(UNITEXT For Physics) Claudio Chiuderi, Marco Velli (Auth.) - Basics of Plasma Astrophysics-Springer-Verlag Mailand (2015) PDF

UNITEXT for Physics
Claudio Chiuderi
Marco Velli
Basics of Plasma
Astrophysics
UNITEXT for Physics
Series editors
Michele Cini, Roma, Italy
Attilio Ferrari, Torino, Italy
Stefano Forte, Milano, Italy
G. Montagna, Pavia, Italy
Oreste Nicrosini, Pavia, Italy
Luca Peliti, Napoli, Italy
Alberto Rotondi, Pavia, Italy
More information about this series at http://www.springer.com/series/13351
Claudio Chiuderi Marco Velli
•
Basics of Plasma
Astrophysics
123
Claudio Chiuderi Marco Velli
Department of Physics and Astronomy Earth Planetary and Space Sciences
University of Firenze University of California
Florence Los Angeles, CA
Italy USA
ISSN 2198-7882 ISSN 2198-7890 (electronic)

ISBN 978-88-470-5279-6 ISBN 978-88-470-5280-2 (eBook)
DOI 10.1007/978-88-470-5280-2
Library of Congress Control Number: 2014947666
Springer Milan Heidelberg New York Dordrecht London
© Springer-Verlag Italia 2015

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
Plasma physics is a discipline that has evolved in an almost explosive way in the
past 80 years. Born as a rather marginal subject, the study of the physics of ionized
gases, it had a first important application to the theory of the propagation of
electromagnetic waves in the Earth’s ionosphere. Its importance for astrophysics
had an early recognition thanks to the pioneering work of Hannes Alfvén, that
prepared the ground for many of the successive developments. A second funda-
mental field of application of plasma physics opened up when the controlled
thermonuclear fusion became one of the leading project in “big science.”
The interactions between the scientific communities engaged in the main two fields
of application, laboratory plasma physics and plasma astrophysics, very limited at
beginning, have constantly increased in time with a reciprocal benefit.
Plasmas are the main constituents of the Universe. Therefore, the interpretation
of the large amount of information provided by modern observations, both from the
ground and from space, requires a working knowledge of plasma physics. The
acquisition of such a working knowledge, however, is a slow process since the
subject is far from our everyday experience and has the disturbing tendency of
becoming complex, even in apparently simple situations. We are convinced that the
study of plasma physics must be started relatively early in the university career, as
soon as the student has mastered the basic concepts of fluid dynamics and
electromagnetic theory.
This book is an outgrowth of the lecture notes of the course taught by the authors
for a number of years at the University of Florence and an expanded version of a
book previously published in Italian. Given the large number of excellent textbook
in plasma physics already available, the obvious question is: why another one? Our
purpose in writing this book has been to address a particular audience, namely
students with a reasonable good background in mathematics and physics at the
undergraduate level who intend to orient their future activity in astrophysics. Our
ambition has been to bring the student from the very basic concepts of plasma
physics up to some of the more active research fields. We have attempted to
maintain a certain balance between mathematical rigor and physical intuition. The
possibility of following explicitly the technical details of calculations allows the
v
vi Preface
student to appreciate the power of mathematics and to acquire the tools necessary to
attack new problems on his own. Mathematics can be boring, but is reassuring. We
have also tried, whenever possible, to provide physical explanations of the phe-
nomena under discussion, in order to develop the physical insight that allows to
understand in a deep and intuitive way the inner workings of nature.
The book is divided in two unequal parts. The first seven chapters cover the
fundamental aspects of plasma physics, presented in a traditional way. Although the
main emphasis of the book is on fluid models and, in particular, on magnetohy-
drodynamics, we have attempted to show the logical path that brings us to those
models. Therefore, we have included a short outline of kinetic theory, also to
prepare the ground for more advanced topics such as the Landau damping and the
collisionless shocks. The advantages and the limits of each model are clearly
spelled out. The long chapter on instabilities contains, apart from a discussion of the
more common cases, such as the Rayleigh–Taylor, Kruskal–Shafranov, and
Kelvin–Helmholtz instabilities, the discussion of certain types of instabilities not
normally present in introductory textbooks.
The last four chapters are more demanding for the reader as they cover the most
advanced material: the properties of collisional and collisionless shocks, magnetic
reconnection, the main features of turbulence and the build-up and maintenance of
cosmical magnetic fields. All these areas are the subject of an intense research effort
worldwide and the students have some trouble finding appropriate description of
these subjects at an introductory level. We have tried to provide such an intro-
duction, hoping to bring the reader to the point where he/she will be able to
approach and understand the more specialized literature. A Section on Problems
and Questions is placed at the end of each Chapter. It contains a small number of
problems that should help the student to verify the degree of understanding
acquired. Solutions are given for all problems.
When writing this book, we have greatly benefited from numerous discussions
and suggestions from our friends and colleagues. We wish to thank them collec-
tively here.
Florence, December 2013 Claudio Chiuderi

Pasadena Marco Velli
Contents
1 An Introduction to Plasma Physics . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Saha’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The Debye Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Fundamental Plasma Parameters . . . . . . . . . . . . . . . . . . . . . . 9
1.4 The Classical Description of Plasmas . . . . . . . . . . . . . . . . . . 14
2 Particle Orbit Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1 Motion in a Uniform, Static Magnetic Field . . . . . . . . . . . . . . 17
2.2 Motion in Orthogonal Electric and Magnetic Fields. . . . . . . . . 19
2.3 Motion in Slowly Variable Magnetic Fields . . . . . . . . . . . . . . 24
2.3.1 Charged Particle Orbits in the Presence of a Magnetic
Fields with a Weak Gradient . . . . . . . . . . . . . . . . . . 24
2.3.2 Magnetic Moment Conservation . . . . . . . . . . . . . . . . 26
2.3.3 Magnetic Mirrors and Magnetic Bottles . . . . . . . . . . . 29
3 Kinetic Theory of Plasmas: An Outline . . . . . . . . . . . . . . . . . . . . 37

3.1 The Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 The Moments of the Distribution Function . . . . . . . . . . . . . . . 41
3.3 Vlasov Equation and Jeans’ Theorem . . . . . . . . . . . . . . . . . . 44
4 Fluid Models . . . . . . . . . . . . . . . ...... . . . . . . . . . . . . . . . . . . 49

4.1 The Case of Neutral Gases. . ...... . . . . . . . . . . . . . . . . . . 49
4.2 The Plasma Case: Two-Fluid Models . . . . . . . . . . . . . . . . . . 56
4.3 The One-Fluid Model . . . . . ...... . . . . . . . . . . . . . . . . . . 59
5 Magnetohydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1 MHD Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1.1 Magnetic Pressure. . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.1.2 The Conservative Form of MHD Equations . . . . . . . . 78
vii
viii Contents
5.2 The Time Evolution of Magnetic Fields. . . . . . . . . . . . . . . . . 80

5.2.1 Rm 1: Magnetic Diffusion . . . . . . . . . . . . . . . . . . 81
5.2.2 Rm 1: Alfvén’s Theorem . . . . . . . . . . . . . . . . . . . 82
5.3 Equilibrium States of Ideal Plasmas . . . . . . . . . . . . . . . . . . . 86
5.3.1 Force-Free Equilibria. . . . . . . . . . . . . . . . . . . . . . . . 87
5.3.2 Equilibria in the Presence of Magnetic Forces . . . . . . 91
5.4 Perturbed Equilibrium States . . . . . . . . . . . . . . . . . . . . . . . . 94
6 Instabilities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.1 Linear Stability of Ideal MHD Equilibria . . . . . . . . . . . . . . . . 100
6.2 Instabilities in the Presence of Gravity. . . . . . . . . . . . . . . . . . 104
6.2.1 Rayleigh-Taylor Instability . . . . . . . . . . . . . . . . . . . . 106
6.2.2 Kruskal-Shafranov Instability: B0 6¼ 0 . . . . . . . . . . . . 109
6.2.3 Parker Instability. . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.2.4 Instabilities in the Presence of Plasma Flows:
Kelvin-Helmholtz Instability. . . . . . . . . . . . . . . . . . . 122
6.3 Instabilities in Cylindrical Geometry . . . . . . . . . . . . . . . . . . . 130
6.3.1 Instability of a Plasma Column. . . . . . . . . . . . . . . . . 130
6.3.2 The Confinement of Solar Coronal Loops . . . . . . . . . 138
6.3.3 The Magnetorotational Instability (MRI) . . . . . . . . . . 143
7 Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.1 The Fourier Representation . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.1.1 Phase Velocity and Group Velocity. . . . . . . . . . . . . . 154
7.2 Waves in the Ideal MHD Regime . . . . . . . . . . . . . . . . . . . . . 155
7.2.1 Magnetic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.2.2 Magnetosonic Waves. . . . . . . . . . . . . . . . . . . . . . . . 158
7.3 Fluid Waves Beyond the MHD Regime. . . . . . . . . . . . . . . . . 161
7.3.1 Intermediate Frequencies: ω . ωce . . . . . . . . . . . . . . . 161
7.4 Waves in Kinetic Regimes: The Case of Landau Damping. . . . 167
8 Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.1 The Jump Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
8.1.1 Contact Discontinuities . . . . . . . . . . . . . . . . . . . . . . 186
8.1.2 Rotational Discontinuities . . . . . . . . . . . . . . . . . . . . 187
8.2 MHD Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
8.2.1 Perpendicular Shocks . . . . . . . . . . . . . . . . . . . . . . . 189
8.2.2 Parallel Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8.2.3 Oblique Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
8.3 Shock Thickness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
8.4 Collisionless Shocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Contents ix
9 Magnetic Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

9.1 Driven Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
9.1.1 The Sweet-Parker Model . . . . . . . . . . . . . . . . . . . . . 210
9.1.2 An Outline of the Petscheck Model. . . . . . . . . . . . . . 214
9.2 Spontaneous Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.2.1 Tearing Mode Instability . . . . . . . . . . . . . . . . . . . . . 217
9.2.2 The Plasmoid Instability of Thin Current Sheets . . . . . 224
10 MHD Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

10.1 Homogeneous and Isotropic Hydrodynamical Turbulence. . . . . 231
10.2 Magnetohydrodynamic Turbulence . . . . . . . . . . . . . . . . . . . . 235
10.3 Turbulence and Coronal Heating . . . . . . . . . . . . . . . . . . . . . 239
11 Build-Up of Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

11.1 The Dynamo Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
11.2 The Anti-Dynamo Theorems . . . . . . . . . . . . . . . . . . . . . . . . 249
11.3 Phenomenological Theory of Astrophysical Dynamos . . . . . . . 252
11.3.1 The Generation of Toroidal Fields
from Poloidal Ones . . . . . . . . . . . . . . . . . . ....... 253
11.3.2 The Generation of Poloidal Fields
from Toroidal Ones. . . . . . . . . . . . . . . . . . . . . . . . . 255
11.4 Mean-field Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . 258
11.4.1 Simple Solutions of the Dynamo Equations . . . . . . . . 263
11.5 The Creation of Magnetic Fields. . . . . . . . . . . . . . . . . . . . . . 271
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Chapter 1
An Introduction to Plasma Physics
Abstract A brief introduction to Plasma Physics is presented, focusing on ion-

ized gases, the plasma constituting the greatest part of the visible (baryonic) Uni-
verse. Basic plasma features such as ionization degree and the condition of quasi-
neutrality are discussed. A heuristic derivation of fundamental plasma parameters,
such as plasma frequency and Debye length are given. Finally, the different possible
approaches to the physical description of plasma dynamics are introduced.
Plasma physics studies the equilibrium and dynamics of globally neutral

collections of charged particles, where the interactions between particles and the
self-consistent long-range electromagnetic fields dominate over the Coulomb force
between nearest neighbors. Plasmas may be thought of as the fourth state of matter,
in the sense that they display a distinguishing set of characteristics clearly separating
them from the traditionally considered physical states of solid, liquid and gaseous
matter. Just as the solid to liquid and liquid to gas phase-transitions generally occur
in materials as heat is added, so a transition to the plasma state requires additional
energy: once the temperature becomes sufficient to ionize atoms, a gas made up of
positively charged ions and electrons is formed. Such charged particles may move
about freely, generating electric currents and charge densities which modify the elec-
tric and magnetic fields present in the system. However, though plasmas are dubbed
the fourth state of matter, the formation of the plasma state does not occur as a classi-
cal phase-transition, rather the properties of the gas change gradually with increasing
ionization. An ionized gas is just an example of a plasma, and is the most frequent
natural plasma present in the universe. Examples range from the Heliosphere, which
is the plasma cavity created by the supersonic expansion of the solar corona into the
interplanetary medium, to the magnetospheres of planets, to the magnetospheres of
neutron stars, to the jets in Active Galactic Nuclei (AGN), to the ionized gases of
the interstellar and intergalactic mediums. The heliospheric plasma, together with
the planetary magnetospheres embedded within, is the only natural plasma for which
accurate measurements of both macroscopic properties, such as velocity, temperature
and densities, as well as microscopic ones, such as the shape of the distribution func-
tions, have been carried out in situ. Approximately 95 % of matter in the Universe
© Springer-Verlag Italia 2015 1

C. Chiuderi and M. Velli, Basics of Plasma Astrophysics,
UNITEXT for Physics, DOI 10.1007/978-88-470-5280-2_1
2 1 An Introduction to Plasma Physics
is found in the plasma state. From this point of view, our environment on Earth is a
fortunate exception.
Even though one often thinks of a plasma as just a collection of charged particles,
it is really the particles and self-consistent electromagnetic fields together that define
the plasma state. A quantum mechanical point of view, where particles, fields, as well
as their excitations may be thought of as (quasi) particles, helps in understanding this
concept. In the case of plasmas these include real particles (electrons, ions, atoms),
the electromagnetic fields (photons) and the plasma waves (plasmons, phonons). The
interactions between the various plasma components, whether they be wave-particle
or wave-wave interactions or particle-particle collisions, may then all be described in
terms of generalized particle interactions. This framework will be helpful in clarifying
important plasma concepts such as collisionless damping, discussed in a subsequent
chapter, even though the description adopted in this text will be classical rather than
quantum mechanical.
According to our first rough definition of a plasma, even the conducting electrons
in a metal constitute a plasma, since the lattice ions form a fixed background and
the dynamics of the electron “gas” is governed by electromagnetic interactions. We
will not discuss such a plasma specifically in this text, which is devoted to the more
common case of ionized gases as produced in the laboratory and in nature.
What degree of ionization is required for a gas to be considered a plasma? Answer-
ing this question requires calculating the conductivity of a partially ionized gas, made
up of electrons, positive ions and neutral atoms. Given the large mass ratio of ions
and neutrals to electrons, the motions of the former may be neglected in the equation
of motion for the electrons (of charge −e and mass m e ),1 which in the presence of a
constant electric field E is:
dv
me = −e E − m e νc v
dt
or
dv e
+ νc v = − E
dt me
with v the electron velocity. The term proportional to the collision frequency νc is
the drag force coming from collisions between electrons and both ions and neutrals.
A stationary state is achieved when
e
v=− E.
m e νc
1 Throughout the text e will denote the proton charge, while a generic charge will be written as e0 .
1 An Introduction to Plasma Physics 3
Electrons carry a current density J = −en e v, and, defining the electric conductivity
σ through J = σ E, we find:
e2 n e
σ = , (1.1)
m e νc
where n e is the electron number density. One may rather generally define the collision
frequency as
v̄
νc = = nσc v̄ = αn; α = σc v̄,
λ
where v̄ is the root mean square (rms) electron speed (typically the electron ther-
mal speed), λ the mean free path, σc the collisional cross section for the interaction
considered (not to be confused with the conductivity, σ ) and n the number density
of scattering centers. In our case the collision frequency νc is the sum of a collision
frequency due to electron-ion interactions νci and one due to electron-neutral inter-
actions νca , νc = νci + νca , with corresponding collision cross sections σc (i), σc (a)
and α i , α a , so that
αa na
νc = α i n i + α a n a = α i n i (1 + ).
αi ni
In terms of the degree of ionization, χ ,
ni 1
χ= = ,
ni + na 1 + (n a /n i )
νc becomes
αa 1 − χ
νc = α i n i (1 + ),
αi χ
and recalling that n e = Z n i , where Z e is the ion charge,
Z e2 1 σmax
σ = = .
m e α 1 + (α /α )(1/χ − 1)
i a i 1 + (α /α i )(1/χ − 1)
a
Here σmax is the maximum conductivity at complete ionization of the plasma, χ = 1.

An estimate of how the ionization degree affects conductivity can be obtained by
asking for what ionization χ0 the conductivity has half its maximum value σ =
σmax /2:
αa
(1/χ0 − 1) = 1.
αi
Since
αa σc (a)
= ,
α i σc (i)
and the cross section for ion-neutral interactions is much smaller than that for ion-
electron scattering (this value is found experimentally to be about 10−2 ) we find
that an ionization degree of only about 1 % is sufficient for the plasma to have a
conductivity half of that of the completely ionized gas. With an ionization degree of
8 % the conductivity reaches a level of about 0.9 σmax .
1.1 Saha’s Equation
The degree of ionization χ depends on the physical parameters, namely density and
temperature, which characterize the thermodynamical equilibrium of the plasma.
Consider an ensemble of particles of energy E m at a temperature T. Their number
density follows from Boltzmann’s formula
n m = gm exp(−E m /kT ),
gm giving the number of states with energy E m and k is Boltzmann’s constant. The
ratio Rlm between the density of states with energies El and E m is therefore given
by :
nl gl
Rlm = = exp[−(El − E m )/kT ].
nm gm
Using i to denote the ionized state and 0 the fundamental state of a neutral atom, it
follows that ni gi
= exp[−I /kT ],
n0 g0
where I is the ionization energy. A good approximation for the number of states g
is found from quantum mechanics to be given by:
3/2
gi m e kT 1 T 3/2
2.4 × 1015 ,
g0 2π 2 ni ni
and Saha’s equation follows
T 3/2
n i /n 0 2.4 × 1015 exp(−I /kT ). (1.2)
ni
For hydrogen the ionization energy is 13.6 eV, and the ratio of ions to neutrals is
given by
T 3/2
n i /n 0 2.4 × 1015 exp(−1.58 × 103 /T ).
ni
1.1 Saha’s Equation 5
Fig. 1.1 χ as a function of T X

for n tot = 1013 cm−3 1
0.8
0.6
0.4
0.2
T (10 3 K )
7 8 9 10 11 12
Introducing the degree of ionization:

ni ni
χ= =
n0 + ni n tot
one finds from Saha’s equation that
1−χ
4.14 × 10−16 n tot T −3/2 exp(1.58 × 103 /T ).
χ2
In the preceding formulae and throughout the text the numerical values for
constants refer to the Gaussian, or mixed cgs, system of units, where all elec-
tromagnetic quantities are calculated using the electromagnetic cgs system, except
the electric charge for which the electrostatic cgs system is used. As a result, the
speed of light c and its powers often appears in formulae, but this inconvenience is
outweighed by the presence of only three fundamental units, length (cm), mass (g)
and time (s), simplifying dimensional analysis considerably. Temperatures are given
in degrees Kelvin (K) except where explicitely stated.
A plot of the degree of ionization for hydrogen as a function of temperature at a
density of n tot = 1013 cm−3 is shown in Fig. 1.1. Remarkably, the ionization of a
hydrogen gas is essentially complete (χ = 1) at temperatures of order 104 K, much
less than the temperature corresponding to the ionization potential T I /k
1.58 × 105 K.
1.2 The Debye Length
A minimal degree of ionization is not the only property required for an ionized
gas to be considered a plasma: another important characteristic is the property of
quasi-neutrality.
A charged particle in a plasma interacts with all other particles directly via the
Coulomb force, an interaction with a long distance range, as the electrostatic potential
falls off only slowly with distance—as 1/r . However, in a plasma charged particles
move freely, so one expects positively charged particles to attract negative charges
and to repel positive ones. As a result a positively charged particle in a plasma will be
surrounded by a volume with prevalently negative charge. This leads to a shielding of
the charge and to an electrostatic potential that nearly vanishes beyond some distance
rs from the charge. We therefore expect the total charge over volumes with a linear
scale greater than rs to vanish. In other words if Q(r ) is the total charge within a
sphere of radius r , quasi-neutrality implies that Q(rs ) 0.
To estimate rs , let’s start by introducing an extra charge e0 > 0 into a purely
hydrogen plasma in some point that we take as the origin, r = 0. An electric field
will appear which by Gauss’ theorem may be written as:
∇ · E = −∇ 2 Φ = 4πq = 4π e(n i − n e ) + 4π e0 δ(r),
where q is the charge density, e, as stated previously, is the proton charge and Φ is
the electrostatic potential defined by
E = −∇Φ
while δ(r) denotes the Dirac δ-function.

If the plasma is in equilibrium at a temperature T , the electron density will be
given by the Boltzmann distribution:
n e = n 0 exp[−(−eΦ)/kT ], (1.3)
where n 0 is the average density in the absence of the extra charge. Similarly, the
protons will be distributed according to
n i = n 0 exp[−(eΦ/kT )]. (1.4)
The equation for the electrostatic potential Φ therefore becomes:
∇ 2 Φ = 4π n 0 e [exp(eΦ/kT ) − exp(−eΦ/kT )] − 4π e0 δ(r)

(1.5)
= 8π n 0 e sinh(eΦ/kT ) − 4π e0 δ(r),
which, assuming eΦ/kT 1 (an approximation we will justify later), can be written
to first order as
∇ 2 Φ = 8π n 0 e (eΦ/kT ) − 4π e0 δ(r) (1.6)
Expressing the laplacian operator explicitly, assuming the space charge to be dis-
tributed isotropically, so that Φ is only a function of the distance from the origin r ,
leads to
1.2 The Debye Length 7
1 ∂ 2 ∂Φ
(r ) = 8π n 0 e(eΦ/kT ) − 4π e0 δ(r)
r 2 ∂r ∂r (1.7)
Φ
= 2 − 4π e0 δ(r),
λ
where we have defined

kT
λ= .
8π e2 n 0
To integrate, observe that the preceding equation may be rewritten as:
d2 rΦ
(r Φ) = 2 − 4π e0 r δ(r)
dr 2 λ
and, since the δ-function term on the right hand side (rhs) now vanishes identically,
exp(−r/λ)
Φ = e0 , (1.8)
r
where we have imposed that the potential become the usual Coulomb potential in a
vacuum, i.e. (n 0 → 0, λ → ∞). The Debye length, λ D , is defined as

kT √ −1/2
λD = 2
= 2λ 6.9 T 1/2 n 0 , (1.9)
4π e n 0
√
(notice the extra 2 factor with respect λ).
Figure 1.2 shows the standard vacuum Coulomb potential (dashed line), together
with the shielded potential, Eq. (1.8) (continuous line), for a case where λ D = 0.07
in (arbitrary) units of length.
Fig. 1.2 The Coulomb

(dashed) and shielded e0
potentials (continuous) as a 100
function of r, shown for
λD = 0.07
λ D = 0.07 in units of r
80
60
40
20
r
0.02 0.04 0.06 0.08 0.1
Fig. 1.3 The total charge Q

Q(r ) as a function of r 1
0.8
0.6
0.4
0.2
1 2 3 4 5 r/λ
The screening effect becomes evident calculating the total charge inside a sphere
of radius r ,
r r
Q(r ) = qd V = 4πqr 2 dr = (−∇ 2 Φ) r 2 dr
V 0 0
r Φ
=− r dr + e0 δ(r)dV
2
0 λ2 V (1.10)
e0 r
= − 2 r exp(−r/λ)dr + e0
λ 0
= e0 (1 + r/λ) exp(−r/λ).
We find that Q(r ) also falls of exponentially in units of λ and, as shown in Fig. 1.3,
it nearly vanishes beyond a distance of a few λs from the bare charge.
The use of the Boltzmann distribution to derive the densities of particles over
regions with a linear dimension of order of the Debye length is only meaningful if
the mean inter-particle distance d̄ is much smaller than that, d̄ << λ D , which also
implies, because d̄ n −1/3 , that the number of particles inside a volume of order
λ3D must be very large:
nλ3D 1. (1.11)
This inequality must be satisfied for the approximation of quasi-neutrality to hold

true, and it is therefore a necessary requirement for an ionized gas to be called a
plasma.
Going back to the ratio eΦ/kT, let’s verify its magnitude in r = d̄, still assuming
d̄ λ:
2
e2 exp(−d̄/λ) e2 1 e2 1/3 d̄
(eΦ/kT )d̄ = n 1.
kT d̄ kT d̄ kT λ
This justifies the consistency of using the approximation eΦ/kT 1 for a plasma.
1.3 Fundamental Plasma Parameters 9
1.3 Fundamental Plasma Parameters
The Debye length introduced in the previous paragraph is only one of a number of
fundamental parameters describing the physical regime (to be better defined later)
in which a plasma can be found. The rms or thermal speed of particles

vT = 3kT /m,
together with the Debye length, may be used to set a characteristic timescale τ p , by
dividing the latter with the former: λ D /vT . The plasma frequency is proportional to
1/τ p , and defined as:
4π e02 n
ωp = , (1.12)
m
where e0 , m, n are the charge, mass and density of the species in question, so that
even for a purely hydrogen plasma there will be a plasma frequency associated with
electrons, ω pe , and one with the protons ω pi . For electrons, in cgs-Gaussian units as
usual, we find
1/2
ω pe 5.64 × 104 n e .
The plasma frequency, obtained here on dimensional grounds, has an important phys-
ical meaning. Consider a plasma in which, in equilibrium, the densities of electrons
and ions are the same n e = n i = n 0 = const. If we perturb the electron density
slightly, n e = n 0 + n e with |n e | << n 0 , and assume the ions to be much heavier
than the electrons, so that they do not move on the timescale over which the electrons
react, the electron equation of motion may be written as (for simplicity, assume a 1D
perturbation of the density only along the x-axis, so that all motion occurs along that
axis):
∂v
me = −eE (1.13)
∂t
with the electric field being determined by Gauss’ law:
∂E
∇ · E = 4πq → = −4π en e
∂x
Keeping in mind that all fluctuations are small so that only first order terms are
retained in n e and v, the continuity equation for electron density may be written as:
∂n e ∂v
+ n0 = 0.
∂t ∂x
Taking the derivative of the equation of motion with respect to both time and space,
and using then Gauss’ law and the continuity equation to eliminate the electric field
and v leads to :
∂ 2 n e 4π e2 n 0
=− n e = −ω2pe n e .
∂t 2 me
The same equation holds also for v, showing that both the electron speed and density
execute harmonic oscillations with frequency ω pe . This ordered motion (as opposed
to random thermal motions) is caused by the electric field that arises as a result of
local violations to quasi-neutrality, and the plasma frequency therefore poses an upper
limit to the frequencies of dynamic motions in the plasma satisfying quasi-neutrality.
Another important characteristic frequency in a plasma is the collision frequency.
To estimate the collision frequency heuristically, consider the properties of different
particle species interacting through the Coulomb (binary) force. We begin by defin-
ing binary collisions between particles as interactions occurring when the distance
between particles decreases below a certain characteristic scale, b, given for example
by the distance at which the kinetic energy of relative motion equals the electrostatic
Coulomb potential energy. This condition is not in contrast with the previous one
requiring that the electrostatic energy be much smaller than the thermal energy when
the distance between particles is equal to the mean interparticle distance d̄, provided
b << d̄. If the colliding particles have charges Z 1 e and Z 2 e, b is determined by:
Z 1 Z 2 e2 3
kT,
b 2
The collision cross-section will therefore be given by:
4π Z 12 Z 22 e4
σc = π b2 = ,
(3k T )2
and the collision frequency:
4π Z 12 Z 22 e4 n
νc = nσc vT = , (1.14)
m 1/2 (3k T )3/2
where now m is the mass of the impacting particles while n is the density of target
particles.
Though the formulae above are only rough approximations, derived heuristically,
they retain the correct dependence on the parameters of the colliding particles. A more
rigorous treatment requires taking into account the fact that in a plasma, collisions can
not be thought of as simply binary interactions; rather, inside the Debye sphere, each
particle interacts with every other. This introduces a numerical constant multiplying
the expressions above, as well as a factor known as the Coulomb logarithm, ln Λ.
The final correct expression for electron-electron collisions is given by:
4π e4 n e
νee = 1.43 1/2
ln Λee
m e (3kTe )3/2 (1.15)
−3/2
3.75n e Te ln Λee .
The analagous expression for two identical ions of charge Z e is:
4π Z 4 e4 n i
νii = 1.43 1/2
ln Λii
m i (kTi )3/2 (1.16)
−3/2
8.76 × 10−3 n i Ti ln Λii ,
where the final numerical expression refers to the proton-proton case.

The Coulomb logarithm is given by
1 3
ln Λ = ln(4π nλ3D ) 8.33 − ln(n) + ln(T ), (1.17)
2 2
displaying only a weak logarithmic dependence on density, temperature (and particle
charge). The numerical value given above corresponds to a plasma made up of protons
and electrons.
For the general case of a plasma with ions of charge Z quasi-neutrality implies
n e = Z n i , leading to the following relation between electron-electron and ion-ion
collision frequencies:
νee m 1/2 T 3/2 1
i i
,
νii me Te Z3
where however we have approximated Λee Λii .

For the case of ion-electron collisions the result depends on which of the particles
is considered the target and which the impinging particle. In the case of electrons
colliding on ion targets we have
4π Z 2 e4 n i
νei = 2 1/2
ln Λei
m e (3kTe )3/2 (1.18)
−3/2
5.26 n i Te ln Λei ,
while for ions colliding on target electrons

24 1/2 m 1/2 T 1/2 4π Z 2 e4 n
e i e
νie = 2 1/2
ln Λie
9π mi Te m i (3kTi ) 3/2
(1.19)
T 1/2
−3/2
2.63 × 10−3
i
n e Ti .
Te
Again, the numerical values for both expressions are given for electron-proton col-
lisions.2
Collisions are also the way through which a plasma thermalizes, i.e. through which
a plasma containing particle populations with different temperatures reaches thermal
equilibrium. In a collision, energy may be transferred from the particle of higher
energy to the lower energy one. Consider then the case of populations of electrons
and ions both out of thermodynamic equilibrium but with comparable energies. It may
be shown that collisions lead to equilibrium among particles of the same species on a
different timescale compared to that required for thermal equilibrium across species.
For collision between identical particles, the characteristic timescale τ required to
reach equilibrium is given by
m 1/2
τee (νee )−1
e
τii .
mi
Thermal equilibrium across species implies collisions between electrons and ions:
in one collision an electron can only lose a fraction of order (m e /m i ) of its energy.
Reaching thermal equilibrium therefore requires a time
m 1/2 m
i i
τei τii τee .
me me
For an electron-proton plasma τei τii τee and electron-proton thermalization

requires substantially longer times than both electron-electron and proton-proton
thermalization. This means that a plasma may survive for long times with electrons
and protons at different temperatures. The timescale given by the inverse of the col-
lision frequency defines a typical time between collisions, and is the characteristic
timescale for the collisional phenomenon under consideration. In the previous para-
graphs we have focused on the timescales of energy exchange. A different timescale
is appropriate for momentum exchange. A detailed discussion may be found, for
example, in Spitzer (1962) [4].
Collision frequencies are associated with disordered motions in a plasma, while
as we have seen, motions at the plasma frequency due to departures from quasi-
neutrality are ordered motions. For this ordered motion to occur, it is necessary that
νei ω pe . From Eq. (1.14) and the definition of ω pe this condition is easily verified.
Indeed νei νee
(nλ3D )−1 1,
ω pe ω pe
from Eq. (1.11).
2 Numerical values given in different texts differ somewhat depending on the way in which the
collisions are calculated, though they are all of the same order of magnitude. The values here are
from Spitzer, “Physics of Fully Ionized Gases” [4].
Almost all natural plasmas are magnetized, in the sense that the charges move in
a domain where a non-negligible magnetic field B is present. Associated with the
magnetic field B are two further characteristic parameters, namely the cyclotron or
Larmor frequency
|Z e|B
ωc = , (1.20)
mc
and the Alfvén speed:

B B −1/2
ca = = 2.18 × 1011 B n i . (1.21)
4πρ 4π m i n i
The Alfvén speed depends only on the mass density of the plasma, while the cyclotron
frequency depends explicitely on particle mass and charge and may be written for
electrons and protons as
|e|B
ωce = 1.76 × 107 B,
mec
and me
ωcp = ωce 9.58 × 103 B.
mp
The plasmas studied in this textbook are classical plasmas, in the sense that the
plasma dynamics is described by non-relativistic classical (non-quantum) mechanics.
When does this approach lose its validity? Relativistic corrections become important
when the thermal energy is such that kT mc2 . For electrons this implies the very
extreme condition
T 6 × 109 K ,
implying that relativistic corrections are very rarely required for plasmas in thermal
equilibrium. Often however there may be particle populations in a plasma with ener-
gies much greater than kT , responsible for so called non thermal processes, and for
these species relativistic corrections may be important.
Quantum Effects are important when the characteristic length scales in the plasma
approach the De Broglie wavelength, λq = / p, where p is the particle momentum
and is the reduced Planck constant = h/2π . For thermal motions this implies

λq = ,
(3mkT )1/2
and quantum-mechanical effects become important when:
d̄ n −1/3 λq
or
2
T n −2/3 2.95 × 10−12 .
3mk
where the numerical value refers to electrons. Quantum effects therefore become
important for low temperatures and/or extremely high densities.
1.4 The Classical Description of Plasmas
The classical, microscopic, description of a plasma considers the plasma as an ensem-

ble of charged particles in a vacuum. Maxwell’s equations are therefore written in
the form
1 ∂B
∇×E =−
c ∂t
1 ∂E 4π
∇×B = + J (1.22)
c ∂t c
∇ · E = 4πq
∇·B =0
where q and J are the total charge and current densities, including possible external
charges and currents. Taking the divergence of the second equation in (1.22) and the
time-derivative of Gauss’ theorem, third equation in (1.22), we obtain the equation
of charge conservation or charge continuity:
∂q
∇· J+ = 0.
∂t
Gauss’ theorem for both the electric and magnetic fields (third and fourth equation
in 1.22) in reality imposes only initial conditions on the dynamics:
1 ∂
0 = ∇ · (∇ × E) = − ∇·B
c ∂t
so that if ∇ · B = 0 is satisfied at any one time it will be satisfied for all subsequent
times. Analogously, taking the time derivative of the equation for ∇ · E and using
conservation of charge we obtain:
∂
(∇ · E − 4π q) = c ∇ · (∇ × B) = 0.
∂t
1.4 The Classical Description of Plasmas 15
This means that Gauss’ theorems for the electric and magnetic fields should not be
included when considering the number of independent equations describing plasma
dynamics, though of course they are often used in applications to specific problems.
In addition, every particle i = 1 . . . N (including ions and electrons) follows
Newton’s equations of motion:
1
m i rï = ei (E + r˙i × B), (1.23)
c
where ei is the charge of the ith particle, r i (t) and ṙ i (t) are the ith particle’s position
vector and velocity—also written v i (t), respectively. At least in principle, complete
knowledge of the microscopic plasma state for an ensemble of N particles therefore
requires solving the 3N equations of motion together with the 6 independent Maxwell
equations (1.22), for a total of 3N +6 equations in the 3N +10 unknowns E, B, j , q.
The missing 4 equations are those determining charge and current densities from the
motions of individual particles:

N
q(r, t) = ei δ[r − r i (t)]
i=1
(1.24)

N
J(r, t) = ei v i (t) δ[r − r i (t)].
i=1
Apart from the complexity of solving such a huge nonlinear system of equa-
tions (clearly impossible in practice), these equations provide at least in principle a
scheme for recovering ab initio the microscopic state of a plasma. However, even if
the solution of such a system were attainable, the information content of the micro-
scopic state would be far too great with respect to any possible experimental or
observational verification. A step forward would be to move from this description
to a statistical one, the best compromise between accuracy and manageability. This
transition corresponds to a loss of detailed but often not essential information, such
as, for example, the trajectories of individual particles. Even a statistical approach
is in many cases too detailed, and one is content with a fluid-type closure, where in
addition to trajectories all information on the detailed distribution of the velocities of
particles in phase space is also lost in favor of a finite number of velocity moments,
typically up to third order. The correct choice of the level of description depends
on the nature of the problem and the type of information required to understand the
dynamics. Each simplification has a cost in the sense that some “resolving power” is
lost, and much of the “art” in solving plasma physics problems resides in the proper
choice for understanding the question at hand.
There is at least one case where the complete description from Eqs. (1.22)–(1.24) is
used, provided the electromagnetic fields in (1.22) are determined by external charges
and currents. In this case the 3N particle motion equations become decoupled, and
we are discussing single particle orbits in given electromagnetic fields. This is called
“Orbit Theory” and belongs only marginally to plasma physics if, as we have said,
by plasmas we mean a system of collectively interacting particles and fields. An
understanding of particle orbit theory is however a pre-requisite for developing a

kinetic plasma theory and gaining a deeper understanding of macroscopic plasma
properties.
Questions and Problems
1.1. Prove that in a head-on collision of an electron with an ion the electron loses an
amount of energy of order m e /m i in favor of the ion.
1.2. Prove that in a plasma permeated by a sufficiently strong electric field a regime of
continuous or runaway acceleration of electrons develops and determine the critical
minimum value for such a field, known as the Dreicer field.
Solutions
1.1. Assume the ion to be originally at rest and call ve and ve the electron speeds
before and after the collision, vi the ion speed after the collision. Write the equations
of conservation for momentum and energy:
m e ve = m e ve + m i vi ,
= 21 m e ve + 21 m i vi .
2 2
2 m e ve
1 2
Solving for ve , vi gives
me − mi 2 me
ve = ve ; vi = ve ,
me + mi me + mi
and therefore
m i vi 2 m e /m i me
=4 4 .
m e ve2 1 + (m e /m i )2 mi
1.2. A runaway regime of continuous acceleration can develop when the electric force
on an electron overcomes the “viscous” drag force due to collisions, in magnitude
e E > m e νei v. Using expression (1.17) for νei and starting from a typical speed v
equal to an electron thermal speed, the result is:
8π Z 2 e3 n i ni
E > ED = 2 × 10−7 V olt/m. (Z = 1)
3kTe Te
Chapter 2
Particle Orbit Theory
Abstract In which particle motion in assigned electric and magnetic fields is

discussed and studied in detail in a few representative cases. Fundamental con-
cepts such as guiding center motion, particle drifts, magnetic moment, and adiabatic
invariants are introduced, and a brief discussion of the mirror force and magnetic
bottles is presented.
This chapter is devoted to a description of the motion of individual charged

particles in assigned magnetic and electric fields, which we assume to be given
as a function of time and position. Such orbits can become quite complex even in
this case, so gaining a general understanding of the main types of motions that may
occur is an important pre-requisite to studying the dynamics of plasmas in their self-
consistent electromagnetic fields. Let’s begin with the simplest case, the motion of
a charged particle in a constant, uniform, magnetic field.
2.1 Motion in a Uniform, Static Magnetic Field
Consider a magnetic field which is constant in space and time, so that with respect
to a Cartesian system of coordinates the field reads B = (0, 0, B). The classical
equation of motion for a particle of mass m and charge e0 in such a field may be
written as
dv e0
m = m r̈ = v × B. (2.1)
dt c
Because the force on the particle is perpendicular to its velocity,
z̈ = 0
and motion along the field occurs at constant speed, ż = v is constant. Another
immediate consequence is found taking the scalar product of the equation of motion
with the velocity ṙ:
m r̈ · ṙ = 0,

18 2 Particle Orbit Theory
implying that the total kinetic energy energy

1
2 m ṙ 2 = W = W + W⊥
is constant. However, since W = 21 mv2 is separately conserved, we find that the

kinetic energy of motion perpendicular to the field W⊥ is also constant.
The trajectory of the particle may be found by integrating the equation of motion,
Eq. (2.1), which rewritten in its three cartesian components reads:
dvx
= Ω vy (2.2a)
dt
dv y
= −Ω vx (2.2b)
dt
dvz
=0 (2.2c)
dt
where
e0 B
Ω= .
mc
The solutions to Eq. (2.2a)–(2.2c) are given by:
vx = v⊥ cos(Ωt + α) (2.3a)
v y = −v⊥ sin(Ωt + α) (2.3b)
vz = v . (2.3c)
Integrating Eq. (2.3a)–(2.3c) once more we obtain the trajectory as a function of time:
v⊥
x = x0 + sin(Ωt + α) (2.4a)
Ω
v⊥
y = y0 + cos(Ωt + α) (2.4b)
Ω
z = z 0 + v t. (2.4c)
The trajectory is a helix which positively charged particles sweep in the clockwise
sense and negative particles in a counter-clockwise sense as seen projected on the
(x, y) plane. The projected orbits are circular with radius R L = v⊥ /|Ω|, known as
the Larmor radius. The gyration frequency |Ω| is known as the cyclotron frequency,
also called the gyrofrequency.
The frequencies ωce and ωci defined in Chap. 1 are the values of |Ω| calculated for
an electron and proton respectively. Let’s proceed now to motion in combined electric
and magnetic fields, starting from the case in which the fields are perpendicular
(Figs. 2.1 and 2.2).
2.2 Motion in Orthogonal Electric and Magnetic Fields 19
Fig. 2.1 Trajectory of a

negatively charged particle in
a uniform magnetic field B
2.2 Motion in Orthogonal Electric and Magnetic Fields
Starting from the same configuration as above, we add a constant, homogeneous

electric field E orthogonal to B:
E = (0, E, 0),
B = (0, 0, B),
with E and B independent of position and time.

To solve Eq. (2.1) now we start from the y component of the equation of motion,
Eq. (2.3c) which becomes:
dv y e0 E E
= −Ω vx + = −Ω(vx − c ). (2.5)
dt m B
It is natural to consider therefore a frame of reference S , moving at the speed c E/B

along the x direction, which is possible as long as E B: in this case the Lorentz
transformation reduces to the Galilean transformation
x = x − c (E/B)t,
y = y,
z = z,
t = t,
Fig. 2.2 Motion in z

orthogonal and constant
electric and magnetic fields
E
y
x
and in the frame S the equation of motion along y becomes
dv y
= −Ω vx ,
dt
identical to Eq. (2.2c). The trajectory written in the new, primed coordinates therefore
coincides with that given by Eq. (2.4a) where y is replaced by y . Returning to the
original frame of reference we can see that the trajectory is just a superposition of the
original helical trajectory and a uniform translation at speed cE/B along the x-axis
as shown in (Fig. 2.2). To interpret this result consider that the effect of the electric
field in the y-direction is to accelerate (or decelerate, depending on the charge) a
particle along that axis. In the presence of a magnetic field, the increase (decrease)
in speed along that direction leads to an increasing (decreasing) centripetal force
deflecting the motion orthogonally to both E and B. This force gradually changes
the direction of motion eventually leading to a change of sign of v y , so that motion
in the y-direction is confined to a finite displacement. On the other hand the different
radius of curvature of the gyrating motion as the particle moves along and against the
direction of the electric field leads to the average uniform displacement orthogonal
to the fields. The velocity associated with this motion is known as the drift velocity
v E and may be written in a coordinate independent way as
E×B
vE = c . (2.6)
B2
When E/B = O(1) relativistic corrections become important, and a different
approach is required if E > B, since vx in this case would become:
E
vx = c + v0x cos(Ωt),
B
which exceeds the speed of light. With E > B one must therefore use the relativistic
equations of motion,
dp v
= e0 E + × B
dt c
p = γ mv
1
γ = .
1 − v 2 /c2
Consider the effects of a Lorentz transformation along the positive x-axis with
speed v on electric and magnetic fields:
E x = E x Bx = Bx (2.7a)
E y = γ (E y − β Bz ) B y = γ (B y + β E z ) (2.7b)
E z = γ (E z + β B y ) Bz = γ (Bz − β E y ), (2.7c)
where β = v/c.
For the configuration considered we have E x = 0, E z = 0, E y = γ (E − β B).
Therefore, for E < B, the choice β = EB makes the y component of the electric field
in the S frame vanish, and only a magnetic field along z = z, Bz = γ (B − β E) =
γ (B − E 2 /B) = B/γ will remain, yielding the helical motion seen before. In the
original frame S the motion will therefore be a superposition of the helical motion
and the translational one. More generally, which kind of Lorentz transformation may
be used to simplify the equations of motion may be understood by recalling the two
Lorentz invariants for the electromagnetic field, E · B and E 2 − B 2 . If the first of
these vanishes, then it will always be possible to find a frame in which either E or B
vanishes, depending on the sign of the second invariant E 2 − B 2 . So if the electric
and magnetic fields are perpendicular, but E > B, a new frame can be found with a
vanishing magnetic field, corresponding to a Lorentz boost along the x direction with
β = B/E. Once the equation of motion is solved in the new frame, one can transform
the solution back to the original frame applying the inverse Lorentz transformation
with β = −B/E along the x axis (see Problem 2.4).
A very important property of the drift speed v E is that it does not depend on the
sign of the charge of the particle in motion. Both electrons and protons move with
the same speed under this drift, and therefore no current is generated by the so called
“E cross B” drift within the plasma. The above discussion may be generalized to the
case of any constant force F orthogonal to the magnetic field B: it will give rise to a
drift which is orthogonal both to the force and the magnetic field, whose magnitude
is obtained by substituting E with F/e0 in Eq. (2.6):
F×B
vF = c (2.8)
e0 B 2
As a consequence, the drift from a non-electric force will give rise to an average
current in the plasma.
So far we have only considered cases in which E = B. However, the case E = B
has particular importance since it is related to the motion of a particle in the field of
an electromagnetic wave. The method of a Lorentz boost, used to treat the cases with
E = B, is no longer applicable and we must solve directly the equation of motion
of special relativity in the laboratory frame. The more familiar non relativistic case
can be easily obtained by taking the appropriate limit.
We shall consider a particle of mass m and charge e0 , initially at rest at the
origin of the coordinates, acted upon by a linearly polarized electromagnetic wave
of frequency ω. We may then choose a frame of reference in which the electric field
lies on the x-axis and the magnetic field on the y-axis, so that the waves propagates
along the z-direction. Thus,
E = (E 0 cos φ) ex , B = (E 0 cos φ) e y ,
with φ = ω(t − z/c).

The equations of relativistic dynamics are
d (mγ v)
= e0 (E + β × B), (2.9)
dt
d (mγ c)
= e0 E · β (2.10)
dt
where β = (v/c). Introducing the quantity ω E = (e0 E 0 /mc), Eqs. (2.9) and (2.10)
give
d
(γβx ) = ω E (1 − βz ) cos φ, (2.11)
dt
d
(γβ y 0) = 0, (2.12)
dt
d
(γβz ) = ω E βx cos φ, (2.13)
dt
dγ
= ω E βx cos φ. (2.14)
dt
Equating Eqs. (2.13) and (2.14) we get
d
[γ (1 − βz )] = 0
dt
and therefore γ (1 − βz ) = K is a constant of motion.

To determine the trajectory of the particle, it is convenient to use the variable φ
in place of t. Then
1 dr 1 dr dφ ω dr ω
β= = = (1 − βz ) = (1 − βz )r ,
c dt c dφ dt c dφ c
where the prime indicates the derivative with respect to φ. Therefore
d d ω ω2 K
(γ β) = [γ (1 − βz )r ] = (1 − βz )r .
dt dt c c
Comparing the last expression with Eq. (2.9) we finally get the following set of
differential equations:
c ωE
x = cos φ
ωK ω
y = 0
1 ωE
z = x cos φ.
K ω
Since r(0) = 0 and ṙ(0) = 0 (the dot indicates the time-derivative), we have
φ(0) = 0; φ̇(0) = ω[1 − βz (0)] = ω; K = K (0) = 1,
and the initial conditions for the system above are:
x(0) = 0, x (0) = ẋ(0)/φ̇(0) = 0 ; y(0) = y (0) = 0 ; z(0) = z (0) = 0.
The second equation shows that y = 0: no motion occurs in the y-direction.

A straightforward integration of the other two equations gives
ω E c
x =− (1 − cos φ), (2.15)
ω ω
ω E 2 c
z= (φ − 1
sin 2φ) = ξ(φ − 1
sin 2φ),
2ω ω 2 2
with ω E 2 c
ξ= .
2ω ω
Inserting the definition of φ into the the expression for z we have
z = ωξ((t − z/c) − (ξ/2) sin 2φ,
or
ωξ ξ
z= t− sin 2φ. (2.16)
1 + ωξ/c 2(1 + ωξ/c)
The velocity along z is therefore a superposition of a constant speed plus a periodic

term. The constant term corresponds to a value of βz
ωξ/c (ω E /2ω)2
βz = = ,
1 + ωξ/c 1 + (ω E /2ω)2
which tends to unity when (ω E /ω) is very large, namely when E 0 is very large or
ω is very small. The first circumstance arises in the field of an intense laser beam
or in linear acclerators. The second one has been considered in connection with
the problem of the acceleration of cosmic rays in the vicinity of a pulsar. In fact,
the pulsar emits a low frequency electromagnetic wave, which in principle could
accelerate particles to very large energies. This model, however, presents a number
of drawbacks and has been abandoned.
The periodic term in Eq. (2.16), coupled with the motion along x, Eq. (2.15),
produces an eight-shaped trajectory in the (x, z) plane of a frame moving along z
with a speed cβz .
2.3 Motion in Slowly Variable Magnetic Fields
Because electric and magnetic fields in space are never uniform or constant in time,
there are in general no exact integrals of motion to which one can resort to simplify the
understanding of charged particle dynamics. However, when particle motion occurs
in slowly variable magnetic fields, either in time or in space, the equation of particle
motion, Eq. (2.1), may be solved approximately. In fact, the systematic expansion
in terms of drift velocities of higher order is allowed if the the ratio of the Larmor
radius R L to the gradient scale L R L /L 1, or alternatively, if the characteristic
frequency of time variation of the field 1/T is much smaller than the gyrofrequency
|Ω|, i.e. |Ω|T 1. When these conditions are satisfied, exact integrals of motion are
replaced by approximate integrals or adiabatic invariants, quantities whose relative
changes are bounded by inequalities more stringent than the ones above.
2.3.1 Charged Particle Orbits in the Presence of a Magnetic Fields

with a Weak Gradient
Equation (2.8) may be used to understand particle motion by use of a local analysis
in a magnetic field with gradients. Consider for example a magnetic field line having
a certain curvature. The particle will locally gyrate around the magnetic field, which
in view of the inequalities expressed above may be considered constant on the scale
of the Larmor radius. However, the particle generally also has a motion parallel
to the field B, and, because of the curvature of the field line, the particle will feel
in its own frame of reference an outward centrifugal force due to this motion F =
(m v2 Rc )/(Rc2 ), where Rc is the local radius of curvature. Equation (2.8) then shows
that as a consequence of curvature a drift motion will arise with velocity
mc v2 (Rc × B) 2 c W
vC = = ( R̂c × B̂), (2.17)
e0 Rc2 B 2 e0 Rc B
where B̂ is the unit vector tangent to the local magnetic field, while R̂c is unit
vector along the radius of curvature. The factor in parentheses is therefore a purely
geometrical term which describes the local geometry of magnetic field lines. This
motion of charged particles is called the curvature drift.
Consider now instead a magnetic field directed along one direction, whose mag-
nitude depends on a coordinate along an axis orthogonal to the magnetic field B
itself. An example is given by a field B = [0, 0, B(y)], where B(y) satisfies the
weak gradient condition
dB/dy
R L 1.
B
2.3 Motion in Slowly Variable Magnetic Fields 25
The Lorentz force around a given point y0 is given by:

e0
F(y) = F(y0 + δy) = B −vx e y + v y ex y +δy ,
c 0
where we are considering motions only in a plane perpendicular to the magnetic

field B.
Denoting y-derivatives with a prime, the field in the neighborhood of this point may
be written B(y0 + δy) = B(y0 ) + δy B (y0 ) = B0 + δy B0 so that
e0 B0
F(y) = −vx e y + v y ex
c
e0 B0
+ −vx e y + v y ex δy (B0 /B0 ). (2.18)
c
For δy of the same order as R L one has by assumption that δy (B0 /B0 ) 1.
Therefore the first term in the equation describes the unperturbed circular motion
while the second term provides the first order correction. We can thus assume that
vx , v y , δy = y − y0 are the same as those obtained from the motion in a constant
magnetic field, which are periodic functions of the phase Φ = Ωt + α. Averaging
the force F over a gyration period, P = 2π/|Ω|, that is calculating
P
1
F = F dt,
P
0
we see that the only non zero contribution comes from the term vx δy B0 on the right
hand side of Eq. (2.18). As a result
P
e0 1 e0 v⊥ 2 B0
F = − v⊥ R L cos2 (Φ) dt B0 e y = − R L B0 e y = − 21 mv⊥ ey .
c P 2c B0
0
In other words, on averaging over a gyroperiod, we are left with a constant force
perpendicular to B 0 which, according to Eq. (2.8), must give rise to a drift orthogonal
to both B 0 and F . Substituting the expression for F in Eq. (2.8) we find:
c W⊥ B0
vG = ex ,
e0 B02
whose general vector expression may be written as:
c W⊥ (B 0 × ∇)|B 0 |
vG = . (2.19)
e0 |B 0 |3
This is known as the gradient drift.

When, as in the preceding discussion, the gradients are small with respect to the
cyclotron radius and the fastest time-scale is given by the gyro-motion one calls the
averaged position of the particle over one gyroperiod the guiding center of the particle.
In terms of our previous discussion of drifts, one could say the particle executes a
gyromotion around its guiding center while the guiding center itself moves following
the drifts in question. In order for the so-called guiding center approximation to be
valid, an additional condition must be established concerning the parallel motion: the
distance a particle moves along the field during one gyroperiod must also be small
respect to the scale of gradients, or v P L.
In these circumstances we may always write for the equations of motion along
and across the magnetic field the equations:
dv
m = F (2.20)
dt
dv ⊥ e0
m = F⊥ + v⊥ × B .
dt c
We may then write v ⊥ = v Ω + v d where v Ω is the velocity of the cyclotron

motion around the guiding center and orthogonal to the local magnetic field while
v d is the drift velocity. We now expand
v d = v 0d + v 1d + v 2d + · · ·
and expand order by order in the small gradient parameter the equations for
the drifts:
dv 0d e0 1
m + ··· = v + · · · × B, (2.21)
dt c d
so that, for example, considering that the time-dependence of the zero-order drift is
a small quantity,
mc dv 0d B
v 1d = − ×
e0 B dt B
and so on and so forth. It is important to remark that quite generally the first order
drift may differ in direction from the zero order drift, and must therefore be included
to understand charged particle trajectories.
2.3.2 Magnetic Moment Conservation
Consider a magnetic field configuration with cylindrical symmetry in which the

magnetic field has a strong axial component and again satisfies the conditions for
weak gradients defined above.
The field may be written in cylindrical coordinates r, θ, z as
B = [Br (r, z), 0, Bz (r, z)]; Bz Br → B = (Br2 + Bz2 )1/2 Bz .
On the other hand the divergence free condition for the magnetic field ∇ · B = 0
may be written as:
1 ∂ ∂ Bz
(r Br ) + = 0,
r ∂r ∂z
which may be integrated to yield
r r
∂ ∂ Bz ∂ B r2
(r Br )dr = − r dr − .
∂r ∂z ∂z 2
0 0
Here we have taken into account the fact that r is of the same order as the Larmor
radius R L and that the magnetic field and the magnetic field gradients vary on a much
larger scale. We therefore obtain:
r ∂B
Br − ,
2 ∂z
where we have now omitted the brackets . The equation of motion of a particle in
the direction parallel to the magnetic field becomes:
dv e0 e0
m = − (v × B)z = − vθ Br .
dt c c
Remembering that the sense of particle gyration is opposite in sign to the charge we
may write
e0
vθ = − v⊥ ,
|e0 |
and therefore
dv e0 e 0 r ∂ B |e |
0 RL ∂ B |e | π R 2 ∂ B
0
=− − v⊥ − − R L |Ω| =− L
,
dt c |e0 | 2 ∂z c 2 ∂z c P ∂z
where as before the gyroperiod P = 2π/|Ω|.

The periodic gyromotion of a charged particle e0 in the plane orthogonal to the
magnetic field is electrically equivalent to a current carrying loop with I = |e0 |/P.
According to Ampère’s law, such a loop corresponds to a magnetic dipole with
magnetic moment μ, given by:
IS |e0 | π R 2L 1 W⊥
μ= = = mv 2 = . (2.22)
c c P 2B ⊥ B
We therefore get, along B, the equation:
dv ∂B
m = −μ ,
dt ∂z
which, multiplying by v leads to
dW dv ∂B dB
= mv = −μv = −μ ,
dt dt ∂z dt
where we have used

d ∂
= + v · ∇.
dt ∂t
On the other hand, since the magnetic field does no work on the particles and there
is no explicit time-dependence of the field itself, the total energy of the particle is
conserved:
dW⊥ dW dB
=− =μ ,
dt dt dt
an expression which may also be found by writing the equation for perpendicular
motion. We may now calculate the time- derivative of μ as follows:
dμ d 1 dW⊥ 1 dB μ dB 1 dB
= (W⊥ /B) = − 2 W⊥ = − 2 (μ B) = 0.
dt dt B dt B dt B dt B dt
The magnetic moment is therefore a conserved quantity, provided gradient length-
scales of the average field are small compared to the particle gyro-radii: this is what
is meant by an adiabatic invariant. In the particle’s frame of reference, the magnetic
field changes, albeight slowly, with time, because of the motion of the particle. We
can therefore expect that conservation of magnetic moment also holds for slow time-
variations of the magnetic field.
Consider then such a slowly varying field, directed along the axis in a cylindrical
coordinate system: B = [0, 0, B(t)], whose typical time variation T is much longer
than the gyroperiod 2π/|Ω|. The time-variation induces an azimuthal electric field
E which appears in the perpendicular components of the equation of motion. Taking
the scalar product of the equation of motion with v⊥ we find:
dv⊥
m · v⊥ = e0 E · v⊥ ,
dt
or
d 1
( m v⊥
2
) = e0 E · v⊥ .
dt 2
Integrating this over the gyroperiod, from time 0 to P = 2π/|Ω|, the left hand side
gives the variation of the kinetic energy of perpendicular motion over one gyroperiod
W⊥ :
P

W⊥ = e0 E · v⊥ dt = e0 E · dr ⊥ = e0 (∇ × E) · dS,
0 S
where dr ⊥ = v⊥ dt while dS is an oriented element of the surface upon which the

cyclotron motion orbit rests.
From Maxwell’s equations, we may rewrite:

∂B
W⊥ = −e0 · dS π R 2L |e0 | Ḃ,
∂t
S
where we have been able to bring < ∂ B/∂t >= Ḃ out of the integral because of the
hypothesis of slow variations of the magnetic field and we have taken into account
the sign coming from the orientation of particle orbits of different sign.
Substituting R L = |v⊥ |/|Ω| we obtain:
2π Ḃ
W⊥ = W⊥ .
|Ω| B
On the other hand, (2π/|Ω|) Ḃ is the change of B over a gyroperiod, i.e. B, so we
can write:
B
W⊥ = W⊥ ,
B
or, recalling Eq. (2.22),
(W⊥ /B) = (μ) = 0, (2.23)
Therefore, we have shown that the magnetic moment is an adiabatic invariant

even in the case of slowly time-varying magnetic fields.
2.3.3 Magnetic Mirrors and Magnetic Bottles
The concept of the conservation of magnetic moment finds application in the study of
magnetic mirrors and magnetic bottles as plasma confinement devices. Let’s consider
again a magnetic field, for example with axial symmetry, which increases in intensity
along the positive direction of the axial direction z as shown in Fig. 2.3.
Since the magnetic field magnitude B increases with z, the constancy of μ also
implies that the kinetic energy in perpendicular motion W⊥ must increase. Con-
servation of energy, W = W + W⊥ = constant, then means that any increase
Fig. 2.3 Trajectory in a

spatially varying magnetic
field
B
in W⊥ must be accompanied by a decrease in W . Thus, it may happen that for

some sufficiently large value of B = B R , W vanishes. In such case, the charged
particle, which cannot continue its trajectory along positive z, is reflected. This kind
of configuration is called a magnetic mirror.
Introducing the pitch angle ϑ, defined as the angle formed by the velocity vector
v with the axial direction z, that is with the direction of the dominant magnetic field
component B, we find v⊥ = v sin ϑ so that the magnetic moment invariance may be
written in the form:
sin2 ϑ sin2 ϑ
1
2 mv
2
= constant i.e. = constant
B B
because of the conservation of total energy 21 mv 2 . By evaluating the constants at the

reflection point we obtain:
B
sin2 ϑ = . (2.24)
BR
Now if the field has a maximum somewhere, Bmax , obviously B R < Bmax must hold:
indeed, if the particle is able to reach the position where B = Bmax with v = 0 it
will continue in the same direction, as B decreases beyond the point B = Bmax and
therefore W increases at the expense of W⊥ . From Eq. (2.24) it follows that we can
write the condition for reflection as
B B
sin2 ϑ = .
BR Bmax
Particles that start off with a pitch angle sin2 ϑ < B/Bmax will not be reflected by
the magnetic mirror.
If we now consider the illustrated magnetic field structure, comprised of two
separate magnetic mirrors encompassing the same axial field, we see that it is possible
Fig. 2.4 A magnetic bottle

configuration
to confine particles with sufficiently large pitch angles in the region between the two
mirrors (Fig. 2.4).
Denoting with B = B0 the lowest value of the magnetic field, the particles that
will be confined will be those with
B0
sin2 ϑ0 = 1/R,
Bmax
the quantity R = Bmax /B0 being called the mirror ratio. The other particles, those
with angles falling witihn a cone defined by ϑ0 , known as the loss cone, will not be
confined. The probability P of losing a particle depends on the ratio between the
solid angles swept by the loss cone and 2π , i.e.
ϑ0
P= 1
2π sin ϑ dϑ dϕ = sin ϑ dϑ = 1 − 1 − 1/R 1/2R for R 1.
Ω0 0
In a real situation, the loss process will never stop, because collisions between charged
particles will continually feed particles into the loss cone. For this reason, the mag-
netic bottle is not an efficient confinement device with regards to possible future
nuclear fusion machines.
A magnetic bottle configuration in which the intensity of the magnetic field varies
slowly with time possesses another adiabatic invariant, the so-called longitudinal
adiabatic invariant J, defined as
s2
J= v ds,
s1
where s1 and s2 are the reflection points for the parallel motion whose position is
assumed to change on time-scales much slower than the transit time inside the bottle.
One may show (see, e.g. Boyd and Sanderson: The Physics of Plasmas, p. 28) that
J is also an adiabatic invariant.
An interesting and important application of the existence of this invariant is con-

nected to understanding the acceleration of cosmic rays. A simple model may be
developed as follows: an ensemble of charged particles finds itself in an environ-
ment where magnetic clouds are present and whose behaviour may be schematized
as being equivalent to moving magnetic mirrors. In the regions between clouds, the
magnetic field has values much smaller than in the clouds, so that particles which start
out with a sufficiently large pitch angle are confined in the region between clouds.
If s0 is a measure of the distance between clouds at a certain time t and s0 that at a
successive moment in time t , the invariance of J implies that
s0
v s0 v s0 or v v ,
s0
where v and v are the average velocities respectively at times t an t . It follows
that the energy W is equal to W = W (s0 /s0 )2 and increases if clouds approach
each other while it decreases if they increase their separation. Since W⊥ = μ B =
constant
(μ is a constant because it is also an adiabatic invariant and B does not change much as
the clouds and particles move around), we see that the total energy W = W +W⊥ also
increases as clouds approach and vice-versa. Since the relative motion of clouds must
be considered a random variable, one might think that on average energy gains and
losses might balance out. This is wrong however, since over a given interval of time
the number of head on collisions between clouds is greater than the number of over-
taking ones. Their relative frequencies are, in fact, proportional to (v +vC )/(v −vC ),
where vC is the speed of the cloud. The net effect is therefore a gain of energy. This
acceleration mechanism for cosmic rays was proposed by Enrico Fermi in 1949 and
is known as the second order Fermi mechanism. It is not very efficient because of the
particles that are scattered into the loss cone. Today it is thought that a much more
efficient mechanism is given by the so-called first order Fermi mechanism where
particles are accelerated as they repeatedly cross magnetic shock waves.
Problems
2.1. Consider a magnetic field configuration with a slight curvature: B ≡ (0, B y (z),
B0z ), and B0z constant while B y and dB y /dz are small quantities. Show that there
is a drift along the x direction with speed
mv2 c
vc = − 2
(dB y /dz).
e0 B0z
and show that this coincides with what one finds from Eq. (2.17).
2.2. Consider the motion of a proton and an electron at the surface of the Earth in
the presence of a horizontal uniform magnetic field with an intensity of B0 0.3 G.
Calculate the drift speed and directions knowing that the magnetic field is directed
from South to North. Assuming that the electron density n e is equal to that of protons
n p calculate thecurrent J resuting from this drift. Finally, show that
J
× B + (n e m e + n p m p )g = 0,
c
where g is the gravitational acceleration vector at the surface of the Earth.
This example shows how, in the presence of gravity, a horizontal magnetic
field induces a current in the plasma whose effect is to suspend the plasma in
the magnetic field, cancelling out gravity. In reality, for a plasma volume of
finite extent, there is a separation of charge induced at the lateral boundaries
by the current J. Show that the generated polarization electric field provides a
downward E cross B drift motion, whose time derivative leads the plasma to
fall through the magnetic field with acceleration g. This points to the fundamen-
tal role of boundary conditions in establishing the correct behavior of a plasma.
2.3. The Earth’s magnetic field may be considered with good approximation to
be a dipole field, whose intensity in the magnetic equatorial plane is given by
B = B0 (R E /R)3 where B0 = 0.3 G, R E is the radius of the Earth and R the
Earth’s radius. Show that because of the gradient drift in the dipole field, Eq. (2.19),
a particle with pitch angle θ = 90◦ carries out a circular orbit around the Earth and
find an expression for this orbit in terms of the Energy of the particle E and its charge
q. Calculate the period of the orbit for a proton and an electron with an energy of
1 keV at a distance of R = 2R E . Compare this drift speed with the one due to the
gravitational field directly and with the gravitational orbital speed at the same height.
2.4. Consider the motion of a charged particle in the presence of orthogonal uniform
electric (along the y-direction) and magnetic (along the z-direction) fields where the
electric field is greater than the magnetic field E > B. Solve the problem by moving
to a frame moving along the x direction with β = B/E. Show that in this frame the
magnetic field vanishes and solve the equation of motion explicitely. Move back to
the original frame of reference and describe the particle orbit in this frame.
Solutions
2.1. Calling Ω = e0 B0z /mc e Ω y = e0 B y /mc, the equations of motion become:
ẍ = Ω ẏ − Ω y ż,
ÿ = −Ω ẋ,
z̈ = Ω y ẋ.
Taking the time derivative and neglecting terms that are quadratic in small quantities
we find:
...
x + Ω 2 ẋ = −Ω y v2 ,
...
y + Ω 2 ẏ = ΩΩ y v .
A particular (non oscillating solution) of this equation is given by
Ω y mv2 c
ẋ = − =− (dB y /dz),
Ω2 2
e0 B0z
providing the requested drift speed.

From the definition of the radius of curvature
∂B RC
=− 2 ,
∂s RC
and taking into acocunt that ∂ B/∂s (dB y /dz)e y , while B Bz ez , one finds that
the above result is equivalent to that found using Eq. (2.17).
2.2. The drift speed is given by Eq. (2.8) where the gravitational acceleration is
orthogonal to the magnetic field, so that
cm p,e g
|v p,e | = = 1.86 × 10−4 cm/s, (electron) = 3.42 × 10−1 cm/s, (proton).
eB
The velocity is towards the East (West) for a proton (electron). The current from this
drift is also directed Eastwards and has a magnitude | J| = cn p (m p + m e )g/B.
2.3. The gradient drift for this case reduces to
cE 1 cE 3R 2
|vG | = |d B/dr | =
q B2 q B0 R 3E
directed towards the West IEast) for protons (electrons). For a particle with the
elctron or proton charge and an energy of 1 keV at a height of R = 2R E we find
|vG | = 6.27 × 106 cm/s.
2.4. In the reference frame moving with vx = cB/E one finds from Eq. (2.7) that
Bz = (Bz − β E y ) = 0 while E y = (E y − β Bz ) = E/ where here β = B/E

and = 1/ 1 − B 2 /E 2 . In the new, primed frame, the relativistic equation of
motions are given by [see Eqs. 2.9) and (2.10)]
d(γ βx ) d(γ βz ) d(γ β y ) dγ

= = 0; = ωE ; = 0,
dt dt dt dt
with ωE = (e0 E /mc) = Ω E / . If the particle starts from rest at the origine
βx = βz = 0, β = β y and γ = (1 − β 2 )−1/2 . From the remaining two equations
we get:
dγ dβ dβ
β + γ = ωE = ωE β 2 + γ ,
dt dt dt
or
dβ dβ ωE
ωE (1 − β 2 ) = γ =⇒ = .
dt dt γ 3
Integrating the last equation we obtain
τ
β = √ =⇒ γ 2 = 1 + τ 2 ,
1 + τ 2
where τ = ωE t . Writing
dy dy
cτ
v y = = ω = cβ
= √ ,
dt E
dτ 1 + τ 2
we see that the particle’s speed increases in the y -direction, i.e. along E , approaching
c when t → ∞.
Integrating the above equation we finally find
c
y = 1 + τ 2 − 1 ,
ωE
which, coupled with x = z = 0, gives the trajectory of the particle in the primed
frame. Transforming back to the original frame by means of a Lorentz transformation
vith V = −B/E, we easily find
c ω E 2 t 2 1/2
x = V t; y= 1+ − 1 .
ωE 4
Chapter 3
Kinetic Theory of Plasmas: An Outline
Abstract A short summary of the kinetic theory of plasmas is presented as a basis

for the development of fluid models. The general kinetic equation is derived and
followed by a discussion of the hierarchy of moment equations and the fluid clo-
sure problem. The Vlasov equation for collisionless plasmas is presented and Jeans’
theorem concerning its solution is proved.
As already remarked in the introductory chapter, when the number of particles

becomes large, a statistical approach to the description of plasma dynamics plasma
provides a good compromise between accuracy and tractability. A complete dis-
cussion of plasma kinetic theory is definitely beyond the scope of this book: we
limit ourselves here to a brief outline of the basic elements of the theory, necessary
prerequisites to the derivation of the fluid models subject of the next chapter.
3.1 The Distribution Function
The complete description of the state of a single particle in classical mechanics

requires knowledge of its position and velocity at all times. Thus, the natural frame-
work to deal with particle dynamics is a six-dimensional space defined by three
spatial coordinates (x, y, z) and three velocity coordinates (vx , v y , vz ) known as
phase space. At any given time t, the dynamical state of a particle is represented by
a point in phase space. The ensemble of all representative points of the particles thus
describes the state of the entire system. The first step towards a statistical approach
is the definition of a density function in phase space, telling us which is the aver-
age number of particles present, at any given time, in an infinitesimal cell of such a
space. This density, generally a function of the six phase space coordinates and of
time, is called the distribution function, f (r, v, t). The average number of particles
with positions lying between r and r + d r and velocities between v and v + dv, will
then be given by:
dN = f (r, v, t) dr dv. (3.1)

38 3 Kinetic Theory of Plasmas: An Outline
When the number of particles is sufficiently large, the distribution function can be
considered a continuous function of its variables and the total number of particles N
can be computed as
N= f (r, v, t) dr dv,
V6
where V6 is the phase space volume. To derive the equation describing the dynamical
evolution of the distribution function, consider the situation in which the total number
N of particles does not change during the evolution of the system. This implies
that we are neglecting, for instance, all ionization or recombination processes, that,
however, could be easily reintroduced at a later stage. Whenever in physics we have a
conserved quantity (N in our case) it is possible to write down a continuity equation.
For example, in non-relativistic mechanics,
mass is a conserved quantity. The mass
contained in a volume V is M = ρ d V and therefore
V

dM ∂ρ
= dV + ρ (v · n)dS = 0,
dt ∂t
V S
where the first integral gives the variation of the mass inside V and the second
represents the mass flux through the surface delimiting V , n being the unit vector
normal to the surface element dS, pointing towards the exterior.1 By transforming
the surface integral into a volume integral, the preceding equation becomes:

∂ρ
+ ∇ · (ρv) dV = 0.
∂t
V
Since V is an arbitrary volume, we necessarily have
∂ρ
+ ∇ · (ρv) = 0.
∂t
The same procedure can be adopted for any conserved quantity, in particular
for N . In this case, however, we must modify the definition of the operator ∇ to take
into account the dependence of the distribution function on the velocities. This can
be achieved by generalizing

3
∂
3
∂
∇ → + = ∇ + ∇v . (3.2)
∂ xi ∂vi
i=1 i=1
1 from now on n will be referred to as the external normal.

3.1 The Distribution Function 39
The conservation law for the number of particles can now be expressed as
∂f
+ ∇ · ( f v) + ∇ v · ( f a) = 0, (3.3)
∂t
where a is the acceleration calculated in the corresponding elementary cell in phase
space. The preceding equation can be simplified by noting that
• In phase space r and v are independent variables. Thus, ∇ · v = 0 and the second
term of (4.30) simplifies into v · ∇ f ,
• The term f ∇ v · a = f ∇ v · (F/m) = 0. In fact, the forces normally coming into
play are velocity independent, with the notable exception of the Lorentz force,
F = (e0 /c)(v × B). However, in this case too we have
e0 ∂(v × B)i
∇v · F = =0
c ∂vi
i
since the i-th component of v × B does not depend on vi .

Keeping all this in mind Eq. (4.30) becomes:
∂f F
+ v · ∇ f + · ∇ v f = 0. (3.4)
∂t m
So far we have implicitly assumed that all the particles occupying the same cell
of phase space are subject to the same acceleration. This is certainly true as far as
collective effects are concerned, i.e. those due to the simultaneous action of all the
particles of the plasma, but may not be true for collisions, namely for interactions
involving only two particles at a time. This suggests separating the collective from
collisional forces. Note that the phase space trajectories generated by these two
types of forces are also very different from one another: collective effects give rise
to forces that are weakly position-dependent and generate smooth trajectories, while
collisions produce abrupt variations of the speed of the interacting particles at the
collision location. We shall therefore write:
F = F sv + F coll ,
where the index sv stands for “slowly varying” and we shall distinguish the collisional
contribution by moving it to the right hand side of the Eq. (3.4). Introducing the
notation: ∂ f
F coll
− · ∇v f = ,
m ∂t coll
we obtain the final form of the equation describing the dynamical evolution of the
distribution function, the so-called kinetic equation,
∂f F ∂ f
+ v · ∇ f + · ∇v f = , (3.5)
∂t m ∂t coll
where it is understood that the forces F appearing on the lhs are only those due to
the collective effects.
Everything said so far is equally valid for a neutral gas and for a plasma. In the
latter case the force term can be written as
1
F = e0 (E + v × B) + f ,
c
where f are the forces of non-electromagnetic origin acting on the plasma, the most
common additional force in astrophysics being gravitational. Equation (9.7) has to
be coupled with the set of Maxwell’s equations (1.22), that determine the electric and
magnetic fields, and with the other equations entering the definition of f . Neglecting
for a moment non-electromagnetic forces write:
∂f e0 1 ∂ f
+ v · ∇ f + (E + v × B) · ∇ v f = . (3.6)
∂t m c ∂t coll
At this point we must specify the nature of the collisions that determine the r.h.s of
the Eqs. (9.7) and (3.6). For each type of collision, or better, for each collision model,
we shall obtain a different kinetic equation. If, for instance, we consider a rarefied
plasma dominated by collective effects, we may neglect collisions and equate the rhs
of Eq. (9.7) to zero, leading to the Vlasov equation:
∂f e0 1
+ v · ∇ f + (E + v × B) · ∇ v f = 0. (3.7)
∂t m c
If we adopt the Boltzmann collision model (binary elastic collisions) we will get
the Boltzmann equation, where the collisional term is given in terms of an integral
that involves the product of two distribution functions. The Boltzmann equation is
particularly important for neutral gases, where binary collisions are dominant, but it
is not the most appropriate to describe plasmas. In fact, as we have already remarked,
inside the Debye sphere every particle interacts with many other particles at the same
time and the deflection suffered by the particle is the result of many small deviations
rather than of a single interaction. The collisional term must therefore be suitably
modified and the resulting equation is called the Fokker-Planck equation.
3.2 The Moments of the Distribution Function 41
3.2 The Moments of the Distribution Function
The analytical solution of the kinetic equations is generally impossible, even in the
simplest case of the Vlasov equation. Unfortunately, their numerical solution also
presents enormous difficulties because f depends on a high number of variables
(seven, in the general case). It is not surprising therefore that the few analytical solu-
tions and the majority of the numerical solutions concern configurations showing
some symmetry property, with the consequent decrease of the number of indepen-
dent variables. Another cause of complexity is due to the fact that in a plasma are
present at least two types of charged particles of opposite signs, typically electrons
and positive ions. As a consequence, we have to deal with a number of kinetic equa-
tions equal to the number of species present. Finally, observe that the linear character
of the lhs of the kinetic equations with respect to f is guaranteed only if the fields E
and B are external, namely if their origin is not tied to the dynamical behaviour of the
plasma particles. In all other cases, the kinetic equations have to be complemented
with Maxwell’s equations, that couple the electromagnetic fields to the distributions
of charges and currents which, in turn, are expressed in terms of the motions of the
charged particles forming the plasma, described by means of the respective distribu-
tion functions. The fields will therefore depend on the distribution functions and the
kinetic equations will have a nonlinear character.
For the sake of discussion, let’s assume that we have managed to solve one of the
kinetic equations, obtaining an explicit analytical form for the distribution function.
Is there any way we can verify that our solution does indeed correspond to some phys-
ically realistic situation? In general, the answer is negative: the distribution function
does not appear to be a directly observable quantity. To determine f experimentally
we should measure the number of particles in a given volume having a well defined
value of the velocity and repeat this measure for all values of the speed and for all
directions. This type of measurement is possible, although not very precisely, for a
limited number of directions and for very rarefied plasmas, conditions that are hard to
realize in a laboratory. In interplanetary space such conditions are indeed fulfilled and
many spacecraft have succeeded in determining the shape of distribution functions
for the plasma particles that flow from the Sun (mostly for electrons, protons, and
some for alpha particles) and fill all of interplanetary space, the so-called solar wind.
These data convincingly demonstrate that the space external to the Earth should be
considered as an extraordinary cosmical laboratory, where experiments unfeasible
in a terrestrial lab are actually performed. If the distribution function cannot be mea-
sured, why is it useful? May we use it to determine other quantities that are more
susceptible to a direct measurement? To answer these questions, we must first ask
ourselves if we really need all the information contained in f . Once again, we are
facing a situation where we must decide whether we can give up as redundant certain
types of information. For instance, if we judge that the essential information concerns
the behavior of our system in ordinary space and not in the whole phase space, we
may decide to give up the information concerning the distribution of the velocities,
by introducing some sort of averaging process over the velocities themselves. As an
example, let’s consider the integral

f (r, v, t)dv ≡ f (r, vv , v y , vz ; t)dvx dv y dvz . (3.8)
Taking into account the definition of f , Eq. (3.1), this integral defines the average
number of particles whose position vector lies between r and r + d r, regardless of
their velocity, or, in other words the numerical density of particles in ordinary space,
that we shall call n(r, t)
n(r, t) = f (r, v, t)dv. (3.9)
In the same way, the average value of any other velocity-dependent quantity Φ(v)
can be defined as

Φ(v) f (r, v, t)dv 1
Φ = = Φ(v) f (r, v, t)dv. (3.10)
f (r, v, t)dv n(r, t)
The factor 1/n appearing in the definition of average values is due to the fact that
f does nor represent the probability of finding a particle in the given phase space
cell, but the average number of particles within the cell. In other words, f is not
normalized to unity, but to the total number of particles, N . Clearly, all average values
are functions of r and t. A special class of functions Φ is particularly interesting:
those that are multilinear products of components of the velocity v, namely Φ =
vi v j . . . vk . They are connected to the definition of the n-th order moment of f
as the integral of the velocity distribution function times the product of n velocity
components:

n-th order moment = (vi v j . . . vk ) f (r, v, t)dv, (3.11)
where the product of components entering the integrand has n factors. Recalling the
definition of an average value, Eq. (3.10), it is obvious that:

(vi v j . . . vk ) f (r, v, t)dv = n(r, t)(vi v j . . . vk ) (3.12)
Therefore, the numerical density is a moment of order zero (in this case Φ = 1).
The average value of the velocity, u(r, t), will be given by

1
u(r, t) = v f (r, v, t)dv. (3.13)
n(r, t)
The components of the average velocity are therefore connected with the first-order
moments of the distribution function. Notice that u(r, t) represents the average speed
3.2 The Moments of the Distribution Function 43
of the particles located in the “point r” and not the velocity of a single (microscopic)
particle, a quantity that has been indicated by v.
In a similar way, it is possible to define the charge density, q(r, t), and the cur-
rent density j (r, t). In fact, if f e and f i are, respectively, the electrons’ and ions’
distribution functions, that for simplicity we shall assume to be singly ionized, we
have:

q(r, t) = e( f i − f e )dv, and j (r, t) = e v( f i − f e )dv. (3.14)
These examples suggest that the moments of the distribution function have a well
defined physical meaning and that they represent quantities that can be measured.
However, as we have already stressed, the averaging process causes a loss of infor-
mation, that related to the behavior of the particles in velocity space. This loss is not
essential in all the cases where the velocities of the single particles are not substan-
tially different from the average speed. If, however, a subset of particles exists which
behaves in a peculiar way with respect to the others, either because of a clumping in
velocity space, such as a particle beam, or because there is a region in velocity space
where particles can become resonant with the oscillations of the electromagnetic
field, the potential effect of these particles might disappear in the averaging process,
with the risk of overlooking important physical effects. We shall return to this point
later on, when discussing the phenomenon known as Landau damping.
As the distribution function evolves in time, the moments of the distribution
functions must also evolve, and their evolution equation, called the general moment
equation is obtained by taking moments of the kinetic equation. In other words,
multiply (3.5) by ψ(v) = vi , v j . . . vk and integrate over all velocities. The first
term, recalling Eq. (3.12), becomes:

∂f ∂ ∂
ψ(v) dv = ψ f dv = (nψ),
∂t ∂t ∂t
since ψ is time independent.

Analogously, since ψ is also independent of r , the second term becomes:

ψ v · ∇ f dv = ∇ · f v ψdv = ∇ · (nv ψ).
In the third term, involving F/m, separate the velocity-independent forces from the
Lorentz force and write (see comments after Eq. (3.3)) F = g + (e0 /c)v × B. For
the first group of forces we have:

1 g
ψ(v)(g · ∇ v f )dv = · ψ (∇ v f )dv
m m

g n
= · ∇ v (ψ f ) − f (∇ v ψ) dv = − g · ∇ v ψ,
m m
where we have used the fact that the fist term in square brackets vanishes if
lim (ψ f ) = 0.
|v|→∞
Acting in the same way on the term (e0 /c)v × B and recalling that the i-th component
of v × B does not contain vi we get:

e0 n e0
ψ(v) (v × B) · (∇ v f )dv = − (v × B) · ∇ v ψ.
mc mc
Collecting the preceding results, we obtain the general moment equation:
∂ n n e0
(nψ) + ∇ · (nv ψ) − g · ∇ v ψ − (v × B) · ∇ v ψ =
∂t m m c
∂ f
= ψ(v) dv. (3.15)
∂t coll
The form of the collisional term on the rhs depends on the properties of particle
collisions, i.e. the collisional model adopted. However, remembering that (∂ f /∂t)coll
represents the temporal variation of f due to collisions, we will assume it may be
written in a similar way to the term ∂ f /∂t representing the temporal variation of f
due to collective forces:
∂ f ∂
ψ(v) dv = (nψ) .
∂t coll ∂t coll
Equation 3.15 with this last form for the collisional term becomes the basis for the
development of fluid plasma models, which are the subject of the next chapter.
3.3 Vlasov Equation and Jeans’ Theorem
While in the case of neutral gases it seldom happens that collisions are negligible,
the Vlasov equation (3.7) has many applications in plasma physics, and in some
sense provides the simplest model for the kinetic behavior of a plasma. The reason
is that while in a system of neutral particles collective effects due to average long
range scale self-consistent forces are generally unimportant (with some important
exceptions discussed further below), they are fundamental to plasma dynamics.
In spite of its simplicity, the solution of the Vlasov equation generally constitutes
a difficult problem, because of the non-linearity arising from the coupling of the
electromagnetic fields to the particles, the latter being sources for the fields which
determine their motion. An extremely important simplification occurs when the fields
E and B entering Eq. (3.7) are external fields. In this case the Vlasov equation
becomes linear in the distribution function f , and a general solution can be found
3.3 Vlasov Equation and Jeans’ Theorem 45
when the system admits one or more constants of motion, a result known as Jeans’
theorem.
To prove Jean’s theorem, let’s introduce the concept of derivative along a curve
and, for the sake of simplicity, let’s consider the case of ordinary geometrical space.
Let g = g(x, y) be the equation of a surface. The differential of g, namely the
variation of g for arbitrary variations of x and y, is given by:
∂g ∂g
dg = dx + dy.
∂x ∂y
Let’s now assume that the variations of x and y are no longer arbitrary, but that
the representative point in the (x, y) plane is constrained to move along the curve
y = y(x), so that dy = (dy/dx) dx. Then
∂g ∂g dy
dg = dx + dx.
∂x ∂ y dx
The quantity
dg ∂g ∂g dy
= + , (3.16)
dx ∂x ∂ y dx
is called the derivative of g along the curve y = y(x).

Let’s now apply this definition to the time derivative of f , a function of (r, v, t),
along the trajectory of the representative point in phase space. The trajectory is given
in parametric form by the functions
r = r(c j , t), v = v(c j , t) j = 1 . . . 6, (3.17)
where the quantities c j are the six integration constants necessary to completely
define the trajectory. By inverting the system of Eq. (3.17), we obtain
c j = c j (r, v, t), j = 1 . . . 6. (3.18)
Any constant of motion can be obviously expressed in terms of the c j and therefore
in terms of the primary variables r, v using the (3.18). Recalling the definition of
derivative along a curve, we write:
df ∂f ∂ f dxi ∂ f dvi
3 3
= + + .
dt ∂t ∂ xi dt ∂vi dt
i=1 i=1
Since dxi /dt = vi and dvi /dt = Fi /m, the preceding equation becomes:
df ∂f F
= + v · ∇ f + · ∇ v f.
dt ∂t m
Therefore, if f is a solution of Vlasov equation we may conclude that (d f /dt) = 0,

i.e. that the distribution function remains constant along a trajectory in phase space.
Let’s now check that the general solution of Vlasov equation can be expressed as
f (r, v, t) = F(c1 , c2 . . . c6 ),
where F is an arbitrary function of its variables. We thus have:
∂F F ∂F ∂c j F
+ v · ∇F + · ∇ v F = + v · ∇c j + · ∇ v c j
∂t m ∂c j ∂t m
j
∂F dc j
= = 0,
∂c j dt trajectory
j
since, by definition, the quantities c j are constants on the trajectory. This proves
Jeans’ theorem: Any function of the constants of motion is a solution of the Vlasov
equation
Of course, the explicit form of the solution is obtained by expressing the c j in terms
of the primary physical variables r and v using Eq. (3.18), once it has been verified
that the solution obtained has the necessary convergence properties for v → ∞.
A simple application of Jeans’ theorem is given by the motion of a particle of
mass m and charge e0 , subject to the action of an electrostatic electric field, whose
potential is Φ(r). The total energy of the particle,
1
E= mv 2 + e0 Φ(r),
2
is a constant of motion and therefore any function of the energy is a solution of Vlasov
equation. In particular, if for r → ∞, where the potential Φ is assumed to vanish,
the system is in thermodynamical equilibrium, with a density n 0 and a temperature
T0 , the distribution function will asymptotically tend to a maxwellian:
m 3/2 1 m 3/2
f0 = n0 exp −( mv 2 /kT0 ) = n 0 exp(−E/kT0 ).
2π kT0 2 2π kT0
Jeans’ theorem now allows us to write down the solution at a generic point as:
m 3/2 1
f (r, v) = n 0 exp −( mv 2 + e0 Φ(r))/kT0 .
2π kT0 2
This form of the distribution function has been already used (without justification)
in the discussion of the Debye length, in Sect. 1.3 of Chap. 1.
The Vlasov equation can sometimes also be used for systems composed of neutral
particles, the most interesting case being that of stellar systems. A system composed
by a very large number of stars, like a galaxy, could be consider as a sort of “gas”,
3.3 Vlasov Equation and Jeans’ Theorem 47
whose “molecules” are represented by the stars. The collective effects are those
produced by the gravitational interactions of the ensemble of stars, the direct colli-
sions between being in general negligible. The equation describing the dynamical
evolution of the system is again Eq. (3.4), where F/m now represent the gravity
acceleration produced at the given point by the collective action of all the stars of the
system. Actually, the description of a stellar system by means of Eq. (3.4) preceded
the formulation of the Vlasov equation for plasmas (1938) [31], thanks to the works
of Jeans (1915) [14] and Chandrasekhar (1942) [15].
We thus see that a very deep analogy exists between the dynamics of a system
of mass points and that of a system of charged particles. The law expressing the
gravitational interaction between two mass points (Newton) is identical to the one for
the electrostatic interaction between two charges (Coulomb), while the Coriolis force
in a rotating reference system is perfectly analogous to the Lorentz force exerted on a
particle by a magnetic field. There are also deep differences, mostly stemming from
the fact that the gravitational interaction is always attractive, while the electrostatic
one can be either attractive or repulsive. As a consequence, in “gravitational plasmas”
the screening effect that gives rise to the concept of Debye length does not exist.
3.1. In interplanetary space distribution functions are often observed, both for protons
and electrons, known as bi-maxwellians, given by:
m 3/2 mv 2 m 3/2 mv 2
f B0 = n c exp − + nh exp − . (3.19)
2π kTc 2kTc 2π kTh 2kTh
n c + n h = n 0 is the total density of particles: n c , Tc are the parameters defining the

“cold” dominant part of the distribution, while n h , Th refer to a population of “hot”
particles of low density but with a much higher value of the temperature. Show that
the temperature of the combined distribution, defined by the second-order moment

m
kT = v 2 f B0 dv,
3n 0
is given by:
nc nh
T0 = Tc + Th .
n0 n0
Using Jeans’ theorem verify, assuming spherical symmetry, with r distance from sun
center, r0 the distance at which density is measured, and a static atmosphere, that the
total density n(r ) varies with distance from the Sun as:
Gm M r0 Gm M r0

n(r ) = n c exp − 1− + n h exp − 1− .
r0 kTc r r0 kTh r
Then show that the temperature varies with heliocentric distance as:
n c Tc Gm M r0 n h Th Gm M r0

T (r ) = exp − 1− + exp − 1− ,
n(r ) r0 kTc r n(r ) r0 kTh r
where M is the solar mass.

3.2. The visible surface of the Sun, the photosphere, has a much lower temperature
than the solar corona and solar wind. Because the sources of solar energy, nuclear
reactions at the center of the Sun, are below the photosphere, the temperature inver-
sion seems to violate the second law of thermodynamics. A possible solution to
this “coronal heating” problem consists in assuming that bi-maxwellian distribution
functions are present already within the solar atmosphere at heights corresponding to
the transition zone between photosphere and corona, with a dominant component at
a temperature Tc 104 K and a minor component say with a density n h /n 0 10−3
and Th 106 K. According to the formula of the preceding problem, the tempera-
ture at a distance of r = r0 7 105 km, corresponding to the transition zone, turns
out to be T0 1.1 × 104 K. Taking first into account only the protons, m = m p ,
show that the temperature at a distance of only r = 1.01 r0 has already risen to
a value of T 106 K. The electrons, because of their lower mass, reach such a
temperature only at a much greater distance: how much greater? In the absence of
collisions, which are the effects that may prevent the formation of free charges inside
the coronal plasma and the solar wind?
Chapter 4
Fluid Models
Abstract The formal derivation of fluid models is presented, starting from the hier-
archy of moment equations stemming from kinetic theory. Starting from the example
of the ideal, neutral gas to establish the procedure, a derivation of the two-fluid plasma
equation is carried out in detail. Finally, the one-fluid plasma equations are derived
and their meaning discussed, with particular emphasis on energy equation and Ohm’s
law.
In Chap. 3 we have shown that the moments of the distribution functions are
related to measurable macroscopic physical quantities. The moments themselves,
however, are defined in terms of the distribution function, so that a knowledge of
the latter would seem to be required to recover such quantities. A considerable
step forward would be achieved if it were possible to find a system of differential
equations containing only the moments. In this case, the moments would be defined
as the solutions of the of such a system, without further reference to the distribution
function, whose knowledge would become unnecessary. This type of approach is the
basis for the development of fluid models, that we are now going to discuss.
4.1 The Case of Neutral Gases
To establish the procedure, let’s start by considering the simple case of a neutral
gas, composed by identical particles of mass m. We shall therefore need just one
distribution function, that we shall still denote by f . The starting point of our program
will obviously be the general moment equation, Eq. (3.15):
∂ n ∂
(nψ) + ∇ · (nv ψ) − F · ∇ v ψ = (nψ) . (4.1)
∂t m ∂t coll
There will be one equation for each moment, which means that we are moving from
a single kinetic equation to a system of equations. It is easy to realize that actually we
shall be faced with an infinite system. In fact, let the function ψ entering Eq. (4.1) be
an n-th order moment, namely the product of n velocity components. Equation (4.1)

50 4 Fluid Models
would then be the equation for the n-th moment. But, the second term of Eq. (4.1)
contains the quantity v ψ that involves moments of order n + 1. We conclude
therefore that the equation for the moment of order n inevitably contains moments
of order n + 1 and therefore the system of moment equations is actually an infinite
chain. It is clear that, unless some way is found to truncate the chain of equations, our
program will turn out to be totally unfeasible. This is the so-called closure problem
that we shall discuss later on.
For the time being, we shall write down the appropriate equation for each moment.
At zeroth-order we put ψ = m in Eq. (4.1). Thus
∂ρ ∂n
+ ∇ · (ρ u) = m , (4.2)
∂t ∂t coll
where we have used the definition of the bulk speed u, Eq. (3.13) .
At first order we use ψ = m vi and write, for the i-th velocity component:
∂ ∂ ∂vi
(nmu i ) + (nmvi vk ) − n Fk
∂t ∂ xk ∂vk
∂ ∂ ∂(nmv )
i
= (nmu i ) + (nmvi vk ) − n Fi = , (4.3)
∂t ∂ xk ∂t coll
where a sum over repeated indexes must be understood. As expected, in Eq. (4.3)
we notice the presence of second order moments, nmvi vk . To make the physical
meaning of those terms clear, let’s write the velocity of a single particle as:
v = u + w, (4.4)
splitting it into an average (or flow) speed, u, and a peculiar speed, w. Since u = v,
the average of Eq. (4.4) gives:
w = 0. (4.5)
Thus, the average of w vanishes, a typical feature of random motions. This suggests
that w could be identified with the velocity of the chaotic thermal motion of the
particles. Inserting the expression Eq. (4.4) in vi vk and noticing that u i = u i ,
since u is a function only of time and spatial coordinates, and the velocity-averaging
process represented by leaves u unaltered, we obtain:
vi vk = (u i + wi )(u k + wk ) = u i u k + wi wk . (4.6)
Thus:
nmvi vk = nmu i u k + nmwi wk = nmu i u k + Pik .
The tensor Pik = nmwi wk is called the pressure tensor.

Since nmu k is simply the momentum density in the k-direction, nmu i u k =
u i (nmu k ) gives the flux of the k-th component of the momentum across the unit
4.1 The Case of Neutral Gases 51
surface whose normal is parallel to the i-axis. Therefore, the first term of Eq. (4.6)
represents the momentum flux connected with ordered motion, whereas the second
term represents the momentum flux connected with disordered or random motion,
that we have already identified with the thermal motion. The diagonal terms of the
Pik tensor, which have the same value in the case of an isotropic medium, correspond
to the normal definition of pressure of a fluid, while the off-diagonal ones are differ-
ent from zero only in the presence of viscous forces, which, in fact, act in a direction
perpendicular to the fluid velocity. These terms can be identified with shear-stresses
in the flow.
The pressure tensor is clearly a symmetric one. It may be useful to separate its
diagonal part from the off-diagonal one, by writing, always in an isotropic medium,
Pik = P δik + ik ,
where the tensor ik is a symmetric tensor with vanishing diagonal terms. Reverting
to Eq. (4.3) we shall therefore write:
∂ ∂ ∂(nmv )
i
(nmu i ) + (nmu i u k + Pik ) − n Fi = . (4.7)
∂t ∂ xk ∂t coll
Moving now to the second-order moments, we can limit ourselves to ψ =

1
2 mvi vi = 21 mv 2 and write:
∂ ∂
(n 21 mv 2 ) + ∇ · (nv 21 mv 2 ) − n F · u = (n 21 mv 2 ) . (4.8)
∂t ∂t coll
Making use once more of the decomposition given by Eq. (4.4), we may express the
term nvk 21 mv 2 as
nvk 21 mv 2 = n( 21 m(u 2 + w 2 )u k + u i Pik + nwk ( 21 mw 2 ).
Moreover, since

n( 21 mv 2 ) = 21 ρu 2 + 1
2 Pii = 21 ρu 2 + 21 (3P),
i
we get:
nvk 21 mv 2 = ( 21 ρu 2 + 23 P)u k + u i Pik + nwk ( 21 mw 2 ).
If the gas obeys the perfect gas equation, P = nkT , the term 23 P gives the internal
energy per unit volume, .1
1We did implicitly assume a monoatomic gas with 3 degrees of freedom; in the general case of f
degrees of freedom we have = P/(γ − 1), with γ = ( f + 2)/ f .
52 4 Fluid Models
Summarizing, the first term on the rhs represents the total energy flux (kinetic
energy of the ordered motion + internal energy) transported by the ordered motions,
the third gives the flux of the internal energy transported by the thermal motions,
namely the heat flux , while the second gives the work done by the pressure forces.
The preceding expression allows us to write Eq. (4.8) as:
∂ 1 2 3 ∂ 1 2 3
ρu + 2 P + ρu + 2 P u k + u i Pik + qk
∂t 2 ∂ xk 2
∂
− n Fi u i = (n 21 mv 2 ) ,
∂t coll
(4.9)
where we have explicitly introduced the heat flux vector, q = nw( 21 mw 2 ).
So far, we have avoided discussing the collisional terms appearing on the rhs of
Eqs. (4.2), (4.7) and (4.9). A rigorous theory should start from a specific collision
model, for instance that of binary elastic collisions by Boltzmann. Adopting such
a collisional term it is possible to show the the laws of conservation, in a single
collision, of the number of particles, of the total momentum and of the total energy,
imply the fulfillment of the following conditions:

∂f
=0
∂t coll

∂f
mv =0 (4.10)
∂t coll

2 ∂f
1
mv = 0.
2 ∂t coll
The importance of these conditions, that reflect the fundamental conservation laws,
is such that their validity is considered to be a necessary condition for adopting a
collisional term different from that of Boltzmann. In other words, the relationships
expressed by Eqs. (4.10), must be verified by any model for the collisional term. The
fulfillment of conditions (4.10) implies that all the terms at the rhs of Eqs. (4.2), (4.7)
and (4.9) are actually zero.
This result can be made plausible by considering, for instance, equation (4.2).
The collisional term represents, apart from the factor m, the temporal variation of
the number of particles in a given phase-space volume caused by collisions. But
the particle number is a conserved quantity in a collision if, as assumed, we neglect
the processes that modify the number of interacting particles, such as ionization or
recombination. Hence, the collisional term of Eq. (4.2) vanishes and we are entitled
to write:
∂ρ
+ ∇ · (ρ u) = 0, (4.11)
∂t
where we recognize the mass continuity equation of fluid mechanics.
In a similar way, the collisional terms of Eqs. (4.7) and (4.9) give, respectively, the
temporal variations of momentum and energy caused by collisions, that both vanish
in elastic encounters.
In Eq. (4.7) we are thus entitled to put the rhs equal to zero and to write, taking
into account Eq. (4.11), the first two terms as:

∂ ∂ ∂u i ∂u i ∂ρ ∂(ρu k )
(ρu i ) + (ρu i u k ) = ρ + uk + ui +
∂t ∂ xk ∂t ∂ xk ∂t ∂ xk

∂u i ∂u i
=ρ + uk .
∂t ∂ xk
But the expression on the rhs is the derivative of u i along the trajectory u = (dr/dt),
a quantity that in fluid mechanics is known as the lagrangian derivative of u i
d ∂
= + (u · ∇). (4.12)
dt ∂t
The final form of Eq. (4.7) is therefore
du i ∂Pik
ρ =− + n Fi . (4.13)
dt ∂ xk
In the preceding expression we recognize the equation of motion of a fluid. In the

case of a perfect fluid, Pik = P δik , and Eq. (4.13) becomes Euler’s equation:
du
ρ = −∇ P + n F, (4.14)
dt
while for a viscous fluid (ik = 0) we get the Navier-Stokes equation:
du i ∂P ∂ik
ρ =− − + n Fi . (4.15)
dt ∂ xi ∂ xk
In both the above equations a distinction has been made between pressure forces and
other types of forces.
Turning now to Eq. (4.9) and putting again the collisional term equal to zero, we
get:
∂ 1 2 3 ∂ 1 2 3
ρu + 2 P = − [ 2 ρu + 2 P]u k + Pu k +u i ik +qk +n Fi u i . (4.16)
∂t 2 ∂ xk
This is the energy equation, which states that the temporal variation of the total
energy contained in a given volume equals the flux of the same quantity across the
surface enclosing the volume, plus the work done by all forces, pressure forces and
forces of a different nature, plus the effect of the heat flux. It is important to notice
that all the terms, with the exception of qk , are proportional to one component of the
54 4 Fluid Models
fluid speed. Hence, these terms refer to energy fluxes connected with macroscopic
motions of matter, i.e. they are convective terms. On the contrary, the term containing
q survives even in static conditions, u = 0, and represents heat conduction.
A useful alternative expression for the energy equation is the following. Let’s
rewrite Eq. (4.16) in the form:

∂ 1 2 ∂ 1 2 3 ∂P + 5 ∂

ρu + ρu u k + Pu k
∂t 2 ∂ xk 2 2 ∂t 3 ∂x
k

∂
=− u i ik + qk + n Fi u i .
∂ xk
Expanding the derivatives and using the equation of continuity, Eq. (4.11), and the
equation of motion, Eq. (4.13), we get:

∂P ∂P ∂u k ∂u i ∂qk
3
+ uk + 53 P = −ik − .
2 ∂t ∂ xk ∂ xk ∂ xk ∂ xk
Using once again Eq. (4.11) to eliminate ∂u k /∂ xk and recalling the definition of
lagrangian derivative, Eq. (4.12), we obtain:

dP P dρ ∂u i ∂qk
3
− 5
= −ik − .
2 dt 3 ρ dt ∂ xk ∂ xk
Multiplying and dividing the lhs by ρ −5/3 we finally arrive at:
d ∂u i ∂qk
2ρ
3 5/3
P ρ −5/3 = −ik − . (4.17)
dt ∂ xk ∂ xk
Equation (4.17) is valid for a monoatomic gas. In the general case we have:
1 d ∂u i ∂qk
ργ P ρ −γ = −ik − . (4.18)
γ −1 dt ∂ xk ∂ xk
Notice that for a perfect gas, with vanishing viscosity and thermal conduction (ik =
0, qk = 0), the preceding equation reduces to
d
P ρ −γ = 0,
dt
which is the adiabatic equation of a prefect gas.
It is easy to generalize Eq. (4.11) to include processes implying a variation of the
total number of particles. The rhs will no longer vanish and will consist of a sum
of positive terms, representing the processes that increase the number of particles,
forcing the density to increase as well, and of negative terms representing a decrease
in particle number.
It is also possible to include in Eq. (4.18), beyond the effects of viscosity and
heat conduction, other dissipative effects, such as radiative losses, of fundamental
importance in astrophysics. In fact, introducing the expression of the entropy per
unit mass of a perfect gas, s = cV ln(P ρ −γ ), and recalling that
R 1
cV = ,
μ γ −1
where cV is the specific heat at constant volume, R the gas constant and μ the mean
molecular weight, we may write the lhs of Eq. (4.18) as ρ T ds/dt, that we recognize
as the expression for the total heat exchanged by the system. This suggests writing
Eq. (4.18) in the symbolic form
ds
ρT = G − P, (4.19)
dt
where G and P represent, respectively, the gains and losses of energy.
We have managed to deduce formally the equations of fluid mechanics, but still
we have not completed our program of obtaining a series of relationships among
the moments that don’t require the previous knowledge of the distribution function.
In fact, it is easily seen that, even in the absence of non- pressure forces (F = 0),
Eqs. (4.11), (4.15) and (4.18) form a system of 5 scalar equations in the 11 unknowns
P, ρ, u, q, ik . Hence, 6 quantities are still defined in terms of the distribution
function. It is clear that nothing can be gained by writing the equations for the
moments of order higher than second, since it is easily seen that the number of
unknowns increases more rapidly than that of equations. Therefore, a scheme must
be established allowing to express higher-order moments in terms of lower-order
ones, to finally end up with an equal number of equations and unknowns.
The simplest case is that of a collision-dominated, neutral gas. Such a system
evolves toward a state of thermodynamical equilibrium described by a maxwellian
distribution function:
m 3/2 1 m(v − u)2
f0 = n exp − 2
2π kT kT
m 3/2 1 mw 2
=n exp − 2 (4.20)
2π kT kT
A distinction must be made between the case in which the macroscopic quantities
n, T and u are constants and that in which they depend on space and time. In the
first case we have a state of thermodynamical equilibrium, in the second of a local
thermodynamical equilibrium, where n(r, t), T (r, t) and u(r, t) are slowly varying
functions over distances of the order of the mean free path and times of the order of
the collision time. In both cases it’s easy to check that ik = 0 and that qk = 0. In
fact, the definition of moments involves integrations over the velocity space only and
the circumstance that T, n and u depend on spatial as well as temporal coordinates
56 4 Fluid Models
becomes irrelevant. Thus Pik = P δik and Eqs. (4.11), (4.15) and (4.18) are sufficient
to determine the 5 unknowns P, ρ, u.
If the gas finds itself in a non-equilibrium condition, it is still possible to make
some progress by adopting a perturbative technique. This amounts to the assumption
that the distribution function can be represented as f = f 0 + f 1 with | f 1 | | f 0 |.
The system of fluid equation can be linearized and first-order expressions for the off-
diagonal terms of the pressure tensor and of the heat flux can be found. A discussion
of this method, called the Chapman-Enskog method, is beyond the scope of this book
(see, however, Exercise 4.1).
4.2 The Plasma Case: Two-Fluid Models
To apply the procedure developed for neutral gases to a plasma we must take into
account that the number of distribution functions increases. Restricting ourselves
for simplicity to the case of a completely ionized hydrogen plasma, composed by
electrons and protons, we will have to deal with two kinetic equations, one for each
species: we shall indicate them by f p and f e . If the only forces coming into play are
those of electromagnetic origin, the two kinetic equations can be written as:

∂ fs es 1 ∂ fs
+ v · ∇ fs + (E + v × B) · ∇ v f s = , s = e, p. (4.21)
∂t ms c ∂t coll
Following the procedure adopted for neutral gases, we obtain the equation equiv-
alent to Eq. (4.2):
∂n s (s) ∂n s
+ ∇ · (n s u ) = , (4.22)
∂t ∂t coll
where, obviously,
(s) 1
ns = f s dv e u = v f s dv.
ns
The equation of motion analogous to Eq. (4.3) becomes:
∂ (s) ∂ es n s (s) (s)

(n s m s u i ) + (n s m s vi vk s ) − es n s E i − (u × B)i = Ri ,
∂t ∂ xk c
where the notation s indicates that the average must be taken with respect to the
(s)
distribution function f s and we have introduced the compact notation Ri for the
collisional term, ∂ f
(s) s
R = ms v dv.
∂t coll
4.2 The Plasma Case: Two-Fluid Models 57
If we now define a pressure tensor for each species:
(s) (s)
Pik = n s m s wi wk = P (s) δik + ik ,
we arrive at the equation of motions for electrons and protons:
∂ (s)
∂ (s) (s) (s)

ns m s ui + n s m s u i u k + Pik
∂t ∂ xk
es n s (s)
(s)
− es n s E i − u × B = Ri . (4.23)
c i
In a similar manner, the two energy equations can be derived:

2
∂ ∂ 1 2
1
2 m s ns u (s) + 23 P (s) + m s n s u (s)
+ 5 (s)
P uk
∂t ∂ xk 2 2

(s) (s) (s) (s)
+ u i ik + qk − es n s E i u i = Q (s) , (4.24)
with
∂ fs
Q (s) = 2 ms v
1 2
dv.
∂t coll
What we did so far is a straightforward extension of the procedure used for neutral
gases, but the collisional terms require closer scrutiny. In fact, unlike the neutral case,
where it has been assumed that all the particles have the same mass, here we must
face cases in which the colliding particles have vastly different masses. For each
species, therefore, the collisional term will be the sum of the contributions coming
from both same species particle and different species particle collisions. To analyze
these cases and to simplify the notation let’s put:

∂ fs
C(s, s ) = (s, s = e, p).
∂t coll
Making use of this notation, the conservation law for total particle number can be
written in the form
C(s, s )dv = 0, (4.25)
valid both for s = s and s = s .

The global momentum and energy conservation give:

m s v C(s, s)dv = 0

2 m s v C(s, s)dv = 0,
1 2
58 4 Fluid Models
for collisions between particles of the same species, while for collisions between
particles of different species we shall have:

m s v C(s, s )dv + m s v C(s , s)dv = 0

1 2
2 m s v C(s, s )dv + 2 m s v C(s , s)dv = 0.
1 2
(4.26)
Thus, the collisional term given by Eq. (4.22) always vanishes, while in those of
Eqs. (4.23) and (4.24), only contributions from particles of different species enter,

∂n s
= 0,
∂t coll

(s)
R = m s v C(s, s )dv (s = s),

(s)
Q = 2 m s v C(s, s )dv
1 2
(s = s).
On the other hand, since the momentum and energy lost by one species are gained
by the other, we have
R(e) = −R( p) e Q (e) = −Q ( p) . (4.27)
The system given by Eq. (4.22) (with no collisional term), Eqs. (4.23) and (4.24) is
the basic system of equations of the so-called two-fluid model. Notice that in general
the temperatures and the heat fluxes of the two fluids will be different:
P (s) (s)
1
T (s) = ; qi = 2 m s wi wk wk .
k ns s
k
The closure problem for the two-fluid model is similar to that of neutral gases. Here
as well, the closure can be obtained by assuming that each fluid is in a state of local
thermodynamical equilibrium at its own temperature, an assumption that allows the
neglect of the viscous and thermal conduction terms. The coupling between the two
species is maintained by the terms that represent the mutual exchange of momen-
tum and energy and, to achieve closure, they must be expressed in terms of the
macroscopic variables. The two-fluid model is a good description of the electro-
proton system when the two species have not yet attained a common thermal equi-
librium. As observed in the Introduction, such a situation may occur in a rarefied
plasma, because of the relative inefficiency of the collisions between different species,
τep τ pp τee . A typical example of a two-fluid system is given by the solar wind,
where, at Earth orbit, the electron temperature is higher than the proton one.
4.3 The One-Fluid Model 59
4.3 The One-Fluid Model
Even in the case where a closure can be obtained, the two-fluid model is still a
complex one. This prompts us to look for a simpler description, at least when the
temperatures of the two fluids are not substantially different. This further simplifi-
cation can be obtained by introducing a fictitious fluid, somehow representative of
the entire plasma. We expect that the equations of such a model, if it can be found,
look similar to those of a neutral gas, with some extra terms to take into account the
electromagnetic effects.
The total numerical density of the plasma, n(r, t), i.e. the total number of particles
per unit volume, irrespective of the sign of the charge, will be given by:
n = ne + n p .
We can further define the following quantities for our one-fluid model;
mass density ρ(r, t) = n p m p + n e m e (4.28a)

charge density q(r, t) = e(n p − n e ) (4.28b)
current density J(r, t) = e(n p u( p) − n e u(e) ) (4.28c)
An equation for ρ can be easily obtained simply by multiplying the continuity

equations, Eq. (4.22), of each species by the respective mass and adding them. Thus,
∂ρ
= ∇ · m e n e u(e) + m p n p u( p) .
∂t
This expression can be cast in the familiar form of the mass continuity equation; by
introducing the vector
m e n e u(e) + m p n p u( p) m e n e u(e) + m p n p u( p)
U= = . (4.29)
m e ne + m p n p ρ
Keeping in mind that n e n p (charge quasi-neutrality!), we see that U, called the

fluid velocity, is a sort of local center-of-mass speed. Using this definition, we finally
obtain
∂ρ
+ ∇ · (ρ U) = 0. (4.30)
∂t
If we now multiply the equations Eq. (4.22) of the two species by the respec-
tive charges, and recall the definitions Eq. (4.28a)–(4.28c), we obtain the charge-
conservation equation:
∂q
+ ∇ · J = 0. (4.31)
∂t
60 4 Fluid Models
Having introduced the velocity U, it is natural to attempt to write the equation of

motion in terms of such a velocity, by adding the two equations of motions Eq. (4.23).
However, this procedure gives rise to a problem connected to the fact that the pressure
(s)
tensors Pik are defined in terms of the velocities w = v − u(s) , whereas, by analogy
to the case of neutral gases that we use as a paradigm, it would be preferable to
define the random speed as w = v − U. Moreover, since the pressures of the two
species have to be added, it would be more logical to refer their peculiar velocities
to the same fluid speed, U, rather than to two different average speeds, u(e) e u( p) .
However, the new definition of the peculiar velocities implies
w s = u(s) − U = 0.
The term m s n s vi vk s then needs to be modified as follows:
(s) (s) (s)

m s n s vi vk = m s n s (wi +Ui )(wk +Uk )s = Pik +m s n s (u i Uk +u k Ui −Ui Uk ),
(s)
where it has to be understood that the tensors Pik are now defined in terms of the
respective w . As a result, new terms appear in Eq. (4.23), that now reads
∂ ∂
(n s m s u i(s) ) + n s m s (u i(s) Uk + u k(s) Ui − Ui Uk ) + Pik
(s)
∂t ∂ xk
es n s (s) (s)
− es n s E i − (u × B)i = Ri , (4.32)
c
where it has been assumed that the only forces acting on the system are those of
electromagnetic origin.
At this stage we define a total pressure tensor as
(e) ( p)
Pik = Pik + Pik ,
add the two equations (4.32) for the two species and, taking into account Eqs. (4.27),
(4.28b), (4.28c) e (4.29), we get:
∂Ui ∂Ui ∂Pik 1

ρ + ρUk =− + q E i + ( J × B)i , (4.33)
∂t ∂ xk ∂ xk c
or, in vector notation,

dU 1
ρ = −∇ · P + q E + ( J × B). (4.34)
dt c
The kinetic temperature of the one-fluid model can be defined by generalizing the
definition used for neutral gases:

2
3
2 nkT = 2 m s w f s dv
1
= 1
2 Pii = 23 P,
s s
obtaining
Pii
T = . (4.35)
3nk
A similar procedure can be adopted with the two energy equations (4.24), provided
that the heat flux vectors, q (s) , are defined in terms of the peculiar velocities w . The
term 21 m s n s vi v 2 s now becomes
(s)
2 m s n s vi
1
v 2 = 23 P (s) Ui + Pik Uk

(s) (s) (s) (s)
+ 21 m s n s u i Uk Uk + 2u k Uk Ui + u i U 2 − 2Ui U 2 + qi
(4.36)
with these new definitions Eqs. (4.24) reads
∂ 1 (s) 3 (s)
∂ 3 (s) (s)
m n
s s (2 u U k − U 2
) + P + P Ui + Pik Uk
∂t 2 k 2 ∂ xi 2

+ 21 m s n s u i(s) Uk Uk + 2u k(s) Uk Ui + u i(s) U 2 − 2Ui U 2

(s) (s)
+ qi − es n s E k u k = Q (s) . (4.37)
Finally, defining the total heat flux vector as q = q (e) + q ( p) and adding the two
energy equations (4.37) we obtain the energy equation for the one-fluid model:
∂ 1 ∂ 1
ρU 2
+ 3
P + Ui ρU 2
+ 5
P + ik U k + qi − Jk E k = 0. (4.38)
∂t 2 2 ∂ xi 2 2
The meaning of the term J · E in the preceding equation becomes clearer if we

write it in the form:

c c ∂E
J·E= (∇ × B) − ·E=
4π 4π ∂t
c c ∂E
=− ∇ · (E × B) − B · (∇ × E) − · E,
4π 4π ∂t
where use has been made of the vector identity
62 4 Fluid Models
∇ · (F × G) = G · (∇ × F) − F · (∇ × G).
If we now use the Maxwell equation for ∇ × E we get the equation describing
electromagnetic energy density

c 1 ∂B ∂E
J·E=− ∇ · E × B) + B· +E·
4π c ∂t ∂t
2
∂ B E 2
= −∇ · S − + ,
∂t 8π 8π
where the Poynting vector S = (c/4π )(E × B), representing electromagnetic energy
flux, has been introduced. This particular form of the term J · E is completely general
and is valid both for ideal and dissipative plasmas. Indeed, the coupling between the
electromagnetic energy equation and the plasma energy equation occurs precisely
via J·E, which when positive provides a loss for the electromagnetic field and a
corresponding source for plasma energy, so that total energy is conserved. Finite
conductivity effects are implicitly contained in E, as we shall see when discussing
the generalized form of Ohm’s law.
Equation (4.38) now becomes
∂ 1 B2 E2 ∂ 1
ρU 2 + 23 P + + + Ui 2 ρU 2 + 25 P + ik Uk + qi + Si = 0.
∂t 2 8π 8π ∂ xi
(4.39)
The physical meaning of this equation is clear: the time variation of the total
energy (including the contributions coming from electric and magnetic fields) is
balanced by the flux terms that also contain the dissipative effects, some of which
appear explicitly while others are implicitly contained in the Poynting vector, as
already remarked. Therefore, even if a magnetic diffusivity term η = c2 /4π σ does
not appear explicitly, the transformation of magnetic energy into different energy
forms, such as thermal energy or kinetic energy of accelerated particles, can still
take place. We may now follow the procedure that allowed us to transform Eq. (4.16)
into Eq. (4.18). Taking into account the equation of motion (4.33) it can be easily
verified that the energy equation can be written as:

1 γ d −γ ∂Ui ∂qk 1
ρ Pρ = −ik − + qUk − Jk E k + (U × B)k .
γ −1 dt ∂ xk ∂ xk c
(4.40)
Notice that the term qU · (U × B) identically vanishes. It has been introduced only
to write the above equation in a form suitable for further developments.
The system of equations formed by Eqs. (4.30), (4.31), (4.33) and (4.38) [or
(4.40)], coupled to Maxwell’s equations for ∇ × E and ∇ × B, is the basis
of the one-fluid model. It consists of 12 scalar equations in the 21 unknowns
ρ, P, U, q, J, E, B, ik , q. It is quite evident that here too we will have to face
the closure problem, namely to find ways to express the last 6 unknowns in terms
of the previous ones. But, even leaving this problem aside, we still need three extra
scalar equations (or one vector equation) to equate the number of equations and
unknowns.
To find the missing vector equation, we notice that the equation of motion for
the one-fluid model was obtained by adding the two vector equations of motions
for the two species. A second independent equation can be obviously obtained by
subtracting them. To proceed, we first multiply each of the equations by es /m s and
then add the resulting equations, which is practically equivalent to subtracting them
because of the opposite sign of the electrical charges. But now the collisional terms
R(s) no longer cancel each other and to close the system we must express their
difference in terms of the fundamental variables of the fluid model. Explicitly,
∂ ( p) ( p)

en p u i − en e u i
∂t
∂ P( p) Pik
(e)
( p) ( p) (e) (e) ik
+ en p u i Uk + u k Ui − en e u i Uk + u k Ui + e −
∂ xk mp me
n ne 2
e n p ( p) n e (e)
p
− e2 + Ei − (u × B)i + u × B)i
mp me c mp me
R ( p) R
(e) 1
( p) 1
=e i − i = e Ri + ,
mp me mp me
where use has been made of Eq. (4.27). As already stated, e p = e, ee = −e .

Recalling the definition Eq. (4.28b) and considering that m e m p e n e n p ,
the preceding expression becomes
∂ (e)
∂ Ji e ∂Pik
+ Ji Uk + Jk Ui −
∂t ∂ xk m e ∂ xk
e2 n e ( p)
e2 n e e Ri
− Ei − m e u( p) + m p u(e) × B = (4.41)
me mem pc i me
The factor (m e u( p) + m p u(e) ) entering the term that contains the magnetic field B
can be rewritten, taking into account Eqs. (4.28b) and (4.29), as:
J J
m e u( p) + m p u(e) = (m e + m p )U + (m e − m p ) m p (U − ).
en e en e
We still need to define the relationship between the collisional terms R(s) and the
fluid variables. A reasonable assumption is to express R(s) in terms of the difference
between the fluid speeds of the two species:

R(s) = −n s m s νs,s u(s) − u(s ) ,
64 4 Fluid Models
where the parameter νs,s represents an average frequency for collisions of the parti-
cles of type s with the particles of type s . This approximation basically corresponds
to envisaging that the force acting between the two species during a collision is of a
viscous nature, so that it vanishes when the fluid speeds of the the species coincide.
Since n e n p , Eq. (4.27) implies that
m e νe, p = m p ν p,e .
Inserting the various expressions we have found into Eq. (4.41) and multiplying it
by (m e /n e e2 ), we finally obtain:
1 Ji m e ∂ Ji ∂
E i + (U × B)i − = 2 + Ji Uk + Jk Ui
c σ e n e ∂t ∂ xk
1 (e)
1 ∂Pik
+ J×B − , (4.42)
en e c i en e ∂ xk
where we have introduced the quantity
e2 n e
σ = ,
m e νep
the plasma electric conductivity [See Eq. (1.1) in the introductory chapter]. 2 Equation
(4.42) is known as the generalized Ohm equation. In fact, whenever all the terms
of the rhs can be neglected, the classic form of Ohm’s law in a moving, conducting
medium is recovered. Equations (4.30), (4.31), (4.33) and (4.38) [or (4.40)] together
with Maxwell’s equations for ∇ × E and ∇ × B and Eq. (4.42) now provide a
consistent system of equations with an equal number of equations and unknowns
provided that we are able to solve the closure problem, i.e. to express the fluid
(e)
quantities ik , Pik , q in terms of the others. There is generally no unique way to
achieve closure, however there are two situations where the system may be closed
without considering any moment higher than third order.
The most radical procedure is to assume that all components of the pressure tensor
Pik , as well as those of the heat flux vector q, vanish. This approximation is known
as the cold plasma model. Clearly, it is a highly idealized situation where all thermal
effect are neglected. In a cold plasma the number of unknowns reduces to 14, while
the number of equations is still 15. However, by inspecting the energy equation in
the form (4.40), we see that the rhs identically vanishes, so that the energy equation
reduces to
1
(qU − J) · E + (U × B) = 0.
c
2 We must however observe that the quantity νep that appears in the definition of R(s) relates
to the transfer of momentum between species, while the collisional frequencies discussed in the
Introduction relate to the transfer of energy. For a more detailed illustration of these aspects, see
Lifshitz and Pitaevskii [8], Chap. 4.
This equation is automatically satisfied provided that Ohm’s law takes the form:
1
E + (U × B) = 0.
c
Thus a cold plasma turns out also to be a necessarily ideal plasma. This is not too
surprising since in the cold plasma approximation we have neglected all effects
associated with the existence of collisions between the microscopic particles of the
plasma.
The other situation in which the system can be easily closed is that of a collisional
plasma, namely a plasma which is always in local thermodynamical equilibrium and
whose distribution function is a maxwellian. In this case we may apply the same
reasonings and approximations developed for neutral gases and conclude that, as a
first approximation, it is possible to write
=0 ; q=0 ; P = 0 ; P (e) P ( p) = 21 P.
In the presence of a strong magnetic field and a low degree of collisionality, plas-
mas may retain anisotropic thermodynamic properties even within the fluid approx-
imation, in the sense that temperatures and pressures in the directions parallel and
orthogonal to the magnetic field will be different. In this limit a fluid closure was
developed by Chew et al. [14], in which the pressure tensor of the plasma, considered
in a one-fluid approximation, is written as
Pi j = P⊥ δi j + (P
− P⊥ )bi b j
where bi = Bi /B is the i-th component of a unit vector along the mean magnetic
field B and subscripts
,⊥ indicate directions parallel and perpendicular to the mag-
netic field respectively. The equations describing the CGL model may be derived by
repeating the procedure described in Sect. 4.3. As far as the equation of continuity
is concerned, there is no change. For the single fluid equation of motion, one can
proceed without separating parallel and perpendicular components for the velocity,
but retaining the general tensor nature of the pressure in Eq. (4.34). Taking the the
scalar product of that equation with b leads to

dU ∇B
ρ = −∇
P
− (P⊥ − P
) .
dt
B
Upon subtraction of the parallel component from Eq. (4.34) we find for the momen-
tum equation orthogonal to the mean field

dU B2 B · ∇B P⊥ − P
ρ = −∇ ⊥ P⊥ + + 1+ 2 = 0.
dt ⊥ 8π 4π ⊥ B /4π
66 4 Fluid Models
Particular care is required in deriving the equation describing internal energy (4.40),
which becomes

1 D P
D P⊥ 1 ∂U j ∂U j
+ + P
+ P⊥ + Pi j = 0. (4.43)
2 Dt Dt 2 ∂x j ∂ xi
A separate independent equation is required for P

, and it is obtained from the general
moment equations by considering the function ψ = 21 mv
2 , with v
= v · b. The
difficulty comes from the fact that ψ now depends not only on velocity, but also on
position and time, through the time and spatial dependence of b. As a consequence,
we find that
∂ψ ∂b
= mv
v · ,
∂t ∂t
∇ψ = mv
(v · ∇ b) and ∇v ψ = mv
b. For simplicity we will consider here only
one particle species. Recall that v
= w
+ U
where U is the mean velocity so
that by definition P
= ρ < w
2 >. Taking the scalar product between U
b and the
momentum equation (4.33) we obtain:
2
ρ DU
Dbi ∂ Pik
− ρU
Ui = −U
bi + qU
E
. (4.44)
2 Dt Dt ∂ xk
Subtracting this equation from Eq. (3.15) for the second order moment, we obtain
the equation for the parallel pressure
D P
∂U j ∂Ui
+ P
+ 2P
bi b j =0 (4.45)
Dt ∂x j ∂x j
and subtracting this from Eq. (4.43), we obtain the equation describing the perpen-
dicular pressure
D P⊥ ∂U j ∂Ui
+ 2P⊥ − P⊥ bi b j = 0. (4.46)
Dt ∂x j ∂x j
These equations replace the adiabatic equation of state valid for an isotropic rarified
plasma and lead to interesting changes in the dynamics in the presence of a strong
magnetic field.
4.1. Starting from the kinetic equation (3.6), put f = f 0 + f 1 with | f 1 | | f 0 |, and
write the collisional term in the approximate form

∂f f1
= −νc ( f − f 0 ) = −νc f 1 = − ,
∂t coll τc
where νc = 1/τc (assumed to be constant) represents the average collision freequency

(see Eq. (1.14)). f 0 is a maxwellian function with constant n and T and u = 0, while
f 1 is a function of the velocity only.
(a) Discuss the meaning of the form adopted for the collisional term.
(b) Consider the stationary case (∂ f /∂t = 0), with the only force acting on the
system being that due to an electric field, E. If E is a first order quantity, namely of
the same order of f 1 , show that the electrical conductivity σ , defined by j = σ E
can be written in the form given by Eq. (1.1):
ne2
σ = .
mνc
Finally, taking Eq. (1.14) into account, deduce the dependence of σ on the density
and the temperature.
4.2. Using the same assumptions of the preceding problem, let n and T be weakly
dependent on the position, so that their spatial gradients could be considered as first
order quantities. The coefficient of thermal conductivity κ is defined by the relation
q = −κ∇T , where q is the heat flux vector. Assuming that the pressure P = nkT
is constant, show that
5Pk
κ=
2mνc
and evaluate the dependence of κ on the density and temperature.

4.3. In a viscous fluid the off-diagonal terms of the pressure tensor do not vanish,
ik = 0. If the mean flow is u = u(z)ex , the viscosity ν is defined by
dU
ρν = −x z .
dz
Using the approximate collisional term of the preceding problems show that
P
ν= τc
ρ
and evaluate the dependence of ν on the density and temperature
Solutions
4.1. (a) Since the collisional term vanishes when f = f 0 and collisions cause the
distribution function to evolve towards the equilibrium distribution f 0 on a timescale
characterized by τc = 1/νc , the form adopted for the collisional term is an acceptable
first order representation.
(b) Within our assumptions, the first order kinetic equation reduces to
68 4 Fluid Models
e E ∂ f0
· = −νc f 1 .
m ∂v
If the electric field is written as E = E ez , from the definition of σ we have:

σE = j = e vz f dv = e vz f 1 dv.
Multiplying the kinetic equation by evz and integrating over the velocities we obtain

e2
σE = E vz2 f 0 dv.
νc kT
Performing the integral we find the required result. The electrical conductivity is
independent of n and scales with the temperature as T 3/2 .
4.2. Though the pressure is constant, both n and T depend on the position. The
equilibrium distribution function f 0 then reads
m 3/2 mv 2 P m 3/2 −5/2 mv 2
f0 = n exp − = T exp − ,
2π kT 2kT k 2π k 2kT
while the kinetic equation is simply given by
v · ∇ f 0 = −νc f 1 .

If T = T (z) we have q = q ez , q = −κ(dT /dz), with q = 21 m v 2 vz f 1 dv, and the
preceding equation becomes
5 mv 2
f 0 vz − = νc f 1 .
2T 2kT 2
Multiplying the above equation by v 2 vz and integrating in dv we obtain the required

result. The thermal conductivity is independent of n and scales with the temperature
as T 5/2 .
4.3. For u = u(z)ex , f 0 takes up the form
m 3/2 m
f0 = n exp{− [(vx − u(z))2 + v 2y + vz2 ]},
2π kT 2kT
and the kinetic equation becomes
mvz du f1
(vx − u) f 0 = − .
kT dz τc
On the other hand,


du
ρν = −m vx vz f dv = −m vx vz ( f 0 + f 1 )dv = −m vx vz f 1 dv,
dz
since ik = 0 for a maxwellian distribution. Multiplying the kinetic equation above
by vx vz and integrating in dv we obtain
kT P
ν= τc = τc .
m ρ
The kinematic viscosity scales as T 5/2 /n.

Chapter 5
Magnetohydrodynamics
Abstract From the one-fluid plasma model the simplifications leading to magne-
tohydrodynamics (MHD) are discussed in detail. The general form of the MHD
equations, both ideal and resistive, is presented and their consequences analyzed, in
particular in connection with the time evolution of magnetic fields. MHD equilibria
are then described, both when magnetic forces are absent (force-free fields) and when
they are present. Finally, the concept of perturbed equilibrium states is introduced.
Although the collisional one-fluid model is the simplest of fluid models for a
plasma, it is still complex enough to make general solutions difficult to recover
in all but the simplest problems. In the fluid description of a plasma one typically
attempts to make further approximations by limiting the range of possible parameters,
namely densities, temperatures, magnetic fields and velocities to those of the specific
plasma under examination, introducing the concept of a plasma regime. The choice
of a regime generally allows one to neglect terms in the equations which are only
important when either the time or spatial scales are far from those of the plasma in
consideration. A very important regime derived from the collisional one-fluid model
is the (classical) magnetohydrodynamic, or MHD, regime.
5.1 MHD Equations
To develop the scalings for magnetohydrodynamics, begin by defining characteristic

length-scales L and timescales τ for electric and magnetic field dynamics, so that
L is the spatial scale over which the fields show a significant variation and τ is the
typical timescale over which the fields change. Let U be a representative value of
the fluid velocity of the plasma. The MHD regime is then defined by the following
conditions:
L
U and U c. (5.1)
τ
The first requirement means that “the typical speed” of electromagnetic phenomena,
that we identify with L/τ , is of the same order of magnitude as the typical speed of

72 5 Magnetohydrodynamics
hydrodynamical phenomena as defined by U.Therefore, since the two classes of phe-

nomena “proceed with the same speed”, the potential for mutual interaction is maxi-
mized. This regime is intermediate between a situation dominated by electromagnetic
effects, where hydrodynamical aspects appear as perturbations, and the symmetrical
one, where hydrodynamics dominates and electromagnetic interactions bring in only
small corrections. The second requirement is that the macroscopic plasma motions
be non-relativistic. If this holds systematically, since individual speeds of the bulk
of particles must also be close to the macroscopic fluid speed, the thermal speed of
the plasma must also be non-relativistic.
To see how the one-fluid equations may be simplified in the MHD regime, we
carry out a dimensional analysis, starting from Maxwell’s equations. Call E, B, Q, J
the characteristic values for the electric and magnetic fields and for the charge and
current densities, respectively. Begin with
1 ∂B
∇×E=− ,
c ∂t
which we write as:
E 1B
.
L cτ
This implies
E 1L U
1. (5.2)
B cτ c
Similarly, the equation for ∇ × B gives:
4π 1 ∂E B 4π E
∇×B= J+ ⇒ J + .
c c ∂t L c τ
Dividing both members by B/L we get:
2
4π L E L 4π L U
1 J + J + .
c B Bτ c B c
The last term is negligible with respect to unity and the equation for ∇ × B reduces to
4π
∇×B= J. (5.3)
c
In the MHD regime it is thus possible to neglect the displacement current, which
makes it clear that we are in a low frequency regime. The displacement current, in fact,
becomes important only when E varies rapidly with time, namely in a high frequency
5.1 MHD Equations 73
regime. The elimination of the displacement current implies that the continuity equa-
tion for the electrical charge must also change. A dimensional analysis of the latter
gives
Q J
+ = 0,
τ L
and, replacing J with the value obtained by a dimensional analysis of (5.3) and Q
with a similar analysis of the equation for ∇ · E ⇒ E/L 4π Q, we obtain
E 1 cB 1
+ = 0.
4π L τ 4π L L
Therefore, the ratio of of the first term to the second one is
2
EL U
1,
Bτ c
implying that in the MHD regime the term with the temporal derivative of the charge
density is negligible: the equation for the charge conservation can be written simply
as ∇ · J = 0, in perfect agreement with Eq. (5.3).
Coming now to the momentum equation (4.43) and performing the usual dimen-
sional analysis, one obtains:
U P 1
ρ − + QE + J B.
τ c
Using the estimates found previously, the ratio of the electric to the magnetic parts
of the Lorentz force becomes
2 2
QE E U
1,
J B/c B c
so that the electric part of the force may be neglected, leading to the form of momen-
tum equation in MHD:
dU 1
ρ = −∇ P + J × B. (5.4)
dt c
To estimate the relative importance of the various terms entering Ohm’s equation
(5.5), that for convenience we repeat here,

1 Ji m e ∂ Ji ∂
Ei + U×B − = 2 + Ji Uk + Jk Ui
c i σ e n e ∂t ∂ xk
(e)
1 1 ∂Pik
+ J×B − , (5.5)
en e c i en e ∂ xk
begin by dividing all terms by the representative value of the electric field E. Notice
that the two terms in square bracket have the same order of magnitude. Defining
1
2
−1
ωτ , cs P/ρ ,
and comparing the various terms in the same order in which they enter the above
equation we obtain:
2 2 2
1 : 1 : ω/ω pe νep /ω pe c/U : ω/ω pe c/U :
2 2 2
ω/ω pe ωce /ω pe c/U : ω/ωcp cs /U ,
where ω pe is the electron plasma frequency, ωce and ωcp are, respectively, the electron
and proton cyclotron frequencies and νep is the electron-proton collision frequency.
We thus see that, in order to neglect the terms in square bracket, we must have

ω/ω pe U/c , (5.6)
to neglect the term proportional to J × B, connected with the so-called Hall effect
we must have
2
ωωce /ω pe U/c ,
2
(5.7)

which may also be rewritten in terms of the proton cyclotron frequency ωcp and the
Alfvén speed ca = B/ 4π m p n p as:
2
ω/ωcp U/ca . (5.8)
Finally, the electron pressure term can be neglected when

2
ω/ωcp U/cs . (5.9)
When all these conditions are satisfied, which is most easily achieved in the low
frequency regime typical of MHD, Ohm’s equation reduces simply to
1 J
E+ U×B= , (5.10)
c σ
which is what is usually known as Ohm’s equation for a resistive plasma. If the
further condition 2
ωνep /ω pe U/c ,
2
(5.11)
is satisfied, the term on the rhs of equation (5.10) can also be neglected. This happens
in the presence of a very high electrical conductivity, in which case we obtain the
form of Ohm’s equation appropriate to an ideal plasma:
1
E + U × B = 0. (5.12)
c
Finally, noticing that the term Q U is of the order of (U/c)2 with respect to J ,
the energy equation (4.40) can be written as

1 d J2
ργ P ρ −γ = . (5.13)
γ −1 dt σ
The fundamental equations of the one-fluid model in the MHD regime are there-
fore Eqs. (4.30), (5.4), (5.13) and Eq. (5.3), with the addition of the equations that
still contain the electric field, namely:
1 ∂B
∇×E=−
c ∂t
1 J
E+ U×B= .
c σ
However by taking the curl of Ohm’s law and eliminating the the electric field via
the equation for ∇ × E one obtains an equation containing only the fluid velocity
and the magnetic field:
1 1 ∂B 1
∇ × E + ∇ × (U × B) = − + ∇ × (U × B)
c c ∂t c

J
=∇×
σ

c
=∇× ∇×B .
4π σ
Defining the magnetic diffusivity,
c2
η= , (5.14)
4π σ
expanding the lhs and remembering that ∇ · B = 0, we obtain
∂B
= ∇ × (U × B) + η∇ 2 B − ∇η × (∇ × B). (5.15)
∂t
This equation is known as the magnetic induction equation, sometimes also called
Faraday’s equation, and allows a further reduction of the number of equations of the
MHD model.
Summarizing, the resistive MHD equations, are:
∂ρ
+ ∇ · (ρ U) = 0 (5.16a)
∂t
dU 1
ρ = −∇ P + 1c J × B = −∇ P + (∇ × B) × B (5.16b)
dt 4π

1 d 4π
ργ P ρ −γ = 2 η J 2 (5.16c)
γ −1 dt c
∂B
= ∇ × (U × B) + η∇ 2 B − ∇η × (∇ × B). (5.16d)
∂t
In the momentum equation we have expressed the current density in terms of B
using Eq. (5.3). When the electrical conductivity is constant the last term of Eqs.
(5.15) and (5.16d) vanishes. The equations for an ideal plasma are easily obtained
by taking η = 0, or, equivalently, by letting σ tend to infinity.
The system (5.16a–5.16d) is a closed system that completely determines the 8
primary unknowns ρ, P, U, B. The other quantities can then be deduced in terms
of the primary ones. The current density is obviously given by
c
J= (∇ × B),
4π
the electric field is obtained from Ohm’s law
1 J
E=− U×B+ ,
c σ
and the charge density from
1
q= (∇ · E).
4π
The electrical conductivity σ is assumed to be a known quantity and depends in
general on density and temperature via the collision frequency, though it will often
be taken as a constant in the text.
By comparing the MHD equations with the corresponding ones for a neutral gas,
we realize that they represent the minimum possible number of equations describing a
plasma as a conducting medium. In spite of the fact that inclusion of electromagnetic
effects introduces three vector quantities (E, B, J) and one scalar quantity (q) when
compared to neutral flow, the primary equations in MHD have only been increased
by three, thanks to the choice of an appropriate regime.
5.1.1 Magnetic Pressure
By means of the MHD equations it is possible to investigate some aspects of the

interactions between the plasma and the magnetic field that are not immediately
apparent.
Let’s start from the momentum equation (5.16b) whose last term can be modified
making use of the vector identity:
∇(F · G) = (F · ∇)G + (G · ∇)F + F × (∇ × G) + G × (∇ × F). (5.17)
Using F = G = B we obtain
1 1 1 2
(∇ × B) × B = (B · ∇)B − ∇ B ,
4π 4π 8π
and Eq. (5.16b) can be written as

dU B2 1
ρ = −∇ P + + (B · ∇)B. (5.18)
dt 8π 4π
The i-th component of the preceding equation is

dUi ∂ B2 1 ∂ Bi ∂ B2 1 ∂(Bi Bk )
ρ =− P+ + Bk =− P+ + ,
dt ∂ xi 8π 4π ∂ xk ∂ xi 8π 4π ∂ xk
namely
dUi ∂
ρ =− Tik , (5.19)
dt ∂ xk
where use has been made of ∇ · B = 0 and a new tensor has been introduced:

B2 1
Tik = P + δik − Bi Bk . (5.20)
8π 4π
Assuming that the z-axis of our reference system is aligned with the magnetic field,
which is always achievable locally by an appropriate rotation of the axes, the tensor
Tik reduces to ⎛ ⎞
B2
⎜ P + 8π 0 0 ⎟
⎜ ⎟
⎜ B 2 ⎟
⎜ 0 P + 0 ⎟.
⎜ 8π ⎟
⎝ B2 ⎠
0 0 P−
8π
We thus see that the presence of a magnetic field introduces an extra isotropic pres-
sure, B 2 /8π , and an anisotropic negative pressure (i.e. a tension) along the field,
−B 2 /4π . To better appreciate the effect of the tension term, let’s examine the case of
a magnetic field given by B = [0, B y (x), Bz (x)], whose field lines are straight lines
(for B y = 0, the field lines are straight lines parallel to the z-axis, while if B y = 0 the
field lines are straight in any given plane x = const., but their inclination changes
when x is varied). For this field the tension term vanishes, which suggests that the
tension is active only when the field lines are curved. In a certain sense, the field lines
of B behave as they were composed of an elastic material: any deformation induces
a tension that tends to restore a configuration with unbent, straight field lines.
The relative importance of plasma kinetic to magnetic pressures is measured by
the parameter β, defined by
P
β= 2 . (5.21)
B /8π
Whenever β 1 the system’s dynamics is dominated by hydrodynamical effects,

while when β 1 magnetic effects are dominant. This is confirmed by the fact that
β can also be written in the form
2 E th
β= ,
3 E mag
where E th /V = 23 P is the thermal energy density and E mag /V = B 2 /8π is the

magnetic energy density.
5.1.2 The Conservative Form of MHD Equations
The continuity equation (5.16a)
∂ρ
+ ∇ · (ρ U) = 0,
∂t
describes mass conservation in differential form. It states that the amount of mass
contained in a given volume varies with time due to mass flux crossing the surface
that encloses the volume. More generally, an equation is said to be conservative if it
is possible to cast it in a form similar to the continuity equation, namely:
∂
+ ∇ · = 0,
∂t
where is the density of some quantity, and can be thought as the flux of the
same quantity. As written, the equation refers to the density of a scalar quantity. Its
generalization to a vector quantity is:
∂i ∂ik
+ = 0,
∂t ∂ xk
where , a tensor, is now the flux of the vector density. The equations of magneto-
hydrodynamics may all be cast in such conservative form. The conservative form of
MHD is useful both for the physical intuition it provides, and, more technically, for
its importance when attempting to solve the equations numerically. The lhs of the
momentum equation can be written as
∂Ui ∂Ui ∂(ρUi ) ∂ρ ∂Ui

ρ + ρUk = − Ui + ρUk
∂t ∂ xk ∂t ∂t ∂ xk
∂(ρUi ) ∂(ρUk ) ∂Ui
= + Ui + ρUk
∂t ∂ xk ∂ xk
∂(ρUi ) ∂(ρUi Uk )
= + ,
∂t ∂ xk
where use has been made of the continuity equation. It is now easy to see that it is
possible to write the momentum equation in conservative form:

∂ ∂ B2 1
(ρ Ui ) + ρ Ui Uk + (P + )δik − Bi Bk = 0. (5.22)
∂t ∂ xk 8π 4π
In this case:
= ρ U; ik = Tik + ρ Ui Uk ,
with the tensor flux Tik given by Eq. (5.20).

The energy equation in the form given by (5.16c) cannot be cast in conservative
form, because it describes how the internal energy varies with time, and it is not
internal energy density that is conserved, but the total energy density. The internal
energy per unit mass of a perfect gas is given by
RP 1 P
W = cv T = cv = ,
μρ γ −1 ρ
and the internal energy per unit volume is therefore = ρ W = P/(γ − 1). It is
easily verified that Eq. (5.16c) can be written as
∂ 4π
+ (U · ∇) + γ (∇ · U) = 2 η J 2 ,
∂t c
showing that the internal energy of a given volume can vary both due to compressive
effects (the term proportional to ∇·U) and to dissipative effects (the term proportional
to η).
To obtain a conservation law we must utilize Eq. (4.39) in the MHD regime, i.e.

∂ P B2 ∂ γ P c
2 ρU + γ + + Ui ( 21 ρU 2 + )+ (E× B)i = 0. (5.23)
1 2
∂t − 1 8π ∂ xi γ − 1 4π
This equation describes the conservation of total energy density and is already written
in conservative form.
Finally, we must consider the induction equation (5.16d). Generally, this equation
is intrinsically non conservative, because of the presence of the dissipative terms
containing η. However, it may be cast in a form that becomes conservative for ideal
plasmas for which η = 0. The i-th component of ∇ × (U × B) is:
∂ ∂
[∇ × (U × B)]i = i jk (klm Ul Bm ) = (Ui B j − Bi U j ),
∂x j ∂x j
where i jk is the antisymmetric tensor and use has been made of
i jk klm = ki j klm = δil δ jm − δim δ jl .
The above relationship makes it possible to write the induction equation as:
∂ Bi ∂
+ (Uk Bi − Ui Bk ) = η∇ 2 B − ∇η × (∇ × B) i , (5.24)
∂t ∂ xk
which is manifestly conservative when η = 0. In the next section we shall prove that
this equation represents the magnetic flux conservation for ideal plasmas.
5.2 The Time Evolution of Magnetic Fields
Consider Farday’s equation assuming that the resistivity η, Eq. (5.14), is a constant,
independent of thermodynamical variables (e.g. density and temperature):
∂B
= ∇ × (U × B) + η∇ 2 B. (5.25)
∂t
If we assume that the velocity field U is known then Eq. (5.25) does not require any
other of the MHD equations for closure, and may be solved for the magnetic field
once an initial field is given at an initial time. This approach to understanding the
evolution of magnetic fields is known as the kinematical approach, and in the rest
of this section we will follow it, disregarding the connection between Eq. (5.25) and
the other MHD equations, in particular the momentum equation (5.16b).
From Eq. (5.25) we see that the temporal variation of B is given by two terms,
each describing a different physical process. The first term, containing the fluid
velocity U, is a convective term, describing how the field changes due to convection
and concentration and rarefaction from the velocity field; the second term, in which
U is absent, is a diffusive term. The corresponding processes occur on completely
different timescales. According to the usual dimensional analysis the convective term
can be evaluated as (U being the magnitude of the velocity)
5.2 The Time Evolution of Magnetic Fields 81
B
with τ f = L/U,
τf
while for the diffusive term one has

B
with τd = L2 /η,
τd
with τ f and τd corresponding to convective and diffusion timescales respectively.

The relative importance of convection and diffusion is measured by the ratio of the
magnitude of the two terms
τd UL
Rm = = ,
τf η
defining the magnetic Reynolds number. If, as often happens in MHD, the typical
convection speed is of the same order of the Alfvén speed, see Eq. (1.21), typical
of magnetic phenomena, the magnetic Reynolds number is called the Lundquist
number, defined by
ca L
S= .
η
Consider now the two limiting cases of small Rm 1 and large Rm 1 Reynolds
numbers.
5.2.1 Rm 1: Magnetic Diffusion
If Rm 1 the convective term can be neglected and the induction equation reduces
to
∂B
= η∇ 2 B.
∂t
This being a linear differential equation for B, it is possible to apply the Fourier
analysis method.1 We therefore write the magnetic field in terms of its Fourier trans-
form
B(r, t) = dk dω B(k, ω) ei(k·r−ωt) ,
and introduce the above representation into the equation for B to obtain

dk dω(iω − ηk 2 )B(k, ω) ei(k·r−ωt) = 0.
1 A short presentation of Fourier analysis will be given in Sect. 7.1, to which we refer for details.
We shall anticipate here a few fundamental concepts, which should however be already known to
the reader.
Since the above equation must be identically satisfied for every B(k, ω), it follows
that ω = −iηk 2 . To take this condition into account, introduce the Dirac δ function
and write
B(r, t) = dk dω δ(ω + iηk 2 )B(k, ω) ei(k·r−ωt) .
Performing the ω integration we get:

dk e−ηk t B(k) eik·r .
2
B(r, t) = (5.26)
This is the Fourier representation of the solution of the diffusive equation: it states
that the Fourier components of a generic magnetic field, given at t = 0 by B(k),
decay exponentially with time. The magnetic energy (B 2 /8π )V (where V is the
volume) decreases because the presence of resistivity transforms magnetic energy
into other forms of energy. A part of the original magnetic energy goes into ther-
mal energy (Joule effect), but in general a fraction may be converted into kinetic
energy from acceleration of the fluid. In fact, if ∂ B/∂t = 0 an electric field is
generated which might accelerate plasma particles. Notice that the components cor-
responding to large values of k, i.e. to small wavelengths, decrease faster. Since small
wavelengths describe rapid spatial variations, the field becomes smoother during its
general decrease.
The diffusive time τd can have enormously different values for different systems.
For instance, in thermonuclear fusion machines τd 10 s, but in the liquid Earth core
its value increases up to 104 years, while in the Sun’s interior it can reach up to1010
years, of the same order or larger than the age of the Sun itself.
5.2.2 Rm 1: Alfvén’s Theorem
When the electrical conductivity is very high and/or the spatial scales are very large
Rm 1, so that the diffusive term in Eq. (5.25) becomes negligible. Such conditions
often arise in natural plasmas (as given by the example of the Sun above), which,
consequently, can be approximated by ideal plasmas, with η = 0, in many respects.
In the latter case, Eq. (5.25) simplifies to
∂B
= ∇ × (U × B). (5.27)
∂t
and its solutions exhibit some important features that we are now going to illustrate.
The most important result is contained in
Alfvén’s theorem: The magnetic flux through any closed line that moves with the
fluid is constant in time.
Fig. 5.1 Surfaces employed S

in the proof of Alfvén’s A
theorem
B
Udt dl
n
C
C’
To demonstrate this theorem, consider, at time t, a closed curve C, that we identify

with the particles that, in that particular moment, lie on it. Because of the motion
of the plasma, the particles will be displaced and, at time t + dt they will define a
different curve C . Since to compute the flux of B through a closed curve we can
utilize any surface having that curve as a boundary, let’s choose at time t a generic
surface S and at time t + dt a surface S , made up of S plus the surface A formed by
the flux lines joining C and C , as shown in Fig. 5.1.
The change in the flux from time t to time t + dt may be written as

d = B · dS − B · dS,
S S
where the field B = B(t + dt, r). Now
∂B
B = B(t + dt, r) = dt + B(t, r)
∂t
while
d = B − B · dS + B · n A,
S
where we have neglected the difference between B and B in calculating the flux
through the surface A, which is already first order in dt because n A = −dl × Udt.
Summarizing, we find that

d = B − B · dS − U × B · dldt.
S C
Using Stokes’ theorem on the curve C and dividing by dt we finally obtain

d ∂B
= − ∇ × (U × B) · dS
dt ∂t
S
Therefore, if Eq. (5.27) is valid, the flux is constant in time and the theorem is
proved.
Let C1 and C2 be two curves connected, at time t, by magnetic field lines, so
as to form a flux tube. The total flux of B across the surface of the tube, B , is
obviously given by the flux across C1 and C2 , since the flux across the lateral surface
of the tube vanishes by construction. From Alfvén’s theorem it follows that B will
remain constant during the whole dynamical evolution of the system. Now let the
area enclosed by C1 and C2 shrink to zero so that the flux tube becomes a single
field line. At this point we may be tempted to say that field lines move together with
the fluid or, as often stated, that the field lines are frozen in the fluid. Intuitive as this
statement may be, a more formal proof is however required, which we are now going
to present To begin write the continuity and induction equations, Eqs. (5.16a) and
(5.27), respectively, in lagrangian form. Remembering that
d ∂
= + U · ∇,
dt ∂t
we have
dρ
= −ρ∇ · U
dt
and
dB
= (B · ∇)U − B(∇ · U),
dt
where use has been made of the vector identity
∇ × (F × G) = F(∇ · G) − G(∇ · F) + (G · ∇)F − (F · ∇)G,
and of the condition ∇ · B = 0. From the two equations above, we get:

d B 1 B B
= [(B · ∇)U − B(∇ · U)] + (∇ · U) = · ∇ U. (5.28)
dt ρ ρ ρ ρ
If the fluid motion is described by a velocity field U(r), the equation that governs
the motion of a generic line element d joining the points r and r + d may be found
simply as:
d(d) d(r + d) dr

= − = U(r + d) − U(r) = (d · ∇)U. (5.29)
dt dt dt
A direct comparison of Eq. (5.28) and (5.29) shows that they are identical. There-
fore the quantity B/ρ evolves precisely as d and it follows that, if d and B are
parallel at a given moment, they will remain so at any later time. We thus recover, in a
quantitative form, the “freezing” condition of field lines in the fluid. To describe this
result in cartoon-like fashion, imagine marking all particles that, at a given moment,
lie on a field line, by painting them in red. During the dynamical evolution, the line
traced by the red particles will be deformed but, according to the property just shown,
it will still be a field line. Alfvén’s theorem thus allows us to identify a line of B
and to follow it in time. In those cases when Alfvén’s theorem does not apply, i.e.
when η = 0, this is not possible. It is true that at any given moment field lines may
still be drawn, but their identification with those drawn at a different time will be no
longer possible. Thus, Alfvén’s theorem endows field lines with a degree of reality
well beyond that of a useful visualization tool. Moreover, since the fluid motion is
considered to be continuous, the lines of B can only be deformed, but not broken,
so that the field topology, i.e. the ensemble of the geometrical properties that are
conserved during deformations, cannot be altered.
In an ideal plasma therefore magnetic field and matter are strongly tied together
and their dynamics depends on by the dominant term in the momentum equation. If
β 1 the motion is determined by pressure forces and matter drags the magnetic
field. On the other hand, if β 1 magnetic forces are the dominant ones and matter
is dragged by the magnetic field.
Having examined the extreme cases Rm 1 and Rm 1, the question arises as
to which of the two is most likely to occur natural plasmas. A table of the estimated
values of S for a certain number of systems is given in Table. 5.1.
As shown in the Table, in most cases S 1 and we may be tempted to neglect
resistive effects all together. However, this would be wrong, as the following con-
siderations show. First, observe that the dimensional analysis used to obtain our
estimates of the relative importance of convective and diffusive terms completely
neglects the vector character of the induction equation. It is true that on average the
convective term dominates over the diffusive one by many orders of magnitude, but
this is clearly untrue where the convective term vanishes or becomes very small. This
may happen close to points where U = 0, where U is parallel to B (U × B = 0) or
finally where ∇ × (U × B) = 0. In these regions the ideal plasma condition may
become locally invalid, and the diffusive term may no longer be neglected. In these
situations, Alfvén’s theorem does not hold, the field topology may change and mag-
netic energy may be transformed into other forms of energy. This is why studies of
the effects of the resistive, or other non-ideal terms in Ohm’s law, are so important:
in fusion machines non-ideal effects may be the cause of instabilities driving the
disruption of the plasma configuration; in astrophysics heating processes and very
Table 5.1 Spatial and temporal scales, Lundquist numbers

System L(cm) τd (s) τa (s) S
Tokamak 102 10−1 10−3 102
Earth nucleus 108 1012 105 107
Sunspot 109 1014 105 109
Solar corona 1011 1018 106 1012
dynamic energy releases are observed in situations where the only apparent source
of energy resides in magnetic fields. But the transformation of magnetic into thermal
energy and other energy forms is only possible when resistive or other non-ideal
effects are at work.
5.3 Equilibrium States of Ideal Plasmas
A static equilibrium state is defined by the conditions U = 0 and ∂/∂t = 0. A system

in equilibrium will remain in that state if no changes occur in the forces acting on the
system or in the boundary conditions. When dealing with MHD equilibrium configu-
rations of an ideal plasma, pressure, magnetic and all other forces of different nature
acting on the system must be taken into account. In astrophysics, the most relevant
force is gravity. Since angular momentum and rotation around gravitational center
are common, centrifugal forces are also often important in non-static equilibria. The
anisotropic nature of magnetic forces considerably contributes to the complexity of
equilibrium structures, as shown by the generalized MHD virial theorem, a simple
generalization to MHD of the well-known theorem of mechanics. One consequence
of this theorem is the impossibility for an isolated plasma to be in equilibrium under
the effects of the pressure and self-consistent magnetic forces only: external forces
are required for confinement.
To demonstrate the theorem, start from the equation
∂
Gik = 0,
∂ xk
that can be deduced from Eq. (4.34) by letting U = 0 and q = 0. Gik incorporates
the entire pressure tensor Pik , including its off-diagonal part, and all the magnetic
terms, namely
B2 Bi Bk B2 Bi Bk
Gik = Pik + δik − = (P + )δik + ik − .
8π 4π 8π 4π
Multiplying by xi , summing over i and integrating in dV over the entire space
leads to
∂
xi ( Gik )dV = 0,
∂ xk
V
which, on integration by parts, becomes

∂ xi
xi Gik dSk − Gik = 0. (5.30)
∂ xk
S V
5.3 Equilibrium States of Ideal Plasmas 87
Here the first term represents the flux of rG across the surface S that encloses V , i.e.
the surface at infinity. For an isolated system, with “good” convergence properties of
the components of Gik , one may assume that the surface integral vanishes. Therefore
the second integral, the volume integral of the trace of G, is also equal to zero. But,
since Tr = 0,
B2
Tr G = 3P + > 0.
8π
Thus Eq. (5.30) cannot be satisfied, which implies that the initial assumption, namely
the existence of an equilibrium under the sole action of pressure and magnetic forces,
is not valid. It must be emphasized that the theorem holds only if the system is
uniquely composed by the plasma and is isolated in three dimensions.
Writing the static equilibrium equation as
1
∇P = J × B, (5.31)
c
we immediately see that
J · ∇ P = B · ∇ P = 0.
Since ∇ P is perpendicular to the surfaces P = constant, we conclude that both

vectors B and J lie on those surfaces. Moreover, Eq. (5.31) shows that even in the
presence of a magnetic field, the magnetic force may vanish. This certainly happens
when J = 0, in which case the field is said to be a potential field. In fact, if J vanishes
we must have ∇ × B = 0 and the field can be represented as the gradient of a scalar
function, the magnetic potential. However, the Lorentz force can also vanish if J = 0
when J and B are parallel to each other. These particular magnetic configurations
are called force-free fields. We shall now examine some of their properties, before
considering cases in which the magnetic force is different from zero.
5.3.1 Force-Free Equilibria
Force-free fields are a common feature of rarefied plasmas whenever the pressure
gradients are sufficiently small to be neglected. The equilibrium equation, J × B ∝
(∇ × B) × B = 0, thus implies:
∇ × B = α B, (5.32)
where, in general, α = α(r).

Notice that Eq. (5.32) does not come from neglecting the pressure term “with
respect to” the magnetic term, and then proceeding to state that the latter is zero.
Force-free equilibria arise when both the pressure gradients and the Lorentz force
separately vanish. However, it is often said in the litterature that magnetic fields are
force-free whenever β 1, reminiscent of the statement above. What is really meant

is that if the plasma pressure is everywhere much smaller than magnetic pressure,
it is unlikely that gradients in the plasma pressure can supply the force required to
balance any significant Lorentz force. In other words, one generally should have
c |∇ P| 4π |∇ P|
= 1.
| J × B| |(∇ × B) × B|
It is clear that if we add to P or to B a constant pressure or a constant magnetic field,

we change the value of β, but not that of the ratio of their gradients. We thus may
have a force-free field also in situations where β > 1.
Returning now to Eq. (5.32), we illustrate some of the properties of force-free
fields. Taking the divergence of Eq. (5.32) we find that
∇ · (∇ × B) = ∇ · (α B) = α∇ · B + B · ∇α = 0,
implying
B · ∇α = 0.
Constancy along magnetic field lines is the only constraint satisfied by the function
α(r). The preceding equation also shows that B must lie on the surfaces α = const..
Writing the vector Eq. (5.32) in terms of components, we immediately see that it
couples different components of B. If α is a constant independent of the coordinates,
it is easy to obtain an equation for a single component of B by taking the curl of
Eq. (5.32):
∇ × ∇ × B = −∇ 2 B = α(∇ × B) = α 2 B. (5.33)
However, since Eq. (5.33) is a differential equation of order higher than Eq. (5.32),
it is necessary to verify that the solutions of Eq. (5.33) also satisfy Eq. (5.32).
As a first example of a force-free field in plane geometry with α = const.,
consider a field given by B = B(x) = [0, B y (x), Bz (x)], which obviously satisfies
the condition ∇ · B = 0. It is easy to check that, with a proper choice of the initial
conditions, the solutions of Eq. (5.33) that are also solutions of Eq. (5.32) can be
written as
B y = B0 cos(α x); Bz = −B0 sin(α x). (5.34)
This is a force-free field whose lines of force are straight lines in every plane x =
const.. However, their inclination with respect to the y axis changes, making a
complete turn when x is increased by 2π/α.
A more interesting example is given by axially symmetric force-free fields, in
which case the fields always have a helical structure. Axial symmetry requires that
the components of B depend only on the radial coordinate r , the distance from
the axis of symmetry, and the condition ∇ · B = 0 implies that Br = 0 and thus
B = B(r ) = [0, Bθ (r ), Bz (r )]. The field lines therefore wind on coaxial cylindrical
surfaces, forming a helical structure. If the pitch of the helix, κ, is defined as the
distance covered in the z direction by a representative points performing a complete

revolution along a field line, it is easy to see that κ(r ) = 2πr Bz (r )/Bθ (r ). In
cylindrical geometry Eq. (5.33) read:

1 d dBθ 1
r + α − 2 Bθ = 0,
2
r dr dr r
and
1 d dBz
r + α 2 Bz = 0.
r dr dr
The solutions of the above equations are:
Bθ = B0 J1 (αr ) e Bz = B0 J0 (αr ), (5.35)
where Jn are the Bessel functions of order n. As expected, the field lines are helices
whose pitch varies with r . On the axis (r = 0) the field is directed along z (J1 (0) = 0);
as r increases, an azimuthal component appears and the helix pitch decreases until,
in correspondence with the first zero of J0 , the axial component vanishes and the
field is totally azimuthal. For larger values of αr , the field is still a helix, but with the
opposite sense of rotation because J0 changes sign. If r is further increased, the first
zero of J1 is encountered and the field is again purely axial, but its sense is opposite
to that of the field on the axis.
An example of a force-free field with α = const. is given by
kr 1
Bθ = B0 ; Bz = −B0 . (5.36)
1 + k 2r 2 1 + k 2r 2
The z-component of ∇ × B is given by
1 d 2k
(∇ × B)z = (r Bθ ) = −B0 = α(r )Bz ,
r dr (1 + k 2 r 2 )2
where
2k
α(r ) = .
1 + k 2r 2
Therefore this field can be considered a force-free field with α = const. at least for
the z-component. It is however easy to verify that the equation for the θ -component
is also satisfied by this choice of α. A peculiar feature of this field is to have a pitch
independent of r .
It is also possible to find simple solutions of Eq. (5.32) in two dimensions. For
instance, it is easily shown that in plane geometry the field
Bx = −(l/k)B0 cos(kx)e−lz
B y = −(1 − l 2 /k 2 )B0 cos(kx)e−lz

Bz = B0 sin(kx)e−lz
is a force-free field with α = (k 2 − l 2 )1/2 = const. This field has been used as a
model for the magnetic arcades observed in the solar corona.
Force-free fields have been the object of remarkable interest. One of the reasons
that justify such an interest is connected with
Woltjer’s theorem: For a given set of boundary conditions, the plasma state with the
minimum magnetic energy is a force-free field with α = constant. To demonstrate
this important theorem we must first introduce the concept of magnetic helicity, a
quantity defined by:
H= A · B dV, (5.37)
V
where A is the potential vector, B = ∇ × A. The evolution equation for A is easily

obtained (see Problem 5.2) by introducing this representation of B into the induction
equation:
∂A
= U × (∇ × A). (5.38)
∂t
An important property of magnetic helicity is to remain constant in time for an
isolated system, provided that A does not vary on the surface S that encloses the
volume V . In fact,
dH ∂A ∂B
= ·B+ · A dV,
dt ∂t ∂t
V
and, making use of the vector identity
∇ · (F × G) = G · (∇ × F) − F · (∇ × G), (5.39)
with F = ∂ A/∂t and G = A, we obtain:

dH ∂A ∂A
= ∇· × A +2 · (∇ × A) dV.
dt ∂t ∂t
V
The first term can be transformed into an integral over the surface S where, by assump-
tion, ∂ A/∂t = 0 while the second term vanishes because of Eq. (5.38). Therefore,
dH
= 0.
dt
By exploiting this property of H we are now in the position to demonstrate Woltjer’s
theorem. Equilibrium states in MHD are generally found by minimizing the magnetic
energy of a system, W B . However, in our case we must take into account the invariance
of H, so that the minimization problem is constrained by the conservation of magnetic
helicity, rather than free. As is well known, this class of problems can be solved by
applying the method of Lagrangian multipliers. This consists in looking for the
minimum of (W B − λH), where λ is a constant, and imposing the condition that
A not vary on the surface S. Writing for convenience λ = α0 /8π the variational
problem becomes:

1
0 = δ(W B − α0 /8π H) = δ (B 2 − α0 A · B) dV
8π
V

1 α0
= ( A · δ B + B · δ A) dV
B · δB −
4π 2
V

1 α0 α0
= (B − A) · δ B − B · δ A dV,
4π 2 2
V
Since
δ B = ∇ × δ A,
making use again of Eq. (5.39) with F = (B − (α0 /2) A) and G = δ A, we obtain

∇ · [B − (α0 /2) A] × δ A dV + [∇ × B − α0 B] · δ A dV = 0
V V
The first integral vanishes since it can be transformed into a surface integral over S
where by assumption δ A = 0. The equilibrium condition thus requires the vanishing
of the second integral for every δ A or
∇ × B = α0 B,
as required by Woltjer’s theorem. It is also possible to show that the configuration

having the minimum magnetic energy, among all those compatible with the given
boundary conditions, corresponds to a potential field α0 = 0.
5.3.2 Equilibria in the Presence of Magnetic Forces
If more general magnetic configurations with non-vanishing magnetic forces are

considered, the equilibrium equation reads:

B2 1
∇ P+ = (B · ∇)B. (5.40)
8π 4π
We shall now briefly discuss some solutions of the preceding equation, restricting
ourselves to cylindrical geometry.
Given a magnetic field B = [0, Bθ (r ), Bz (r )] Eq. (5.40) reduces to

d Bθ2 + Bz2 1 Bθ2
P+ =− . (5.41)
dr 8π 4π r
We have only one differential equation for the three unknown functions P(r ), Bθ (r )
and Bz (r ). If any two of them are prescribed, together with the appropriate boundary
conditions, the third can be found from Eq. (5.41). It is therefore clear that an infinite
number of solutions exists; we shall illustrate some examples of magnetic configura-
tions particularly interesting for the plasma confinement in fusion machines, usually
called pinches.
• Theta-pinch, Bθ = 0.
Such a configuration can be produced by currents flowing in the azimuthal direction
θ on the surface of a plasma column. In this case the solution of Eq. (5.40) is simply
given by
B2
P+ = const.
8π
The value of the constant is determined by the radial boundary conditions. If, for
instance, the plasma column has a radius a and is surrounded by a vacuum (or by a
medium of negligible pressure) and outside of the column a constant axial magnetic
field is present, Bz = B0 , the equilibrium condition becomes
B2 B2
P+ = 0.
8π 8π
We can easily figure out why this is an equilibrium configuration: the force J × B,
directed towards the axis of the cylinder, pinches the plasma and compensates the
push produced by the internal pressure. Notice that this solution is valid also in
configurations that do not show an axial symmetry: it suffices that (B · ∇)B = 0,
which happens, for instance, when the field lines are straight.
• Zeta-pinch, Bz = 0.
This configuration is obtained from the preceding one by switching the roles of J
and B. We thus assume an axial current flow in a column af radius a, surrounded
by a vacuum. We may envisage such a current as formed by many wires in which
parallel currents flow. Since, as is well known, parallel currents attract each other,
the joint effect will again be the squeezing of the plasma column. The equilibrium
condition now reads:
dP 1 d
=− Bθ (r Bθ ). (5.42)
dr 4π r dr
Multiplying both members of the equation by r 2 and integrating between 0 and a,

we obtain:
a a
dP 1 d
r2 dr = − r Bθ (r Bθ )dr.
dr 4π dr
0 0
Performing the integrations and taking into account the boundary conditions we get:
a
1
2 r Pdr = [a Bθ (a)]2 .
8π
0
If the plasma can be considered a perfect gas, P = nkT , the lhs can be written as
a a
1 kT N T
2πr (nkT )dr = 2πr ndr = ,
π π π
0 0
where the temperature T is assumed to be constant inside the plasma column and we
have introduced the quantity N , representing the number of particles per unit length
of the column (linear density):
a
N = 2πr ndr.
0
The integral on the rhs can be transformed by noticing that the intensity of the current
flowing in the plasma column is:
a
I = 2πr Jz dr,
0
and that
c 1 d
Jz = (r Bθ ),
4π r dr
so that finally
2I
= a Bθ (a).
c
Making use of the above relations we obtain:
I 2 = 2kT N c2 ,
known as the Bennet’s relation. Observe that this relation is independent of the
details of the pressure profile. The latter can be deduced directly from the equilibrium
equation (5.42), for instance by assuming that Jz is constant inside the plasma column
and zero outside. In this case Bθ is equal to
2I
Bθ = r r ≤ a,
a2
and the pressure profile is given by
2
1 I r2
P(r ) = 1− 2 .
π a a
5.4 Perturbed Equilibrium States
The existence of an equilibrium state is not by itself a guarantee that the equilib-
rium will not change with time: an equilibrium results from a precise balance of the
forces acting on the system, which in turn implies a well defined set of values of the
parameters characterizing the system. If one or more of those values are changed,
the equilibrium cannot be maintained and the system evolves dynamically. If only
small perturbations of the equilibrium parameters are considered, there are essen-
tially two possibilities: either the resulting force is such as to push to restore the
system to equilibrium or it causes a further displacement from it. In the first case, an
oscillatory regime sets in and the amplitude of the perturbations remains small: the
equilibrium is then said to be stable. In the second case, the amplitude of the pertur-
bation increases and the equilibrium is said to be unstable. The intermediate case, in
which perturbations do not alter the balance of forces, corresponds to a marginal or
neutral equilibrium.
In principle, to determine the existence of an equilibrium one should check that
all the forces acting on the system balance, i.e. the resultant of all forces should van-
ish. In practice, this is often not possible and to establish the equilibrium condition
only the dominant forces are taken into account. Such an “equilibrium” therefore
does not represent the real situation, where other, smaller, forces are present. As a
consequence, equilibrium conditions are never perfectly satisfied and reality is bet-
ter represented by a (theoretical) equilibrium with perturbations, a situation which
makes a stability analysis of the system a requirement. In fact, if we want to set up an
experimental or observational check of the actual existence of an equilibrium state,
we are forced to restrict our check to stable, or moderately unstable, systems. By this
we mean that the rate of change with time of the parameters that characterize the
system must be such that their values will not be substantially altered during their
measurement. It follows that the physical concept of equilibrium (not the mathe-
matical one!) makes sense only when referred to a well defined timescale. In other
words, we are not interested in finding equilibria that will last for eternity: we only
5.4 Perturbed Equilibrium States 95
want to verify the existence of states that do not undergo substantial changes over
what we consider the relevant timescales.
The analysis of the dynamical evolution of an equilibrium subject to small per-
turbations, the so-called linear stability analysis, can be performed by using two
different methods. The first, the normal modes method, basically consists in the
study of the dynamics of the perturbed system. This not only determines whether a
system is stable or not, but also characterizes the dynamics of small perturbations,
namely the characteristic frequencies of oscillation in the case of stable systems or the
growth rates of the amplitude of perturbations in the unstable case. The assumption
that perturbations are small allows significant mathematical simplification coming
from linearization of the equations. In the stable case perturbations remain small and
the solutions found are valid also for long time spans. In the unstable case, pertur-
bations grow, typicall exponentially, and the validity of solutions is limited to short
time periods.
The second method, called the energy method, is a variational method generalizing
the well known result of classical mechanics that equilibrium states correspond to
energy extrema, with stable equilibria localized at energy minima and unstable ones
at energy maxima. The energy method establishes if a system is stable or not, but
does not yield any detailed information on the system’s dynamics.
It is important to keep in mind that a system can be considered stable only if
stability is achieved for any possible perturbation, provided that it is compatible
with the constraints imposed on the system, but must be considered unstable even if
a single growing perturbation is found.
In the following we shall utilize only the method of normal modes, that basically
consists of the following steps.
• All f entering the system of equations that governs the system’s dynamics, are
expanded as
f = f0 + f1 ,
with 1. f 0 corresponds to the equilibrium state and f 1 to the perturbation.

• This representation is introduced into the equations and terms of order higher than
the first one in are discarded.
• The equation are written order by order, with terms of zeroth order separated
from those of first order and the resulting equations are separately solved. The
first order equations, evidently linear in the quantities f 1 , determine the dynamics
of perturbations. The coefficients of those linear equations are functions of the
unperturbed quantities and of the other parameters that may enter in the starting
set of equations.
This scheme is equally valid for any plasma model, both kinetic and fluid.
5.1. Show how a dimensional estimate of the Hall term in Eq. (5.5) leads to the
inequality Eq. (5.8).
5.2. Justify Eq. (5.38) using the gauge invariance properties
√ of Maxwell’s equations.
5.3. Introducing the (Elsasser) variables: Z ± = U ± B/( 4πρ), show that the MHD
equations for an incompressible ideal plasma can be written as:

∂ Z± 1 B2
+ (Z ∓ · ∇)Z ± = − ∇ P + .
∂t ρ 8π
5.4. (a) A plasma column with density n = 1015 cm−3 is placed inside a medium
of negligible density, or “vacuum”, where a constant magnetic field, B, is present.
Determine the value of B necessary to confine a plasma whose temperature is 5 ×
106 K . (b) If in a Zeta-pinch with a linear density of 1018 particles/cm circulates a
current of 106 A, evaluate the plasma temperature (beware of the units!).
5.5. Compute the helicity for the following field configurations: (a) a Theta-pinch,
(b) a Zeta-pinch, (c) the field given by Eq. (5.36). Show that the answer to question
(c) applies to any helical field of constant pitch.
5.6. Compute the helicity of a force-free field with constant α in plane and in cylin-
drical geometries.
Solutions
5.1. Proceeding by dividing the Hall term with the electric field magnitude, or equiv-
alently multiplying it by c/(UB), we obtain
2
1 c cm p B 2 1 ω ca
J×B =ω = ,
en e c i UB eB n e m p U 2 ωci U
from which inequality Eq. (5.8) follows trivially.

5.2. If we add to the potential vector A the gradient of a scalar function the magnetic
field associated with A does not change. In fact, if A = A + ∇ψ, we clearly have
B = ∇ × A = ∇ × A . If A is replaced by A in Eq. (5.25), taking into account that
∇ × ∇ψ = 0, we immediately obtain the requested result.
5.3. In an incompressible plasma the density can be considered to be uniform, in
which case the continuity equation is automatically respected and r ho may therefore
be passed into any differential operator as all√ its derivatives vanish. Therefore, we
divide all terms in the induction equation by 4πρ, while we divide the momentum
equation by ρ. We then simply add the momentum and induction equation to obtain
the equation for Z + , while subtracting the induction equation from the momentum
5.4 Perturbed Equilibrium States 97
equation we obtain the equation for Z − :
∂ Z± 1 B2
+ (Z ∓ · ∇)Z ± = − ∇ P + .
∂t ρ 8π
5.4. (a) B 4.16 × 103 G. (b) T 36 × 106 K .

5.5. From B = ∇ × A we deduce the following relations:
r r
1
Az = − Bθ dr ; Aθ = r Bz dr.
r
0 0
In the definition of the helicity, the integrand is: A · B = Aθ Bθ + A z Bz , which turns

out to vanish for all cases. Moreover, for any helical field,
r r r r
Bθ Bθ Bθ
A· B = r Bz dr − Bz Bθ dr = κ Bθ dr − κ Bθ dr,
r 2πr 2πr
0 0 0 0
which vanishes for κ = const.

5.6. In plane geometry we have A · B = (2B02 /α) cos2 (α x/2). Therefore H = 0
and has the same sign of α. The helicity normalized to the volume is H/V = B02 /α.
In cylindrical geometry, making use of the properties of Bessel functions [see e.g.
Abramowitz and Stegun, Handbook of Mathematical Functions], we have
r
B0 B0
Aθ = r J0 (αr )dr = J1 (αr ),
r α
0
r
B0
A z = −B0 J1 (αr )dr = J0 (αr ) − 1 .
α
0
Therefore,
B02 2
A· B = J0 (αr ) + J12 (αr ) − J0 (αr ) .
α
In this case, it can be shown, always making use of the properties of Bessel functions,
that the normalized helicity depends on the cylinder’s radius, but its sign is still defined
by that of α.
Chapter 6
Instabilities
Abstract A survey of the most important linear instabilities of the fluid plasma
described by MHD is presented. We start by considering instabilities in planar geom-
etry when gravity is present, including the well-known Rayleigh-Taylor and Kelvin-
Helmholtz instabilities, and discuss the stabilizing effects of magnetic fields in these
cases. The instabilities stemming from the combined effects of currents, current gra-
dients, and pressure in the ideal plasmas are then described, focusing on cylindrical
geometry. Finally, the Parker or buoyancy instability of the galactic magnetic field
and the magneto-rotational instability of accretion disks, topics of contemporary
astrophysical relevance, are discussed in some detail.
In the preceding Chapter we have emphasized the role and importance of stability
analysis for understanding possible plasma equilibrium states in the MHD limit and
illustrated one of the possible approaches to the problem. This chapter is devoted to
the study of the linear stability of systems described by the MHD equations using that
approach, consisting in solving the eigenvalue problem and searching for the growth
rates of small perturbations. As a result of our analysis, we shall be able to say if the
perturbed system will show a tendency to return to the equilibrium position, giving
rise to oscillatory dynamics, or to move away from it. However, nothing can be said
in general about the final state of the system, which might be a different (stable)
equilibrium state, but also may not exist, leading to a dynamical evolution followed
by drastic changes.
An example taken from astrophysics is that of a white dwarf with a mass in
excess of the Chandrasekhar limit. If the composition and the equation of state
are those usually assumed for white dwarfs, a stable configuration simply does not
exist. However, if the composition changes into that typical of neutron stars and
the equation of state is accordingly modified, a new stable equilibrium is finally
achieved. The composition change is due to the neutronization process, a process
completely extraneous to the physical mechanisms determining the equilibrium of a
white dwarf.
The importance of stability analysis increases with the number of degrees of
freedom of the system. In a system with one degree of freedom, its energy will be a
function of only one parameter and equilibria will generically correspond to energy

100 6 Instabilities
maxima or minima, so that the number of stable equilibria corresponds to a fraction

1/2 of the total. In a system with two degrees of freedom, the energy becomes a
surface in the two-parameter space and there will be four possible extrema, namely
an absolute maximum (a mountain top), an absolute minimum (a valley) and two
saddle points (two mountain passes). These latter states are unstable since only for
a subset of perturbations along a particular direction will the system return to the
saddle-equilibrium, while any noise will lead to instability. Therefore, over four
possible equilibrium configurations, only one is stable. One might be tempted to
generalize assuming that for a system with n degrees of freedom only a fraction 1/2n
of the total number of equilibria is stable, a conjecture for which there is no proof.
Nonetheless it is a reasonable intuition that in a system with a very large number
of degrees of freedom, such as a fluid or a plasma, most equilibrium configurations
might be unstable, making stability analysis a requirement.
6.1 Linear Stability of Ideal MHD Equilibria
Consider the ideal MHD equations:
∂ρ
+ ∇ · (ρU) = 0
∂t
dU ∂U 1
ρ = ρ( + (U · ∇) U)) = −∇ P + (∇ × B) × B + f (6.1)
dt ∂t 4π
d
(Pρ −γ ) = 0
dt
∂B
= ∇ × (U × B)
∂t
In agreement with the assumption of small perturbations, expand any physical
variable h, entering the preceding equations as:
h = h 0 + h 1 and 1,
where h 0 represents the equilibrium value and h 1 the perturbation. In the unper-
turbed state for static equilibria, the velocity vanishes and consequently U = U 1
is a first order quantity. Taking all this into account, the following equations result
from an order by order expansion:
Zeroth order:
∂ρ0 ∂ B0
= = 0,
∂t ∂t
−γ
P0 ρ0 = const.
6.1 Linear Stability of Ideal MHD Equilibria 101
1
0 = −∇ P0 + (∇ × B 0 ) × B 0 + f 0 . (6.2)
4π
First order:
∂ρ1
+ (U · ∇) ρ0 + ρ0 (∇ · U) = 0, (6.3)
∂t
∂U 1
ρ0 = −∇ P1 + [(∇ × B 0 ) × B 1 + (∇ × B 1 ) × B 0 ] + f 1 (6.4)
∂t 4π
−γ d P1 −γ −1 dρ1
ρ0 − γ P0 ρ0 = 0, (6.5)
dt dt
∂ B1
= ∇ × (U × B 0 ). (6.6)
∂t
Introducing the sound speed cs ,

P0
cs2 = γ ,
ρ0
and making use of Eq. (6.3), Eq. (6.5) can be writen as

∂ P1
+ (U · ∇)P0 + cs2 (∇ · U)ρ0 = 0, (6.7)
∂t
The linearized MHD equations can be simplified by introducing the concept of
lagrangian displacement , ξ , and by properly choosing initial conditions. The instan-
taneous position of a fluid element may be characterized by:
r(t) = r 0 + ξ (r 0 , t),
where, thanks to the linearization, ξ may also be taken to be a small, first order,
quantity. Therefore,
dr dξ ∂ξ ∂ξ
U= = = + (U · ∇)ξ .
dt dt ∂t ∂t
Inserting this expression in Eq. (6.3) we obtain:
∂
(ρ1 + ∇ · (ρ0 ξ )) = 0,
∂t
102 6 Instabilities
or, upon integration in time
ρ1 + ∇ · (ρ0 ξ ) = const.
The initial conditions have been chosen, without loss of generality, so that all
perturbed quantities except ξ̇ (r 0 , 0) vanish everywhere at t = 0.1 With this choice,
the constant of the preceding equation vanishes and therefore
ρ1 = −(ξ · ∇) ρ0 − ρ0 (∇ · ξ ). (6.8)
Applying the same procedure to Eqs. (6.5) and (6.6) we obtain:
P1 = −(ξ · ∇)P0 − ρ0 cs2 (∇ · ξ ), (6.9)
and
B 1 = ∇ × (ξ × B 0 ), (6.10)
while the momentum equation, Eq, (6.4), becomes
∂ 2ξ
ρ0 = F(ξ ), (6.11)
∂t 2
where the “force per unit volume”, F(ξ ) is given by:
1
F(ξ ) = −∇ P1 + [(∇ × B 0 ) × B 1 + (∇ × B 1 ) × B 0 ] + f 1 , (6.12)
4π
with P1 , and B1 given by Eqs. (6.9) and (6.10).
Equation (6.11) is the starting point for the application of the method of normal
modes: it describes the temporal evolution of displacements from the equilibrium
position under the action of the force F(ξ ). F(ξ ) is a second order differential
operator [∇ × B 1 = ∇ × ∇ × (ξ × B 0 )] involving only spatial coordinates, which
may be written formally as an operator F̂ r acting on [ξ (r, t)]
F(ξ ) = F̂ r [ξ (r, t)].
Equation (6.11) is a partial differential equation in the independent variables r, t

and a standard method of solution for this type of equations, in particular when
the coefficients do not depend on the independent variables, is that of the Fourier
transform (that will be discussed in more detail in Chap. 7). Presently, since the
equilibrium may have spatial gradients but does not depend explicitly on time, it
1Think of the case of a pendulum: it is completely equivalent to move the mass from the equilibrium
position and let it go or to leave it there and give it a push.
6.1 Linear Stability of Ideal MHD Equilibria 103
is useful to carry out the Fourier transform with respect to time of ξ (r, t), ξ̃ (r, ω),
defined by
∞
1
ξ̃ (r, ω) = ξ (r, t) eiωt dt, (6.13)
2π
−∞
which implies that

∞
ξ (r, t) = ξ̃ (r, ω) e−iωt dω. (6.14)
−∞

as can be easily verified multiplying Eq. (6.14) by eiω t , integrating in dt and taking
into account that
∞

e−i(ω −ω)t dt = 2π δ(ω − ω).
−∞
Using Eq. (6.14), the equation of motion, Eq. (6.11), becomes:
∞ ∞
−iωt
− ρ0 ω ξ̃ (r, ω) e
2
dω = F̂ r [ξ̃ (r, ω)] e−iωt dω.
−∞ −∞
Because the coefficients of the unknown function ξ and its derivatives are functions of
the equilibrium quantities, which by definition are time-independent, the preceding
expression can be written as:
∞ ∞
−iωt
−ρ0 ω ξ̃ (r, ω) e
2
dω = F̂ r ξ̃ (r, ω) e−iωt dω .
−∞ −∞

If we multiply both members by eiω t and integrate over dt we finally get
− ρ0 ω2 ξ̃ (r, ω ) = F̂ r [ξ̃ (r, ω )]. (6.15)

We have obtained an equation for the Fourier transform of ξ (r, t) which has the
same form of Eq. (6.11), with the operator ∂ 2 ξ /∂t 2 replaced by the multiplicative
factor −ω2 . In this form this becomes an eigenvalue equation, the eigenvalues being
here represented by −ρ0 ω2 . Once this equation has been solved, the unknown
function ξ (r, t) can be obtained by the simple application of the inverse Fourier
transform, as defined by Eq. (6.14).
The Fourier transform can be applied also to those spatial coordinates, ignorable
coordinates, that do not appear explicitly in the coefficients of the equations, i.e. in
104 6 Instabilities
the unperturbed quantities. This happens when the equilibrium state exhibits some
symmetry property. Let s be the set of ignorable coordinates, r the set of the remain-
ing coordinates and k a vector with non vanishing components only in the directions
of the ignorable coordinates. The Fourier representation of ξ (r, t) then has the form

ξ (r, t) = ξ (r , s, t) = ξ̃ (r , k, ω) ei(k·s−ωt) dω dk. (6.16)
Repeating the procedure adopted for the time transform, it is easy to realize that
the application of the Fourier transform method is equivalent to the substitution
(restricted to ignorable coordinates)
∂
→ −iω, ∇ → ik, (6.17)
dt
and to the replacement of the unknown function with its Fourier transform.
It is possible to show that in the absence of dissipative terms the operator F̂ r
is a hermitian or self-adjoint operator, which implies that the eigenvalues are real.
Thus, in ideal MHD, ω2 is a real number. If ω2 is positive, ω must be real and ξ ,
whose components are proportional to e−iωt represents an oscillation around the
equilibrium position. A negative ω2 however implies a purely imaginary ω, with
consequent growth of perturbation amplitudes and therefore instability. Therefore,
the stability of an equilibrium will be completely determined by the sign of ω2 .
6.2 Instabilities in the Presence of Gravity
The most common force acting on astrophysical plasmas (apart from pressure gra-
dients and magnetic field forces) is gravity. This means that the term f entering the
second equation of (6.1), i.e. the momentum equation, has the form ρg where g is
the local value of the gravitational acceleration. For the sake of simplicity, we shall
assume that the plasma contribution to the mass of the system is small, i.e. the plasma
is not self-gravitating but is rather immersed in the gravitational potential of a more
extended mass (such as for example, the plasma of a stellar corona). Therefore, g
will be considered an external field, independent of the plasma perturbations. In the
linearization process therefore one has g 1 = 0 and Eq. (6.12) becomes:
1
F(ξ ) = −∇ P1 + [(∇ × B 0 ) × B 1 + (∇ × B 1 ) × B 0 ] + ρ1 g, (6.18)
4π
with ρ1 , P1 , and B1 given by Eqs. (6.8), (6.9) and (6.10), respectively.
Moreover, we shall assume that the plasma perturbations are incompressible and
that g and B 0 are constant. Because incompressible perturbations can be viewed as a
subset of all possible perturbations, if a system is unstable to such modes it will also
6.2 Instabilities in the Presence of Gravity 105
be unstable more generally. We can choose our system of reference with g directed
along the negative z-axis. In this frame, the equilibrium equation, ∇ P0 = ρ0 g,
implies that both P0 and ρ0 are functions of z only and that
d P0 (z)
P0 ≡ = −ρ0 (z)g.
dz
The x and y coordinates are therefore ignorable. If the system is assumed to be

homogeneous in the y-direction, we may restrict our attention to perturbations of
the type ξ = [ξx , 0, ξz ] and Fourier expand along x all the quantities entering the
equations, writing for the single Fourier component (see Eq. (6.16)):

ξ (r, ω) = ξ̃ (z, k, ω)eikx dk.
The vectors k = [k, 0, 0] and g define the coordinate plane (x, z). In the reference
system [x, y, z] the vector B 0 can have any orientation, but we can safely put B0z = 0,
since it is easy to realize that a (constant) B0z has no influence on the linear stability
of the system. We shall therefore assume that B 0 lies in the (x, y) plane, B 0 =
[B0x , B0y , 0]. Making use of Eqs. (6.8), (6.9) and (6.10) we may cast Eq. (6.15) in
the form:
1
− ω2 ρ0 ξ = ∇(ξ · ∇ P0 ) − g(ξ · ∇ρ0 ) + ∇ × [∇ × (ξ × B 0 )] × B 0 , (6.19)
4π
where, to simplify the notation, we have written ξ instead of ξ̃ (z, k, ω) and we have
used the incompressibility condition ∇ · ξ = 0. Observe that now the vector ∇ is
represented by ∇ = [ik, 0, ∂/∂z].
The magnetic term in Eq. (6.19) can be transformed into:
1 1
∇ × [∇ × (ξ × B 0 )] × B 0 = ∇[∇ · (ξ × B 0 )] − ∇ 2 (ξ × B 0 ) × B 0 ,
4π 4π
and a long, but straightforward, calculation gives simply:
2
B0x (−k 2 ξz + ξz ) ez ,
where primes indicate derivatives with respect to z.

The final equation for ξ = ξ̃ (z, k, ω) is thus seen to be:
− ω2 ρ0 ξ = ∇(ξz P0 ) − g(ξz ρ0 ) + (B0x

2
/4π )(−k 2 ξz + ξz )ez , (6.20)
whose x and z components are:
−ω2 ρ0 ξx = ik(ξz P0 ) (6.21a)

106 6 Instabilities
−ω2 ρ0 ξz = (ξz P0 ) + (B0x

2
/4π )(−k 2 ξz + ξz ) + g(ρ0 ξz ). (6.21b)
Equations 6.21a, 6.21b must be complemented by the incompressibility condition,
ikξx + ξz = 0. (6.22)
We now combine the latter expression with Eq. (6.21a) to eliminate (ξz P0 ) and
finally obtain:
B2 k2
ω2 (ρ0 ξz ) − ρ0 k 2 ξz = 0x (ξz − k 2 ξz ) + k 2 g(ρ0 ξz )
4π
(k · B 0 )2
= (ξz − k 2 ξz ) + k 2 g(ρ0 ξz ). (6.23)
4π
We shall now turn to the illustration of some consequences of the eigenvalue equation
just found.
6.2.1 Rayleigh-Taylor Instability

Consider first the development of instability when the equilibrium magnetic field
vanishes, known as the Rayleigh-Taylor instability. An approach which quickly leads
to the stability conditions consists multiplying Eq. (6.23) by ξz and integrating in in
dz between −∞ and ∞, to obtain:
⎡ ∞ ⎤
∞ ∞
2⎣
ω
[ρ0 ξz ] ξz dz − k 2
ρ0 ξz dz ⎦ = k g
2 2
ρ0 ξz2 dz.
−∞ −∞ −∞
Performing an integration by parts on the first integral in square brackets and assum-
ing that ξz and/or ξz vanish for z = ±∞ leads to:
∞
ρ0 ξz2 dz
−∞
ω = −k g
2 2
∞ . (6.24)
ρ0 [ξz2 + k 2 ξz2 ] dz
−∞
The same equation can be obtained by applying the energy method, which, as we have
remarked, is a variational method. Equation (6.24) shows that all the terms appearing
under the integral sign are positive definite, except ρ0 . The sufficient condition to
have stability, i.e. ω2 > 0 is therefore ρ0 < 0 for every z, namely a density which
always decreases with height. At first sight, one may think that this condition is not
a necessary one. In fact, ω2 > 0 does not imply ρ0 < 0 everywhere, since ρ0 is
weighted in the integral by ξz2 and if ρ0 is positive only in a certain range, the integral
at the numerator of Eq. (6.24) could still be negative, Nonetheless, it is possible to

show that the condition ρ0 < 0 for every z is also a necessary one, as we shall now
see.
First observe that the expression for ω2 given by Eq. (6.24) is a presently only
formal solution of Eq. (6.23), since it contains the quantities ξz and ξz that remain
unknown until Eqs. (6.21) have been explicitly solved. In the theory of calculus of
variations it is proved that, if in Eq. (6.24) the functions ξz and ξz , solutions of the
original equation, are replaced by arbitrary functions (satisfying the correct boundary
conditions, in this case convergence properties at infinity), Eq. (6.24) gives us a value
ω̃2 larger than the correct one (namely the one we would have obtained by using the
exact solutions):
ω̃2 ω2 ,
where the equal sign holds only when the correct solutions of the problems are
inserted into the integral.
Assume now that ρ0 > 0 only in a finite interval, with ω2 still positive. The
variational property mentioned above tells us that we are free to choose arbitrary
functions that are different from zero only within that interval. In this case, we would
obtain a negative value for ω̃2 . But then, according to the previous inequality, ω2
would also be negative, contrary to our assumption. It is therefore impossible to have
ω2 > 0 even for a limited interval over which ρ0 > 0, Therefore ρ0 < 0 for every z
is also a necessary condition for stability.
A simple and interesting case is that of two incompressible fluids of different
densities lying one atop of the other, i.e.
ρ = ρ1 = const., z > 0
ρ = ρ2 = const., z < 0.
In this model the derivative of the density is discontinuous in z = 0 and can be

represented as: ρ0 = (ρ1 − ρ2 ) δ(z). Consequently, Eq. (6.23) can be written, for
z = 0,
ξz = k 2 ξ,
whose solutions are:
ξz = ξz (0)e−kz , z > 0 (6.25)

ξz = ξz (0)e , z < 0.
kz
(6.26)
We are now in a position to explicitly evaluate ω2 by using Eq. (6.24). The integral
in the numerator gives simply
(ρ1 − ρ2 ) ξz2 (0).

108 6 Instabilities
The integral in the denominator can be evaluated as follows:
∞ 0− ∞
[ξz2 + k 2 ξz2 ] dz = ρ2 [ξz + k ξz ] dz + ρ1 [ξz2 + k 2 ξz2 ] dz
2 2 2
−∞ −∞ 0+
∞
= 2(ρ1 + ρ2 ) k 2 e−2kz dz = k(ρ1 + ρ2 ) ξz2 (0),
0+
where the correct expressions for ξz and ξz have been used in each of the integration
intervals. Therefore:
ρ1 − ρ2
ω2 = −kg . (6.27)
ρ1 + ρ2
An instability will set in whenever the fluid of higher density lies above that of lower
density, in agreement with the general criterion discussed above. When ρ1 < ρ2 the
system will be stable and the frequency of the oscillation will be given by Eq. (6.27).
The maximum value of ω is reached when ρ1 ρ2 , in which case

ω= kg.
This is the first example of a dispersion relation, a relationship that expresses ω as a

function of the wavevector k. As we shall discuss in more detail in the next Chapter,
the stable oscillations described√ above are an example of dispersive waves with a
phase velocity vϕ ≡ ω/k = 21 g/k.
The dispersion relation derived above is also valid for surface ocean waves, that
develop at the interface between the water and the atmosphere. The description of air
as an incompressible fluid, may throw some legitimate doubt on the applicability of
our result to sea waves. However, every fluid actually behaves as an incompressible
one when the velocity of the fluid particles is negligible with respect to the sound
speed. Notice that the dispersion relation we have deduced refers to the case of
deep water, since we have imposed as a boundary condition the vanishing of ξz for
z → −∞. If the water has a depth h the dispersion relation is modified by the fact
that boundary conditions must be imposed in z = −h.
Even if the linear analysis does not determine the complete temporal evolution
of the system, it is easy to visualize the initial phases of such evolution. If in the
(unstable) equilibrium the dense gas lies above the light one, any perturbation will
deform the originally flat surface separating the two fluids into a wavy surface with
portions of lighter gas rising above of the equilibrium surface (z = 0) and portions
of the heavier gas descending below that level. The distance between two successive
regions of dense gas is of the order of λ = 2π/k. In the subsequent phases the vertical
extension of those regions increases and the gas structure is modified until a complete
overturning of the initial configuration is reached, with the lighter gas lying above
Fig. 6.1 Numerical

simulation of the
development of the
Rayleigh-Taylor instability
the denser one. The study of the nonlinear phase of the Rayleigh-Taylor instability
and, more generally of any instability, can only be made by means of numerical
simulations. An example of the results of these simulations for the Rayleigh-Taylor
instability is shown in Fig. 6.1.
6.2.2 Kruskal-Shafranov Instability: B0 = 0
If gravity and magnetic fields are both present, a series of new dynamical situations
arise. The basic physical mechanism underlying these new phenomena is actually
simple and is due to the fact that, while plasma density and pressure are connected
directly via the equation of state, the magnetic field only modifies the pressure, but
not the density. The nature of the gravitational instability of the previous Section is
therefore altered and, since in ideal plasmas the magnetic field is coupled to matter, the
magnetic structure is altered as well. As an example, consider the case of a magnetic
field occupying only a limited portion of space, with matter present everywhere. An
astrophysical example of this situation is encountered when the magnetic field of a
star is confined below the visible surface of the star.
Let then B be a constant horizontal field confined below the plane z = 0 and
assume that its value is rapidly decreasing, going to zero at the top of a thin layer above
that plane. Let’s further assume that the gas temperature is the same everywhere. With
the above assumptions, the local equilibrium condition requires that:
d
P + B 2 /8π = −ρg.
dz
110 6 Instabilities
There will therefore be an abrupt variation of magnetic pressure crossing the sur-
face that separates the magnetized region of the atmosphere from the field-free
one. Such a variation must be compensated by a corresponding increase of the gas
pressure, P = B 2 /8π, in order to maintain equilibrium in the atmosphere across
the discontinuity. The latter pressure variation, however, necessarily induces a den-
sity variation in the gas, ρ = (m/k B T )P = (m/k B T )(B 2 /8π ), which will
become denser with respect to that of the underlying region. The system is therefore
subject to a Rayleigh-Taylor instability and regions of dense and unmagnetized gas
will start to flow downward. In turn, the magnetized plasma will start rising, drag-
ging the field with it. In the final state, the field will be concentrated in a series of
vertical layers emerging from the original separation surface. This dynamics is often
described by saying that the magnetic field tends to be buoyant.
To derive the dispersion relation, will shall again consider a situation with two
incompressible fluids of different density, lying one on top of the other. Following the
procedure already adopted for the case of a vanishing field, we start from Eq. (6.23)
where we now keep the term proportional to (k · B 0 )2 .
For z = 0 Eq. (6.23) becomes:
(k · B 0 )2
[ω2 ρ0 − ](ξz − k 2 ξz ) = 0,
4π
with ρ0 = ρ1 or ρ0 = ρ2 , for z ≷ 0, respectively. Therefore, no value of ω exists for
which the first parenthesis is identically equal to zero, irrespective of the value of z.
Thus, the equation for ξz is still the one of the unmagnetized case:
ξz = k 2 ξ,
whose solutions are given by Eq. (6.25). Proceeding as in the previous case we easily
find
∞ ∞
(k · B 0 )2
ω 2
ρ0 (ξz2 + k 2 ξz2 ) dz = −k g (ρ1 − ρ2 )ξz (0) +
2 2
(ξz2 + k 2 ξz2 ) dz.
4π
−∞ −∞
(6.28)
The integrals entering Eq. (6.28) can be evaluated using Eq. (6.25), and the dispersion
relation is:
ρ1 − ρ2 (k · B 0 )2
ω2 = −|k| g +2 , (6.29)
ρ1 + ρ2 4π(ρ1 + ρ2 )
where k has been written as |k| to make the expression of ω independent of a particular
frame of reference.
In the limit B 0 → 0 we find again the condition for the Rayleigh-Taylor instability,
while in the case of a homogeneous plasma, ρ1 = ρ2 = ρ0 , we have
(k · B 0 )2
ω2 = ,
4πρ0
which, as we shall see in the following, corresponds to the propagation of Alfvén

waves. Equation (6.29) shows that the presence of a magnetic field has a stabilizing
effect. In fact, even if ρ = ρ1 − ρ2 > 0, ω2 can still be positive, provided that the
second term dominates the rhs of Eq. (6.29), i.e. when
2π B2
λ= < 0 cos2 θ,
|k| gρ
where θ is the angle between k and B 0 . The most efficient stabilization is reached
when k is parallel to B 0 , while the stabilizing effect vanishes when k and B 0 are
mutually orthogonal. This is easily understood since, in the first case the perturbation
deforms the field lines and the magnetic tension resists the deformation, while in the
second case the field lines remain straight and the tension cannot act.
A case of particular interest is one in which the plasma is confined to the region
z > 0, while in the region z < 0 only a magnetic field is present. In this situation,
by applying the previously adopted procedure again, we find:
(k · B 0 )2
ω2 = −|k|g + ,
4πρ1
which shows how a plasma could be sustained against gravity by a magnetic field
alone. This could possibly explain the existence and stability of solar prominences,
i.e. structures composed by a plasma denser than that of the surrounding environment.
Note however, that as long as there is a direction along which the perturbation fronts
can propagate in an unstable way, the system will be unstable. In the present case,
nothing can stop modes with wave-vectors orthogonal to the magnetic field from
growing. Interestingly, introducing a gradient, or shear in the magnetic field with
heights, which might lead to stabilization (because in this case there would be no
fixed wave-vector orthogonal to the field at all heights) does not completely stabilize
the instability.
6.2.3 Parker Instability
A further example of an instability in the presence of gravity is provided by the

so-called Parker instability, an interesting astrophysical application of the theory
of plasma instabilities. We shall now discuss it in some detail because it permits
a good appreciation of the technicalities involved in the theory, but also because it
emphasizes the necessity of a close comparison with observations to assess the actual
interest of the results obtained.
112 6 Instabilities
The problem that prompted this study, as discussed by Parker in 1966 [15], con-
cerns the structure of the magnetic field of the Galaxy and the influence of the field
on galactic configuration and dynamics. The essential observational data are the
following:
• the Galaxy is permeated by a large scale magnetic field, whose indicative value is
a few times 10−6 G.
• the field appears to be confined in the galactic plane.
• the magnetic field (as well as the cosmic rays trapped in it) cannot be confined by
the weak pressure of the external extragalactic medium.
In this situation it is reasonable to assume that the confinement within the galactic
disc is due to the weight of the self-gravitating interstellar gas that forms the disc
itself. The freezing of the field within the plasma might then confine the field inside
the disc. The gas confinement due to the gravity of the galactic nucleus itself can be
ruled out since it can be shown that it would produce a number of effects that are not
observed. To check the correctness of the above assumption, let’s first determine the
properties of the galactic disc in equilibrium.
We are going to adopt a simplified two-dimensional model, in which the magnetic
field is parallel to the galactic disc. We identify this direction with the y-axis, B 0 =
B0 (z)e y , the z-axis being the direction of gravitational acceleration.
The equilibrium equation thus reads:
d B2
P0 + 0 = −ρ(z)g,
dz 8π
where, to keep things as simple as possible, g has been assumed to be constant. To be

rigorous, we should include in the equilibrium equation also the pressure generated
by the cosmic rays. This contribution was included in Parker’s original treatment,
but here we shall neglect it, since its absence does not alter the physics of our system.
Moreover, we shall assume that the interstellar gas is isothermal, which allows us
to write:
k B T0
P0 = ρ0 = c02 ρ0 ,
m̄
and that the following relation holds
B02
= α P0 = αc02 ρ0 , α = const. (6.30)
8π
The last assumption implies that the Alfvén speed, ca2 = B02 /(4πρ0 ) is constant. In
the above framework, the solution of the equilibrium equation is
ρ0 (z) = ρ0 (0) exp(−z/H ),

where H is the scale height,

c02 (1 + α)
H= .
g
Equation (11.17) further implies that
B0 (z) = B0 (0) exp(−z/2H ).
Summarizing, we have
1 dρ0 1 d P0 2 dB0 1
= = =− . (6.31)
ρ0 dz P0 dz B0 dz H
Observations suggest that H is of the order of some hundreds of parsecs (H =

1 ÷ 3×1020 cm) and that c0 is of the order of a few km/sec (c0 = 3 ÷ 8×108 cm sec−1 )
corresponding to temperatures of the order of thousands of degrees, with g = 1 ÷
3 × 10−9 cm sec−2 . From the definition of H we find that it is likely that α 1.
The observed magnetic field is about B0 = 1 ÷ 5 × 10−6 G and Eq. (6.30) suggests
values of ρ of the order of a few hydrogen atoms per cm3 , close to the observed
values. The proposed model thus seems to provide a reasonable representation of the
equilibrium configuration of the interstellar gas-magnetic field system. The critical
point, of course, is that of the stability of such equilibrium, discussed below.
It is convenient to start from the linearized equations, Eqs. (6.3)–(6.6), rather than
use Eq. (6.23). In the following, we shall express B 1 (y, z, t) in terms of its vector
potential A1 (y, z, t) : B 1 = ∇ × A1 .
Since B1x = 0 and both y and t are ignorable coordinates, the equilibrium quan-
tities will be functions of z only and we can write:
A1 (y, z, t) = a(z)ei(ky−ωt) ex , (6.32)
and
B 1 (y, z, t) = b(z)ei(ky−ωt) . (6.33)
Since
∂ A1x ∂ A1x
B1y = , B1z = − ,
∂z ∂y
we have:
da
by = ≡ a (z), bz = −ika. (6.34)
dz
114 6 Instabilities
Taking into account that Ux = iωξx = 0 and adopting a Fourier representation of

the type given by Eq. (6.32) for all first-order quantities, from Eq. (6.6) we obtain
∇ × ( A1 − ξ × B 0 ) = 0.
The quantity in round brackets can be expressed as the gradient of a potential, which,
using the gauge invariance of Maxwell’s equations, can be safely taken equal to zero.
We shall thus write
A1 = ξ × B 0 ,
whose z-component provides the relation:
a 1 a
ξz = − , ξ (z) = − a + . (6.35)
B0 B0 2H
Equation (6.35) suggests expressing all first-order quantities in terms of a(z).

Equation (6.8), taking into account Eqs. (6.31) and (6.35), then becomes
ρ0 a ρ0 a ρ0 a
ρ1 = − − ρ0 ikξ y + a + = a − − ρ0 ikξ y .
H B0 B0 2H B0 2H
In a similar way, Eq. (6.9) can be cast in the form
ρ0 c02 a
P1 = γ a + γ /2 − 1 − γρ0 c02 (ikξ y ).
B0 H
To eliminate ξ y , consider first the y-component of the momentum equation, Eq.

(6.11).Writing the force term, Eq. (6.12), as
1
F(ξ ) = −∇ P1 + [(∇ × B 0 ) × B 1 + (∇ × B 1 ) × B 0 ] + ρ1 g
4π
1 1
= −∇(P1 + B 0 · b) + [B 0 · ∇)b + (b · ∇)B 0 ] + ρ1 g,
8π 4π
we find that the y-component of the momentum equation reduces to
B0 a
ω2 ρ0 ξ y = ik P1 − ik .
8π H
Making now use of the preceding expression for P1 , we find:
kc02 a
Ω 2ξy = i [γ a + (γ /2 − 1 − α) ], (6.36)
B0 H
where Ω 2 has been defined as
Ω 2 = ω2 − γ k 2 c02 .
Equation (6.36) allows us to deduce two explicit expressions for ρ1 and P1 in terms
of a and its derivative:
ρ0 2 a
Ω 2 ρ1 = ω a − [ω2 /2 + (1 − α − γ )k 2 c02 ] , (6.37)
B0 H
ρ0 c02 2 a
Ω 2 P1 = γ ω a + [(γ /2 − 1)ω2 − αγ k 2 c02 ] . (6.38)
B0 H
Finally, the z-component of the momentum equation, multiplied by Ω 2 , gives:
B0 2 B0 2 2 (1 + α)c02 2
ω2 ρ0 (Ω 2 ξz ) = [(Ω 2 P1 ) + Ω a )] − k Ω a+ (Ω ρ1 ),
4π 4π H
where g has been expressed in terms of H , according to its definition.
Inserting in the preceding equation the previously deduced expressions for Ω 2 ρ1
and Ω 2 P1 and performing the derivatives, we arrive at the final form of the differential
equation for a:
N
a (z) = k 2 a(z), (6.39)
D
where
ω 4 1 ω 2 (1 + α)(1 + α − γ ) − αγ /2
N= − (γ + 2α) 1 + 2 2 + 2αγ − ,
kc0 4k H kc0 k2 H 2
(6.40)
and
ω 2
D = 2αγ − (γ + 2α) . (6.41)
kc0
The boundary conditions to be imposed on a(z) are such that a should vanish at the
origin and remain finite for all values of z. This implies that N /D must be a negative
quantity, in which case the solution of Eq. (6.39) can be written as:
a(z) = ( b̄ H ) sin K z, (6.42)
where
N

K 2 = k 2 ,
D
116 6 Instabilities
and the constant of integration has been written as b̄ H . The explicit appearance of
emphasizes that a is a first-order quantity and b̄ is an arbitrary constant with the
dimensions of a magnetic field to be specified later on.
Since we are interested in the unstable case, ω2 < 0, let’s put ω = i/τ , τ
representing the growth time of the instability and define
ω 2 1 H 2 1 2
ν2 = − 2 2 = 2 2 2 = .
k c0 τ k c0 c0 τ kH
Since H/c0 represents the transit time over the scale height for a sound wave, n =
H/(c0 τ ) is the growth rate of the instability measured in units of the inverse of that
transit time. For the unstable solutions, the term
D = 2αγ + (γ + 2α)ν 2 > 0,
and the boundary conditions reduce to the requirement that N < 0. If x = 1/(k 2 H 2 ),
unstable solutions exist when
N (x, n 2 ) = q x 2 − r x + 2αγ < 0, (6.43)
with
1 2 2
q= n (4n + 2α + γ ),
4
r = − n 2 (2α + γ ),
= (1 + α)(1 + α − γ ) − αγ /2 = (1 + α)2 − γ (1 + 3α/2).
Thus, a necessary and sufficient condition to satisfy Eq. (6.43) is that r > 0,
which implies that > 0. In addition, x must lie between the two positive solutions
of the equation N = 0,

1
x= r ± r 2 − 8αγ q , (6.44)
2q
which requires that
r 2 > 8αγ q.
On the other hand, the definitions of q and r allow to write the preceding expression
as:
n 4 − 2λn 2 + μ2 > 0, (6.45)

with
(2α + γ )( + αγ )
λ= , μ= .
(γ − 2α)2 γ − 2α
Equation (6.45) is satisfied for n = 0 which means that marginal stability states
exist for any value of α and γ . According to Eq. (6.45), the maximum value of n 2
corresponds to the smallest of the two positive solutions of the equation n 4 − 2λn 2 +
μ2 = 0,

n 2max = λ − λ2 − μ2 .
We shall verify in the following that in the cases of physical interest λ2 μ2 ; the
square root can then be expanded as
μ2
n 2max ,
2λ
or, explicitly

n max . (6.46)
[2(γ + 2α)( + αγ )]1/2
However, noticing that n = n max implies r 2 = 8αγ q from Eq. (6.44), we find
xmax ≡ x(n max ) = r/2q. Inserting this result into the definition of N (see Eq.(6.43))
we find
N (x max , n max ) = 0 → K = 0.
In this case the solution would have an infinite vertical wavelength, which makes
apparent that the n max should be considered a limit rather than an effectively reach-
able value.
To summarize the preceding discussion, we have found that the gas-magnetic field
system is always unstable, irrespective of the values of α e γ , provided that > 0.
The last condition can be cast in the form
α(α + 1/2)
γ −1< . (6.47)
1 + 3α/2
In the absence of a field, α = 0, and, given that the γ parameter in the gas is larger
than unity, the preceding instability condition can never be satisfied and the system
is stable. When a magnetic field is present Eq. (6.47) gives only an upper limit for γ .
Typically in the interstellar medium γ 1 and Eq. (6.47) is always satisfied, for all
α values. The presence of a magnetic field thus inevitably generates an instability,
known as the Parker instability.
To assess the real relevance of this instability for the structure and dynamics of
the interstellar gas in the Galaxy, it is necessary to evaluate its growth time in the
118 6 Instabilities
prevailing galactic conditions and compare it with the typical galactic timescales,
which fall in the range of 108 ÷ 109 years.
Let’s choose the following set of typical values: B0 = 5 × 10−6 G, ρ0 = 5 ×
10 −24 g cm−3 (about 3 hydrogen atoms per cm3 ), g = 1.3 × 10−9 cm sec−2 and
γ = 1.
The minimum growth time will then be:
H c0 1 + α
τmin = .
c0 n max g n max
Using Eq. (11.17) to express c0 as
B0 ca
c0 = √ =√ ,
8πρ0 α 2α
from Eqs. (6.31) and (6.46) (with γ = 1) we get:
ca (1 + α) 2(2α + 3) 1/2
τmin = . (6.48)
g α (2α + 1)
With the typical values chosen above, Eq. (6.48) shows that τmin is less than 108
years for α > 0.5.
It also shows that τmin is a decreasing function of α, not a surprising result since
the instability is due to the presence of a magnetic field, whose importance grows
with increasing α. As an example, if α = 0.6, Eq. (6.48) gives τmin 8 × 107 years.
Moreover, the sound speed turns out to be c0 5.8 km sec−1 , corresponding to a
temperature of T0 4, 000 K and the scale height H 130 pc. All these values
agree with the average observed properties of the galactic disc.
As already remarked, one should also include the effects connected with the
presence of the cosmic rays and in particular of their pressure. It is possible to show
that those effects act in the same direction of those associated with the magnetic
field and therefore they contribute to the development of the instability. Therefore,
to generate growth rates in the physically interesting range, values of α lower than
those previously computed are actually sufficient.
In summary, we may conclude that the presence of a magnetic field will produce
an instability whose growth times are significantly shorter the the Galaxy’s lifetime.
Let’s now turn to the investigation of the consequences of the Parker instability on
the structure and evolution of the galactic disc. To simplify the discussion, we shall
again neglect the cosmic rays and focus our attention on a “cold” gas, c0 0. This
assumption seems to give rise to a problem when it is inserted into the expressions
for N e D (Eqs. (6.40) and (6.41)), but the difficulty is only an apparent one, as can
be seen by multiplying those expressions by k 4 c04 and noticing that αc02 = 21 ca2 =
B02 /(8πρ0 ) = 0. The solution for a(z) is still given by Eq. (6.42), with
4 4
K 2 = 41 k 2 s 2 1 − 2 (1 + 4/x) − 4 x ,
s s
where
τ
s= and x = 1/(k 2 H2 ).
H/ca
The quantity s is simply the growth time measured in units of the transit time of the
scale height H at the Alfvén speed. If it is reasonably assumed that s 1, K 21 ks.
The complete expression for the perturbed vector potential, A1 (y, z, t),
remembering that the physical parameters are real quantities, will then be given
by:
A1 (y, z, t) = [ b̄H exp(t/τ ) Re(expiky ) sin K z] ex

= [ b̄H exp(t/τ ) cos ky sin K z] ex .
From this, we may easily compute the magnetic field components, that turn out to be:
B y (y, z, t) = B0 (z) + b̄H K exp(t/τ ) cos ky cos K z,
and
Bz (y, z, t) = b̄H k exp(t/τ ) sin ky sin K z.
The field lines are defined by the equation
dz Bz b̄H k exp(t/τ ) sin ky sin K z b̄H k exp(t/τ ) sin ky sin K z

= = ,
dy By B0 (z) + b̄H K exp(t/τ ) cos ky cos K z B0 (z)
where in the denominator the term proportional to has been neglected with respect
to B0 . Let’s now integrate the preceding equation in the neighborhood of the point
z 0 where the field line crosses the z-axis (y = 0). We get
z y
B0 (z)
dz = b̄H k exp(t/τ ) sin ky dy = b̄H exp(t/τ )(1 − cos ky).
sin K z
z0 0
Since the rhs is of the order , the lhs must be of the same order, which allows us to
replace B0 (z) and sin K z with their values in z = z 0 . In this way we obtain:
z − z 0 H exp(t/τ )(sin K z 0 )(1 − cos ky),
where the arbitrary constant has been written as b̄ = B0 (z 0 ). The instability therefore
bends the initially straight field lines forming a succession of crests and troughs.
The effects of the instability are well illustrated by the behaviour of the perturbed
density, ρ1 (y, z, t), that we are now going to determine. Write first the physical
120 6 Instabilities
components of the displacement in the neighborhood of the point z = z 0 in agreement

with what was already done when computing the shape of the field lines.
From Eq. (6.35) we get
A1
ξz = − = − H exp(t/τ ) cos ky sin K z 0 ,
B0
while Eq. (6.36) gives us
ξ y = (kca2 τ 2 ) exp(t/τ ) sin K z 0 [Re(i expiky )]

= −(kca τ )2 exp(t/τ ) sin K z 0 sin ky = − H (s 2 k H ) exp(t/τ ) sin ky sin K z 0 .
This shows that for s 1 the motion of the plasma is essentially horizontal, since
ξ y ξz .
Accordingly, we approximate the continuity equation (6.3) as:
∂ξ y
ρ1 −ρ0 = H (sk H )2 exp(t/τ ) cos ky sin K z 0 . (6.49)
∂y
Figure 6.2 shows both the profile of a field line and the perturbed density in
arbitrary units.
As apparent from the figure, the density increases in correspondence of the troughs
of the field line and decreases in correspondence of the crests, a result that can
pictorially be described by saying that the instability forces the plasma to slide along
the field lines and to accumulate in the lowest points.
It is not difficult to understand the physical mechanism responsible for this behav-
ior. In equilibrium, gravity holds matter, together with the magnetic field frozen in
it. When the instability sets in, the magnetic field is distorted and portions of the
field are vertically displaced. Matter attached to the field should also move up, but
the action of gravity produces a downward fall of the portions of plasma along the
lines of B. As a result, the gravitational effects on the plasma sitting at the crests’
location weaken, which favors a further distortion of the field lines. At the same time,
the gravitational action becomes stronger on portions of plasma lying at the bottom
Fig. 6.2 A magnetic field

line (upper curve) and the
perturbed density ρ1 (lower
curve) as functions of ky
(arbitrary units)
ky
3 0 3
2 2 2 2
Fig. 6.3 Sketch of a

condensation at the bottom of
a valley of B
of the troughs of B, giving a further help to the development of the instability. The
final configuration thus features a series of condensations placed all along the disc,
as schematically shown in Fig. 6.3.
The picture is consistent with the observations that show that the interstellar
gas is not uniformly distributed, but tends to condensate into clouds, denser than
the surrounding medium. If the magnetic field is absent, the formation of stable
gravitationally induced condensations is only possible if their mass is larger than the
so-called Jeans mass.2
3 1/2 c03
M > MJ = 1/2
. (6.50)
4π G 3/2 ρ 0
If we use the values for the interstellar gas previously adopted, M J turns out to
be of the order of 106 M , which therefore should be a lower limit for the mass of
interstellar clouds. On the contrary, clouds are observed with masses of the order
of 104 M or less. It is therefore clear that the Parker instability cannot be the only
cause of their formation: nevertheless it may contribute to the solution of the problem
of cloud formation. In fact, let’s look again at the configuration sketched in Fig. 6.3
and imagine that the condensations are actually a series of infinitely long cylinders,
parallel to the x-axis, i.e. normal to the plane of the figure. The plasma they contain
feels the action of a horizontal magnetic field and of a vertical gravitational field. In
this configuration the charged particles inside the plasma will move with the speed
given by Eq. (2.8)
c mg × B 0
v= ,
e0 B02
where m and e0 are the mass and the charge of the single particle. This motion gives
rise to an electric current flowing along the cylinder. The current density will be
given by:
g g
J = en 0 (v p − ve ) = cn 0 (m p + m e ) c n 0 m p .
B0 B0
2 The formation of a stable condensation requires that the gravitational force be dominant over the
pressure (see Exercise 6.2)
122 6 Instabilities
Let’s now consider a segment of our cylinder of length and cross section S =
πr 2 , whose volume will evidently be S. Let m 0 be the mass contained in it, n 0 m p =
m 0 /S. The value of the electric current, I , is computed by evaluating the flux of the
current density, J. In our case we shall have
g m0 g
I = SJ = c ≡ c m,
B0 B0
where m is the mass of the cylinder per unit length. The currents flowing in the
cylinders are all mutually parallel and therefore two cylinders placed at a distance d
will attract each other with a force (per unit length) given by:
2I 2 2 m2 g 2
FB = = .
c2 d B02 d
If we now compute the gravitational force per unit length between the same two
cylinders, it is easy to show that
Gm2
FG = π .
d
It follows that the ratio
FB 2 g2
= ,
FG π G B02
can be larger than unity in the region where the galactic magnetic field is weak and
the gravitational field is strong. If, for instance, we consider the situation at a height
of 100 parsec above the central galactic plane, where g can be estimated to be of
the order of 3 × 10−9 cm sec−2 , and B0 = 5 × 10−6 G, we find
FB
3.
FG
In conclusion, we have seen how the combined action of gravity and magnetic
fields, apart from giving rise to an instability, can generate an attractive force which
adds to the plasma self-gravity and may justify the existence of clouds of mass
substantially lower than the Jeans mass, as those that are actually observed.
6.2.4 Instabilities in the Presence of Plasma Flows:

Kelvin-Helmholtz Instability
In the preceding Sections we have always assumed that the unperturbed state was
a static one, namely that the velocity was vanishing at the zeroth-order U 0 = 0.
However, in many circumstances, both in the lab and in astrophysics, mass motions
are present in the unperturbed state, a classical example being that of a fluid in laminar
motion where the different layers do not move with the same speed. Therefore, the
study of the possible instabilities that might arise in this or similar situations is
clearly of interest. Obviously, viscosity should play a role in the system dynamics,
since its action tends to reduce velocity differences between different fluid layers. In
the following, however, we shall concentrate only on ideal plasmas, thus neglecting
all viscous effects as well as any other effects that may modify the system’s dynamics
(for instance, those connected with surface tension). The most important instability
that develops in a system of superposed fluids when a velocity gradient is present
in the direction normal to the flow is known as the Kelvin-Helmholtz instability. If
the velocity stratification is parallel to the direction of gravity, the Kelvin-Helmholtz
instability can be thought of as a generalization of the Rayleigh-Taylor instability to
a dynamical case.
To reduce the algebraic complexity of the treatment, we shall be concerned with
a case analogous to that already examined for the Rayleigh-Taylor instability: two
homogeneous superposed fluids of different densities moving with constant and
parallel velocities, but whose speeds are different on the two sides of the separating
surface. The x-axis will be chosen in the direction of the velocities, the z = 0 plane
will define the surface separating the fluids and the z-axis coincides with the direction
of the gravity assumed to be constant, g = −gez . We shall further assume that the
two fluids are incompressible and that the system is completely homogeneous in the
y direction.
We shall first consider a purely hydrodynamic system (B = 0), and then add the
effects connected with the presence of a magnetic field.
Let the unperturbed state be characterized by:
ρ0 = ρ+ (z > 0); ρ0 = ρ− (z < 0),
and
U 0 = U+ e x (z > 0); U 0 = U− e x (z < 0).
At the beginning, it turns out to be convenient to write down the equations in a

general way, i.e. without considering separately the regions z ≷ 0, a distinction that
will be made later on. As a result derivatives (with respect to z) of the discontinuous
quantities ρ0 and U 0 will appear, in spite of the fact that those derivatives actually
vanish in each of the two half-spaces defined by the plane z = 0.
The only relevant equation at zeroth-order is the momentum equation, which in
our case is identical to the one for the static case, i.e.
0 = −∇ P0 + ρ0 g.
124 6 Instabilities
Let u e ρ1 be the velocity and density perturbations, respectively. If the perturbed

quantities are assumed to be proportional to exp[i(kx −ωt)], we obtain the following
first-order equations:
• Continuity equation
−iωρ1 + ∇ · (ρ0 u + ρ1 U 0 ) = ρ0 u z + ikU0 ρ1 ,
where the incompressibility condition, ∇ · U 0 = ∇ · u = 0, has been used and the

prime indicates a derivation with respect to z. Defining Ω as
Ω = ω − kU0 ,
from the preceding expression for ρ1 we obtain
ρ0
ρ1 = i uz. (6.51)
Ω
• Momentum equation
−iρ0 Ω u + ρ0 U0 u z ex = −∇ P1 ,
whose components are:
− iρ0 Ωu x + ρ0 U0 u z = −ik P1 , (6.52)
− iρ0 Ωu z = −P1 − ρ1 g (6.53)
To these equations we must add the incompressibility condition,
iku x + u z = 0, (6.54)
and a further condition expressing the fact the the separating surface, even if distorted,
remains the border between the two fluids, which means that the particles lying on
that surface will move with it. This translates into the fact that the vertical velocity
must be considered as the lagrangian derivative of the position of the point located
on the separating surface. Therefore:
dζ ∂ζ
uz = = + (U 0 · ∇)ζ = −iΩζ,
dt ∂t
where ζ represents the vertical deformation of the plane z = 0. Since ζ is a continuous
quantity, the preceding relation shows that
uz
is a continuous function across the separating surface.
Ω
It is now possible to find an equation for u z only, by eliminating u x by means of

Eq. (6.54), ρ1 by using Eq. (6.51) and P1 by deriving Eq. (6.52) and inserting the
result in Eq. (6.53). Finally, we get:
uz
k 2 ρ0 Ωu z − [ρ0 (Ωu z + U0 ku z )] = gk 2 ρ0 . (6.55)
Ω
As already remarked, this equation is valid for all values of z. To impose the
boundary conditions, which in this case are simply the continuity of u z /Ω, we
integrate the preceding equation between − and + and notice that all terms that
are not derivatives with respect to z do not contribute. Consider for example the first
term on the lhs of Eq. (6.55). We have to evaluate
0
ρ0 Ωu z dz = ρ0 Ωu z dz + ρ0 Ωu z dz.
− − 0
But ρ0 Ω is a constant quantity in each of the two integrals which implies
0
ρ0 Ω u z dz = [ρ0 Ω u z ] → 0 for → 0,
−
where u z is an average value of u z in the integration interval. The same conclusions

hold for the other integral. If we define 0 as 0 ( f ) = f () − f (−), the result of
the integration can be expressed as
uz
0 [ρ0 (Ωu z + U0 ku z )] = −gk 2 0 (ρ0 ) ,
Ω
where the continuity of u z /Ω has been taken into account. Moreover, 0 (U0 ku z ) = 0
since U0 = 0 in each of the two half-spaces. In conclusion:
uz
0 [ρ0 Ωu z ] = −gk 2 0 (ρ0 ) . (6.56)
Ω
Equation (6.56) must be considered as a boundary condition valid in z = 0, while
Eq. (6.55) holds for every z = 0.
Let’s now write Eq. (6.55) explicitly in the two regions z > 0 e z < 0. Since in
those regions ρ0 = U0 = 0, Eq. (6.55) assumes the simple form:
u z − k 2 u z = 0,
whose solutions, convergent in z = ±∞, can be written in the form
u z = AΩe+kz (z < 0),

126 6 Instabilities
u z = AΩe−kz (z > 0),
A being an arbitrary constant that expresses the continuity of u z /Ω in z = 0.

Making use of the solutions just found, we can write the condition given by Eq.
(6.56) explicitly:
ρ+ (ω − kU+ )2 + (ω − kU− )2 = gk(ρ− − ρ+ ). (6.57)
The assumption ρ− > ρ+ ensures that in the absence of flows the system is
Rayleigh-Taylor stable (see Eq. (6.27)). Defining α as
ρ+ ρ−
α+ = e α− = → α+ + α− = 1,
ρ+ + ρ− ρ+ + ρ−
and expanding the preceding equation, we finally get
ω2 − 2k(α+ U+ + α− U− )ω + k 2 (α+ U+2 + α− U−2 ) − gk(α− − α+ ) = 0.
If the discriminant of this equation is negative, ω has a non-vanishing imaginary part

and the solution is therefore unstable. Evaluating explicitly the discriminant, we see
that the instability condition can be written as
gk(α− − α+ ) − k 2 α+ α− (U− − U+ )2 < 0.
The latter expression allows us to draw a few interesting conclusions. In the first
place, the discriminant vanishes for k = 0 and in this case ω is real and no instability
arises. We also see that we always have an instability when
g(α− − α+ )
k> .
α+ α− (U− − U+ )2
Therefore, to have an instability, there is no need of strong velocity gradients. What-

ever the value of U− − U+ might be, even very small, sufficiently large values of k
always exist, i.e. sufficiently small wavelengths, that make the system unstable.
Figures 6.4 and 6.5 show the results of a numerical simulation of the nonlinear
development of the Kelvin-Helmholtz instability and the direct observation of the
same instability in the atmosphere.
Let’s now examine effects that might inhibit the development of the Kelvin-
Helmholtz instability. For an unmagnetized plasma or fluid, it is clear that the pres-
ence of surface tension, which contrasts the distortion of the separating surface, may
contribute, within certain limits, to the stabilization of the system. In the case of a
magnetized plasma, the same stabilizing role can be played by a horizontal magnetic
field, coplanar with the velocity. Even without carrying out an explicit calculation, it
easy to realize that a magnetic field normal to U cannot influence the development
of the instability. In fact, the rippling of the separating surface can only displace the
Fig. 6.4 Numerical

simulation of the
development of the
Kelvin-Helmholtz instability
Fig. 6.5 The

Kelvin-Helmholtz observed
in the atmosphere
lines of force without deforming them. A different picture arises when a magnetic
field parallel to the velocity is present: the field lines are bent, with the consequent
birth of a tension that opposes the distortion.
We shall therefore concentrate on a configuration where, in the unperturbed state,
a magnetic field, B0 = B0 ex , is present in both fluids.To simplify the discussion,
we shall assume that the field is constant everywhere. The momentum equation now
includes also the magnetic term:
B0 · b 1
−iρ0 Ω u + ρ0 U0 u z ex = −∇ P1 + + (B0 · ∇)b.
4π 4π
128 6 Instabilities
The discussion of the preceding case has shown that all the terms containing U0 drop
out of the final expressions: we shall therefore omit them from now on. The two
components of the momentum equation now read:
B0
−iρ0 Ωu x = −ikΠ + i bx ,
4π
B0
−iρ0 Ωu z = −Π + i bz − ρ1 g,
4π
where the quantity Π is defined as
B0 · b
Π = P1 + .
4π
The induction equation can now be written as:
B0
−iΩ b = ik u.
4π
The conditions ∇ · b = 0 and ∇ · u = 0, i.e. ikbx + bz = 0 and iku x + u z = 0,

can be used to eliminate from the equations the quantities u x and bx and to write the
z-component of the induction equation as:
Ωbz = −k B0 u z , (6.58)
while the x-component is simply the derivative of the preceding relation. The
x-component of the momentum equation (multiplied by k) becomes
k 2 B02
ρ0 Ω − u = −ik 2 Π, (6.59)
Ω 4π z
and the z-component reads
k 2 B02 u
− gρ0 = −iΠ ,
z
ρ0 Ω 2 − (6.60)
4π Ω
where Eq. (6.51) has been used. Notice that in the preceding equations explicitly
contain the term u z /Ω, which, as the purely hydrodynamical case has shown, is
continuous across the separating surface. Taking the derivative of Eq. (6.59) and
inserting the result into Eq. (6.60), we finally obtain
k 2 B02 u z k 2 B02 u

− gρ0
z
ρ0 Ω 2 − = k 2 ρ0 Ω 2 − . (6.61)
4π Ω 4π Ω
To impose the boundary conditions, we integrate Eq. (6.61) between − and + and
obtain
k 2 B02 u z u
z
0 (ρ0 Ω 2 − = −k 2 g0 (ρ0 ) . (6.62)
4π Ω Ω
In each of the two half-spaces z > 0 and z < 0, u z satisfies the equation
u z − k 2 u z = 0,
and therefore, once again, we shall have
u z = AΩe+kz (z < 0),
u z = AΩe−kz (z > 0).
We now insert these expressions into Eq. (6.62) and obtain
B02
ρ+ (ω − kU+ )2 + ρ− (ω − kU− )2 = gk(ρ− − ρ+ ) + k 2 ,
2π
which, compared to Eq. (6.57), clearly shows the effect of the magnetic field. Fol-
lowing the procedure adopted before, we find the following equation for ω
ω2 − 2k(α+ U+ + α− U− )ω + k 2 (α+ U+2 + α− U−2 )

k 2 B02
− gk(α− − α+ ) − = 0. (6.63)
2π(ρ+ + ρ− )
The instability condition (i.e. the discriminant of Eq. (6.63) < 0) now becomes
k 2 B02
gk(α− − α+ ) − k 2 α+ α− (U− − U+ )2 + < 0.
2π(ρ+ + ρ− )
This expression clearly shows the stabilizing effect of the magnetic field. In fact, if
B02
α+ α− (U− − U+ )2
2π(ρ+ + ρ− )
the Kelvin-Helmholtz instability cannot develop.

130 6 Instabilities
6.3 Instabilities in Cylindrical Geometry
All instabilities discussed so far shared the common assumption that the system
could be descibed in the framework af a planar geometry. However, several situations
occur where the use of a cylindrical geometry is more appropriate. For instance,
in laboratory plasmas, the machines aiming at controlled thermonuclear fusion of
the type known as zeta-pinch are substantially cylinders. The Tokamak, the most
common fusion machine at present time, has a toroidal geometry, which in a first
approximation can be treated as a cylindrical geometry (assuming a periodicity in the
axial direction) if the ratio between the major and minor axes, the so-called “aspect
ratio”, is very large. In astrophysics it is often useful to describe the magnetic field in
terms of “flux tubes”, which identifies the cylindrical geometry as the most natural
one to describe systems such as stellar jets or solar coronal loops.
In the following we shall discuss two examples in same detail, one of relevance to
laboratory plasmas and the other to solar physics. In neither case a claim of realism
is made. They are rather illustrations of the methods used to study the stability of
cylindrical systems.
6.3.1 Instability of a Plasma Column
Our basic schematic model is a plasma column of radius a and infinite length with no
gravity effects. In this situation, all quantities will depend only on the coordinate r ,
the distance from the axis of the cylinder. The only non-vanishing components of
magnetic field will be Bθ and B0z . Equation ∇ · B = 0 implies
1 ∂ const.
(r Br ) = 0 → Br = = 0,
r ∂r r
if we require the field be finite on the cylinder’s axis. As already remarked in Chap. 5,
the field is expected to show a helical structure.
The equilibrium equation (see Eq. 5.41)
B2 1
0 = −∇ P0 + 0 + (B 0 · ∇)B 0 ,
8π 4π
in cylindrical geometry reduces to the r -component only:
d 2
B 2 + Boz 2
1 B0θ
P0 + 0θ + = 0, (6.64)
dr 8π 4π r
which shows that the equilibrium is completely determined by the profiles of the
pressure and of the magnetic field. The density, that does not enter the equilibrium
equation since gravity has been neglected, can be chosen at will. Of course, boundary
6.3 Instabilities in Cylindrical Geometry 131
conditions appropriate to the model under study must complement the preceding
equation. Actually, as we shall see, the boundary conditions will fix the dispersion
relation and consequently the stability conditions of the system.
The model we have chosen, represents, to a first approximation, the behaviour
of a θ -pinch: a plasma column contained inside a cylindrical, rigid and perfectly
conducting wall. However, the plasma is not in direct contact with the wall, but is
separated from it by a vacuum zone. The presence of a region devoid of plasma is
necessary to keep the plasma off the conducting wall, where it would recombine and
loose all its plasma properties. In the following we shall denote by a the radius of
the plasma column and by b (a) the radius of the conducting wall.
If B̃ is the magnetic field in the vacuum zone external to the plasma column, the
boundary conditions at the surface of the plasma column are

n· B = n · B̃ = 0, (6.65)
S− S+
B2 B̃2
P+ = , (6.66)
8π S− 8π S+
where n is a unit vector normal to the plasma surface pointing towards the exterior3
and the notations S− and S+ indicate the the quantities in square brackets must
be computed, respectively, inside and outside of the plasma-vacuum interface. We
emphasize here that the boundary conditions must be applied to the instantaneous
configuration of the separating surface, not to the initial configuration, since the
instability distorts the plasma column. For instance, the external normal, n, that
enters Eq. (6.65), equals er in the equilibrium configuration, but has a different
direction at later times.
These boundary conditions are easy to interpret. If the first were not satisfied,
i.e. if a component of the magnetic field normal to the separating surface were be
present, the vacuum region simply would not exist, since nothing could prevent the
plasma from freely flowing along such a component. If the second condition were
not verified, a big pressure difference would arise and the separating surface would
simply be swept away.
Even if the conditions expressed by Eqs. (6.65) and (6.66) are sufficient to deter-
mine the solution of the problem, it is useful to add the condition that expresses the
continuity of the tangential component of the electric field across a surface separating
two different media:
1 1
n × [E + U × B] S− = n × [Ẽ + U × B̃] S+ , (6.67)
c c
where Ẽ is the electric field in the vacuum region.
3 From now on n will be referred to as the external normal.

132 6 Instabilities
The algebra of stability problems has the disappointing tendency to become very
complex: we shall therefore introduce a few simplifying assumptions, that, however,
do not obscure the physics involved. Let’s thus assume that the unperturbed magnetic
field, B0 , is constant and directed along the z-axis, B 0 ≡ (0, 0, B0 ). Equation (6.64)
then implies that P0 = const. Such a configuration can be thought as generated
by a system of azimuthal currents circulating in the plasma column, actually in a
surface layer, whose thickness tends to zero in an ideal plasma. In fact, in the interior
of the column J ∝ ∇ × B 0 = 0, while on the surface currents must necessarily
flow to comply with the boundary conditions. Equation (6.65) constrains only the
radial component of the field and nothing forbids the presence of discontinuities
in the tangential components Bθ and B0z . To take this into account we shall write:
Jθ = Jθ (r )δ(r − a) and Jz = Jz (r )δ(r − a). Since:
c dB0z
Jθ = Jθ (r )δ(r − a) = − ,
4π dr
we may integrate the preceding equation and obtain:
a+ a+
c dB0z
Jθ (r )δ(r − a)dr = Jθ (a) = − dr
4π dr
a− a−
c c
= − [B0z (a + ) − B0z (a − )] = − [B̃z (a) − B0 ].
4π 4π
If the z-component of the magnetic field is discontinuous, it follows that Jθ (a) cannot
vanish. In a similar way, integrating
c 1 d(r Bθ )
Jz = Jz (r )δ(r − a) = ,
4π r dr
we obtain
c
Jz (a) = [Bθ (a + ) − Bθ (a − )] = B̃θ (a),
4π
and thus also Jz (a) = 0 when a discontinuity of Bθ is present.

The vacuum field will be determined by the currents flowing in the plasma column.
In particular, if I is value of the current globally flowing in the column’s surface layer,
we shall have B̃θ = (2 I )/r. If we further assume that B̃z = 0, a discontinuity of Bz
will be present on the surface of the plasma column and therefore Jθ (a) = 0. The
z-component of the vacuum field is not influenced, since only Jz contributes to the
current I .
The dynamics of the system is described by the momentum equation, Eq. (6.11),
∂ 2ξ
ρ0 = F(ξ ),
∂t 2
with
1
F(ξ ) = −∇ P1 + [(∇ × B 0 ) × B 1 + (∇ × B 1 ) × B 0 ],
4π
where, in our case,
P1 = −ρ0 cs2 (∇ · ξ ) (6.68)
and
B1 = ∇ × (ξ × B 0 ). (6.69)
Equation (6.11) thus becomes:
∂ 2ξ 1
ρ0 = γ P0 ∇(∇ · ξ ) + (∇ × B 1 ) × B 0 ]. (6.70)
∂t 2 4π
Since the zeroth-order terms are constant, we can Fourier expand with respect to
the variables t, θ and z. A similar expansion, however, cannot be made with respect
to r , since in cylindrical coordinates the differential operators, besides derivatives,
also contain functions of r . For instance, ∇ ≡ [∂/∂r, (1/r )∂/∂θ, ∂/∂z]. We shall
therefore write

ξ ≡ ξr (r ), ξθ (r ), ξz (r ) ei(mθ +kz) e−iωt ,
where m must be an integer to ensure the periodicity in the θ coordinate. To simplify

the calculations we shall assume that m = 0, i.e. we shall restrict our study to
perturbations that preserve the axial symmetry. Using the preceding expressions, it
is easy to find:
B 1 = ik B0 ξr er + ik B0 ξθ eθ + B0 [ikξz − (∇ · ξ )]ez . (6.71)
In cylindrical coordinates ∇ · ξ is written as
1 d
∇·ξ = (r ξr ) + ikξz = f (r ) + ikξz ,
r dr
where
1 d
f (r ) = (r ξr ).
r dr
Writing down the components of Eq. (6.70) explicitly we get:
d B2 d f
−ω2 ρ0 ξr = γ P0 [ f (r ) + ikξz ] + 0 ( − k 2 ξr )
dr 4π dr
134 6 Instabilities
k 2 B02
−ω2 ρ0 ξθ = − ξθ
4π
−ω2 ρ0 ξz = ikγ P0 [ f (r ) + ikξz ].
Recalling the definitions of the characteristic speeds
P0 B02
cs2 = γ , ca2 = ,
ρ0 4πρ0
the preceding equations become:
df dξz
(k 2 ca2 − ω2 )ξr = (cs2 + ca2 ) + ikcs2 (6.72a)
dr dr
(k 2 ca2 − ω2 )ξθ = 0 (6.72b)
(k 2 cs2 − ω )ξz =
2
ikcs2 f. (6.72c)
The equation for ξθ is uncoupled from the others and its solutions are either ξθ = 0,
or ω2 = k 2 ca2 . Inserting the latter relation into the other two equations, it is easily
seen that in this case ξz = const. and ξr is a linear function of r . If ξr = ξz = 0, this
solution represents pure torsional oscillations.
Going back to the equations for ξr and ξz and putting ξθ = 0, it is possible to
eliminate ξr , and obtain an equation for ξz only:
d 2 ξz 1 dξz
+ − K 2 ξz = 0, (6.73)
dr 2 r dr
with
(k 2 ca2 − ω2 )(k 2 cs2 − ω2 ) (ω/k)4

K2 = = k 2
1 + . (6.74)
k 2 cs2 ca2 − ω2 (cs2 + ca2 ) cs2 ca2 − (ω/k)2 (cs2 + ca2 )
This is the well known equation for the Bessel functions of zero order. The choice
of a specific Bessel function depends on the sign of K 2 . If K 2 > 0, the solution,
compatible with the condition that ξz (0) is a finite quantity, is
ξz = I0 (K r ), (6.75)
where I0 is the modified Bessel function of the first kind. If instead, if K 2 < 0 the
Bessel function of the first kind, J0 (K r ), must be chosen. Assuming that K 2 > 0
and inserting the solution given by Eq. (6.75) in Eqs. (5.16a) and (5.16c) we find
K cs2 (k 2 ca2 − ω2 ) − ω2 ca2

ξr = I0 (K r ), (6.76)
ik cs2 (k 2 ca2 − ω2 )
where the prime indicates derivation with respect to the argument of the function.
In conclusion, we have shown that ξz ∝ I0 (K r ) and ξr ∝ I0 (K r ) = I1 (K r ),
but this does not solve completely the problem since the eigenvalue ω2 , entering the
definition of K , is still unknown. To find it, it is necessary to impose the boundary
conditions, obviously in their linearized form. The linearization operation requires
a certain amount of care, since first-order terms may arise from the perturbations
of the various quantities involved, but also as a consequence of the distortion of the
surface that separates the plasma from the vacuum. In fact, a generic quantity f (r, t)
entering the boundary conditions must be written in the form
f (r, t) = f 0 (r, t) + f 1 (r, t) f 0 (r 0 ) + ξ · ∇ f 0 (r0 , t) + f 1 (r 0 , t).
Equation (6.66) thus becomes
2
1 1 B̃
P1 + B0 · B1 = B̃0 · B̃1 + (ξ · ∇) 0 .
4π 4π 8π
To apply the boundary conditions it turns out to be convenient to use Eqs. (6.66) and
(6.67). Starting from the latter, we express the vacuum electric and magnetic fields,
Ẽ and B̃, in terms of the potential vector. Since in the equilibrium state the vacuum
electric field vanishes, Ẽ is a first-order quantity and therefore
1 ∂ Ã
Ẽ = − ,
c ∂t
and Ã is a first-order quantity as well. The total magnetic field in vacuum will be
B̃ = B̃0 + B̃1 = B̃0 + ∇ × Ã.
Taking into account the fact that E, U = ∂ξ /∂t and Ẽ are first-order quantities
and that for an ideal plasma the lhs of Eq. (6.67) vanishes, this condition reduces to
∂A
n0 × = n0 × ∂ξ /∂t × B̃0 .
∂t
The last expression can be integrated with respect to time giving
n0 × A = n0 × (ξ × B̃0 ) = (n0 · B̃0 )ξ − (n0 · ξ )B̃0 = −(n0 · ξ )B̃0 , (6.77)
since B̃0 , whose only component is in the θ direction, is perpendicular to n0 = er , On

the other hand, the linearized version of this boundary condition, given by Eq. (6.66),
gives
1 1
−γ P0 (∇ · ξ ) + B0 · [B 1 + (ξ · ∇)B0 ] = B̃0 · [B̃1 + (ξ · ∇)B̃0 ],
4π 4π
136 6 Instabilities
where use has been made of the equilibrium condition
B02 B̃ 2
P0 + = 0.
8π 8π
From Eq. (6.71) we get
B0 d
B1z = − (r ξr ),
r dr
which allows us to write the preceding boundary condition as:
B02 d 1 dB̃0
− γ P0 (∇ · ξ ) − (r ξr ) = B̃0 · [∇ × Ã + ξr ]. (6.78)
4π r dr 4π dr
On the other hand, Eq. (6.77) implies that Ã does not have components along eθ and
therefore we may put
Ã = q(r )er + ξr B̃0 ez ,
where q(r ) is an arbitrary function. Thus,

∂q ∂
B̃1 = ∇ × Ã = − (ξr B̃0 ) eθ (6.79)
∂z ∂r
Moreover, since ∇ × B̃1 = 0 (the vacuum is current-free) we have

∂ 2q ∂ 2q ∂2
∇ × (∇ × Ã) = − 2 er + − − 2 (ξr B̃0 ) ez = 0.
∂z ∂r ∂z ∂r
Equating the r, z components of the preceding equation to zero and integrating once
we get:
∂q
= g(r ),
∂z
and
∂q ∂
− (ξr B̃0 ) = h(z),
∂z ∂r
where g and h are arbitrary functions of their arguments. Combining the two equations
above we obtain
∂
g(r ) − (ξr B̃0 ) = h(z).
∂r
But the lhs of the preceding equation is a function of r only and therefore h(z) =
const. From Eq. (6.79) we now find that B̃1 = h(z)eθ is a constant vector. Actually
it is a null vector, since initially B̃1 = 0. Equation (6.78) can now be cast in the form
2
γ ω2 P0 B02 (k 2 cs2 − ω2 ) B̃
2
ξ z − 2
ξr = − 0 ξr ,
ikcs 4πikcs 4π r
where we have taken into account that B̃0 ∝ 1/r when calculating the derivative
of B̃0 on the rhs At this point all we have to do is to insert in the last equation the
expressions previously found for ξr and ξz and evaluate the result at the (initial!)
plasma-vacuum interface:
B̃02 K I0 (K a)
[cs2 ω2 − ca2 (k2cs2 − ω2 )]I0 (K a) = − c2
2 a
.
B0 a Io (K a)
The last equation allows us, finally, to deduce the dispersion relation
cã 2 K a Io (K a)
ω2 = k 2 ca2 − , (6.80)
a2 I0 (K a)
where the notation cã 2 = B̃02 /(4πρ0 ) has been introduced.

Obtaining the dispersion relation has required a considerable amount of work, in
spite of the many simplifying assumptions adopted. Moreover, Eq. (6.80) is still an
implicit definition of ω2 , since this quantity enters also the definition of K , so that
the effective value of ω2 can only be obtained by numerical means. An exception
to this statement is provided by the incompressible limit of Eq. (6.6) which can be
simply obtained by letting γ (and therefore cs2 ) tend to infinity. In this limit K = k
and Eq. (6.80) becomes an explicit definition of ω2 .
The dispersion relation we have found shows that a plasma column is unstable to
perturbations with axial symmetry, provided that
B̃02 K a Io (K a)
B02 .
(ka)2 I0 (K a)
We may further observe that when K is real, the quantity [K a Io (K a)/I0 (K a)] > 0
and increases with increasing |K |. From the definition of the latter quantity we
clearly see that when B02 = 0 and therefore ca2 = 0, it is always possible to find
solutions of Eq. (6.80) satisfying the instability condition. It follows that, without a
longitudinal field inside the column, the plasma is always unstable. The introduction
of a longitudinal field does stabilize the system, at least in a certain range of values
of B0 . In fact, the equilibrium condition, B02 = B̃02 − 8π P0 , imposes un upper limit
to B0 , namely
B02 < B̃02 .

138 6 Instabilities
Fig. 6.6 Instability of a

plasma column with m = 0
(sausage instability)
If this further limitation is inserted into the stability condition we get

K a I (K a)
o
(ka)2 > ,
I0 (K a)
namely a limitation of the values of k that allow the system to be stable. As shown in
Fig. 6.6, an axially symmetric perturbation creates a series of necks and bulges along
the column, which give the name to this instability, known as the sausage instability.
These features are easily interpreted from a physical point of view. In the neigh-
bourhood of the necks the vacuum region is is closer to the cylinder’s axis and the
value of B̃0 ∝ 1/r is larger than around the bulges. The force J × B̃ 0 is therefore
stronger and tends to reduce even more the section of the column. The introduction of
a longitudinal field in the column’s interior increases the internal magnetic pressure,
which tries to resist the squeeze and therefore stabilizes, at least partially, the system.
The discussions of the modes with m = 0 develops along similar lines with
increased technical difficulties. Of particular interest are the m = 1 modes, see
Fig. 6.7.
In this case the instability is due to the fact that in the regions where the plasma
column is concave towards the exterior, the lines of force of B̃0 get closer increasing
the value of the vacuum field, while the converse happens when the plasma column
is convex. In this case too the presence of a longitudinal magnetic field opposes the
instability because of the tension the deformation induces on the field lines.
6.3.2 The Confinement of Solar Coronal Loops
The case we have just discussed illustrates how a plasma column can be confined
by the magnetic pressure present in a vacuum region surrounding the plasma itself.
Of course, this is not the only way to obtain a confinement. A possible alternative
is to use the gas pressure of an external medium to confine a magnetic flux tube,
Fig. 6.7 Instability of a

plasma column with m = 1
(kink instability)
whose interior contains both matter and field. An astrophysical example of such a
configuration is provided by solar coronal loops, shown in Fig. 6.8.
The structures we see in the figure are (unresolved) magnetic flux tubes, immersed
in a plasma whose temperature is typically higher than the internal temperature of
the tubes, while the external magnetic field is typically weaker than the internal
one. The figure suggests that a loop should be considered as a sort of half-torus,
obtained by cutting in two the entire torus and by anchoring the bases to the solar
photosphere. Given that the aspect ratio of the torus is clearly very large, we may
Fig. 6.8 Solar coronal loops observed in the ultraviolet (photo TRACE-NASA)
140 6 Instabilities
neglect the toroidal effects and model the loop as a straight cylinder of finite length,
L and radius a. If we impose some periodicity conditions in the z-direction, we may
somehow mimic the photospheric boundary conditions.
We are now going to discuss a model of this “cylindrical” loop, without claiming
it to be a realistic model. We rather consider this to be a useful example of the
determination of a magnetic structure in an astrophysical context. We shall assume
that the gas pressure has a minimum on the axis of the cylinder and connects smoothly
to the pressure Pc (assumed to be constant) of the external coronal plasma. We shall
further assume that outside of the tube, i.e. for r > a, the magnetic field is force-free,
∇ × B e = α B e , α = const.
The equilibrium of the plasma column is still defined by the equation
1 1
∇P = J×B= (∇ × B) × B.
c 4π
We now write the current density as J = J + J⊥ where J and J⊥ are the
components parallel and perpendicular to B, respectively, and observe that only the
J⊥ component enters the equilibrium equation. We therefore need another equation
to define J , which, obviously, can be always written in the form [16]
4π
J = λB.
c
λ is in general a function of r and t, but in the following we shall consider it to be a
constant. In cylindrical coordinates the equilibrium equation reads (see Eq. (5.41)):
Bθ2
B0z B0z + Bθ Bθ + = −4π P , (6.81)
r
where the prime indicates the derivative with respect to r . Using the definition of J
we may write
c
B · (∇ × B) = B · J = λB 2 ,
4π
and evaluating explicitly the product B · (∇ × B) we get
Bθ B0z
B0z Bθ − Bθ B0z

+ = λ(Bθ2 + B0z
2
). (6.82)
r
2 and introduce the ratio of the field com-
If we divide the preceding equation by B0z
ponents
Bθ (r )
w(r ) = ,
B0z (r )
Equation (6.82) becomes:

w
w + − λw 2 = λ. (6.83)
r
Equation (6.83), the Riccati equation, is well known in the theory of differential
equations. Its solution is obtained by defining an auxiliary function u connected to
w by the relation
u
w=− .
λu
In terms of u, Eq. (6.83) becomes
u
u + + λ2 u = 0,
r
namely the Bessel equation, whose general solution, not diverging in r = 0, is
u = C1 J0 (λr ), C1 being an arbitrary constant. The general solution for w(r ) is
therefore
J1 (λr )
w(r ) = ,
J0 (λr )
from which we obtain
Bθ (r ) = B0 (r )J1 (λr ) and B0z (r ) = B0 (r )J0 (λr ), (6.84)
B0 (r ) being an arbitrary function. To determine B0 , we insert the expressions just

found for the field components in the equilibrium equation, Eq. (6.81), and get
d 2 d P/dr
B (r ) = −8π 2 (6.85)
dr 0 J0 (λr ) + J12 (λr )
This equation has to be complemented by the boundary conditions for r = a, which

are simply the continuity of the field components and of their derivatives. When
r > a, B is a force-free field with constant α and its component are a linear com-
bination of Bessel functions of the first and second kind, J0 e Yo . In this case, the
functions Y , previously discarded because of the request of regularity in r = 0,
cannot be neglected, since the origin is no longer included in the region where B
is defined. Taking into account that J0 = −J1 and that the same relation holds for
are:
the Y functions, the continuity conditions for B0z , Bθ and B0z
B0 (a)J0 (λa) = C2 J0 (αa) + C3 Y0 (λa) (6.86a)

B0 (a)J1 (λa) = C2 J1 (αa) + C3 Y1 (λa) (6.86b)
142 6 Instabilities
λ
B0 (a)J1 (λa) = C2 J1 (αa) + C3 Y1 (λa) (6.86c)
α
From the direct comparison of Eqs. (6.86b) and (6.86c) we deduce that λ = α and
the Eq. (6.86a) thus becomes

B0 (a) − C2 J0 (αa) = C3 Y0 (αa).
Since J0 and Y0 are two independent solutions of the Bessel equation, the preceding
relation can only be satisfied by
C2 = B0 (a) and C3 = 0.
In summary, the complete solution is
Bθ (r ) = B0 (r )J1 (αr )
B0z (r ) = B0 (r )J0 (αr ), (6.87)
where the function B02 (r ) is the solution of Eq. (6.85)
a
P (s) ds
B02 (r ) = B02 (a) + 8π r <a
J02 (αs) + J12 (αs)
r
B0 (r ) = B0 (a) r a. (6.88)
Equation (6.88) shows that in our case the equilibrium is determined by the pres-
sure profile only. An interesting feature of the field defined by Eq. (6.87) is that, in
spite of the fact that it is not a force-free field (P (s) = 0), its field lines perfectly
coincide with those of a force-free field with the same value of α. It follows that the
knowledge of the profiles of the field lines does not allow to distinguish a force-free
field from a more general one. This remark should be kept in mind when it is tried to
guess the nature of a magnetic field from the reconstruction of its field lines based
on observations.
We have succeeded in determining the equilibrium structure of a system that, at
least to a certain extent, mimics a solar coronal loop. The computation of the sta-
bility properties of this field is rather complex and can be accomplished by the use
of techniques developed for laboratory plasmas, [17] well beyond the scope of this
introductory text. We shall therefore restrict ourselves to the mention of a necessary
stability criterion, very often utilized in the studies of the stability of laboratory plas-
mas, the so-called Suydam criterion. In its version valid for a cylindrical geometry,
it states that the system is locally stable when
d P(r ) B 2 (r ) d Bθ (r ) 2
−r z ln .
dr 32π dr r B0z (r )
In our case, the Suydam criterion is always satisfied for a monotonically increasing
pressure profile, while it only defines a restricted stability region for profiles present-
ing a maximum in r < a. It must be emphasized that even if the Suydam criterion is
satisfied, instabilities involving global deformations of the plasma structure cannot
be excluded.
6.3.3 The Magnetorotational Instability (MRI)
Rotation and angular momentum and its transport play a fundamental role in astro-
physical objects. Transport of angular momentum is fundamental in the formation
and evolution of accretion disks, one of the natural steps in the process of gravita-
tional collapse. Accretion disks can have widely different characteristic scales, with
radii spanning dimensions from a hundred astronomical units (AU) or more for stel-
lar disks leading to planetary systems all the way up to hundreds and thousands of
kiloparsec for galaxies, and down to an astronomical unit or less around black holes.
In many of these cases the rotating, accreting gas can become sufficiently hot as to
become at least partially ionized (either heated by the process of accretion itself or by
an outside radiation source) and must therefore be considered a plasma. In order for
such a plasma to continue to fall towards the accretion center some form of viscosity
must be present, which allows the removal of the angular momentum of the falling
accreting gas, otherwise the accretion itself would stall due to the centrifugal barrier
coming from the conservation of angular momentum. However, accreting plasma is
usually so rarified that collisions may be completely neglected, so that the collisional
viscosity can not be considered a viable mechanism for accretion.
A possible solution to this problem may come from instabilities naturally develop-
ing in a rotating plasma. There is a vast litterature on the instability of rotating fluids
in hydrodynamics: a classic example is given by the Rayleigh instability, according
to which a fluid rotating around an axis is unstable if its specific angular momentum
decreases with distance from the rotation access.
A simple way to derive the Rayleigh criterion comes from perturbing the motion
of a fluid element found at a distance r from the center of the disk, and evaluating how
the force balance between the centrifugal force and pressure gradients are altered by
the time the element reaches a new distance r , supposing that in the corresponding
motion the fluid element preserves its angular momentum (we consider also for
simplicity the density to be uniform). If only pressure and angular momentum play
a role in defining the equilibrium and we assume cylindrical symmetry around the
rotation axis, the following relation must hold in r :
1 ∂P U 2 (r )
= .
ρ ∂r r
On the other hand in the new position r angular momentum conservation for the
fluid element leads it to reach a new velocity U 2 = r 2 U 2 /r 2 around the axis, so
144 6 Instabilities
that for r > r the fluid element will feel a restoring force towards its original radius
only if
1 ∂P U 2 (r ) U 2 r 2U 2
r
=
> = 3 ,
ρ ∂r r r r
or in other words if the rotational velocity profile satisfies U (r )r > U (r )r . In terms

of the angular velocity Ω(r ) = U (r )/r and the specific angular momentum r 2 Ω the
stability condition therefore becomes
∂[r 2 Ω(r )]
> 0.
∂r
This condition is satisfied, as may be easily verified, by Keplerian orbits around
a point-like center of gravity, so that a simple hydrodynamic instability leading to
an effective viscosity allowing accretion is not generally found. However, as is often
found in plasma dynamics, the magnetic field turns out to play a fundamental role
in this case, in a way which is counter-intuitive. We have seen previously that the
magnetic field tends to have a stabilizing effect on those instabilities, such as the
Kelvin-Helmholtz or Rayleigh-Taylor instabilities, where the driving force is not
electromagnetic in nature. This is due to the fact that deformation of magnetic lines
of force by the unstable perturbing motions acts as a stabilizing energy sink for the
motions themselves. However, in the case of a plasma which is orbiting around a
central object, the magnetic field leads instead to a new type of instability, called the
magnetorotational instability or MRI for short [21].
Consider a plasma in equilibrium rotating around a mass M, with an angular
velocity given by Ω(r ) = (G M/r 3 )1/2 , in which an equilibrium magnetic field
orthogonal to the equatorial plane is present Bez . The geometry of the problem is
therefore cylindrical with coordinates (r, θ, z) and we assume the equilibrium to be
independent of the angular coordinate θ .
Consider now small deformations of the field arising from a radial velocity fluc-
tuation, and suppose that is such as to shift one plasma fluid element closer to the
center, and another one, at a different height but very close to the previous one and
belonging to the same magnetic field line, further away from the center as shown
in Fig. 6.9. This kind of perturbation would be typical, for example, of an Alfvén
wave propagating vertically along the field. As a result of this velocity fluctuation,
the plasma element finding itself closer to the center of gravity (labelled a, in the
Figure), will tend to increase its angular velocity, while the other one will tend to
slow down. On the other hand the magnetic tension created in the field line will tend
to brake the inner plasma parcel’s (a) motion and accelerate that of the plasma ele-
ment (b). Because of the loss of angular momentum, the inner parcel will tend to fall
still closer in, while the second, accelerated one will tend to move further away, and
the perturbation amplitude will increase. The net result will be a transfer of angular
momentum, mediated by the magnetic field, from the internal regions of the disk to
(r) b
Fig. 6.9 Deformation of a magnetic field line B in the magnetorotational instability. For a weak
magnetic field, the breaking of the inner plasma parcel a and the acceleration of the outer plasma
parcel b due to the magnetic field causes a to fall further and b to escape further out, leading to
instabiliy and growth of the perturbation and magnetic field
the external ones, together with a transfer of the kinetic energy of the rotating plasma
into magnetic field energy, leading to a growth of the magnetic field intensity.
Consider now how the instability develops: take small perturbations by expanding,
as usual, with the velocity field U = r Ω eθ +u1 and the magnetic field B = Bez +b1 .
For simplicity we take incompressible plasma perturbations with axial symmetry
(∂/∂θ = 0) and fluctuations in the form u1 = u(r ) exp(−iωt + ik z z + ikr r )
and similarly for the field b1 . If we consider fluctuations localized in space but
with a large wavenumber (such as the perturbation considered in the example), the
condition |k|r 1 must hold and we may neglect radial derivatives of the envelopes
u(r), b(r ) with respect to those in the exponential. This means assuming to lowest
order that u(r ), b(r ) are constants. On the other hand, care must be taken in the
linearization to include the radial derivatives of the average Keplerian motion, i.e. of
terms proportional to ∂(Ωr )/∂r . Upon linearization of the r, θ components of the
momentum and magnetic induction equations (considering the ideal MHD limit)
and taking account of the appropriate expressions for the differential operators in
cylindrical geometry, one finds:
ik z Bbr
−iωρ0 u r − 2ρ0 Ωu θ = −ikr P̄1 + , (6.89a)
4π
∂Ω ik z Bbθ
−iωρ0 u θ + ρ0 2Ω + r ur = , (6.89b)
∂r 4π
−iωρ0 u z = −ik z P̄1 , (6.89c)
ωbr = −k z Bu r (6.89d)
∂Ω
ωbθ = −k z Bu θ + ibr r , (6.89e)
∂r
146 6 Instabilities
where P̄1 is the total perturbed pressure, P1 + Bbz /4π. It is possible to eliminate P̄1 ,
by combining the equations along r and along z. u z is instead eliminated using the
fact that the velocity fluctations are taken to be incompressible:
1 ∂(r u r ) ∂u z
+ = 0.
r ∂r ∂z
If now we also ask that k z kr , is is easy to see that it is also possible to neglect
the fluctuations along z of velocity, magnetic field and pressure. Eqs. (6.89a)–(6.89e)
therefore become
ik z Bbr
−iωρ0 u r − 2ρ0 Ωu θ = + , (6.90a)
4π
κ2 ik z Bbθ
−iωρ0 u θ + ρ0 ur = , (6.90b)
2Ω 4π
ωbr = −k z Bu r (6.90c)
∂Ω
ωbθ = −k z Bu θ + ibr r , (6.90d)
∂r
where we have written
∂Ω κ2
2Ω + r = ,
∂r 2Ω
introducing the epicyclic frequency
∂Ω 2
κ 2 = 4Ω 2 + .
∂ lnr
The dispersion relation describing the MRI instability may therefore be found by
calculating the determinant of the linear system (6.90a)–(6.90d), and it is a simple
algebraic calculation that leads to:

∂Ω 2
ω4 − κ 2 + 2(k z ca )2 ) ω2 + (k z ca )2 (k z ca )2 + = 0. (6.91)
∂ lnr
In the absence of a magnetic field ca = 0, Eq. (6.91) shows the existence of

oscillations at the epicyclic frequency ω2 = κ 2 , which correspond to the small
oscillations arising from elliptical orbits neighboring the circular orbits of the original
keplerian disk. If on the other hand we allow the rotational terms in the dispersion
relation to vanish we obtain Alfvén waves propagating along the unperturbed average
field B (the quartic comes from the existence of waves with independent polarizations
along r and along θ ).
The simultaneous presence of both rotational and magnetic terms leads however
to the existence of unstable solutions, i.e. ω2 < 0, as long as
∂Ω(r ) ∂Ω 2 (r )
<0 and (k z ca )2 < −r . (6.92)
∂r ∂r
The proof is left as a problem (Problem 6.3).
In a Keplerian disk, the gradient ∂Ω(r )/∂r = −3Ω/2r , so that orbiting ionized
plasmas tend to be generally unstable. It is important to emphasize that the MRI
instability exists even for extremely weak fields, and that in this case it does not
depend much on the field geometry. In addition, the MRI instability has growth rates
which can reach a significant fraction of the original rotation frequency, up to a
maximum value
1 ∂Ω 2 (r )
|ωm | = .
2 ∂ lnr
The nonlinear evolution of the instability can lead to developed magnetohydro-
dynamic turbulence and a consequent effective viscosity in accretion disks. Detailed
studies of the nonlinear evolution of the MRI, the nature of the turbulence and its
saturation in different accreting systems are still at the forefront of contemporary
astrophysical research.
6.1 Consider a configuration of the theta-pinch type, but with the plasma column
in direct contact with a rigid and perfectly conducting wall at a distance a from
the cylinder’s axis. Deduce the dispersion relation for the m = 0 modes, assuming
cs2 < ca2 . What changes if cs2 > ca2 ?
6.2 Deduce the condition for the gravitational collapse of a mass M, Eq. (6.50).
6.3 Derive the magneto-rotational instability condition, Eq. (6.92). Show that in the
keplerian case such a condition becomes (k z ca )2 < 3Ω 2 .
Solutions
6.1 The equation for the perturbed quantities is still Eq. (6.73), but the boundary
conditions (6.65) now require ξr (0) = ξr (a) = 0. To determine the explicit form
of the dispersion relation we need to know the sign of K 2 . However, the solutions
corresponding to K 2 > 0 do not satisfy the new boundary conditions, since the
function I0 is a monotonically increasing function of its argument. Thus, K 2 < 0.
Assuming that cs2 < ca2 from Eq. (6.74) we see that this condition can be satisfied
either if ω2 > k 2 ca2 or if k 2 cs2 ca2 /(cs2 + ca2 ) < ω2 < k 2 cs2 . Therefore, the solution
for ξz is proportional to J0 (|K |r ) and that for ξr is proportional to J0 (|K |r ), i.e. to
148 6 Instabilities
J1 (|K |r ). The allowed frequencies are thus given by the condition
J1 (|K |a) = 0,
namely |K |a = j1n , n = 0, 1 . . ., where j1n is the n−th zero of the function

J1 . Writing explicitly the preceding condition we find that the frequencies are the
solutions of the equation:
ω4 − ω2 (cs2 + ca2 )(k 2 + j1n

2
/a 2 ) + k 2 cs2 ca2 (k 2 + j1n
2
/a 2 ) = 0.
If cs2 > ca2 , K 2 < 0 when either ω2 > k 2 cs2 or k 2 cs2 ca2 /(cs2 + ca2 ) < ω2 < k 2 ca2 ,
while the final equation for ω remains the same, as one could have predicted from
the symmetry of that equation with respect to the exchange of cs2 with ca2 ).
6.2 To form a gravitational condensation the gravitational force must dominate over
the pressure force, namely
d P(r ) Gm(r )ρ(r ) P0 GM

< ⇒ < .
dr r2 ρ0 R
Introducing the sound speed, c02 = P0 /ρ0 , and expressing the mass as a function of
the density we obtain:
3 1/2 c0
R > RJ = ( ) ,
4π (Gρ0 )1/2
or, in terms of masses,
4π 3 c03
M > MJ = ρ0 R 3J = ( )1/2 1/2
.
3 4π G 3/2 ρ 0
6.3 Introducing the quantities ωa2 = k z2 ca2 and ω12 = ωa2 − ω2 , Eq. (6.91) can be
written as
ω14 + κ 2 ω12 − 4ωa2 Ω 2 = 0,
which has only one positive root. Unstable solutions (ω2 < 0) must have ω12 > ωa2 .
If this condition is introduced into the positive root of the preceding equation, we
find
∂Ω 2
ωa2 < 4Ω 2 − κ 2 = −r .
∂r
In the keplerian case we have Ω 2 = G M/r 3 , ∂Ω 2 /∂r = −3Ω/r and the

instability condition becomes ωa2 < 3Ω 2 .
Chapter 7
Waves
Abstract The propagation of small amplitude, linear waves in plasmas is presented,

starting from the MHD regime and the Alfvén and magnetoacoustic modes. High
frequency effects leading to kinetic corrections are then discussed before introducing
waves in completely kinetic regimes, where the phenomenon of Landau damping for
ion-acoustic modes is treated in detail.
Waves are generated when a stable equilibrium is perturbed. The dynamics of such
a system, as already remarked, is characterized by oscillations. This means that the
value of any physical quantity at a given point oscillates around its equilibrium
value. If a stone is thrown in a pond, the initially flat, horizontal water surface is
warped, at first close to the point where the stone has fallen and then in increasingly
larger zones. This example suggests defining a wave a propagating disturbance. The
fluid particles within the perturbation region oscillate with a certain speed and the
perturbed zone itself spreads at a different speed. It is the latter, called the propagation
speed, that controls the dynamics of the system. Notice that in the preceding example
the fluid particles oscillate essentially in the vertical direction, while the propagation
speed is directed along the water surface: in this case the wave is called transverse.
Longitudinal waves, in which the two speeds are parallel to each other, also exist.
The best known longitudinal waves are sound waves.
In this Chapter we shall be concerned with “small amplitude” waves, or linear
waves, meaning that the values of any oscillating quantity do not differ markedly from
the equilibrium values. In the preceding example, this means that the local increase
or decrease of the water level due to the oscillatory motion of the fluid particles is
small with respect to the pond’s depth, but does not mean that the propagation speed
is small. It does mean however that all terms in the equations in which perturbations
appear at second or higher powers may be neglected when compared to first order
terms.
The starting point for the treatment of linear waves is the system of differential
equations describing the dynamics of the physical system under study. This could
be the system of MHD equations, but waves can exist in all regimes, from fluid to
kinetic. A perturbative analysis applied to these equations reduces them to a system

150 7 Waves
of linear equations, to which the method of Fourier analysis can be applied. It is

therefore appropriate, before delving into the discussion of waves in plasmas, to
review the fundamental aspects of this method.
7.1 The Fourier Representation
Consider a generic function f (x), where x is a spatial coordinate, and its Fourier
transform, f˜(k), defined by:
∞
1
f˜(k) = f (x)e−ikx dx. (7.1)
2π
−∞
In the context of the Fourier method, the original function f (x) is also called the
inverse Fourier transform of f˜(k). Clearly (see (6.14)):
∞
f (x) = f˜(k)eikx dk. (7.2)
−∞
To illustrate the relationship between a function and its Fourier transform, consider
the special case where f (x) is an infinitely extended, constant amplitude plane wave
of wavelength λ0 = 2π/k0 :
f (x) = A eik0 x .
Its Fourier transform is given by:
∞
1
f˜(k) = Aei(k0 −k)x dx = Aδ(k0 − k).
2π
−∞
We thus see that while the function does not identify any particular region of space,
its transform is localized in k-space with infinite precision.
A “wave packet”, is a function which oscillates as above but in which the envelope
A is not constant, but localized in x-space. As an example, a Gaussian wave-packet
has the form
f (x) = f 0 e−a
2x2
eik0 x , (7.3)
representing an oscillation of wavelength λ = 2π/k0 whose amplitude is modulated

by a gaussian function. The parameter a gives us a measure of the width of the
7.1 The Fourier Representation 151
gaussian: at x0 = 1/a the value of the amplitude has dropped to f 0 /e. The Fourier
transform of Eq. (7.3) is given by:
f 0 −(k−k0 )2 /4a 2
f˜(k) = √ e , (7.4)
2a π
which shows that the Fourier transform of a gaussian wave-packet is still a gaussian,
whose width, defined by the range over which the amplitude remains greater than
a fraction 1/e of its maximum value, is now k − k0 = 2a. A more general way to
define the “localization region” of a wave-packet is to write:
(Δx)2 = (x − x)2 = x 2 − 2xx + x 2

= x 2 − x2 = x 2 , (7.5)
since x = 0. Thus, for a gaussian wave-packet,

∞
x 2 e−a
2x2
dx
(Δx)2 z = x 2 = −∞
∞ . (7.6)
−a 2 x 2
e dx
−∞
Defining the localization in k-space in the same way, (Δk)2 , and performing
the integrals, it is easily found that for Gaussian wave-packets
Δx 2 Δk 2 = 1/4,
while in the more general case of a non-gaussian packet, it is possible to show that
the preceding equality changes into
Δx 2 Δk 2 ≥ 1/4. (7.7)
The above results show that to a more localized function in the x-space, cor-
responds a less localized transform in k-space and viceversa. This property of the
Fourier transforms is strongly reminiscent of the Uncertainty Principle of Heisenberg,
and is, in fact, equivalent to it: if f (x) is the wave-function for the position of a par-
ticle, recalling that in quantum mechanics momentum is the conjugate operator with
respect to position, defined as p = k, one finds, mutiplying Eq. (7.7) by , that
ΔxΔp .
All preceding expressions are easily generalized to three dimensions plus time
for a function f = f (r, t):
∞
f (r, t) = f˜(k, ω)ei(k·r−ωt) dk dω, (7.8)
−∞
152 7 Waves
and
∞
1
f˜(k, ω) = f (r, t)e−i(k·r−ωt) dr dt, (7.9)
(2π )4
−∞
where k and ω are real quantities.

Equation (7.8) illustrates the physical meaning of the Fourier transform: the given
function is considered a superposition of “elementary waves”, represented by the
factor exp[i(k · r − ωt)], whose amplitude is given by f˜(k, ω). The quantity
ω
Φ = k · r − ωt = k r · ek − t (7.10)
k
is called the “phase” of the wave.
Equation (7.10) shows that, in each elementary wave, space and time coordinates
appear only in the combination r · ek − (ω/k)t. The planes Φ = const. are therefore
seen to move in the direction of ek at the phase velocity:
ω
vf = ek . (7.11)
k
This may be summarized qualitatively by saying that each elementary wave moves
with its own phase velocity, given by Eq. (7.11), without changing its amplitude: at
any particular time the profile of f (r, t) is “reconstructed” by adding the contribu-
tions of all elementary waves. Since in general the phase speed is different for the
various elementary waves, the “reconstructed” profile at time t will be modified with
respect to the initial profile. This phenomenon is called “dispersion”. If it happens
that the phase speed is the same for all elementary waves, or, equivalently, if ω is a
linear function of k (see Eq. (7.11)), the profile will propagate without distortion.
In describing the above procedure the dynamics of the phenomenon underlying
wave propagation and the meaning of the parameter ω have been left unspecified. We
have implied that ω is a function of k but have not specified whence this relationship
comes from. In the study of the propagation of small amplitude waves however,
f (r, t) is the solution of a linear, homogeneous differential equation in the variables r
and t. This equation will, in general, contain the operators ∇ and ∂/∂t. If f (r, t) is
represented in the form of Eq. (7.8), we immediately realize that:
∞
∂f
= (−iω f˜)ei(k·r−ωt) dk dω,
∂t
−∞
and, similarly:
∞
∇f = (ik f˜)ei(k·r−ωt) dk dω.
−∞
If we assume that the equilibrium is also homogeneous in space, so that the linear
system has constant coefficients, in the space of Fourier transforms the operators ∇ e
∂/∂t become simply the multipliers ik e −iω. This result holds true also for repeated
applications of the above operators, ∂ 2 f /∂t 2 → −ω2 f˜, or for the vector calculus
formulas when f is a vector quantity, ∇ · f → ik · f˜ and so on.
As a result, the differential homogeneous equation in (r, t) space, that we sym-
bolically write as:
D̃(∇, ∂/∂t) f = 0, (7.12)
where D̃ is a given functional, in the transformed space (k, ω) becomes:
D̃(ik, −iω) f˜ = 0. (7.13)
Therefore, in the transformed space we have to solve an algebraic equation in place of

a differential equation, and the condition for the existence of non vanishing solutions
to Eq. (7.13) may simply be written as
D̃(ik, −iω) = 0.
The preceding equation, known as the dispersion relation, provides the connection
between ω and k. In general, the dispersion relation has a finite number of discrete
solutions, known as normal modes:
ω = ωα (k), α = 1, 2, . . . , N . (7.14)
The condition D̃(ik, −iω) = 0 can be formally introduced into Eq. (7.8) by
writing:

N
f˜(k, ω) = f˜α (k)δ[ω − ωα (k)],
α=1
with the quantities ωα given by Eq. (7.14). Performing the ω integration in Eq. (7.8),
we see that the general solution of our problem can be cast in the form:
N ∞

f (r, t) = f˜α (k)ei[k·r−ωα (k) t] dk. (7.15)
α=1−∞
To simplify the notation, the quantity f˜α (k, ωα (k)) has been written as f˜α (k).
The preceding results are easily generalized to vector quantities. For instance
Eq. (7.12) becomes:
D̃(∇, ∂/∂t) f = 0,
154 7 Waves
where D̃ is a functional tensor and Eq. (7.13) is now:
D̃(ik, −iω) f˜ = 0.
The dispersion relation, i.e. the condition that non-vanishing solutions exist, is now
given by
D(k, ω) = Det[ D̃(ik, −iω)] = 0. (7.16)
To each solution of the dispersion relation is associated an eigenvector f˜ that char-

acterize that particular propagation mode.
7.1.1 Phase Velocity and Group Velocity
The phase velocity has been defined as the propagation speed of elementary waves,
v ph = (ω/k)ek . This velocity is not directly associated with any physical effect,
and specifically can not be associated with the transmission of signals (information)
or the transfer of energy (or energy flux propagation). In fact, elementary waves
have a constant amplitude and are infinitely extended (not localized) in space. Their
“motion” is not physically observable, since as time proceeds the wave remains
identical to itself modulo a phase (which repeats periodically). Therefore, even a
phase velocity larger than the speed of light, c, would not violate the principle of
relativity stating that the speed of light provides an upper limit to the speed of any
signal.
A completely different situation is that of a wave packet, with the wave-
perturbation localized in a particular region of space. The motion of a packet is
observable and this motion is associated with the propagation of information and
energy. The propagation speed of a wave packet is known as the group velocity.
To find an expression for the group velocity, consider a wave train of finite length,
composed of an exponential and a modulating amplitude, for example a gaussian
packet. Further assume that the packet’s “spectrum”, i.e. the ensemble of wave vectors
representing it in k-space, peaks around a particular value, k0 (for the case of a
gaussian wave-packet this implies a/k0 1). The normal mode associated with
k0 will thus be “dominant” and in the integral of Eq. (7.15) only k values close to
k0 will contribute. We may therefore expand the quantity ωα (k) in the integral in a
Taylor series centered around k0 . Restricting ourselves to first order, we may write,
dropping the index α,
∂ω
ω(k) ω(k0 ) + (k − k0 )i = ω0 + (k − k0 ) · v g ,
∂ki k0
i
where
∂ω
vg = . (7.17)
∂k k0
Equation (7.15) (considering only one term in the summation) can now be
written as
∞
f (r, t) = f˜(k)ei(k−k0 )·(r−v g t) dk eik0 ·r−ω0 t .
−∞
The expression in square brackets represents a generic function, A, of the variable

(r − v g t), so that we finally obtain
f (r, t) = A(r − v g t) eik0 ·r−ω0 t .
The last equation shows that for each wave mode, namely for each value of the
parameter α in Eq. (7.15), the solution is an infinite plane wave, whose wave vector
is k0 and whose frequency is ω0 , propagating with the phase speed v ph = (ω0 /k0 )ek ,
having an amplitude modulated by the function A(r − v g t), which propagates with
the group velocity v g = (∂ω/∂ k)k0 . Because the group velocity can be associated
with energy propagation, the relationship vg < c must hold quite generally.
7.2 Waves in the Ideal MHD Regime
The starting point for the study of ideal MHD waves is the system of linearized
MHD equations (4.2), (6.9), (6.10), (6.4) and (6.12). If the only forces present are
magnetic and pressure forces, we may put f 1 = 0 in Eq. (6.12). If we further
assume a uniform unperturbed state, we may Fourier transform with respect to both
space and time coordinates. The system of equations for the transformed quantities
(where for simplicity the “tilde” has been omitted since no confusion may arise)
becomes:
ρ1 = −iρ0 (k · ξ )
P1 = −iρ0 cs2 (k · ξ )
B 1 = ik × (ξ × B 0 )
i
−ω2 ρ0 ξ = −ik P1 + [(k × B 1 ) × B 0 ] (7.18)
4π
156 7 Waves
Introducing the expressions for P1 and B1 into the last equation, an equation con-
taining only the vector displacement ξ is obtained:
1
ω2 ρ0 ξ = ρ0 cs2 k(k · ξ ) + {k × [k × (ξ × B 0 )] × B 0 } (7.19)
4π
Writing B 0 = B0 eb and introducing the Alfvén speed, ca2 = B0 /(4πρ0 ), we finally

get
ω2 ξ = cs2 k(k · ξ ) + ca2 {k × [k × (ξ × eb )] × eb } (7.20)
7.2.1 Magnetic Waves
Consider first the simple case where plasma compressibility, represented here by the
presence of cs2 , can be neglected, i.e. cs ca Expanding the triple vector products
entering Eq. (7.20) we obtain the following expression:
ω2 ξ = ca2 {(k · eb )2 ξ + [(k · ξ ) − (k · eb )(ξ · eb )]k − (k · ξ )(k · eb )eb }. (7.21)
Taking the scalar product of eb with Eq. (7.21) we see that (unless ω2 = 0)
ξ · eb = 0.
The displacements, and therefore the velocities, of the particles for oscillations with
non-vanishing frequencies are perpendicular to the direction of the unperturbed mag-
netic field B 0 . Using this condition in Eq. (7.21) and taking the scalar product with k
we get
(ω2 − k 2 ca2 )(k · ξ ) = 0. (7.22)
The preceding equation has two possible solutions: k · ξ = 0 or ω2 = k 2 ca2 , that will
now be examined in turn.
7.2.1.1 k · ξ = 0, Alfvén Waves
The condition k · ξ = 0, which in ordinary space is equivalent to ∇ · ξ = 0, implies

that the perturbations are incompressible. Clearly, this does not mean that the plasma
is incompressible, but simply that the perturbations do not produce density variations.
If the above condition is introduced into Eq. (7.21) we obtain the dispersion relation:
ω2 = (k · eb )2 ca2 = k 2 ca2 cos2 θ, (7.23)

7.2 Waves in the Ideal MHD Regime 157
where θ is the angle between the propagation vector k and the direction of the
unperturbed magnetic field B 0 . Taking the scalar product of eb with the third equa-
tion (7.18) we see that
B 1 · eb = 0, (7.24)
showing that magnetic field perturbations are also perpendicular to the magnetic
field B0 . These waves, called Alfvén waves, are transverse waves, both for the dis-
placement and magnetic field perturbation.
The dispersion relation for Alfvén waves shows that their phase velocity
v ph = (ω/k)ek = ±(ca cos θ )ek
depends on the propagation angle with respect to B 0 . However, since the dispersion
relation (7.23) contains only the component of k along B 0 , the group velocity is
directed along eb ,
v g = ± ca eb .
Thus, even if the Alfvén wave-front advances in a direction different from that of
B0 , the energy associated with the wave propagates either along or opposite to the
unperturbed magnetic field B0 . The waves are therefore intrinsically anisotropic in
their propagation.
Using again the expression for B 1 given by Eq. (7.18), the condition given by
Eq. (7.24) and the dispersion relation (7.23), it is easy to derive a relationship between
the velocity U = (∂ξ /∂t) → −iωξ and the magnetic field perturbation B 1 :
B1 U
=± . (7.25)
B0 ca
This relation characterizes Alfvén waves propagating in either direction (along or

against the mean field) exactly and shows that the kinetic and the magnetic energy
of the waves is the same
E kin 1
ρ0 U 2
= 2 2 = 1,
E mag B1 /8π
i.e. that kinetic and magnetic field energies are in equipartition. To understand the
physical origin of Alfvén waves, consider the linearized form of the magnetic force,
F1:
F 1 ∝ (k × B 1 ) × B 0 = (k · B 0 )B 1 − (B 0 · B 1 )k.
The first term on the rhs derives from magnetic tension, while the second is connected
with magnetic pressure (it is the linearized form of the term ∇(B 2 /8π )). Since the
latter term vanishes in this case because of Eq. (7.24), we find that Alfvén waves
158 7 Waves
are caused by perturbations in the magnetic tension. Therefore they bear a strong
resemblance to the (transverse) oscillations of stretched elastic string, such as the
vibrations of chords in a musical instrument. It is known that the √phase velocity of
the waves propagating on a vibrating string is given by u ph = T /σ , where T is
the tension of the string and σ is the linear mass density, i.e. the mass per unit length.
Substituting σ = ρ0 S, where S is the section of the string and the magnetic tension
T = (B02 /4π )S, we get u ph = ca . This result clarifies and expands the observation
made in Chap. 5 of the analogy between magnetic lines of force and elastic strings.
An important feature of Alfvén waves is that they may be a solution not only of
the linearized but also of the full MHD equations. Consider a homogeneous plasma
(i.e. with constant P0 , ρ0 , B 0 ) and a perturbation B = B 0 + B 1 , without assuming
that |B1 | |B0 |. If the following conditions are satisfied
B1 U
|B 0 + B 1 | = const. , =± , (7.26)
B0 ca
it is easy to show (see Exercise 7.1) that this perturbation is an exact solution of
the MHD equations. This means that if fluctuations of velocity and magnetic field
obeying the preceding relation are observed, even of large amplitude, they can be
identified as nonlinear Alfvén waves. Waves of this nature have been observed in the
solar wind.
7.2.1.2 k · ξ = 0, Compressible Alfvén Waves
If k · ξ = 0 the solution of Eq. (7.22) becomes simply
ω2 = k 2 ca2 . (7.27)
These are compressible waves with identical isotropic phase and group velocities:
v ph = v g = ±ca ek .
Since the condition ξ · eb = 0 is still valid, but now ξ · ek = 0, it follows that ξ is
a vector normal to B0 lying in the plane defined by B 0 and k. It is interesting to note
that for propagation perpendicular to B 0 (θ = π/2), ξ is directed along k and the
waves becomes a longitudinal wave. For propagation parallel to B 0 , (θ = 0, ek eb ),
ξ · eb = 0 implies ξ · ek = 0 and the wave becomes transverse and incompressible,
and thus indistinguishable from an Alfvén wave.
7.2.2 Magnetosonic Waves
Let’s now restore the effects of the plasma compressibility and utilize the full form
of Eq. (7.20). Expanding the triple vector products, Eq. (7.21) becomes:
ω2 ξ = cs2 (k·ξ )k+ca2 {(k·eb )2 ξ +[(k·ξ )−(k·eb )(ξ ·eb )]k−(k·ξ )(k·eb )eb }. (7.28)
Following the procedure already employed for the magnetic waves, let’s take the
scalar product of the preceding equation with eb and with k, to obtain
ω2 (ξ · eb ) = cs2 (k · ξ )(k · eb ), (7.29)
and
[ω2 − k 2 (cs2 + ca2 )](k · ξ ) = −k 2 ca2 (k · eb )(ξ · eb ). (7.30)
If (k · ξ ) = 0, Eq. (7.29) shows that (ξ · eb ) = 0 as well. Introducing these constraints

in Eq. (7.28), we obtain again the dispersion relation for the Alfvén waves, Eq. (7.23),
ω2 = k 2 ca2 cos2 θ . These incompressible waves are therefore a solution even when
the medium compressibility is taken into account.
If (k · ξ ) = 0, we multiply Eq. (7.30) by (k · eb ) and use Eq. (7.29) to obtain the
following equation:
ω4 − ω2 k 2 (cs2 + ca2 ) + cs2 ca2 k 4 cos2 θ = 0. (7.31)
The two solutions (in ω2 ) provide the dispersion relations of the magnetosonic waves:
ω 2
1 2
= (c + ca2 ) ± cs4 + ca4 − 2cs2 ca2 cos 2θ . (7.32)
k 2 s
The two modes corresponding, respectively, to the plus or minus sign in Eq. (7.32)
are known as the fast and slow magnetosonic wave. The Alfvén wave has a phase
velocity intermediate between the two.
The propagation characteristics depend on the ratio cs2 /ca2 . In fact, for propagation
parallel to the magnetic field B 0 , ω/k tends to cs for the fast wave and to ca for the
slow one, if cs > ca . The situation is just the opposite if cs < ca . To summarize,
we may say that when θ → 0 the phase speed of the fast wave tends to the higher
of ca and ca , while the slow wave tends to the lower of the two. When θ → π/2,
the phase speed of the slow wave tends to zero, while that of the fast wave tends to
(cs2 + ca2 )1/2 . It is interesting to remark, by taking the product of the squares of the
phase speeds of the fast and slow modes (v F and v S respectively) from Eq. (7.32),
that it is always true that v 2F v S2 = cs2 ca2 cos 2 θ i.e. the product is equal to the square
of the sound speed times the square of the Alfvén wave phase speed. Obtaining the
group velocities of the waves requires taking the gradient of expression Eq. (7.32)
with respect to the wave-vector. Since the algebra is somewhat involved the results
are best summarized using polar plots of the group velocities as a function of the
angle between velocity and the magnetic field, known as the Friedrichs diagram.
This is shown, together with the polar plots of the phase-speeds (as a function of the
angle between wave-vector and magnetic field), for the fast, slow and Alfvén waves
in two cases, one for a sound speed below the Alfvén speed (plasma β < 1, Fig. 7.1)
and one for the opposite case (plasma β > 1, Fig. 7.2).
160 7 Waves
F F
S
B
. A S S
.
A B
Fig. 7.1 Phase and group velocities of MHD waves for a plasma in which cs /ca = 0.6 (ca has
been normalized to 1). The left hand panel shows, in polar coordinates, the phase speeds of the fast
(F), slow (S), and Alfvén (A) waves as a function of the angle between the wave-vector and the
magnetic field, aligned with the x-axis. The right hand panel shows, again in polar coordinates,
the magnitude of the group velocities for the fast (F), slow (S), and Alfvén waves (the thick dot in
x = ±1), again as a function of the angle between the group velocity and the magnetic field
F F
. .
A
S B A S S A B
Fig. 7.2 Phase and group velocities of MHD waves for a plasma in which cs /ca = 1.5 (ca has
been normalized to 1). The left hand panel shows, in polar coordinates, the phase speeds of the
fast (F), slow (S), and Alfvén (A) waves as a function of the angle between the wave-vector and
the magnetic field, aligned with the x-axis. The right hand panel shows, again in polar coordinates,
the magnitude of the group velocities for the fast (F), slow (S), and Alfvén waves (the thick dot in
x = ±1), as a function of the angle between the group velocity and the magnetic field
For both cases one can see that the fast mode wave is almost isotropic, with
only modest changes in speed between the perpendicular and parallel propagation
as described above. In the figures, the Alfvén speed is normalized to 1 so the Alfvén
wave is described by circles centered in x = ±0.5 and radius 0.5 in the polar
phase speed plot. The group velocity on the other hand is given by the dark dots,
identifying vg = ca = ±1. The main difference between β < 1 and β > 1 lies
in the curves that are tangent in parallel propagation: these are the fast and Alfvén
for β < 1 and the slow and Alfvén for β > 1. Also, note the strange shape of
the Friedrichs diagram for the slow wave. In parallel propagation, it is equal to the
sound speed, but as the angle between wave-vector and field increases, the group
velocity reaches a maximum angle of propagation with respect to the magnetic field,
typically a fairly small angle, and then returns to parallel propagation as the wave-
vector become perpendicular to the field. The point at which the cusp forms on the
x-axis, corresponding to orthogonal field and wave-vector, is called the cusp speed
vC . It shows that even though the slow wave has a vanishing phase speed at orthogonal
propagation, like the Alfvén speed it retains a finite group velocity parallel to the
magnetic field. It is left as an exercise to calculate its value
ca cs
vC = .
(cs2 + ca2 )
7.3 Fluid Waves Beyond the MHD Regime
The preceding Section was devoted to a discussion of wave modes in the MHD
regime. The solutions found are exact, but, of course, their validity is limited to the
range of validity of the equations from which they have been derived. Recalling the
discussion in Chap. 5, where we have shown that the MHD regime is a low frequency
regime, it is natural to ask how the wave modes change with increasing frequency, but
remaining within a framework of an infinite conductivity, ideal plasma model. As a
general rule, we may say that the MHD equations are valid for frequencies lower than
the lowest of the characteristic frequencies of the plasma. For instance, in the physical
conditions prevailing in the quiet solar corona, n e n p 108 cm−3 , B 10 G,
we get ω pe 5.6 × 108 s−1 , ω pi 1.3 × 107 s−1 , ωce 1.8 × 108 s−1 , ωci
9.6×104 s−1 . Thus, in the case of magnetosonic waves for which ω ∝ k, ω can easily
reach values taking them beyond the validity limit of MHD equations for sufficiently
large k’s.
7.3.1 Intermediate Frequencies: ω ωce
An intuition on the changes occurring when the frequencies becomes higher can be
obtained recalling the comparison of the orders of magnitude of the terms entering
the generalized Ohm’s equation (5.5), restricting our attention to the “cold plasma”
case, where the terms connected with pressure gradients are absent. It is easy to
verify that the first term that must be kept when the frequency increases is the Hall
term, namely the term proportional to J × B. In Chap. 5 it was shown that the ratio
of this term to the ideal one, (1/c)(U × B), is (c/U)2 ω(ωce /ω2pe ). Considering
again the example of the solar corona and assuming U 0.01c, we see that the two
terms are comparable when ω ωci . An analytical study of the waves supported
162 7 Waves
by a cold plasma for propagation at arbitrary angles with respect to the magnetic
field is possible but algebraically involved. Here we shall limit ourselves to the case
of parallel propagation, a simplification which still allows understanding how wave
modes change when moving beyond MHD to high frequencies (close to the proton
cyclotron frequency).
Consider as usual B 0 = B0 eb , and take k = keb so that the divergenceless
condition ∇ · B 1 = 0, or equivalently k · B 1 = 0, implies that eb · B 1 = 0. Fourier
transforming the linearized MHD equations, but without introducing the lagrangian
displacement ξ , leads to:
ca2 kc2 kc2

ωU = − (k × B 1 ) × eb = − a [B 1 − (eb · B 1 )eb ] = − a B 1 .
B0 B0 B0
The generalized Ohm equation, keeping only the dominant Hall term, reads:
i B0 k 2 c
ω B 1 + keb × (U × B 0 ) = eb × [(eb × B 1 ) × eb ].
4π n 0 e
Using the preceding expression for U and expanding the vector products, we finally
obtain:
(ω2 − k 2 ca2 )B 1 − i(ω/ωci )k 2 ca2 (eb × B 1 ) = 0. (7.33)
Choosing the z-axis along B 0 (eb = ez ) and writing the components of Eq. (7.33)
explicitly these become:
(ω2 − k 2 ca2 )B1x + i (ω/ωci )k 2 ca2 B1y = 0

(ω2 − k 2 ca2 )B1y − i (ω/ωci )k 2 ca2 B1x = 0
(ω2 − k 2 ca2 )B1z = 0 (7.34)
If B1z = 0 we must have ω2 = k 2 ca2 , from which it follows that B1x = B1y = 0.
If, on the contrary, B1z = 0, the solution of the problem is obtained by requiring
that the determinant of the coefficients of the first two equations of the system must
vanish (7.34):
(ω2 − k 2 ca2 )2 − (ω/ωci )2 k 4 ca4

= ω2 − k 2 ca2 + (ω/ωci )k 2 ca2 ω2 − k 2 ca2 − (ω/ωci )k 2 ca2 = 0.
The system then splits into two separate second degree equations, each having only
one positive root. Consider first the equation:
ω2 − k 2 ca2 + (ω/ωci )k 2 ca2 = 0,

7.3 Fluid Waves Beyond the MHD Regime 163
whose positive root is:

1 2 2
ω= −k ca + 2 .
k 4 ca4 + 4k 2 ca2 ωci
2ωci
The limiting values for small and large k’s are:
ω k ca (k → 0), and ω ωci (k → ∞).
Turning now to the solution of equation:
ω2 − k 2 ca2 − (ω/ωci )k 2 ca2 = 0,
we have

1 2 2
ω= k ca + 2 ,
k 4 ca4 + 4k 2 ca2 ωci
2ωci
with limiting values,
k 2 ca2
ω k ca (k → 0) but ω (k → ∞).
ωci
We thus see the wave modes obey the same dispersion relation for k → 0, namely
for low frequencies, in agreement with the results found for parallel propagating
MHD waves [see Eqs. (7.23) and (7.27)]. However, for large values of k, the two
dispersion relations are considerably different: in the first case, the frequency remains
finite and tends to ωci , while in the second case it diverges quadratically with k.
However, our treatment cannot be extended to exceedingly large values of k, and
consequently of ω, since it only takes into account the Hall term in the equation
for B and it is easy to see that this approximation ceases to be valid for values of ω
considerably higher than ωci .
Before moving to still higher frequencies, it is interesting to determine the char-
acteristic features of the two wave modes just found. If we consider the first solution
when k → ∞ and replace ω with ωci in Eq. (7.34), we get
i B1y
−B1x + i B1y = 0 namely = 1.
B1x
This relation implies that we are in the presence of circularly polarized waves that
appear to rotate clockwise when we look in a direction antiparallel to B 0 . To verify
this statement, remember that the quantities B1x and B1y , entering the preceding
equations, are actually the Fourier components of the respective physical quantities.
Introducing again the notation f˜ for the Fourier transform of a generic quantity f ,
164 7 Waves
we may write the condition found above as
B̃1y = −i B̃1x .
But,
B1x ≡ Re( B̃1x ) = |B1 | cos ωt
and
B1y ≡ Re( B̃1y ) = −Re(i B̃1x ) = −|B1 | sin ωt,
where |B1 | is the magnitude of the vector B 1 . Recalling Maxwell’s equation for
∇× E, (ez × Ẽ ∝ B 1 ), we see that the same relationship holds among the components
of Ẽ. Therefore
E x = |E| cos ωt ; E y = −|E| sin ωt.
At time t = 0 the field is directed along x. As time increases, E x decreases but

remains positive, while E y takes up negative values. This shows that the vector E
rotates in the sense indicated above. If we recall the discussion of Chap. 2, we see that
this is the sense of rotation of the positive particles under the action of the magnetic
field B0 . Therefore, the sense of rotation of the wave’s electric field coincides with
that of positive ions, thus justifying the name of ion cyclotron waves given to this
mode. It is clear that this situation enhances the interaction between the wave and
the positive ions, thus favoring the birth of resonance phenomena.
Repeating the procedure for the second solution, we get:
i B1y
B1x + i B1y = 0 or = −1,
B1x
and in this case the sense of rotation agrees with that of the electrons. It is not
surprising therefore that the exact theory for this type of waves predicts that when
k → ∞, the frequency tends to ωce and that a resonance might take place between
the wave and the electrons. For intermediate values of ω, ωci ω ωce , it is
possible to show that this so-called whistler wave obeys the dispersion relation
k 2 c2 ωce
ω
ω2pe
7.3.1.1 High Frequencies: ω ω pe
Once again we shall restrict ourselves to a simple case, that of a cold unmagnetized
plasma. At sufficiently high frequencies, we can no longer neglect the displacement
current in Maxwell’s equation for ∇ × B. On the other hand ions, because of their
larger mass, are unable to follow the oscillations over the fast timescales ω−1 .
Therefore they can be approximated by a fixed background, their only function being
to provide the charge required to ensure plasma quasi-neutrality. Our description will
thus refer to the electronic component only. The relevant equations are:
e
−iωu = − E,
me
ω
k×E = B1,
c
4π ω
ik × B 1 = J 1 − i E, (7.35)
c c
where u is the electrons’ velocity and J 1 = −en 0 u. Notice that all quantities entering
the preceding equations are Fourier transforms. By combining the above equations,
we arrive without difficulty at the system:
(ω2pe − ω2 + k 2 c2 )E = c2 (k · E)k. (7.36)
Taking the scalar product of the preceding expression with k, we get:
(ω2pe − ω2 )(k · E) = 0, (7.37)
that shows that two types of waves are possible, depending on whether (k · E) is
different from zero or not.
In the first case
ω = ω pe ,
and, introducing the above equality in Eq. (7.36), we see that E is directed along k.
On the other hand, the first of Eq. (7.35) tells us that also u is parallel to k. We
therefore have a longitudinal wave with a vanishing magnetic field, B 1 = 0, as
shown by the second of Eq. (7.35). These electrostatic waves are known as plasma
waves or Langmuir waves. In the cold plasma limit so far adopted, these waves do
not propagate and are therefore stationary oscillations, caused by the reaction of the
plasma to local violations of charge neutrality: since (k · E) = 0, the equation for
∇ · E, ik · E = 4πq = −4π e n 1 confirms that a non-vanishing charge density is
present.
The introduction of thermal effects modifies the picture, as shown by the intro-
duction of the term proportional to the pressure gradient in the equation of motion
of the electron fluid. In this case:
∂u
m e n0 = −∇ P1 − en 0 E = γ cs2 ∇ρ1 − en 0 E,
∂t
166 7 Waves
where use has been made of the definition

kB T
cs2 = .
me
The Boltzmann constant has been written as k B to avoid confusion with the magnitude
of the wave vector k. Using the continuity equation to eliminate ρ1 and Fourier
transforming, we get:
ω2 u = γ k 2 cs2 u − i eω E.
Since we are dealing with longitudinal waves, the only non-vanishing component of
the preceding equation is along ek . Therefore:
e
(ω2 − γ k 2 cs2 )u = −i ωE = −ω2pe u,
me
the third of Eq. (7.35) (with B 1 = 0) has been used together with the definition of
J 1 = −en 0 u. We thus arrive at the dispersion relation:
ω2 = ω2pe + γ k 2 cs2 .
Therefore, in a “warm” plasma, the wave propagates, but only for frequencies higher
than the plasma frequency, ω > ω pe . When ω ω pe , the phase speed tends to γ cs .
The value of γ is connected with the number of degrees of freedom, s, of the plasma
particles, γ = 1 + 2/s. In our case the particle motion is strictly one-dimensional,
since the oscillations take place in the k-direction and we have neglected the effects
of collisions that may scatter the electrons in other directions. Therefore, s = 1 and
γ = 3. The final form of the dispersion relation thus reads:
kB T
ω2 = ω2pe + 3k 2 cs2 = ω2pe + 3k 2 ( ). (7.38)
me
Consider now the second solution of Eq. (7.37),
k · E = 0.
In this case Eq. (7.36) immediately yields the dispersion relation
ω2 = ω2pe + k 2 c2 . (7.39)
Now the wave’s magnetic field is given by B 1 = (ω/c)(k× E) = 0 and is orthogonal

to both E e k. This solution thus represents transverse electromagnetic waves, the
generalization of the well known electromagnetic waves that propagate in vacuum.
Their phase velocity is
ω
v ph = = c 1 + (ω2pe /k 2 c2 ),
k
greater than the speed of light c. As already remarked, this does not constitute a
problem. However, we may verify that the group velocity vg < c. In fact,
∂ω k c2
vg = = c2 = < c.
∂k ω v ph
These waves also propagate only for frequencies ω > ω pe , a circumstance of great
relevance for radio broadcasting. In fact, low-frequency waves propagating in the
terrestrial atmosphere eventually reach its farthermost layer, formed by an ionized
gas (the so-called ionosphere) and may find themselves in a situation where their
frequency is lower than the local plasma frequency, ω < ω pe . The propagation is
then stopped and the waves undergo total reflection back towards the Earth’s surface.
This phenomenon allows the connection by radio of points on the terrestrial surface
that are not “in sight” of each other as first demonstrated by Marconi.
7.4 Waves in Kinetic Regimes: The Case of Landau Damping
We shall now examine in some detail a phenomenon of great interest that opens the
way to a better understanding of the physics of plasmas at microscopic level and of
the connection between kinetic and fluid models.
So far, we have been concerned with the description of waves starting from fluid
models in different regimes: MHD, non-MHD one-fluid. When fluid models cease
to be adequate, we must go back to the more general kinetic description and ana-
lyze wave phenomena in this context. In 1946 Landau attacked the problem of the
propagation of electrostatic waves in a collisionless plasma and showed that the
amplitude of such waves must eventually decrease with time. This result is very
surprising, since we are used to thinking that wave damping is caused by the action
of dissipative processes that, in turn, depend on the presence of collisions.
As we have already seen in Chap. 3, a non-collisional plasma is described by the
Vlasov equation, that therefore will be chosen as the basis of our treatment. Assume
that the plasma is homogeneous, with a vanishing magnetic field at equilibrium,
and that the ions form as before a fixed background providing charge neutrality.
Given these assumptions, consider again only the electron component, whose dis-
tribution is denoted by f (r, v, t). In agreement with the procedure already adopted
for the fluid case, consider an equilibrium state, represented by f 0 and a pertur-
bation f 1 , with | f 1 | | f 0 |, around which the Vlasov equation for the electrons
will be linearized. In the unperturbed situation, the electric field must vanish, other-
wise the plasma particles would be permanently accelerated without the possibility of
168 7 Waves
reaching an equilibrium state. It is then clear that any arbitrary function of the velocity,
f 0 = f 0 (v), would be a solution of Vlasov’s equation at the zeroth order:
∂ f0 ∂ f0
+v· = 0.
∂t ∂r
In the following we shall assume that f 0 is the maxwellian distribution function:
m 3/2 mv 2
f0 = n0 exp − .
2π k B T 2k B T
The total electron distribution function is written as:
f (r, v, t) = f 0 (v) + f 1 (r, v, t),
and the linearized form of the Vlasov equation will then be:
∂ f1 ∂ f1 e E ∂ f0
+v· − · = 0, (7.40)
∂t ∂r m ∂v
where m is the electron mass, the electric field being a first order quantity.
The preceding equation must be coupled with that for the electric field:

∇ · E(r, t) = 4 πq = −4π e f 1 (r, v, t)dv.
Finally, because of the assumption of homogeneity, we can Fourier transform with

respect to the spatial coordinates and write:

f 1 (r, v, t) = f˜1 (k, v, t) eik·.r dk.

E(r, t) = Ẽ(k, v, t) eik·.r dk.
The Fourier transform of Eq. (7.40) now reads:
∂ f˜1 e Ẽ ∂ f 0
+ i k · v f˜1 − · = 0. (7.41)
∂t m ∂v
Since we are restricting to the case of longitudinal waves,
∇ × E = 0, ⇒ i k × Ẽ = 0,
7.4 Waves in Kinetic Regimes: The Case of Landau Damping 169
we are allowed to choose a reference system where
k = [k, 0, 0], Ẽ = [ Ẽ, 0, 0 ], v = [ u, v y , vz ].
Unless explicitly indicated, in the following we shall always assume that k > 0.
The formal solution of the problem is found by solving first the equations for he
Fourier transforms
∂ f˜1 (k, v, t) e Ẽ ∂ f 0 (v)

+ i ku f˜1 (k, v, t) − =0 (7.42)
∂t m ∂u
and

i k Ẽ(k, t) = −4π e f˜1 (k, v, t)dv, (7.43)
and then transforming back to the coordinates in physical space.

As far as time dependence is concerned, one might Fourier transform with respect
to this variable as well, since the unperturbed state is stationary. However, if we
assume that our perturbation is applied at time t = 0 it is preferable, following the
original treatment by Landau, to perform a Laplace transform with respect to time.
Let’s begin by defining and reviewing some aspects of the Laplace transform.
Given a generic function of the time g(t), we define its Laplace transform as:
∞
ĝ( p) = g(t) e− p t dt,
0
where p is a complex number with Re p > 0. It may be shown that the inverse
Laplace transform is obtained by an integration along a line parallel to the imaginary
axis, lying to the right of all the singularities of ĝ( p):
σ+i∞
1
g(t) = ĝ( p) e p t d p.
2π i
σ −i∞
Also notice that the Laplace transform is formally equivalent to a Fourier transform in
which the substitution −iω → p has been made. The use of the Laplace transform
turns out to be especially useful when the solution of an initial value differential
equation is sought. Consider for instance the first order equation:
dg
+ a = b(t) ; g(t = 0) = g(0)
dt
170 7 Waves
with a = const. and b(t) a known function of t. Laplace transforming the preceding
equation we get:
∞ ∞ ∞
dg − p t −p t
e dt + a e dt = b(t)e− p t dt.
dt
0 0 0
An integration by parts of the first term gives:

∞

g(t)e− p t
+ p ĝ( p) + a ĝ( p) = b̂( p),
0
and finally
−g(0) + (a + p) ĝ( p) = b̂( p).
Notice that the initial condition for g(t) is automatically included in the solution
for ĝ( p). The function g(t), solution of the original equation and satisfying the given
initial condition, is obtained by performing the inverse Laplace transform of ĝ( p).
Returning to our problem, take the Laplace transform of Eq. (7.42):
e ∂ f0
p fˆ1 (k, v, p) − f˜1 (k, v, t = 0) + ik u fˆ1 (k, v, p) − Ê(k, p) = 0,
m ∂u
To simplify the notations we shall omit the explicit indication of the variables k and
p in fˆ1 and Ê and drop the index 1 in f 1 . The preceding equation now reads:
e ∂ f0
( p + ik u) fˆ = f˜(0) + Ê . (7.44)
m ∂u
Repeating the procedure with Eq. (7.43) and using of Eq. (7.44) we obtain:

f˜(0) e2 (∂ f 0 /∂u)
i k Ê = −4π e dv − 4π Ê dv,
p + iku m p + iku
that we write in the form

4π e2 (∂ f 0 /∂u) 4π e f˜(0)
1−i dv Ê = i dv. (7.45)
mk p + iku k p + iku
Introduce the notation:

1 m 1/2 mu 2
F0 (u) = f 0 (v) dv y dvz = exp − , (7.46)
n0 2π k B t 2k B T
if f 0 is a maxwellian. If we further define:
∞
4π e2 n 0 (dF0 /du)
D(k, p) = 1 − i du
mk p + iku
−∞
∞
ω2pe (dF0 /du)
=1−i du, (7.47)
k p + iku
−∞
where n 0 is the electron density in the unperturbed state, we finally arrive at an

explicit expression for Ê(k, p):

4π e 1 f˜(k, v, t = 0)
Ê(k, p) = i dv. (7.48)
k D(k, p) p + iku
To obtain E(r, t) the inverse Laplace and Fourier transform of Eq. (7.48) must
be taken and this introduces a few difficulties, starting with the inverse Laplace
transform of Ê(k, p), namely
σ+i∞
1
Ẽ(k, t) = Ê(k, p) e p t d p. (7.49)
2π i
σ −i∞
To find the value of the parameter σ , we must know the position of all the singularities
of Ê(k, p) in the complex p-plane. However, if f˜(k, v, t = 0) and (dF0 /du) are
analytic functions of u, it is possible to show that the only singularities of Ê(k, p)
are the poles corresponding to the zeroes of D(k, p), which we shall assume to be
all distinct, for simplicity. Some of these poles could have a positive real part, which
would imply that σ > 0. In this case the function under the integral in Eq. (7.49)
would contain a factor eσ t , diverging when t → ∞. However, the integration path
in the complex p-plane can be modified in the way shown in Fig. 7.3. Notice that the
vertical segments of the integration path (where Re p = −α) do not contribute to
the integral when t → +∞, because of the factor e−α t . The contributions from the
oppositely directed horizontal segments cancel out as well. Thus the only part left is
the contribution from the paths (of infinitesimal radius) circling the poles:

1
Ẽ(k, t) = Ê(k, p) e p t d p.
2π i
If the solutions of the equation
D(k, p j ) = 0, (7.50)
172 7 Waves
Fig. 7.3 a Original (a)

integration path; b Modified
integration path
(b)
are indicated by p j , ( j = 1, 2 . . . N ), writing
p j (k) = γ j (k) − i ω j (k),
the following expression for Ẽ(k, t) is obtained:

N
Ẽ(k, t) = R j eγ j t e−iω j t , (7.51)
j=1
where R j is a quantity known in the theory of functions of complex variable as the

residue of Ê(k, p) in p = p j .
The electric field is therefore given by a superposition of waves, whose amplitude
increases or decreases exponentially with time, according to the sign of γ . All terms
with γ j < 0 represent damped oscillations, the more so the highest the value of |γ j |.
When the time elapsed from the initial excitation is sufficiently long, the dominant
term will be the one corresponding to the pole closest to the imaginary axis of p.
The terms with γ j > 0, on the other hand, represent amplified oscillations that very
quickly get out of the linear regime. To stay within the framework of the present
theory of small amplitude waves, we must assume that, even if poles with Re p > 0
do exist, they lie close to the imaginary axis of p. In both cases, we are concerned
only with poles of D(k, p) whose γ is sufficiently small, i.e. |γ /ω| 1.
Going back to the dispersion relation of our low-amplitude electrostatic waves,
Eq. (7.50), if we examine the expression for D(k, p), Eq. (7.47), we see that the
integrand has a pole for u = i p/k. Originally, the Laplace transform had been
defined for Re p > 0, which implies that the pole lies in the half-space Im u > 0,
external to the integration path that runs along the real axis of u. However, when
we deformed the integration path of the inverse Laplace transform, we assumed that
Re p = α < 0. Therefore, we must find an analytical continuation of the integral
entering Eq. (7.47) when Re p < 0. Such a continuation is obtained, once again, by
deforming the integration path so that the pole remains located above the new path,
as shown in Fig. 7.4.
floatbelowskip-3pc
Making use of this prescription, known as Landau prescription, we can define
D(k, p) for any value of p, in particular for Re p = γ → 0, which is the most
interesting case for as, as remarked above. By applying the above procedure and
recalling the definition of principal part of an integral in the complex plane:
∞ − ∞
f (z) f (z) f (z)
P dz = lim dz + dz ,
z − z0 →0 z − z0 z − z0
−∞ −∞
we may write the integral appearing in Eq. (7.47) as
(a) (b)
Fig. 7.4 The integration path according to the Landau prescription. a Original path, b Modified
path
174 7 Waves
∞ ∞
(dF0 /du) 1 F0
du = du
p + iku ik u − i p/k
−∞ −∞
∞
1 F0
= P du + i π F0 (i p/k)
ik u − i p/k
−∞
∞
F0 π
=P du + F0 (i p/k). (7.52)
p + iku k
−∞
The dispersion relation will be given by the solution of the equation:
∞
ω2pe F0 π
D(k, p) = 1 − i P du + F0 (i p/k)
k p + iku k
−∞
∞
F0 π
= 1 + ω2pe P du − i F (i p/k) = 0, (7.53)
( p + iku)2 k2 0
−∞
where an integration by parts has been carried out. For small k-values we take a
series expansion of ( p + iku)−2 to get
∞
ω2pe 2iku 3k 2 u 2 π ω2pe
1+ F0 1 − − + · · · du − i F (i p/k) = 0,
p2 p p2 k2 0
−∞
Note that the imaginary terms in the integrand are odd functions of u and therefore
the corresponding integrals vanish, since F0 is an even function of the same variable.
The preceding equation therefore reduces to:
k 2 c2 π ω2pe
− p 2 ω2pe 1 − 3 2s − i 2 p 2 F0 (i p/k), (7.54)
p k
where use has been made of the integrals:
∞ ∞
kB T
F0 du = 1, u 2 F0 du = = cs2 .
m
−∞ −∞
In the limit k → 0, the dispersion relation reduces to
− p 2 = ω2 − γ 2 + 2iγ ω = ω2pe ,
since F0 and all its derivatives converge so fast when the argument tends to infinity
that the last term of Eq. (7.54) tends to zero. In this limit, γ = 0 and as a consequence
the wave’s electric field, given by Eq. (7.51), is constant in time. The wave frequency
coincides with that of a Langmuir wave, ω = ±ω pe .
Waves whose amplitude is time-dependent appear when the corrections connected
with finite values of k are taken into account. At the lowest order, we may keep the
term in k 2 at the rhs of Eq. (7.54), but replace p 2 with −ω2pe , namely with the solution
of zeroth-order in k 2 . From the real part of Eq. (7.54) we get:
ω2 = ω2pe + 3k 2 cs2 , (7.55)
which is the dispersion relation of Langmuir waves corrected for thermal effects [see
Eq. (7.38)].
Finally, the imaginary part of Eq. (7.54) gives us:
π ω3pe
γ = F (ω pe /k). (7.56)
2 k2 0
Equation (7.56) is one of the most fundamental results of the theory of collisionless
plasmas. Before illustrating the physical mechanism underlying the above result, it
is important to notice that the sign of γ depends upon the sign of the first derivative
of the equilibrium distribution function, taken at the position where the speed of the
particles equals the phase velocity of the wave. The damping is thus related to the
resonant interaction of a particular group of particles with the electrostatic wave. It is
therefore not surprising that this phenomenon does not show up in a fluid description,
where the “anomalous” behaviour of a group of peculiar particles is cancelled by
the process of averaging over the velocities that lies at the basis of those models.
Since the (maxwellian) equilibrium function is a decreasing function of the velocity
for u > 0, γ will have negative values and the wave will be subject to the so-called
Landau damping.
It could be objected that F0 (u) has a positive derivative for u < 0, which might
suggest that an amplification of the wave amplitude is also a possibility. However,
when u < 0 the resonant particles have k < 0, contrary to what we have so far
assumed. We must then repeat the whole procedure for this new case and it’s easy to
realize that now the pole of D(k, p) is located in half-plane Im u < 0 and that when
γ → 0 the integration path must be modified to maintain the pole under the path.
The half circle around the pole is now travelled in a clockwise sense and this implies
a sign change in the imaginary terms of Eqs. (7.53) and (7.54). Equation (7.56) now
reads:
π ω3pe
γ =− F (−ω pe /|k|),
2 k2 0
176 7 Waves
and the sign of γ is still negative. We thus conclude that if the particles obey a
maxwellian distribution function the electrostatic waves are always damped. The
explicit form of Eq. (7.56) for a maxwellian distribution function can be written in
the following form:

π ω pe 1 3
γ =− exp − − , (7.57)
8 |kλ D |3 2(k λ D )2 2
where use has been made of
ω pe λ D = cs ,
and the term proportional to k 2 in Eq. (7.55) has been retained, because of the strong
dependence of γ on (kλ D ).
The damping appears to be very modest for small values of (kλ D ), namely for
wavelengths large with respect to λ D , but it increases rapidly with the decrease of λ,
reaching a maximum for kλ D 0.6. Notice that Eq. (7.57) has been obtained by a
series expansion in powers of the parameter kλ D 1 which loses its validity close
to one. If Eq. (7.53) is exactly solved (by a numerical procedure), the maximum
disappears and γ is found to be a monotonically increasing function of kλ D . The
conclusion is that when the latter quantity is of the order or larger than unity, the
waves are so strongly damped that in practice they do not exist.
The preceding discussion sheds some light on the physical process underlying
Landau damping. Imagine observing the phenomenon from a reference system mov-
ing at the wave’s phase speed. In that system both the electric field and its associated
electrostatic potential would appear time-independent. A particle whose total energy
is less than the maximum of the electrostatic potential energy of the wave will be
reflected at the points marked by A and B in Fig. 7.5 and will therefore be trapped
between two successive maxima of the potential.
In our system of reference, a particle moving at a speed slightly lower than the
wave’s phase speed will be seen to slowly drift towards the left to be eventually
Fig. 7.5 Potential -e

energy,−eΦ, as a function of
x in the wave’s reference
system
A B
reflected at the point A, where it will acquire energy from the wave. Conversely, a
particle moving at a speed slightly higher than the wave phase speed will drift to the
right and will reflect at the point B, transferring part of its energy to the wave. If the
distribution function is a decreasing function of the velocity close to the phase speed
of the wave, the number of particles with speeds u < ω/k will be larger then that
of the particles having u > ω/k. The absorption of energy from the wave will thus
be larger than the transmission of energy to the wave and the final result will be a
damping of the wave.
If the distribution function presents a secondary maximum, for instance for k > 0,
and if the phase speed of the wave falls in the region where F0 (u) > 0, γ becomes
positive and the wave amplitude grows feeding on the energy of the particles. Such
a situation may arise when in a thermal plasma (maxwellian distribution) a beam of
particles with speeds substantially higher than the thermal speed of the plasma is also
present. As already remarked, an exponential growth of the wave amplitude forces
the system out of the linear regime, thus invalidating our treatment. However, it easy
to guess what will be the state of the system in the non linear phase. The growth of
the wave amplitude will slow down the particles: particles with u > ω/k will be
moved to regions where u < ω/k, and this will eventually “cut” the secondary bump
of the distribution function, resulting in a saturation of the wave amplitude.
Finally, notice that our physical explanation of the linear phase of Landau damping
only applies for times shorter than the “bouncing time” of the particle in the wave’s
potential well. In fact, a particle reflected at the point A will eventually reach point
B where it will lose energy to the wave, thus partially compensating the damping.
The bouncing frequency ωb can be evaluated to a first approximation as follows.
In the wave reference system, let the electric field be given by E = E 0 sin(kx).
The corresponding potential will then be Φ = (E 0 /k) cos(kx) = Φ0 cos(kx) and
the potential energy of a test electron will be V = −eΦ = −(eE 0 /k) cos(kx), as
shown in Fig. 7.5.
The equation of motion of the test electron, with energy slightly higher than the
minimum of V , will then be:
m ẍ = −eE = −eE 0 sin(kx) −eE 0 k x,
from which it follows that the particles trapped close to the region of minimum
potential energy will harmonically oscillate with a frequency

ωb = eE 0 k/m e = k eΦ0 /m e .
The ratio between the bouncing time, τb and the wave period τ pe for ω ω pe , is
therefore

τb ω pe k B T
1,
τ pe kcs eΦ0
178 7 Waves
where we have taken into account both of Eq. (7.55) and of the basic condition for
plasmas eΦ0 /kT 1.
We thus see that the linear phase of the Landau damping might last for many wave
periods.
7.1. Show that in a homogeneous plasma an incompressible perturbation with B =

B 0 + B 1 , with an arbitrary B 1 , satisfying the conditions given by Eq. (7.26) is an
exact solution of the full MHD equations. Show also that the solution corresponds
to a transverse wave that propagates with speed ca .
7.2. Calculate the polarization of slow and fast magnetosonic waves, ξ s, f , prop-
agating at an arbitrary angle θ with respect to the magnetic field and show that
ξ f · ξ s = 0 always holds. The eigenmodes of the slow, fast and Alfvén waves are
therefore mutually orthogonal.
7.3. Calculate the limit of the group velocity of the slow mode wave when the angle
of propagation is close to 90◦ . First, expand the expression for the square of the
frequency when θ π/2 − with small , then calculate the components of the
group velocity, and then take the limit − > 0.
Solutions
√ √
7.1. Let B 0 = B0 ez , ca = B0 / 4πρ0 and b = B 1 / 4πρ0 . The momentum and
induction equations are:
∂U ∂b
− ca = (b · ∇)b − (U · ∇)U,
∂t ∂z
∂b ∂U
− ca = (b · ∇)U − (U · ∇)b,
∂t ∂z
where the following conditions have been used: ∇ · U = 0, P0 = const. and

|B 0 + B 1 | = const.
If the characteristic relations for the Alfvén waves hold, namely b = ± U, the rhs
of the preceding equations vanish identically. Deriving the first equation with respect
to t and making use of the second we get
∂2U ∂2U
− ca2 2 = 0.
∂t 2 ∂z
7.2. The polarizations of the slow and fast waves are obtained solving the equation
for the eigenvectors obtained by inserting the value of the frequency, Eq. (7.32), into
Eqs. (7.29) and (7.30). If y is the coordinate orthogonal to B in the k, B plane, we
have
ξ 2 − c2 k 2 cos2 θ
ωs, s
y f
= (7.58)
ξz s, f cs2 k 2 sin θ cos θ
The quantity ξ f · ξ s can be written as

ξ ξ
y y
ξ f · ξ s = (ξz )s (ξz ) f +1 .
ξz s ξz f
ξ
y
Using the previously found expressions for ξz and the properties of the solutions
s, f
of Eq. (7.31) we find the required result.
7.3. Expanding Eq. 7.32 for θ = π/2 − and taking the minus sign for the slow
wave leads immediately to

1 2 cs2 ca2
ωs2 /k 2 (cs2 + ca2 ) 1 − 1−4 .
2 (cs2 + ca2 )2
Expanding the square root, we find that
cs2 ca2
ωs2 /k 2 2 ,
(cs2 + ca2 )
however k 2 2 is nothing but the square of the parallel wave-vector k x2 . So the group
velocity in the y direction vanishes, while in the x-direction taking the derivative
with respect to k x yields
∂ω k x cs2 ca2 cs ca

∂k x ω(cs + ca )
2 2
(cs2 + ca2 )
which is the desired result.

Chapter 8
Shocks
Abstract Large amplitude waves, or shocks, are the subject of this chapter. Shocks
in neutral gases are briefly reviewed. MHD shocks are then treated in detail while a
brief discussion of collisionless shocks is presented in the final section.
So far we have restricted our attention to small amplitude waves. However, situations
may arise where the equilibrium state undergoes substantial changes that can no
longer be considered as small perturbations. The whole procedure of linearization
therefore becomes meaningless and a new approach is required. The most impor-
tant qualitative change when the amplitude of perturbations, or “waves”, in a fluid
becomes large is the appearance of spatial discontinuities in the physical quantities
characterizing the fluid. These discontinuities are usually referred to as shock waves
or simply shocks.
Shocks are generated both in plasmas and neutral gases, where they are produced
by supersonic motions. This can be understood by considering the effect of the motion
of a body through a gas, that we shall assume to be a perfect gas. The body generates
a disturbance that propagates in the gas at the sound speed cs = (γ P/ρ)1/2 . If the
velocity of the object is less than cs (subsonic motion), the perturbation, namely
the compressive sound wave, will always move ahead of the body which caused it.
However, if the motion of the object is supersonic the perturbation will lag behind
and the unperturbed gas will not be “alerted” of the arrival of the body by the sound
wave, as was the case for subsonic motions. The physical parameters of the gas will
then be subject to abrupt variations, whose amplitude may became extremely large,
thus forbidding a treatment in terms of the small perturbations of the linear theory.
The region where such variations take place is called shock front. The structure of a
shock front is determined by a delicate balance among steepening, arising from the
nonlinear terms (convective derivative in a fluid, and convective derivatives and field
aligned gradients in MHD), and the smoothing effect associated with viscosities and
resistivities, which depend on collisions or other irreversible processes which allow
heating of the gas across the shock.
As remarked several times, the dissipative effects connected to thermal conduc-
tivity, viscosity (and resistivity) are generally negligible in gases and plasmas. How-
ever, they may become important in regions of strong gradients, as can be seen by

182 8 Shocks
examining the mathematical structure of the dissipative terms in the fluid equations.
These are given by the product of a coefficient (usually depending on the thermo-
dynamical parameters) times the gradient of a fluid variable. In the case of thermal
effects we have the coefficient of thermal conductivity multiplied by the temperature
gradient, viscous effects are described by the viscosity coefficient multiplied by the
velocity gradient, and so on.
It is therefore clear that if the gradients of the fluid variables become large, the
importance of the dissipative terms will accordingly increase, at least locally. On the
other hand, the formation of strong gradients is a typical nonlinear effect, coming
into play when the amplitude of perturbations increases. This statement can be easily
verified by considering the acceleration of a fluid particle, appearing on the lhs of
the one-dimensional equation of motion:
dU ∂U ∂U
= +U .
dt ∂t ∂x
The second term is manifestly nonlinear and, to illustrate its effects, assume for
example to have a sinusoidal velocity perturbation so that U ∝ sin(kx), and dU/dx ∝
cos(kx). Thus:
∂U
U ∝ sin(kx) cos(kx) ∝ sin(2kx)
∂x
and this shows that the nonlinear term generates quantities that vary on a spatial
scale 1/2k, smaller than the scale of the gradients of U, namely 1/k. If U is given by a
Fourier expansion containing a finite number of terms of non negligible amplitude, we
easily realize that the product of that series by the corresponding series of derivatives
will eventually create a disturbance with increasingly strong spatial gradients until
their further growth is stopped by the counteracting effect of the dissipative terms.
The balance, in the case of small dissipative coefficients, requires large values of
the gradients across the shock front, which in turn demands a small shock thickness.
Since dissipative coefficients are generally very small in gases and plasmas, the
shocks can be practically considered as discontinuities.
This qualitative introductory discussion suggests that shocks may be described as
a transition zone (of vanishingly small thickness) between a region of undisturbed
flow, that will be identified by the index 1, and a region that has been subject to
the action of the shock, identified by the index 2. These regions are often indicated
as the zone ahead of and behind the shock or upstream and downstream of the
shock. It is usually assumed that both regions are homogeneous, which is not true
in practice, since the region behind the shock relaxes in time to a state different
from that prevailing immediately after the passage of the shock. The homogeneity
assumption makes sense only in the framework of a local analysis, namely for times
shorter than the typical relaxation times and for distances small with respect to the
macroscopic scale of the system.
8 Shocks 183
The equations describing a discontinuity are simply a series of relationships that

connect the values of the fluid (and electromagnetic) variables in the two homo-
geneous regions upstream and downstream of the shock, where it is reasonable to
assume that the ideal fluid (or MHD) equations are valid. These relationships are
known as the Rankine-Hugoniot relations or jump conditions. In the spirit of the
local analysis we just mentioned, we may consider the situation as stationary and
adopt a frame of reference attached to the shock front. In this system, the fluid
enters the shock with a speed U 1 and leaves with a speed U 2 . In the system that
better corresponds to the physical situation, the so-called “laboratory system”, the
plasma velocity in Region 1 is zero and the shock front moves into that region with
a velocity U shock = −U 1 , while in Region 2 the plasma moves with the speed
U + U shock = U 2 − U 1 .
The treatment of shocks in plasmas is considerably more complex than in neutral
gases, because of the larger number of effective degrees of freedom of a plasma. In
fact, let’s consider the motion of a piston in a neutral gas producing a compressive
wave in the direction of motion. If transverse motions are also present, they do not
alter the situation and the system has therefore only one effective degree of freedom.
But if we consider the motion of a conducting piston in an ideal magnetized plasma,
any transverse motion of the piston would drag the magnetic field, generating a
magnetic component of the perturbation. A magnetized plasma described by MHD
has three effective degrees of freedom, one associated with each of the propagating
modes. The situation is reminiscent of that encountered when studying the linear
MHD waves, where we had an incompressible transverse mode (the Alfvén waves)
and two “mixed” modes (the magnetosonic waves) whose nature is intermediate
between a sound wave and a magnetic wave. In the nonlinear case as well, we have
three waves with different propagation speeds, generally called fast, intermediate
and slow. The waves propagating at the intermediate speed are purely transverse and
do not produce shocks, while the other two exhibit both longitudinal and transverse
components and give rise to shocks.
Before discussing the formation of discontinuities within magnetohydrodynam-
ics, it is important to mention that the mechanisms leading to the formation of such
discontinuities, typically nonlinear steepening, may also lead to the formation of
gradients that are too strong for the fluid description of the plasma to remain valid.
In fluid shocks the thickness of the discontinuity, as we shall see, is typically set by
the collisional particle mean free path, or taking into account magnetic fields, the
particle gyration radius. If however the scale of gradients turns out to be smaller
than such length-scales, then phenomena other than those described by MHD, for
example wave-particle interactions, must mediate the dissipation of energy required
across the discontinuity. In fact, in most space and astrophysical plasmas, the shocks
that form are collisionless, i.e. the shock structure is not well described by MHD.
A detailed discussion of the physics of collisionless shocks is beyond the scope of
this work: we will however return to them briefly at the end of the chapter.
184 8 Shocks
8.1 The Jump Conditions
Consider a plane shock front of vanishing thickness (i.e. a discontinuity occurring

along only one direction with all quantities homogeneous in planes orthogonal to
this direction) moving in a direction normal to the the front itself and identify such
a direction with the unit vector ex . The planes x = const. will thus be parallel to the
shock front. With the assumed symmetry, all quantities will depend on the variable x
only.
Although there may be discontinuities in some of the fluid quantities across the
front, the dynamics must not violate the fundamental conservation laws of mass
momentum and energy, which maintain their validity. To derive relationships between
the quantities downstream and upstream of the shock we therefore take the MHD
equations in conservative form and integrate them in a right cylindrical volume V
containing the shock front (the classic pill box) defined as follows. The bases of
the volume, A1 and A2 straddle the shock, while the side, of height, h, is parallel
to the shock normal direction and will eventually tend to zero. From the continuity
equation, we get:

∂ρ ∂ρ
dV + ∇ · (ρ U) dV = dV + (ρ U) · ns dS = 0,
∂t ∂t
V S
where ns is the unit vector normal to the surface S pointing outwards from the volume
itself. In our reference frame therefore ns = ± ex . The volume integral tends to zero
when h → 0 (and V → 0 as well). The surface integral splits into the two integrals
on the base surfaces, because the contribution from the sides vanishes as the cylinder
shrinks to zero,

(ρU) · ex dS − (ρU) · ex dS = 0.
A1 A2
Since this relation has to remain valid for any area of the base surfaces, in the limit
h → 0 we obtain:
(ρUx )1 = (ρUx )2 ,
where the indices 1 and 2 indicate that the quantities in brackets are evaluated at the
upstream and downstream bases, respectively. In the theory of shocks, it is customary
to introduce the notation:

G ≡ G2 − G1 ,
8.1 The Jump Conditions 185
to denote the jump in any quantity G (scalar or vector). The first jump condition,
deduced from the continuity equation, thus reads:

ρ Ux = ρ U · ex = 0.
More generally, without making reference to any particular coordinate system, if n

is the unit vector normal to the shock front, we write

ρ(U · n) = 0. (8.1)
Notice that this jump condition can be obtained directly from the (stationary)
continuity equation by applying the formal substitution (∇ ·) → (n ·) and setting the
jump of the quantity obtained in this way to zero. Following the same procedure, the
jump conditions deriving from the momentum and energy equations (5.22 and 5.23)
may be found and read:
B2 B
ρ U(U · n) + P + n− (B · n) = 0, (8.2)
8π 4π
and
γP c
2 ρU + (U · n) + (E × B) · n = 0
1 2
(8.3)
γ −1 4π
If we now adopt the following representation for the magnetic field
B = Bn + Bt = Bn n + Bt ,
thus separating the component normal to the plane containing the discontinuity from
the “tangential” component lying on that plane, and analogous representations for
the other vectors, Maxwell’s equations ∇ · B = 0 and ∇ × E = −∂B/∂t = 0 (with
the replacement ∇ → n) provide the conditions

Bn = 0,
and

Et = 0,
namely the conservation relations already known from electrodynamics. Using the
equation E = −(1/c)(U × B) to eliminate the electric field, the jump conditions can
be cast in the following form:

ρUn = 0, (8.4)
186 8 Shocks
B2
ρUn2 + P + t = 0, (8.5)
8π
Bn
ρUn U t − Bt = 0, (8.6)
4π
γP Bn2 + Bt2 Bn
1
ρU 2
+ + U n − (U · B) =0 (8.7)
2 γ −1 4π 4π

Bn = 0, (8.8)

Et = 0, (8.9)

U n × Bt + U t × Bn = 0, (8.10)
The presence of a discontinuity in the flow does not necessarily imply that a
shock wave is present. Discontinuities may be classified on the basis of the jump
on whether Un = 0 or Un = 0.
conditions described above, and their nature depends
In the first
case, if there is no jump in density, ρ = 0, no discontinuity is present,
while if ρ = 0 and there is a jump in density, but no flow across it, we have a
so-called contact discontinuity. In the second case in which there is flow upstream
and (Un = 0), if there is no jump in density ρ = 0 we are in the presence of a
is a flow upstream, Un = 0, and a finite jump
rotational discontinuity. Only if there
in density across the discontinuity, ρ = 0, the discontinuity is classified as a shock.
8.1.1 Contact Discontinuities
In a contact discontinuity (Un = 0) no matter flows across the discontinuity,

but a
density jump is present. If Bn = 0, Eq. (8.10) implies that U t = 0, while from
Eq. (8.6) it follows that Bt = 0. Finally, Eq. (8.5) shows that P = 0. As a
consequence, the pressure and all components of B and U are continuous across
the discontinuity and the density is the only quantity with a discontinuity. From the
equation of state it follows that the temperature must also abruptly change across the
discontinuity. This discontinuity is therefore nothing but the passage between two
plasma states in pressure equilibrium but different entropies (a hotter, lower density
plasma in contact with a cooler, higher density plasma).
Bn = 0 the jump conditions show that it is also possible to have U t = 0,
If
Bt = 0 and
B2
P + t = 0.
8π
In this circumstance we speak of a tangential discontinuity. In this type of flow, the
velocity and the magnetic field are parallel to the surface of discontinuity, but their
value and/or direction change across that surface, while the total pressure remains
unaltered. Generally a current flows along the interface between the two plasma
states, and if the tangential velocity doesn’t vanish and has a finite discontinuity, not
only is the interface a current sheet, but also a vorticity sheet. The entropy may or may
not be different, depending on whether the density jumps across the discontinuity as
well.
An astrophysical example of a tangential discontinuity is provided by the
magnetopause, i.e. the surface separating the terrestrial magnetic field, the mag-
netosphere, from the solar wind. In the absence of processes violating the Alfvén
theorem, namely of phenomena of magnetic reconnection (which will be discussed
in the next Chapter), the normal components of the velocity and of the magnetic
field turn out to be negligible (Un 0 and Bn 0) and the magnetosphere is
said to be “closed”, because the solar wind and the magnetic field associated with
it cannot penetrate inside. In this situation the magnetopause behaves practically as
a tangential discontinuity and has the same pressure in the Sunward and Earthward
directions. Similar configurations have been observed also in the magnetospheres of
other planets.
8.1.2 Rotational Discontinuities
Rotational discontinuities are characterized by a non-vanishing

normal velocity and
no density jump across the discontinuity: Un = 0, but ρ = 0. Eqs. (8.4), (8.5) and
(8.6) now imply that

Un = 0, (8.11)
B2
P + t = 0, (8.12)
8π

U t − Q = 0, (8.13)
where we have introduced the vector

Bn
Q= Bt .
4πρ Un
In terms of Q Eq. (8.7) may now be written as:

γ P B2
1 2
2 Ut + + t − Q · U t = 0. (8.14)
γ −1ρ 4πρ

The absence of a jump for a generic vector A, A = 0, implies also that
2
(A) = 0. Applying this to Eq. (8.13), it follows that :

0 = (U t − Q)2 = Ut2 + Q2 − 2Q · U t .
188 8 Shocks
Now eliminate Q · U t between the preceding equation and Eq. (8.14) to obtain:
γ P B2 B2 /4πρ
+ t 1 − 21 n 2 = 0. (8.15)
γ −1ρ 4πρ Un
From the continuity of both Bn and Un Eq. (8.10) may be rewritten as:
Bn2
Un n × Bt = Bn n × U t = Bn n × Q = n × Bt ,
4πρ Un
or

Bn2 n × Bt
n × Bt Un − = (Un2 − Bn2 /4πρ) = 0.
4πρ Un Un

Since Un = 0 and n × Bt = 0 it follows that
Bn2
Un2 = ≡ can
2
, (8.16)
4πρ
2 = B2 /(4πρ) is the Alfvén speed based on the normal magnetic field.
where can n
Therefore the rotational discontinuity advances with the Alfvén speed corresponding
to the magnetic field along the normal direction. Introducing Eq. (8.16) into Eq. (8.15)
we get
γ P B2
+ t = 0.
γ −1ρ 8πρ
By comparing the preceding expression with Eq. (8.12), we conclude that they are
compatible only if

P =0 and Bt2 = 0.
This type of discontinuity is characterized by the fact that no density or pres-

sure variations are associated with the
discontinuity, the transverse component of
the magnetic field does change, Bt = 0, but its magnitude remains unaltered,
2
Bt = 0. Thus, both the magnetic field and the velocity rotate when the disconti-
nuity is crossed, thus justifying the name given to it. The discontinuity itself moves
with respect to the unperturbed fluid at the speed Un = ±can , and depending on the
sign the correlation between the tangential velocity and magnetic field discontinuities
changes accordingly. We therefore recover the result of Chap. 7 (Exercise 7.1) con-
cerning
large amplitude, circularly polarized, Alfvén waves. For small amplitudes,
Ut and Bt can be identified with the perturbations U1 and B1 of the linear theory
and the rotational discontinuity simply becomes an incompressible Alfvén wave. In

fact the rotational discontinuity may be viewed as nothing but a propagating, large
amplitude Alfvén wave steepened to a discontinuous wave front.
Discontinuities of this type are observed when the magnetic field associated with
the solar wind has a direction opposite to the terrestrial one. In this case, magnetic
reconnection becomes important, with the effect of generating non negligible com-
ponents of the velocity and of the magnetic field along the normal to the surface
of discontinuity (Un = 0, Bn = 0). The wind and the field that goes along with it
penetrate the magnetosphere, which is then said to be “open” and the magnetopause
becomes a rotational discontinuity.
8.2 MHD Shocks
According to the classification introduced above, MHD shocks correspond

to jumps

in both density and the velocity normal to the shock front, Un = 0 and ρ = 0.
In this case the jump conditions can not be easily simplified, but the treatment can
be made more tractable by a clever choice of an appropriate reference frame, as we
shall show in the following. To begin, let’s examine the two simple cases of:
• Perpendicular shocks, characterized by the vanishing of the magnetic field normal
to the plane of the shock, B1n = 0. As we shall see, in this case it is possible to
choose a reference frame where U1t = 0 and consequently U 1 · B1 = 0, which
justifies the name given to these discontinuities.
• Parallel shocks, for which both U 1 and B1 are parallel to the shock normal n.
The general case of an arbitrary angle between U 1 and B1 is known as an oblique
shock.
8.2.1 Perpendicular Shocks
So far we have only defined the axis normal to the shock plane but not a coordinate
system within the plane itself. In the case of a perpendicular shock when the magnetic
field lies completely in the plane, it is natural to choose one of the axes in the direction
n = ex , and choose the z-axis in the direction
of B1 . Identify n with the x-axis, of B1 .
In this frame, therefore, B1 = 0, 0, B . Eqs. (8.4) and (8.6) then show that U t = 0,
which means that the transverse speed remains unaltered when the discontinuity is
crossed. We can therefore move to a new frame moving at a constant speed
parallel
to U t , where obviously the transverse speed will vanish, U 1 = U, 0, 0 . The jump
conditions now read:

ρU = 0. (8.17)
190 8 Shocks
B2
ρU 2 + P + = 0, (8.18)
8π
γ P B2 1 2 cs2
2 U + γ − 1 ρ + 4πρ = 2 U + γ − 1 + ca = 0,
1 2 2
(8.19)
B
= 0. (8.20)
ρ
Le condition expressed by Eq. (8.20) simply reflects the “freezing in” condition of
the magnetic field in the plasma [see Eq. (5.28)].
If we let B = 0 in the above relations we recover the Rankine-Hugoniot relations
for a hydrodynamic shock. In this case it is useful, making use of the jump conditions,
to express the ratios of the values of the various physical parameters upstream and
downstream of the shock in terms of the upstream Mach number,
U1
M= ,
cs1
where cs1 is the upstream sound speed. A long algebraic calculation gives:
ρ2 U1 (γ + 1)M 2
= = , (8.21)
ρ1 U2 2 + (γ − 1)M 2
P2 2γ M 2 − (γ − 1)
= . (8.22)
P1 γ +1
It is also possible to show that, as a consequence of the Second Principle of Ther-

modynamics, which requires that S2 S1 a hydrodynamic shock must always be
compressive, ρ2 ρ1 , P2 P1 and that M 1. Also, M2 = U2 /cs2 1. Thus, in
the frame of the shock, the upstream fluid motion is supersonic, while downstream
of the shock the gas motion is subsonic . From the preceding equations it follows
that in the strong shock limit, M 1, the pressure ratio P2 /P1 ∝ M 2 becomes
arbitrarily large,1 while the density ratio tends to the finite value (γ + 1)/(γ − 1).
Consequently, the ratio of temperatures can also assume very large values for strong
shocks. The presence of very high temperatures behind the shock front can initi-
ate ionization processes, so that a plasma can be generated as a consequence of
the passage of a gas through a shock front, even if the gas ahead of the shock is
neutral.
In a hydrodynamic shock part of the kinetic energy of the incoming flux is
transformed into thermal energy. In fact, from Eq. (8.18) with B = 0 we get:
γ P2

1 − P1 /P2 = 21 ρ1 U12 1 − U1 /U2 ,

γ −1
1 It is customary to use P2 /P1 as a measure of the shock strength.

8.2 MHD Shocks 191
and finally, making use of Eqs. (8.21) and (8.22) (with M = 1):
γ P2 2 γ −1 1
= 21 ρ1 U12 − .
γ −1 γ + 1 γ (γ + 1) M 2
The expression in brackets on the rhs is a decreasing function of M 2 that approaches

2/(γ + 1) when M 1. Thus for a strong shock,
2(γ − 1) 1
P2 ρ1 U12 ,
γ (γ + 1) 2
which shows that a strong shock transforms a fraction γ (γ2+1) of the initial kinetic
energy into thermal energy. For γ = 5/3, 9/20, almost one half, of the kinetic energy
is converted into heat.
When a magnetic field is present, the relations equivalent to Eqs. (8.17)–(8.20)
become considerably more complicated. Introducing, in addition to M, the parame-
ters:
ρ2
X= ,
ρ1
and
P1 2 cs1 2
β= = ,
B12 /8π γ ca1
a long calculation leads to the following expressions:
B2 U1
= = X0 , (8.23)
B1 U2
P2 γ M 2 (X0 − 1) X02 − 1
=1+ − , (8.24)
P1 X0 β
where X0 is the positive solution of the equation for the density jump:
f (X) ≡ aX 2 + bX − c = 0,
with
a = 2(2 − γ ), b = γ [2β + (γ − 1)βM 2 + 2], c = γ (γ + 1)β M 2
Since 1 γ 2, it is easy to verify that the only positive root of the preceding
equation is
1
X0 = −b + b2 + 4ac .
2a
192 8 Shocks
Using the expressions for the coefficients, it is easily verified that when β 1, X0
can be written as:
c ac2 (γ + 1)M 2
X0 ≈ − 3 = − O(1/β).
b b 2 + (γ − 1)M 2
The hydrodynamic case corresponds to the limit β → ∞ and the preceding expres-
sion shows that the effect of a magnetic field is that of reducing the value of X0 with
respect to the hydrodynamic case. Since it is possible to show that in any case the
shock is compressive, X0 1, Eq. (8.24) shows that the shock strength is reduced
by the presence of a magnetic field. This is a consequence of the fact that part of the
kinetic energy of the incoming flux can be converted not only into heat, but also into
magnetic energy.
The function f (X) exhibits a single minimum for X = −b/(2a) < 0. Thus
f (X) < 0 for 0 X X0 and, since X0 1, it follows that f (X = 1) 0.
Explicitly evaluating this expression, we find:
2 c2
M2 1 + = 1 + a2 .
βγ cs
Thus, U12 cs2 + ca2 , namely the upwind flow, or the shock’s speed in the lab system,
must be greater than the propagation speed of a fast magnetosonic wave.
8.2.2 Parallel Shocks
B1 = B1n n and B1t = 0, a possible solution of

In a completely parallel shock, with
the jump conditions is given by Bt = 0, i.e. B2 = B1 . Since in addition U1t = 0
and Eq. (8.10) guarantees that Ut = 0, it also follows that U 2 is parallel to n. In
this situation we have both Bn = 0 and Bt = 0 and all magnetic terms drop out of
the jump conditions, which thus turn out to be identical to those of a hydrodynamic
shock.
However, the preceding solution is not unique. In fact,as already remarked, in
an MHD shock part of the upstream kinetic energy can be converted into magnetic
energy. So there are possible solutions with |B2 | > |B1 |. But, the normal component
of B does not change crossing the shock front and this means that, during the crossing,
a tangential component of B that was absent in the upstream region may be generated.
In other words, the shock “switches on” a magnetic field (or better a field component),
which justifies the name of switch-on shocks given to this kind of waves. We shall
return to this point below.
8.2 MHD Shocks 193
8.2.3 Oblique Shocks
In the general case, B1 will have both normal and tangential components to the shock
plane. Adopting again the reference frame used for perpendicular shocks, i.e. the x-
axis in the direction of n and the z-axis along the tangential component of B1 , we
can write:

B1 ≡ Bx , 0, B1z . (8.25)
Because the normal component of the field is continuous, the component Bx refers
indifferently to the upstream or downstream region [see Eq. (8.8)]. As for the velocity
vector, U 1 will in general have all components different from zero,

U 1 ≡ U1x , U1y , U1z .
The discussion of oblique shocks is greatly simplified by the choice of an appro-

priate frame of reference, called the de Hoffmann-Teller frame (hereafter referred to
as dHT frame), where the vectors B and U are parallel to each other both upstream
and downstream of the shock. To show that such a frame exists, write Eqs. (8.6) and
(8.10) as:
1
(ρUn ) U t = Bn Bt , (8.26)
4π
and
Bt
(ρUn ) = Bn U t , (8.27)
ρ

where ρUn and Bn are conserved quantities. Eliminating U t between the above
equations, we get
Bt Bn2
(ρUn )2 = B t.
ρ 4π

Recalling the definition of the symbol , wee see that the preceding equation implies
(ρU )2 B2 (ρU )2 B2
n n
− n B1t = − n B2t .
ρ1 4π ρ2 4π
Since ρ1 = ρ2 , we
conclude
that the vectors B1t and B2t are parallel to each other
and to the vector Bt , which,
in turn, is parallel to U t , as shown by Eq. (8.26). The
parallelism of B1t and U t translates, in our reference frame, into the fact that the
y-component of U is continuous across the shock U1y = U2y ≡ Uy . It is therefore
clear that in a reference frame moving at a speed V = Uy ey with respect to the
194 8 Shocks
original frame, the y-components of the velocities will vanish. In the new system
therefore:

U 1 ≡ U1x , 0, U1z and U 2 ≡ U2x , 0, U2z .
The situation is illustrated in Fig. 8.1.

For the sake of simplicity, we have continued to use the same notation for the
components of the vectors B and U adopted in the original frame, but we must
remember that they now refer to the new frame. The vectors B and U are now
coplanar and the problem has been reduced to a two-dimensional one in the coordinate
plane (x, z). Consider now a further reference frame moving along the z-axis with
speed W = W ez , where W is to be determined. In this system, calling the velocity
components U1i
, (i = x, z), we have:
U1x = U1x , U1z = U1z − W ,
while the components of B remain unaltered. An appropriate choice for W then leads
to U and B becoming parallel to each other. This happens when:

U1z U1z − W B1z

= = ,
U1x U1x B1x
namely when
B1z
W = U1z − U1x .
B1x
We have thus shown that with the appropriate choice of two kinematic (Galilean)
transformations, it is possible to find a reference frame where B1 and U 1 are parallel
to each other. This is the dHT frame. Note that the method used here would not work
for perpendicular shocks, since for them B1x = 0.
Fig. 8.1 The vectors B1t ,

B2t , U 1t , U 2t and U t in the
new frame
8.2 MHD Shocks 195
The advantage of the dHT frame is clear: in fact, not only are the vectors U and B
parallel upstream of the shock, but they maintain that property also in the downstream
region, as shown by Eq. (8.10), that now reads:

U2x B2z − B2x U2z = U1x B1z − B1x U1z .
The rhs of this equation vanishes, due to the choice of the reference system: the lhs
therefore vanishes as well and the vectors U and B stay parallel when crossing the
shock. Moreover, since both upstream and downstream of the shock the plasma can
be considered ideal, in the dHT frame the electric field, E = −(1/c)(U × B), is
equal to zero and the condition (8.3) reduces to
γ P/ρ cs2
1 2
2U + (ρUx ) = 0 =⇒ 21 U 2 + = 0. (8.28)
γ −1 γ −1
The terms containing the magnetic field drop out from the energy jump condition,
that now is identical to that of a hydrodynamic shock. The other jump conditions
are still those given by Eqs. (8.4)–(8.10), that however turn out to be significantly
simplified in the new reference frame:

ρUx = 0, (8.29)
2 Bz2
ρUx + P + = 0, (8.30)
8π
Bx Bz
ρUx Uz − = 0, (8.31)
4π
1 2 cs2
U + = 0, (8.32)
2 γ −1

Bx = 0, (8.33)
U2x B2z − B2x U2z = U1x B1z − B1x U1z . (8.34)
Once more, though we have used the same notation for the components of B and U, we
have to keep in mind that they are now evaluated in the de Hoffman-Teller frame. To
deduce the value of the same quantities in the original frame of reference, we must
apply the two Galileian transformations in the opposite sense. Eqs. (8.29)–(8.34)
are seven relations among the unknown quantities (ρ, cs , Ux , Uz , Bx , Bz ) evaluated
upstream and downstream of the shock, a total of 12 quantities. The usual way of
proceeding is to fix ρ1 , cs1 , B1x , B1z and the ratio of the densities in order to obtain
a system of seven relations connecting the seven remaining unknowns, which will
be therefore expressed in terms of the five quantities whose values have been fixed.
Introducing the Alfvén speed,
B1
ca = ca1 ≡ √ ,
4πρ1
196 8 Shocks
a long algebraic calculation allows the preceding relations to be cast in the form:
U2x ρ1
= = X0−1 , (8.35a)
U1x ρ2
U2z U 2 − ca2
= 21 , (8.35b)
U1z U1 − X0 ca2
B2x
= 1, (8.35c)
B1x
B2z (U12 − ca2 )X0
= , (8.35d)
B1z U12 − X0 ca2
P2 cs2 γ − 1 U12 − U22
= X0 = X0 1 + 2
, (8.35e)
P1 cs1 2 cs1
where X0 is a positive solution of the third degree equation:

2
U12 − X ca2 2
X cs1 + 21 U12 cos2 θ X (γ − 1) − (γ + 1)

+ 21 ca2 U12 X sin2 θ γ + X (2 − γ ) U12

+ X X (γ − 1) − (γ + 1) = 0, (8.36)
and θ is the angle between B (or U) and the normal to the shock plane, ex .
To the three solutions of Eq. (8.36) correspond three types of waves, that are
characterized by the value of the propagation speed along the normal to the shock
plane, i.e. by the value of U1x = U1 cos θ . Ordering the solutions for increasing
values of U1x , we will have a slow shock, an intermediate (or alfvenic) shock and
a fast shock. These shocks correspond to the three linear MHD waves discussed in
Chap. 7. In fact, in the limit X → 1, when the discontinuity becomes infinitesimal,
Eq. (8.36) factorizes as follows:

2
U1x − ca2 cos2 θ U1x
4
− cs1
2
+ ca2 U1x
2
+ cs1
2 2
ca cos2 θ = 0,
and the Alfvén wave and the slow and fast magnetosonic waves of the linear theory
are recovered.
The intermediate shock is not really a shock at all, but a rotational discontinuity.
1 = ca , one of the solutions of Eq. (8.36) is simply X0 = 1, or, equivalently,
If U
ρ = 0. But U1x = 0 so we are back to the already discussed situation characterizing
rotational discontinuities. In the dHT frame of reference the x-components of U
and B are continuous, while their z-components change sign when crossing the
discontinuity.
The study of the general properties of oblique shocks involves cumbersome alge-
bra: therefore we shall only mention that it is possible to show that oblique shocks
8.2 MHD Shocks 197
are always compressive, X0 > 1 and that the sign of the tangential component of the
magnetic field is conserved, so that the ratio B2z /B1z is always positive. Eq. (8.35d)
then clearly shows that we have two possibilities:
U12 ca2 < X0 ca2 , or U12 X0 ca2 > ca2 .
In the first case Eq. (8.35d) implies that B2z < B1z . This is the fundamental
property of slow shocks: the magnetic field lines are refracted towards the shock
normal and thus the magnitude of B decreases when crossing the shock front.
In the second case (fast shocks), B2z > B1z and therefore the magnetic lines are
refracted away from the shock normal and the magnitude of B increases.
Two interesting cases occur when the equal sign holds in the preceding relations.
In the case of a slow shock, U1 = ca , and Eq. (8.35d), for X0 > 1, implies B1yz = 0,
but B2z = 0. The crossing of the shock turns off the tangential component of B: we
have a so-called switch-off shock. In the case of a fast shock the condition U12 = X0 ca2
is equivalent to the statement that the upstream speed (or shock speed in the laboratory
frame) must be higher than the Alfvén speed (X0 > 1). In tis case, Eq. (8.35d) tells
us that B1z = 0 (parallel shock), but B2z = 0 so that crossing the shock a tangential
component of B appears, and we recover the switch-on shock, already encountered
in the discussion of parallel shocks (Fig. 8.2).
Eliminating P2 between Eqs. (8.30) and (8.32), it is possible to deduce the fol-
lowing relationship::
2
B2z c2
2
= tan2 θ2 = (X0 − 1) (γ + 1) − (γ − 1)X0 − 2 s2 ,
B2x ca
where θ2 is the angle between B2 and the shock normal. Since the lhs is a positive
definite quantity, we obtain a limitation of the values of X0 that may possibly give
rise to a switch-on shock:
γ + 1 − 2cs2 /ca2
1 < X0 .
γ −1
We thus see that to have such type of a shock ca > cs must hold. The maximum
value of the compression ratio is obtained for ca cs , in which case we have
Fig. 8.2 The refraction of the (a) (b)

lines of B for the oblique
MHD shocks: a slow shock,
b fast shock B2 B2
B1
198 8 Shocks
X0 = (γ + 1)/(γ − 1), i.e. the same result found for a hydrodynamic shock. The
value of θ2 varies from zero for X0 = 1 to a maximum of 4(1 − cs2 /ca2 )/(γ − 1)2 for
X0 = (γ − cs2 /ca2 )/(γ − 1) < (γ + 1)/(γ − 1).
8.3 Shock Thickness
In the preceding sections shocks have been considered as discontinuities, i.e. as

geometrical surfaces of zero thickness. In reality shocks are thin transition layers
between two regions of different flow characteristics, with the thickness of a shock
resulting from a delicate balance between nonlinear effects, that tend to steepen the
wave front, and dissipative effects that oppose that tendency. That balance allows an
estimation of the thickness of the shock. The relevant dissipative effects to be taken
into account are due to viscosity and thermal conduction for neutral gases, plus Joule
heating in the case of collisional plasmas. The equations of momentum and energy
must therefore be modified to include viscous, thermal and resistive effects. The use
of a simple one-fluid description to determine the structure of a shock, however, has
to be taken with caution. In fact, especially in the case of a strong shock, the computed
thickness could easily become comparable with the ion mean free path or the ion
Larmor radius, a situation well beyond the limits of a fluid model. But even in a weak
shock it may happen that the electrons are heated well before the ions and thermal
equilibrium between species is reached only after a considerable time. A correct
treatment should therefore allow for different electron and ion temperatures.
As we have seen, the jump conditions determine the flow parameters behind the
shock in terms of the flow characteristics ahead of the shock. Thus, the total rate
of dissipation of energy, which occurs within the shock thickness δ, is in principle
known. Given a particular dissipation mechanism it is then possible to estimate the
shock thickness.
Consider first a hydrodynamic shock and assume that viscosity is the dominant
dissipative effect. Starting from the energy equation for a viscous fluid, it is possible
to show that (see e.g. Landau and Lihschitz [7]) the rate of dissipation of kinetic
energy, Ekin due to viscosity can be expressed as

dEkin ρν ∂Ui ∂Uj 2
=− + dV , (8.37)
dt 2 ∂xj ∂xi
where V is the volume and ν and ρ are the kinematic viscosity and the density, respec-
tively. Introducing the kinetic energy per unit volume, Ē = E/V , and performing a
dimensional analysis of the preceding expression we obtain

Ē ρ ν
U 2
,

t 2 δ
8.3 Shock Thickness 199
and, since
t δ/U1 ,
(
U)2

Ē ρ ν ,
2 U1 δ
from which
ρ ν(
U)2
δ . (8.38)
2 U1
Ē
In a strong hydrodynamic shock we have shown that the fraction of the initial
kinetic energy converted into heat by viscosity is2

Ē Ē ρ1 U12 .
On the other hand
(
U)2 = (U1 − U2 )2 = U12 (1 − U2 /U1 )2 U12 ,
where use has been made of Eq. (8.21). Inserting the preceding two expressions in
Eq. (8.38) we finally get
ν
δ ,
U1
which shows that in the shock the Reynolds number is of the order of unity. Recalling
the result of Problem 4.3, we write ν (P/ρ)τc . As a measure of P within the shock
we use P = 21 (P1 + P2 ) ≈ P2 /2 in a strong shock, in which case we also have
P2
≈ U12 −→ U1 ≈ P2 /ρ2 .
ρ2
Inserting this value in the estimate for δ we find

(i)
δ τc P2 /ρ2 vth τc = λ.
We thus see that the shock thickness is comparable to the ion mean free path, as
anticipated.
Consider now a strong MHD perpendicular shock and estimate the shock thickness
assuming the dominant dissipative effect is Joule heating. Then the energy dissipation
2In this formula and in the following ones in this Section we drop all factors containing γ , typically
of order one.
200 8 Shocks
rate will be proportional to j2 /σ , with j = (c/4π )(∇ × B). The energy dissipation
rate per unit volume will be

Ē U1
Ē 1 c 2 B2 − B1 2
,

t δ σ 4π δ
which gives
1 c 2 (B2 − B1 )2 η (B2 − B1 )2
δ = ,
σ 4π U1
Ē 4π U1
Ē
with η the magnetic diffusivity. If we use

Ē ≈ 2 ρ1 U1
1 2 and recall Eq. (8.20),

B/ρ = 0, we obtain for a strong shock
(B2 − B1 )2 = B12 (1 − ρ2 /ρ1 )2 B12
and
2η B12 /8π
δ .
U1 21 ρ1 U12
The preceding relation is actually a condition on the magnetic Reynolds number

inside the shock as can be seen by writing it in the form:
U1 δ B2 /8π
RM = ≈ 2 11 2.
η 2 ρ1 U1
Thus, contrary to the hydrodynamical case, the magnetic Reynolds number within
the shock is not always of the order unity, but is proportional to the upstream ratio
of magnetic and kinetic energies.
8.4 Collisionless Shocks
As mentioned at the beginning of the chapter, most shocks generated in space and
astrophysical plasmas are collisionless, in the sense that the processes governing
shock thickness, structure and dynamics are not the classical dissipation processes
due to collisional viscosities and resistivities, but involve wave-particle interac-
tions which mediate energy dissipation. The notion of collisionless shocks became
accepted by the early sixties in space, though a similar problem also confronted
researchers involved in laboratory experiments. In the solar system, the mean free
path of particles can be up to a few Astronomical Units (AU) yet shocks are observed
8.4 Collisionless Shocks 201
in front of the magnetospheres of most planets as well as traveling in the solar wind
away from the sun.
The first solutions proposed for the existence of collisionless shocks was asso-
ciated with processes of anomalous dissipation, in particular anomalous resistiv-
ity, generated by the intense currents present in within the steepened gradients of
the shock and associated with the relative motions of the current carrying particles
(mostly electrons and protons). The plasma supporting such intense currents is gener-
ally unstable and these instabilities generate waves, which can mediate the transport
of momentum and energies between particle species and within each species between
particle populations with different speeds. For example, current carrying electrons
can generate waves that may then be absorbed by ions, on scales much smaller than
the mean free path. The current instabilities were first though to involve the gen-
eration of ion-sound waves (discussed in the previous chapter section on Landau
damping) but more generally the wave-types depend in detail on the geometry of the
shock and the magnetic field configuration.
Experimental studies of shocks in space and laboratory plasmas have shown that
the characteristics of observed shocks can be strongly variable and depend crucially
on the angle between the magnetic field and the shock normal (i.e. whether the
shock is quasi-parallel or quasi-perpendicular), the ratio of plasma to magnetic
field pressures upstream of the shock, and the magnetosonic Mach number, i.e. the
ratio of the upstream flow speed to the speed of the (fast) magnetoacoustic wave
propagating in the direction of the shock normal. The main characteristic feature of
the first group of shocks is that ions reflected by the shock front can freely propagate
back up into the upstream region. Such shocks correspond to magnetic field-shock
normal angles below 45◦ . In the second case, reflected ions remain close to the
shock and can gain energy thanks to the inductive electric field tangential to the
shock surface but perpendicular to the magnetic field.
Low Mach number shocks then dissipate the required energy through anomalous
resistivity mechanisms within the current carrying layer. Right hand fast magneto-
acoustic/whistler waves have phase and group velocities that increase with decreas-
ing wavelength beyond the fluid regime. Therefore, steepened fast mode shocks are
expected to radiate short wavelength waves, and hence energy, into the un-shocked
oncoming flow. The shortest wavelength capable of standing in the flow then forms
a precursor wave-train that has been observed at these “sub-critical” shocks (but
the process occurs for all Mach numbers). Above a critical Mach number, anom-
alous resistivity within the layer carrying the limited shock current is unable to
convert the required amount of energy from directed bulk flow into thermal energy.
At such “super-critical” quasi-perpendicular shocks, a fraction of the incident ions
are reflected by the steep shock ramp as described above. They gyrate around the
magnetic field and gain energy due to acceleration by the transverse motional electric
field. Returning to the shock layer they have sufficient energy to pass through into
the downstream shocked region. The separation of ions into two groups, crossing
the front directly and after reflection, results in the dispersal of particles in veloc-
ity space. Reflected particles are separated from the bulk ion population due to an
increase in peculiar velocity relative to the bulk motion. This process corresponds to
202 8 Shocks
the kinetic heating required by the shock jump conditions and it ensures the major
part of energy dissipation necessary for directed energy transfer to thermal energy
of plasma ion population. The process of ion thermalization takes place on rather
large scale downstream of the shock front. The spatial length of the transition to
ion thermal equilibrium can be treated in a similar fashion to that of the shock front
thickness in collisional shocks. Reflection occurs on sufficiently smaller scales than
thermalization due to a combination of magnetic forces and the electrostatic cross-
shock potential, which results directly from the leading electron pressure gradient
term in the generalized Ohms Law. In more detailed two-fluid descriptions, the quasi-
perpendicular shock has fine structure that depends upon the characteristics of the
nonlinear shock profile.
At high Mach number ion reflection appears to play a crucial role for dissipation
processes also within the quasi-parallel shock. However, in the quasi-parallel shock,
there is no single, monotonic shock ramp that develops as a combined result of
reflection and gyration. At the quasi-parallel shock the characteristic length scales
depend on the coupling between waves and energetic particles, which in turn depends
on the mechanisms for extracting energetic particles from the thermal distribution.
Spacecraft observations of the quasi-parallel shock shows they they have intrinsically
unsteady behaviors with strong magnetic pulsations, and the pulsation shock layer is
embedded in turbulence between the upstream foreshock and the downstream flow
region. The structure of such shocks is studied mostly via computer simulation and
is still the study of considerable research. In the Earth’s magnetosphere, beacause
of the changing direction of the Earth’s magnetic field towards the direction of the
incoming solar wind, there are both regions of quasi-perpendicular and quasi-parallel
shocks, and this is most likely true of general bow shocks in astrophysics.
We refer the readers to two reviews on quasi-parallel and quasi perpendicular
shocks for a summary of contemporary understanding and research on this topic
(Burgess and Scholer [18]; Krasnoselskikh et al. [19]).
Chapter 9
Magnetic Reconnection
Abstract The phenomenon of magnetic reconnection and its importance and role
in many plasma dynamic phenomena are introduced and described in detail. The
stationary reconnection models of Sweet-Parker and Petschek are described before
analyzing the resistive instabilities of current sheets and, in particular, the tearing
mode. The chapter concludes with reconnection instabilities of thin sheets, or plas-
moid instabilities, and the consequent general nonstationary behavior of reconnection
processes.
The previous chapters have been devoted to the dynamics of ideal plasmas, neglecting
all effects connected with their finite resistivity; in this chapter we shall be concerned
precisely with those effects. Recalling the discussion of Chap. 5, rewrite the induction
equation describing the temporal variation of B, [see Eq. (5.25)]:
∂B
= ∇ × (U × B) + η∇ 2 B. (9.1)
∂t
The two terms on the rhs vary over different timescales which can be defined as
τ f = L/U (the fluid or convective scale) and τd = L2 /η (the diffusive or resistive
scale), where U is a typical value of the fluid velocity, L is the spatial scale of
variation of the magnetic field and η = (c2 /4π σ ) is the plasma magnetic diffusivity.
The relative importance of the two terms is measured by the value of the magnetic
Reynolds number, Rm = τd /τ f = (UL)/η.
The choice of the timescales is, at least to a certain extent, arbitrary. If, as often
the case, one is interested in dynamical situations dominated by the magnetic field, a
logical choice would be U = ca . As far as the scale L is concerned, it seems natural
at first sight to use the “macroscopic” scale typical of the fluid variables, L = L.
With these choices τ f becomes identical to τa , the time needed to cover the distance

204 9 Magnetic Reconnection
L at the Alfvén speed; the magnetic Reynolds number is then called the Lundquist
number, S:
Lca
S= . (9.2)
η
Table 5.1 in Chap. 5 displays the characteristic values of the spatial and temporal
scales of some plasmas of interest for the laboratory and for astrophysics, together
with the corresponding Lundquist numbers. It is important to distinguish the diffusive
times shown in the Table with the effective lifetimes of the corresponding magnetic
fields. In fact, if these two scales were equal, we should conclude that a sunspot
survives for a few million years, while its effective lifetime is never longer than
a few months. The Earth’s magnetic field, on the other hand, should have been
extinguished a long time ago. We shall return to this point in Chap. 11, when dealing
with the problem of the build-up and maintenance of magnetic fields.
The simple considerations above force us to conclude that there must be processes
present in nature, different from simple diffusion, that modify the decay of magnetic
fields. The existence of such mechanisms is extremely important both for laboratory
and astrophysical plasmas in all circumstances where magnetic energy is converted
into different forms of energy, in particular into thermal energy, as, for instance, in
connection with the problem of the heating of the solar corona, to which we shall
return in the following chapter.
So given the huge values of the Lundquist number, that seem to indicate that the
ideal description of a plasma should be perfectly adequate, what processes might
actually cause magnetic field dissipation and/or regeneration on such vastly faster
timescales? First, observe that the ideal limit of the induction equation can be obtained
not only by letting the magnetic diffusivity tend to zero, but also by allowing the
spatial scale L to become arbitraryly large, because of the different dependence on
the scale of the two terms of that equation. Secondly the induction equation has
a vector character, i.e. spatial anisotropy, which is lost in the simple dimensional
analyses on which the definition of the Lundquist number is based. So, for example,
the local importance of the resistive term is increased in regions where the convective
term vanishes, as, for instance, when the velocity U is parallel to B. In such regions,
the use of the macroscopic scale L is not justified and a local scale, typically of the
same order of the dimensions of the regions of vanishing convective term (potentially
much smaller than L), seems better suited. The value of S then decreases locally, the
Alfvén theorem ceases to be valid and the magnetic field topology may vary, giving
access to configurations of lower energy unattainable in an ideal MHD regime. In
such circumstances, we expect that magnetic fields dynamics develop on timescales
intermediate between τd and τ f , as we shall verify in the following.
Before considering processes violating Alfvén’s theorem, consider once more the
purely diffusive case, S 1. In this situation the Eq. (9.1) reduces to
∂B
= η∇ 2 B.
∂t
9 Magnetic Reconnection 205
Let’s assume a constant η and a unidirectional magnetic field given by B = B(x, t)e y ,
with B(−x, t) = − B(x, t). The current density will then have the form J =
J (x, t)ez . We shall assume that at the time t = 0 the current is localized in a narrow
region around x = 0 and write
J (x, 0) = J0 e−x
2 /a 2
,
The equation ∇ × B = (4π/c) J then reduces to
∂B 4π
= J, (9.3)
∂x c
so that
x 2π 3/2 J a
4π J0
e−x
2 /a 2 0
B(x, 0) = dx = erf(x/a) = B0 erf(x/a),
c c
0
where
z
2
e−x dx
2
erf(z) = √
π
0
is the error function and B0 is the asymptotic value of B when x → ∞.

The temporal evolution of J (x, t) can be found by deriving Eq. (9.3):
∂J ∂2 J
=η 2,
∂t ∂x
which is identical to the equation for B(x, t).

In Chap. 6 we have seen that the solution of the preceding equation can be found
via Fourier transforms. Recalling Eq. (5.26), we may write:

dk e−ηk t J (k) ei(k x) ,
2
J (x, t) = (9.4)
where J (k) is the Fourier transform of the current distribution at time t = 0, namely
J0 a a2k2
J (k) = √ exp − .
2 π 4
Using this expression and performing the integral in Eq. (9.4) we get
x2
J (x, t) = J0 a (4η t + a 2 )−1/2 exp − .
4η t + a 2
In terms of the non-dimensional variables ξ = x/a e τ = 4η t/a 2 the current density

reads:
exp[−ξ 2 /(1 + τ )]
J (ξ, τ ) = J0 ,
(1 + τ )1/2
and the magnetic field,
B(ξ, τ ) = B0 erf[ξ/(1 + τ )1/2 ].
The profiles of J e B as functions of ξ for different values of τ are shown in Figs. 9.1
and 9.2.
From the figures it is seen that the current distribution widens as time progresses,
while the value of B and that of its gradient decrease. The magnetic energy per unit
Fig. 9.1 J (ξ, τ ) as a function J( , )/J0

of ξ , for τ = 0, 2, 10, 50 1
0.8
=0
0.6
0.4
0.2
=50
-4 -2 2 4
Fig. 9.2 B(ξ, τ ) as a function B( , )/B0

of ξ , for τ = 0, 2, 10, 50
1 =0
0.5
= 50
-4 -2 2 4
-0.5
-1
9 Magnetic Reconnection 207
length in the y-direction varies with time as
∞ ∞
∂ B2 1 ∂B
dx = B dx.
∂t 8π 4π ∂t
−∞ −∞
Inserting the expression of ∂ B/∂t given by the induction equation in the preceding
integral and performing an integration by parts, it is easy to obtain:
∞ ∞ ∞ ∞
∂ B2 η ∂2 B η ∂ B 2 1
dx = B 2 dx = − dx = − J 2 dx.
∂t 8π 4π ∂x 4π ∂x σ
−∞ −∞ −∞ −∞
As expected, the magnetic energy is dissipated as a result of the finite value of the
electrical conductivity and converted into heat. Even when the Lundquist number is
large, and in the presence of flows, the configuration discussed above, consisting in
a so-called neutral sheet or current sheet, and schematically represented in Fig. 9.3,
may be considered a fundamental starting point to study resistive effects.
The magnetic field has only one component B y = 0, but its magnitude depends
on the x-coordinate. Let B y (x) be an odd function of x, (B y (0) = 0) tending to
±B0 when x → ±∞. Such a field is that generated by a homogeneous current sheet
directed along z, of infinite extension in the y-direction and piling up around x = 0.
The Lorentz force J × B pushes the plasma towards the neutral sheet from both sides.
In the absence of resistivity the action of the Lorentz force is opposed by pressure
gradients, which counteract the natural tendency of the sheet to collapses on itself.
But the presence of a finite magnetic diffusivity, no matter how small, breaks the
frozen-in condition and allows the decoupling of field lines from the motion of the
plasma, eventually giving rise to a configuration similar to that shown in Fig. 9.4.
This transition can be “pictorially” described by saying that two lines of force
are first “cut” and then “reconnected” to field lines of opposite polarity, a descrip-
Fig. 9.3 Neutral sheet: lines 1.5

of B. The solid line represents
B y (x). The scale on the 1
vertical axis refers to both the
values of the x-coordinate
0.5
and of the magnetic field
-2 -1 1 2
-0.5
-1
-1.5
Fig. 9.4 Magnetic field lines x

after reconnection
Fig. 9.5 The formation of

magnetic islands
tion at the origin of the name magnetic reconnection usually given to this class of
processes. If the magnetic reconnection of field lines occurs in several points, the
final configuration resembles the one shown in Fig. 9.5.
The field shows two distinct topologies: closed field lines, so-called magnetic
islands, and open field lines, separated by two limiting curves, known as separatrices.
The initial neutral line, i.e. the line corresponding to B y = 0, breaks into a discrete
succession of null points, said to be of the O-type when they are located within
closed field lines, and of X-type when they lie at the intersection of separatrices. To
get an intuitive understanding of why this final configuration is reached, think of the
initial current sheet as formed by a continuous collection of parallel line currents.
Since parallel currents attract each other, such a system turns out to be unstable and
the sheet evolves toward a configuration of distinct parallel currents, encircled by
magnetic field lines. It is possible to show that the energy of the final configuration
is lower than that of the initial one.
When discussing resistive processes it is necessary to distinguish two broad cate-
gories of models referring to different physical situations. In the first, named driven
reconnection reconnection is “driven” by appropriate plasma flows, while in the sec-
ond, known as spontaneous reconnection reconnection takes place in the absence
of driving plasma motions, rather, the motions are initiated self-consistently by the
reconnection process itself.
9.1 Driven Reconnection 209
9.1 Driven Reconnection
Consider the situation where non-magnetic forces, for instance pressure gradients,
induce systematic converging flows bringing together regions with different magnetic
structure. Since resistivity has practically no effect at large spatial scales, the plasma
can be considered as ideal and the field lines are dragged by the motion of the
plasma. If the magnetic field carried by the colliding fluxes has opposite directions,
a current sheet will be formed, caused by the polarity inversion of B. Inside this very
narrow sheet B ≈ 0, and resistivity plays a fundamental role allowing the breaking
and subsequent reconnection of magnetic field lines. The relaxation of the magnetic
tension in the newly bent field lines pushes the plasma out of the resistive region, thus
reducing the local total pressure, and this in turn attracts new plasma from the outside.
A stationary state will be established if the magnetic flux entering the resistive region
per unit time will be exactly balanced by the annihilation and diffusion rates of the
field. However, notice that while it is possible to destroy the magnetic field locally,
the mass of the plasma itself is conserved, so mass has to flow out of the resistive
region. The problem is well illustrated by the following model due to Parker.
Consider the two-dimensional incompressible stationary flow of a plasma
described by the system of equations:
1 B2
ρ(v · ∇)v = −∇ P + J × B = −∇ P + + (B · ∇)B (9.5)
4π 8π
1 1 η
E + v × B = J = ∇ × B, (9.6)
c σ c
where
∇ · B = 0; ∇ · v = 0.
Notice that in this chapter the fluid velocity will be indicated by v, instead of U. One
way to satisfy the condition of vanishing divergence for B and v is by taking the
following structure for the vector fields in −L ≤ x, y ≤ L:
x y
B ≡ [0, B(x), 0], v ≡ [−v0 , v0 , 0],
a a
with constants v0 and a. This flow has a stagnation point (v = 0) in [x = y = 0].
Equation (9.6) shows that the electric field is directed along z and, since in a stationary
situation ∇ × E = 0, E = E ez with E = const.
With this field structure, Eq. (9.6) reduces to
x dB
cE − v0 B = η , (9.7)
a dx
a first order linear equation for B.
Fig. 9.6 Profile of B/B0 , the B/B0

dashed lines indicate the 0.8
approximate solutions for 0.7
small and large values of x
0.6
0.5
0.4
0.3
0.2
0.1
1 2 3 4 5 6 x/ 7
The behavior of the solution may be qualitatively inferred by noticing that for
large values of x we expect that dB/dx → 0 and thus B ≈ E(c/v0 )(a/x), while for
x → 0, the second term on the lhs will be negligible and therefore B ≈ E(cx/η).
The two approximations give the same value of B for x = x0 = (aη/v0 )1/2 . The
fact that the magnetic field produced by a current flowing in a wire at a point external
to the current at a distance x from the wire’s axis is proportional to 1/x, suggests
considering x0 as an estimate of the thickness of the current sheet.
These conclusions are confirmed by the exact analytical solution of Eq. (9.7).
Using standard solution methods for first order linear differential equations one
obtains:
x/σ
−(x/σ )2 2
B = B0 e eu du,
0
where ca √
B0 = 2E ; σ = (2ηa/v0 )1/2 = 2 x0 .
v0 σ
A plot of the exact solution, together with the approximations for small ad large
values of x is shown in Fig. 9.6.
9.1.1 The Sweet-Parker Model
The Sweet-Parker model illustrates in a schematic, extremely simple, way the basic
features of stationary reconnection, without specifying the precise analytical form of
B and v. It is a two-dimensional, incompressible model similar to the one we have
just described: B and v have non vanishing components only in the (x, y) plane,
while E is a constant vector, directed along z. The equation governing the dynamics
is still Eq. (9.6). It is assumed that reconnection takes place in a narrow current sheet,
whose length and thickness are 2L and 2 , respectively. The sheet separates two
regions of oppositely directed magnetic field, as shown in Fig. 9.7.
The magnetic field vanishes in the center of the resistive layer. Outside of it, the
plasma can be considered as ideal and therefore the magnetic field, Bi is frozen
into the plasma and is convected towards the diffusive region with the speed of the
flowing plasma, vi . The assumption of stationarity is a very stringent one: it implies
a perfect agreement between the entrance speed of the plasma and the velocity of
diffusion and annihilation of the magnetic field brought into the region by the plasma
from opposite directions. This implies that the incoming speed vi cannot be fixed
at will. On the other hand the diffusion velocity is controlled by the gradient of
B ≈ [Bi − B(0)]/ = Bi / and time-independence then imposes limitations also
on the thickness of the diffusive layer, .
The model is defined by the entrance values of vi and Bi and by the corresponding
outflowing values, vo and Bo , in addition to the characteristic dimensions of the
resistive layer, L and and the value of magnetic diffusivity η. If the latter quantity
is assumed to be known, the quantities to be determined are six, while, as we shall
see, the equations connecting them are only four. We must therefore fix the values
of two of the unknown quantities: fixing Bi and L is a useful choice.
We can now estimate the values of the current density, as given by Eq. (9.6) on
the border of the resistive region (where J = 0) and in its center, where B 0),
obtaining, respectively
vi
E= Bi ,
c
and
Jc
E= ,
σ
where the index c indicates values referring to the center of the resistive layer. Since,
on the other hand, J = (c/4π )(∇ × B) → Jc (c/4π )Bi / , we may eliminate E
and Jc from the preceding equations and obtain
η
vi = . (9.8)

Fig. 9.7 Sketch of the Bi

Sweet-Parker model
vi vi
2L
v u, Bu v u, Bu
2L
vi vi
Bi
Mass conservation, requiring equality of the ingoing mass flux, 4Lρvi , with that of
the outgoing one, 4 ρvo , may be written as:
vi
Lvi = vo → =L , (9.9)
vo
where vo is the velocity of the plasma leaving the diffusive layer. On the other hand,
the magnetic flux conservation law tells us that
vi
vi Bi = vo Bo → Bo = Bi = Bi , (9.10)
vo L
where Eq. (9.9) has been used to obtain the last equality, which can be also deduced
from a dimensional analysis of ∇ · B = 0. Eqs. (9.8), (9.9) and (9.10) allow us to
express all quantities in terms of the outflow velocity vo . To determine the latter
and close the system, consider the momentum equation and estimate the order of
magnitude of the various terms entering it. For the moment, we shall assume that
the pressure is the same everywhere and thus neglect the term ∇ P in the momentum
equation. We have already seen that the current density is approximately equal to
Jc (c/4π )Bi / , which implies that the magnetic force along the resistive layer is
Jc Bo /c (1/4π )(Bi Bo / ). This force accelerates the plasma within the resistive
layer, bringing its velocity from zero, at the stagnation point, to the final value vo
when it leaves the diffusive region. The momentum Eq. (9.5) therefore tells us that
vo2 1 Bi Bo
ρ .
L 4π
Making use of the last equality in Eq. (9.10), we finally get
Bi
vo = √ ≡ cai ,
4πρ
dove cai is the Alfvén speed of the incoming flux.

In conclusion, the plasma enters the resistive layer with a velocity vi and is expelled
from it at a speed cai . The ratio of these two speeds, denoted by Ri , may be considered
as a non-dimensional measure of the reconnection rate. The index i indicates that Ri
depends only on the parameters of the ingoing flux. Ri can be conveniently expressed
in terms of the Lundquist number, making use of Eqs. (9.8) and (9.9):
vi η 1/2
Ri ≡ =( ) = S −1/2 . (9.11)
cai Lcai
Notice that the length entering the definition of S is the length of the sheet, L.
Moreover
= L S −1/2 and Bo = Bi S −1/2 .
Since S 1, vi cai , L and Bo Bi .

Notice also that the length , giving the spatial scale in the direction of the thickness
of the resistive layer, has been defined without any reference to the scale of the
magnetic field in the same direction. However, it seems natural to assume that the
two scales are correlated and this suggests identifying the scale with the thickness
of the current sheet generating the magnetic field. In fact, since the dissipation is
proportional to η J 2 , resistivity acts wherever a current flows. These considerations
will turn out to be useful in the following.
In the version of the Sweet-Parker model so far presented, it has been assumed,
without justification, that the plasma pressure was the same everywhere, mostly
to underline the fact that the outgoing plasma acceleration can be produced under
the action of the sole magnetic tension. The addition of a pressure gradient alters
the outflow speed and consequently the reconnection rate. To estimate the effects
connected with the presence of pressure gradients, we first observe that in our long
and thin resistive layer, the acceleration and the magnetic tension can be neglected
in the y-component of the momentum equation, that therefore reduces to
∂ B2
+ P = 0.
∂ y 8π
Integrating the preceding equation between 0 and we get:
Bi2
Pc = Pi + .
8π
The pressure at the center of the resistive layer is therefore larger than that at the
entrance, contrary to what assumed so far. Let’s now look at the momentum equation
in the x-direction:
∂vx ∂vx2 1 ∂P
ρvx = 1
= J By − .
∂x 2 ∂x c ∂x
Recalling that J = (c/4π )(∂ Bx /∂ y) ≈ (c/4π )(2Bi /2 ) and integrating the preced-
ing equation between 0 and L we obtain:
L
v2 1 Bi
ρ o = B y dx − (Po − Pc ).
2 4π
0
If B y is assumed to vary linearly between zero in x = 0 and Bo in x = L, i.e.

B y (x) = (Bo /L) x. an estimate of the integral is easily obtained, allowing us to
write:
Bo Bi L 2(Po − Pc ) 2 Bo L 2(Po − Pc ) 2(Po − Pc )

vu2 = − = cai − = cai
2
− ,
4πρ ρ Bi ρ ρ
where Eq. (9.10) has been used. If the central pressure, Pc , is expressed in terms of
Pi we arrive at:
2(Pi − Po )
vu2 = 2cai
2
+ . (9.12)
ρ
Defining once again the reconnection rate as vi /vo and using Eq. (9.12) we obtain:
vi cai −1/2
Ri ≡ = S .
vo vo
If the outflow velocity is less than cai , namely if the exit pressure is sufficiently
larger than the entrance one, and precisely if ΔP = Po − Pi > ρcai 2 = B 2 /4π , the
i
reconnection rate turns out to be larger than that of the Sweet-Parker model which,
as we have seen, equals S −1/2 .
9.1.2 An Outline of the Petscheck Model
The weakest point of the Sweet-Parker model (and of its generalizations as well)
is the modest value of the reconnection rate. This seems to preclude the possibility
of identifying magnetic reconnection as the mechanism capable of transforming
the magnetic energy in other forms of energy, such as thermal energy or kinetic
energy of accelerated particles. On the other hand, in the laboratory and especially in
astrophysics, phenomena are observed that necessarily imply such transformations.
A typical example is given by solar flares that exhibit a sudden increase of the
emissivity over the whole electromagnetic spectrum, localized in regions where the
magnetic field is particularly intense, the so-called active regions. This phenomenon
is associated with the presence of accelerated particles and, occasionally, the ejection
of large amounts of matter. The total energy released by a large flare is estimated to
be of the order of 1032 erg, on time scales of the order of hours. A careful analysis of
the magnetic configuration in the flare’s site before and after the flaring brings to the
conclusion that the only conceivable source of the energy released during the flare
in various forms is the magnetic energy stored in the magnetic field prior to the flare.
Observations further indicate that most of the energy is emitted during the initial
phase of the flare, on timescales of the order of minutes or less, implying a high
conversion efficiency of the magnetic energy. It is therefore natural to inquire if the
limited theoretical efficiency of the process could not be connected to the particular
geometrical configuration chosen and if other configurations exist allowing more
substantial reconnection rates.
To try to answer this question, let’s first notice that the analysis of Sweet ans Parker
is a local analysis, in the sense that it focuses only on the magnetic configuration in
the immediate surroundings of the diffusive layer, a region whose dimensions are
generally small as compared to the global size of the system. On the other hand, it
seems likely that the input values of the velocity and of the magnetic field, vi and
Bi , are actually determined by the prevailing conditions in a region whose spatial
scale is larger than the one so far analyzed. This suggests to extend our analysis to
a wider region, where the spatial scale and the characteristic values of v and B are,
respectively, L e , ve and be , the index e denoting the fact that those values are, in
fact, referring to a region external to the Sweet-Parker sheet. As a consequence, we
shall have a new value of the Lundquist number Se = L e cae /η, and, according to
the previously introduced definition, a new reconnection rate, Re = ve /cae . Then,
the problem to solve is the determination of vi in terms of ve and the evaluation of
Re in terms of Se .
A first relationship is immediately obtained from the magnetic flux conservation,
ve Be = vi Bi , that can be cast in the form:
Ri cae vi B2
= = e2 . (9.13)
Re cai ve Bi
The ratio of the spatial scales is easily found by using Eqs. (9.13), (9.11) and (9.9):
Li Si cae Si Be Si Ri 1/2 1 −3/2 −1/2

= = = = R Re (9.14)
Le cai Se Se Bi Se Re Se i
and
Li 1 −3/2 −1/2 1 −1/2 −1/2
= = Ri Ri Re = R Re . (9.15)
Le Li Le Se Se i
If, starting from a particular configuration, it is possible to determine the ratio Bi /Be ,
Eqs. (9.13), (9.14) and (9.15) give us the possibility to deduce the dimensions of the
diffusive region as functions of Re and Se . Moreover, if the model provides further
relations among the quantities defining the model, for instance between Be , Bi and
L e /L i , we may arrive at an estimate of Re in terms of Se alone, as in the case
of the Sweet-Parker model. The prototype of this kind of models is that proposed
by Petschek in 1964. The basic starting point of this model is the deep analogy
between the interaction of a supersonic gas flow with an obstacle and that of a plasma
impinging on the diffusive region with a speed larger than a characteristic velocity
of the system. In both cases shocks are formed and this allows the dissipation to take
place not only in the resistive region, but also on the shock fronts, thus increasing
the efficiency of the energy conversion.
The detailed treatment of the Petschek model falls beyond the scope of this book:
we therefore will restrict ourselves to the simple description of its main √results. The
characteristic velocity mentioned above is that of a “slow” shock, Bn / 4πρ, where
Bn is the magnetic field component normal to the shock plane. For a geometry of
the Sweet-Parker type, it can be shown that a system of four steady shocks forms,
emanating from the vertices of the resistive region. The dissipation is mostly con-
centrated on the shock fronts. For a specific heats ratio γ = 5/3, two fifths of the
incoming magnetic energy are converted into heat and three fifths into kinetic energy
of accelerated particles. The maximal reconnection rate turns out to be:
Re (max) ∝ (ln Se )−1 ,
much larger than the Sweet-Parker one for most plasmas. Moreover, since the depen-
dence of Re on Se is only logarithmic, the reconnection rate is only weakly sensitive
even to large variations of Se .
In spite of these positive features, the Petschek model does not seem to be able to
correctly describe the physics of the resistive processes, except when an anomalous
resistivity manifests itself in small localized regions, as demonstrated by numerical
simulations. Notice that the resistivity is only one of the parameters capable of
invalidating the frozen-flux theorem: in the generalized Ohm equation, besides the
collisional resistive term, at least the Hall term and the gradient of the electronic
pressure term are present. Numerical simulations have proved that the Hall effect
can indeed help establishing a reconnection regime faster than Sweet-Parker. Without
going into details, a way to understand this speed-up of the reconnection rate is to
realize that the latter depends on the fact that the evacuation of the reconnected
plasma from the Sweet-Parker layer is caused by waves generated by the relaxation
of the tension of the reconnected magnetic field lines, that proceeds at the Alfvén
speed. On the other hand, if the resistivity is sufficiently small, the current sheet can
shrink to dimensions comparable to the gyration radius of the ions, whose dynamics
will become decoupled from that of the electrons, still frozen in the magnetic field.
At those scales, as shown in Chap. 7, the Hall term is no longer negligible. As we
have seen, the Hall effect introduces, among the possible modes, a whistler wave,
obeying a quadratic dispersion relation in the wave vector k that can propagate at a
speed larger than the Alfvén velocity. The emission of whistler waves from current
sheet in Hall-MHD simulations shows that these waves may induce reconnection
rates faster than those predicted by the Sweet-Parker model.
It is natural to enquire if other ways to speed-up the reconnection process exist,
even within the framework of a classical collisional resistivity, also by considering
configurations different from the isolated and steady current sheet so far considered.
The last few years have witnessed a growing evidence of the existence of a more
complex dynamics in the observed or numerically simulated current sheets. They
tend to break and to form multiple magnetic islands, or plasmoids, which in turn
may coalesce or move away from each other, forming increasingly narrow current
sheets. This calls for the study of the possible instabilities of a current layer, the
subject of a subsequent section.
9.2 Spontaneous Reconnection
The reconnection processes so far considered assumed the presence of appropriate

material flows dragging the field towards one or more dissipative regions. The origin
of those flows was left unspecified and the possibility that the system would eventually
9.2 Spontaneous Reconnection 217
reach a steady state was implicitly assumed. At this point it seems natural to inquire
whether a resistive process could develop spontaneously without the need of imposing
a velocity field from the outside. In this case we speak of resistive instabilities: they
can have a great influence both in laboratory and astrophysical plasmas, provided
that their reconnection rates are considerably higher than the rate of pure diffusion.
The latter, as we know, develop on timescales of the order of τd = L ca /η = S τa ,
where τa = L/ca , is the time needed to cover the distance L at the Alfvén speed.
Since in practice all plasmas have S 1 (see Table 5.1), the diffusive times are
generally too long to be of interest. As we shall see, resistive instabilities develop
on timescales proportional to τd (τa /τd ) p = τd S − p , with 0 < p < 1, which might
allow for growth rates γ = τ −1 sufficiently fast to be of physical interest.
The origin of these instabilities is connected with the tendency of homogeneous
current sheets to evolve into a series of current filaments, which in turn diffuse and
release magnetic energy. This evolution may originate from several causes. Density
variations at small spatial scales may cause the formation of local structures, but
have little effect on the global field structure. These instabilities are known as the
gravitational modes and rippling modes and may act as local sources of turbulence.
The most important instability is known as the tearing instability. Like all instabilities
it is characterized by a wavelength λ (or by a wave number k = 2π/λ) and by a
growth time τ (or by a growth rate γ = 1/τ ). It is a long wavelength instability,
capable of producing important global changes in the field structure, with a growth
rate given by
τ 2/5
(k L)−2/5 .
a
τ τd
τd
The tearing instability develops even in the absence of a neutral line, since the
dynamics associated to it is substantially unaltered by the presence of a field com-
ponent parallel to the initial current sheet. As we shall see, in this case the instability
is localized close to the points where
k · B 0 = 0.
The tearing instability was first analyzed in a fundamental paper by Furth, Killeen
and Rosenbluth in 1963 [20]. We shall now illustrate the most relevant points of that
paper.
9.2.1 Tearing Mode Instability
The model to be discussed has the following characteristic features. In the unper-
turbed state the velocity vanishes and the magnetic field is given by:
B 0 = B0x (y) ex + B0z ez ,

with B0x (0) = 0, B0x (± ∞) = ± B̄. The Alfvén speed is defined by
B̄ 2
ca = .
4πρ0
Even if not strictly necessary, we shall assume for simplicity that B0z is constant and
that the plasma is incompressible, ρ = ρ0 = const. At zeroth order the induction
equation reduces to
∇ × (η∇ × B0 ) = 0.
If we further assume that the resistivity η is also constant, we see that the preceding
equation implies that
d2 B0x
= 0,
dy 2
a condition that would impose too strong a limitation on the form of the equilibrium
state. In reality, the unperturbed state will not be an equilibrium state in a strict
sense, but will evolve on time scales of the order of τd = L 2 /η, exceedingly long
with respect to the dynamical evolution time of the instability, which, as already
remarked, develops on time scales intermediate between τa and τd . In other words,
the regime under consideration is defined by
τa τ τd or γ τa 1 γ τd .
In our configuration the current density is given by:

c c
J0 = (∇ × B 0 ) = − B ez ,
4π 4π 0x
where the prime indicates the derivative with respect to y,
The starting point are the resistive MHD equations. Following the usual procedure,
we shall consider velocity and field perturbations given by:
B 1 = B1x (x, y) ex + B1y (x, y) e y ; v 1 = v1x (x, y)ex + v1y (x, y)e y .
By linearizing all the equations we get:
∇ · v1 = 0 continuity equation,
∂v 1 1
ρ0 = − ∇ P1 + J 0 × B 1 + J 1 × B 0 momentum equation,
∂t c
∂ B1
= ∇ × v 1 × B 0 + η∇ 2 B 1 induction equation,
∂t
together with the condition ∇ · B 1 = 0.
The effect of the instability will be the generation of a magnetic field with a
component B1y = 0, that was absent in the equilibrium field and of plasma flows in
the x, y plane dragging the magnetic field B 0 towards the dissipative region on both
sides of the line y = 0. It then follows that the y-component of the plasma velocity
v1y is an odd function of y, with v1y < 0 for y > 0. This observation will turn out
to be useful in the following.
By taking the curl of the momentum equation to eliminate the pressure term we
obtain the system:
1
γρ0 (∇ × v 1 ) = ∇ × J 0 × B1 + J 1 × B0 , (9.16a)
c

γ B 1 = ∇ × v 1 × B 0 + η∇ 2 B 1 , (9.16b)
in addition to the equations ∇ · B 1 = 0 and ∇ · v 1 = 0. At this point it turns out to be

convenient to perform a Fourier expansion with respect to the ignorable coordinates
x and t. Since we aim at determining the growth rate of the instability, we shall
replace −iω with γ in the temporal dependence of the Fourier expansions. A generic
first order quantity f 1 (x, y, t) will thus be represented by:
f 1 (x, y, t) = f (y)eikx+γ t .
γ > 0 corresponds to an unstable situation. We therefore have:
v1x = vx (y)eikx+γ t ; v1y = v y (y)eikx+γ t ,
and an analogous representation for B 1 . The two equation for the divergence of B 1
and v 1 that now read:
ikbx (y) + by (y) = 0 and ikvx (y) + v y (y) = 0,
allow us to eliminate vx e bx from the system.

Referring to the z-component of Eq. (9.16a), introduce in the rhs the definition of
J in terms of B. A long, although straightforward, calculation produces the following
equation
ik
γ (v − k 2 v) = B0x (b − k 2 b) − B0x

b (9.17)
4πρ0
where, to simplify the notation we have replaced v y e b y with v and b, respectively.

Finally, the y-component of Eq. (9.16b) gives us
γ b = ik B0x v + η(b − k 2 b). (9.18)

The fourth order differential system formed by Eqs. (9.17) and (9.18) for the
unknowns functions b and v is the basic system for the analysis of the tearing insta-
bility.
Let’s first observe that the effect of resistivity will be confined, as usual, to a thin
layer, of thickness , close to y = 0, where the x-component of the equilibrium
field vanishes. Notice that the z-component of the field, Boz , has disappeared from
the system and therefore does not influence the development of the instability. The
magnetic field, however, does not vanish in y = 0, where in fact B = Boz ez , but, in
that plane, k · B 0 = 0, as anticipated. Moreover, Eq. (9.18) tells us that within the
resistive layer b (≈ γ b/η) has to be very large because of the smallness of η.
Outside the resistive layer the plasma behaves as an ideal plasma and this allows
us to put η = 0 almost everywhere. At first sight then, it would seem possible to
consider the term proportional to η in Eq. (9.18) as a perturbation applied to the ideal
system. A more detailed analysis, however, shows that actually we are dealing with a
singular perturbation, meaning that when we move from a situation in which η = 0
to one having η = 0 the order of the differential system changes from second to
fourth order and this gives rise to difficulties with the number of initial conditions to
be imposed on the system. The approach to be followed with this kind of problems
is based on boundary layer theory, which basically consists in the separation of the
region of integration into two distinct domains: that external to the boundary layer,
where we put η = 0 and the internal one where resistivity has to be taken into account.
Let a be the spatial scale of the external region, namely the macroscopic scale for
the variation of the magnetic field in the direction transverse to the extension of the
resistive layer, which is connected to the tickness of the current layer generating B 0 .
The quantities τa and τd will now be defined as τa = a/ca and τd = a 2 /η. The
arguments presented in the section on the Sweet-Parker model allow us to identify
a with the quantity of that section.
We expect a to be much larger than the thickness of the boundary layer, δ, an
unknown quantity whose value will be determined together with the actual solution
of the problem. Making use of the boundary layer technique, we shall first solve
the external problem and evaluate the unknown functions and their derivatives at the
border of the boundary layer. Next, we shall solve the internal problem and connect
smoothly the internal and external solutions at the border of the boundary layer. As
we shall see, the conditions that allow such a smooth connection determine both the
thickness of the resistive layer δ and the value of the growth rate γ . This technique
turns out to be very efficient since in the external region only the solution of a second
order equation is required, while in the internal one the smallness of the spatial
scale usually allows important simplifications to be introduced. Furthermore, this
procedure is essential whenever the problem has to be solved numerically. In fact, it
would be impossible to solve the resistive problem everywhere, because of the huge
value of the ratio a/δ which would force the use of an extremely fine integration
grid with a consequent prohibitive number of integration points. On the contrary, the
boundary layer technique greatly reduces the number of integration points, both in
the external region, where the spatial scale is large, and in the internal one, of modest
extension.
The actual solution of the system of Eqs. (9.17) and (9.18) requires the knowledge
of a specific profile for B0x (y) and presents substantial technical difficulties even for
relatively simple profiles, such as the so-called Harris profile: B0x = B̄ tanh(y/a).
However, it is possible to find the correct expressions for δ and γ by using simple
dimensional considerations and heuristic arguments, as we shall now do,
The diffusive layer extends in y from −δ/2 to δ/2 and in its interior the resistive
term in Eq. (9.18) will dominate over the convective one, while the latter will be the
dominant one outside. The point y = δ/2 can therefore be taken as the point where
the two terms have the same order of magnitude. Since, as already anticipated, the
regime of interest is one of “long”wavelengths, λ a δ, in the internal region
the term b b/δ 2 will be much larger than k 2 b b/λ2 , which, in turn, will be
much larger than B0x b since the equilibrium field varies slowly on the resistive scale.
In a similar way, the term v1y will be dominant with respect to k 2 v. Equation (9.17)
thus reduces to:

ik B0x
γ v = b for − δ/2 < y < δ/2 (9.19)
4πρ0
and Eq. (9.18) becomes simply:

γ b = ik B0x v + ηb (9.20)
Equations (9.19) and (9.20) are the equations for the internal region.
To determine the external solution, let η → 0 in Eq. (9.18) and eliminate v between
this equation and Eq. (9.17). The resulting equation reads

B0x B̄ 2 B B
b − k 2 b − b − (γ τa )2 (ka)−2 0x
b − 0x b = 0.
B0x B0x B0x B0x
Putting
γ τd
γ τa = ,
S
and keeping in mind that the growth rate of pour instability γ = 1/τ must have a
finite value also for S 1, we see that the terms proportional to γ τa can be neglected
and therefore the equation for the magnetic field in the ideal case is:

B0x
b − k 2 b − b = 0. (9.21)
B0x
If the resistivity could be neglected everywhere, or equivalently, if Eq. (9.21)

were valid for all values of y, we would get a solution continuous in y = 0, but
with a discontinuous derivative in that point. To illustrate this general property of
= 0, which, as already noted,
the solutions of Eq. (9.21), let’s consider the case B0x
would be a rigorous equilibrium solution for a constant η. The ideal equation then
reduces to
b − k 2 b = 0,
whose solution, with the appropriate convergence properties at ±∞, is:
b(y) = b(0)e−|ky| ,
exhibits a discontinuous derivative in y = 0.

The jump of the solution going from the external to the internal region is normally
expressed in terms of the non dimensional parameter
b (δ/2) − b (−δ/2)
Δ = a .
b(δ/2)
The assumed continuity of both b(y) and its derivative in y = δ/2, implies that the
value of Δ can be computed using only the external solution. In general Δ must
be computed numerically, but can be given an analytical expression for particular
choices of the profile of B0x . An important case is that of the already mentioned
Harris profile, B0x = B̄ tanh(y/a) for which it is possible to show that
1 − (ka)2
Δ = 2 .
ka
To procedure to obtain solutions continuous together with their first derivative is the
following. We first put η = 0 only in the region |y| > δ/2. In the internal region,
where η = 0, the second derivative of b must be sufficiently large to eliminate the
discontinuity of the first derivative. In that region we may write:
b (δ/2) − b (−δ/2) b (δ/2) − b (−δ/2) b(δ/2) Δ b(δ/2)

b (y) = lim ≈ = ,
δ→0 δ b(δ/2) δ aδ
(9.22)
An approximate estimate of Eq. (9.19) at y = δ/2 is obtained making use of
Eq. (9.22), namely
v(δ/2)
v (δ/2) − .
δ2
The presence of the minus sign in the preceding expression is due to the fact that v is
an odd function of y while v (δ/2) > 0, as can be realized considering, for instance,
a function of the type sin(y). We finally get:
v ik B b
Δ ,
0x
−γ =
δ2 4πρ0 aδ
where all the functions are evaluated in y = δ/2.

On the other hand, Eq. (9.20), always for the external solution, gives
γ b ik B0x v,
and, combining together the last two equations,
(k B0x )2 δ
γ2 = Δ,
4πρ0 a
(0) = (δ/2)B (0) we finally arrive at:
Since B0x (δ/2) B0x (0) + (δ/2)B0x 0x
(0))2 3
1 (k B0x δ (ka) 2 δ 3
γ2 Δ =A Δ , (9.23)
4 4πρ0 a τa a
with
a B (0) 2
0x
A=
2 B̄
This equation shows that in order to have a real γ (and thus an instability) we must
have Δ > 0.
Let’s now turn to the internal equation, which, using Eq. (9.22), can be written as
Δ
γ b(δ/2) ηb η b(δ/2),
aδ
or
Δ δ Δ
γ η implying .
aδ a γ τd
By eliminating δ/a between the above equation and Eq. (9.23) we obtain an estimate
of the growth rate:
γ τd A1/5 (ka)2/5 S 2/5 (Δ )4/5 . (9.24)
Alternatively, if we eliminate γ between the same equations, we obtain an estimate

of the thickness of the resistive layer:
δ
A−1/5 (ka)−2/5 S −2/5 (Δ )1/5 . (9.25)
a
If the values of A and Δ just found are used for the Harris current sheet, B0x =
tanh(y/a), and the diffusive time, instead of the Alfvén time, is used to normalize
the times, we arrive at the dispersion relation
γ τa 22/5 (ka)−2/5 S −3/5 (1 − k 2 a 2 )4/5 . (9.26)
A more detailed analysis shows that the approximations used limit the validity
of our result to wavelengths satisfying the condition kaS 1/4 1. The maximum
growth rate, obtained for kaS 1/4 1, scales as γ τa S −1/2 . Notice that such a
growth rate, associated with a timescale which is orders of magnitude smaller than
the diffusive timescale for large Lundquist numbers, can still be extremely slow.
Consider the solar flare phenomenon, which has timescales of an hour or less, with
a coronal Lundquist number of S ≥ 1013 : for a typical Alfvén crossing time say of
10 s, the timescale associated with the fastest growing tearing mode becomes over
a year! A possible way out of this timescale problem may come from considering
current sheets with a thickness which is much smaller than the macroscopic scale,
as, for example, was found in the Sweet Parker current sheet.
9.2.2 The Plasmoid Instability of Thin Current Sheets
The analysis carried out for the tearing mode illustrated that current layers are unsta-
ble under quite general conditions, over timescales scaling as τ/τa ∼ S 1/2 . This
timescale seems to be similar to that associated with the reconnection time for the
Sweet-Parker current sheet discussed previously (Eq. (9.11)), and this might lead to
believe there is a connection between the two results. Careful consideration however
shows this coincidence to be accidental: the internal, dissipation layer of the tearing
mode was given by Eq. (9.25), which together with the relation ka S −1/4 , leads
to δ/a ∼ S −3/10 , or identifying the current shear scale a with a macroscopic scale
L, leads to a singular current layer which is thicker than the Sweet-Parker sheet
a/L ∼ S −1/2 . This suggests that Sweet-Parker’s stationary reconnection solution
might also be unstable to tearing modes, which might have even faster growth rates
than that of the classical tearing mode.
To see if this might happen, one must start by considering the same normalizations
for both the tearing mode and the Sweet-Parker model, specifically as concerns the
Lundquist number, which was based on the current sheet length L for the Sweet-
Parker case, but normalized to a current sheet thickness a for the tearing mode. As a
consequence, the definitions of τa and τd as well as S = τd /τa were different. Using
an asterisk to denote the values as defined in the section on the tearing mode, and
leaving the Sweet-Parker values as they are, we find that:
a aca 1/2 L Lca 1/2 a 3/2

τ ∗ = τa∗ (S ∗ )1/2 = = .
ca η ca η L
Now remembering that the Sweet-Parker aspect ratio is a/L ∼ S −1/2 , the fol-
lowing hypothetical result is found for a tearing mode on the Sweet-Parker current
sheet:
τ ∗ ∼ τa S 1/2 S −3/4 = τ S −1/4 . (9.27)
This means that if the Sweet-Parker current sheet were found to be unstable, the
instability would grow on a timescale which would tend to vanish as the Lundquist
number S → ∞. Before discussing this result we analyze the tearing instability for
the Sweet-Parker current sheet following the reasoning of Loureiro et al. (2007) [22].
Again, we must linearize the equation of motion and the induction equation,
but this time including the background flow in the equilibrium of the Sweet-Parker
model and therefore generalizing the Eqs. (9.17) and (9.18). In this case, the linearized
equations take the form
∂v 1 1
ρ0 (∇ × [ + v 1 · ∇v 0 + v 0 · ∇v 1 ] = ∇ × J 0 × B 1 + J 1 × B 0 , (9.28a)
∂t c
∂ B1
= ∇ × v 1 × B 0 + ∇ × v 0 × B 1 + η∇ 2 B 1 , (9.28b)
∂t
where v 0 is the background flow, defined by v 0 ≡ [v0 x/l, −v0 y/l, 0]. Proceeding
as in the static case, taking the z component of Eqs. (9.28a) and (9.28b), care must
be taken to include the spatial gradients of v 0 . Differentiating the equation of motion
along the x direction, and using incompressibility, one finds
∂ 2 v0 2 B0x ∂ 2 1 ∂b1y ∂ 2 B0x

ρ0 + v 0 · ∇ ∇⊥ v1y + ρ0 ∇⊥ v1y = ∇⊥ b1y − (9.29a)
∂t l 4π ∂ x 4π ∂ x ∂ y 2
∂b1y ∂b1y v0
+ v 0 · ∇b1y = B0x − b1y + η∇ 2 b1y , (9.29b)
∂t ∂y l
where ∇⊥ 2 = ∂ 2 /∂ x 2 + ∂ 2 /∂ y 2 and nothing has yet been stated about the time-
dependence of perturbations. The equations above depend explicitly on both the

coordinates y, as before, but also x, because of the background flow. It is still pos-
sible, however, to eliminate this dependence completely by still taking the ansatz
v1y = v(y, t)eikx (and the same for B1y ) but allowing for a time dependence in the
wavevector k = k(t). This is because the background acceleration along the current
sheet has the effect of stretching the wavelength in the x direction, so that one expects
k(t) = k0 exp(−v0 t/a). In this way the equations for v, b reduce, as previously, to
having only an explicit y dependence. Now if we search for eigenmodes with growth
rates γ , i.e. solutions of the form v(y, t) = v(y)eγ t , with γ v0 /a, terms pro-
portional to v0 /a in the linearized equations may be neglected, and Eqs. (9.17) and
(9.18) are recovered.
In this way, the instability of the Sweet-Parker current sheet has been shown
more rigorously, since the scale of the equilibrium is precisely the thickness of the
Sweet-Parker current sheet x0 = (aη/v0 )1/2 , so, as long as the Δ is positive for
this equilbrium, the sheet is unstable with the fast timescale given by (9.27). We will
not calculate Δ explicitly for the Sweet-Parker equilibrium, but the result indeed
shows that the Sweet-Parker current sheet is unstable to this super-tearing mode or
Fig. 9.8 Island formation in a Sweet-Parker current sheet with Lundquist S 105 after about 2
Alfvén times showing multiple island formation. Top, current intensity (peaking at x-points and
separatrices). Bottom, density and magnetic field lines. Units arbitrary, color contours increasing
from blue to red
plasmoid instability, so called for its fast evolution leading to the formation of many
islands, as shown in Fig. 9.8.
From the linear point of view, the existence of instabilities with a growth rate
which increases with increasing Lundquist number is problematical. In ideal MHD,
magnetic reconnection is prohibited, and the finding of an “infinitely unstable” mode
in the limit of infinite Lundquist number points to a strong singularity in the behaviour
of the MHD equations. Diffusive terms change the order of the equations, even in
the linearized equations, and though it would not be too surprising to find a different
physical behaviour between systems described by ideal MHD and those described
by the resistive equations with S → ∞, the presence in the latter case of infinitely
fast growing instabilities has only one possible interpretation: the un-realizability of
the corresponding equilibrium configurations (one could also argue that an infinite
growth rate is non-causal and unphysical). Pucci and Velli (2013) [23] have shown
that in the asymptotic S → ∞ regime an effectively “ideal” tearing mode survives,
in the sense that its growth rate does not depend on the Lundquist number.
The instability also defines a maximum critical aspect ratio (or ratio between cur-
rent sheet width and length) L/a ∼ S 1/3 , above which any current configuration
must be intrinsically unstable and turbulent, and for which a laminar configuration
is impossible to attain. In other words, if the MHD configuration leads to thinning
sheets, an instability with an ideal growth rate arises before the current sheet col-
lapses to the Sweet-Parker aspect ratio. This type of instability naturally leas to the
formation of many islands and resulting nonlinear dynamics, which are most appro-
priately described within a theory of magnetohydrodynamic turbulence, which we
will discuss in the following chapter.
9.1. Compute the values of density, temperature, and magnetic field for the plasmas of
Table 5.1. Then compute the values of the resistivity and verify that the corresponding
Lundquist numbers are those given in the Table.
9.2. Show that for the Harris current sheet B0x = B̄ tanh(y/a)
1 − (ka)2
Δ = 2 .
ka
Hint
Show that Eq. (9.21) can be written as
2
b − k 2 b + sech2 (y/a) b = 0.
a2
Look for solutions of the above equation of the form:
tanh(y/a)
b = b0 e∓ky 1 ± ,
ka
for y > 0 and y < 0, respectively.
Show that the solutions present a discontinuity in y = 0
b (0+ ) − b (0− )
Δ =
b(0)
that has the required value.

Chapter 10
MHD Turbulence
Abstract Turbulent plasmas are extremely common in the astrophysical context.

The phenomenological theory of isotropic and homogeneous hydrodynamical tur-
bulence is presented and the fundamental results of Kolmogorov deduced. Their
generalization to plasmas in the MHD regime is then discussed. The relevance of
turbulence theory for the problem of solar coronal heating is the subject of the last
section.
As illustrated in the preceding Chapters the dissipative coefficients, both resistive

and viscous, are as a rule extremely small in astrophysical plasmas. Equivalently, the
Reynolds and Lundquist numbers are extremely large. We know from hydrodynamics
that in these circumstances fluids enter a regime where vortices of different sizes and
lasting over different times, smaller vortices typically lasting less than larger ones,
appear in the flow, and generally chaotic, unpredictable motions result. When the
chaotic motions, associated with the vortices, are present over a wide range of spatial
scales we speak of a regime of developed turbulence. The distinctive feature of this
regime is that energy appears to be transported from the largest spatial scales to those
sufficiently small to allow dissipation to set in. Indeed, the energy dissipation rate
as measured experimentally in many different fluids of varying Reynolds numbers
appears to be independent from the viscosity of the fluid, i.e., it reaches a finite value
in the limit of zero viscosity.
The appearance of stronger and stronger gradients in the flow, already remarked
in Chap. 8, is clearly borne out from inspection of the equation of motion of a viscous
fluid, described by the Navier-Stokes equation,
∂U 1
+ (U · ∇)U = − ∇ P + ν∇ 2 U, (10.1)
∂t ρ
where incompressibility has been assumed, ∇ · U = 0, and ν is the kinematic

viscosity. The importance of the nonlinear convective term, (U · ∇)U, increases
when the spatial scale l decreases, (but not faster than 1/l), and the nonlinearity of
the term leads to the formation of harmonics in Fourier space and sharper gradients.
For compressible motions, the convective derivative acting on sound waves is the
cause of the formation of increasingly steep wave fronts. In the incompressible case,

230 10 MHD Turbulence
the same term remains responsible for the generation of harmonics and reduced
spatial scales. However, in this case vortex structures are formed, rather than steep
wave fronts. On the other hand, the importance of the viscous term, ν∇ 2 U, increases
as 1/l 2 , and therefore, at sufficiently small scales, the latter always dominates. Let L
be the spatial scale where energy is injected into the system, and ld the spatial scale
where the viscous term (or, in general, the dissipative terms) is no longer negligible.
If the scales L and ld are separated by many orders of magnitude, L ld , as usually
happens in astrophysical plasmas, we speak of completely developed turbulence.
In natural plasmas energy sources for turbulence are present either in connection
with global motions or as a consequence of instabilities arising in non-potential
magnetic configurations. If the spatial and temporal scales involved are such that
the plasma can be described by the MHD equations, we speak of MHD turbulence.
MHD turbulence plays a fundamental role in the dynamics of astrophysical plasmas
in very different environments and over a very wide range of scales, from kilometers
to kiloparsecs and beyond. The understanding of turbulence generated by nonlinear
dynamics of plasmas and magnetic fields is essential to attack problems such as the
physics of the outer layers of convective stars, like the Sun, the heating of the solar
corona and the acceleration of the solar wind, or the dynamics of the molecular clouds
and of the star-forming regions, where turbulence generates the viscosity needed
to remove angular momentum. The interaction of turbulence with higher energy
particles is a determining factor in understanding the properties of cosmic rays.
In this chapter we describe some of the main results of the modern theory of
MHD turbulence, taking a phenomenological approach, based on invariance princi-
ples and on a dimensional analysis of the equations. For simplicity, we shall start from
the fundamental results established by Kolmogorov for homogeneous and isotropic
hydrodynamical turbulence. These results remain the basis for the further general-
izations required to include the effects arising from the presence of magnetic fields.
Even if hydrodynamics and magnetohydrodynamics share several common fea-
tures, the development of turbulence in a magnetized plasma is considerably different
from the purely hydrodynamical case, because of the intrinsic anisotropy created by
the magnetic field, which allows the existence of Alfvén waves propagating along
field lines. We shall restrict our study to the case of an incompressible plasma, even
though strictly incompressible plasmas don’t really exist, though they remain a good
approximation as long as the fluctuations are subsonic or subalfvénic, i.e. are suf-
ficiently smaller than the characteristic speeds of propagation in the plasma. Our
analysis must therefore be understood as a model for the evolution of the fluctu-
ations at small Mach numbers, as remarked in Chap. 7. We shall then discuss the
specific case of the solar corona and of the solar wind. In the latter environment,
MHD turbulence has been observed and studied in great detail over the last fifty
years, thanks to the many spacecraft launched into interplanetary space.
10.1 Homogeneous and Isotropic Hydrodynamical Turbulence 231
10.1 Homogeneous and Isotropic Hydrodynamical Turbulence
The observations of the spectra of the fluctuations of velocity, magnetic field and
density measured indirectly in distant astrophysical environments, or directly in situ
in the solar wind, have produced a general consensus on the nature of the processes
operating in those contexts. The emerging picture is one where energy is injected at
macroscopic scales and transferred to structures of increasingly reduced scales, until
these are so small that the dissipative terms become comparable to the other terms,
no matter how small the viscous and resistive coefficients might be.
The source of free energy at the injection scale may be connected, for instance, to
the kinetic energy associated with the velocity difference between two streams, as in
the Kelvin-Helmholtz instability, or from the difference of thermal energy between
two fluid layers, as in convection, or from the free energy stored in a non-potential
magnetic field, as in the case of the tearing instability. As a consequence of the non-
linear terms of the equations, the energy is re-distributed over different scales, in a
way that does not correspond to an equipartition, but rather to a power-law energy
distribution over spatial scales. The observed spectral index of this distribution is
close to 1.67. According to the theory originally proposed by Kolmogorov in 1941,
there are two properties that characterize a completely developed turbulence. First:
the dissipation within the fluid is independent of the exact value of the viscosity and
does not tend to zero when the viscosity vanishes. Second: the energy transfer from
the injection scales to the dissipative ones does not take place directly, but through
a nonlinear cascade which transfers the energy from larger to smaller eddies. This
transfer is due to the interactions between those modes that, at each step, have compa-
rable scales. Already at scales slightly smaller than the injection ones, the nonlinear
interactions that cause the cascade lose memory of the macroscopic anisotropies at
the so-called injection scales, so that the turbulence acquires a homogeneous and
isotropic character, at least in a statistical sense.
To put these considerations in a more quantitative form, we start again from the
Navier-Stokes Eq. (10.1). The nonlinear term appears to be due to the convective
derivative, however the pressure here is an unknown, and its role is to ensure incom-
pressibility of the flow. One way to remove it, assuming a constant density, is to take
the curl of that equation. Defining the vorticity ω = ∇ × U, we get
∂ω
= ∇ × (U × ω) + ν∇ 2 ω. (10.2)
∂t
Notice that Eq. (10.2) is apparently identical to Eq. (5.25), with the vorticity ω replac-
ing B and the viscosity replacing the magnetic diffusivity η. However, the analogy
is not complete, since Eq. (10.2) depends only on one vector field, i.e. U, ω being
defined in terms of U, while two independent vector fields, U and B, enter Eq. (5.25).
The Reynolds number, that measures the ratio between the convective and the
viscous terms on the rhs of Eq. (10.2) at the macroscopic scales or injection scales,
is defined by R = U L/ν, where U e L are a typical value of the velocity and its
scale of variation, respectively. It is therefore clear that if the viscous term dominates
R 1 and the fluid motions will be laminar, while in the opposite limit, R 1, the
motions will be turbulent with extremely irregular variation of the velocity at every
point. We may thus identify U with the average, over very long times, of the effective
velocity at each point of the fluid. In other words, the velocity of each fluid particle
can be written as U = U 0 + w, where w refers to the fluctuating part of the velocity,
while |U 0 | U. In an intuitive way, the flow can be described as the superposition of
turbulent eddies of different dimension, or scale, the latter being defined as the order
of magnitude of the distance that separates two points with substantially different
speeds, ΔU |w|. If is the scale just defined and w the corresponding velocity
variation, we may associate a Reynolds number R = w /ν with each scale.
In spite of the complexity and chaotic nature of the turbulent motions, a large
amount of information concerning their characteristics can be deduced in a sim-
ple way making use of dimensional considerations as shown by Kolmogorov and
Obukhov. When a turbulent flow sets in, vortices of large dimensions first appear,
then giving rise to vortices of increasingly smaller size. In the large vortices, can
be identified with L and, on this scale, ΔU U and consequently R L R. For
these vortices viscosity is not important and no dissipation of energy takes place.
The subsequent decrease of the size of the vortices, reduces down to values where
dissipation sets in. This happens at the scale d , the dissipative scale, corresponding
to a local Reynolds number Rd 1. The energy, injected in kinetic form at the scale
L, is transferred unaltered to smaller scales until the dissipative scale is reached,
where it is converted into heat.
The fact that the energy transfer takes place in a conservative way follows directly
from the properties of the nonlinear terms of the Navier-Stokes equation, namely the
convective derivative and the pressure gradient. As mentioned above, the role of
the pressure term is to ensure incompressibility of the fluid, ∇ · U = 0. It can be
calculated by taking the divergence of the Navier-Stokes equation and imposing
incompressibility, which leads to the Poisson-type equation for the pressure:
P
∇2 = −∇ · (U · ∇U). (10.3)
ρ
Therefore, the pressure field necessary to maintain the fluid’s incompressibility is also
a nonlinear function of the convective derivative. Introducing the Fourier transform
of the velocity field in three spatial dimensions as:

U(x, t) = d 3 k U k (t)ei k·x , (10.4)
and applying a Fourier transform to both the Navier-Stokes equation and the pressure
Eq. (10.3), the pressure may then be eliminated using the latter, to obtain

∂Uik
= Ailm Ulp Umq dq − νk 2 Uik , (10.5)
∂t
p+q=k
where the tensor Ailm is given by

k i kl
Ailm = −ikm δil − 2 .
k
The fluid energy is conserved by nonlinear interactions among modes of different

wavenumbers in the following sense. For every set of vectors k, p, q satisfying
p + q = k, the nonlinear term associated with the interactions between them is such
that the time derivatives of the energies in each mode, defined as E k = |U k |2 /2, when
calculated from the interaction with the other two modes, obeys the relationship:
∂ E k /∂t + ∂ E p /∂t + ∂ E q /∂t = 0.
In other word, the nonlinear interactions conserve energy between each triad of
interacting modes. Because the full nonlinear term is nothing but a sum over triads,
the nonlinear terms conserve energy, but redistribute it over different wavenumbers.
According to Kolmogorov, therefore, energy flows in wavenumber space until it
reaches the smallest scales where it can be dissipated. Since viscous dissipation only
takes place at the scale d , all quantities characterizing the fluid at scales > d
don’t depend on viscosity. This is also true of the energy per unit mass transferred
from one scale to the next per unit time that we shall indicate by ε. The dimensions of
ε are evidently [l 2 t −3 ]. In a steady situation we can estimate ε both at the injection
scale and at the dissipation scale. At injection ε can depend only on the density ρ, on
the macroscopic velocity U0 U and on the scale L. A simple dimensional estimate
shows that ε can be written as h
U3
ε . (10.6)
L
Similarly, at the dissipation scale we may write:
wd3
ε ,
d
from which follows:

U 3 L
. (10.7)
wd d
Equation (10.6) allows us to define the concept of turbulent viscosity. Recalling

the expression for energy dissipation in a viscous fluid, Eq. (8.37), we see that the
energy dissipation rate per unit mass, our ε, can be written as

1 dE 1 ν̃ ∂Ui ∂U j 2
ε= =− + dV,
M dt 2V ∂x j ∂ xi
where ν̃ is the viscosity relevant to the case under study. A dimensional estimate of
the above expression shows that ε ν̃ (U/L)2 . A comparison with Eq. (10.6) then
shows that
ν̃ U L = ν R. (10.8)
The ratio of the turbulent viscosity to the kinematic viscosity is thus equal to
R 1: the turbulent viscosity exceeds the kinematical one by a very large factor.
The condition Rd 1 can now be used to obtain an estimate of the ratio between
the dissipative scale d and the injection scale L. By combining this condition with
Eq. (10.7) we get:
ν ν L 1/3 4/3 ν 1/3 L 4/3

d = = =⇒ d = L = ,
wd U d U R
and finally
d
= R −3/4 ,
L
which shows that the scale d is much smaller than the injection scale L. The inter-
mediate scales, L d , are associated with the so-called inertial range where
a straightforward transfer of energy from one scale to the next takes place. In this
range it is easy to evaluate the order of magnitude of the turbulent velocity associated
with a particular scale . In fact, if for each scale , we write ε ∝ ν̃(w /)2 and use
Eq. (10.8) we obtain
ε ∝ w3 / =⇒ w ∝ (ε)1/3 ,
known as the Kolmogorov-Obukhov law. The relationships just found can be used to
define the timescale for the transfer of energy from one scale to the next, τ ∗ , that in
this case coincides with the so-called turnover time: τ ∗ = τnl ≡ /w .
The preceding results can be cast in a different form by introducing the spectral
density (per unit mass), E(k), where k is the wavenumber associated with the scale
, k 1/. Let’s write now the total energy per unit mass as

E/m ∝ E(k)dk,
from which we deduce that the dimensions of E(k) are [l 3 t −2 ]. On the other hand,
we know that in the inertial range E(k) can depend only on ε and k. Therefore
E(k) ∝ ε2/3 k −5/3 , (10.9)
which is the celebrated Kolmogorov spectrum.

Actually, Kolmogorov proved a more rigorous result, the so-called four-fifths law.
To describe it, it is first necessary define the velocity increments calculated between
two points of the fluid separated by a distance given by the vector r:
δU(r) = U(x + r) − U(x). (10.10)
The concepts of isotropy and homogeneity are summarized by the statistical prop-
erties of powers of velocity increments δU (called the structure functions of the
relevant order n)

Sin (r ) = δU in (10.11)
The structure functions depend, for an isotropic and homogeneous turbulence, only
on the distance between the two points, r , and not on the origin chosen for the
increments x (homogeneity), nor on the direction of the vector r (isotropy). Statistical
averages performed over the components of δU can consist either in time averages
(in a steady state case) or spatial averages (in a region where the turbulence can
be considered homogeneous, in which case frequency structure functions must not
depend on the origin in time). Structure functions are the quantities that are generally
used in the experimental study of the properties of turbulence, both in terrestrial labs
and when in situ measurements are performed in space.
Kolmogorov’s law is connected to the dependence of the third-order structure
function on scale and establishes that, under the assumptions of homogeneity,
isotropy and vanishing viscosity, the longitudinal components (i.e. along r) of the
third order structure function is proportional to both the energy dissipated in the fluid
and the scale l
4
3
Slong (l) = − l. (10.12)
5
A more detailed discussion of hydrodynamic turbulence can be found in the
textbooks by Landau and Lifschitz [7], Frisch [9].
10.2 Magnetohydrodynamic Turbulence
Consider now a plasma described by the MHD equations, immersed in a uniform

magnetic field B 0 , and suppose that velocity and magnetic field fluctuations are
present in the form of Alfvén waves of arbitrary amplitude. The first thing to remem-
ber, as we have seen in the chapter on waves, is that Alfvén waves propagating in
one direction along the field are an exact, nonlinear
√solution of the MHD equations.
In terms of the Elsasser variables, z ± = u ± b/ 4πρ the incompressible MHD
equations may be written:
∂z± 1
∓ ca · ∇z± = − ∇ p T − z∓ · ∇z± . (10.13)

∂t ρ
√
where ca = B 0 / ρ is the Alfvén velocity and p T = p + |B 0 + b|2 /8π is the total
pressure, this time satisfying the Poisson-type equation
pT
∇2 = −∇ · (z∓ · ∇z± ) (10.14)
ρ
which guarantees the incompressibility of z± , again found by taking the divergence

of Eq. (10.13). In the above we have always assumed a constant density.
Taking the Fourier transform in space, Eq. (10.5) is generalized to

∂zk±
∓ ik·ca zk± = −iP(k) zp∓ zq± d3 q (10.15)
∂t
p+q=k

k i kl
P(k)ilm = km δil − 2 . (10.16)
k
The difference between hydrodynamics and magnetohydrodynamics now appears

clearly, in the sense that in MHD the nonlinear interactions couple two different fields
z± , which are nothing but Alfvén waves propagating in opposite directions along the
mean magnetic field B 0 . If we imagine ensembles of localized wave-packets trav-
eling along the field in opposite directions, it is clear that the nonlinear interactions
among them will be limited, in time, to the effective duration over which they cross
paths, and this, in general, will slow down the nonlinear cascade. A different way of
understanding this slowing down effect consists in moving to an “interaction” repre-
sentation of the fields, by incorporating the Alfvén wave-like temporal oscillations
into the definition of the fields themselves:
zk± (t) ≡ zk± (t) exp(∓ik·ca t),
so that Eq. (10.15) becomes

∂zk±
= −iP(k) zp∓ zq± exp ∓2i p·ca t d3 q (10.17)

∂t
p+q=k
In the hydrodynamic case both z fields simplify into the simple velocity fluctuations
u, and the kernel in the convolution integral does not oscillate in time, (in other words,
the hydrodynamic interactions of vortices in homogeneous turbulence is always
resonant). In MHD however the kernel oscillates in time, decreasing the intensity of
nonlinear interactions, and there are no resonant interactions between modes unless
the vectors p and ca are orthogonal. This introduces a directional anisotropy in
the nonlinear cascade due to the presence of the mean field which has important
implications when generalizing Kolmogorov’s law to MHD.
10.2 Magnetohydrodynamic Turbulence 237
This MHD effect which slows down the cascade was introduced, even in the
absence of a mean field B0 , by Iroshnikov and Kraichnan independently in the early
sixties. The dimensional argument may be made as follows: at any given scale in the
inertial range, the Elsasser field on scales larger than the one in consideration may
be seen as a mean stochastic field cas , upon which wave packets at the smaller scale
under consideration propagate while interacting nonlinearly to give rise to packets
on smaller scales still, even though there still is a global isotropy in a statistical sense.
This means that the cascade is still delayed by this Alfvén effect. A new time-scale
τa l/cas therefore appears and the nonlinear transfer time τ ∗ will no longer be
equal to the turnover time

τnl l/u(l) l 4πρ/b(l) l/z + (l) l/z − (l),
where we have generalized from hydrodynamics to MHD by introducing the rms

values of magnetic field fluctuatins, b(l), and supposed there is equipartition between
velocity fluctuations, magnetic field fluctuations in velocity units and Elsasser field
fluctuations. The collisions between Alfvén wave-packets on a scale l is limited by the
crossing time τa , and supposing that the collisions between individual wave-packets
are independent stochastic events, the value of the effective time τ ∗ increases by a
factor
τnl
τ ∗ τnl .
τa
Substituting this value in the equation for the energy flux Equation, we find
1 u(l)2 1 u(l)4
(l) ∼ ∗
(10.18)
2 τ 2 lcas
from which the energy on scale l may be found as E(l) (cas l)1/2 , or in terms of
energy epctra E k (cas )1/2 k −3/2 known as the Iroshnikov-Kraichnan spectrum of
MHD turbulence.
In the above we have assumed that the amplitudes of the fields z ± were about the
same. However, from the MHD Eq. (10.13) it is easily seen that the turnover time
for the fields z ± depend on the amplitude of the opposite field z ∓ :
τnl± = l/z∓ . (10.19)
In the case of magnetohydrodynamics it may be shown that the global integral

invariants are different from the ones in hydrodynamics: in MHD nonlinear interac-
tions separately conserve the energies per unit mass of the individual Elsasser fields
E ± = 1/2|z± |2 (corresponding physically to the conservation of total energy in the
fluctuations as well as the mixed or cross-helicity Hm = <u · b>). This implies
that separate cascades with independent energy fluxes + and − are possible.
However, repeating the phenomenological arguments from Iroshnikov and Kraich-
nan separately on the fields z + and z − we find the extremely interesting result that
the two fluxes + and − are identical:
+ − 3 + − s
k = k = k E k E k /ca . (10.20)
Note that for equal amplitudes in the fields we are once again led to the Iroshnikov-
Kraichnan spectrum. If however the large scale amplitudes of the two fields are
initially different, Eq. (10.20) leads to a new conclusion: because the energy fluxes,
and therefore the dissipation rates, of the two fields are identical, for long times the
field with initially lower energies will tend to be depleted with respect to the dominant
one. This will eventually quench the turbulence, slowing down nonlinear interactions
and leaving a nonlinear Alfvén wave, with a well-defined power spectrum, as an exact
solution of the MHD equations, which in the presence of an even small mean field
will end up propagating along the field in one direction. This evolution towards a
configuration in which only one Elsasser field dominates is called dynamic alignment,
since it involves an increase in the correlation between velocity and magnetic field
fluctuations with time (corresponding to the decrease towards zero of one of the
Elsasser fields). Incompressible MHD simulations have indeed shown that dynamic
alignment occurs, while observations of the evolution of Alfvénic turbulence in space,
in the solar wind, show that for the case of turbulence propagating away from the sun
the opposite happens: there is a tendency for the development of the Elsasser mode
corresponding to Alfvén waves propagating towards the sun, with respect to those
propagating away, as the distance from the Sun increases, so that at large distances
the two fields z± become equivalent.
We have said nothing at present about the possible directional anisotropy of the
cascade in the presence of a mean field, yet we have seen that the nonlinear fluxes
depend on the angle between the wave-vectors and the mean field via the correspond-
ing Alfv’en times. If we imagine taking wave-packets in 3D that are elongated along
the mean field, so that the dominant wave-vectors are quasi-perpendicular to the
field, i.e. k·ca 0 then propagation effects among wave-packets become negligible
and we should return to a hydrodynamic or Kolmogorov type cascade (provided
there is equipartition between the fields z± ). In cases where the mean field is strong
B0 |z± |, and remaining within the approximation where characteristic scales
along the field are much greater than those transverse to the mean field, the effective
reduction of nonlinear interactions along the field can be such as to quench parallel
nonlinear interactions entirely, and a further simplification of the MHD equations,
to so called reduced MHD or RMHD is possible. These equations describe two-
dimensional fluctuating fields, polarized in planes orthogonal to the mean magnetic
field and whose nonlinear interactions are confined in this orthogonal plane. Com-
munication across planes occurs only via the propagation along the mean field B0 .
Reduced MHD corresponds therefore to a series of 2D incompressible MHD planes,
coupled together by the linear propagation of the fluctuations as Alfvén waves along
B0 from one plane to another. In this approximation the spectral anisotropy that may
develop can be calculated precisely. In the 2D planes where nonlinear interactions
occur via the 2D incompressible MHD equations, we expect a cascade completely
analogous to a Kolmogorov cascade with hydrodynamic type spectra. However, the
10.2 Magnetohydrodynamic Turbulence 239
evolution at a scale l⊥ in two different planes separated by a distance L along the

field will be completely independent only if the propagation time between the two
planes, τa = l /ca , is longer than the nonlinear time τ ∗ (l⊥ ). There will therefore be
an anisotropy set by the region where τa = τnl (l⊥ ). Considering the Kolmogorov
value for the nonlinear time τ ∗ = τnl , this region will be defined by:
l l⊥ l⊥
∼ 1/3 (10.21)
ca u(l⊥ ) l⊥
which in terms of parallel and perpendicular wavevectors k|| , k⊥ may also be

written as u 1/3
k||c = k⊥ 2/3 k⊥0 .
ca
This condition is known as the critical balance condition for k⊥ and k|| . A spectrum
will develop in the parallel direction as a consequence of uncorrelated perpendicular
planes only for wavevectors satisfying k ≤ k||c .
Magnetohydrodynamic turbulence in more general regimes is still an extremely
active area of research, with vast astrophysical applications: we mention here the
problems of stellar convection (where the role of the magnetic field is secondary, but
the problem of maintaining and generating the field is fundamental, (see the chapter
on dynamo theory), the heating and dynamics of the interstellar medium, the stability
of molecular clouds, viscosity in accretion disks, and the heating of stellar coronae
and acceleration of stellar winds. Let us briefly discuss this latter problem within
the framework of MHD turbulence, with specific application to the case of our star,
the Sun.
10.3 Turbulence and Coronal Heating
The solar corona is a hot, very low density plasma with a temperature of about 106 K.
All stars, with the possible exceptions of A and B type, have external atmospheric
regions where the temperature is much higher than that of the respective photo-
sphere. The photosphere-corona interface is an extremely complex and dynamic
region, developing over a few thousand kilometers in the chromosphere, where the
temperature rises slowly, and then through the very thin transition region where
the temperature jumps from about 104 K to above 106 K over just a few hundred
kilometers. The chromosphere, transition region and corona are characterized by a
substantial energy loss: from the chromosphere and the transition region this occurs
mostly via radiation in visible and EUV, while losses from the corona occur through
radiation, conduction back towards the photosphere, and in the open magnetic field
energy flux transported into the solar wind. A mechanical energy source is required to
keep this temperature imbalance active, since by the second law of thermodynamics
heat can not flow from the colder photosphere upwards. So there must be a viable
mechanism allowing energy to be transported between different regions in the solar
atmosphere, and then releasing and dissipating it in the higher temperature corona.
The source is found in the abundant mechanical energy of the turbulent photospheric
velocity fluctuations, which must be transported upwards and then dissipated within
one or two solar radii and most probably within a few tens of thousand of kilometers
above.
White light images of the solar corona show that is highly structured, with fine
loops and plumes jetting outwards. EUV and x-ray images show how the strongest
energy is released above photospheric regions where the magnetic field is strong,
sunspots and active regions, so that the association of coronal heating with the
photospheric magnetic field is a strong one. Even dark coronal regions, known as
coronal holes, and usually confined to the polar regions of the sun, show remarkably
high temperatures. Their relative darkness is due to the presence of open magnetic
fields, leading to smaller densities as these regions give rise to the expanding solar
wind. An estimate of the energy flux required to sustain the coronal energy losses is
about 107 erg/cm2 /sec for active regions, 8 105 − 106 erg/cm2 /sec for the
quiet sun and = 5 105 − 8 105 erg/cm2 /sec, for coronal holes (Fig. 10.1).
It is natural to imagine that the solar magnetic field might play an essential role
both in the transfer of energy upwards from the photosphere into the chromosphere
and corona and in the subsequent dissipation of this energy. Photospheric motions
buffet and displace the footpoints of magnetic field lines extending into the corona,
therefore launching fast slow and Alfvén waves upwards with a wide range of peri-
ods, ranging from a few hours (lifetime of supergranulation cells) to a few minutes
(granulation cells, solar 5 min oscillations) and below. The Lundquist number cal-
culated for typical densities and magnetic fields in the solar corona and scales of
order 104 km turns out to be huge, around 1012 , so the only way that dissipation
can occur in the corona is that dynamical evolution involve the formation of small
scales. Applying MHD turbulence theory to the solar corona and solar wind how-
ever requires taking into account the characteristics of the system, which involve the
boundary conditions (how photospheric motions couple to the atmosphere above),
the global magnetic filed topology (whether the coronal field is open or closed) and
finally the gravitational stratification and inhomogeneous nature of the corona, which
is finely structured by the magnetic field itself.
The simplest way to consider the photosphere-corona interface is to completely
neglect any details of chromospheric dynamics and consider it a sharp boundary
between two separate ideal plasmas with different densities and temperatures (but
essentially constant pressure across the boundary). The density of the photosphere is
so much greater than that of the corona that photospheric inertia is strong enough to
neglect the back reaction of the field on the plasma. This means that one can assume
that the photospheric velocity field is a given assigned flow, determined by convective
motions below. In the corona however, it is the magnetic field that dominates, i.e. the
plasma from photosphere to corona goes from β 1 to β 1. Using n to denote the
normal to the photosphere, which points radially outwards from the sun, the energy
flux associated with photospheric motions is dominated by the Poynting vector
10.3 Turbulence and Coronal Heating 241
Fig. 10.1 An image of the solar corona superposing three different images taken in differ-
ent wavelengths. Different colors correspond to different temperatures: blue 6 ×105 K , green
1, 3 × 106 K , and red 2 × 106 K (Courtesy NASA-SDO)
c
S= (E × B), (10.22)
4π
where E is the Electric field. Eliminating it using the ideal Ohm’s law we find
S · n = (b⊥
2
(u ph · n) − B0 (b⊥ ph · u ph ))/4π, (10.23)
where the photospheric velocity field is given by U = u ph and the magnetic field at
the photosphere - corona boundary is given by B = B0 n + b⊥ ph (⊥ ph indicates vec-
tor components parallel to the photospheric surface). In this approximation, we have
neglected the energy flux associated with density fluctuations, present in magnetoa-
coustic waves, or the possibility that plasma jets (observed to exist, i.e. spicules) may
contribute significantly to coronal energy balance. The reason we neglect the com-
pressible component of plasma motions is that in the limit of very small β slow modes
essentially become sound waves ducted along the magnetic field, which steepen to
form shocks already in the chromosphere, before arriving into the corona. Fast waves
Fig. 10.2 Cartesian Photosph.

approximation of a coronal
loop
Bo
Photosph.
feel the strong density gradients with height and suffer total reflection, except for
quasi-parallel propagation where they become identical to Alfvén waves.
Equation (10.23) shows that energy flux into the corona is made up of two terms.
The first term is the energy flux associated with emerging or sub-merging magnetic
field, and either carried into the corona or away from it by the radial component of the
velocity field. The second term, proportional to the scalar product of the velocity and
magnetic field components parallel to the photosphere times the radial component
of the magnetic field, describes the energy flux associated with waves along the field
crossing into the corona, or simply the dragging of that field by photospheric motions.
It is hard to separate the contributions from the two terms observationally, though
they typically are of the same order of magnitude. In what follows, we will limit our
consideration to the problem of an assigned velocity field parallel to the photospheric
surface, neglecting for simplicity the dynamics associated with emerging flux.
The coronal response to photospheric motions depends strongly on the frequency
of the motions themselves as well as the coronal magnetic field configuration, for
example if it opens into the heliosphere or closes back down into the photosphere.
Consider the case of the closed solar corona, and consider a configuration of a
coronal loop of length L, transverse cross section diameter l: observationally one
has l/L 1. This allows us to approximate the coronal loop in cartesian geometry,
neglecting the curvature of the field and assuming that it is simply a box delimitated
on top and bottom by photospheric planes, as shown in Fig. 10.2. The transverse,
incompressible oscillations driven by photospheric motions are then made up of
Alfvén waves which, in the linear approximation, obey the equations
∂ u⊥ B0 ∂ b⊥
ρ = , (10.24)
∂t 4π ∂z
∂ b⊥ ∂ u⊥
= B0 , (10.25)
∂t ∂z
where B0 is the strong, uniform axial magnetic field and u⊥ , b⊥ are the transverse
oscillation of velocity and magnetic field. We have neglected gravitational stratifi-
cation along the loop in the z direction, because the scale height at a temperature
of 106 K is of the same order of magnitude of the height L of the loop itself. These
equations are solved assigning the velocity field u⊥ = u ph at the boundaries in

z = 0, L, and supposing that at t = 0 the transverse field vanishes everywhere,
i.e. the field lines are all initially straight and perpendicular to the photosphere. To
simplify the problem further, consider photospheric motions to be assigned only on
one of the two boundaries, while the other will be kept anchored with so that u⊥ = 0,
in z = 0, while u⊥ = u0ph (x, y) cos(ω ph t) in z = L. ω ph = 2π/T ph is a typical
frequency associated with photospheric motions, with T ≥ 3m, much longer than
the propagation time of an alfvén wave along a loop which is of the order of tens of
seconds or less.
A solution of Eqs. (10.24) and (10.25) with the assigned initial and boundary
conditions is therefore given by
sin(ω ph z/ca0 )
u⊥ = u0ph (x, y) cos(ω ph t) , (10.26)
sin(ω ph L/ca0 )
b⊥ cos(ω ph z/ca0 )
√ = u0ph (x, y) sin(ω ph t) , (10.27)
4πρ sin(ω ph L/ca0 )
√
where ca0 = B0 / ρc is the coronal Alfvén speed.
This is a valid solution only if the frequencies associated with photospheric
motions are such that sin(ω ph L/ca0 ) = jπ , with j an integer (positive, nega-
tive, or 0), because of the denominators in Eqs. (10.26) and (10.27). The zeros of
the denominators identify frequencies where the photospheric driver is in resonance
with Alfvén wave oscillations in the loop, in which case linear solutions give rise to a
secular growth of the amplitude of oscillations with time. In other words, at resonant
frequencies a coronal loop may store significant amounts of energies, that will be
saturated or limited only by dissipative processes: so the idea comes naturally that
the photospheric driving of closed coronal loops will lead to the spontaneous devel-
opment of MHD turbulence in terms of Alfvénic-type fluctuations within the loop,
which may provide an explanation for the heating of the loop itself. This because we
have identified a process which leads to a secular increase of energy in the loop, with
amplitudes of the fields u⊥ , b⊥ growing without limit; if nonlinear interactions and
a cascade did not develop, the saturation of growth due simply to the linear resistive
and viscous damping with coronal values of the Lundquist and Reynolds numbers
would lead to enormous oscillations, with field amplitudes much greater than those
observed, albeit indirectly, in the corona itself. The heating of the loop and saturation
of amplitudes will therefore occur via a nonlinear cascade.
As mentioned above, the characteristic propagation time of Alfvén waves along a
loop, a few seconds or tens of seconds, is typically much smaller than that associated
with the variations of photospheric motions. It is therefore important to distinguish
the resonances with j > 0 from the resonance at null frequency j = 0, correspond-
ing to stationary photospheric flows (i.e. stationary as measured using the coronal
crossing time clock). This case, for which ω ph t → 0, can be found taking the corre-
sponding limit of Eqs. (10.26) and (10.27): we find that sin(ω ph L/ca0 ) ω ph L/ca0 ,
sin(ω ph z/ca0 ) ω ph z/ca0 , from which we obtain velocity field fluctuations
u⊥ = u0ph (x, y)z/L , (10.28)
while for the magnetic field,
b⊥ ca0
√ = u0ph (x, y) t. (10.29)
4πρ L
Therefore, with stationary photospheric flows, the coronal velocity field remains
limited in amplitude with growing time, while the transverse magnetic field grows
secularly. Another way of interpreting this result is to consider the original coronal
field lines as strings, which are tied on the plane where the velocity vanishes, but
are dragged about by the stationary flow on the other photospheric plane. It is this
dragging which leads to a secular growth of the transverse field. Parker, in 1972,
[24], proposed that this kind of coronal dynamo might be the main cause of coronal
heating. It is important to point out that the slow, stationary photospheric motions are
not homogeneous but depend on position, therefore the growing perturbed coronal
magnetic field will not be in equilibrium, so a coronal dynamics will ensue where the
stresses in the transverse field attempt to relax towards an equilibrium through the
action of the Lorentz force on the plasma. Because the plasma β of the corona is very
small, the corresponding coronal equilibrium configurations would be expected to be
force-free states. Parker conjectures however that for generic motions, the force-free
equilibrium states attainable could not be continuous, but would contain singularities
in the topology of the field, or current sheets, whose stability we have discussed in
the previous chapter. In the language of turbulence, Parker conjecture that nonlinear
interaction would naturally lead to extremely thin structures, whose destruction via
reconnection would lead to coronal heating.
The energy flux into the corona resulting from the field-line dragging by stationary
flows results grows in time without bound (within the linear approximation consid-
ered here) as can be understood by calculating the Poynting flux using expressions
Eqs. (10.28) and (10.29): S = ca0 ρ|uph 0 |2 c t/L. Parker estimated the time it would
a0
take for such flux to reach a value that could balance coronal energy losses, S = ,
and from this estimate calculated the corresponding amplitude of fields and types of
current sheets that would have to develop. For a typical active region, Parker found
a time required to reach a stationary flux of 107 ergs/cm2 /sec, starting from typi-
cal supergranular estimates of the photospheric velocity field, u ph 0.5 km/s, an
axial magnetic field B0 100 Gauss, and a loop of length L 105 km, of about
t = 5.0 104 sec, compatible with the correlation time of supergranulation itself.
Numerical simulations of this problem (within the approximations we have made)
indeed prove that the coronal volume develops a regime of MHD turbulence where the
energy contained in magnetic field oscillations dominate the kinetic energy, and dis-
sipation does occur through the spontaneous development of turbulent current sheets,
resulting in a dissipation rate that does not depend on the Lundquist number. The
energy spectra however differ from those estimated using the models of Iroshnikov–
Kraichnan or Kolmogorov. As may be expected, detailed studies of the problem in
geometries closer to those of the real Sun and with a more appropriate treatment of
the photosphere-corona boundaries are the subject of intense and active research.
10.1. Derive Eq. (10.5), by multiplying the Navier-Stokes equation (10.1) and the
Poisson equation for the pressure (10.3) by e−i k·x , and using the definition of the
Fourier transformed velocity field (10.4). Recall that

d 3 xe−i(k−q)·x = (2π )3 δ 3 (k − p)
where δ is the Dirac distribution.
10.2. Show that Eqs. (10.26) and (10.27) are solutions of Eqs. (10.24) and (10.25).
10.3. Show that a solution of the system (10.24) and (10.25) in the case where
sin(ω ph L/B0 ) = jπ with j a relative integer of any value, is given, for the velocity
field, by

z
ca0 t
ω ph z
u⊥ = (−1) j u0ph (x, y) cos ω ph t − sin ω ph t sin . (10.30)
L L ca0
Find the corresponding solution for the magnetic field fluctuation which satisfies
the initial condition b⊥ = 0 everywhere as well as the boundary conditions. This
problem shows that the resonances with non-zero frequencies lead to secular growth
of both the magnetic and velocity fields.
Solutions
10.3. To eliminate the magnetic field from Eqs. (10.24), (10.25), differentiate
Eq. (10.24) with respect to time and Eq. (10.25) with respect to z. The magnetic
field may then be found by solving Eq. (10.25) with the given solution for u⊥ :

B0
ω ph z B0 t ω ph z
b⊥ = (−1) j u0ph (x, y) sin ω ph t 1 − cos + cos cos ω ph t .

ω ph L ca0 L ca0
(10.31)
Chapter 11
Build-Up of Magnetic Fields
Abstract The problem of the build-up and maintenance of cosmical magnetic fields
is presented. The basic phenomenological theory of astrophysical dynamos is out-
lined. The more formal approach of mean-field electrodynamic is also exposed.
Simple solutions of the dynamo equations are discussed. Finally, the problem of the
creation of magnetic fields in an unmagnetized plasma is briefly considered.
11.1 The Dynamo Problem
Magnetic fields appear to be ubiquitous in the Universe: almost all celestial bodies,
from planets and stars to galaxies and galaxy clusters host magnetic fields of different
intensities. The magnetic field on Earth is about one gauss and most of the planets,
with the exception of Venus and possibly Mars, also exhibit magnetic fields of varying
strength. The average magnetic field of the Sun is of the order of a few gauss,
but locally it can be much stronger, reaching values up to a few thousand gauss in
sunspots. Magnetic fields of normal stars also reach values around 1,000 G, but those
of compact stars may be extremely intense by comparison: the field of a magnetized
white dwarf is typically of the order of a few million gauss and the magnetic fields
of pulsars fall in the range of 1012 G. The galactic magnetic field, on the contrary,
is extremely weak, of the order of a few microgauss. To a first approximation, the
external fields of planets and stars are well approximated by magnetic dipoles, but,
as we shall see, the internal fields are likely to have a very different structure. At first,
one might think that magnetic fields in different contexts throughout the Universe
are all evolved remnants of primordial fields, present since the very beginning of the
cosmic history. A little thought, however, shows that this assumption must be false.
Cosmic plasmas are not perfect conductors and are therefore subject to ohmic decay,
even if their resistivity is very small, as shown by the induction equation describing
the time evolution of magnetic fields (see Eq. (5.25)):
∂B
= ∇ × (U × B) + η∇ 2 B. (11.1)
∂t

248 11 Build-Up of Magnetic Fields
The two terms on the right are associated with very different timescales given,
respectively, by τ f = L/U, the fluid or convective scale and τd = L2 /η , the diffusive
or dissipative scale. Here U is a typical value of the fluid speed, L is the scale of the
spatial variation of the magnetic field and η = (c2 /4πσ) is the magnetic diffusivity
of the plasma which we will take here to be uniform.
Using the typical dimensions of the system under consideration for L, τd turns
out to be extremely long for stars (τd 1010 years) and galaxies (τd ≥ 1026 years),
but much smaller, about 105 years for the Earth. The problem of how to maintain
the magnetic field, however, is not limited to ours and other planets, since the field
of the Sun is known to reverse its sign over a period of about eleven years (yielding
an average 22 year Solar cycle) and such a rapid oscillation (with respect to the
computed τd ) requires a dynamic explanation. Another aspect of the problem is that
the existence of a primordial magnetic field at the beginning of the Universe is itself
an unanswered question. In the absence of a primordial field, when, how and on
which timescales has the magnetic field been generated?
The solution to the problem of the build-up of magnetic fields can only come from
the convective term of the induction equation that couples fluid motion and magnetic
field and that, at least in principle, might not only contrast but overcome the decay
due to the diffusive term. The effect of velocity on magnetic fields, which defines
dynamo action, may result in a reduction of the spatial scales entering the definition
of the dissipative scale, thus allowing changes to occur over timescales considerably
shorter than τd .
It is often useful to describe the magnetic field in terms of the vector potential A,
related to B by
B = ∇ × A,
a representation made possible by the fact that ∇ · B = 0. A further gauge condition,

i.e. ∇ · A = 0, is necessary to completely specify A. As discussed in Chap. 5, A is
defined modulus the gradient of an arbitrary function φ, and thanks to this gauge
invariance B remains the same.
Expressing B in terms of A in Eq. (11.1) we obtain
∂ A
∇× − U × (∇ × A) − η∇ 2 A = 0,
∂t
which implies that
∂A
= U × (∇ × A) + η∇ 2 A + ∇φ. (11.2)
∂t
Taking the divergence of the above equation and using the gauge condition ∇ · A = 0
we get:
∇ 2 φ = ∇ · (U × B).
11.1 The Dynamo Problem 249
The gauge condition therefore eliminates the arbitrariness of φ.

Equations (11.1) and (11.2) constitute the basis for the discussion of the dynamo
problem.
The complete solution of the problem would require the induction equation to be
solved at the same time and in parallel with the momentum equation, an extremely
difficult non linear problem. At the heart of the dynamo problem lies understanding
which velocity fields, sustained by the available forces, are capable of supporting
a growing, oscillating or steady magnetic field against ohmic decay. To limit the
difficulties, the problem is often restricted to the kinematic dynamo problem, where
the induction equation is actually decoupled from the equation of motion by assuming
a particular form for the velocity field. Even so, the determination of a flow with the
appropriate characteristics is not an easy task. We begin by examining the simplest
possible flow configurations. Unfortunately, as we shall see, none of these basic flows
are able to generate or sustain a magnetic field.
11.2 The Anti-Dynamo Theorems
A certain number of two- and three-dimensional theorems can be proven that deny
the possibility of producing a magnetic field lasting in time. Common assumptions
to all theorems are the incompressibility of the flow, ∇ · U = 0, and the vanishing
of U, B and A at infinity. We shall now illustrate the most relevant ones.
Theorem 1 In cartesian coordinates [x, y, z] it is impossible to generate by dynamo

action a z-independent field that vanishes at infinity.
Assume z-independence both for U and B. Then write
B = B z ez + B ⊥ = B z ez + ∇ × (Aez ),
and
U = Uz ez + U ⊥ ,
with A = A(x, y). Equations (11.1) and (11.2) then transform into
∂ Bz
+ (U ⊥ · ∇)Bz = η∇ 2 Bz + (B ⊥ · ∇)Uz ,
∂t
∂A
+ (U ⊥ · ∇)A = η∇ 2 A.
∂t
Now multiply the last equation by A and integrate over the whole space

1 ∂
A dV = −
2
A(U ⊥ · ∇)AdV − η (∇ A)2 dV.
2 ∂t
But
1 1 1
A(U ⊥ · ∇)A = (U ⊥ · ∇)(A2 ) = ∇ · (A2 U ⊥ ) − A2 (∇ · U ⊥ ).
2 2 2
The last term of the preceding equation vanishes (incompressibility!) and therefore
the first integral on the rhs can be transformed into a surface integral at infinity where
all the fields vanish. The integral of A2 therefore decays to zero (unless A = const.
in which case B would be identically zero). Since A decays in time, B ⊥ must also
tend to zero. But then the equation for Bz no longer has a source term, so that Bz will
vanish as well. This theorem states that dynamos are intrinsically three-dimensional,
and no 2D field can be maintained by dynamo action.
Theorem 2 In cartesian coordinates [x, y, z] a planar flow cannot produce a
dynamo effect.
The velocity field now has the form U = [Ux (x, y, z, t), U y (x, y, z, t), 0]. Notice
that the field components are allowed to depend on z. The z-component of Eq. (11.1)
is
∂ Bz
+ (U · ∇)Bz = η∇ 2 Bz ,
∂t
since Uz = 0. Multiplying this equation by B z , integrating over the volume and
repeating the reasoning of the previous theorem, we see that the integral of the
convective term (U · ∇)Bz vanishes. Thus,

1 ∂
B 2z dV = −η (∇ B z )2 dV,
2 ∂t
showing that B z must tend to zero in time. If B z = 0, it is easy to see that the vector
potential becomes simply A = A(x, y)ez , and the remaining field components are
given by Bx = (∂ A/∂ y) and B y = −(∂ A/∂x). Moreover, defining Q = U × B, we
have Q = (U · ∇) A. Taking the curl of Eq. (11.1) and evaluating the z-component
we get:
∂
(∇ × B)z = [∇ × (∇ × Q)]z + η∇ 2 (∇ × B)z .
∂t
But,
∂ By ∂ Bx ∂2 A ∂2 A
(∇ × B)z = − = + ≡ ∇⊥
2
A,
∂x ∂y ∂x 2 ∂ y2
and
[∇ × (∇ × Q)]z = −∇ 2 Q z = −∇⊥
2
(U · ∇)A,
Introducing these expressions in the induction equation , we finally obtain

11.2 The Anti-Dynamo Theorems 251
∂A
∇⊥
2
+ (U · ∇)A − η∇ 2 A = 0.
∂t
Since this equation is linear in A, we can Fourier transform it. The operator ∇⊥2
becomes −k⊥ = −(k x + k y ) and the Fourier transform of the expression in square
2 2 2
brackets must therefore vanish. This implies
∂A
+ (U · ∇)A = η∇ 2 A,
∂t
namely the same equation already encountered in Theorem 1. Following the same
procedure we can prove that A → 0: the magnetic field therefore decays. No dynamo
action is possible for planar flows.
Moving now to a more realistic axisymmetric configuration, we can prove that
Theorem 3 An axisymmetric magnetic field vanishing at infinity cannot be main-

tained by dynamo action
This is Cowling’s celebrated theorem (1934) [25]. The proof, which is different
from the original one, is analogous to that of Theorem 1. Any axisymmetric vector
field can be represented in a cylindrical coordinate system (r, θ, z) as the sum of
a toroidal (i.e. azimuthal) field and a poloidal field lying in the meridional planes.
The latter, in turn, is the sum of the radial and vertical components of the field.
Adopting such a representation for the magnetic field B = Bθ eθ + B p , it is easy
to see that the poloidal magnetic field can be expressed as B p = ∇ × (Aeθ ), with
A = A(r, z, t). Because of the assumed axial symmetry (∂/∂θ) = 0, ∇ · A = 0 and
the θ−component of Eq. (11.2) becomes
∂A 1 A
+ U p · ∇(r A) = η ∇ 2 A − 2 . (11.3)
∂t r r
We now multiply the above equation by r 2 A and integrate over the volume. Again,
the integral of the convective term can be transformed into

1
∇ · (r 2 A2 U)dV,
2
which vanishes if A has “good” convergence properties at infinity. We therefore

arrive at

1 ∂
(r A )dV = −η |∇(r A)|2 dV,
2 2
2 ∂t
which shows that (r A) must decay in time, unless r A = const. But in this case A
would diverge on the axis and therefore also B p . After A has gone to zero, B p = 0.
On the other hand, using the toroidal-poloidal representation of B, we find that the
θ-component of Eq. (11.1) is
∂ Bθ Bθ Br Bθ
= (B p · ∇) Uθ − (U p · ∇) Bθ + Ur − Uθ + η ∇ 2 Bθ − 2 ,
∂t r r r
or equivalently,
∂ Bθ B U Bθ
θ θ
+ r (U · ∇) = r (B p · ∇) + η ∇ 2 Bθ − 2 . (11.4)
∂t r r r
If B p = 0 the above equation reduces to
∂ Bθ B Bθ
θ
+ r (U · ∇) = η ∇ 2 Bθ − 2 .
∂t r r
Multiplying by Bθ /r 2 and integrating over the volume we obtain:

1 ∂ 1
(Bθ /r ) dV +
2
∇ · [(Bθ /r ) U]dV = −η
2
|∇(Bθ /r )|2 dv,
2 ∂t 2
and, discarding the second integral, we conclude that also Bθ has to decay, unless
Bθ ∝ r which is unacceptable since Bθ is assumed to vanish at infinity. Steady
axisymmetric dynamos do not exist.
Finally, it is also possible to prove that purely toroidal flows cannot generate
steady magnetic fields.
The above theorems, and in particular Cowling’s theorem, show that even rela-
tively simple field topologies, such as the approximate dipole structure of the Earth’s
magnetic field or of a sunspot, are not directly amenable to simple symmetric fluid
motions, which must have a more complicated structure. These rigorous results pre-
vented substantial progresses on dynamo theory for a long time.
11.3 Phenomenological Theory of Astrophysical Dynamos
It is clear that understanding the structure of cosmic magnetic fields and of the
systems of currents producing them requires clever interpretations of the available
data. Just to summarize the most relevant features deduced from observation, we
first remark that all magnetized bodies rotate. If the magnetic fields configurations
may be approximately described by dipoles at lowest order, their axes turn out to
be more or less aligned with the rotation axes, although there are exceptions to
this. Terrestrial planets are composed of a non-conducting solid mantle, enclosing
highly conducting liquid iron cores, where convective motions carry heat from the
center, most likely composed of solid iron, to the mantle and outwards. Jupiter, a
gaseous planet, is mainly composed of an alloy of metallic hydrogen and helium. All
the magnetic decay timescales of these planets are very short compared with their
lifetimes, spanning from 104 to 108 years. The Sun is known to rotate and to possess
a convection zone extending from about 0.7 R almost to the photosphere.
11.3 Phenomenological Theory of Astrophysical Dynamos 253
Unfortunately, in most cases we cannot observe the motions that generate the
external magnetic fields we see in celestial bodies. This is certainly true for the Earth
and the planets, where observations can at most suggest the pattern of the internal
motions. The situation with the Sun has greatly improved with the advent of the
helioseismology, the detailed study of the spectrum of the oscillations seen on the
surface, interpreted as global oscillation modes. A notable exception to the shortage
of direct observations is provided by the galactic disc, where we are able to actually
see the the dynamo mechanisms at work. An important conclusion, based on all the
available evidence, both direct and indirect, is that, independently of the form of the
external magnetic field, the internal field is essentially azimuthal, as we are now
going to illustrate, limiting for simplicity our discussion to the representative cases
of the Earth, the Sun and the Galaxy.
11.3.1 The Generation of Toroidal Fields from Poloidal Ones
The dipole field of the Earth, measured on the surface, has a strength of about 0.3 G
at the equator and 0.6 G at the poles. If one extrapolates these values to the surface
of the core, r = Rc R E /2, where R E is the radius of the Earth, one arrives at
B p (Rc ) 5 G. On the other hand, a careful analysis of the observed anomalies
of the terrestrial magnetic field suggests that the surface of the molten metal core
does not rotate uniformly. Its angular velocity, Ω, is higher at the poles than at
the equator, which implies that Ω is a decreasing function of the distance from the
rotation axis. It is clear that any non-uniform rotation of the core would stretch the
magnetic field lines and drag them out of the meridional planes, thus generating a
toroidal component of the magnetic field that was originally absent. It is also clear
that the generated toroidal field would have opposite directions in the northern and
southern hemispheres.
To be more quantitative, consider again the azimuthal component of the induction
Eq. (11.4):

∂ Bθ Bθ Uθ
r ∂t + (U · ∇) = r (B · ∇) + η ∇ 2 Bθ − Br 2θ
r r r (11.5)
= r (B · ∇Ω) + η ∇ 2 Bθ − Br 2θ ,
where the angular velocity Ω = Uθ /r has been explicitly introduced. The left hand
side of the equation is simply the lagrangian derivative of Bθ /r , multiplied by the
factor r . The first term on the right could, in principle, produce an increase of the
azimuthal field, while the last one is as usual the diffusive term responsible for
the field’s decay. Consider a case in which Bθ = 0 at t = 0, so that B = B p .
Equation (11.5) then becomes
∂ Bθ
= r (B p · ∇Ω),
∂t
Fig. 11.1 Sketch illustrating

the generation of a toroidal
field from a purely poloidal
one
showing how a toroidal field can be generated from a purely poloidal one, see
Fig. 11.1.
The existence of differential rotation, ∇Ω
= 0, thus appears to be a necessary
condition for the growth of toroidal fields. Given an appropriate Ω profile, Bθ would
increase with time until either the decay term sets in or the field grows sufficiently
that a back reaction on the flow via the Lorentz force begins to inhibit differential
rotation, preventing any further growth. In the case of the Earth, since the value
of the core’s lag with respect to the mantle is known, it is possible to estimate the
core-mantle speed difference at the core radius, which turns out to be very small,
only 3 × 10−2 cm/s, corresponding to a lag of a complete turn every 2 × 103 years.
On the other hand, the decay time of the terrestrial magnetic field is of the order of
105 years, during which period the differential rotation progresses five times or more
around the core, thus allowing a substantial growth of the azimuthal field.
According to Bullard [26] we can envisage that the increase of Bθ stops when the
Lorentz force, ( j × B)/c = [(∇ × B)× B]/4π (B p /4π Rc )Bθ , balances the effect
of the Coriolis force that acts on the convective motions, 2 ρ Ω E × v c 2 ρ Ω E vc ,
where Ω E is the angular velocity of the Earth, ρ the density and v c the typical speed
of convective motions. Therefore
Bθ
(B p /4π Rc )Bθ 2 ρ Ω E vc or 8π ρ Ω E vc Rc /B 2p .
Bp
Using ρ 10 g cm−3 , vc 10−2 cm s−1 and Ω E 7 × 10−5 s−1 we find Bθ /B p

200, which proves that indeed the toroidal component is the dominant part of the
internal terrestrial field. In spite of this, that field is inevitably hidden, because the
poloidal currents associated with it cannot flow upwards through the non-conducting
mantle.
As mentioned previously, a lot more is known about the Sun, as a result of helio-
seismologic observations. From the detailed study of the spectrum of the global
oscillations and, in particular, from the rotational splitting of the low order modes,
we are now able to infer the behavior of the angular velocity of the Sun from the
surface to the bottom of the convection zone and, with a somewhat larger uncertainty
all the way down to ≈0.2R . Figure 11.2 shows the profile of solar rotation rate as
a function of the radius from the sun’s center, as deduced form the analysis of the
solar oscillations.
Fig. 11.2 Internal rotation of the Sun from helioseismology
The visible solar surface does not rotate rigidly, with the photospheric angular
velocity, unlike the terrestrial case, being higher at the equator than the poles. Moving
inward, the angular velocities at all latitudes are seen to converge towards a common
value, so that the rotation appears to be uniform below ≈0.6R .
The general scenario, therefore, is similar to that of the Earth’s interior and we
expect to find a strong toroidal field in the convection zone. The magnetic pressure
inside the flux tubes increases, making them buoyant and forcing them to emerge
through the photosphere in the form of bipolar magnetic regions, whose strongest
manifestations are sunspots. Given the mechanism that generates the toroidal field,
we expect the leading magnetic spots to have different polarities in the northern and
southern hemispheres. This is precisely what observations of bipolar sunspots show,
thus confirming the correctness of our deductions. Finally, in the gaseous galactic
disc the main magnetic field is observed to be azimuthal. The poloidal component,
which is certainly present, is sufficiently small to be hidden by the fluctuations of
the much stronger toroidal field.
11.3.2 The Generation of Poloidal Fields from Toroidal Ones
So far, we have seen that differential rotation is capable of generating a toroidal

field from a poloidal one. But this not only does not solve the dynamo problem,
but rather complicates it. The fact is that the fields that we see are, in most cases,
poloidal rather than toroidal: the general mean dipolar field of the Earth, apart from
local and secular variations, is substantially steady, maintaining both its strength and
direction for periods of 105 years or longer; the mean solar field, to a first rough
approximation, can also be approximated by a poloidal field which however varies

with time, reversing sign every 11 years (so that a complete cycle, i.e. the complete
regeneration of the initial conditions, takes 22 years). Given the presence in both cases
of a strong internal toroidal field, we see that a viable dynamo theory should provide
mechanisms capable of generating both steady and dynamical poloidal fields from
an initial toroidal one: Cowling’s theorem however shows that the vector potential
A that generates a poloidal field must decay in time and this seems to preclude the
possibility of a growing poloidal field from a toroidal one.
A brilliant way out of the problem was envisaged by Parker [27] in 1955. Parker’s
starting point was a careful analysis of convective motions, that, as we have seen,
are present both in the terrestrial and in the solar cases. Consider a rising blob of
plasma in the convective layer of a rotating body: it is naturally subject to the action
of the Coriolis force. Because of the density stratification, it must expand as it rises
so that its velocity developes non vanishing tangential components. In a spherical
coordinate system, with Ω defining the polar axis, Ω = Ω(cos θ er − sin θ eθ ), the
tangential components of the Coriolis force, −2 ρ(Ω × U) are
2ρ Ω[Uφ cos θ eθ − (Uθ cos θ + Ur sin θ)eφ ]
To simplify the discussion let’s neglect the radial motion completely by putting
Ur = 0. The motion of a fluid particle in the plane tangent to the sphere at a given
point is also subject to a Coriolis force which in this case is simply
2ρ Ω cos θ(Uφ eθ − Uθ eφ ).
The effect of such a force on a diverging flow (corresponding to a rising blob of

plasma) can be easily found from the preceding formula considering, in turn, all
possible signs for the components Uθ and Uφ . In the northern hemisphere, θ < π/2,
the fluid rotates in a clockwise direction with respect to the local vertical, while an
anti-clockwise vortex forms in the opposite hemisphere. Of course, for a descending
blob, giving rise to a convergent flow, the senses of rotation are the opposite. Such
cyclonic motions are well known in meteorology, where they control the local climate
and occasionally develop into hurricanes. If these cyclonic motions carry magnetic
flux tubes with them, the twist converts toroidal fields into poloidal ones. Let’s in
fact consider a piece of a toroidal magnetic field line and assume that the freezing-in
condition applies. A rising blob of plasma will carry the field with it forming a loop,
approximately shaped as the greek capital letter Ω, that, for a π/2 twist, will lie in
the meridional plane. A local, small-scale poloidal field has thus been formed from
the original toroidal field, see Fig. 11.3
As we have seen, the sense of rotation of the twisting loop changes going from
one hemisphere to the other, but the direction of the toroidal field changes as well, so
that the circulation sense of the poloidal fields of the newly formed loops is the same
in both hemispheres. The same conclusion can be reached also by considering the
current associated with the magnetic loops generated by the twisting motions. The
current is parallel to the toroidal field in the northern hemisphere and antiparallel in
Fig. 11.3 Sketch of the

formation of a local poloidal
loop from a toroidal field.
a Segment of a toroidal field
line. b Field distortion due to
a rising flow. c π/2 twist due
to the Coriolis force. d Final
configuration after a magnetic
reconnection
the southern one, but because of the reversal of the toroidal field, the sense of the
currents is the same in both hemispheres.
Each convective cell thus generates a small scale poloidal field. If their number
is large enough, two adjacent loops may come in close contact and their oppositely
directed fields may reconnect forcing the loops to coalesce and to form larger struc-
tures. At the end of the process a poloidal field of the same scale of the toroidal field
results.
So far we have considered only rising motions, but evidently sinking motions
are also present, which would tend to produce a poloidal field of opposite sense.
Therefore, to have a net effect, some asymmetry must exist. There can be several
causes for such an asymmetry, the most obvious being the density stratification which
forces rising cells to expand and falling ones to contract. Magnetic buoyancy also
tends to favor rising motions. The existence of motions of overall definite handedness
thus appear to be a necessary ingredient for the generation of poloidal fields from
toroidal ones.
Since the rate of generation of B p must somehow be proportional to the toroidal
component Bθ , the simplest possible model of poloidal field generation is obtained,
as suggested by Parker, by adding a source term αBθ to the rhs of Eq. (11.3), where
α has the dimensions of a velocity and, in general is a function of position and time
α = α(r, t). Equations (11.3) now reads
∂ Ap 1 Ap
+ U p · ∇(r A p ) = αBθ + η ∇ 2 A p − 2 . (11.6)
∂t r r
The term αBθ , responsible for the so-called α-effect, is equivalent to an electric
field or mean electromotive force Eθ = αBθ /c, that counteracts the decay due to
the resistive term. Depending on the relative importance of these terms, the poloidal
field may grow or reach a steady state. Recalling our previous considerations, we see
that in order to have the same sign of the current, jθ = σEθ = (σ/c)αBθ , in both
hemispheres α has to change sign across the equator.
11.4 Mean-field Electrodynamics
The pioneering approach of Parker prompted the plasma community to look for a
more formal justification for the appearance of the crucial electromotive force term
in the induction Eq. (11.6). This was achieved by the so-called mean-field theory first
proposed by Steenbeck, Krause and Rädler in 1966,1 that we shall now illustrate,
closely following their treatment.
Basically, the theory relates the time evolution of the mean magnetic field to the
statistical properties of the turbulent velocity field, v. Turbulence is usually con-
sidered to be incompressible and isotropic, i.e. invariant for arbitrary rotations of
the axes. However, as already remarked, the turbulent velocity field should posses a
definite handedness, which implies that it is not invariant for reflections of the axes
with respect to a plane. Reflection invariance, in fact, would mean that right-handed
motions are as probable as left-handed ones, contrary to our assumption. A velocity
field lacking reflection invariance is said to have a non vanishing helicity, h, defined
as the average value of v · (∇ × v):
h = v · (∇ × v)
Here and in the following, the average F of a certain quantity F has to be intended
as an ensemble average, defined as the expectation value of F in an ensemble of
identical systems. Notice that the averaging operation commutes with both space
and time derivatives.
It is a general characteristic of homogeneous and isotropic turbulence that the
small-scale fluctuations at a given point, r, have a correlation with another quantity
at a different point, r , only if the distance between the two points is not too large: in
other words, correlations vanish if |r − r | > λcorr , where λcorr is the correlation
length. Similarly, the correlations at different times are zero if if they are taken
between times separated by t − t > τcorr , where τcorr is the correlation time. Thus,
to determine the correlation between any two fluctuating quantities, the knowledge
of those quantities is required only in the neighborhood of the point considered. If
the initial and boundary conditions are outside of that neighborhood, they will not
influence the value of the correlation.
Vectors can be classified into two different classes according to their transfor-
mation properties for reflections with respect to a plane: polar vectors (or simply
vectors) change their sign upon reflection, whereas axial vectors don’t. It is clear
that axial vectors must be connected with operations where handedness is involved,
like cross products or curls. The velocity, electric field and current density are vectors,
while the magnetic field is an axial vector.
The scalar product of a vector and an axial vector produces a quantity that changes
sign under reflections: such a quantity is called a pseudo-scalar. An example of a
1An exposition in english of the original german papers can be found in the book by Krause and
Rädler [13].
11.4 Mean-field Electrodynamics 259
pseudo-scalar is the helicity. If h

= 0 the turbulent velocity field is not reflection
invariant. In fact, if the turbulence were reflection invariant, on the one hand h should
remain unaltered under reflections since all the properties of the velocity field do not
change, while on the other it should change sign since it is a pseudo-scalar, Hence,
h = 0 for such a field.
Let’s now consider a small-scale turbulent velocity field, u, which is statistically
stationary, homogeneous and pseudo-isotropic, meaning that it is rotation invariant
but not reflection invariant. Such a turbulence will both produce a fluctuating mag-
netic field, b on a small-scale l, and sustain a mean magnetic field, B0 , on a much
larger scale L, so that the total field will be B = B 0 + b. The induction equation
(11.1) then becomes:
∂
(B 0 + b) = ∇ × [(U 0 + u) × (B 0 + b)] + η∇ 2 (B 0 + b), (11.7)
∂t
where U0 is a large-scale velocity field. We now average this equation on a scale
= O(l). Taking into account that u = b = 0, we obtain
∂ B0
= ∇ × [U 0 × B 0 + u × b ] + η∇ 2 B 0 . (11.8)
∂t
Subtracting Eq. (11.4) from Eq. (11.7) we get:
∂b
= ∇ × [U 0 × b + u × B 0 + u × b − u × b ] + η∇ 2 b. (11.9)
∂t
If U 0 , B 0 and u are known, the above equation in principle determines b, even if the
actual solution is hard to obtain. To proceed, let’s evaluate the order of magnitude of
the various terms entering Eq. (11.9). The large scale velocity field can be removed by
adopting a reference system moving with the local value of U 0 . The orders of magni-
tude of the other terms at the rhs are, respectively: O(u B0 /), O(ub/), O(ηb/2 ). If
the small-scale magnetic Reynolds number (u/η) 1, the term ∇×[u×b−u×b ]
can be neglected and Eq. (11.9) becomes a linear equation.2 The large scale field B0
is determined by Eq. (11.8), where the term u × b appears to be the only source
term capable of opposing the field decay. To allow the growth of the field this term
must therefore be kept. This procedure is known as the quasi-linear approximation
or first-order smoothing.
Taking into account all the above approximations, the final form of our system of
equations is:
∂ B0
= ∇ × u × b + η∇ 2 B 0 , (11.10)
∂t
2 The assumption on the magnetic Reynolds number may be difficult to justify in astrophysical
conditions. In spite of this it is customary to accept it.
and
∂b
= ∇ × (u × B 0 ) + η∇ 2 b. (11.11)
∂t
Cosmic plasmas are, as a rule, excellent conductors: we shall therefore start by
looking for the solutions of Eq. (11.11) with η = 0, which, in this high-conductivity
limit, can be written as
t
b(r, t) = ∇ × [u(r, t ) × B 0 (r, t )]dt . (11.12)
−∞
According to our previous discussion, we identify the source term in Eq. (11.10)
with an electromotive force and write
t
1 1
E(r, t) = u × b = u(r, t) × ∇ × [u(r, t ) × B 0 (r, t )] dt .
c c
−∞
As illustrated above, the integrand is different from zero only if t − t < τcorr and
in this time interval the large scale field B0 can be assumed to be constant, B0 (r, t )
B0 (r, t). Changing the integration variable into τ = t − t , we finally have
∞
1
E(r, t) = u(r, t) × ∇ × [u(r, t − τ ) × B 0 (r, t)] dτ . (11.13)
c
0
Since
∇ × (u × B0 ) = (Bo · ∇)u − (u · ∇)B0 (for ∇ · B0 = ∇ · u = 0),
the i-th component of Eq. (11.13), in a right-handed cartesian coordinate system

[x1 , x2 , x3 ], can be written as
∂ B0k
c Ei (r, t) = i jk w jkl B 0l − i jk w jl , (11.14)
∂xl
where we have introduced the notations:

∞ ∞
∂u k (r, t − τ )
w jkl = u j (r, t) dτ and w jl = u j r, t)u l (r, t − τ ) dτ .
∂xl
0 0
and i jk is the totally antisymmetric tensor.3
3 i jk vanishes if at least two of the indices are equal, equals +1 for a permutation of the indices
even with respect to the fundamental permutation 123 and −1 for an odd permutation.
The average quantities wi jk and wil reflect the assumed properties of the turbulent
velocity field: for an isotropic turbulence they must remain unchanged for arbitrary
rotations of the axes. It is then easy to deduce a series of important properties of
these quantities.
First of all, wi jk = 0 if at least two indices are equal. To show this, consider a
rotation of π/2 around the x1 axis. If ei are the unit vectors along the coordinate axes,
as a result of the rotation e2 → e3 , u 2 → u 3 so that w112 transforms into w113 . But
a rotation of −π/2 around the same axis would give e2 → −e3 and w112 → −w113 .
Thus, w112 = w113 , but also w112 = −w113 which implies w112 = w113 = 0.
Similarly, considering rotations of ±π/2 around the x3 axis, we would find w111 =
w222 = 0. This procedure can be applied to other values of the indices and shows
that the indices of a non vanishing wi jl must all be different. On the contrary, by the
same means, we can prove that wil = 0 if i
= l and that w11 = w22 = w33 . If η̃ is
their common value, we have
1
η̃ = w11 = w22 = w33 = (w11 + w22 + w33 ).
3
Taking into account the above properties, we obtain for the components of the
electromotive force E:
∂ B02 ∂ B03
c E1 = (w213 − w312 )B01 − η̃( − )
∂x3 ∂x2
= (w213 − w312 )B01 − η̃(∇ × B0 )1
c E2 = (w321 − w123 )B02 − η̃(∇ × B0 )2
c E3 = (w132 − w231 )B03 − η̃(∇ × B0 )3
Exploiting again the invariance properties of wi jk , it is easy to show that the

coefficients of B0i in the preceding equations are all equal. Denoting by α their
common value, we obtain
α = w213 − w312 = w321 − w123 = w132 − w231

1
= (w213 + w321 + w132 − w312 − w123 − w231 ).
3
We have thus shown that the electromotive force E can be written as
c E = αB0 − η̃(∇ × B0 ), (11.15)
recovering the phenomenological term introduced by Parker with the addition of an

extra term proportional to ∇ × B0 . The effect of the term proportional to η̃ is easily
appreciated if we introduce the expression for u × b , as given by Eq. (11.15), into
the equation for the mean magnetic field (11.10) that becomes
∂ B0
= α(∇ × B0 ) + (η + η̃)∇ 2 B 0 , (11.16)
∂t
showing that the presence of turbulence has increased the magnetic diffusivity.
Writing the expressions above, we have implicitly assumed that both α and η̃
are constants, when they should in general be functions of r and t. In particular
we know that α has different signs in the two hemispheres (see the comment after
Eq. (11.6)): therefore the term containing α in Eq. (11.16) should read ∇ × (αB0 ).
Moreover, since in general neither η nor η̃ are space-independent, an extra term
−∇(η + η̃) × (∇ × B0 ) should be added to the rhs of Eq. (11.16): our conclusions
concerning the effect of turbulence thus remain unaltered.
Introducing the explicit expressions of the wi jk and wik in the definitions of α
and η̃, we obtain:
∞
1
α=− u(r, t) · [∇ × u(r, t − τ )] dτ , (11.17)
3
0
and
∞
1
η̃ = u(r, t) · u(r, t − τ ) dτ . (11.18)
3
0
An order of magnitude estimate of α and η̃ gives:
1 1
α − h τcorr and η̃ u 2 τcorr .
3 3
Summarizing, the mean field theory, applied to an axisymmetric configuration,
results in the following form of the basic dynamo equations, Eqs. (11.1) and (11.2):
∂A 1 1
+ (U p · ∇)(r A) = αBθ + η0 (∇ 2 − 2 )A, (11.19)
∂t r r
∂ Bθ B 1
θ
+ r (U p · ∇) = r (B p · ∇Ω) + ∇ × (αB p ) + η0 (∇ 2 − 2 )Bθ , (11.20)
∂t r r
where for simplicity η and η̃ are assumed to be constant and η0 = η + η̃ is the total
diffusivity.
Equation (11.20) shows that a toroidal field can be generated from a poloidal
one not only by a nonuniform angular velocity (ω-effect), but also by the presence
of the term proportional to α (α-effect). On the other hand, the generation of a
poloidal field from a toroidal one, that completes the dynamo cycle, only depends
on α. Accordingly, we speak of α ω-dynamos when in the first phase the ω-effect
dominates and of α2 -dynamos when the α-effect dominates.
11.4.1 Simple Solutions of the Dynamo Equations
The dynamo equations are far too complex to allow a detailed analytical treatment,
even within the kinematic dynamo theory. Models with a reasonable degree of realism
can only be based on numerical simulations. However, the study of highly simplified
configurations can help to understand the general behavior of the solutions of the
dynamo equations. In the following we shall therefore limit ourselves to consider
solutions in simple geometries, but reminiscent of the more realistic ones encountered
in astrophysical contexts.
Slab Geometry
If we restrict ourselves to a thin spherical shell, the solution of the dynamo equa-
tions is considerably simplified in cartesian geometry by setting up a coordinate
system where the z-axis is directed along the local vertical and the y-axis is directed
eastwards, i.e. in the toroidal direction. In a right-handed coordinate system there-
fore, the x-axis points south. The large scale velocity field is assumed to be purely
toroidal U = U (z)e y with a vertical shear. Moreover the problem is considered to
be y-independent, ∂/∂ y = 0. The magnetic field can be represented as
∂A ∂A
B0 ≡ − , By , ,
∂z ∂x
with A = A(x, z)e y

In this configuration the dynamo equations are:
∂A
= αB y + η0 ∇ 2 A,
∂t (11.21)
∂ By ∂ A dU
= − α∇ 2 A + η0 ∇ 2 B y .
∂t ∂x dz
α is considered to be constant: the cases of α > 0 and α < 0 can be treated separately.
We look for solutions in the form of waves: A ∝ exp[i(k x x + k z z) + pt)], where
k x is assumed to be positive and p is in general complex, p = γ − iω, as one
could have anticipated from the fact that the system of dynamo equations (11.21)
is not self-adjoint. For simplicity assume a linear shear for U, U (z) = U z, with
U = const.
Start from the case of an α ω-dynamo and therefore neglect the α term in the
equation for B y . In this framework, the dispersion relation may be written:
( p + η0 k 2 )2 = ik x αU ,
or ik αU 1/2
p = (ik x αU )1/2 − η0 k 2 = −η0 k 2 1 −
x
.
η02 k 4
Since αU can be positive or negative we have:

ik αU 1/2 1 ± i |k x αU | 1/2
x
= √ = (1 ± i) |N D |),
η0 k 4
2
2 η0 k 4
2
where we have introduced the dynamo number, N D :
αU
ND = ,
2η02 k 3
and the upper sign refers to positive N D values. Notice that the properties of the
solution are not affected by the value of k z . Thus,

p = −η0 k 2 (1 − |N D |) ∓ i η0 k 2 |N D |
Growing solutions require that γ > 0, which implies |N D | > 1. The sign of N D
√ the direction of propagation of the waves. In fact, when N D > 0, ω =
determines
−η0 k 2 |N D | < 0,
|ω|
A ∝ exp ik x (x + t)
kx
and the wave therefore propagates in the negative x−direction (i.e. northward), while
for N D < 0, ω > 0 and the wave propagates southward.4 These wave-like solutions
of the dynamo equations have been called by Parker migrating dynamo waves.
We turn now to the case of α2 −dynamos and thus neglect the effect of the velocity
shear in the dynamo equations (11.21) by putting U = 0. The dispersion relation in
this case turns out to be:
p = ±kα − η0 k 2 .
Since p is real there are no migrating waves. We have growing or decaying solutions
according to whether
|α|
≷ 1,
η0 k
with the appropriate choice of sign. The limiting case |α| = η0 k produces a steady
state dynamo.
The interest of this slab model resides in its simplicity. Its relevance to real astro-
physical situations, typically requiring a spherical geometry, is marginal: for instance,
4 If k x < 0, the whole reasoning can be repeated with the same final results.
in the case of the Sun, a plane geometry α ω-dynamo could be only used to describe
a thin spherical shell. In spite of this, there are a few interesting consequences. As
already remarked, α changes sign across the equator: in the case of the Sun, α > 0
in the northern hemisphere. On the other hand, sunspots are known to form in both
hemispheres at about 45◦ of latitude and to migrate towards the equator during the
solar cycle. If we want to connect this observational fact to the properties of Parker’s
migrating dynamo waves, we must admit that the toroidal velocity must increase
with depth, making U < 0. Then N D ∝ αU < 0 in the northern hemisphere, while
N D > 0 in the southern one. In this situation, the waves migrate towards the equator
in both hemispheres, in agreement with the observed behavior of sunspots.
Another known quantity is the duration of the solar cycle requiring 22 years for
a complete restoration of the original poloidal field. If we tentatively identify the
cyclic behavior of the solar magnetic field with the constant-amplitude solution of
the planar model, |N D | = 1, we have |ω| = η0 k 2 . Since the dynamo action takes
place within the convection zone, whose depth is h 2 × 105 km, the vertical
wavelength must be less than 2h or k z π/ h. Let λ = 2π/k x be the horizontal
wavelength and assume λ h. Then
4π 2 π2 π2
k 2 = k x2 + k z2 + ,
λ2 h2 h2
or
2π 2h 2
T = ,
|ω| πη0
where T is the wave’s period. To reproduce the observed 22-years period η0 should
be 4 ×1011 cm2 s−1 . Since in the solar convection zone the molecular diffusivity,
η, is of the order 105 cm2 s−1 , we conclude that the dominant term in η0 must be
the turbulent diffusivity η̃. The next question is whether η̃ is compatible with the
required value of ≈1012 cm2 s−1 . Reasonable estimates based on the properties of
the solar turbulent convective eddies give values close to those quoted above. It seems
therefore that even a simplified dynamo model is able to reproduce the basic features
of solar activity. However, more realistic models might give a completely different
picture and the preceding discussion should be taken cautiously.
Cylindrical Geometry
This geometry is the relevant one to discuss magnetic field generation in the Galaxy,
where the field is mostly confined in the galactic disc, whose radius R 10 kpc is
much greater than its thickness h 300 pc, so that h/R 10−2 . This suggests that
the galactic disc can be treated as a thin disc, which implies that in cylindrical coor-
dinates the vertical gradients are much larger than the radial ones. As a consequence,
to lowest order, r -derivatives are dropped and only z-derivatives are kept. An order
of magnitude estimate of the condition ∇ · B = 0 implies that Bz /Br h/R 1,
which allows us to eliminate Bz . The dynamo equations therefore reduce to
∂ Br ∂ ∂ 2 Br
= − (αBθ ) + η0 , (11.22)
∂t ∂z ∂z 2
∂ Bθ ∂Ω ∂ ∂ 2 Bθ
= Br r + (αBθ ) + η0 . (11.23)
∂t ∂r ∂z ∂z 2
The total magnetic diffusivity, η0 , has been taken to be constant, while α = α(z), is
an unspecified odd function of z. Introducing the vector potential A = A(z)eθ , so
that
∂A
Br = − ,
∂z
and noticing that in the disc U = U eθ , with U = r Ω almost constant, we find that
(r ∂Ω/∂r ) = −Ω. Replace now Eq. (11.22) with
∂A ∂2 A
= (αB) + η0 2 ,
∂t ∂z
and Eq. (11.23) with
∂B ∂A ∂ ∂2 B
=Ω + (αB) + η0 2 ,
∂t ∂z ∂z ∂z
where, to simplify the notations, we have written Bθ ≡ B. Assuming B , A ∝ e pt ,

where, as before, p is in general a complex quantity, the following eigenvalue system
results:
∂2 A
p A = α B + η0 2 , (11.24)
∂z
∂A ∂ ∂2 B
pB = Ω + (αB) + η0 2 . (11.25)
∂z ∂z ∂z
It is worth noticing that the thin-disc approximation, that leads to the preceding
system of equations and controls the z–dependence of the field, does not eliminate
completely the r –dependence, since Ω, h and possibly α, still depend on the radial
variable. As a result, the eigenvalues, p, also depend on r : i.e. the growth rates change
with radius.5
If α is an antisymmetric function of z, the solutions of the above system of
equations can be classified according to their symmetry properties for z → −z, into
odd or dipole solutions :
5A discussion of this point, which is beyond the aim of this introductory book, can be found in
Zeldovich et al. (1983) [10].
B(z) = −B(−z), A(z) = A(−z),
or even or quadrupole solutions:
B(z) = B(−z), A(z) = −A(−z).
Obviously, we must supplement Eqs. (11.24) and (11.25) with the boundary con-
ditions in z = ±h. So-called vacuum boundary conditions that require
∂A
Bθ (±h) = B(±h) = 0 and Br (±h) = 0 or equivalently =0
∂z z=±h
are usually chosen. Since the galactic field seems to be even, the quadrupole solution
is the appropriate one.
In spite of the apparent simplicity, the reduced set of dynamo equations (11.24)
and (11.25) is complicated to solve analytically even with the simplest forms of α(z).
Examples of analytical solutions can be found e.g. in Parker (1979) [11] or in Moffat
(1978) [12].
In the following we shall limit ourselves to a few elementary cases, whose only
merit is to give some flavor of the more realistic numerical solutions. To simplify
the notation, we first introduce, as in Zeldovich et al. (1983) [10], the following
non-dimensional quantities:
z h2 α(ζ)
ζ= , τ= t, α̃(ζ) = with α0 ≡ αmax ,
h η0 α0
and
α0 h Ωh 2
Rα = , RΩ = .
η0 η0
The non-dimensional form of the dynamo equations reads:
∂A ∂2 A
= Rα [α̃h B)] + , (11.26)
∂τ ∂ζ 2
∂(h B) ∂A ∂ ∂ 2 (h B)
= RΩ + Rα [α̃(h B)] + , (11.27)
∂τ ∂ζ ∂ζ ∂ζ 2
with the boundary conditions:

∂A
B(±1) = 0, = 0.
∂ζ ±1
[The extra factor h is due to the fact that Br = −∂ A/∂z = −(1/ h)(∂ A/∂ζ)]
• α = Ω = 0: decay modes.
When no sources are present, the even solution of the system (11.26), (11.27) is
easily found by writing A(ζ, τ ) = a(ζ) f (τ ). Using this representation in (11.26),
we get
d f /dτ d2 a/dζ 2
= = −γ,
f a
or
A ∝ e−γτ sin (γ 1/2 ζ).
The boundary conditions in ζ = ±1 give
cos (γ 1/2 ) = 0 and therefore γ = (n + 21 )2 π 2 , n = 0, 1, 2, . . .
Clearly A = 0 is also a solution of Eq. (11.26). In the same way we find that
B = 0 or B ∝ e−γτ cos (γ 1/2 ζ), γ = (n + 21 )2 π 2 , n = 0, 1, 2, . . .
• α = 0, Ω
= 0: differential rotation only.
The equation for A is identical to that of the preceding example, therefore:
A = A0 e−γτ sin (γ 1/2 ζ), γ = (n + 21 )2 π 2 , n = 0, 1, 2, . . . .
The equation for B becomes:
∂B RΩ ∂ A ∂ 2 B γ 1/2 A0 RΩ −γτ ∂2 B
= + = e cos (γ 1/2
ζ) + ,
∂τ h ∂ζ ∂ζ 2 h ∂ζ 2
which suggests writing B as B(ζ, τ ) = b(τ ) cos (γ 1/2 ζ). Inserting this expression
in the preceding equation we find
db(τ )
+ γb(τ ) = γ 1/2 A0 RΩ e−γτ ,
dτ
a linear differential equation whose general solution is:
b(τ ) = e−γτ [b(0) + (γ 1/2 A0 RΩ / h)τ ].
If we consider a situation in which initially the azimuthal field vanishes, b(0) = 0,

then
B ≡ Bθ = (γ 1/2 A0 RΩ / h)τ e−γτ cos (γ 1/2 ζ).
On the other hand at τ = 0 the field is purely radial (neglecting the small Bz )
∂A
Br (0, ζ) = −(1/ h) = −(γ 1/2 A0 / h) cos (γ 1/2 ζ),
∂ζ
so that the quantity (γ 1/2 A0 / h) is simply the magnitude of the initial field at the
midplane of the disc, B0 = |B(0, 0)|. Finally,
B ≡ Bθ = B0 RΩ τ e−γτ cos (γ 1/2 ζ).
The azimuthal field increases linearly with time until the damping factor becomes
dominant, bringing both Br and Bθ to zero. The maximum value of the azimuthal
field, (B0 RΩ e−γ ), is reached at τ = 1.
• α
= 0, Ω = 0: α2 -dynamo.
In this case it is possible to show (see Zeldovich et al 1983) that the problem can
be reduced to the following equation:
∂2 B ∂ α̃B
− γ B = iRα .
∂ζ 2 ∂ζ
Writing B as

B = ψ exp 1
2 iRα αdζ ,
the preceding equation can be reduced to a Schrödinger equation with a generally

complex potential:
∂2ψ
+ (E − V )ψ = 0,
∂ζ 2
∂ α̃
E = −γ, V = 41 Rα2 α̃2 + 21 iRα .
∂ζ
If α̃ is chosen to be, as in Parker (1979), α̃ = 1 for ζ > 0 and α̃ = −1 for ζ < 0,

the eigenvalues γ turn out to be
γ = 41 Rα2 − k 2 ,
where k is given by the solutions of the equation
k cot k = 21 Rα .
Leaving aside the problem of actually solving the dynamo equations in realistic
geometries, it is important to estimate at least approximately the amount of field
growth that can be expected in the conditions typical of our Galaxy. To do so, we
shall completely neglect the boundary conditions and consider the system (11.24) and
(11.25), where we drop the α–term in the equation for B, since the galactic dynamo
seems to be of the αω type. If we look for solutions of the form A, B ∝ exp ( pt + ikz)
and assume that both α and Ω are constants and k > 0, the dispersion relation is
given by
( p + η0 k 2 )2 = ikαΩ,
namely
1 ± i√
p = −η0 k 2 + √ kαΩ,
2
where the positive sign refers to positive values of αΩ. Introducing the dynamo
number N D = |αΩ|/(2k 3 η02 ), we have

p = −η0 k 2 (1 − N D ) ± iη0 k 2 N D ,
which implies that growing solutions require N D > 1. Taking into account the
thickness of the galactic disc, h, which implies that k ≥ k0 = π/2h, we realize that
the critical value for growth is:
4|αΩ|h 3
≥ N D > 1.
π 3 η02
Let’s now evaluate N D for our Galaxy. The observational data are the following:
h = 300 pc 9 × 1020 cm, Ω = 9 × 10−16 s−1 , the average turbulent velocity in
the interstellar medium, ū = u 2 106 cm s−1 . The typical collision time, τcorr ,
is of the order of 107 y 3.16 × 1014 s and the correlation length, λcorr = ū τcorr
3.16 × 1020 cm. Using these data to estimate the values of α and η̃ we obtain (see
Sect. 11.4):
1
η̃ ū 2 τcorr 1.0 × 1026
3
1 1 ū 2 1
α − h τcorr τcorr ū 3.3 × 105
3 3 λcorr 3
Since in the interstellar gas the dominant diffusivity is the turbulent one, we put
η0 η̃ and we obtain:
4|αΩ|h 3
2.6,
π 3 η̃ 2
slightly above the critical value of 1. With these values the growth rate turns out to
be: π 2
p −η̃ (1 − N D ) 1.9 × 10−16 s−1 ,
2h
and the amplification factor is e p t e6 t , with t expressed in units of a billion years.

During a period of 3 billion years, thus shorter than the lifetime of the Galaxy, that
factor would be e18 6.6 × 107 , which means that an initial field of the order of
2 × 10−14 G would be amplified to the actual value of ≈ 10−6 G.
11.5 The Creation of Magnetic Fields
Let’s go back to the remark made in the first Section of this Chapter: we cannot be
sure that primordial seed fields existed immediately after the Big Bang. If such a
primordial field was not there, how is it possible to generate a field different from
zero in subsequent epochs?
At first sight this seems to be an impossible task in view of the evolution equation
for B, Eq. (11.1):
∂B
= ∇ × (U × B) + η∇ 2 B.
∂t
If the diffusivity, η, is identically zero (ideal plasma), Alfvén’s theorem implies
the strict conservation of flux and a field that is identically zero at the beginning will
remain so for all times. But, even if η
= 0, B = 0 is a solution to the evolution
equation at all times compatible with the initial null condition and, since the solution
is unique, no magnetic field can be generated, independently of the form of U.
However, this particular form of the evolution equation is not the most general
one. To obtain it, we used an approximate form of the generalized Ohm equation,
Eq. (5.5),
1 Ji m e ∂ Ji ∂
E i + (U × B)i − = 2 + Ji Uk + Jk Ui
c σ e n e ∂t ∂xk
1 1 ∂Pik
(e)
+ J×B − ,
en e c i en e ∂xk
dropping all the rhs terms. The conditions under which this is allowed are specified
in Eqs. (5.6)–(5.9). When B = 0, ωcp = 0 and the condition expressed by (5.9) is
certainly violated, so that the corresponding term in the generalized Ohm equation
must be included. Thus, the approximate form of the Ohm equation to be used is
1 1
E+ U×B=− ∇ Pe ,
c en e
where the usual resistive term has been dropped.

Taking the curl of the preceding equation and eliminating the electric field by
using
1 ∂B
∇×E=− ,
c ∂t
we get:
∂B c ∇P
e
= ∇ × (U × B) + ∇ × ,
∂t e ne
namely
∂B c ∇n e × ∇ Pe
= ∇ × (U × B) − . (11.28)
∂t e n 2e
The last term in this equation is known as the Biermann battery term. Since it does
not depend on B, it can act as a source term, even when B = 0 at t = 0. In spite of
its promising look, however, it does not solve the problem, unless some additional
conditions are imposed. In fact, Pe is usually considered to be a unique function of
n e (barotropic assumption), so that ∇ Pe is parallel to ∇n e and the Biermann term
vanishes. To avoid this, some mechanism has to be found that breaks the barotropic
assumption. This could be provided by the propagation of shocks of limited spa-
tial extent, not an unlikely event in a turbulent medium like the one present in the
primordial Universe or in a protogalaxy. The basic idea is the following. If a shock
propagates in an essentially pressureless medium, behind the shock the density sud-
denly increases to relax to the the pre-shock value after a certain time. The pressure
follows the same path, but when n e has returned to the initial value, Pe
= 0 since the
gas has been heated by the shock. However, in the regions unaffected by the shock
the pressure is still zero, so that to the same value of density correspond different
values of the pressure. Pe therefore is not a single-valued function of n e and so the
Biermann term does not vanish.
The growth of seed magnetic fields from an initial field-free situation based on
Eq. (11.28) has been the object of numerical investigations (Kulsrud et al. 1997) [28]
that prove that the Biermann battery could indeed create seed fields of the order of
10−21 G prior to the formation of galaxies, when it can be assumed that the plasma
is completely ionized.
Once the seed field has been created, it can be amplified by the Biermann battery,
even if the plasma is only partially ionized. To see this, let’s write n e = n i =
χ(n i + n n ), where n n is the density of neutrals. If the temperature has the same value
for the three plasma components then
Pe ne ne ne m p
= = = .
P ne + ni + nn (1 + χ)(n i + n n ) (1 + χ)ρ
11.5 The Creation of Magnetic Fields 273
We now express Pe in terms of P by using the above equation, multiply Eq. (11.28)
by e/m p c, introduce the vector Ω c = e B/m p c and finally obtain
∂Ω c ∇ρ × ∇ P
= ∇ × (U × Ω c ) − , (11.29)
∂t (1 + χ)ρ2
where it has been assumed that χ is constant. If we compare this equation with the
well-known equation for the fluid vorticity ω = ∇ × U:
∂ω ∇ρ × ∇ P
= ∇ × (U × ω) + ,
∂t ρ2
we see the two equations are identical, apart from the factor −1/(1 + χ). This shows
that, for identical initial conditions,
ω
Ωc = − ,
1+χ
everywhere and for all times, a relationship confirmed by numerical simulations that
also prove that, starting from an initial field ≈ 10−21 G the Biermann battery is
capable of amplifying the field up to ≈ 10−18 G. Once this stage is reached, other
mechanisms, like those discussed earlier, can boost the field to the present value of
≈ 10−6 G.
References
The following short list includes some textbooks that may be useful to integrate
our treatment. We have also indicated a selected list of original papers dealing with
astrophysical application discussed in the text.
1. T. J. M. Boyd and Sanderson, J.J., 2003, The Physics of Plasmas, Cambridge University Press,
Cambridge, UK.
2. Goedbloed, H., Poedts, S., 2004, Principles of Magnetohydrodynamics, Cambridge University
Press, Cambridge, UK.
3. Gurnett, D. A., Battacharjee, A., 2005, Introduction to Plasma Physics, Cambridge University
Press, Cambridge, UK.
4. Spitzer, L., 1962, Physics of Fully Ionized Gases, Interscience, New York, USA.
5. Kulsrud, R. M., 2004, Plasma Physics for Astrophysics, Princeton University Press, Princeton,
USA
6. Celnikier, L. M., 1989, Basics of Cosmic Structures, Ed. Frontières, Gif-sur-Yvette, France.
7. Landau, L. D., Lifshitz, E. M., 1987, A Course in Theoretical Physics, Vol. 6: Fluid Mechanics,
Pergamon Press, Oxford, UK.
8. Lifshitz, E. M., Pitaevskii, L.P., 2002, A Course in Theoretical Physics, Vol.10: Physical Ki-
netics, Pergamon Press, Oxford, UK.
9. Frisch, U., 1995, Turbulence: The Legacy of A. N. Kolmogorov, Cambridge University Press,
Cambridge, UK.
10. Zeldovich, A.B., Ruzmaikin, A.A. and Sokoloff, D.D., 1983, Magnetic Fields in Astrophysics,
Gordon and Breach, New York, USA
11. Parker, E.N., 1979, Cosmical Magnetic Fields, Glarendon Press, Oxford, UK
12. Moffat, H.K., 1978, Magnetic field Generation in Electrically Conducting Fluids, Cambridge
University Press, Cambridge, UK.
13. Krause, F. and Rädler, K.-H., 1980, Mean-field Electrodynamics and Dynamo Theory, Perga-
mon Press, New York.
14. Jeans, J. H., 1915, On the theory of star-streaming and the structure of the universe, Monthly
Notices of the Royal Astronomical Society76, 70–84.
15. Chandrasekhar, S., 1942, Principles of Stellar Dynamics, University of Chicago Press, Chicago.

UNITEXT for Physics, DOI 10.1007/978-88-470-5280-2
276 References
Original Papers
16. Chew, G.F., Goldberger, M. L., Low, F. E., 1956, The Boltzmann Equation and the One-Fluid
Hydromagnetic Equations in the Absence of Particle Collsions, Proceedings Royal Society
London, A236, 112.
17. Parker, E. N., 1966, The Dynamical State of the Interstellar Gas and Field, Astrophysical
Journal, 145, 811, IoP Publishing Ltd., Philadelphia, USA.
18. Chiuderi, C., Giachetti, R., Van Hoven G., 1977, The Structure of Coronal Magnetic Loops. I
- Equilibrium Theory, Solar Physics, 54, 107, D. Reidel Publishing Co., Dordrecht, Holland.
19. Giachetti, R., Van Hoven G, Chiuderi, C., 1977, The Structure of Coronal Magnetic Loops. II -
MHF Stability Theory, Solar Physics, 55, 371, D. Reidel Publishing Co., Dordrecht, Holland.
20. Burgess, D., Scholer, M., 2013, Microphysics of Quasi-parallel Shocks in Collisionless Plas-
mas, Space Sci Rev 178, 513–533, Springer Science, Dordrecht, Holland.
21. Krasnoselskikh, V., Balikhin, M. et al., 2013, The Dynamic Quasiperpendicular Shock: Cluster
Discoveries, Space Sci Rev 178, 535–598, Springer Science, Dordrecht, Holland.
22. Furth, H. P., Killeen, J., Rosenbluth, M. N., 1963, Finite-Resistivity Instabilities of a Sheet
Pinch, Physics of Fluids, 6, 459, American Institute of Physics, Melville, USA.
23. Balbus, S. A., Hawley, J. F., 1991, A Powerful Local Shear Instability in Weakly Magnetized
Disks, Astrophysical Journal, 376, 214, IoP Publishing Ltd., Philadelphia, USA.
24. Loureiro, N.F., Schekochihin, A.A., Cowley, S.C., 2007, Instability of Current Sheets and
Formation of Plasmoid Chains, Physics of Plasmas 14, 100703, American Institute of Physics,
USA.
25. Pucci, F. and Velli, M., 2013, Reconnection of quasi-singular current sheets: the “ideal” tearing
mode, Astrophysical Journal, 780, L19, IoP Publishing Ltd., Philadelphia, USA.
26. Parker, E.N., 1972, Topological Dissipation and the Small-Scale Fields in Turbulent Gases,
Astrophysical Journal, 174, 499, IoP Publishing Ltd., Philadelphia, USA.
27. Cowling, T.G., 1934, The stability of Gaseous stars, Montly Notices of Royal Astronomical
Society, 94, 39, Wiley & Sons Inc.
28. Bullard, E.C., 1949, The Magnetic Field within the Earth, Proceedings Royal Society London,
A199, 433.
29. Parker, E.N., 1955, Hydromagnetic Dynamo Models, Astrophysical Journal, 121, 491, IoP
Publishing Ltd., USA.
30. Kulsrud, R.M and Anderson, S.W., 1997, The spectrum of random magnetic fields in the mean
field dynamo. Theory of the Galactic magnetic field, Astrophysical Journal, 396, 606, IoP
Publishing Ltd., USA.
31. Vlaosv, A.A., 1938, On Vibration Properties of Electron Gas, J Exp Theor Phys 8(3), 291. (in
Russian).
Index
Symbols D
β parameter, 78, 85 De Hoffman-Teller frame, 193
Debye length, 6, 7, 9
Developed turbulence, 229
A Dispersion relation, 108, 153
Adiabatic invariants Dissipative scale, 232
longitudinal, 32 Distribution function, 38
magnetic moment, 28 general moment equation, 44
Alfvén average values, 42
speed, 13 moments, 42
theorem, 82 Drift motions
waves, 157 curvature drift, 24
Alfvén waves, 157 gradient drift, 26
Anti-dynamo theorems, 249 Dynamo, 247
Axial vectors, 259 α ω-dynamos, 262
α 2 -dynamos, 262
Parker’s α-term, 257
B Biermann battery, 272
Bennet’s relation, 94 cyclonic motions, 256
Biermann battery, 272 cylindrical geometry, 265
Boltzmann equation, 40 dynamo number, 264, 270
Bouncing frequency, 177 electromotive force, 257, 260, 261
galactic dynamo, 266
generation of poloidal fields, 255, 257
C
generation of toroidal fields , 253, 254
Cold plasma model, 64
kinematic dynamo, 249
Collision frequency, 2, 10, 12
mean-field electrodynamics, 258
Collisional plasma, 65
migrating dynamo waves, 264
Compressible Alfvén waves, 158
slab geometry, 263
Contact discontinuity, 186
Coriolis force, 256
Corona solare, 239
buchi coronali, 240 E
riscaldamento della, 240 Electrical conductivity, 3, 64, 67
Coulomb logarithm, 10 Electromagnetic waves, 166
Cowling’s theorem, 251 Electron cyclotron waves, 164
Cyclotron frequency, 13, 18 Euler equation, 53
UNITEXT for Physics, DOI 10.1007/978-88-470-5280-2
278 Index
F Kinetic equation, 40
Fermi acceleration, 32 Kolmogorov spectrum, 234
Fluid models Kolmogorov-Obukhov law, 234
neutral gases, 49 Kruskal-Shafranov instability, 109
one-fluid model, 59
two-fluid models, 56
closure problem, 50 L
Fokker-Planck equation, 40 Lagrangian displacement, 101
Force-free fields Landau damping, 175
cylindrical geometry, 89 Landau prescription, 173
equilibria, 87 Langmuir waves, 165
plane geometry, 88 Laplace transform, 169
Woltjer’s theorem, 90 Larmor frequency, 13
Lundquist number, 81, 204
G
Generalized Ohm equation, 64 M
Group velocity, 154 Magnetic bottle, 30
Magnetic diffusivity, 75
Magnetic field
H galactic, 247, 265
Hall term, 74, 161 planetary, 247
Heat flux, 52 poloidal-toroidal decomposition, 251
Helicity, 258 solar, 247
Helioseismology, 253, 254 terrestrial, 247, 253
Hydrodynamic shocks, 190 Magnetic mirror, 30
Hydrodynamical turbulence, 231 Magnetic pressure, 77
Magnetic Reynolds number, 81
Magnetic tension, 78
I Magnetosonic waves, 159
Ideal plasma, 75 Mean-field electrodynamics, 258
Instabilità MHD
magnetorotazionale, 143 equations in conservative form, 79
Instabilities Alfvén’s theorem, 82
in cylindrical geometry, 130 frozen-in condition, 84
θ-pinch, m = 0, 130 magnetic diffusion, 81
θ-pinch, m = 1, 130 resistive equations, 76
coronal loops, 130 MHD regime, 71
Instability Mirror ratio, 31
Kelvin-Helmholtz, 122
Kruskal-Shafranov, 109
Parker, 111 N
Rayleigh-Taylor, 106 Navier-Stokes equation, 53, 229
Ion cyclotron waves, 164 Nonlinear cascade, 231
Normal modes, 153
Normal modes method, 95
J
Jeans mass, 121
Jeans’ theorem O
for stellar systems, 47 Oblique shocks, 193, 195
Jump conditions, 183, 186
P
K Parallel shocks, 192
Kelvin-Helmholtz instability, 122 Parker instability, 111
Index 279
Particle motions T
electromagnetic wave, 21 Tangential discontinuity, 186
constant magnetic field, 17 Theta-pinch, 92, 133
crossed E, B fields, 19 Turbulence
Perpendicular shocks, 189 pseudo-scalar, 259
Phase velocity, 152, 154 turbulent diffusivity, 262
Pitch angle, 30 Turbulent viscosity, 233
Plasma frequency, 9
Plasma waves, 165
Polar vectors, 259 V
Pressure tensor, 50 Virial theorem, 87
Pseudo-scalars, 259 Vlasov equation, 40, 44, 167
Vorticity, 231
R W
Rayleigh-Taylor instability, 106 Waves
Resistive plasma, 76 Alfvén, 157
Rotational discontinuities, 187 compressible Alfvén, 158
Rotational discontinuity, 186 electromagnetic, 166
electron cyclotron, 164
Fourier representation, 150
general wave equation, 156
S ioncyclotron, 164
Saha equation, 4 Langmuir, 165
Shock front, 181 magnetosonic, 159
Shocks, 181 whistler, 164
Solar wind, 41 Woltjer’s theorem, 90
Sound speed, 101
Suydam criterion, 142
Switch-off shock., 197 Z
Switch-on shock, 197 Zeta-pinch, 92

(UNITEXT For Physics) Claudio Chiuderi, Marco Velli (Auth.) - Basics of Plasma Astrophysics-Springer-Verlag Mailand (2015) PDF

Uploaded by

Copyright:

Available Formats

(UNITEXT For Physics) Claudio Chiuderi, Marco Velli (Auth.) - Basics of Plasma Astrophysics-Springer-Verlag Mailand (2015) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(UNITEXT For Physics) Claudio Chiuderi, Marco Velli (Auth.) - Basics of Plasma Astrophysics-Springer-Verlag Mailand (2015) PDF

Uploaded by

Copyright:

Available Formats

UNITEXT for Physics

ISSN 2198-7882 ISSN 2198-7890 (electronic)

Library of Congress Control Number: 2014947666

Springer Milan Heidelberg New York Dordrecht London

© Springer-Verlag Italia 2015

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Florence, December 2013 Claudio Chiuderi

1 An Introduction to Plasma Physics . . . . . . . . . . . . . . . . . . . . . . . 1

2 Particle Orbit Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Kinetic Theory of Plasmas: An Outline . . . . . . . . . . . . . . . . . . . . 37

4 Fluid Models . . . . . . . . . . . . . . . ...... . . . . . . . . . . . . . . . . . . 49

5.2 The Time Evolution of Magnetic Fields. . . . . . . . . . . . . . . . . 80

9 Magnetic Reconnection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

10 MHD Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

11 Build-Up of Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

Abstract A brief introduction to Plasma Physics is presented, focusing on ion-

Plasma physics studies the equilibrium and dynamics of globally neutral

© Springer-Verlag Italia 2015 1

In terms of the degree of ionization, χ ,

and recalling that n e = Z n i , where Z e is the ion charge,

Here σmax is the maximum conductivity at complete ionization of the plasma, χ = 1.

1.1 Saha’s Equation

and Saha’s equation follows

Fig. 1.1 χ as a function of T X

Introducing the degree of ionization:

one finds from Saha’s equation that

1.2 The Debye Length

∇ · E = −∇ 2 Φ = 4πq = 4π e(n i − n e ) + 4π e0 δ(r),

while δ(r) denotes the Dirac δ-function.

n i = n 0 exp[−(eΦ/kT )]. (1.4)

The equation for the electrostatic potential Φ therefore becomes:

∇ 2 Φ = 4π n 0 e [exp(eΦ/kT ) − exp(−eΦ/kT )] − 4π e0 δ(r)

∇ 2 Φ = 8π n 0 e (eΦ/kT ) − 4π e0 δ(r) (1.6)

To integrate, observe that the preceding equation may be rewritten as:

Fig. 1.2 The Coulomb

Fig. 1.3 The total charge Q

This inequality must be satisfied for the approximation of quasi-neutrality to hold

1.3 Fundamental Plasma Parameters

and the collision frequency:

The final correct expression for electron-electron collisions is given by:

The analagous expression for two identical ions of charge Z e is:

where the final numerical expression refers to the proton-proton case.

where however we have approximated Λee Λii .

while for ions colliding on target electrons

For an electron-proton plasma τei  τii  τee and electron-proton thermalization

from Eq. (1.11).

and quantum-mechanical effects become important when:

1.4 The Classical Description of Plasmas

The classical, microscopic, description of a plasma considers the plasma as an ensem-

understanding of particle orbit theory is however a pre-requisite for developing a

Questions and Problems

Solving for ve , vi gives

Abstract In which particle motion in assigned electric and magnetic fields is

This chapter is devoted to a description of the motion of individual charged

2.1 Motion in a Uniform, Static Magnetic Field

© Springer-Verlag Italia 2015 17

implying that the total kinetic energy energy

is constant. However, since W = 21 mv2 is separately conserved, we find that the

For an electron-proton plasma τei τii τee and electron-proton thermalization

B = [Br (r, z), 0, Bz (r, z)]; Bz Br → B = (Br2 + Bz2 )1/2 Bz .

where M is the solar mass.