8.6) implies
~
1~ 2 ~ = − ∇P − ∇Φ
∇(~v ) − ~v × Ω ~ ; (8.23)
2 ρ

let now ~l be a unit tangent vector to a flow line; if we multiply


(8.23) with ~l, the first term quantifies the change of ~v2 /2 along the
flow line; the second term vanishes, because it is perpendicular to
~v and thus also to ~l, and this implies that
~v2
B≡ +w+Φ (8.24)
2
is constant along flow lines; this is Bernoulli’s equation,
B = constant along flow lines ; (8.25)

• we have merely used that ds = 0, i.e. such flows can be called


adiabatic because no heat is exchanged between flow lines; if the
flow is furthermore isentropic, i.e. if s = const. holds everywhere
and thus all flow lines have the same value of s, then
s = s(P, ρ) = const. (8.26)
means that P must be a function of ρ alone, i.e. the flow is then
also barotropic, and the circulation Γ is conserved;

8.1.3 Hydrostatic Equlibrium


• in the hydrostatic case, ~v = 0, and Euler’s equation simplifies to
~ = −ρ∇Φ
∇P ~ , (8.27)
and Poisson’s equation
~ 2 Φ = 4πGρ
∇ (8.28)
relates the gravitational potential to the density; taking the diver-
gence of (8.27) yields
~ 
 
 ∇P
~ · 
∇  ρ  = −4πGρ ; (8.29)

only assumptions on mechanical, but not on thermodynamical


equilibrium entered here; the curl of (8.27) shows that
~ × (ρ∇Φ)
0=∇ ~ = ∇ρ
~ × ∇Φ
~ , (8.30)
which shows that the gradients of ρ and Φ are then parallel to each
other, i.e. ρ and Φ have the same iso-surfaces, i.e. equipotential
surfaces are then also iso-density contours;
8.1. FLOWS OF IDEAL FLUIDS 93

• in the spherically symmetric limiting case, we finally have


2
~ = dP ~er ,
∇P ~ · f~ = 1 d(r fr )
∇ with fr ≡ f~ · ~er , (8.31)
dr r2 dr
and thus
1 d r2 dP
!
= −4πGρ ; (8.32)
r2 dr ρ dr
if there is a barotropic relation, P = P(ρ), this is an ordinary,
second-order differential equation for the density ρ which can
directly be integrated;
• Euler’s equation in its form (8.32) is often applied to self-gravitating
systems such as galaxy clusters where the gravitational potential
is largely caused by the dark matter, which requires us to separate
the gas density ρgas from the dark-matter density ρDM ,

1 d r2 dP
!
= −4πGρDM ; (8.33)
r2 dr ρgas dr

• with the equation of state for an ideal gas,


ρgas
P = nkT = kT , (8.34)
m
where m is the (mean) mass of a gas particle, we find
Z r
r2 k d(ρgas kT )
= −4πG r02 dr0 ρDM = −GM(r) , (8.35)
ρgas m dr 0

where M(r) is the dark mass enclosed in a sphere of radius r; thus,


kr2
!
dρgas dT
M(r) = − T + ρgas =
mGρgas dr dr
r2 kT d ln ρgas d ln T
!
= − + =
mG dr dr
rkT d ln ρgas d ln T
!
= − + , (8.36)
mG d ln r d ln r
i.e. if the two logarithmic gradients can be estimated or determined
e.g. from X-ray observations, the dark mass can be found;

8.1.4 Curl-Free and Incompressible Flows


~ × ~v = 0 = Ω,
• if the velocity field is curl-free, ∇ ~ the velocity field
~ and
can be written as the gradient of a velocity potential, ~v = ∇ψ,
Euler’s equation (8.6) then reads
~
∂(∇ψ) 1~ 2 ~
∇P
+ ∇~
v =− ~ ,
− ∇Φ (8.37)
∂t 2 ρ
94 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

or, if we combine the gradients (cf. 8.22)

∂ψ ~v2
!
~
∇ +w+ +Φ =0; (8.38)
∂t 2

the quantity
∂ψ ~v2
B0 ≡ +w+ +Φ (8.39)
∂t 2
can then only be a function of time,

B0 = f (t) , (8.40)

which can be set to zero because it can be absorbed into the velocity
~ thus,
potential ψ without changing the relation ~v = ∇ψ;

∂ψ ~v2
B =
0
+w+ +Φ=0, (8.41)
∂t 2
which generalises Bernoulli’s equation for such cases in which the
~ instead of stationary, ∂~v/∂t = 0;
velocity field is curl-free, ~v = ∇ψ,

• finally, if ρ = const. in addition, like for incompressible fluids, its


~ = 0, and the vorticity equation (8.8) implies
gradient vanishes, ∇ρ

~ × ~v)
∂(∇ ~ × ~v × (∇
~ × ~v) ,
h i
=∇ (8.42)
∂t
i.e. the flow is then described solely by one equation for the velo-
~ · ~v = 0;
city field, because the continuity equation shrinks to ∇
~ × ~v = 0, i.e. if the flow is
• if the velocity field is also curl-free, ∇
incompressible and curl-free, ∇ ~ · ~v = 0 and ~v = ∇ψ
~ imply that the
velocity potential ψ has to satisfy the Laplace equation,

~ 2ψ = 0 ;
∇ (8.43)

8.2 Flows of Viscous Fluids

8.2.1 Vorticity; Incompressible Flows


• Euler’s equation for viscous fluids was

∂~v ~
~ v = − ∇P + η ∇
~ 2~v + 1 ζ + η ∇(
 
+ (~v · ∇)~ ~ ∇
~ · ~v) ; (8.44)
∂t ρ ρ ρ 3
taking the curl of this equation and using the vector identity

~ × (~v · ∇)~
~ v = −∇ ~ × ~v × (∇
~ × ~v) ,
h i h i
∇ (8.45)
8.2. FLOWS OF VISCOUS FLUIDS 95

we obtain the equation for the vorticity in viscous fluids,


∂Ω~
−∇ ~ = η∇
~ × (~v × Ω) ~ 2Ω
~ (8.46)
∂t ρ
~
∇ρ ~ 2~v − ζ + η ∇(
   
+ × ~ − η∇
∇P ~ ∇
~ · ~v) ;
ρ2 3
here, the first term on the right-hand side describes the diffusion
of vorticity caused by the viscosity η, and the second term arises
exclusively through the density gradient;
• thus, if the flow is incompressible, ρ = const., the second term on
the right-hand side of (8.47) drops out, and the equation is reduced
to
∂Ω~
−∇ ~ = η∇
~ × (~v × Ω) ~ 2Ω
~ ; (8.47)
∂t ρ
in incompressible fluids, the divergence of ~v must also vanish,
~ · ~v = 0, thus

~ × (~v × Ω)
∇ ~ = (Ω ~ v − (~v · ∇)
~ · ∇)~ ~ Ω~ , (8.48)
which allows us to write the vorticity equation as
~
∂Ω ~ v + (~v · ∇)
~ · ∇)~
− (Ω ~ Ω~ = η∇~ 2Ω
~ , (8.49)
∂t ρ
~
or, if we identify the total time derivative of Ω,
~
dΩ
− (Ω ~ v = η∇
~ · ∇)~ ~ 2Ω
~ ; (8.50)
dt ρ

~ · ~v = 0 for
• the divergence of Euler’s equation implies, with ∇
incompressible fluids,
~2
~ v = −∇ P ;
~ · (~v · ∇)~
h i
∇ (8.51)
ρ
together with ∇ ~ ·~v = 0, these equations (8.50) and (8.51) determine
the flow of incompressible, viscous fluids: ρ is constant, the curl
~ × ~v = Ω,
of ~v is the vorticity, ∇ ~ and the divergence of ~v vanishes;
this determines the velocity field, and the pressure P follows from
(8.51);

8.2.2 The Reynolds Number


• the only dimensional physical parameter in equation (8.50) for the
vorticity of a viscous, incompressible flow is
η
ν≡ , (8.52)
ρ
96 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

whose dimension is
erg s cm3 cm2
= , (8.53)
cm3 g s

thus squared length over time; a body characterised by a geometri-


cal dimension L moving with velocity u through a viscous fluid
thus introduces the length scale L and the time scale L/u; from
them and ν, we can form the following dimension-less quantity:

L2 1 uL uLρ
· ≡R= = , (8.54)
L/u ν ν η

which is the so-called Reynolds number;

• since no other dimensional physical quantities occur, the velocity


~v of the fluid can be scaled with the velocity u of the body and the
spatial coordinate ~x with the length scale L; in this way, dimension-
less quantities emerge whose change is described by quantities
which otherwise contain only the Reynolds number; bodies of
different size but otherwise equal shape embedded into flows which
are scaled as described here thus create self-similar flows as long as
the Reynolds number remains the same; conversely, the Reynolds
number classifies such self-similar solutions of the flow equations;
the transition to ideal fluids is characterised by R → ∞;

8.3 Sound Waves in Ideal Fluids

8.3.1 Linear Perturbations


• we now consider small perturbations of a fluid which is otherwi-
se flowing according to a background solution characterised by
density ρ0 and pressure P0 ; we further transform into a coordinate
frame in which the unperturbed fluid is locally at rest, thus ~v0 = 0;

• let the flow be ideal for now; the perturbations be small, and we
can consequently linearise Euler’s equation around the background
solution, i.e. we write ~v = 0+~v0 , P = P0 +P0 , ρ = ρ0 +ρ0 and neglect
terms of higher than first order in ~v0 , P0 and ρ0 ; this transforms the
continuity equation to

∂ρ ~ ∂(ρ0 + ρ0 ) ~
+ ∇ · (ρ~v) = 0 = + ∇ · (ρ0~v0 ) ; (8.55)
∂t ∂t
the background density ρ0 must also satisfy the continuity equation,
which reads
∂ρ0
=0 (8.56)
∂t
8.3. SOUND WAVES IN IDEAL FLUIDS 97

because we are in a frame locally co-moving with the fluid, ~v0 = 0,


and thus the linearised continuity equation reads

∂ρ0 ~
+ ∇ · (ρ0~v0 ) = 0 ; (8.57)
∂t

• the linearised Euler equation reads

~ 0
∂~v0 ∇P
+ =0 (8.58)
∂t ρ0

if terms of higher than first order in the perturbations are ignored


again;

• a relation between density and pressure is established as follows:


since the ideal fluid flows adiabatically, we can set

∂P
!
P =
0
ρ0 (8.59)
∂ρ s

for the pressure perturbations, where the subscript s denotes that


the derivative must be taken at constant entropy; if we further
neglect the density gradient of the background solution on the
~ 0 = 0, the continuity equation
length scale of the perturbation, ∇ρ
requires
∂P0 ∂P ~ 0
!
+ ρ0 ∇ · ~v = 0 ; (8.60)
∂t ∂ρ s

• this equation connects the curl-free part of ~v0 to the evolution of


~ we first have
the pressure; if we set ~v0 = ∇ψ,

∂P0 ∂P
!
+ ρ0 ~ 2ψ = 0 ,
∇ (8.61)
∂t ∂ρ s

and the linearised Euler equation further requires

∂ψ ∇P~ 0
~
∇ + =0, (8.62)
∂t ρ0

which means that, up to an irrelevant constant, the pressure pertur-


bation is
∂ψ
P0 = −ρ0 ; (8.63)
∂t
we thus find the wave equation

∂2 ψ ∂P ~ 2
!
− ∇ ψ=0; (8.64)
∂t2 ∂ρ s
98 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

8.3.2 Sound Speed


• this is the usual d’Alembert equation, whose solutions are arbi-
trary functions f (x) which propagate with a velocity cs in either
direction, f (x ± cs t), where
! #1/2
∂P
"
cs ≡ (8.65)
∂ρ s
is the sound speed; obviously,
∂2 f (x ± cs t) ∂[±cs f 0 (x ± cs t)]
= = c2s f 00 (x ± cs t) (8.66)
∂t2 ∂t
and
∂2 f (x ± cs t)
= f 00 (x ± cs t) , (8.67)
∂x 2

if the primes denote derivatives of f (x) with respect to its argument;


these solutions represent arbitrarily shaped waves propagating at
the sound speed cs into the positive or negative x direction;
• the condition imposed during the derivation of this wave equation
was that the perturbations are small; since
∂ψ
v0x = = f 0 (x ± cs t) (8.68)
∂x
as well as
∂ψ
P0 = −ρ0 = ∓ρ0 f 0 (x ± cs t) (8.69)
∂t
and
P0 = c2s ρ0 , (8.70)
we find
P0 cs ρ0
v0x = ∓ =∓ ; (8.71)
ρ0 cs ρ0
thus, this condition is satisfied as long as |v|  cs , i.e. for sub-sonic
flows;
• sound waves describe how small perturbations propagate through
~ they oscillate in propagation directi-
the fluid; because of ~v0 = ∇ψ,
on, i.e. they are longitudinal waves; however, we have to take into
account that entropy is conserved in an ideal fluid,
ds
=0 (8.72)
dt
along flow lines, which means that entropy perturbations can on-
ly propagate with the flow velocity ~v since they have to remain
constant along the flow lines; the same holds for the circulation of
barotropic fluids;
• in viscous fluids, sound waves are dissipated, i.e. they are conver-
ted to heat;
8.4. SUPERSONIC FLOWS 99

8.4 Supersonic Flows

8.4.1 Mach’s Cone; the Laval Nozzle


• perturbations propagate with the sound speed into all directions
relative to the fluid, i.e. relative to a coordinate frame which is
co-moving with the fluid; within a given time t, they reach all
points around their origin with radius cs t;

• if the fluid is flowing with velocity ~v, it depends on the modulus |~v|
where perturbations may propagate to; the situation for |~v| < cs is
illustrated in a diagram (to be inserted);

• sound waves can still propagate into all directions, but the situation
changes according to another diagram (to be inserted) if |~v| > cs ;
then, as seen from the laboratory frame, sound waves can only
reach points within a cone with the half opening angle
cs
α = arcsin ; (8.73)
v
this means that sound waves cannot reach an area in the direction
of the flow because they are passed by the flow; from the point
of view of a body which is at rest in the laboratory frame, this
implies that the flow is meeting with the body “blindly”, without
having been “informed” about its presence by sound waves; this
has far-reaching implications;

• as an example for the (steady) transition from sub- to supersonic


flow, we consider a nozzle with variable circular cross section A;
mass conservation requires

ρvA = const. (8.74)

or
d(ρv) dA
=− ; (8.75)
ρv A
from Euler’s equation, we obtain in the stationary case

d~v ~
~v · ∇P 1 dP
~v · =− =− , (8.76)
dt ρ ρ dt
for ∂P/∂t = 0 due to the assumed stationarity; in the rotationally-
symmetric and thus effectively one-dimensional case considered,
this implies
dv 1 dP
v = − ⇒
dt ρ dt
1 ∂P c2
!
dP
vdv = − =− dρ = − s dρ , (8.77)
ρ ρ ∂ρ s ρ
100 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

and thus
dρ vdv
=− 2 ; (8.78)
ρ cs
• with this result, we obtain

vdρ + ρdv v2 dv
!
dA dρ dv
=− =− − =− 1− 2 (8.79)
A ρv ρ v cs v
or
c2s
!
dA dρ
=− 1− 2 ; (8.80)
A ρ v
• we first consider a flow which enters with v < cs into a nozzle
which widens in the flow direction; in this case, dA/A > 0 and

c2s
!
1− 2 >0, (8.81)
v

thus dv/v < 0; therefore, the flow decelerates, but becomes denser,
dρ/ρ > 0;

• the reverse case happens if the nozzle narrows in the flow direction,
dA/A < 0; then, dv/v > 0 and the flow accelerates;

• exactly the opposite occurs if the flow is supersonic on entrance,


v > cs , since then
dA dv
>0 ⇒ >0 (8.82)
A v
and
dA dv
<0 ⇒ <0, (8.83)
A v
i.e. a supersonic flow accelerates in a widening nozzle; we now
consider a nozzle which is assembled as follows (graphic to be
inserted); if sub-sonic gas is entering the nozzle, v < cs , it ac-
celerates; if it remains subsonic up to the smallest cross section,
its velocity decreases again; however, if it reaches v = cs at the
narrowest cross section of the nozzle, the gas accelerates further
beyond the sound speed;

8.4.2 Spherical Accretion


• now we consider the example of spherical accretion, i.e. of a body
of mass M with radius R which is embedded in a gas cloud from
which it attracts mass; the continuity equation requires in this
spherically-symmetric case
1 ∂ 2
(r ρv) = 0 (8.84)
r2 ∂r
8.4. SUPERSONIC FLOWS 101

if the accretion flow is stationary and thus ∂ρ/∂t = 0; the quantity


r2 ρv is therefore spatially constant, and we set it equal to the
accretion rate,
4πr2 ρv ≡ − Ṁ , (8.85)
i.e. the mass accepted by the central body per unit time, which
is defined with a negative sign because v < 0; likewise for a
stationary, spherically-symmetric flow, Euler’s equation says

~ v=v dv 1 dP GM
(~v · ∇)~ =− − , (8.86)
dr ρ dr r

because the gravitational potential is

GM dΦ ~ = GM ;
Φ=− ⇒ = ~er · ∇Φ (8.87)
r dr r2

• we now further assume a polytropic equation-of-state, i.e. P ∝ ργ


with the adiabatic index 1 ≤ γ ≤ 5/3; moreover, the pressure
gradient can be written as

dP dP dρ dρ
= = c2s ; (8.88)
dr dρ dr dr

• we now use these three equations and rewrite them; first, we can
conclude from the continuity equation

1 d(r2 ρv) 1 d(r2 v)


" #
2 dρ
=0 = 2 ρ +r v
r2 dr r dr dr
ρ d(r v)
2

= 2 +v , (8.89)
r dr dr
thus
1 dρ 1 d(r2 v)
=− 2 ; (8.90)
ρ dr r v dr

• with this result, Euler’s equation can be written as

dv c2s dρ dv c2 d(r2 v) GM
v + = v − 2s + 2 =0, (8.91)
dr ρ dr dr r v dr r

where we can use


dv 1 dv2
v = (8.92)
dr 2 dr
to arrive at the more convenient equation

c2s dv2 2c2s r


! !
1 GM
1− 2 =− 2 1− ; (8.93)
2 v dr r GM
102 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

• far away from the central mass, the term


2c2s r
!
1− (8.94)
GM
is negative, thus the right-hand side of (8.93) is positive; since
d(v2 )/dr < 0 in order to have inflowing gas, the factor 1 − c2s /v2
must be negative, and hence the flow at large radii must be sub-
sonic, v < cs ;
• the gas flows towards the accreting object and will reach a point at
sufficiently small rc where
2c2s r
!
1− =0, (8.95)
GM
from which the critical radius rc can be read off to be
GM
rc = ; (8.96)
2c2s
inserting the sound speed from
γ
dP [.P0 (ρ/ρ0 ) ] γP0 γ−1
c2s = = = γ ρ
dρ dρ ρ0
γP γnkT γkT
= = = (8.97)
ρ ρ m
yields
GMm
rc = ; (8.98)
2γkT
• this critical radius is typically far beyond the radius of the central
object, i.e. the accretion flow passes into the supersonic regime
there;
• Euler’s equation can be integrated; first,
!γ #
ρ γP0
Z Z " Z
dr dP dr d dρ
= P0 = γ dr ργ−2
ρ dr ρ dr ρ0 ρ dr
γ−1
! 0 γ−1
γP0 d ρ γP0 ρ
Z
= γ dr = , (8.99)
ρ0 dr γ − 1 (γ − 1)ργ0
then, we can use the squared sound speed
γP0 γ−1
c2s = ρ (8.100)
ργ0
from (8.97) to find
c2
Z
dr dP
= s , (8.101)
ρ dr γ−1
where c2s is a function of radius r, of course;
8.4. SUPERSONIC FLOWS 103

• thus, Euler’s equation implies


v2 c2 GM
+ s − = const. ≡ C ; (8.102)
2 γ−1 r

• since v → 0 at r → ∞, the constant must equal


c2s (∞)
C= , (8.103)
γ−1
but on the other hand we must satisfy at the critical radius
GM
v = cs at r = rc = , (8.104)
2c2s
so that we can conclude from (8.102)
c2s (∞) c2s (rc ) c2s (rc )
= + − 2c2s (rc ) , (8.105)
γ−1 2 γ−1
or
c2 (∞)
!
1 1
c2s −2+ = s , (8.106)
γ−1 2 γ−1
and thus
" #1/2 s
1 2
cs (rc ) = cs (∞) = cs ; (8.107)
1 − 3/2(γ − 1) 5 − 3γ

• because of the proportionality c2s ∝ ργ−1 , the density follows from


" #2/(γ−1)
cs (rc )
ρ(rc ) = ρ(∞) ; (8.108)
cs (∞)
this also specifies the accretion rate, since
Ṁ = 4πr2 ρv = const. = 4πrc2 ρ(rc )cs (rc ) , (8.109)
and therefore
!1/(γ−1) !1/2
2 2
Ṁ = 4πrc2 ρ(∞) cs (∞)
5 − 3γ 5 − 3γ
!1/(γ−1)+1/2−2
ρ(∞) 2
= πG M 32 2
cs (∞) 5 − 3γ
!(5−3γ)/[2(γ−1)]
ρ(∞) 2
= πG M 32 2
; (8.110)
cs (∞) 5 − 3γ

• the solution for the velocity finally follows from mass conservati-
on,
−4πr2 ρv = Ṁ ⇒ (8.111)
" #2/(γ−1)
− Ṁ − Ṁ cs (∞)
v = = ;
4πr2 ρ(r) 4πr2 ρ(∞) cs (r)
104 KAPITEL 8. FLOWS OF IDEAL AND VISCOUS FLUIDS

• inserting this into equation (8.102) yields the solution for v(r),
and mass conservation then yields ρ(r); indeed, the maximum
accretion rate at given r is reached exactly when that radius is
critical, r = rc ; i.e. if the flow velocity reaches the sound velocity
there;
Kapitel 9

Shock Waves and the Sedov


Solution
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
9.1 Steepening of Sound Waves Gas Dynamics”, chapter 15 and
17; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophy-
9.1.1 Formation of a Discontinuity sical Processes”, sections 8.10–
8.12; Landau, Lifshitz, “Theore-
• as long as perturbations remain small in the sense that δρ 
tical Physics, Vol VI: Hydrodyna-
ρ, δP  P and δv  v, they propagate with the sound speed
mics”, chapter IX
and retain their initial (arbitrary) shape, as guaranteed by the
d’Alembert equation governing them;
• in a polytropic gas, the pressure is P ∝ ργ and thus the sound speed
cs ∝ ρ(γ−1)/2 increases if the density of the medium increases;
waves thus run faster in denser regions and can even pass less
dense parts of the waves;
• the density varies along the wave; in denser regions, the wave
propagates faster, which makes the wave front steepen; however,
since the density must be unique at any location, this behaviour
must lead to discontinuities, i.e. to sharp density jumps;
• we analyse this behaviour in the simple case of an isentropic flow,
which means s = const. and the same everywhere in the fluid; in
this case, the pressure is a function of density alone, P = P(ρ), and
likewise the velocity along the wave can be written as a function
of density alone, v = v(ρ); for a wave propagating into the positive
x direction, the continuity equation requires
∂ρ ∂(ρv) ∂ρ d(ρv) ∂ρ
+ =0= + ; (9.1)
∂t ∂x ∂t dρ ∂x
• similarly, Euler’s equation requires
∂v ∂v 1 ∂P ∂v 1 dP ∂v
!
+v + =0= + v+ ; (9.2)
∂t ∂x ρ ∂x ∂t ρ dv ∂x

105
106 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

• the total change in density is

∂ρ ∂ρ
dρ = dt + dx , (9.3)
∂t ∂x
i.e. the change in x at constant ρ is
!−1
∂x ∂ρ ∂ρ
! !
d(ρv) dv
=− = =v+ρ , (9.4)
∂t ρ ∂t ∂x dρ dρ

where (9.1) was used;

• the change of x at constant v follows similarly,

∂v ∂v
dv = 0 = dt + dx ⇒
∂t ∂x
!−1
∂x ∂v ∂v
! !
= − , (9.5)
∂t v ∂t ∂x

or, with the help of Euler’s equation (9.2),

∂x
!
1 dP
=v+ ; (9.6)
∂t v ρ dv

• however, since ρ is a function of v alone, the change of x with


constant ρ must equal the change of x with constant v, and therefore
the two partial derivatives (9.4) and (9.6) must equal,

dv 1 dP 1 dP dρ
v+ρ =v+ =v+ ; (9.7)
dρ ρ dv ρ dρ dv

• from this, we conclude


!2
dv c2s
= 2 , (9.8)
dρ ρ

hence the velocity can be expressed as


Z Z
cs dρ dP
v=± =± ; (9.9)
ρ ρcs

• for our polytropic equation of state,



ρ ρkT
P = P0 and P= , (9.10)
ρ0 m

and the squared sound speed is



∂P ργ−1 γ ρ γkT
c2s = = γP0 γ = P0 = , (9.11)
∂ρ ρ0 ρ ρ0 m
9.1. STEEPENING OF SOUND WAVES 107

thus c2s ∝ T , and according to (9.10) the temperature scales with


the density as

T ∝ ργ−1 or ρ ∝ T 1/(γ−1) , (9.12)

which can be used to express the density in terms of the sound


speed,
!2/(γ−1)
cs
ρ = ρ0 ; (9.13)
c0

• the velocity follows from (9.6), which can be combined with (9.8)
to find x,

∂x
!
1 dP 1 dP dρ
=v+ =v+ = v ± cs , (9.14)
∂t v ρ dv ρ dρ dv

which yields
x(v, t) = (v ± cs )t + f (v) , (9.15)
where f (v) is an arbitrary function of v to be specified by the
boundary conditions;

• with
 !2/(γ−1) 
ρ
!
dρ  cs
= d ln = d ln 

ρ ρ0

c0
2 cs 2 dcs
= d ln = , (9.16)
γ−1 c0 γ − 1 cs

the sound speed can be rewritten from (9.9)

γ−1
Z
cs dρ 2(cs − c0 )
v=± =± ⇒ cs = c0 ± v ; (9.17)
ρ γ−1 2

• putting this into (9.15) finally yields

γ+1
!
x = ±c0 + v t + f (v) ; (9.18)
2

9.1.2 Specific Example

• to give an example, we study an infinitely long pipe along the


x direction which is closed from the left side with a piston and
filled with gas from the right side; until t = 0, the piston be at
rest at x = 0, and it be accelerated into the pipe with a constant
acceleration a afterwards;
108 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

• the velocity and position of the piston are thus


a
vp = at , xp = t2 , (9.19)
2
and since the velocity of the gas has to equal the velocity of the
piston vp at the piston’s position xp , we find from (9.18) at the
position of the piston

γ+1
!
a2
t = c0 + at t + f (v) ; (9.20)
2 2

using t = v/a, we can solve for f (v),


 γv  v c0 v γv2
f (v) = − c0 + =− − , (9.21)
2 a a 2a
which gives us the general solution for the relation between x and
t to the right of the piston

γ+1 c0 v γv2
!
x = c0 + v t− − ; (9.22)
2 a 2a

• the velocity at position x and time t is thus determined by

γv2 c0 γ − 1
!
+ − t v − (c0 t − x) = 0 , (9.23)
2a a 2

which can be solved for v to yield


 s 
!2
1  γ + 1 γ+1
 ! 
v =  at − c0 + 2γa(c0 t − x) ;

at − c0 ±
γ 2 2
(9.24)
this is the velocity of the gas for all points x to the right of the
piston, i.e. for x ≥ at2 /2;

• for x = 0 and t = 0, the velocity must vanish, which selects from


the two branches of the solution (9.24) the one with the positive
sign,
 s 
!2
1  γ + 1 γ+1
 ! 
v =  at − c0 + 2γa(c0 t − x) ;

at − c0 ±
γ 2 2
(9.25)

• a discontinuity is formed where the velocity changes suddenly,


which is where
∂v ∂x
! !
=∞ ⇒ =0 (9.26)
∂x t ∂v t
9.1. STEEPENING OF SOUND WAVES 109

holds; physically, this means that parts of the wave with different
velocities meet at the same position; differentiating (9.22) with
respect to v, we see that this happens in our example when
γ+1 c0 γv
t− − =0 (9.27)
2 a a
at the time
2
tc = (c0 + γv) ; (9.28)
a(γ + 1)
there, the velocity drops to zero, v = 0, and thus
2c0
tc = ; (9.29)
a(γ + 1)
• specifically, setting γ = 5/3 and a = c0 /τ with a characteristic
acceleration time scale τ, the gas velocity is
 s
! !2 !
3c0  4t
 4t 10 t x 
v= −1 + −1 +  , (9.30)


3 τ c0 τ 

5 3τ 3τ
where x must obey
at2 c0 t2 c0 τ  t 2 x 1  t 2
x≥ = = ⇒ ≥ ; (9.31)
2 2τ 2 τ c0 τ 2 τ
• now, we define
x θ2 t v
− ≡ξ; ≡θ; ≡η, (9.32)
c0 τ 2 τ c0
which enables us to write (9.22) as
 s
!2 !
θ
! 2
3  4θ
 4θ 10
η =  −1 + −1 + θ − − ξ  ,
 
(9.33)
5 3 3 3 2
where ξ ≥ 0 is now the distance from the piston in units of c0 τ;
• the discontinuity occurs at the time
3τ 3
tc = or θc = ; (9.34)
4 4
since the velocity must vanish there,
v=0=η, (9.35)
the position of the discontinuity is determined by
θc2 15
θc − = ξc ⇒ ξc = ; (9.36)
2 32
• the sound speed is
γ−1  η
cs = c0 + v = c0 1 + (9.37)
2 3
and thus the density is
ρ  η 3
= 1+ ; (9.38)
ρ0 3
110 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

9.2 Shock Waves

9.2.1 The Shock Jump Conditions


• shock waves occur when a supersonic flow hits an obstacle, for
instance because it impinges on a solid, resting body, or because
a body moves at supersonic speeds through the fluid; in the im-
mediate vicinity of the body, the velocity must drop tp zero, and
because the flow is supersonic, no information on the obstacle can
propagate upstream against the flow;
• we approximate the discontinuity as a plane and consider a super-
sonic fluid flow perpendicularly hitting the plane from the left; let
ρ1 , P1 and v1 be the density, pressure and velocity of the fluid to
the left of the discontinuity, and ρ2 , P2 and v2 be to its right;
• deriving Euler’s equations, we had identified the following cur-
rents:
ρ~v2 mass current
~v

2
+ w ρ~v energy current (9.39)
ρv v + Pδi j
i j
momentum current
• when specialised to the situation considered here, in which the
discontinuity is perpendicular to the flow along the x axis, the
conservation of these currents requires the following conditions
ρ1 v1 = ρ2 v2
~v21 ~v22
! !
+ w1 ρ1~v1 = + w2 ρ2~v2
2 2
ρ1 v21 + P1 = ρ2 v22 + P2 ; (9.40)

• the enthalpy and the sound speed for a polytropic gas are
γ P γP
w= , c2s = ; (9.41)
γ−1ρ ρ
• writing the velocity left of the discontinuity as
v1 = M1 cs (9.42)
with the Mach number M1 > 1, the equations (9.40) can be re-
written in the following way
ρ2 (γ + 1)M21 v1
= =
ρ1 (γ + 1) + (γ − 1)(M1 − 1) v2
2

P2 (γ + 1) + 2γ(M21 − 1)
= , (9.43)
P1 γ+1
and for an ideal gas with P = ρkT/m, the temperature ratio is
T 2 P2 ρ1
= ; (9.44)
T 1 P1 ρ2
9.2. SHOCK WAVES 111

• by construction, the Mach number M1 > 1, which implies ρ2 > ρ1


because

(γ + 1) + (γ − 1)(M21 − 1) = 2 + (γ + 1)M21 − 2M21


= 2(1 − M21 ) + (γ + 1)M21
< (γ + 1)M21 , (9.45)

and therefore the density ratio from (9.43) is larger than unity;
correspondingly, v2 < v1 , P2 > P1 and T 2 > T 1 ;

• in the limiting case of a highly supersonic flow, M1 → ∞ and

ρ2 γ + 1
lim = ; (9.46)
M1 →∞ ρ1 γ − 1

for a gas with adiabatic index γ = 5/3, the maximum density ratio
is therefore
γ+1
=4, (9.47)
γ−1
which is called the maximum shock strength; in the same limit,

P2 T2
→∞, →∞; (9.48)
P1 T1

9.2.2 Propagation of a One-Dimensional Shock Front

• now we consider a fluid pipe with a piston, which remains at rest


at x = 0 until t = 0 and is then instantly accelerated to a velocity u
into the positive x direction;

• a discontinuity forms at t = 0 which propagates with a velocity vs


to the right; then, there exists a region ahead of the shock where the
density, pressure and temperature still have their original values
ρ1 , P1 and T 1 ; in the region between the shock and the piston, the
gas moves with the velocity u of the piston; the difference of the
velocities between the two regions is thus u;

• in order to use the jump conditions derived before, we need to


transform into a coordinate frame in which the shock is at rest; in
this (primed) frame, the gas velocities are obviously

v01 = −vs ; v02 = u − vs = u + v01 , (9.49)

and the velocity difference must remain the same, of course,

v02 − v01 = u (9.50)


112 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

• eliminating the Mach number M1 , the jump conditions (9.43) for


the density ρ and the pressure P can be combined to read
ρ2 P2 (γ + 1) + P1 (γ − 1)
= ; (9.51)
ρ1 P2 (γ − 1) + P1 (γ + 1)
the matter current is j = ρ1 v01 , thus
j2 = ρ21 v02
1 = ρ2 v2 ,
2 02
(9.52)
and therefore
1  2 02  1h i
j2 = ρ1 v1 + ρ22 v02
2 = j2 + (P2 − P1 )ρ1 + ρ1 ρ2 v02
2 , (9.53)
2 2
where we have used the jump condition for the momentum from
(9.40); we thus obtain

2 ρ1
" #
1 2
j =
2
j + (P2 − P1 )ρ1 + j (9.54)
2 ρ2
or
P1 − P2
j2 = ρ1 ρ2 ; (9.55)
ρ1 − ρ2
• the velocity difference can be written as follows:
ρ1 v02 − ρ1 v01 1 ρ1 0
!
v02 − v01 = u= = ρ2 v − ρ1 v01
ρ1 ρ1 ρ2 2
ρ1 − ρ2
= j ; (9.56)
ρ1 ρ2
with (9.55), this turns into
#1/2
(P1 − P2 )(ρ1 − ρ2 )
"
u= (9.57)
ρ1 ρ2

• according to (9.51), the density ratio is


ρ1 − ρ2 1 ρ1
!
= −1
ρ1 ρ2 ρ1 ρ2
1 P1 (γ + 1) + P2 (γ − 1)
" #
= −1
ρ1 P1 (γ − 1) + P2 (γ + 1)
2 P1 − P2
= , (9.58)
ρ1 P1 (γ − 1) + P2 (γ + 1)
which allows us to write the velocity difference as
s
2
u = cs |1 − π| , (9.59)
γ(γ − 1) + πγ(γ + 1)

where π ≡ P2 /P1 is the pressure ratio, and we have expressed the


ratio P1 /ρ1 in terms of the sound speed cs in the unshocked gas
ahead of the piston from (9.41), P1 /ρ1 = c2s /γ;
9.2. SHOCK WAVES 113

• this can be re-written as a quadratic equation for the pressure ratio


π,
γ(γ + 1)u2 γ(γ − 1)u2
" # " #
π −π 2+
2
+ 1− =0 (9.60)
2c2s 2c2s
which has the solutions
γ(γ + 1)u2
π = 1+ (9.61)
4c2s
s
!2
γ(γ + 1)u2 γ(γ − 1)u2
!
± 1+ − 1− ;
4c2s 2c2s
the pressure ratio needs to exceed unity, π ≥ 1, which excludes
the negative branch; the solution for the pressure ratio can thus be
simplified to
s
γ(γ + 1)u 2
γu (γ + 1)2 u2
π=1+ + 1 + ; (9.62)
4c2s cs 16c2s
note that, if the piston is at rest, u = 0 and π = 1, as expected;
• equations (9.56) and (9.58) together yield v01 ,
1−π
u = 2v01 , (9.63)
γ − 1 + π(γ + 1)
which can be simplified by means of (9.59) to read
cs p
v01 = − p γ − 1 + π(γ + 1) ; (9.64)

this is the velocity of the unshocked gas in the rest frame of the
shock front; obviously, the velocity of the shock front in the rest
frame of the unshocked gas is
vs = −v01 , (9.65)
hence (9.65) also yields the (negative) velocity of the shock front
in our laboratory system;
• the physical conditions in the unshocked gas, expressed by (ρ1 , P1 , T 1 )
and the velocity u of the piston, first yield π ≡ P2 /P1 from (9.62),
from which ρ2 and T 2 immediately follow using the shock jump
conditions; the velocity of the shock relative to the velocity of the
piston is given by (9.64); if u  cs , we find
γ(γ + 1)u2
π≈ , (9.66)
2c2s
and the shock velocity becomes
γ+1
vs ≈ u&u, (9.67)
2
i.e. for a gas with γ = 5/3, the shock moves ∼ 33% faster than the
piston;
114 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

9.2.3 The Width of a Shock

• one of our assumptions was that the energy is conserved across the
discontinuity of a shock front; this is not necessarily so because it
can be transported away due to radiative losses, friction, diffusion
of fast particles and similar processes;

• if the energy remains conserved, the shock front is called adiaba-


tic; otherwise, viscous and thermal effect need to be taken into
consideration; energy conservation then demands

~v2
!
const. = ρ~v + w − κ∇T ~ (9.68)
2
~ · ~v) + 2 η~v(∇
+ ζ~v(∇ ~ · ~v) − 2η(~v · ∇)~
~ v,
3

which can be specialised to our case of a one-dimensional shock


front,

v2 ∂T ∂v
! !
4
ρv +w −κ − η − ζ v = const. ; (9.69)
2 ∂x 3 ∂x

• we now assume that η dominates over ζ and put in coarse ap-


proximation ζ = 0; moreover, the viscous friction term must be
comparable to the kinetic energy if it is to play an important role;
thus
v2 4η ∂v
ρv ≈ v ; (9.70)
2 3 ∂x
if the velocity changes by ∆v across a distance ∆x, we estimate

∂v ∆v ∆v
ηv ≈ ηv ≈ ρvν , (9.71)
∂x ∆x ∆x

from which we can estimate ∆x,

v2 4η ∆v 8ν ∆v
ρ ≈ ρν ⇒ ∆x ≈ ; (9.72)
2 3 ∆x 3v v

across a strong shock, ∆v ≈ v, which yields the shock width


∆x ≈ ; (9.73)
3v

since the viscosity is ν ≈ λv, the shock with turns out to be ∆x ≈


8λ/3, i.e. it is a factor of order unity times the mean-free path
length;
9.3. THE SEDOV SOLUTION 115

9.3 The Sedov Solution

9.3.1 Dimensional Analysis

• one example for a shock wave is given by an explosion, i.e. by an


event in which in very short time energy is being released within a
very small volume; we consider such an event under the following
simplifying assumptions: (1) the shock is very strong, meaning
that the pressure of the surrounding medium can be neglected,
P1  P2 ; (2) the energy E is released instantaneously; and (3) the
energy of the surrounding material is negligible compared to E,
i.e. the explosion energy dominates that of the surroundings; and
finally (4) the gas be polytropic with an adiabatic index γ;

• under these conditions, our shock jump condition for the density is

ρ1 P1 (γ + 1) + P2 (γ − 1) γ − 1
= ≈ , (9.74)
ρ2 P1 (γ − 1) + P2 (γ + 1) γ + 1

which implies that ρ1 and ρ2 are completely determined by one


another; the behaviour of the shock must thus be entirely determi-
ned by the explosion energy E and the surrounding matter density
ρ1 ;

• if we now consider the shock at a time t when it has reached the


radius R(t), the only quantity with the dimension of a length, which
can be formed from E, t and ρ1 is
!1/5
Et2
; (9.75)
ρ1

which makes us set


!1/5
Et2
R(t) = R0 (9.76)
ρ1

with a dimension-less constant R0 which remains to be determined;

• the shock velocity is obviously


!1/5
dR E 2t2/5−1 2 R
vs = = R0 = ; (9.77)
dt ρ1 5 5t

• we now use the jump conditions which we had obtained for the
piston in the tube; first, the velocity of the “piston” is, according
to (9.67),
2vs
u= , (9.78)
γ+1
116 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

from which (9.66) yields for the pressure within (“behind”) the
shock
!−1
γ(γ + 1) u2 2γv2s γP1 2ρ1 v2s
P2 = P1 = P 1 = , (9.79)
2 c2s γ + 1 ρ1 γ+1
where the sound speed (9.41) in the surrounding, unshocked medi-
um was inserted;

• our earlier expression (9.74) for the density shows that the density
inside the shock remains constant, because ρ1 is constant; since
the shock velocity (9.77) drops with time like

vs ∝ t2/5−1 = t−3/5 , (9.80)

the pressure inside the shock drops like

P2 ∝ t−6/5 , (9.81)

and the velocity of the gas behind the shock is

u ∝ vs ∝ t−3/5 ; (9.82)

• we can interpret these relations as follows: a shock wave driven


by the release of the energy E, which propagates outward with the
time-dependent radius R(t), collects material with mass

M ≈ ρ1 R3 ; (9.83)

this material is accelerated from zero to a velocity ≈ R/t, such that


the kinetic energy
R5
ρ1 2 (9.84)
t
is put into the collected material; the energy of the material behind
(“within”) the shock is thus approximately

R5
ρ1 Ṙ2 R3 ≈ ρ1 ; (9.85)
t2
equating this to the energy E, we immediately find
!1/5
Et2
R= , (9.86)
ρ1
i.e. the scaling relation (9.76) simply expresses energy conservati-
on within the shock;

• we now know how the velocity, the radius, the pressure and the
density at the shock; they are completely determined by the release
of an amount of energy E into surrounding material with the
density ρ1 whose energy can be neglected;
9.3. THE SEDOV SOLUTION 117

9.3.2 Similarity Solution

• the external radius of such an explosion is given by (9.86), which


suggestes introducing
r
ξ≡ (9.87)
R
as a dimension-less radial variable; we will now use ξ to express
the radius in v(r, t), ρ(r, t) and P(r, t) and solve the hydrodynamic
equations to determine the properties of the gas everywhere within
the shock;

• the velocity at the shock is given by (9.76); imitating this behaviour,


we put
2r
v(ξ, t) = V(ξ) (9.88)
5t
with a dimension-less function V(ξ) which needs to be determined;
the gas velocity at the inner rim of the shock, given by (9.78),
requires that V(ξ) satisfy the boundary condition

2
V(1) = ; (9.89)
γ+1

• similarly, we use the ansatz

ρ = ρ1G(ξ) (9.90)

for the density and must, because of (9.74), satisfy the boundary
condition
γ+1
G(1) = (9.91)
γ−1
for the as yet unknown function G(ξ);

• we finally express the pressure by the sound speed, using (9.42)


together with (9.74), (9.77) and (9.79) to write
!2
γP2 2γρ1 2 2γ(γ − 1) 2 R
c2s = = v = , (9.92)
ρ2 (γ + 1)ρ2 s (γ + 1)2 5 t

which justifies the ansatz

4 r2
c2s = Z(ξ) , (9.93)
25 t2
where Z(ξ) must satisfy the boundary condition

2γ(γ − 1)
Z(1) = ; (9.94)
(γ + 1)2
118 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION

• now, we can use energy conservation again to relate Z and V;


the energy E within the shock must remain constant because no
energy can flow at supersonic velocities; since the solutions, ho-
wever, have to be self-similar because they are expressed by the
dimension-less radius ξ, energy must also be conserved within any
other sphere with ξ , 1, because ξ , 1 gives the shock position
at another than the considered instant of time; consequently, we
can set up the energy balance for any sphere and require that the
energy remain constant;
• per time interval dt, a sphere with radius r loses the energy
v2
!
4πr ρv
2
+ w dt , (9.95)
2
while it gains energy by growing by an amount
!
2 2r
4πr vs dt = 4πr
2
dt , (9.96)
5t
incorporating the additional amount of energy
v2
! !
2 2r
4πr dt  + ρ; (9.97)
5t 2
the energy balance then implies
v2 2r v2
! !
v +w = + , (9.98)
2 5t 2
and since the enthalpy is
+P γ P c2
w= = = s , (9.99)
ρ γ−1ρ γ−1
the thermal energy per unit mass is
 1 P c2s
ε= = = ; (9.100)
ρ γ − 1 ρ γ(γ − 1)

• this implies with (9.98)


v2 c2s 2r v2 c2s
! !
v + = + , (9.101)
2 γ−1 5t 2 γ(γ − 1)
where we now insert the ansätze for v and c2s , (9.88) and (9.93);
the result can be written as
V2 V2
!
Z Z
V + = + , (9.102)
γ−1 2 γ(γ − 1) 2
from which follows
γ(γ − 1)(1 − V)V 2
Z= ; (9.103)
2(γV − 1)
9.3. THE SEDOV SOLUTION 119

• the hydrodynamic equations which we now have to solve are of


course the continuity, Euler, and energy conservation equations;
when specialised to our radially symmetric problem, they read
∂ρ ∂(ρv) 2ρv
+ + = 0
∂t ∂r r
∂v ∂v 1 ∂P
+v = −
∂t ∂r ρ ∂r
∂ ∂
! !
P
+v ln γ = 0 , (9.104)
∂t ∂r ρ
where the conservation of entropy was used instead of the energy
conservation equation;
• inserting G and V into the continuity equation and using
∂ ∂ξ d 2ξ d
= =− (9.105)
∂t ∂t dξ 5t dξ
yields after some straightforward manipulation
ξG0
ξV 0 − (1 − V) + 3V = 0 (9.106)
G
• noticing that
P 1−γ c2s 1−γ 4r2 ρ1
1−γ
P
= ρ = ρ = ZG1−γ (9.107)
ργ ρ γ 25γt2
and dropping irrelevant constants, the logarithm in (9.104) can be
re-written as
r2
!
P
ln γ = ln 2 ZG 1−γ
, (9.108)
ρ t
and substituting this into the entropy-conservation equation (9.104)
yields
ξZ 0 ξG0 5 − 2V
+ (1 − γ) + =0; (9.109)
Z G 1−V
• eliminating ξG0 /G from the continuity equation (9.106), this latter
equation becomes
ξZ 0 1 − γ  5 − 2V
+ 3V + ξV 0 + =0, (9.110)
Z 1−V 1−V
which is supplemented by (9.103), which implies
ξZ 0 γ
!
2 1
= − − ξV 0 ; (9.111)
Z V 1 − V γV − 1
taken together, (9.110) and (9.111) yield the single equation for V,
ξV 0 γ(1 − 3γ)V 2 + (8γ − 1)V − 5
= ; (9.112)
V γ(γ + 1)V 2 − 2(γ + 1)V + 2
this ordinary first-order differential equation can directly be inte-
grated in closed form after separating variables, using the boundary
condition (9.89); having found V(ξ), Z follows from (9.103), and
G from (9.107);
120 KAPITEL 9. SHOCK WAVES AND THE SEDOV SOLUTION
Kapitel 10

Instabilities, Convection, Heat


Conduction, Turbulence
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
10.1 Rayleigh-Taylor Instability Gas Dynamics”, chapter 8; Pad-
manabhan, “Theoretical Astro-
• instabilities occur in many different forms in hydrodynamical physics, Vol. I: Astrophysical
systems; their analysis always proceeds according to the same Processes”, sections 8.13–8.15;
scheme: one starts from an equilibrium configuration, perturbs it Landau, Lifshitz, “Theoretical
slightly, i.e. in linear approximation, decomposes the perturbations Physics, Vol VI: Hydrodyna-
into plane waves, ∝ exp[i(ωt − ~k · ~x)] and derives the dispersion re- mics”, chapter III
lation ω(k); imaginary frequencies signal the onset of instabilities;
• we first study the situation in which two fluids with the densities
ρ1 and ρ2 are separated by a plane; we choose that plane to be the
x-y plane and assume that gravity is directed into the negative z
direction with the acceleration g, corresponding to the gravitational
potential Φ = gz; finally, the perturbation of the separating plane
be described by a function ζ(x), i.e. perturbations are assumed to
be independent of y, without loss of generality;
• since ~v is assumed to be small, we can neglect the curl term in
Euler’s equation (8.6) and assume that the velocity is the gradient
of a velocity potential ψ; then, we can use Bernoulli’s law in the
form (8.41), where we ignore the term ~v2 /2 because it is of second
order in ~v; thus,
∂ψ
+ w + gz = 0 ; (10.1)
∂t
• for an incompressible fluid, γ → ∞ and the enthalpy turns into
γ P P
w= → ; (10.2)
γ−1ρ ρ
multiplying (10.1) with ρ thus yields
∂ψ ∂ψ
ρ + P + ρgz = 0 ⇒ P = −ρgz − ρ ; (10.3)
∂t ∂t

121
122KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

• pressure equilibrium at the boundary then implies


∂ψ ∂ψ

ρ1 gζ + ρ1 = ρ2 gζ + ρ2 , (10.4)
∂t 1 ∂t 2

which constrains the boundary by

∂ψ ∂ψ
!
1
ζ= ρ2 − ρ1 ; (10.5)
g(ρ1 − ρ2 ) ∂t 2 ∂t 1
since the velocity must be continuous across the boundary, the
velocity components in z direction must also agree on both sides
of the boundary, thus
∂ψ ∂ψ

= at z = ζ ; (10.6)
∂z 2 ∂z 1

• finally, the velocity components in z direction can to first order be


identified with the displacement of the boundary,
∂ζ ∂ψ
= ; (10.7)
∂t ∂z
differentiating the equation (10.5) for the boundary with respect to
t, this yields

∂ψ ∂2 ψ ∂2 ψ

g(ρ1 − ρ2 ) = ρ2 2 − ρ1 2 (10.8)
∂z ∂t 2 ∂t 1

~ v = 0, implies that the velocity potential ψ


• the incompressibility, ∇·~
~ 2 ψ = 0, and since ψ is independent
satisfies Laplace’s equation, ∇
of y by construction, we can set

ψ = f (z) cos(ωt − kx) ; (10.9)

the Laplace equation then demands

d2 f
cos(ωt − kx) − f k2 cos(ωt − kx) = 0 , (10.10)
dz2
thus f satisfies an oscillator equation with the usual exponential
solutions,
d2 f
− k2 f = 0 ⇒ f ∝ e±kz ; (10.11)
dz2
• let now h1 and h2 be the heights of the layers, then the velocity
needs to vanish at both z = −h1 and z = h2 ; this specifies the
solution

ψ1 = A1 cosh [k(z + h1 )] cos(kx − ωt) ,


ψ2 = A2 cosh [k(z − h2 )] cos(kx − ωt) ; (10.12)
10.1. RAYLEIGH-TAYLOR INSTABILITY 123

if we insert these solutions into the equation (10.8) for the pressure
balance, we find
g(ρ1 − ρ2 )A1 sinh[k(z + h1 )] k cos(kx − ωt) =
− ρ1 A1 cosh[k(z + h1 )]ω2 cos(kx − ωt)
+ ρ2 A2 cosh[k(z − h2 )]ω2 cos(kx − ωt) ; (10.13)
from (10.13), we obtain the ratio A1 /A2 ,
A2 g(ρ1 − ρ2 ) k sinh[k(z + h1 )] + ρ1 ω2 cosh[k(z + h1 )]
= ,
A1 ω2 ρ2 cosh[k(z − h2 )]
(10.14)
and a similar expression follows if ψ1 and ψ2 are swapped; equa-
ting both yields the dispersion relation
kg(ρ1 − ρ2 )
ω2 = ; (10.15)
ρ1 coth kh1 + ρ2 coth kh2

• if ω2 > 0 as required for a stable situation, ρ1 > ρ2 is obviously


necessary, which means that the specifically heavier fluid must
lie below the specifically lighter fluid; in this case, perturbations
propagate as waves along the boundary between the fluids;
• if the density of the upper fluid is small compared to the lower, we
can approximate ρ2 = 0 and have
kgρ1
ω2 = = kg tanh kh1 , (10.16)
ρ1 coth kh1
which gives the frequency of waves on a fluid under the influence
of gravity whose surface is being perturbed; in the limiting case of
low depth, h1  k−1 , this simplifies to
ω2 ≈ k2 gh1 , (10.17)
while for large depth, h1  k−1 and
ω2 ≈ kg ; (10.18)
these are the limiting cases of waves on shallow or deep water;
• if ρ2 > 0 and kh1  1 as well as kh2  1, we find
ρ1 − ρ2
ω2 ≈ kg ; (10.19)
ρ1 + ρ2
this is the limiting case of very deep layers; in the limit of long
waves, kh1  1 and kh2  1, the dispersion relation becomes
g(ρ1 − ρ2 )h1 h2
ω2 ≈ k2 ; (10.20)
ρ1 h2 + ρ2 h1

• this linear stability analysis shows that ω becomes imaginary if


the specifically heavier fluid lies on top of the specifically lighter
one, ρ1 < ρ2 , which is intuitively obvious;
124KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

10.2 Kelvin-Helmholtz Instability


• we now consider a situation in which one fluid flows with a velocity
~v parallel to the surface of a fluid at rest, for instance like wind
over a lake;

• we choose the coordinate system such that the z axis is orthogonal


to the boundary surface and direction of motion (i.e. the direction
of ~v) points into the x direction;

• writing Euler’s equation in the form

d~v ~
∇P
=− (10.21)
dt ρ

and taking the divergence, we see that the pressure must satisfy
Laplace’s equation
~ 2P = 0
∇ (10.22)
~ · ~v = 0 and ρ = const.;
if the fluid is incompressible, ∇

• as before, the ansatz δP = f (z) exp[i(kx − ωt)] leads to the oscilla-


tor equation
d2 f (z)
2
− k2 f (z) = 0 , (10.23)
dz
with the solutions f (z) ∝ exp(±kz); the exponentially growing
solution ∝ exp(kz) is ruled out physically because it diverges at
large distances from the boundary surface, and thus the pressure
perturbation must behave as

δP2 ∝ ekz ei(kx−ωt) (10.24)

above the boundary surface;

• to first order in the velocity perturbation δvz along the z axis,


Euler’s equation reads

∂δvz ∂δvz k δP2


+v = , (10.25)
∂t ∂x ρ2

which, with δvz ∝ exp[i(kx − ωt)], implies

kδP2
− iωδvz + ikvδvz = , (10.26)
ρ2

and thus
kδP2
δvz = ; (10.27)
iρ2 (kv − ω)
10.2. KELVIN-HELMHOLTZ INSTABILITY 125

• let now again ζ(x, t) be the boundary surface between the two
fluids, then we must have
dζ ∂ζ ∂ζ
= δvz = +v (10.28)
dt ∂t ∂x
to linear order, and the ansatz ζ ∝ exp[i(kx − ωt)] implies
− iωζ + ikvζ = δvz = i(kv − ω)ζ , (10.29)
thus the pressure perturbation δP2 from (10.27) can be written as
ρ2 ζ
δP2 = − (kv − ω)2 ; (10.30)
k
• on the other side of the boundary surface, we have v = 0 and must
choose the solution f (z) ∝ exp(kz); inserted into Euler’s equation,
this yields
∂δvz k δP1 k δP1 ik δP1
=− ⇒ δvz = = ; (10.31)
∂t ρ1 iρ1 ω ρ1 ω
• the boundary surface then satisfies the equation
∂ζ
= −iωζ = δvz , (10.32)
∂t
or
ζρ1 ω2
δP1 = ; (10.33)
k
• there must be pressure balance at the boundary, δP1 = δP2 , there-
fore
ρ1 ω2 + ρ2 (kv − ω)2 = 0 = (ρ1 + ρ2 )ω2 − 2ρ2 kω + ρ2 k2 ω2 , (10.34)
and this equation has the solutions
q
2ρ2 kv ± 4ρ22 k2 v2 − 4(ρ1 + ρ2 )ρ2 k2 v2
ω =
2(ρ1 + ρ2 )
kv √
= ρ2 ± i ρ1 ρ2 ;

(10.35)
ρ1 + ρ2
• for ρ1 , 0 , ρ2 , ω always has an imaginary part, i.e. the pertur-
bation grows; this so-called Kelvin-Helmholtz instability can be
damped by buoyancy forces, if ρ1 > ρ2 ; of course, a gravitational
field must then be taken into consideration;
• the time scale for the perturbation to grow is obviously given by
ρ1 + ρ2
=(ω) −1 = √
 
; (10.36)
kv ρ1 ρ2
if ρ1  ρ2 ,
ρ2
r
ω ≈ ±ikv (10.37)
ρ1
follows, which holds for example for wind blowing over water;
126KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

10.3 Thermal Instability


• we assume that the system under consideration gains energy by
heating and loses energy by cooling; the net cooling rate, i.e. the
net energy loss per unit time, be

L(ρ, T ) ; (10.38)

if the system is in thermal equilibrium, we require L(ρ, T ) = 0,


which implicitly defines a relation between ρ and T ; for thermal
bremsstrahlung, for example,

L(ρ, T ) = C ρ T − (heating) ; (10.39)

• this cooling function L(ρ, T ) can adopt various forms, in particular


because cooling processes are often related to thermal occupati-
on numbers and atomic or molecular excitations; because of the
Boltzmann factor, sometimes small temperature changes can give
rise to large changes in occupation numbers, and the atomic or
molecular energy levels introduce discrete thresholds; the curve
L(ρ, T ) = 0 thus typically looks as shown in the figure (to be
inserted);
• at the same time, let the system be in pressure equilibrium with
its surroundings, i.e. the pressure P be externally regulated; for an
ideal gas, we have
P ∝ ρT , (10.40)
such that pressure equilibrium may be represented by a straight
inclined line; in this example, the curve P = const. intersects the
curve L = 0 in three points where both mechanical and thermal
equilibrium are possible;
• if the system moves along the curve P = const., it gets out of
thermal equilibrium; the idea behind that is that mechanical is
usually established much faster than thermal equilibrium such
that a system more likely stays in mechanical than in thermal
equilibrium if ρ and T are changed, and thus it remains on the
curve P = const.;
• if, at the point A, the density is increased, the system moves to-
wards the lower right along the dashed line and thus into the regime
where L > 0; here, the energy losses are larger than the energy
gain, the temperature decreases further, the density grows further,
and the system moves further away from thermal equilibrium, it is
thermally unstable;
• if the same happens in point B, the cooling function becomes
negative, L < 0, the system heats up and moves back into the
equilibrium point;
10.3. THERMAL INSTABILITY 127

• we now consider a simple model for this thermal instability; the


fundamental equations are

∂ρ ~
+ ∇ · (ρ~v) = 0
∂t
~ 2
∂~v ∇v ~
+ − ~v × (∇~ × ~v) = − ∇P
∂t 2 ρ
∂s
" #
T ~
+ (~v · ∇)s = −L(ρ, T ) , (10.41)
∂t

where the specific entropy s is


!
P
s = cv ln γ (10.42)
ρ

up to a constant which is irrelevant here; L is the net energy loss


per unit mass, and the equation of state is

ρkT
P= = νRT (10.43)
m

with the mol number ν and the gas constant R; the specific heat
capacities at constant volume and constant pressure are

R
cv = (10.44)
γ−1

and

γR
!
1
cp = cv + R = R +1 = = γcv (10.45)
γ−1 γ−1

• the equilibrium state has ρ = ρ0 , T = T 0 , L(ρ0 , T 0 ) = 0 and ~v = 0;


we perturb this state by small deviations δρ, δT , δ~v and linearise
in these perturbations; this first yields

∂δρ ~ ∂δ~v ~
∇δP
+ ∇ · (ρ0 δ~v) = 0 , =− (10.46)
∂t ∂t ρ0

and, by substituting into the partial time derivative of the first


equation the divergence of ρ0 times the second equation,

∂2 δρ ~ ∂δ~v ∂2 δρ ~ 2
!
+ ∇ · ρ0 = − ∇ δP = 0 (10.47)
∂t2 ∂t ∂t2

we first allow perturbations with δP , 0 and ask later for the


conditions for instability under constant pressure;
128KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

• up to the irrelevant constant, the entropy is


!
P
s = cv ln γ = cv (ln P − γ ln ρ)
ρ
= cv ln(P0 + δP) − γ ln(ρ0 + δρ)
 

δP δρ
( " !# " !#)
= cv ln P0 1 + − γ ln ρ0 1 +
P0 ρ0
δP δρ
" #
= cv ln P0 − γ ln ρ0 + −γ
P0 ρ0
δP δρ δP δρ
!
= s0 + cv −γ = s0 + cv − cp ; (10.48)
P0 ρ0 P0 ρ0

• this allows us to write the left-hand side of the entropy equation


(10.41) as
∂ δP δρ
! !
(T 0 + δT ) ~
+ δ~v · ∇ s0 + cv − cp (10.49)
∂t P0 ρ0
∂ δP δρ ∂s0
! !
= T0 ~
+ δ~v · ∇ s0 + cv − cp + δT ;
∂t P0 ρ0 ∂t

• equilibriums requires that


∂ ∂s0
!
T0 ~ s0 = 0
+ δ~v · ∇ ⇒ T0 =0 (10.50)
∂t ∂t
because of ~v0 = 0; this reduces the left-hand side of (10.41) to
∂ δP δρ
!
T0 cv − cp ; (10.51)
∂t P0 ρ0

• on the right-hand side, we have


∂L ∂L
− L(ρ0 + δρ, T 0 + δT ) = −L(ρ0 , T 0 ) − δρ − δT ; (10.52)
| {z } ∂ρ ∂T
=0

since ρ and T are furthermore connected through the equation-of-


state, we can write
∂L ∂L
! !
dL = dρ + dT
∂ρ T ∂T ρ
∂L ∂ρ ∂ρ ∂L
! " ! ! # !
= dP + dT + dT
∂ρ T ∂P T ∂T P ∂T ρ
∂L ∂L
! !
= dT P + dT ρ ; (10.53)
∂T P ∂T ρ

at constant pressure or constant density ρ,


dρ dP
dT P = −T , dT ρ = −T , (10.54)
ρ P
10.3. THERMAL INSTABILITY 129

and thus
cv ∂δP cv ∂δρ ∂L δρ ∂L δP
! !
− = − ; (10.55)
P0 ∂t ρ0 ∂t ∂T P ρ0 ∂T ρ P0

• if we apply the Laplacian to this equation and use (10.47) for


~ 2 δP, we find

∂ ∂ δρ P0 ~ 2
" 2 #
cv 2 − cp ∇ δρ =
∂t ∂t ρ0
∂L P0 ~ 2 ∂L ∂2 δρ
! !
∇ δρ − ; (10.56)
∂T ρ ρ0 ∂T P ∂t2
dividing by cv , recalling that
cp P0
=γ, γ = c2s , (10.57)
cv ρ0
and introducing the abbreviations
1 ∂L 1 ∂L
! !
Np ≡ , Nv ≡ , (10.58)
cp ∂T P cv ∂T ρ
we find the relation
∂ ∂2 δρ ~ 2 δρ − Nv ∂ δρ ;
2
!
2~ 2
− cs ∇ δρ = N p c2s ∇ (10.59)
∂t ∂t ∂t
• again, we expand the perturbations into plane waves, δρ ∝ exp[i(kx−
ωt)], and thus transform (10.59) to
∂ h 2 2  i  
cs k − ω2 δρ = Nv ω2 − N p c2s k2 δρ (10.60)
∂t
which yields
(c2s k2 − ω2 )iω = Nv ω2 − N p c2s k2 ; (10.61)

• this cubic dispersion relation is difficult to solve; in the limiting


case of small wave lengths, c2s k2  ω2 , we can approximate
iω ≈ −N p ⇒ ω ≈ iN p , (10.62)
which indicates an unstable solution if N p < 0, since then
δρ ∝ e−N p t (10.63)
grows exponentially; thermal instability thus sets in when
∂L
!
<0; (10.64)
∂T P
• in the opposite limiting case of very large wave length,
iω ≈ −Nv , (10.65)
i.e. instability then requires that
∂L
!
<0; (10.66)
∂T ρ
130KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

10.4 Heat Conduction and Convection

10.4.1 Heat conduction


• a system can be in mechanical equilibrium, but out of thermal
equilibrium; the most straightforward example is a star which
is being kept in mechanical equilibrium by the balance between
gravity and the pressure gradient, but nonetheless continuously
radiates energy; the entropy equation reads in this case

~ ) + σi j ∂v ;
i
ds ~
ρT = ∇ · (κ∇T (10.67)
dt ∂x j

• if the velocity gradient is too small to drive currents, the second


term on the right-hand side can be neglected; from
!
dq
dq = T ds and cp = (10.68)
dT P

follows
cp dT = T ds ⇒ ds = cp d ln T ; (10.69)
with that, the entropy equation (10.67) can be reduced to

dT ~ 2T dT ~ 2T κ
ρcp = κ∇ or = χ∇ with χ ≡ ; (10.70)
dt dt ρcp

• in analogy to radiative energy transport, we define a conductive


opacity κcond through the conductive energy current
c ~ ~ ,
F~cond = − ∇(aT 4 ) ≡ −κ∇T (10.71)
3ρκcond

from which we obtain the relation between heat conductivity κ and


conductive opacity κcond ,

4caT 3
κ= ; (10.72)
3ρκcond

• if both radiative and conductive energy transport is present, the


effective opacity is

1 1 1 κrad κcond
= + ⇒ κeff = ; (10.73)
κeff κrad κcond κrad + κcond

• when we discussed the heat conductivity, we saw that it can be


written as
nv
κ= cv λ , (10.74)
3
10.4. HEAT CONDUCTION AND CONVECTION 131

where λ is the mead free path of the fluid particles; if we consider


electrons whose mean free path is determined by ions,

1
λ= , (10.75)
ni σ
where ni and σ are the number density and the scattering cross
section of the ions;

• typically, an electron will approach an ion up to a distance ri where


the kinetic and potential energies equal,

mv2 Ze2 2Ze2


≈ ⇒ ri ≈ ; (10.76)
2 ri mv2
the cross section is then

σ ≈ πri2 , (10.77)

and we obtain the expression

ne ve m2e v4e m2e


! !
1 ne
κ= cv = cv v5e (10.78)
3 4πni Z 2 e4 3 4πZ 2 e4 ni

for the heat conductivity of electrons scattered by ions;

• in a thermal electron gas, cv = 3k/2 and

3kT e
ve = , (10.79)
me

which yields the heat conductivity


!5/2
m2e
! !
k ne 3kT e
κ= (10.80)
2 4πZ 2 e4 ni me

for classical (non-degenerate) electrons; if we identify the Thom-


son cross section σT , we can alternatively write
√ ! !5/2
3 3kc ne kT e
κ= 2 (10.81)
Z σT ni me

with the obvious unit


erg
[κ] = ; (10.82)
cm s K
numerically,
! !5/2
ne kT e
κ ≈ 5.5 × 10 Z
12 −2
; (10.83)
ni 1 keV
132KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

10.4.2 Convection
• if the temperature gradient is too large, convection sets in: then,
warm, rising bubbles cannot any more cool and return to their
original locations, but continue to rise; we consider the situation,
in which a volume V(P, s) characterised by the pressure P and the
specific entropy s rises against the gravitational force;
• we ignore again thermal compared to mechanical adaptation pro-
cesses because they are typically slower and assume that the bubble
with volume V(P, s) rises by an amount ∆z, where its volume is
V(P0 , s); there, its buoyancy force is determined by the volume
V(P0 , s0 ) which the bubble would adopt if it had the specific entro-
py of its new environment; the situation is stable if the actual
volume V(P0 , s) is smaller than the volume V(P0 , s0 ), because then
gravity will dominate the buoyancy force, and the bubble will then
sink down again; we thus have the condition
V(P0 , s0 ) > V(P0 , s) (10.84)
for stability;
• with
ds
s = s + ∆z
0
(10.85)
dz z
and because of
∂V T ∂V
! !
cp dT = T ds ⇒ = >0, (10.86)
∂s P cp ∂T P

we find
∂V ∂V
! !
ds
V(P , s ) = V(P , s) +
0 0 0
∆s = V(P , s) +
0
∆z ;
∂s P ∂s P dz z
(10.87)
thus, the stability condition is satisfied if

ds
>0 (10.88)
dz z

• in an ideal gas with adiabatic index γ, the entropy is


s = cv ln PT γ/(γ−1)
h i
(10.89)
up to an irrelevant constant; thus
ds ∂s dP ∂s dT γ d ln T
" #
d ln P
= + = cv + ; (10.90)
dz ∂P dz ∂T dz dz γ − 1 dz
the condition (10.88) then shows that the temperature gradient
must satisfy
d ln T γ − 1 d ln P
− < (10.91)
d ln z γ d ln z
10.5. TURBULENCE 133

for the gas stratification to be stable against convection; the quan-


tity
γ−1
≡ ∇ad (10.92)
γ
is often called the adiabatic temperature gradient (“nabla adiaba-
tic”); using this, we stability condition is written
d ln T
− ≡ ∇ < ∇ad ; (10.93)
d ln P

• when convection sets in, it is a very efficient means of transporting


heat; viscosity hinders the convective energy transport;

10.5 Turbulence
• hydrodynamical flows with large Reynolds numbers turn out to
be highly unstable; for high viscosity (low Reynolds number),
stable solutions of the Navier-Stokes equation exist which develop
instabilities above a critical Reynolds number
uL
R= & Rcr ; (10.94)
ν

• a full analysis of such instabilities is very difficult and in general


an unsoved problem; turbulence sets in, in the course of which
energy is being transported from large to small scales until it is
dissipated by the production of viscous heat on sufficiently small
scales;

• let λ be the size of an eddy and vλ the typical rotational velocity


across the eddy; let further λ be the energy per unit mass in such
an eddy; then, the energy flow through an eddy of that size is

v2λ v3
!−1
λ
!
˙ ≈ ≈ λ (10.95)
2 vλ λ
|{z} |{z}
typical energy time scale

• energy is being fed into the turbulent cascade on the macroscopic


scale L where the typical velocity is u; from there, the energy cas-
cades through the turbulent eddies to progressively smaller scales
until it is finally viscously dissipated on a scale λv ; in between,
i.e. on scales λ satisfying

λs < λ < L , (10.96)

the energy flow ˙ must be independent of scale because the energy


cannot be accumulated anywhere; therefore, we conclude from
134KAPITEL 10. INSTABILITIES, CONVECTION, HEAT CONDUCTION, TUR

(10.95) that the typical eddy velocity must change with the eddy
scale λ as
vλ ∝ λ1/3 (10.97)
or  λ 1/3
vλ ≈ u ; (10.98)
L
the largest eddies thus carry the highest velocities, but the smallest
have the highest vorticity,
vλ u
Ω≈ ≈ 2 1/3 ; (10.99)
λ (λ L)
• to estimate the scale λv , we compare the viscous dissipation with
the specific energy flow ˙ ; the viscous heating rate is approximately
!2/3
 v 2
λ v3λ
hv ≈ ησ ≈ η
2
≈η λ−4/3 = η˙ 2/3 λ−4/3 ; (10.100)
λ λ
therefore, hv is negligibly small on large scales, but if the heating
rate becomes of order the energy flow rate,
hv ≈ ρ˙ , (10.101)
viscous dissipation sets in; this happens on a length scale λv given
by
!3/4
η
η˙ λv ≈ ρ˙ ⇒ λv =
2/3 −4/3
(10.102)
ρ˙ 1/3
or, because of
u3
˙ ≈ , (10.103)
L
the viscous scale can be transformed to
!3/4
ηL1/3  ν 3/4 L
λv = =L = 3/4 , (10.104)
ρu uL R
where R is the Reynolds number on the scale L;
• finally, we consider the correlation function of the eddy velocity
vλ , or rather its Fourier transform, the power spectrum; since
vλ ≈ (˙ λ)1/3 , (10.105)
the correlation function scales as
ξv ∝ (˙ λ)2/3 , (10.106)
while its Fourier transform Pv will then scale as
 2/3
Pv ∝ λ3 ξv ∝ k−3 ˙ k−1 ∝ ˙ 2/3 k−11/3 ; (10.107)
the “power” per logarithmic k interval will thus scale as
k2 Pv ∝ ˙ 2/3 k−5/3 , (10.108)
which is the Kolmogorov turbulence spectrum;
Kapitel 11

Collision-Less Plasmas
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
11.1 Basic Concepts Gas Dynamics”, chapters 28 and
29; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophysi-
11.1.1 Shielding; the Debye length cal Processes”, sections 9.1–9.4;
Ishimaru, “Basic Principles of
• we would like to describe plasmas as fluids; this requires that Plasma Physics”, chapters 1–4
collisions be random and short-ranged, such that equilibrium can
be locally established sufficiently fast; the fundamental difference
between plasma physics and the hydrodynamics of neutral fluids
is the long-ranged Coulomb interaction between the particles;
• a plasma consists of electrons and ions of charge Ze; the existence
of two different types of charge allows shielding on a characteristic
length scale; to calculate it, we put into a plasma with charge
density ±en± a point charge q; Poisson’s equation for the electro-
static potential then reads
~ 2 Φ = 4πe(n+ − n− ) + 4πqδD (~x)
∇ (11.1)
if the charge is placed at the coordinate origin;
• in thermal equilibrium, the charge densities are
!
∓eΦ
n± ∝ exp , (11.2)
kT
so that we can write
" ! !#
eΦ eΦ
n+ − n− = n̄ exp − exp − , (11.3)
kT kT
with the mean particle number density n̄;
• sufficiently far from the central charge, the argument of the expo-
nential

1 (11.4)
kT

135
136 KAPITEL 11. COLLISION-LESS PLASMAS

so that we can approximate


!
2eΦ
n+ − n− ≈ n̄ (11.5)
kT

and write Poisson’s equation as


2
~ 2 Φ ≈ 8πe n̄ Φ + 4πqδD (~x) ,
∇ (11.6)
kT
which has the solution
 √ 
q 2r 
Φ = exp −  ,

(11.7)
r λD

which contains the Debye length defined by


!1/2  T 1/2  n̄ −1/2
kT
λD ≡ = 6.9 cm ; (11.8)
4πe2 n̄ K cm−3

this is the characteristic length scale for the shielding of a charge


in a plasma;

• a plasma can be considered ideal if it contains sufficiently many


particles within the Debye length, since then the interaction energy
is small compared to the thermal energy; to see that, we compare

e2
with kT ; (11.9)

the first expression is the mean potential energy of the interacting
charges, the second is a measure for their thermal energy; r̄ is the
mean separation of the particles, given by
!1/3
4π 3 3
r̄ n ≈ 1 ⇒ r̄ ≈ ; (11.10)
3 4πn

we thus obtain from the comparison between the two energies


!1/3 !2/3
e2 4πn 4πne2 3 1 r̄2
= = 1 (11.11)
kT 3 3kT 4πn 3 λ2D

for λD  r̄;

11.1.2 The plasma frequency


• the mean thermal velocity of the electrons is given by
kT
hv2 i ≈ , (11.12)
me
11.2. THE DIELECTRIC TENSOR 137

which means that an electron passes the Debye length in the time
r
λD
r
kT me me
tD ≈ p ≈ 2
= ; (11.13)
hv2 i 4πne kT 4πne2

this is the time sale on which the thermal motion of the electrons
can compensate charge displacements by shielding;
• the time tD can be transformed into a characteristic frequency for
plasma oscillations,
s
1 4πne2  n 1/2
ωp ≈ = ≈ 5.6 × 104 Hz , (11.14)
tD me cm−3

which is the frequency with which charge inhomogeneities can


oscillate against each other; with λD and ωp , we now have two
essential parameters for describing plasmas at hand;

11.2 The Dielectric Tensor

11.2.1 Polarisation and dielectric displacement


• external electric fields E~ polarise media, i.e. they induce in these
media a charge displacement which can be described by a polarised
~ is defined by
charge density ρpol ; the polarisation P
~ ·P
∇ ~ = −ρpol ; (11.15)

the Maxwell equation in vacuum,


~ · E~ = 4πρ ,
∇ (11.16)

changes to
~ · E~ = 4π(ρ + ρpol ) = 4πρ − 4π∇
∇ ~ ·P
~; (11.17)
~ ≡ E~ +4πP,
the dielectric displacement, D ~ is defined as an auxiliary
field which satisfies
~ ·D
∇ ~ = 4πρ ; (11.18)

• due to charge conservation, the charge density must satisfy


∂ρpol
~ · ~jpol = 0 ;
+∇ (11.19)
∂t
if we substitute (11.15) here, we find
~
 ∂P ~
∂P
 
~ + ~jpol  = 0 or ~jpol =

∇ · − ; (11.20)
∂t ∂t
138 KAPITEL 11. COLLISION-LESS PLASMAS

• the Maxwell equation

∂ ~ 4π   1  ∂E~ ∂ ~  4π
 
~ ×B
~= 1 E ~j + ~jpol =  P
∇ + + 4π  + ~j (11.21)
c ∂t c c ∂t ∂t c

can then be written as


~
∇ ~ = 1 ∂D + 4π ~j
~ ×B (11.22)
c ∂t c

where ~j is the external (unpolarised) current;

11.2.2 Structure of the dielectric tensor


• we now need a relation between the external field E~ and the pola-
~ for sufficiently weak fields, we assume this relation to
risation P;
be linear and write for the Fourier-transformed quantities

D̂i (ω, ~k) = ˆ ij (ω, ~k) Ê j (ω, ~k) , (11.23)

defining the dielectric tensor ˆ ij ;

• due to the Fourier convolution theorem, the multiplication (11.23)


in Fourier space corresponds to a convolution in real space; this
~ x, t) can also be influenced by fields at earlier times
means that D(~
~ x − δ~x, t − δt);
and other places, E(~

• since the fields must remain real in real space, the dielectric tensor
must satisfy the symmetry relation

ˆ ij (−ω, −~k) = ˆ ∗ij (ω, ~k) (11.24)

in Fourier space;

• the principal-axis directions of the tensor ˆ ij can only depend on


the vector ~k, so that we can start from the ansatz

ki k j
ˆ ij (ω, ~k) = Âδij + B̂ (11.25)
k2

with functions  and B̂ depending on (ω, ~k); obviously,

ki k j j
!
k = ki (11.26)
k2

is parallel to ~k, while

ki k j j
!
δij − 2 k = ki − ki = 0 (11.27)
k
11.3. DISPERSION RELATIONS 139

and thus perpendicular to ~k; therefore, we split the tensor ˆ ij into


two parts,
ki k j ki k j
!
ˆ j = δ j − 2 ˆt + 2 ˆl ,
i i
(11.28)
k k
which we call the transversal and the longitudinal components of
ˆ ij , with  = ˆt and B̂ = ˆl − ˆt according to the ansatz (11.25);

• of course, ˆt and ˆl are generally functions of ω and k which also
need to satisfy the symmetry condition (11.24),

ˆl,t (−ω, k) = ˆl,t∗ (ω, k) ; (11.29)

• we have neglected in this derivation of the form of the dielectric


tensor ˆ ij that preferred macroscopic directions may exist in the
plasma, e.g. due to magnetic fields ordered on large scales; if they
exist, they must also be built into the dielectric tensor;

• in the limit of long waves, |~k| → 0, the dielectric tensor tends to

ˆ ij → δij ˆt , (11.30)

and thus turns into the “normal” scalar dielectricity;

11.3 Dispersion Relations

11.3.1 General form of the dispersion relations

• the dielectric tensor determines which kinds of wave can propagate


through the plasma; we shall now derive the dispersion relations
between the wave vectors ~k and the frequencies ω of such possible
waves;

• if the fields are decomposed into Fourier modes ∝ exp[i(~k · ~x − ωt)],


Maxwell’s equations in the plasma read

~ˆ ~ˆ
~ˆ = − ωD ,
~k × E~ˆ = ω B , ~k × B
c c
ˆ
~k · D ˆ
~ = 0 , ~k · B
~=0, (11.31)

if we neglect external charge densities and currents;

• combining the curl of the first equation (11.31) with the second
yields
~ˆ = ω ~k × B
~ˆ = − ω D
2
~k × (~k × E) ~ˆ , (11.32)
c c2
140 KAPITEL 11. COLLISION-LESS PLASMAS

and if we expand the double vector product, we can write the


displacement vector as
ω2 ~ˆ ˆ ~ˆ ;
2
D = k2 E~ − ~k(~k · E) (11.33)
c
we now use the dielectric tensor to write D̂i = ˆ ij Ê j and find,
written in components
ω2 i j
2
ˆ j Ê = k j k j Ê i − ki k j Ê j , (11.34)
c
or
ki k j ω2 i j
!
δij − 2 − 2 2 ˆ j Ê = 0 ; (11.35)
k ck
• the expression in brackets can be represented by a matrix; equa-
ˆ
tion (11.35) has non-trivial solutions E~ , 0 if and only if the
determinant of that matrix vanishes,
ki k j ω2 i
!
det δ j − 2 − 2 2 ˆ j = 0
i
(11.36)
k ck

11.3.2 Transversal and longitudinal waves


• this condition defines a relation between ~k and ω which must be
satisfied for waves allowed to propagate through the plasma; we
now insert ˆ ij from (11.28) and find

ki k j ω2 ki k j ki k j
( " ! #)
det δ j − 2 − 2 2 δ j − 2 ˆt + 2 ˆl = 0 , (11.37)
i i
k ck k k
or
ki k j ω2 ω2 ki k j
" ! ! #
det δij − 2 1 − ˆt 2 2 − 2 4 ˆl = 0 ; (11.38)
k ck c k

• for transversal waves, ki Ê i = 0, and the corresponding terms


disappear from the matrix equation (11.35); what remains is
ω2
!
1 − ˆt 2 2 Ê i = 0 , (11.39)
ck
which implies the dispersion relation
c2 k2
ω2 = ; (11.40)
ˆt
ˆ
if, conversely, E~ is parallel to ~k, terms proportional to
ki k j
!
δj − 2
i
(11.41)
k
11.4. LONGITUDINAL WAVES 141

do not contribute, and the dispersion relation is reduced to

ω2 ki k j Ê j
ˆl = 0 , (11.42)
c2 k4
which generally demands that ˆl = 0; in order to understand this
condition, we need to determine the form of ˆl ;

11.4 Longitudinal Waves

11.4.1 The longitudinal dielectricity


• in order to determine the form of ˆl , we return to the collision-less
Boltzmann equation; we neglect the motion of the ions because of
their lower velocities and write

f = f0 + δ f , (11.43)

i.e. we expect that sufficiently weak fields E~ and B~ will change the
phase-space distribution function only little away from a homoge-
neous and isotropic distribution function f0 ; to first approximation,
Boltzmann’s equation then reads
∂δ f ~
v ~ · ∂ f0 = 0 ;
!
~ f − e E~ + × B
+ ~v · ∇δ (11.44)
∂t c ∂~p
for an isotropic distribution f0 , we must further have
∂ f0
k ~v (11.45)
∂~p
because no other direction is defined, thus

~ · ∂ f0 = 0 ,
(~v × B) (11.46)
∂~p
and Boltzmann’s equation in linear approximation shrinks to
∂δ f ~ f = eE~ · ∂ f0 ;
+ ~v · ∇δ (11.47)
∂t ∂~p

• decomposing E~ and the perturbation δ f into plane waves ∝ exp[i(~k·


~x − ωt)], (11.47) yields
∂ f0
− iωδ f + i~v · ~kδ f = eE~ , (11.48)
∂~p
which can be solved for δ f ,

eE~ ∂ f0
δf = · ; (11.49)
i(~k · ~v − ω) ∂~p
142 KAPITEL 11. COLLISION-LESS PLASMAS

• charge and current densities are exclusively caused by δ f because


f0 is a homogeneous, isotropic and stationary equilibrium distribu-
tion; therefore,
Z Z
3 ~
ρ = −e d p δ f , j = −e d3 p~vδ f ; (11.50)

these quantities are then also proportional to the phase factor


exp[i(~k · ~x −ωt)], and the polarisation equations (11.15) and (11.20)
can be written
~ˆ = −ρ̂ , −iωP
i~k · P ~ˆ = ~ˆj ; (11.51)
with δ f from (11.49), we find

ˆ ∂ f0
Z
~ ˆ
~ 1
ik · P = e2
d3 p E~ · ; (11.52)
∂~p i(~k · ~v − ω)

• this integral has a pole at ω = ~k · ~v and is therefore ill-defined; we


remedy that by writing
1 1
→ (11.53)
i(~k · ~v − ω) i(~k · ~v − ω − iδ)
which has no poles any more on the real axis, and later sending δ
to zero;

• for longitudinal waves, we have 4πP ~ˆ = D~ˆ − E~ˆ = (ˆl − 1)E~ˆ and E~ˆ =
Ê~k/k; inserting this into (11.52) allows us to write the longitudinal
part ˆl of the dielectricity as

4πe2 ∂ f0
Z
1
ˆl = 1 − 2 d3 p ~k · ; (11.54)
k ∂~p (~k · ~v − ω − iδ)

• if we now place the coordinate system such that ~k points into the
positive ~x direction, the integral can be split up; we then have

~k · ∂ f0 = k d f0 (11.55)
∂~p dp x
and
∂ f0
Z
1
d3 p~k ·
∂~p i(~k · ~v − ω − iδ)
d f¯(p x )
Z
1
= k dp x , (11.56)
dp x i(kv x − ω − iδ)
with the definition
Z
f¯(p x ) ≡ dpy dpz f0 (~p) ; (11.57)
11.4. LONGITUDINAL WAVES 143

11.4.2 Landau Damping


• obviously, the longitudinal dielectricity ˆl has a real and an imagi-
nary part; the latter implies dissipation of electrical energy, as we
shall shortly see; to begin with, the dissipation follows from
˙
∂  E~ 2  ~ ~ E~ · E~ ~ ~˙
 
Q =   + E · j = +E·P
∂t 8π 4π
E~  ~˙ ˙ ~˙
 E~ · D
= ~
E + 4πP = , (11.58)
4π 4π

where the missing hats indicate that E~ and D


~ are to be taken as
functions of ~x and t here;

• we consider the contribution of individual Fourier modes (ω, ~k) to


the dissipation Q, i.e. we set
ˆ ~ ~ˆ ei(~k·~x−ωt) ,
E~ = E~ ei(k·~x−ωt) , ~ =D
D (11.59)

and thus Q will be the dissipation per Fourier mode (ω, ~k);

• in order to arrive at a real quantity, we need to replace E~ by


(E~ + E~ ∗ )/2 to obtain the real part of the E~ field which is defined
complex here, and the same for D; ~

~ˆ = ˆl E,
• since, for longitudinal waves, D ~ˆ we can write

~˙ = − iω ˆl E~ − ˆl∗ E~ ∗ ,
 
D (11.60)
2
where the minus sign on the second term comes from the change
in sign in the phase factor exp[i(~k · ~x − ωt)] due to the complex
conjugation of the E~ field; inserting this into Q from (11.58) gives
iω  ~ ~ ∗   ~
E + E ˆl E − ˆl∗ E~ ∗ ;

Q=− (11.61)
16π

averaging this over time eliminates the products E~ · E~ and E~ ∗ · E~ ∗


because they vary with the phase factor exp(−2iωt), while the
mixed terms become independent of time; thus,
iω  ~ˆ ~ˆ ∗ ˆ ˆ
 iω  ~ˆ 2
Q=− ˆl E · E − ˆl∗ E~ ∗ · E~ = − ˆl − ˆl∗ |E| ; (11.62)
16π 16π
the remaining expression in brackets is

ˆl − ˆl∗ = 2i=ˆl , (11.63)

and so we find
ω ~ˆ 2 ;
Q= =ˆl |E| (11.64)

144 KAPITEL 11. COLLISION-LESS PLASMAS

• the imaginary part of ˆl follows from (11.54),

1 ~k · ~v − ω + iδ δ
= == = ,
~k · ~v − ω − iδ (~k · ~v − ω)2 + δ2 (~k · ~v − ω)2 + δ2
(11.65)
and in the limit δ → 0 this turns into a Dirac delta function,
δ
→ πδD (~k · ~v − ω) ; (11.66)
(~k · ~v − ω)2 + δ2
therefore, the imaginary part of ˆl is

4π2 e2 d f¯
Z
=ˆl = − 2 dp x k δD (kv x − ω)
k dp x

4π2 e2 m d f¯
= − (11.67)
k2 dp x p =ωm/k

x

• the damping then turns into

ˆ~ 2 πme ω d f¯
2

Q = −|E| ; (11.68)
k2 dp x px =ωm/k

this is Landau damping, which is caused by the fact that electrons


which are slightly faster than the wave are slowed down, electrons
which are slightly slower than the wave are accelerated, and since
the velocity distribution is typically monotonically decreasing,
more electrons need to be accelerated than decelerated, and thus
the wave loses energy;

11.5 Waves in a Thermal Plasma

11.5.1 Longitudinal and transversal dielectricities


• in a thermal plasma, f0 is the Maxwell distribution

p2
!

f0 = exp − dp x dpy dpz , (11.69)
(2πmkT )3/2 2mkT

therefore the integrated distribution f¯ is

n̄ e−px /(2mkT )
2
! 2

¯f (p x ) = n̄(2πmkT ) exp − p x = , (11.70)


(2πmkT )3/2 2mkT (2πmkT )1/2
and its derivative is

−n̄p x e−px /(2mkT )


2
d f¯
= √ ; (11.71)
dp x 2π(mkT )3/2
11.5. WAVES IN A THERMAL PLASMA 145

• with these expressions, the longitudinal dielectricity can be trans-


formed to read
e−px /(2mkT )
2
4πe2 ∞
Z
n̄p x
ˆl = 1 + dp x √ ; (11.72)
k −∞ 2π(mkT )3/2 kpmx − ω − iδ
introducing the Debye length λD (11.8) and using v x = p x /m, we
find
e−mvx /(2kT )
Z ∞ 2
1 mv x
ˆl = 1 + 2 dv x √ , (11.73)
kλD −∞ 2πmkT kv x − ω − iδ
and with the mean thermal electron velocity
r
kT vx
ve = and z ≡ , (11.74)
m ve
we get the expression
Z ∞ 2
1 ze−z 1
ˆl = 1+ √ dz ≡ 1+ 2 2 W(x) , (11.75)
2π(kλD )2 −∞ z − x − iy k λD
where the abbreviations
ω δ
x≡ and y ≡ (11.76)
kve kve
were used and the function
z e−z /2
Z ∞ 2
1
W(x) ≡ √ dz (11.77)
2π −∞ z−x
was defined for x ∈ C;

• this function W(x) is well-defined and analytic for x in the upper


half of the complex plane, i.e. for =x > 0, and it can analytically
be continued into the lower half-plane by integrating over a closed
(rectifiable and positively oriented Jordan) curve C in C whose
interiour encloses x;

• we thus choose the contour such that it runs along the real axis
from −R to R, possibly including x in a small extension if =x ≤ 0,
and closing along a half-circle in the upper complex plane from
R to −R; because of the steep exponential drop of the integrand,
the integral along that half-circle will not contribute in the limit
R → ∞, and the integral along the closed contour C will equal the
integral along the real axis,

z e−z /2 z e−z /2
I 2 Z ∞ 2

dz → dz , (11.78)
C z−x −∞ z−x
with =z = 0, as required for evaluating (11.75);
146 KAPITEL 11. COLLISION-LESS PLASMAS

• beginning with =x > 0, we can integrate over z along the real axis;
for doing so, we first substitute the identity
Z ∞
1
=i dt exp [−i(z − x)t] (11.79)
z−x 0

into (11.77), which yields


Z ∞ Z ∞
i
dz ze−z /2+i(z−x)t
2
W(x) = √ dt
2π Z0 −∞
∞ Z ∞
i
dz ze−z /2+izt ; (11.80)
2
= √ dt e−ixt
2π 0 −∞

carrying out the z integration first, and splitting the remaining


exponential exp(−ixt) into trigonometric functions, yields
Z ∞
dt t (cos xt − i sin xt) e−t /2
2
W(x) =
0
π −x2 /2
Z x r
−x2 /2 y2 /2
= 1 − xe dy e +i xe ; (11.81)
0 2

• series expansions of W(x) are useful for practical calculations; for


small x, |x| < 1,

π −x2 /2 π
r r
x4 x4
W(x) ≈ 1 − x + + i
2
xe ≈ 1 − x2 + + i x (11.82)
3 2 3 2

while, for large x,

π −x2 /2
r
1 3
W(x) ≈ − 2 − 4 + i xe ; (11.83)
x x 2

• substituting this back into (11.75) and expanding the abbreviations


(11.76), we find the longitudinal dielectricity

(kve )2 3(kve )4
"
1
ˆl = 1− +
(kλD )2 ω2 ω4

πω −ω2 /(2k2 v2e )
#
− i√ e
2kve
 ω 2 3(kve )2
!
p
≈ 1− 1+
ω ω2

πωω2p −ω2 /(2k2 v2 )
+ i√ e e ; (11.84)
2(kve ) 3

in the limiting case ω  kve ; we have inserted the plasma frequen-


cy ωp = ve /λD in the first term here;
11.5. WAVES IN A THERMAL PLASMA 147

• in the opposite limiting case ω  kve ,



ω2 πω ω2
" !#
1
ˆl = 1 + 1− +i√ 1− 2 2
(kλD )2 (kve )2 2kve 2k ve

ωp 
!2  !2
ω πω 

≈ 1+ 1 − +i√  ; (11.85)
kve kve 2kve

• a similar calculation leads to the transversal dielectricity

ω2p ω
" ! #
ˆt = 1 + 2 W −1 ; (11.86)
ω kve

• if ions need to be taken into account in addition to the electrons,


the dielectricities are summed according to
X
ˆ − 1 = (ˆi − 1) ; (11.87)
i=species

11.5.2 Dispersion Measure and Damping


• the dispersion relation for transversal waves was given by (11.40);
in the high-frequency limit,

ω2p
ˆt ≈ 1 − (11.88)
ω2
because then x  1 and F(x) ≈ −1, and thus the dispersion
relation becomes

ω2 − ω2p = k2 c2 ⇒ ω2 = ω2p + k2 c2 ; (11.89)

• the group velocity of such transversal waves is


q
ω 2 − ω2 c
s
∂ω kc 2 p ω2p
cg = = = =c 1− 2 ; (11.90)
∂k ω ω ω
the propagation time of such waves is this

ω2p  L 2πe2
Z Z   Z
dl
∆tω = dlc 1 +  = + dl n , (11.91)

≈ 
cg 2ω2 c mcω2

where the quantity Z


dl n ≡ DM (11.92)

is called the dispersion measure (which is the column density of


the free electrons); here, we have approximated ω  ωp , which
should not be confused with the assumption ω  kve !
148 KAPITEL 11. COLLISION-LESS PLASMAS

• for ω < ωp , the dispersion relation implies


q
i ω2p − ω2
k= , (11.93)
c
i.e. the transversal waves are damped; ω  kve and ω < ωp =
ve /λD are possible if both ve and λD are small, which is the case
in sufficiently cold plasmas; in the ionosphere of the Earth, n ≈
106 cm−3 , thus
λD ≈ 0.11 cm (11.94)
and
ve ≈ 6.4 × 106 cm s−1 , (11.95)
thus
ωp ≈ 60 MHz , (11.96)
which corresponds to a wavelength of
c
λ≈ ≈ 5m , (11.97)
ωp

where all numbers are given assuming T = 273 K;

• the dispersion relation for longitudinal waves requires ˆl = 0, as


was shown above [see (11.42)]; to lowest order in kve /ω,

ω2p
ˆl ≈ 1 − ⇒ ω = ωp ; (11.98)
ω2
to next higher order, setting the real part of ˆl to zero,

ω2p
3(kve )2 !
" #
<ˆl ≈ 1 − 2 1 + =0, (11.99)
ω ω2

implies
ω2p 3ω4p k2 λ2D
+ −1=0; (11.100)
ω2 ω4
this quadradatic equation in ω2 has the solutions

ω2p  q 
ω =
2
1± 1+ 12k2 λ2D (11.101)
2
or  
ω ≈ ωp 1 + 3k2 λ2D ; (11.102)
Kapitel 12

Magneto-Hydrodynamics
further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
12.1 The Magneto-Hydrodynamic Equations Gas Dynamics”, chapter 21 and
27; Padmanabhan, “Theoretical
Astrophysics, Vol. I: Astrophysi-
12.1.1 Assumptions
cal Processes”, section 9.6
• magneto-hydrodynamics is built upon several assumptions which
go significantly beyond hydrodynamics; this begins the fact that
plasmas consist of ions and electrons which should be described as
two fluids which are coupled to each other; this is simplified by the
assumption that ions and electrons may be coupled to each other
so tightly by the electrostatic attraction that they can be treated as
a single fluid;

• secondly, the usual assumption of hydrodynamics plays a critical


role that the mean free path of electrons and ions is very small
compared to any other relevant scales;

• thirdly, there is usually a small drift velocity ~vdrift between the


electrons and ions,
vdrift = ~ve − ~vi , (12.1)
which causes an electric current, which in turn induces a magnetic
field;

• finally, it is assumed that the plasma flows non-relativistically,


such that terms of higher than linear order in v/c can be neglected;

• we thus get to the following mathematical assumptions: in the rest


frame of the plasma, the charge density ρ0 be negligible compared
to the current density ~j0 ,

|~j0 |
ρ0  ; (12.2)
c

149
150 KAPITEL 12. MAGNETO-HYDRODYNAMICS

• the current density be related to the electric field E~ 0 through Ohm’s


law,
~j0 = σE~ 0 , (12.3)
with the conductivity σ, which is assumed to be very high (“ideal”),
such that weak fields can be responsible for significant currents;

• because of the non-relativistic velocities, we neglect the displace-


ment current,
1 ∂E~ 0 4π~j0
 ; (12.4)
c ∂t c
the equations of the magnetic field thus read

~ ·B
∇ ~0 = 0 , ∇ ~ 0 = 4π ~j0 ;
~ ×B (12.5)
c

• the rest frame of the plasma and the observer’s laboratory frame
are related by the Lorentz transformation, which can be written, to
lowest order in v/c, as

~j · ~v
ρ0 = ρ − , ~j0 = ~j − ρ~v ; (12.6)
c2

since we have assimed ρ0  ~j0 /c, we also have ρ  ~j/c, and


because
~ · E~ 0 = 4πρ0 , ∇
∇ ~ 0 = 4π ~j0 ,
~ ×B (12.7)
c
~  | B|;
this also implies |E| ~

• thus, the assumptions of magneto-hydrodynamics lead to the con-


ditions

|~j| ~ ~
~ , ∂E  j , |~v|  1 ;

ρ ~  | B|
, |E| (12.8)
c ∂t c c

from here, we can now derive the equations of magneto-hydrodynamics;

12.1.2 The induction equation


• from Maxwell’s equation

~
1 ∂B ~ × E~
= −∇ (12.9)
c ∂t
and Ohm’s law

~
v ~ 
 
~j ≈ j = σE = σ E~ +
~0 ~0
 × B  , (12.10)
c
12.1. THE MAGNETO-HYDRODYNAMIC EQUATIONS 151

we conclude
~j ~v × B
~
E~ = − (12.11)
σ c
and
~
∂B  ~j ~v × B
~ 
 
~
= −c∇ ×  −  , (12.12)
∂t σ c
but we also need to satisfy

~j = c ∇~ ×B
~, (12.13)

and thus
~
∂B c2 ~ ~
!
~
= −∇ × ∇×B +∇ ~ × (~v × B)
~
∂t 4πσ
2 h
~ − c ∇(
~ × (~v × B) ~ ∇~ · B)
~ −∇ ~ 2B
~
i
= ∇
4πσ
~ ~ c2 ~ 2 ~
= ∇ × (~v × B) + ∇ B, (12.14)
4πσ

because of ∇~ ·B ~ = 0; this induction equation determines the


evolution of the magnetic field embedded into a plasma flow with
the velocity ~v; moreover, we have assumed that σ is spatially
~ = 0;
constant, ∇σ

12.1.3 Euler’s equation


• we now need equations for the back reaction of the magnetic field
on the flow of the plasma; first, there is the continuity equation for
the mass density ρ, which is of course unalteredly valid,
∂ρ ~
+ ∇ · (ρ~v) = 0 ; (12.15)
∂t
Euler’s equation which describes the conservation of momentum
or, more precisely, the transport of the specific momentum density,
must be modified because of the Lorentz equation; the Lorentz
force on a charge e is
e ~ ;
(~v × B) (12.16)
c
multiplying that with the number density of charges, the product
e~v turns into the current density ~j, which is

~j = c ∇~ ×B
~, (12.17)

and thus Euler’s equation is modified to read
d~v ~ + 1 (∇
~ × B)
~ ×B
~;
ρ = −∇P (12.18)
dt 4π
152 KAPITEL 12. MAGNETO-HYDRODYNAMICS

the Lorentz force density on the right-hand side can also be re-
written,

∂Bm
!
1 ~ ~ ~ = 1
(∇ × B) × B  
i jl
Bk
4π 4π jk m ∂xl
1  il  ∂Bm
= − δ δkm − δim δlk Bk l
4π ! ∂x
1 ∂B i
∂Bk
= Bk k − Bk
4π ∂x ∂xi
1 ~ ~ ~ 1 ~ ~2
= ( B · ∇) B − ∇ B ; (12.19)
4π 8π

the first term specifies how B~ changes along B,


~ i.e. it quantifies
the tension of the magnetic field lines, which obviously tend to
be as straight as possible; the second term is the gradient of the
magnetic energy density and augments the pressure gradient in
Euler’s equation;

• we had seen in normal, viscous hydrodynamics that Euler’s equa-


tion can be written in the form

∂(ρvi ) ∂T i j
+ =0, (12.20)
∂t ∂x j

with the stress-energy tensor

T i j = ρvi v j + Pδi j − σi j (12.21)

and the shear tensor

∂vi ∂v j 2 i j ~
!
σ =η
ij
+ ~ · ~v ;
− δ ∇ · ~v − ζδi j ∇ (12.22)
∂x j ∂xi 3

in presence of the magnetic field, this changes to

~ 2 i j 
 
1  i j B
ij
T →T −  B B − δ  ,
ij
(12.23)
4π 2

i.e. the stress-energy tensor of the viscous fluid is augmented by


the stress-energy tensor of the magnetic field;

• together with an equation of state P = P(ρ), the induction equation,


the continuity equation and Euler’s equation determine the flow;
these are eight equations for the eight unknowns ρ, P, ~v and B; ~
if the magnetic field is known, the current follows from equation
(12.17), and the electric field from (12.11);
12.1. THE MAGNETO-HYDRODYNAMIC EQUATIONS 153

12.1.4 Energy and entropy


• the entropy equation, which read

ds ∂vi ~ ~ ),
ρT = σi j j + ∇ · (κ∇T (12.24)
dt ∂x
needs to be augmented by the production of Ohmic heat; per unit
time, the induction current ~j0 in the rest frame of the fluid dissipates
the energy

~02 ~2 2
~j0 · E~ 0 = j ≈ j = c ~ ×B
2
~ ,

∇ (12.25)
σ σ 16π2 σ
which must be added to the right-hand side of the entropy equation,

ds ∂vi ~ ~ )+ c
2 
~ ×B
2
~ ;
ρT = σi j j + ∇ · (κ∇T ∇ (12.26)
dt ∂x 16π2 σ

• if we need to express this by the specific energy density ε instead


of the specific entropy density s, we start from the energy con-
servation equation of viscous hydrodynamics and augments that
analogously; this yields

∂  ρ~v2 ~ 2 
 
B ~ · ~q = 0
+ ε +  + ∇ (12.27)
∂t 2


with the extended energy current density vector

~v2 ∂T
!
c
q =ρ
i
+ w vi − κ − σij v j +  ijk E j Bk , (12.28)
2 ∂xi 4π
which now contains the Poynting vector

 ~j ~v × B~ 
 
c c
S~ = E~ × B
~=  − ~
 × B
4π 4π σ c
c2 ~ ~ ~ − 1 (~v × B) ~ ×B~,
= (∇ × B) × B (12.29)
16π σ 2 4π
and this yields the energy current density

~v2 ∂T
!
q = ρ
i
+ w vi − κ − σij v j (12.30)
2 ∂xi
c2 h ~ ~ + 1 B
~ × B)
ii h
~ × (~v × B)
ii
~ ;
− B × (∇
16π2 σ 4π

• for incompressible flows with ∇~ · ~v = 0, these equations simplify


somewhat; then, we first have to satisfy

~ ·B
∇ ~=0, ~ · ~v = 0 ,
∇ (12.31)
154 KAPITEL 12. MAGNETO-HYDRODYNAMICS

next, the induction equation,

~
∂B 2
= ∇ ~ + c ∇
~ × (~v × B) ~ 2B
~
∂t 4πσ
~ · B)
~ − B(
~ ∇~ · ~v) + ( B ~ v − (~v · ∇)
~ · ∇)~ ~ B~+ c2 ~ 2 ~
= ~v(∇ ∇ B
4πσ
2
~ v − (~v · ∇)
~ · ∇)~
= (B ~ B~+ c ∇~ 2B
~, (12.32)
4πσ
thus
~
∂B 2
~ B
+ (~v · ∇) ~ = (B ~ v+ c ∇
~ · ∇)~ ~ 2B
~, (12.33)
∂t 4πσ
~ · ~v = 0
and Euler’s equation reads with ∇

∂~v ~ 2 
 
1
~ v=− ∇~ P +  + 1 ( B
B ~ B
~ · ∇) ~ 2~v , (12.34)
~ + ν∇
+ (~v · ∇)~

∂t ρ 8π 4πρ

where ν = η/ρ is the specific viscosity per unit mass; likewise, the
viscosity tensor σij in the energy conservation equation simplifies;

12.1.5 Magnetic advection and diffusion

• two terms determine the temporal change of the magnetic field in


the induction equation,

~ × (~v × B)
~ c2 ~ 2 ~
∇ and ∇ B; (12.35)
4πσ

~ × (~v × B),
• the first term, ∇ ~ determines the transport of the magnetic
field with the fluid flow; it is called advection term; its order-of-
magnitude is
vB
, (12.36)
L
if L is a typical length scale of the flow;

• the second term, c2 ∇ ~ 2 B/(4πσ),


~ determines the diffusion of the
magnetic field due to the finite conductivity of the plasma; its
magnitude is of order
c2 B
, (12.37)
4πσ L2
and it vanishes if the conductivity is ideally (infinitely) large;

• the ratio of the two orders-of-magnitude,

advection 4πσ L2 vB 4πσvL


= 2 = ≡ RM , (12.38)
diffusion c B L c2
12.2. GENERATION OF MAGNETIC FIELDS 155

is called the magnetic Reynolds number; obviously, the diffusion


can be neglected if RM  1, and the induction equation simplifies
to
∂B~
~ × (~v × B)
−∇ ~ =0 (12.39)
∂t
in this case, there is no diffusion, and the magnetic field is “frozen”
into the plasma; the physical reason for that is that, if the conduc-
tivity is ideally high, σ → ∞, each motion of the magnetic field
with respect to the plasma immediately induces strong currents
which counter-act their origin, i.e. the motion of the field; this is
the typical case in astrophysical plasmas;

• in the opposite limit, RM  1, which occurs of the conductivity is


small, the induction equation reads

~
∂B c2 ~ 2 ~
= ∇ B; (12.40)
∂t 4πσ

the magnetic diffusion time scale therefore is

L2
τdiff ≈ 4πσ ; (12.41)
c2

plasmas in the laboratory are typically characterised by RM  1,


while astrophysical plasmas typically have RM  1;

12.2 Generation of Magnetic Fields

• the induction equation constains no source term, i.e. it only descri-


bes how existing magnetic fields change, but if B~ = 0 initially, this
remains so; this is a consequence of the assumption that ions and
electrons are ideally (tightly) coupled to each other;

• if that is not the case, the motions of the electrons and the ions need
to be considered separately, in particular with different velocities
~ve and ~vi ; then, the two separate Euler equations for the electrons
and the ions are

d~ve ~v
!
ne me = −∇P~ e − ne e E~ + × B e ~ ,
~ − ne me ∇Φ
dt c
d~vi ~v
!
ni mi = −∇P~ i − ni e E~ + × B
i ~ ; (12.42)
~ − ni mi ∇Φ
dt c

we divide these equations by ne me and ni mi and subtract the second


156 KAPITEL 12. MAGNETO-HYDRODYNAMICS

from the first; this yields

d(~ve − ~vi ) ~ e
∇P ~ i
∇P
= − +
dt ne me ni mi
e ~ ~ve ~
!
− E+ ×B
me c
e ~ ~vi ~
!
− E+ ×B ; (12.43)
mi c

• since the ion mass mi is much larger than the electron mass me , but
ne = ni ≡ n, equation (12.43) can be approximated by

d(~ve − ~vi ) ~ e
∇P e ~ ~ve ~
!
=− − E+ ×B ; (12.44)
dt nme me c

• this equation must be accomplished by a phenomenological colli-


sion term through which different electron and ion velocities can
be justified or produced in the first place; introducing a collision
time τ, we write

d(~ve − ~vi ) ~ e
∇P e ~ ~ve ~
!
~ve − ~vi
=− − E+ ×B − ; (12.45)
dt nme me c dt

• the current is
~j = eni~vi − ene~ve = en(~vi − ~ve ) , (12.46)

where we emphasise the implicit assumption of singly-charged


ions;

• we now assume that the relative drift velocity is constant,


d(~vi − ~ve )
=0, (12.47)
dt
which also means that the current density is constant; in such a
situation, the electric field following from (12.44) is
~ e ~ve
∇P me
E~ = − − ×B~− (~vi − ~ve )
ene c eτ
~ e ~ve
∇P me ~j
= − − ×B~+ ; (12.48)
ene c ne2 τ
this implies
~ ~ e ~ve ~j 
 
∂B ~ × E~ = c∇
 ∇P
~ ×  ~− m e
= −c∇  en + c × B
∂t ne2 τ 

e

~ e
c ~ ∇P ~
= ∇× +∇ ~ − me c ∇
~ × (~ve × B) ~ × j ; (12.49)
e ne e2 τ n
12.3. AMBIPOLAR DIFFUSION 157

• now, we have
~ ~
∇ ~ e × ∇n
~ × ∇Pe = ∇P (12.50)
ne n2
~ × ∇P
because ∇ ~ e vanishes identically, and

~ ~ ~ ~
~ × j = ~j × ∇n + ∇ × j

n n2 n
~
=
c ~
∇ × (∇ ~ + ~j × ∇n ;
~ × B) (12.51)
4πn n2
if the current is flowing along the gradient in the number density
of the electrons, the latter term vanishes identically, and we obtain
the modified induction equation
~
∂B c ~ 2
= 2 (P ~ ~
e × ∇n) + ∇ × (~
~ + me c ∇
ve × B) ~ 2B
~; (12.52)
∂t en 4πne2 τ
inserting the definition
ne2 τ
σ≡ (12.53)
me
of the conductivity, this equation is identical to the previous form
of the induction equation, except for the first term,
c ~ ~ ,
(Pe × ∇n) (12.54)
en2
whcih now appears as the source of the magnetic field; mecha-
nisms like this for creating magnetic fields are called battery
mechanisms;

12.3 Ambipolar Diffusion

12.3.1 Scattering cross section


• in a mixture of neutral particles and plasma, the magnetic field
can be thought of “frozen into” the plasma; collisions between the
plasma and neutral particles then creates a friction force between
the plasma and the neutral fluid, which causes a diffusion of the
magnetic field; this diffusion process is called “ambipolar”;
• in order to determine it, we first need a cross section σ for the col-
lisions, or, more conveniently, the velocity-averaged cross section
hσvi; we adopt two limiting cases for it;
• if v is very large, we may approximate the cross section by its
geometrical value; if ri and rn are the effective radii of the ions and
the neutral particles, respectively, we have
σ = π(ri + rn )2 (12.55)
158 KAPITEL 12. MAGNETO-HYDRODYNAMICS

and thus
hσvi = hviσ = hviπ(ri + rn )2 ; (12.56)
the relative velocity during the interaction is

v = |~vi − ~vn | ; (12.57)

• if v is sufficiently small, the ion can polarise the neutral particle,


by which the cross section is enlarged; the electric field of the ion,
assumed singly charged, is
Ze
E~ i = 2 ~er , (12.58)
r
while it should be possible to approximate the electric field of the
polarised neutral particle as a dipole field,

2~p
E~ n = − 3 , (12.59)
r

where ~p = αE~ i is the plarised dipole moment with a parameter α;

• the dipole field of the neutral particle exerts the force


2 2
Zeα Z e
F~ = ZeE~ n = −2Ze 5 ~er = −2α 5 ~er (12.60)
r r
on the ion, which has the potential

α Z 2 e2
U= ; (12.61)
2 r4
the motion of the ion past the neutral particle can be characterised
by the minimum separation r0 , the impact parameter b and the
velocity v∞ at infinity;

• angular-momentum conservation requires

µv∞ b = µr0 v0 (12.62)

with the reduced mass


mi mn
µ≡ ; (12.63)
mi + mn
in addition, energy conservation demands

µ 2 µ αZ 2 e2
v∞ = v20 − ; (12.64)
2 2 2r04

from that, and because of


bv∞
v0 = , (12.65)
r
12.3. AMBIPOLAR DIFFUSION 159

we obtain a quartic equation for the minimum separation,


αZ 2 e2
r04 = b2 r02 − , (12.66)
µv2∞
or s
b2
b4 αZ 2 e2
2
r0,± = ± − ; (12.67)
2 4 µv2∞
2
both roots are mathematically possible, but physically, only r0,+ is
relevant because in the limiting case of vanishing coupling α, the
minimum radius must equal the impact parameter, r0 = b, since
then the ion is not scattered at all; r0 will become minimal of the
impact parameter is
!1/4
4αZ 2 e2
b0 = ; (12.68)
µv2∞
since the force between the ion and the neutral particle decreases
very steeply with r, by far the strongest effect occurs for close
encounters; thus, we estimate the cross section as
2πZe α
r
σ = πb0 =
2
(12.69)
v∞ µ
obviously, hσv∞ i is independent of the asymptotic velocity v∞ ,
α
r
hσv∞ i = 2πZe ; (12.70)
µ

12.3.2 Friction force; diffusion coefficient


• a single collision transfers the momentum
|∆~p| = µ|~vi − ~ve | , (12.71)
and the scattering rate per volume is ni nn hσv∞ i, thus the momen-
tum transfer per time and volume is
f~r = ni nn hσv∞ iµ(~vi − ~ve ) , (12.72)
which corresponds to a friction force;
• with
ρi ρn = ni mi nn mn = (mi + mn )ni nn µ , (12.73)
this can be cast into the form
f~r = γρi ρn (~vi − ~vn ) , (12.74)
where γ is the friction coefficient
hσv∞ i
γ≡ ; (12.75)
mi + mn
160 KAPITEL 12. MAGNETO-HYDRODYNAMICS

• in the two limiting cases of very small or very large relative velo-
city, we find

π(ri + rn )2
γ = |~vi − ~vn | or
mi + mn
α
r
2πZe
γ = (12.76)
mi + mn µ

• if the Lorentz force and the friction force balance each other, the
relative drift velocity between the ions and the neutral particles is
correspondingly established; the Lorentz force density is

~j × B
~ 1 ~ ~
f~L = = ~,
(∇ × B) × B (12.77)
c 4π
from which follows
1 ~ ~ ~;
γρi ρn (~vi − ~vn ) = (∇ × B) × B (12.78)

or
~ × B)
(∇ ~ ×B ~
~vd ≡ ~vi − ~vn = ; (12.79)
4πγρi ρn
this is of order
B2
vd ≈ ; (12.80)
4πγρi ρn L
• the magnetic field, which is thought to be “frozen” into the flow
of the plasma, satisfies the equation

~
∂B ~ × (B
+∇ ~ × ~vi ) = 0 ; (12.81)
∂t
in order to calculate the flow of the magnetic field relative to the
neutral particles, we replace ~vi by ~vd , which leads to the equation

~ ~ × B)
~ ×B ~ 
 
∂B ~ ×  B (∇
+∇  ~ × 4πγρ ρ  = 0 ;

(12.82)
∂t i n

the second term is of order


B3
, (12.83)
4πγρi ρn L2
and corresponds to a diffusion coefficient

B2
D≈ ≈ vd L ; (12.84)
4πγρi ρn
Kapitel 13

Waves in Magnetised Plasmas


further reading: Shu, “The Phy-
sics of Astrophysics, Vol II:
13.1 Waves in magnetised cold plasmas Gas Dynamics”, chapter 22; Pad-
manabhan, “Theoretical Astro-
13.1.1 The dielectric tensor physics, Vol. I: Astrophysical
Processes”, sections 9.7; Ishima-
• we study the propagation of electromagnetic waves in a plasma in ru, “Basic Principles of Plasma
which random particle motion is negligible, whose temperature is Physics”, chapter 5
thus low, and which can thus be considered as cold;

• the equation of motion of an electron in such a plasma with external


magnetic field B~ 0 is then only determined by the Lorentz force,

d~v ~v ~
m = −eE~ − e × B 0 , (13.1)
dt c
where E~ is the internal electric field;

• we now assume that the spatial change of the electric field is


negligible and decompose ~v and E~ into harmonic contributions
varying with time as exp(−iωt); then, each monochromatic velocity
mode is determined by
e ~ ~v ~
!
− iω~v = − E + × B0 , (13.2)
m c
which implies the matrix equation
 e i k j e i
iωδij −  B v = E ; (13.3)
mc jk 0 m

• the term in parentheses on the left-hand side can be represented


by the matrix

ω3 −ω2 
   
 iω 0 0   0
 0 iω 0  −  −ω
3
0 ω1  ≡ M , (13.4)
   
0 0 iω ω2 −ω1 0

161
162 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

where we have identified the Larmor frequency


eB0
ωB ≡ (13.5)
mc
and used the definition
eB0 Bi0 Bi
ωi ≡ = ωB 0 ; (13.6)
mc B0 B0

• the inverse of the matrix (13.4) is


1
M −1 = − (13.7)
det M
ω − ω21
2
−ω1 ω2 − iωω3 −ω1 ω3 − iωω2
 
 
−ω ω − iωω ω2 − ω22 −ω1 ω3 − iωω2  ,
 
 1 2 3
−ω1 ω3 − iωω2 −ω1 ω3 − iωω2 ω2 − ω23
with
det M = −iω(ω2 − ω2B ) , (13.8)
which can be written in components as
 j −1 −i  
Mk = ω δ
2 j
− ω j
ωk − iω ω
j l
; (13.9)
ω(ω2 − ω2B ) k kl

• this allows us to write the solution for the velocity as


e
~v = M −1 E~ , (13.10)
m
which yields, with the inverse matrix (13.9), the velocity com-
ponents v j
−ie  
vj = ω 2 j
E − ω j
ωk E k
− iω E ω
j k l
, (13.11)
mω(ω2 − ω2B ) kl

or, written in vector notation,


ω2B
" #
−ieω iω
E~ − 2 (E~ · ~eB )~eB − E~ × ~eB ,
B
~v = (13.12)
m(ω2 − ω2B ) ω ω
~ 0 /B0 is the unit vector in the direction of the external
where ~eB ≡ B
magnetic field B ~ 0;

• because of E~ + 4πP
~ = D,
~ the polarisation is
~ ~
~ = D−E ;
P (13.13)

its temporal change is the current density
~ ~ ~
~j = −ene~v = ∂P = −iωP
~ = −iω D − E , (13.14)
∂t 4π
again under the assumption of a harmonic dependence on time,
∝ e−iωt ;
13.1. WAVES IN MAGNETISED COLD PLASMAS 163

• therefore, we find for the dielectric displacement

~ = 4πi ~j + E~ = E~
D (13.15)
ω
4πe2 ne ω ~ ω2B ~
" #
iωB ~
+ − E − 2 (E · ~eB )~eB − E × ~eB ;
m(ω2 − ω2B ) ω ω

if we can write this in the form D j = kj E k , we can read off the
dielectric tensor; we thus obtain

4πe2 ne 4πe2 ne ω2B j


" #
k = δk 1 −
j j
+ e eBk
m(ω2 − ω2B ) m(ω2 − ω2B ) ω2 B
4πie2 ne ωB j l
+  e , (13.16)
m(ω2 − ω2B ) ω kl B

which we abbreviate as

kj ≡ ⊥ δkj + (k − ⊥ )eBj eBk + igklj elB ; (13.17)

• if we identify the plasma frequency there,


r
4πe2 ne
ωp = , (13.18)
m
we obtain the expressions

ω2p
⊥ = 1 − ,
ω2 − ω2B
ω2p ω2B
k = ⊥ + 2
ω − ω2B ω2
ω2p ω2B ω2p
!
= 1− 2 1− 2 =1− 2 ,
ω − ω2B ω ω
ωp ωB
2
g = ; (13.19)
ω2 − ω2B ω

13.1.2 Contribution by ions


• if ions need to be taken into consideration, these quantities change
according to

⊥ − 1 → (⊥ − 1)e + (⊥ − 1)i ,


k − 1 → (k − 1)e + (k − 1)i ,
g → ge + gi ; (13.20)

there, the plasma and Larmor frequencies of the electrons and the
ions will then have to be distinguished;
164 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

• the larmor frequency of the ions is


ZeB0 Zme
ωBi = = ωBe  ωBe ; (13.21)
mi c mi
the plasma frequency of the ions is given by

4πZ 2 e2 ni 4πZ 2 e2 ne Zme 2


ω2pi = = = ω , (13.22)
mi mi mi pe
therefore its ratio to the electron plasma frequency is

ωpi
r
Zme
= 1; (13.23)
ωpe mi

• thus, the contribution of the ions to the longitudinal dielectrici-


ty k is negligible, but not necessarily their contribution to the
perpendicular dielectricity ⊥ ; the ratio

ωpi  2 2 2
2
ω2 −ω2Bi  ωpi  ω − ω2Be

ω2 =  2  2 ≈1 (13.24)
2 pe 2 ωpe ω − ω2Bi
ω −ωBe

only if

ω − ω2Be 1
2
Zme
≈ , with f ≡ (13.25)
ω − f 2 ω2Be f
2
mi

holds; this is the case if

− f (ω2 − ω2Be ) ≈ ω2 − f 2 ω2Be (13.26)

is satisfied, or
f − f2 2
ω2 ≈ ω ≈ f (1 − f )ω2Be ; (13.27)
1 + f Be
similarly, the contributions of ions and electrons to g are compara-
bly large;

13.1.3 General dispersion relation


• with the expressions (13.17) and (13.19), the general dispersion
relation
k j kk ω2 j
!
det δk − 2 − 2 2 k = 0
j
(13.28)
k kc
leads to the equation
!4 !2
kc kc
A +B +C = 0 , (13.29)
ω ω
13.1. WAVES IN MAGNETISED COLD PLASMAS 165

with the coefficient


k j kk eB j k j eBk kk ki k j ekB
A ≡  jk = ⊥ + ( k − ⊥ ) + igi jk ; (13.30)
k2 k2 k2
if θ denotes the angle between the magnetic field and the wave,
~eB · ~k = cos θ, this implies

A = ⊥ + (k − ⊥ ) cos2 θ = ⊥ sin2 θ + k cos2 θ , (13.31)

because
ki k j ek ~k  ~k
 
i jk 2 B = ·  × ~eB  = 0 ;

(13.32)
k k k
similarly, we find

B = −⊥ k (1 + cos2 θ) − (⊥2 − g2 ) sin2 θ (13.33)

and
C = k (⊥2 − g2 ) , (13.34)
and, by definition, A equals the longitudinal dielectricity l ;

13.1.4 Wave propagation parallel to the magnetic field


• if ~k k B,
~ the angle θ = 0 and thus

A = k , B = −2⊥ k , C = k (⊥2 − g2 ) (13.35)

the roots of the dispersion relation (13.29) are then


!2
kc 1  q 
= 2⊥ k ± 4⊥2 k2 − 4k2 (⊥2 − g2 )
ω 2k
ω2p ω2p ωB
= ⊥ ± g = 1 − 2 ±
ω − ω2B ω2 − ω2B ω
ω2p  ωB 
= 1− 2 1 ∓
ω − ω2B ω
ω2p
= 1− ; (13.36)
(ω ± ωB )ω

• in order to determine what kind of waves are described by these


relations, we return to the equation (11.35)
ki k j ω2 i j
!
δj − 2 − 2 2 j E = 0 ;
i
(13.37)
k kc

for longitudinal waves, E~ k ~k, we have


ki k j Ek j j k
i Ek k
 ij E j = ⊥ E i + (k − ⊥ ) + ig jk , (13.38)
k2 k k k
166 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

~ k ~k; this can obviously be reduced to


since we had also assumed B

 ij E j = ⊥ E i + (k − ⊥ )E i = k E i , (13.39)

and, in addition, we have

ki k j j ki k j Ek j
!
δj − 2 E = E − 2
i i
=0; (13.40)
k k k

thus, longitudinal waves do not satisfy the dispersion relation


(13.37), which implies that (13.37) describes transversal waves;

• for transversal waves, E j k j = 0, and thus

ω2 k
!
jk
E − 2 2 ⊥ E + ig jk E
i i i
=0 (13.41)
kc k

or, in vector notation,

~k 
  
1
E~ − ⊥ E~ + ig E~ ×  = 0 ;
 
(13.42)
⊥ ± g  k

if we turn the z axis into the direction of ~k, the components E x and
Ey of E~ read

1  
Ex − ⊥ E x + igEy = 0 and
⊥ ± g
1  
Ey − ⊥ Ey − igE x = 0 , (13.43)
⊥ ± g

which is solved if
E x = ±iEy , (13.44)
characterising circularly polarised light;

13.1.5 Faraday rotation


• the two solutions (13.36) for the dispersion relation thus describe
the propagation of left- and right-circular polarised transversal
waves which obey different dispersion relations; left- and right-
circular polarised light thus propagates differently along the ma-
gnetic field; the polarisation direction of linearly polarised light is
thus rotated; this effect is called Faraday rotation;

• in the limit of ω  ωp , we have

ω2 ω2p  ω2  ω2p  ωB 


   
k±2 = 2  ,

1 −  ≈ 2 1 − 2 1 ∓ (13.45)
c (ω ± ωB )ω c ω ω
13.1. WAVES IN MAGNETISED COLD PLASMAS 167

or
ω  ω2p  ωB  ω ω2p ω2p ωB
 
k± ≈ 1 − 1∓  = − ±
c 2ω2 ω c 2ωc 2ω2 c
≡ k0 ± ∆k ; (13.46)
the first term k0 corresponds to the wave vector in the unmagnetised
medium, which the second term ∆k causes a phase shift between
left- and right-circular polarised light, and therefore to a rotation
of linear polarisation by the angle
ωp ωB
Z 2
4πe2 ne eB dz
Z Z
ψ = ∆kdz = dz =
2ω2 c m mc 2ω2 c
2πe3
Z
= dz ne B ; (13.47)
m2 c2 ω2
obviously, the Faraday rotation is proportional to ω−2 or, equiva-
lently, to the squared wave length λ2 ; the expression
Z
dz ne B ≡ RM (13.48)

is called the rotation measure;

13.1.6 Wave propagation perpendicular to the magne-


tic field
• in this case, ~k ⊥ B,
~ thus θ = π/2, and the coefficients A, B and C
turn into
A = ⊥ , B = −⊥ k − ⊥2 + g2 , C = k (⊥2 − g2 ) ; (13.49)
these imply the solutions for the dispersion relation
k
!2 !
kc C 1 C
= + ± ⊥ k −
ω 2 2⊥ k 2⊥ k
C g 2
= k or = ⊥ − , (13.50)
⊥ k ⊥
where
C
B = −⊥ k − (13.51)
k
was used;
• the first of these solutions is a dispersion relation which is obvious-
ly independent of B; ~ perpendicular to the magnetic field, wave
propagation is thus possible as in an unmagnetised plasma; these
waves are transversal and polarised in the direction of B; ~ the other
dispersion relation corresponds to waves with longitudinal and
transversal components;
168 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

13.2 Hydro-Magnetic Waves

13.2.1 Linearised perturbation equations


• now we consider, in a way very similar to the treatment of sound
waves in a neutral fluid, the propagation of waves in a magnetised
plasma; for simplicity, we assume that dissipation and heat conduc-
tion are unimportant, ζ = η = κ = 0, and that the conductivity be
infinite, σ−1 = 0; then, the equations of magneto-hydrodynamics
read
∂B~
~ ·B
∇ ~=0, =∇ ~ , ∂ρ + ∇
~ × (~v × B) ~ · (ρ~v) = 0 ,
∂t ∂t
∂~v ~
~ v = − ∇P + 1 (∇
+ (~v · ∇)~ ~ × B)
~ ×B ~; (13.52)
∂t ρ 4πρ
the energy conservation equation is not relevant for the following
considerations;
• we assume that an equilibrium solution exists,
~0 ,
B ρ0 , P0 , ~v0 = 0 , (13.53)
which is perturbed by small quantities
~,
δB δρ , δP , δ~v ; (13.54)

• in absence of dissipation, entropy is conserved along flow lines;


we further assume isentropic flow, thus s = const. everywhere in
the flow;
• we now linearise the MHD equations and obtain, to first order in
the perturbations
~
∂δ B
~ · δB
∇ ~=0, ~ × (δ~v × B)
=∇ ~ ,
∂t
∂δρ ~
+ ∇ · (ρδ~v) = 0 ; (13.55)
∂t
with ρ = const. for the homogeneous equilibrium solution, the last
equation implies
∂δρ ~ · δ~v = 0 ;
+ ρ∇ (13.56)
∂t
finally, Euler’s equation reads to first order in the perturbations
∂δ~v ~ + δP)
∇(P
= −
∂t ρ + δρ
1 nh
~ × (B
~ + δ B)
~ × B
i 
~ + δB
~
o
+ ∇
4π(ρ + δρ)
~
∇δP ~ × δ B)
(∇ ~ ×B~
≈ − + , (13.57)
ρ 4πρ
13.2. HYDRO-MAGNETIC WAVES 169

again under the assumption that the equilbrium solution is homo-


~ =0=∇
geneous, thus ∇P ~ × B;
~

• again, we decompose the perturbations Q into plane waves,

δQ ∝ exp i(~k · ~x − ωt) ,


h i
(13.58)

and find

i~k · δ B
~=0, ~ = i~k × (δ~v × B)
−iωδ B ~ (13.59)

for the magnetic field and

i~kδP (i~k × δ B)
~ ×B
~
− iωδρ + iρ~kδ~v = 0 , −iωδ~v = − + (13.60)
ρ 4πρ

for the velocity;

• the pressure perturbation can be related to the density perturbation


as usual,
δP = c2s δρ , (13.61)
where cs is the sound speed in the neutral gas, and this allows us
to write Euler’s equation as

ic2s ~ (i~k × δ B)
~ ×B
~
− iωδ~v = − kδρ + ; (13.62)
ρ 4πρ

• without loss of generality, we can now rotate the coordinate frame


such that ~k points along the positive x axis and that B
~ falls into the
x-y plane; further, we denote the angle between ~k and B ~ with ψ;

• first, the continuity equation yields

~k · δ~v kδv x
δρ = ρ =ρ ; (13.63)
ω ω
the phase velocity of the wave is
ω
ck ≡ , (13.64)
k
and thus
δv x
δρ = ρ ; (13.65)
ck

• the induction equation turns into

δv x By − δvy Bx δvz Bx
δBx = 0 , δBy = , δBz = − (13.66)
ck ck
170 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

(the latter because of δBx = 0); and Euler’s equation finally reads,
also in components
c2s kδρ kBy δBy
−ωδv x + = − ,
ρ 4πρ
kBx δBy
−ωδvy = ,
4πρ
kBx δBz
−ωδvz = ; (13.67)
4πρ

• with (13.64) and (13.65), we obtain from here the two sets of
equations
Bx δBz
ck δBz = −δvz Bx , ck δvz = − (13.68)
4πρ
and
Bx δBy
ck δBy = δv x By − δvy Bx , ck δvy = − ,
4πρ
c2s c2s By δBy
!
ck δv x − δv x = δv x ck − = ; (13.69)
ck ck 4πρ

• the first set (13.68) of equations contains only δBz and δvz ; the
second set (13.69) couples δv x and δBx to δvy and δBy ; we can thus
distinguish waves which are purely transversal and are described
by (13.68), and other waves, which have longitudinal and trans-
versal components; since the density perturbations are caused by
velocity perturbations δv x , only longitudinal waves are responsible
for them;

13.2.2 Alfvén waves


• the first set (13.68) of equations, describing transversal waves,
yields
δvz B2x ω Bx
ck δvz = ⇒ ck = = p (13.70)
4πρck k 4πρ
if we eliminate δBz between the first and the second equation;
~ · ~k, this can also be written as
since Bx = B
~ · ~k
B
ω= p ≡ cA k cos ψ , (13.71)
4πρ
where the Alfvén velocity
!1/2
B2
cA = (13.72)
4πρ
was identified;
13.2. HYDRO-MAGNETIC WAVES 171

• the phase velocity of Alfvén waves is

cA cos ψ , (13.73)

their group velocity is

∂ω ~
B
= p ; (13.74)
∂~k 4πρ

accordingly, Alfvén waves travel along ~k with the phase velocity


cA cos ψ, which depends on the angle between ~k and B;
~ furthermo-
re, they are transversal and transport physical quantities (energy
~ independent of ~k!
and such) along the magnetic field B,

13.2.3 Slow and fast hydro-magnetic waves


• we now set δvz = 0 and consider waves which obey the second set
(13.69) of equations; if we eliminate
δv x By − δvy Bx
δBy = (13.75)
ck
from the second and third equations, we find
Bx  
c2k δvy = − δv x By − δvy Bx ,
4πρ
  By  
c2k − c2s δv x = δv x By − δvy Bx ; (13.76)
4πρ
further eliminating
!−1
Bx By δv x 2 B2 Bx By δv x
δvy = − ck − x =− (13.77)
4πρ 4πρ 4πρc2k − B2x

from the second equation yields

  B2y δv x B2x B2y δv x


ck − cs δv x =
2 2
+ ; (13.78)
4πρ 4πρ(4πρc2k − B2x )

• δv x cancels on both sides, and we find

B2y B2x
  !
c2k − c2s = 1+
4πρ 4πρc2k − B2x
B2y c2k
= ⇒
4πρ c2k − B2x /(4πρ)
B2x c2k B2y
  !
2 2 2
ck − cs ck − = ; (13.79)
4πρ 4πρ
172 KAPITEL 13. WAVES IN MAGNETISED PLASMAS

• this quadratic equation in c2k has the solutions


s
!2
c2s B c2s 2
B2 B2 c2
c2k,± = + ± + − x s (13.80)
2 8πρ 2 8πρ 4πρ
s
!2
cs + cA
2 2
c2s + c2A
= ± − c2s c2A cos2 ψ ;
2 2

thus, a fast and a slow wave mode are possible;

• we first consider the case in which B ~ is (almost) parallel to ~k, hence


ψ  1 and cos ψ ≈ 1 − ψ /2; then,
2 2

 r 
ω 2
1  2 
c2k = ≈ c2s + c2A ±

c2s − c2A 

k2 2
1 2  ( c2 or
= c + cA ± |cs − cA | =
2 2 2 s
(13.81)
2 s c2A

to first order in ψ; accordingly, the fast wave propagates with the


faster of the sound and the Alfvén speeds, the slow wave with the
slower;

• in this approximation, By ≈ ψB, and the Alfvén mode has

c3A δB Bx δBy
δv x = ψ2 , δvy = − , (13.82)
c2A − c2s B 4πρcA

which can be approximated as


δB
δv x ≈ 0 , δvy ≈ cA ; (13.83)
B
the acoustic mode has
Bx δBy c2A δB
By δBy = 0 , δvy = − ≈− ; (13.84)
4πρcs cs B

~ ⊥ ~k, we find
• if, finally, cos2 ψ = 0 or B
1 2 
c2k,± = cs + c2A ± c2s + c2A , (13.85)
2
thus q
ck = c2s + c2A (13.86)
for the fast and ck = 0 for the slow MHD wave;
Kapitel 14

Jeans Equations and Jeans


Theorem
further reading: Binney, Tremai-
ne, “Galactic Dynamics”, secti-
14.1 Collision-less motion in a gravitational ons 4.1–4.4
field

14.1.1 Motion in a gravitational field


• particles in a gas or a fluid move almost unaccelerated until they
meet another particle, which forces them to change their state of
motion abruptly; we had based our treatment of hydrodynamics on
the assumption that the collisions occur on much smaller length
scales λ than those that characterise the extent of the entire system,
L; in plasma physics, we had seen that the shielding of charges
on length scales λD allows a hydrodynamical treatment despite
the inifite range of electrostatic interactions, provided there are
sufficiently many particles in a volume ≈ λ3D ; in these cases, the
interactions are effectively extremely short-ranged; likewise, we
had assumed in our treatment of local thermodynamical equilibri-
um that the mean free path of the photons is much smaller than
the characteristic dimensions of the system under consideration;
• studying the motion of many point masses such as stars in a gra-
vitational field, we encounter a fundamental change: the forces
between the particles are long-ranged and cannot be shielded; a
single star in a galaxy, for instance, thus experiences not only the
attraction of its nearest neighbours, but of a large fraction of all
stars in the entire galaxy;
• let us consider now a two-dimensional system, such as a galactic
disk, which we shall assume to be infinitely extended for now and
in whose centre we assume a star; the disk be randomly covered
by stars such that their mean number density is spatially constant;

173
174 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

• in a circular ring of radius r and width dr, we find

dN = 2πrdr n (14.1)

stars, whose gravitational force on the star in the centre is

Gm2
dF = 2πrdr n (14.2)
r2
of the mass of all stars is assumed to be the same, for simplicity;

• of course, the directions of all forces cancel in the mean, but the
contribution of arbitrarily distant rings diverges logarithmically,
Z Z
dr
dF = 2πGnm 2
= 2πGnm2 ln r ; (14.3)
r
thus, the structure of the entire stellar system is important for the
dynamics of the stars in the gravitational field;

• in the spirit of our distinction of microscopic and macroscopic


forces, which we had made when introducing hydrodynamics, the
forces in a system which is dominated by self-gravity are also
macroscopic; therefore, the collision terms, which describe the
interaction on a microscopic scale, can be neglected here at least
to first order of approximation;

• thus, we begin our treatment of self-gravitating systems with the


collision-less Boltzmann equation,

d f (~x,~v, t) ∂ f ˙ ~ ∂f
= + ~x · ∇ f + ~v˙ · =0; (14.4)
dt ∂t ∂~v

14.1.2 The relaxation time scale


• before we turn to a detailed study of Eq. (14.4) in a gravitational
field, we investigate approximately how the trajectory of a star
through a galaxy which is composed of individual stars deviates
from the trajectory through a hypothetical, “smooth” galaxy;

• we consider the passage of a star by another star employing Born’s


approximation, i.e. we integrate the deflection along a straight
trajectory passing the deflecting star at an impact parameter b;
the perpendicular force at the location x along the hypothetical,
straight trajectory is


2

~ ⊥ Φ| = − √ Gm Gm2 b
F⊥ = | − ∇ = ,

(14.5)
∂b b2 + x2 (b2 + x2 )3/2

where Φ is the gravitational potential;


14.1. COLLISION-LESS MOTION IN A GRAVITATIONAL FIELD175

• with x ≈ vt, we have

Gm2
"  vt 2 #−3/2
F⊥ ≈ 2 1 + , (14.6)
b b
and Newton’s second law mv̇⊥ = F⊥ thus implies

Gm ∞
Z "  vt 2 #−3/2
δv⊥ ≈ 1+ dt
b2 −∞ b
2Gm ∞
Z
2Gm
= (1 + τ2 )−3/2 dτ = ; (14.7)
bv 0 bv

• let N be the number of stars in the galaxy and R be its radius, then
the fiducial test star experiences
N 2N
δN = 2πbδb n = 2πbδb = 2 bδb (14.8)
πR 2 R
such encounters with an impact parameter between b and b + δb;
the mean quadratic velocity change is thus
!2
2Nbδb 2Gm 8NG2 m2 δb
δv⊥ ≈
2
= ; (14.9)
R2 bv R2 v2 b

• integration this expression, we need to take into account that the


assumption of Born’s approximation requires that
2Gm Gm
δv⊥ . v ⇒ .v ⇒ b & bmin = , (14.10)
bv v2
and thus we obtain
Z ∞ !2 !2
2Gm 2Gm
∆v⊥ =
2
δv⊥ ≈ 8N
2 R
ln b|bmin ≡ 8N ln Λ ,
bmin bv bv
(14.11)
where
R Rv2
ln Λ ≡ ln = ln ; (14.12)
bmin Gm
is the so-called Coulomb logarithm;
• a typical velocity for the stars in a galaxy of mass M = Nm is,
according to the virial theorem,
GMm GNm
v2 ≈ ⇒ R≈ ; (14.13)
R v2
from which we obtain
∆v2⊥ 8 ln Λ
≈ ; (14.14)
v2 N
this shows by what relative amount the star’s velocity is changed
during one passage through the galaxy;
176 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

• the Coulomb logarithm ln Λ follows from

R Rv2
ln Λ = ln = ln ≈ ln N , (14.15)
bmin Gm

i.e. the relative velocity change is approximated by

∆v2⊥ 8 ln N
≈ ; (14.16)
v2 N

• after ncross passages through the galaxy, the total relative velocity
change will approximately be

8 ln N
ncross ; (14.17)
N
if this should be of order unity, the number of passages needs to be

N
ncross ≈ ; (14.18)
8 ln N
one passage takes approximately the time

R
tcross ≈ , (14.19)
v
i.e. a complete velocity change needs the relaxation time

R N
trelax ≈ ; (14.20)
v 8 ln N

• in a galaxy, we have

10 kpc
tcross ≈ ≈ 5 × 107 yr , (14.21)
200 km s−1

and N ≈ 1011 , thus the relaxation time is

trelax ≈ 3 × 1016 yr , (14.22)

which is much more than the age of the Universe; this illustrates
that in many astrophysically relevant systems, the collision-less
Boltzmann equation can be used;

• in a globular cluster, on the other hand, N ≈ 105 and tcross ≈ 105 yr,
and thus
trelax ≈ 108 yr , (14.23)
which is short compared to the life time of the globular cluster; in
such cases, therefore, collisions do play a role;
14.2. THE JEANS EQUATIONS 177

14.2 The Jeans Equations

14.2.1 Moments of Boltzmann’s equation

• we thus begin again with Boltzmann’s collision-less equation, in


which the right-hand side is set to zero,

df
=0 (14.24)
dt
consider f (~x,~v, t) as a function of position, velocity and time, and
replace the time derivative of the velocity according to Newton’s
second law
F~ ~
~v˙ = = −∇Φ (14.25)
m
to obtain
∂f ~ · ∂f = 0 ;
~ f − ∇Φ
+ ~v · ∇ (14.26)
∂t ∂~v
• as several times in the course of this lecture, we now form moments
of equation (14.26) by integrating over velocity space,

∂ ∂f
Z Z Z
3 3 ~ ~
d v f + d v~v · ∇ f − ∇Φ · d3 v =0; (14.27)
∂t ∂~v

the last term here leads to boundary terms which vanish under the
assumption that there are no infinitely fast point masses,

f (~x,~v, t) → 0 for |~v| → ∞ ; (14.28)

it is a divergence in velocity space to which Gauß’ theorem can be


applied;

• also, the gradient can be pulled out of the integral in the second
term, and this yields

∂n ~
Z
+ ∇ · d3 v f~v = 0 ; (14.29)
∂t

the mean velocity is defined as


Z
1
h~vi = d3 v, f~v , (14.30)
n

and so we find the continuitiy equation for the point masses in the
form
∂n ~
+ ∇ · (nh~vi) = 0 , (14.31)
∂t
as expected;
178 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

• the second moment of Boltzmann’s equation is taken by multiply-


ing the equation with ~v prior to integration over velocity space; in
this way, we obtain, written in components

∂ 3 ∂f i j ∂Φ ∂f
Z Z Z
d v fv + d v iv v − i
3 j
d3 v i v j = 0 ; (14.32)
∂t ∂x ∂x ∂v

partial integration of the third term yields

∂f ∂v j
Z Z
d v i vj = −
3
d3 v f = −nδij ; (14.33)
∂v ∂vi

this allows us to write (14.32) as

∂(nhv j i) ∂(nhvi v j i) ∂Φ
+ +n j =0, (14.34)
∂t ∂x i ∂x
where Z
1
hv v i ≡
i j
d3 v f vi v j (14.35)
n
is the correlation matrix of the velocity components;

• we now multiply the continuity equation (14.31) with v j ,

∂n ∂(nhvi i)
vj + vj =0, (14.36)
∂t ∂xi
subtract it from (14.34) and use the continuity equation to obtain

∂hv j i j ∂(nhv i)
i
∂(nhvi v j i) ∂Φ
n − hv i + = −n j ; (14.37)
∂t ∂x i ∂x i ∂x

• the velocity-correlation matrix, hvi v j i, can be re-written to read

hvi v j i = h(vi − hvi i)(v j − hv j i)i + hvi ihv j i


≡ (σ2 )i j + hvi ihv j i , (14.38)

with which we can cast (14.37) into the form

∂hv j i ∂n ∂hvi i
n − hvi ihv j i i − nhv j i i
∂t ∂x ∂x
  ∂n ∂  
+ (σ2 )i j + hvi ihv j i + n (σ 2 ij
) + hv i
ihv j
i
∂xi ∂xi
∂Φ
= −n j (14.39)
∂x
which we can reduce to

∂hv j i ∂hv j i ∂Φ ∂  
n + nhvi i i = −n j − i (σ2 )i j n ; (14.40)
∂t ∂x ∂x ∂x
14.2. THE JEANS EQUATIONS 179

• for convenience, we now abbreviate

vi ≡ hvi i (14.41)

because only averaged velocities and no velocities of individual


particles remain, and we obtain the two equations
∂n ~
+ ∇ · (n~v) = 0 ,
∂t
∂vi ∂vi ∂Φ ∂  
+ v j j = − i − j (σ2 )i j n ; (14.42)
∂t ∂x ∂x ∂x
these are the Jeans equations which were derived for the first time
by Maxwell, but first applied to stellar-dynamical problems by
Sir James Jeans; obviously, the second equation corresponds to
Euler’s equation in ideal hydrodynamics, where the divergence of
the tensor (σ2 )i j n takes the role of the pressure gradient,

~
∇P 1 ∂P ∂  
= δij j → j (σ2 )i j n ; (14.43)
ρ ρ ∂x ∂x

14.2.2 Jeans equations in cylindrical and spherical coor-


dinates
• it is useful for many applications to write the distribution function
f as a function not of cartesian, but of curvilinear coordinates, such
as cylindrical or spherical coordinates; for instance, in cylindrical
coordinates, we first have

x = r cos φ , y = r sin φ , (14.44)


ẋ = ṙ cos φ − rφ̇ sin φ , ẏ = ṙ sin φ + rφ̇ cos φ

and thus in the plane perpendicular to the z axis

cos φ − sin φ
! !
~v = ṙ + rφ̇ = ṙ~er + rφ̇~eφ (14.45)
sin φ cos φ

as well as

ẍ = r̈ cos φ − 2ṙφ̇ sin φ − rφ̈ sin φ − rφ̇2 cos φ ,


ÿ = r̈ sin φ + 2ṙφ̇ cos φ + rφ̈ cos φ − rφ̇2 cos φ , (14.46)

and thus
~a = (r̈ − rφ̇2 )~er + (2ṙφ̇ + rφ̈)~eφ ; (14.47)
the gradient in cylindrical coordinates is

~ = ~er ∂ + ~eφ ∂ + ~ez ∂ ,


∇ (14.48)
∂r r ∂φ ∂z
180 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

and we obtain
vr = ṙ , vφ = rφ̇ , vz = ż ,
∂Φ
ar = r̈ − rφ̇2 = − ⇒
∂r
∂Φ vφ
2
∂Φ
v̇r = − + rφ̇ = −
2
+ ,
∂r ∂r r
1 ∂Φ
aφ = 2ṙφ̇ + rφ̈ = − ⇒
r ∂φ
1 ∂Φ vr vφ ∂Φ
v̇φ = ṙφ̇ + rφ̈ = − − , az = z̈ = v̇z = − ;
r ∂φ r ∂z
• this implies the collision-less Boltzmann equation in cylindrical
coordinates,
df ∂f ∂ f vφ ∂ f ∂f
= + vr + + vz
dt ∂t ∂r r ∂φ ∂z
 vφ ∂Φ  ∂ f vr vφ 1 ∂Φ ∂ f
 2 
∂Φ ∂ f
!
+  − − + −
∂r ∂vr r ∂φ ∂vφ ∂z ∂z

r r
= 0; (14.49)
in the same way, we can transform Boltzmann’s equation to sphe-
rical coordinates;
• from Boltzmann’s equation in spherical coordinates, we find after
integration over vr and under the practically important assumption
hvφ i = 0 = hvθ i (14.50)
the equation
d(nσ2r ) n h 2 i dΦ
+ 2σr − (σ2θ + σ2φ ) = −n , (14.51)
dr r dr
where σ2r,θ,φ are the velocity dispersions
Z
1
σr,θ,φ ≡
2
d3 v v2r,θ,φ ; (14.52)
n

14.2.3 Application: the mass of a galaxy


• as an application of the Jeans equations, we consider a galaxy in
spherical coordinates whose polar and azimuthal velocity dispersi-
ons are assumed to be equal,
σ2θ = σ2φ ; (14.53)
we introduce the anisotropy parameter
σ2θ σ2φ
β≡1− 2 =1− 2 , (14.54)
σr σr
which is typically β ≥ 0;
14.3. THE VIRIAL EQUATIONS 181

• the spherically-symmetric Jeans equation then reads

d(nσ2r ) 2nβ 2 dΦ
+ σr = −n ; (14.55)
dr r dr
for a spherically-symmetric system, we can write

dΦ GM(r)
= , (14.56)
dr r2
and thus we find

dσ2r 2nβ 2
" #
GM(r) r 2 dn
= vc = − σr
2
+n + σ
r2 n dr dr r r
d ln σ2r
" #
2 d ln n
= −σr + + 2β ; (14.57)
d ln r d ln r

here, vc is the orbital velocity on a circular orbit with radius r


around the centre of the galaxy; given an assumption for β, such
as β = 0, this equation allows us to determine the mass of a galaxy,
for instance if the surface-brightness profile is used as a measure
for d ln n/d ln r and the profile of the radial velocity dispersion is
measurable;

14.3 The Virial Equations

14.3.1 The tensor of potential energy

• we return to the Jeans equation in the form (14.34)

∂(nhv j i) ∂(nhvi v j i) ∂Φ
+ +n =0; (14.58)
∂t ∂x i ∂x j

multiplication with the particle mass m and the spatial coordinates


xk , followed by integration over d3 x yields

k ∂(ρhv ∂Φ ∂(ρhvi v j i)
Z j Z Z
i)
3
d xx =− d x 3
xk ; − d3 x xk ρ
∂t ∂x j ∂xi
(14.59)
the second term on the right-hand side is Chandrasekhar’s tensor
of the potential energy,

∂Φ
Z
W ij ≡ d3 x x i ρ , (14.60)
∂x j

whose trace is the system’s potential energy;


182 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

• the properties of W ij become clearer if we insert the potential


explicitly,
ρ(~x0 )
Z
Φ = −G d3 x0 , (14.61)
|~x − ~x0 |
and use the gradient

∂Φ (x j − x0 j )ρ(~x0 )
Z
=G d3 x0 ; (14.62)
∂x j |~x − ~x0 |3

then, Chandrasekhar’s tensor becomes

xi (x j − x0 j )
Z Z
W j = −G d x d3 x0 ρ(~x)ρ(~x0 )
i 3
; (14.63)
|~x − ~x0 |3

we now swap ~x and ~x0 and change the order of integrations, obtain

x0i (x j − x0 j )
Z Z
W ij = +G 3
d x d3 x0 ρ(~x)ρ(~x0 ) ; (14.64)
|~x − ~x0 |3

and add this to the previous expression (14.63) to find

(xi − x0i )(x j − x0 j )


Z Z
G
W ij =− d x 3
d3 x0 ρ(~x)ρ(~x0 ) ; (14.65)
2 |~x − ~x0 |3

first, this shows that W ij is manifestly symmetric, W ij = Wij , and its


trace is the potential energy,

3 0 ρ(~x)ρ(~x0 ) 1
Z Z Z
G
Wi = −
i 3
d x d x = d3 x ρ(~x)Φ(~x) ,
2 |~x − ~x0 | 2
(14.66)
as claimed;

14.3.2 The tensor virial theorem

• now we return to the first term on the right-hand side of the spatial
integral (14.58),

∂(ρhvi v j i) ∂(xk ρhvi v j i)


Z Z
d x 3
xk = d3 x
∂xi ∂xi
∂xk
Z
− d3 x ρhvi v j i i ; (14.67)
∂x

the first term on the right-hand side is a divergence and vanishes


upon integration over a closed system; the second is the tensor of
the kinetic energy, multiplied by −2,
Z
1
i
Kj ≡ d3 x ρhvi v j i ; (14.68)
2
14.3. THE VIRIAL EQUATIONS 183

we insert the definition (14.38) of (σ2 )ij into this tensor and obtain

1 1
K ij = T ij + Πij , (14.69)
2 2
where T ij and Πij are the tensors
Z Z
Tj ≡i
d x ρv v j , Π j ≡
3 i i
d3 x ρ(σ2 )ij , (14.70)

where the mean velocity hvi i was written as vi again;


• obviously, the tensor T ij corresponds to the stress-energy tensor in
ideal hydrodynamics up to the pressure term, while Πij describes
the momentum transport by unordered motion and thus a form of
anisotropic pressure;
• on the left-hand side of the spatial integral (14.58), the term
∂(ρhv j i)
Z
d3 x x k (14.71)
∂t
remains; we use its symmetry to write it as

k ∂(ρhv i) j ∂(ρhv i)
j k
Z " #
1 3
d x x +x (14.72)
2 ∂t ∂t
or
1∂
Z  
d3 x ρ x k v j + x j v k ; (14.73)
2 ∂t
the partial time derivative can again be replaced by a total time
derivative because the convective derivative ~v · ∇ ~ vanishes when
applied to the volume integral, and this finally yields
Z
1d  
d3 x ρ xi v j + x j vi = T ij + Πij + W ij , (14.74)
2 dt
where the symmetry of the three tensors T ij , Πij and W ij was used
again;
• from our earlier considerations on the tensor virial theorem, we
know that
d
Z   d2 I ij
d x ρ x v j + x jv = 2 ,
3 i i
(14.75)
dt dt
where I ij is the tensor of second moments of the mass distribution,
Z
i
Ij ≡ d3 x ρxi x j ; (14.76)

thus, we obtain the tensor virial theorem for collision-less systems,


d2 I ij  
= 2 T ij + Πij + W ij ; (14.77)
dt2
184 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM

• taking the trace of this equation leads us back to the ordinary


(scalar) virial theorem, if the mass distribution is static,
d2 Iii
=0 ⇒ T ii + Πii + Wii = 0 ; (14.78)
dt2
now, the trace of the sum
Z
T ii + Πii = 2Kii = d3 x ρv2 (14.79)

is twice the kinetic energy, and Wii is the total potential energy, as
we saw before; thus
2K = −W , (14.80)
which is the scalar virial theorem;

14.4 The Jeans Theorem


• an integral of motion in a gravitational field Φ is a quantity f
which satisfies
d f (~x,~v)
=0, (14.81)
dt
where ~x(t) and ~v(t) are characterising arbitrary orbits; by means of
the equation of motion, ~v˙ = −∇Φ, ~ this can be cast into the form

d f (~x,~v) ˙ ~ ∂f ~ · ∂f = 0 ;
~ f − ∇Φ
= ~x · ∇ f + ~v˙ · = ~v · ∇ (14.82)
dt ∂~v ∂~v

• in comparison to the collision-less Boltzmann equation, we see


that f is an integral of motion if and only if it is a stationary
solution of Boltzmann’s equation,
∂f
=0; (14.83)
∂t

• this leads us to Jeans’ theorem: Any stationary solution of the


collision-less Boltzmann equation depends on the phase-space
coordinates only through integrals of motion, and conversely any
function depending only on integrals of motion is a stationary
solution of the collision-less Boltzmann equation;

• the proof of the first statement has already been given; as to the
second statement, let Ii , 1 ≤ i ≤ n be n integrals of motion and
f (I1 , I2 , . . . , In ) an arbitrary function thereof; then,
df ∂ f dIi
= =0, (14.84)
dt ∂Ii dt
and f solves the collision-less Boltzmann equation;
14.4. THE JEANS THEOREM 185

• an orbit in a gravitational potential always has six integrals of mo-


tion; namely, let the orbit be specified by ~x(t) and ~v(t), then it can
be traced back to the initial point ~x0 , ~v0 by means of the equations
of motion; these six numbers can be considered as integrals of
motion;

• one distinguishes isolating and non-isolating integrals of motion;


isolating integrals such as the energy E or the angular momentum
~L constrain the orbits; an orbit passing through the point ~x0 , ~v0
in phase space is part of a subspace S n of phase space defined
by the 6 − n conditions I1 = const., . . . , I6−n = const.; an integral
I(~x,~v) is isolating for this orbit if an n-dimensional subspace S n
exists in which no point comes arbitrarily close to the hypersurface
I(~x,~v) = I(~x0 ,~v0 );

• isolating integrals are extraordinarily important, non-isolating or-


bits have no practical importance;
186 KAPITEL 14. JEANS EQUATIONS AND JEANS THEOREM
Kapitel 15

Equilibrium, Stability and


Disks
further reading: Binney, Tremai-
ne, “Galactic Dynamics”, secti-
15.1 The Isothermal Sphere ons 4.5, 4.7, 5.1 and 5.3

15.1.1 Phase-space distribution function


• spherical systems which are independent of time have orbits with at
least the four integrals of motion energy E and angular momentum
~L; the Jeans theorem then tells us that any (non-negative) function
f (E, ~L) is a stationary solution of the collision-less Boltzmann
equation and may thus represent a stable, self-gravitating system;
• the gravitational potential caused by a system with phase-space
distribution function f is determined by Poisson’s equation,
Z
~
∇ Φ = 4πGρ = 4πGm d3 v f ;
2
(15.1)

• if the system is isotropic also in velocity space, it cannot depend


on the direction of angular momentum, which implies
f (E, ~L) = f (E, L) ; (15.2)

• using the Laplacian operator in spherical symmetry,


!
~ 1 d 2d
∇ = 2
2
r , (15.3)
r dr dr
the equation for the gravitational potential
mv2
! Z !
1 d 2 dΦ
r = 4πGm d v f 3
+ mΦ, m|~x × ~v| (15.4)
r2 dr dr 2
follows as the fundamental equation for self-gravitating spherical
systems in equilibrium;

187
188 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS

• it is often convenient to re-scale the Potential Φ and the energy E


by subtracting a constant Φ0 ,

E v2
ψ ≡ −Φ + Φ0 , E ≡ − + Φ0 = ψ − ; (15.5)
m 2

• we now consider a simple example, in which f does not depend


on L,

ψ − v2 /2
!
n1 E/σ2 n1
f (E) = e = exp , (15.6)
(2πσ2 )3/2 (2πσ2 )3/2 σ2

with a normalising constant n1 ;

• integration over all velocities yields



4πn1 eψ/σ 4πn1 eψ/σ π(2σ2 )3/2
2 Z ∞ 2
2 −v2 /(2σ2 )
dv v e =
(2πσ2 )3/2 0 (2πσ2 )3/2 4
= n1 eψ/σ ≡ n ,
2
(15.7)

where n is the number density of particles;

15.1.2 Isothermality
• Poisson’s equation for this system reads
!
1 d 2 dψ
= −4πGnm = −4πGmρ1 eψ/σ ,
2

2
r (15.8)
r dr dr

or, using
n
ψ = σ2 ln = σ2 (ln n − ln n1 ) , (15.9)
n1
we find an equation for the number density n,
!
1 d 2 dn 4πG
r = − n, (15.10)
r2 dr dr σ2

which can of course also be considered as an equation for the


density ρ = nm;

• in (ideal) hydrodynamics, we had derived the equation

kT r d ln ρgas d ln T
!
M(r) = − + (15.11)
mG d ln r d ln r

for a hydrostatic, spherical system; if this is isothermal, dT/dr = 0,


we first have
d ln ρgas mG r 0 02
Z
r 2
=− dr r ρgas (r0 ) (15.12)
dr kT 0
15.1. THE ISOTHERMAL SPHERE 189

if we consider the mass as only contributed by the gas (without


any dark matter); differentiation with respect to r and division by
r2 yields
1 d 2 d ln ρgas
!
4πGm
2
r =− ρgas ; (15.13)
r dr dr kT
this equation is identical with (15.10) which we have just derived
from the Jeans theorem, if we set
kT
σ2 ≡ ; (15.14)
m
thus, the corresponding stellar-dynamical model is called the iso-
thermal sphere;

• the mean-squared velocity in the isothermal sphere is


 2
dv v4 exp 2σ
Z R −v
1 2
hv i =
2
d vv f = R
3 2
 2
n dv v2 exp 2σ
−v
2
√ 
3 π 1 −5/2

8 2σ2 3
= √  −3/2 = · 2σ = 3σ2 ; (15.15)
π 1 2
4 2σ2

any individual velocity component thus has the mean square

hv2x i = hv2y i = hv2z i = σ2 ; (15.16)

15.1.3 Singular and non-singular solutions


• one solution of the equation (15.10) for the density of the isother-
mal sphere follows from the ansatz

n = Cr−α ; (15.17)

on the left-hand side of the equation, it yields


α
− , (15.18)
r2
and thus the ansatz indeed is a solution if α = 2 and C = σ2 /(2πG),

σ2
ρ = mn = ; (15.19)
2πGr2
this solution is called the singular isothermal sphere;

• another solution which avoids the central singularity can be found


numerically; for doing so, we conveniently introduce the dimension-
less quantities
r ρ
x≡ , y≡ , (15.20)
r0 ρ0
190 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS

where ρ0 is meant to be the finite central density; then, the equation


for the scaled density y is

d 2 d ln y
!
4πG
x = − 2 ρ0 r02 yx2 ; (15.21)
dx dx σ

• if we set the scale radius r0 to


s
9σ2
r0 ≡ , (15.22)
4πGρ0

the equation simplifies to

d 2 d ln y
!
x = −9yx2 ; (15.23)
dx dx

• this equation can be integrated with the boundary conditions



dy
y(0) = 1 , =0; (15.24)
dx 0
within x . 2, i.e. r . 2r0 , the numerical result can be approximated
by
1
y(x) ≈ ; (15.25)
(1 + x2 )3/2

15.2 Equilibrium and Relaxation


• is there an equilibrium state of a self-gravitating system, which
corresponds to an entropy maximum? the entropy
Z
S ∝− d3 xd3 p p ln p (15.26)
phase space

is maximised if and only if p is the distribution function of the


isothermal sphere; however, the isothermal sphere has infinite
mass and energy and can thus not be an exact description of a
thermodynamical equilibrium state; this implies that there is no
thermodynamical equilibrium of a self-gravitating system, and that
self-gravitating systems cannot have stable final configurations,
but at best long-lived transient states!

• if we populate a narrow region in phase space with N stars, their


orbits will have slightly different initial conditions; as time pro-
ceeds, they will progressively evolve away from each other and
thus occupy a growing part of phase space; this phase mixing cau-
ses the averaged phase-space distribution f¯ to decrease, because
15.3. STABILITY 191

the averaged phase-space density is progressively diluted; thus,


the macroscopic entropy
Z
S̄ ∝ − d3 xd3 v f¯ ln f¯ (15.27)

does indeed increase;

• this process of phase mixing is in fact hardly different from the


thermodynamical trend to equilibrium; there, too, the increase of
entropy is caused by macroscopically averaging over processes
which are otherwise reversible;

• if the potential is changed while the particles are moving through


it, energy can be transported from particles to others; if, for ex-
ample, the system contracts while a star approaches its centre,
the potential deepens and the star looses energy; other stars can
gain considerable amounts of energy; this process is called violent
relaxation (Lynden-Bell);

15.3 Stability

15.3.1 Linear analysis and the Jeans swindle


• as before in hydrodynamics, we consider an equilibrium soluti-
on f0 , Φ0 of the coupled system of the collision-less Boltzmann
equation and Poisson’s equation,

∂f ~ ∂f = 0 ,
~ f − ∇Φ
+ ~v · ∇
∂t Z∂~v
~ 2 Φ = 4πGm d3 v f ;
∇ (15.28)

in equilibrium, ∂ f /∂t = 0;

• then, we perturb f0 , Φ0 by small amounts δ f , δΦ and linearise the


equations in δ f , δΦ:

∂δ f ~ 0 ∂δ f − ∇δΦ
~ f − ∇Φ ~ ∂ f0 = 0 ,
+ ~v · ∇δ
∂t ∂~v Z ∂~v
~ 2 δΦ = 4πGm d3 v δ f ;
∇ (15.29)

• as an equilibrium solution, we adopt an infinitely extended, homo-


geneous distribution f0 , which implies a density ρ0 and a potential
Φ0 given by
~ 2 Φ0 = 4πGρ0 ;
∇ (15.30)
192 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS

due to the infinite extend and the homogeneity, we must have

~ 0=0,
∇Φ (15.31)

which is possible only if ρ0 = 0; we invoke the “Jeans swindle” to


set Φ0 = 0; this is practically permissible if we study perturbations
which are small in scale compared to possible scales in ρ0 ;

• with the “Jeans swindle”, the linearised equations read

∂δ f ~ ∂ f0 = 0 ,
~ f − ∇δΦ
+ ~v · ∇δ
∂t Z ∂~v
~ 2 δΦ = 4πGm d3 v δ f ;
∇ (15.32)

again, we decompose the solution into plane waves,


~ ~
δ f = δ fv (~v)ei(k·~x−ωt) , δΦ = δΦv ei(k·~x−ωt) (15.33)

and obtain as a condition for δ f to solve the equations

∂ f0
−iωδ fv + i~v · ~kδ fv − iδΦv~k · =0
Z ∂~v
−k2 δΦv = 4πGm d3 v δ fv ; (15.34)

• the first equation yields

∂ f0 1
δ fv = δΦv~k · (15.35)
∂~v ~k · ~v − ω

which, when inserted into the second, yields

δΦv~k · ∂ f0
Z
∂~v
− k δΦv = 4πGm
2
dv 3
; (15.36)
~k · ~v − ω

since δΦv does not depend on ~v, we find

4πGm
Z ~k · ∂ f0
∂~v
1+ dv
3
=0 (15.37)
k2 ~k · ~v − ω

• this correponds exactly to the longitudinal dielectricity l from


plasma physics,

4πe2
Z ~k · ∂ f0
∂~p
l = 1 − 2 3
d p (15.38)
k ~k · ~v − ω

and thus leads to Landau damping, exactly as in plasma physics;


15.3. STABILITY 193

15.3.2 Jeans length and Jeans mass


• the boundary between stable an unstable solutions is defined by
ω = 0; if we suppose a thermal system with a Maxwellian velocity
distribution,
n0
e−v /(2σ ) ,
2 2
f0 (~v) = 2 3/2
(15.39)
(2πσ )
we have
∂ f0 ~v
= − f0 (~v) 2 ; (15.40)
∂~v σ
if ~k is chosen parallel to the positive x axis, the condition
Z −v2x /(2σ2 )
4πGmn0 3 kv x e
1− 2 2 dv (15.41)
k σ (2πσ2 )3/2 kv x − ω
follows; for ω = 0, the integral is
Z
d3 v e−vx /(2σ ) = (2πσ2 )3/2 ,
2 2
(15.42)

and we find
4πGρ0 4πGρ0
1 − 2 2 = 0 ⇒ k2 (ω = 0) ≡ kJ2 = ; (15.43)
kσ σ2
instability sets in for smaller k or wave lengths larger than λJ =
2π/kJ ; the quantity

2π 2πσ πσ
λJ ≡ = p = √ (15.44)
kJ 4πGρ0 Gρ0
is called the Jeans length;
• the Jeans length defines the volume
√ !3
πσ
λJ = √
3
(15.45)
Gρ0
and thus the mass
!1/3 !1/3
MJ GMλ2J
MJ ≈ ρ0 λJ3
⇒ λJ ≈ =
ρ0 πσ2
GM
⇒ λJ ≈ ; (15.46)
πσ2
according to the virial theorem,
GM GM
σ2 ≈ ⇒ R≈ 2 ; (15.47)
R σ
the radius of the system is thus comparable to the Jeans length;
this means that the assumption of homogeneity on the scale of
the Jeans length cannot be satisfied and that the nature of the
instability needs to be studied for each system in detail once its
geometry is specified; nonetheless, the Jeans length defines an
order of magnitude estimate for the boundary between stability
and instability;
194 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS

15.4 The rigidly rotating disk

15.4.1 Equations for the two-dimensional system


• as an example for a rotating system with flat geometry, we consider
an infinitely thin disk (with thickness zero) which is rigidly rotating
around the z axis with an angular velocity Ω ~ = Ω~ez ; the disk thus
fills the x-y plane and have a surface-mass density Σ0 ;

• we consider perturbations in the plane of the disk and neglect


warps or twists; furthermore, we transform into a co-rotating coor-
dinate frame and study the disk in the (simpler) fluid approximati-
on;

• then, the continuity equation, Euler’s and Poisson’s equations read

∂Σ ~
+ ∇ · (Σ~v) = 0 ,
∂t
∂~v ~
~ v = − ∇P − ∇Φ
+ (~v · ∇)~ ~ − 2Ω ~ × ~v + Ω
~ 2~r ,
∂t Σ
∇~ 2 Φ = 4πGΣδD (z) ; (15.48)

here, we had to take into account in Euler’s equation that Coriolis


and centrifugal forces occur in the co-rotating coordinate frame;

• the physical quantities occuring here are two-dimensional,~v(x, y, t),


Σ(x, y, t) and so on, and for the pressure we assume a barotropic
equation-of-state,

P(x, y, t) = P[Σ(x, y, t)] ; (15.49)

• the unperturbed quantities are obviously ~v = 0 and Σ = Σ0 as


well as P0 = P(Σ0 ); this trivially satisfies the continuity equation,
Euler’s equation reads

~ 0 = Ω2~r ,
∇Φ (15.50)

and Poisson’s equation is

~ 2 Φ0 = 4πGΣ0 δD (z) ;
∇ (15.51)

• since no direction can be preferred on a homogeneous disk, ∇Φ~ 0


must point along the z axis, which contradicts Euler’s equation;
thus, there is no gravitational force yet to balance the centrifu-
gal force; therefore, we assume that the disk is embedded into a
surrounding gravitational field which compensates the centrifugal
force, such as the halo of a galaxy;
15.4. THE RIGIDLY ROTATING DISK 195

15.4.2 Analysis of perturbations


• we now perturb the disk by small amounts δΣ, δ~v and so on and
linearise the equations in the perturbations; this implies
∂δΣ ~ · δ~v = 0 ,
+ Σ0 ∇
∂t
∂δ~v c2 ~ ~ ~ × δ~v ,
= − s ∇δΣ − ∇δΦ − 2Ω
∂t Σ0
~ 2 δΦ = 4πGδΣδD (z) ,
∇ (15.52)

where the sound velocity was introduced as


!
dP(Σ)
cs =
2
; (15.53)
dΣ Σ0

• as usual, we decompose the perturbations into plane waves,


~
δ~v = δ~vA ei(k·~x−ωt) ,
~
δΦ = δΦA ei(k·~x−ωt) ,
~
δΣ = δΣA ei(k·~x−ωt) (15.54)

valid at z = 0, and turn the x axis into the direction of ~k;


• we first consider Poisson’s equation; for z = 0, we have

δΦ = δΦA ei(kx−ωt) , (15.55)

while
~
∇δΦ =0 (15.56)
must hold otherwise; this is achieved by

δΦ = δΦA ei(kx−ωt)−|kz| , (15.57)

where k = k x can have either sign;


• we now integrate Poisson’s equation along the z direction from −ζ
to ζ and then take the limit ζ → 0; due to the continuity of
∂2 δΦ ∂2 δΦ
and (15.58)
∂x2 ∂y2

at z = 0, these two terms disappear from ∇~ 2 δΦ, but what remains


is
∂δΦ ζ !
Z ζ
∂2 δΦ

lim dz = lim = 4πGδΣ ; (15.59)
ζ→0 −ζ ∂z2 ζ→0 ∂z −ζ

however, we also have


∂δΦ ζ

lim = −2|k|δΦA ei(kx−ωt) , (15.60)
ζ→0 ∂z −ζ
196 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS

and so δΦA is related with δΣA through


− 2|k|δΦA = 4πGδΣA ; (15.61)
thus, we find
2πGδΣA i(kx−ωt)−|kz|
δΦ = − e ; (15.62)
|k|
• putting these equations into the continuity and Euler’s equations,
we find
iΣ0 kδvAx = iωδΣA ,
c2 2πGk
−iωδvAx = −i s kδΣA + i δΣA + 2ΩδvAy ,
Σ0 |k|
−iωδvAy = −2ΩδvAx ; (15.63)
these are three linear equations for the three variables δΣA , δvAx
and δvAy ; writing them in the form
 
  −iω  ikΣ0 0   δΣ 
A 
2  
 ik cs − 2πG −iω −2Ω   δvAx  = 0
 
Σ0 |k|
(15.64)
δvAy
   
0 2Ω −iω
makes it immediately obvious that non-trivial solutions exist if
and only if the determinant of the matrix vanishes, hence
c2s 2πG
" ! #
iω(ω − 4Ω ) + ikΣ0 ik
2 2
− (iω) = 0 ; (15.65)
Σ0 |k|
this means that either ω = 0 or
ω2 = 4Ω2 + k2 c2s − 2πG|k|Σ0 ; (15.66)
this dispersion relation describes the modes of the perturbed, ri-
gidly rotating disk; the modes are stable for ω2 ≥ 0 and unstable
for ω2 < 0;

15.4.3 Toomre’s criterion


• if Ω = 0, which is boring for a rotating disk, the disk is unstable if
2πGΣ0
|k| < kJ ≡ , (15.67)
c2s
where kJ is the Jeans wave number for the disk; if the sound speed
can be arbitrarily low, cs → 0, perturbations are unstable for
arbitrarily large k; their growth rate is then
γ = 2πGΣ0 |k| (15.68)
and the growth is exponential, ∝ eγt ; obviously, small perturbations
with |k| → ∞, λ → 0, grow particularly violently, i.e. the cold,
non-rotating disk fragments violently on small scales;
15.4. THE RIGIDLY ROTATING DISK 197

• this is not suppressed by rotation either; for cs → 0, ω2 < 0 for

2Ω2
|k| > , (15.69)
πGΣ0
i.e. even then the instability sets in on the smallest scales;

• pressure and rotation are obviously not able individually to stabili-


se the disk, since for Ω = 0, the disk is unstable for perturbations
with small, and for P = 0 = cs for large wave lengths; however,
pressure and rotation can be stabilising if they act together, since
then the dispersion relation has a minimum where
∂ h 2 2 i !
|k| cs − 2πGΣ0 |k| + 4Ω2 − ω2 = 0 , (15.70)
∂k
which yields
πGΣ0 kJ
2|k|c2s = 2πGΣ0 ⇒ |k| = = ; (15.71)
c2s 2

obviously, the disk is stable if and only if ω2 ≥ 0 at this wave


number because it is then positive for all wave numbers; thus, the
condition for global stability is

πGΣ0 (πGΣ0 )2 πGΣ0


!
4Ω − 2πGΣ0 2 +
2
= 4Ω − 2
> 0 (15.72)
cs c2s c2s
or
cs Ω π
> ≈ 1.57 ; (15.73)
GΣ0 2
this is Toomre’s criterion, which can also be applied to collision-
less systems (recall that we had adopted the fluid approximation!);
then,
cs Ω
& 1.68 (15.74)
GΣ0
is the condition for stability;
198 KAPITEL 15. EQUILIBRIUM, STABILITY AND DISKS
Kapitel 16

Dynamical Friction,
Fokker-Planck Approximation
further reading: Binney, Tremai-
ne, “Galactic Dynamics”, secti-
16.1 Dynamical Friction ons 7.1 and 8.3

16.1.1 Deflection of point masses


• an interesting effect occurs if a mass M moves through an environ-
ment of masses m which are homogeneously distributed around
the mass M; although the motion of the masses can be considered
collision-less, a deceleration occurs which is called dynamcial
friction;

• let ~vm and ~v M be the velocities of one of the masses m and of the
mass M, respectively; ~xm and ~x M are their locations; further,

~r ≡ ~xm − ~x M (16.1)

is the distance vector from M to m, and

~v ≡ ~r˙ = ~vm − ~v M (16.2)

is the relative velocity of m with respect to M; the system of two


point masses obeys the equation of motion around a fixed force
centre of a single body with the reduced mass,
 mM  GMm α
~r¨ = − 2 ~er ≡ − 2 ~er , (16.3)
m+M r r
where ~er is the unit vector in radial direction away from M;

• obviously, the change of ~v equals the difference of the changes in


~vm and ~v M ,

∆~v = ∆~vm − ∆~v M , (16.4)

199
200KAPITEL 16. DYNAMICAL FRICTION, FOKKER-PLANCK APPROXIMAT

and, furthermore, the velocity of the centre of mass is unchanged,

~ ≡ m~xm + M~x M
X ⇒ ~˙ = m~vm + M~v M ,
X (16.5)

and thus
~˙ = 0
∆X ⇒ m∆~vm + M∆~v M = 0 ; (16.6)
consequently then, ∆~vm = ∆~v + ∆~v M , and thus
m
m(∆~v + ∆~v M ) = −M∆~v M ⇒ ∆~v M = − ∆~v ; (16.7)
m+M
we shall now determine ∆~v;

• the fictitous particle with the reduced mass, Mm/(M + m), now
describes a hyperbolic orbit around the (resting) centre of force;
from the Kepler problem of celestial mechanics, we know that the
complete scattering angle is given by
θ 1
sin = , (16.8)
2 
where  is the orbit’s eccentricity;

• this implies that the cosine of (half) the scattering angle is


r r
θ 2 θ 1 1√ 2
cos = 1 − sin = 1− 2 =  −1, (16.9)
2 2  
and thus
θ 1
tan = √ ; (16.10)
2 2 − 1
• generally, the treatment of the Kepler problem shows that the ec-
centricity is related to energy E and angular momentum ~L through
s
2L2 E
 = 1+ 2 ; (16.11)
αµ

if the impact parameter is b, the angular momentum is


mM
L= bv ≡ µbv (16.12)
m+M
and the energy is

mM v2 µ 2
E= ≡ v (16.13)
m+M 2 2
because we need to insert the reduced mass
mM
µ≡ (16.14)
m+M
16.1. DYNAMICAL FRICTION 201

for the mass of the fictitous particle whose motion we study; since
α ≡ GMm = Gµ(m + M) , (16.15)
the eccentricity is
s s
2µv2 (bµv)2 b2 v 4
 = 1+ = 1 + , (16.16)
2µ[Gµ(M + m)]2 G2 (M + m)2
and thus the tangent of (half) the scattering angle is
θ 1 G(M + m)
tan = q = ; (16.17)
2 b2 v4 bv2
G2 (M+m)2

16.1.2 Velocity changes


• because of energy conservation, the velocity of the fictitous mass is
v also for t → ∞, and because of the deflection by the angle θ, the
velocity components parallel and perpendicular to the asymptotic
incoming direction are
2 tan θ/2
v⊥ = −v sin θ = −v (16.18)
1 + tan2 θ/2
2G(M + m) 1
= −v 2 h G(M+m) i2
bv 1 + bv2
G(M + m) b2 v4
= −2
bv b2 v4 + G2 (M + m)2
 #2 −1
bv3 bv2
 " 
= −2 1 + G(M + m) 
 
;
 
G(M + m) 
 

the velocity change of the mass M perpendicular to its initial


direction of motion is thus
m
∆v M⊥ = − v⊥ (16.19)
M+m
 #2 −1
2bmv3  bv2
" 
= 1+
 
;
 
G(M + m)2  G(M + m) 

 

• in parallel direction, we have


vk = v cos θ ⇒ ∆vk = v(cos θ − 1) = −v(1 − cos θ) (16.20)
or
1 − tan2 θ/2 2 tan2 θ/2
!
∆vk = −v 1 − = −v
1 + tan2 θ/2 1 + tan2 θ/2
−2v −2v
= = i ; (16.21)
1 + tan−2 θ/2 1 + bv2 2
h
G(M+m)
202KAPITEL 16. DYNAMICAL FRICTION, FOKKER-PLANCK APPROXIMAT

the velocity change in parallel direction is thus

m 2mv 1
∆v Mk = vk = h 2 i2 ; (16.22)
M+m M + m 1 + bv
G(M+m)

• if the mass M is moving through a homogeneous “sea” of masses


m, all perpendicular deflections must cancel, while the parallel ve-
locity changes must add up; therefore, the mass M will experience
a steady deceleration;

16.1.3 Chandrasekhar’s formula


• let f (~vm ) be the phase-space density of the stars with mass m which
constitute the “sea” of masses; then, the rate at which the masse M
encounters collisions with stars with an impact parameter between
b and b + db is
2πbdb v f (~vm )d3 vm ; (16.23)
these collisions change the velocity of M by

d~v M
= ~v f (~vm )d3 vm (16.24)
dt
 #2 −1
bmax
bv2
Z "
2mv  
1+
 
× db 2πb
 
m+M G(M + m) 

 
0  

where bmax is the maximum possible impact parameter, which


could be defined by the physical size of the cloud of stars m;

• the integral
Z bmax Z bmax
b b
db = db
1 + ab2
h i2
0 1+ bv2
G(M+m)
0

Z 1+ab2max
1 dβ
= , (16.25)
2a 1 β

where β ≡ 1 + ab2 was set; thus,


#2
bmax
1 G(M + m)
Z "
b
db = (16.26)
v2
i2
2
h
0 1+ bv2
G(M+m)
b2max v4
" #
× ln 1 + 2 ,
G (M + m)2

where #2
v2
"
a≡ (16.27)
G(M + m)
16.1. DYNAMICAL FRICTION 203

was used; let further


bmax v2
Λ≡ , (16.28)
G(M + m)
then the rate of change for the velocity of M by encounters with
stars of mass m and velocity ~vm is
#2
d~v M 2πmv G(M + m)
"
= ~v f (~vm )d vm
3
ln(1 + Λ2 )
dt m+M v2
= 2πG2 ln(1 + Λ2 )m(M + m)
~vm − ~v M
× f (~vm )d3 vm , (16.29)
|~vm − ~v M |3
where we have inserted ~v = ~vm − ~v M ;

• the quantity Λ is typically Λ  1, thus

ln(1 + Λ2 ) ≈ ln Λ2 = 2 ln Λ ; (16.30)

hence, we replace from now on ln(1 + Λ2 ) ≈ 2 ln Λ; typical values


for this so-called Coulomb logarithm are

5 . ln Λ . 20 ; (16.31)

• for stars which are isotropically distributed in velocity space, the


integral over velocity space is determined by
~vm − ~v M ~v M vM
Z ∞ Z
dvm vm f (vm )
2
=− 3 dvm v2m f (vm ) , (16.32)
0 v
|~ m − ~
v M |3
v M 0

in perfect analogy to the gravitational field of a collection of mass


points with mean mass density ρ;

• together, this implies Chandrasekhar’s formula for dynamical fric-


tion,
d~v M ~v M vM
Z
= −16π ln ΛG m(M + m) 3
2 2
dvm v2m f (vm ) ; (16.33)
dt vM 0

• if v M is small compared to the typical velocity of the stars m, the


remaining integral can be approximated by
Z vM
v3
dvm v2m f (vm ) ≈ M f (0) ; (16.34)
0 3
then,
d~v M 16π2G2
=− ln Λm(M + m) f (0)~v M ; (16.35)
dt 3
the friction is then proportional to ~v M like for Stokes-type friction;
204KAPITEL 16. DYNAMICAL FRICTION, FOKKER-PLANCK APPROXIMAT

• for sufficiently large v M , the integral converges to a constant, which


implies that the friction force becomes proportional to v−2 M;

• for a Maxwellian velocity distribution,


n0
e−v /(2σ ) ,
2 2
f (vm ) = 2 3/2
(16.36)
(2πσ )
the friction force becomes
d~v M 4πG2 ln Λ(M + m)ρ0
" #
2X −X2
=− erf(X) − √ e ~v M , (16.37)
dt v3M π

where X ≡ v M /( 2σ) scales the velocity v M ; here, we have set
ρ0 ≡ mn0 ;
• if M  m, (M + m) ≈ M, and the friction only depends on the
density ρ0 of the scatteres stars, but not on their mass any more,
d~v M 4πG2 ln Λρ0 M
" #
2X −X2
=− erf(X) − √ e ~v M ; (16.38)
dt v3M π
similarly, in this case the friction force os proportional to M 2
because the deceleration is proportional to M;

16.2 Fokker-Planck Approximation

16.2.1 The master equation


• so far, we have considered the collision-less Boltzmann equation,
df
=0; (16.39)
dt
in presence of collisions, it has to be augmented by collision terms
on the right-hand side,
df
= C[ f ] , (16.40)
dt
which can be written as a functional of the phase-space distribution
function f ; explicitly, this equation reads
∂f ˙ ~ ~ · ∂ f = C[ f ] ;
+ ~x · ∇ f − ∇Φ (16.41)
∂t ∂~v
• collisions transport particles from one position w~ in phase space t
~ + ∆~
another position w w; within a time interval ∆t, this may happen
with a probability
ψ(~w, ∆~w)d6 ∆~
w∆t ; (16.42)
like in hydrodynamics, we distinguish between long-ranged, “smooth”
forces short-ranged collisions;
16.2. FOKKER-PLANCK APPROXIMATION 205

~ are
• the losses at w
∂ f (~
w)
Z
= − f (~
w) d6 ∆~
w ψ(~
w, ∆~
w) , (16.43)
∂t −
while the gain is
∂ f (~
w)
Z
= − f (~
w − ∆~
w) d6 ∆~
w ψ(~
w − ∆~
w, ∆~
w) ; (16.44)
∂t +
their sum is the total change
∂ f (~
w) ∂ f (~ w) ∂ f (~
w)

= + = C[ f ] , (16.45)
∂t ∂t + ∂t −
which yields the so-called master equation
df
= (16.46)
Zdt
d6 ∆~
w ψ(~
w − ∆~
w, ∆~
w) f (~
w − ∆~
w) − ψ(~
w, ∆~
w) f (~
w) ;
 

16.2.2 The Fokker-Planck equation


• most collisions will change the velocity only by a small amount;
we had seen in the beginning that
2Gm Gm 2v bmin
δv⊥ ≈ = 2 = 2v , (16.47)
bv v b b
and thus
δv⊥ bmin
≈ ; (16.48)
v b
the relative velocity change thus decreases like b−1 while the num-
ber of collisions increases with b proportional to b2 ; thus most
collisions cause only small velocity changes;
w
• we can use that to simplify the master equation; for small δv, ∆~
is also small, and the first term under the integral in the master
equation can be expanded into a Taylor series,
ψ(~
w − ∆~ w, ∆~w) f (~w − ∆~ w) ≈ ψ(~
w, ∆~ w) f (~
w) −
∂ 
ψ(~w, ∆~ w) f (~
w) ∆w +
 i
∂wi
1 ∂2 
ψ(~
w, ∆~ w) f (~
w) ∆wi ∆w j ,

(16.49)
2 ∂w ∂w
i j

if we stop after the second order; now, we can integrate over ∆~ w,


which yields the scattering term

" Z #
C[ f ] = − i f (~ w) d ∆w ∆w ψ(~
6 i
w, ∆~ w) (16.50)
∂w
1 ∂2
" Z #
+ w) d ∆w ∆w ∆w ψ(~
f (~ 6 i j
w, ∆~w) ;
2 ∂wi ∂w j
206KAPITEL 16. DYNAMICAL FRICTION, FOKKER-PLANCK APPROXIMAT

• we introduce the diffusion coefficients in phase space,


Z
i
D(∆w ) ≡ d6 ∆w ∆wi ψ(~w, ∆~
w) (16.51)

and Z
D(∆w ∆w ) ≡
i j
d6 ∆w ∆wi ∆w j ψ(~
w, ∆~
w) (16.52)

which enable us to bring the master equation into the Fokker-


Planck form
∂f ˙ ~ ~ · ∂f ∂ h i
+ ~x · ∇ f − ∇Φ = − f (~w )D(∆w i
) (16.53)
∂t ∂~v ∂∆wi
∂2 h i
− f w
(~ )D(∆w i
∆w j
) ;
∂∆wi ∂∆w j

• the great advantage of the Fokker-Planck approach is that the


diffusion approximation in phase space depends only on the local
phase space coordinates w ~ of a test particle, such that the integro-
differential master equation turns into a pure differential equation;
the diffusion coefficients can now further be approximated by
several simplifying assumptions;

You might also like