Lecture notes

NAST008: Cosmic electrodynamics

S
S

February 6, 2017
2
Contents

1 Basics of statistical physics 11

1.1 Liouvilles theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Statistical description of a physical system . . . . . . . . . . . . . . . . . 12
1.3 Reduced distribution functions . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Boltzmann equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.1 BBGKY hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Vlasov approximation . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.3 Fokker-Planck approximation . . . . . . . . . . . . . . . . . . . . 18
1.5 Fluid equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.1 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.2 Euler equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.3 Energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.4 Closing the equations set . . . . . . . . . . . . . . . . . . . . . . 20
1.6 Thermodynamical equilibria . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6.1 Statistical ensembles . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.2 Maxwell-Boltzmann distribution . . . . . . . . . . . . . . . . . . 24
1.6.3 Chemical equilibrium, Sahas equation . . . . . . . . . . . . . . . 27

2 Plasma 29

3 Charged particle motion in electromagnetic fields 33

3.1 Homogeneous magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Homogeneous electric field . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Relativistic E-B drift . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.2 Drift in the field of general force . . . . . . . . . . . . . . . . . . 41
3.3 Inhomogeneous magnetic field . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Inhomogeneous electric field . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5 Drift currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 Guiding centre motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.7 Magnetic mirrors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.8 Adiabatic invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.8.1 First adiabatic invariant . . . . . . . . . . . . . . . . . . . . . . . 63
3.8.2 Second adiabatic invariant . . . . . . . . . . . . . . . . . . . . . . 64

3
CONTENTS 4

3.8.3 Third adiabatic invariant . . . . . . . . . . . . . . . . . . . . . . 66

3.9 van Allen radiation belts . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4 Plasma as fluid 71
4.1 Double-fluid approximation in physics of plasmas . . . . . . . . . . . . . 71
4.1.1 Plasma approximation . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2 Highly ionised plasmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3 Single-fluid model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4 Approximations to plasma fluid . . . . . . . . . . . . . . . . . . . . . . . 77
4.4.1 Force-free and ideal MHD . . . . . . . . . . . . . . . . . . . . . . 78
4.4.2 Axisymmetrical systems, stream functions . . . . . . . . . . . . . 79
4.5 Drifts in the plasma fluid . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.5.1 Drifts perpendicular to B . . . . . . . . . . . . . . . . . . . . . . 82
4.5.2 Drifts parallel to B . . . . . . . . . . . . . . . . . . . . . . . . . . 82

5 Waves in plasmas 85
5.1 Linear waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2 Example: Acoustic waves in fluids . . . . . . . . . . . . . . . . . . . . . 86
5.3 Types of plasma waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.4 Plasma oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.5 Electron plasma wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.6 Ion waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.7 Electrostatic waves in background magnetic field . . . . . . . . . . . . . 97
5.7.1 Electron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.7.2 Ion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.8 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.8.1 In vacuum: Light waves . . . . . . . . . . . . . . . . . . . . . . . 102
5.8.2 In plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.8.3 Complex of electromagnetic waves in presence of B 0 . . . . . . . 106
5.9 MHD waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.10 Alfven wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.11 General MHD waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.11.1 Mode conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

6 Diusion 121
6.1 Diusion in weakly ionised plasmas . . . . . . . . . . . . . . . . . . . . . 121
6.2 Diusion in highly ionised plasmas . . . . . . . . . . . . . . . . . . . . . 125
6.3 Diusion in the magnetic field . . . . . . . . . . . . . . . . . . . . . . . . 126

7 Equilibrium and stability 129

7.1 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.1.1 Diusion of the magnetic field into plasmas . . . . . . . . . . . . 131
7.2 Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.2.1 Two-stream instability . . . . . . . . . . . . . . . . . . . . . . . . 133
5 CONTENTS

7.2.2 Gravitational instability . . . . . . . . . . . . . . . . . . . . . . . 135

7.2.3 Universal instability . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2.4 Landau damping . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

Appendices

Appendix A Guiding centre motion 149

A.1 Alternative derivation of (3.133) . . . . . . . . . . . . . . . . . . . . . . 149
A.2 Derivation of equations (3.137)(3.139) . . . . . . . . . . . . . . . . . . . 150
CONTENTS 6
ToDos

Liouville theorem

Eqs. (3.137) to (3.139) to be derived in the appendix.

Section 4.4

Relativistic E-B drift in the Appendix fully including the intermediate steps.

7
CONTENTS 8
Preface

Blah
Korektury: Jan Kara, L
ydia Stofanov
a

Literature
T. Padmanabhan: Theoretical astrophysics I, Cambridge University Press (2000),
ISBN-13: 978-0521566322

F. Shu: Physics of Astrophysics, vol. II, Gas dynamics, University Science Books
(1992), ISBN-13: 978-0935702651

J. Binney & S. Tremaine: Galactic dynamics, Princeton University Press (2008),

ISBN-13: 978-0691130279

F. F. Chen: Introduction to Plasma Physics and Controlled Fusion, Springer

(2010), ISBN-13: 978-1441932013

P. Kulhanek: Teorie plazmatu, studijn text pro FJFI CVUT

(2009), online
http://www.aldebaran.cz/studium/fpla.pdf

9
CONTENTS 10
Chapter 1

Basics of statistical physics

1.1 Liouvilles theorem

. . . We will show that volume of the set t , which equals t d6N q is equal to volume of
0 . At the first step, let us write

6N
d q= d6N q t , (1.1)
t t

which can be viewed just as trivial renaming of the integration variables. Next, let
us consider the theorem about substitution of variables (find out proper mathematical
terminology), which states:
1) Let be a regular one-to-one mapping from RN to RN defined on an open set
G RN ;
2) let M (G) RN ;
3) let f be a mapping RN R1 defined (skoro vsude) on M .
Then,  
 D 
f (y)dy = 
f ((x)) det  dx . (1.2)
M 1 (M ) Dx 

We will apply (1.2) to the right-hand side of equation (1.1) with f 1, y q t , x

q, (x) q t (q), M = t and 1 (M ) = 0 , which yields
 
 Dq t 
6N
d q= 6N t
d q =  
det Dq  dq . (1.3)
t t 0

Let us further denote  

 Dq t 
Jt det  , (1.4)
Dq 
which is the Jacobian of the transformation q q t . Apparently, the Liouvilles theorem
is valid provided Jt = 1. In practise, we will show that dJt /dt = 0 which, together with
trivial identity J0 Jt=0 = 1, implies the required property of the Jacobian. Let

11
1.2. STATISTICAL DESCRIPTION OF A PHYSICAL SYSTEM 12

us denote element of the Jacobi matrix Jij qit /qj . Its time derivative may be
expressed as
dJij qt qit qkt
= i = , (1.5)
dt qj qkt qj
K

where we denote
qit dqit /dt.
In equation 1.5 we have exchanged order of derivatives
with respect to t and qj which is possible provided these two are continuous. Time
derivative of the Jacobian Jt can be written with help of eq. 1.5:

dJ J dJij J qt
= = Jkj ti . (1.6)
dt Jij dt Jij qk
i,j i,j,k

Equation (1.6) can be simplified due to identity (see Box ??)

J
Jkj = ik J (1.7)
Jij
j

to
dJ qt
i
=J . (1.8)
dt qit
i

Finally, let us rewrite (1.8) in terms of canonical coordinates and momenta and consider
Hamiltons equations to evaluate rit and pti :

3N ( ) 3N (
)
dJ rit pti 2H 2H
=J + =J t t (1.9)
dt rit pti rit pti pi ri
i=1 i=1

1.2 Statistical description of a physical system

Probability density is constant along each state evolutionary track through the phase
space, i.e. each particular state keeps in time its probability of being realised (found /
measured). This, together with an assumptions required for the Liouvilles theorem to
be valid (i.e. Hamiltonian nature of the system) allows to formulate a Lioville equation
for temporal evolution of the distribution function DN .
Lets assume an arbitarary fixed subset, , of the phase space. Probability to find
the system in some state in is P () = DN d. Liouvilles theorem states that
the fluid of states is incompressible, hence, the only way how the probability P () can
change is due to flow of the states through the boundary of , i.e.:
I

DN d = DN q d (1.10)
t

where d is an (outward) oriented element of the surface . The minus sign means
that the probability decreases when the state flows parallel to d, i.e. out from .
Provided some physically plausible mathematical conditions are fulfilled (provide an
exact formulation of these conditions), we may exchange derivative and integration
13 1.3. REDUCED DISTRIBUTION FUNCTIONS

operators on the left-hand side of eq. (1.10); on the right-hand side, we apply Gauss
theorem, yielding
DN
d = (DN q) d . (1.11)
t
Integral equality (1.11) holds for an arbitary set which implies equality of integrands.
Therefore, we obtain a continuity equation for the distribution function,
DN
= (DN q)
, (1.12)
t
which can be further rewritten as
6N [ ] 3N [
( )]
DN DN qi DN H DN H H H
= qi + DN = + DN .
t qi qi ri pi pi ri ri pi pi ri
i=1 i=1
(1.13)
The terms in () brackets cancel out provided H is continuously dierentiable with
respect to ri and pi . Using the Poisson brackets notation, the Liouville equation is
usually written as:
DN
= {H, DN }P . (1.14)
t
(Mention the analogous von Neumann equation from the quantum theory.)

1.3 Reduced distribution functions

Distribution function, DN holds a maximum information of a particular (usually very
complex) physical system. This maximum information may often be more than what we
need. In some circumstances, it is convenient to define reduced distribution functions:

f (r, p) DN (r, p, r 2 , p2 , . . . , r N , pN ) dN1 , (1.15)

where
d3 r 2 d3 p2 . . . d3 r N d3 pN
dN1 (1.16)
(N 1)!
and the integration is taken over the whole definition space of variables r 2 . . . pN . Def-
inition (1.15) generally gives dierent results when dierent pairs (r i , pi ) are excluded
from integration (which means that the result should keep some mark according to the
excluded particle). The definition (1.15), however, is usually considered for specific sys-
tems of undistinguishable particles which imply that DN as well as H are symmetric
with respect to permutations of individual particles (exchange of pairs of (r i , pi ) and
(r j , pj )). For these systems, integration (1.15) gives identical functions regardless of
over which N 1 pairs of coordinates and momenta it is performed. Function f (r, p) is
usually called single-particle distribution function and it gives a probability density of
finding any particle at position r with momentum p. Note that the single-particle distri-
bution function really holds considerably reduced information with respect to DN . We
1.3. REDUCED DISTRIBUTION FUNCTIONS 14

could add that to the meaning of f (r, p) that . . . and we have no information about po-
sitions and momenta of the other particles. Single-particle distribution function holds
sucient amount of information to calculate mean values of physical quantities which
may be written in a form:

N
A(r 1 , p1 , . . . , r N , pN ) = a(r i , pi ) , (1.17)
i=1

i.e. it is a sum of N identical functions. The mean value is then

A A DN dN = a(r, p) f (r, p) dr dp . (1.18)

Similarly to (1.15) we may define two-particle distribution function,

f2 (r 1 , p1 , r 2 , p2 ) DN dN2 , (1.19)

which gives the probability density of finding some particle at position r 1 with mo-
mentum p1 and any other particle at position r 2 with momentum p2 . Mean values
of physical quantities which depend on combinations of positions and momenta of two
particles,
B(r 1 , p1 , . . . , r N , pN ) = b(r i , pi , r j , pj ) , (1.20)
i,j

can be calculated through integration of the two-particle distribution function:

B = b(r 1 , p1 , r 2 , p2 ) f2 (r 1 , p1 , r 2 , p2 ) dr 1 dp1 dr 2 dp2 . (1.21)

(Be more precise in definition (1.19) in terms of limits of i and j which is related to
presence (or not) of factor 12 in (1.21).)
An important example of physical quantity which often takes the form of (1.17)
or (1.20) or combination of both is total energy of the system which is defined by
Hamiltonian. Lets introduce its specific (i.e. not general!) form:

N
p2i
N N
N
H= + V (r i ) + w(r i , r j ) . (1.22)
2m
i=1 i=1 j=1 i=1
j=i

Here, first sum corresponds to system of free particles; first plus second sum defines a
system of non-interacting particles (of equal mass, m, which is an inevitable consequence
of their undistinguishability) in external potential V (r). The third term is a possible
form of interaction term. The mean value of energy of the statistically described system
determined by Hamiltonian (1.22) is

H = h(r, p) f (r, p) d3 r d3 p + w(r 1 , r 2 ) f2 (r 1 , p1 , r 2 , p2 ) d3 r 1 d3 p1 d3 r 2 d3 p2 ,
(1.23)
15 1.4. BOLTZMANN EQUATION

where we denote
p2
h(r, p) + V (r) . (1.24)
2m
In the following, we will use a particular example of the interaction term which
describes mutual gravitational interaction among particles:

Gm2
w(r i , r j ) = (1.25)
|r i r j |

with G representing the gravitational constant.

1.4 Boltzmann equation

Similarly to the Liovilles equation (1.14) which defines evolution of the N-particle dis-
tribution function, DN , we may derive a partial dierential equation for evolution of the
single-particle distribution function, f (r, p). The procedure lies in integration of (1.14)
over the whole definition space of positions and momenta of all, but one particles. More
precisely, lets define r r 1 and p p1 and perform the integration over all coordinates
and momenta except for r and p:
N (
)
DN DN h DN h w(r i , r j ) DN
dN1 = dN1 + dN1 .
t pi r i r i pi r i pi
i=1 i,j
(1.26)
Lets first consider the left-hand side of equation (1.26). Provided the conditions of a
thorem of derivation of integral with respect to parameter are fulfilled (put them down
explicitely here?), we may write

DN f (r, p)
dN1 = DN dN1 = . (1.27)
t t t

Considering the second term in (1.26), we first split it into three parts (i = 1, i = 2 and
i > 2):
N (
)
DN h DN h DN h DN h
dN1 = dN1 dN1 +
pi r i r i pi p r r p
i=1
( )
DN h DN h
dN1 +
p2 r 2 r 2 p2
N ( )
DN h DN h
dN1 .
pi r i r i pi
i=3
(1.28)

The specific and very important property of the first two terms on the right-hand side
of eq. (1.28) is that (i) h is function of just r and p and (ii) the partial derivation is
1.4. BOLTZMANN EQUATION 16

with respect to these quantities over which we do not integrate. Hence, we may put the
terms h/r and h/p out of the integrals and, under the assumptions of derivative
of integral with respect to a parameter, we may write:

DN h DN h h f h f
dN1 dN1 = . (1.29)
p r r p r p p r
Finally, we will show that the second (and analogically any addent in the subsequent
sum) is zero. Lets reorganise the diernetials in dN1 and evaluate a sub-integral
( )
DN h DN h
I1 d3 r 2 d3 p2 . (1.30)
p2 r 2 r 2 p2
Performing integration per partes with respect to variable p2 and r 2 in the first and
second term, respectively, we will get
[ ]
h h
I1 = DN d r2
3
DN d3 r 2 d3 p2
r 2 p2 r 2 p2
[ ]
h 3 h
DN d p2 + DN d3 r 2 d3 p2 (1.31)
p2 r2 p2 r 2

the notation of individual terms in (1.31) has to be taken somewhat fuzzy. More pre-
cisely, one should perform the integration over components of vectors r 2 and p2 . In the
first and third term on the right-hand side of (1.31), the argument of squared brackets
is to be evaluated in appropriate infinity. Hence, these terms are both zero as DN goes
to zero at infinity which is a necessary condition for the normalisation (??) to be ful-
filled. The second and fourth term in (1.31) cancel out provided the function h(r, p is
smoothly dierentiable with respect to both arguments (which is a condition for possible
exchange of the order of dierentiation).
As mentioned above, each individual part of the sum in the third right-hand side
term of eq. (1.28) gives zero after the manipulation analogical to that presented above.
Equation (1.26) then reduces to

f h f h f w(r i , r j ) DN
+ = dN1 . (1.32)
t p r r p r i pi
i,j

The right-hand term of eq. (1.32) could be simplified, using a procedure similar to that
presented above. It can be easily shown that only N 1 terms with i = 1 are non-zero
and they are all identical due to symmetries of DN . With an alternate notation r r 2
and p p2 we write

f h f h f w(r, r ) DN d3 r d3 p d3 r 3 d3 p3 . . . d3 r N d3 pN
+ = (N 1)
t p r r p r p (N 1)!

w(r, r ) DN 3 3
= d r d p dN2
r p

w(r, r ) f2 (r, p, r , p ) 3 3
= d rd p . (1.33)
r p
17 1.4. BOLTZMANN EQUATION

In the last step we, again, assumed possibility to exchange order of integration and dier-
entiation with respect to p. The right-hand side of eq. (1.33) covers mutual interaction
of the particles and is usually called the collision term, being denoted as (f /t)coll .
Using the Hamilton canonical equations, we may write one of several standard forms
of Boltzmann equation:
( )
f f f f
+ r + p = . (1.34)
t r p t coll

Keep in mind that this is just a notation and the collision term still involves integral of
two-body distribution function. This is an principial problem which makes it impossible
to solve the equation. There are several ways how to overcome this obstacle.

1.4.1 BBGKY hierarchy

Deriving partial dierential equation for two-body distribution function, f2 , in a similar
way, i.e. integrating the Liouville equation over 6(N 2) variables may seem to be a
straightforward way how to make the system of equations closed. Unfortunately, the
problem would reappear in the equation for f2 we would obtain an integral term
including three-body distribution function. The procedure may be repeated again and
again, leading to a closed set of N 1 equations. This set is called the BBGKY
hierarchy (after scientists who independently studied this problem: Bogolyubov, Born,
Green, Kirkwood and Yvon) but in its completeness, it is no more suitable for solving
than the Liouville equation.
A general class of approximations with an intent to make a smaller set of equations
closed lies in assumption that the higher-order distribution function can be written as a
combination of lower-order distribution functions, e.g. f2 (r, p, r , p ) = f (r, p) f (r , p ).

1.4.2 Vlasov approximation

The simplest, above presented approximation is, actually, often used and is the basis of
the Vlasov equation. Lets start without the approximation and rewrite f2 in a following
way:
f2 (r, p, r , p ) = f (r, p) f (r , p ) + g2 (r, p, r , p ) . (1.35)
The first term on the right han side of formula (1.35), a product of two single particle
distribution functions, represents probablility density of two uncorrelated events (find-
ing one particle at position r with momentum p and another one at position r and
momentum p . Function g2 is then a correlation function which gives the probablity
density of finding one particle at position r and momentum p due to finding another
particle at position r and momentum p and vice versa.
(Describe transition from p to v.)

f f f
+v (V (r) + (r)) =0 (1.36)
t r v
(We ma define as the mean value, i.e. drop the sign from eq. 1.36.)
1.5. FLUID EQUATIONS 18

1.4.3 Fokker-Planck approximation

1.5 Fluid equations

Some variant of Boltzmann or Vlasov equation appears to be suitable for solving various
physical systems. Nevertheless, in other cases it is either too dicult, or unnecessary
to consider all information held by the distribution function. Fluid equations, which
subsequently allow various heuristic modifications are quite commonly used. Usually,
they are derived from the Vlasov equation in the form (1.36).
The basic quantities which stand in the fluid equations are defined as various, linearly
independent moments of the single-particle distribution function:

m
u mv f (r, v) d3 v . (1.37)
E 2 m |v u|
1 2

The quantities , u and E are mean values of density, velocity and specific (per unit
mass and unit volume) internal energy of the statistically described system. They are
functions of position and time, but we have dropped the dependency on velocity v.
The fluid equations are obtained by multuplying of Boltzmann or Vlasov equation by
m, v and 12 m E and subsequent integration over the velocity space. They are sometimes
called moments (momenta?) of the Boltzmann/Vlasov equation.

1.5.1 Continuity equation

The simplest to derive is continuity equation which starts from

f 3 f 3 f 3
m d v+ mv d v m [V (r) + (r)] d v=0. (1.38)
t r v

Under certain assumptions on the form of f (r, v), we may exchange the order of inte-
gration and derivation with respect to t and r in the first and secon term, respectively.
In the last term, we may put all but the f /v term out of the integral:

f 3
m f d3 v + m v d3 v m [V (r) + (r)] d v=0. (1.39)
t r v

First two integrals are by definion and u, the third integral evaluates to f (r, v) in
infinity which is zero due to the normalisation condition. Hence, the zeroth moment of
the Boltzmann/Vlasov equation reads:

+ ( u) = 0 . (1.40)
t
19 1.5. FLUID EQUATIONS

1.5.2 Euler equation

First moment of the Vlasov equation is sometimes called Euler equation or fluid equation
of motion. It is a vector equation and for the sake of better clarity, we will write it in
an index notation, using the Einstein summation convention:

f 3 f 3 f 3
m vi d v + m v i vj d v m vi [V (r) + (r)] d v=0. (1.41)
t rj rj vj
Lets assume that derivation with respect to t and rj may be exchanged with integration
over the velocity space. We will get

ui f 3
+ (vi vj ) m [V (r) + (r)] vi d v=0, (1.42)
t rj rj vj

where for arbitrary function X we denote X m X f d3 v. It can be easily shown
(e.g. by means of integration per partes) that

f 3
m vi d v = ij (1.43)
vj
and
vi vj = ui uj + wi wj , (1.44)
where wi vi ui represents random deviations of particles motion from the mean
bulk velocity. The symmetric tensor wi wj can be split into its trace, P 13 |w|2 ,
and traceless tensor ij 13 |w|2 ij wi wj . The newly introduced quantities P and
ij are called pressure and stress tensor, respectively. Using them, we can rewrite Euler
equation to the form

( ui ) + ( ui uj + P ij ij ) = (V (r) + (r)) (1.45)
t rj ri
and with further small alterations we obtain
ui ui P ij
ui + ui ( uj ) + + uj = (V (r) + (r)) + . (1.46)
t rj t rj ri ri rj
The sum of the frist two terms on the left-hand side of eq. (1.46) is zero as it is the
left-hand side of the continuity equation (1.40) multiplied by ui . Hence, the standard
form of the Euler equation reads:
ui ui 1 P 1 ij
+ uj = (V (r) + (r)) + . (1.47)
t rj ri ri rj
Lets put down also a common vector notation of the Euler equation
u 1 1
+ (u )u = f P + , (1.48)
t
where we have replaced gradient of potential by the force density f . Lets put stress
on meaning of the term (u )u, which can be inferred from comparison of eqs. (1.47)
and (1.48).
1.5. FLUID EQUATIONS 20

1.5.3 Energy equation

Energy ballance equation will be obtained multiplying eq. (1.36) with 12 m|v|2 and inte-
grating over the velocity space. We will be more brief than in the previous two cases as
the steps done are qualitatively the same. Hence, after some algebra, we will obtain
[1 ] [1 ]
2 (u 2
+ w 2
) + 2 (u i + w i )|u + w|2
= ui (V + ) , (1.49)
t ri ri
where we denote u2 |u|2 for the sake of brevity. Introducing the (conductive) heat
flow, F 12 w w2 and keeping in mind that E 21 w2 23 P , we rewrite (1.49) to
the form
(1 ) [1 2 ]
2 (u2 ) + E + 2 u ui + uj (P ij ij ) + E ui + Fi = ui (V + ) ,
t ri ri
(1.50)
which can be further simplified by subtracting the scalar product of the Euler equa-
tion (1.45) and ui ,
1 2 1 2 P ij
( u ) + ( u ui ) = ui (V + ) ui + ui , (1.51)
t 2 ri 2 ri ri rj
yielding
ui Fi
( E) + ( E ui ) = P +, (1.52)
t ri ri ri
where ij ui /rj is a local dissipation measure. Equation (1.52) can be rewritten
with help of the continuity equation into vector form
dE
= P u F + . (1.53)
dt
Sometimes, there are extra terms added rather heuristically to the right-hand side of the
energy ballance equation, e.g. divergence of radiative heat transfer, in order to match
the considered physical system. Note, however, that rigorous addition of such terms
from the first principles is usually not trivial.

1.5.4 Closing the equations set

We have obtained a set of five linearly independent momentum equations,

+ ( u) = 0 , (1.54)
t
u 1 1
+ (u )u = f P + , (1.55)
t
dE
= P u F + (1.56)
dt
for 14 independent variables, , u (3 components), P , (5 independent components), F
(3 components) and . Such a description of a physical problem is clearly not solvable.
Hence, various approximations are used to overcome this problem.
21 1.6. THERMODYNAMICAL EQUILIBRIA

Big simplification, just sucient to close the problem from the mathematical point
of view, is achieved for isotropic systems with f (r, v) = f (r, v) (check corretness of
this condition), which imply ij = 0 and Fi = 0.
Quite often, a so-called Navi`ere-Stokes approximation is used, assuming the stress
tensor to be proportional to the tensor of deformation,
[ ]
ui uj 3 uk
ij = Dij = + ij , (1.57)
rj ri 2 rk

which reduces the number of independent variables by 5. The constant of proportion-

ality, , is called dynamical viscosity. (Check definition of various deformation tensors
and definition of the Navi`ere-Stokes approximation.)
Another step towards a closed set of equations may lie in heuristic introduction of
temperature, T , equation of ideal gas, P = n KB T and an assumption

F = T . (1.58)

With being a constant determining the heat conduction.1

Finally, if an internal potential is not neglected, some variant of Poisson equation,
e.g.
(V + ) = 4 G , (1.59)
completes the set.
(Comment on dx/dt vs. x/t; maybe already in Sec. 1.5.1)

1.6 Thermodynamical equilibria

So far, we attempted to downscale the problem of evolution of the statistically described
system to a space (much) less-dimensional than the phase space. There is, however, a
good reason to describe the system in its full complexity in a specific case when
it is in a thermodynamical equilibtium. We postulate that the distribution function
corresponding to the equilibrium state is the one for which the statistical entropy,

S(D) KB Tr (D ln D) , (1.60)

reaches a maximum value. (Where to add a section on the statistical entropy? Here,
or to Sec. 1.2?) For example, lets consider a system with finite number, N , of possible
states. The distribution function (probabilty in this case) for each state i is then D(i) =
1/N . Nevertheless, the definition of the equilibrium system is more general and can be
applied also to statistical systems with some additional information. In particular, we
usually consider a given mean value, A Tr(AD), of some physical quantity which
can be assigned to the state of the system (i.e. measured in the real world) to play the
role of the constraint on the distribution function.
1
Quite often the conductive heat flow, F , is, actually, assumed to be zero or negligible but, on the
other hand, radiative heat flow is considered. In such a case, equation of radiative transfer, formally
similar to (1.58) is introduced with being an opacity.
1.6. THERMODYNAMICAL EQUILIBRIA 22

Lets have a set of n physical quantities Ai , mean values of which are given. We are
looking for the form of the distribution function which maximises value of S(D) and
fulfills n + 1 constraints Ai = Tr(Ai D) and TrD = 1. A possible way to solve this task
is via variation: { }
1 n
S(D) + i [Tr(Ai D) Ai ] = 0 , (1.61)
KB
i=0
where i are Lagrange multipliers and A0 1. Using the explicit form of A0 and S(D)
equation (1.61) leads to

n
Tr [(ln D + 1)D] + 0 Tr(D) + i Tr(Ai D) = 0 , (1.62)
i=1

which may be further rewritten as

n
Tr [ln(D)D] + 0 Tr(D) + i Tr(Ai D) = 0 , (1.63)
i=1

where 0 0 1. Finally, dropping prime from 0 , we obtain:

[( ) ]
n
Tr ln D + i Ai D = 0 . (1.64)
i=0

It can be shown (do it!) that, in order to fulfill relation (1.64) for arbitrary variation
D, the term in round brackets has to be zero, i.e.
[ n ]

D = exp i Ai . (1.65)
i=0

(Where does the minus sign come from? Perhaps, we should have minus sign al-
ready in eq. 1.61.) Hence, we have an explicit form of the distribution function (func-
tions/operators Ai are given) with n + 1 parameters i which are related to given values
of Ai . Let us express 0 for convenience. First, rewrite (1.65) to
[ n ]

D = e0 exp i Ai . (1.66)
i=1

Then, consider that A0 = TrD = 1 and apply the operator Tr to equation (1.66)
which gives { [ n ]}

e0 Tr exp i Ai =1. (1.67)
i=1
Instead of 0 itself, let us introduce so-called partition function (also known as patition
sum): { [ n ]}

Z e0 = Tr exp i Ai . (1.68)
i=1
23 1.6. THERMODYNAMICAL EQUILIBRIA

The equilibrium distribution function 1.65 can be then rewritten as

[ n ]
1
D = exp i Ai . (1.69)
Z
i=1

Partition function Z keeps a lot of information about the system. It can be easily shown
that, e.g.
ln Z 2 ln Z
= Ai and = Ai Aj Ai Aj . (1.70)
i i j
Similarly, we may write relation for entropy S and the Lagrange multipliers:
S
= KB i . (1.71)
Ai

(Can the equilibrium state (of isolated system) be reached starting from non-equilibrium
one?)

1.6.1 Statistical ensembles

The statistical systems appear often static from a human perspective (despite micro-
scopic motions, which are not directly observable) and this allow us to describe these
systems by a set of macroscopic variables. Such systems may be described by statis-
tical ensembles, which are assumed to be in a statitical equilibrium, and given various
macroscopic constraints, these ensembles depend on a few observable parameters. Three
important thermodynamic ensembles were defined for a fixed volume .

Microcanonical ensemble is a statistical ensemble, where the total energy of the

system and the number of particle in the system are known (they each have a
particular fixed values). The system is totally isolated, no exchange of the energy
or particles with the environment is allowed. Each member of the ensemble has
the same total energy and the same total number of particles.

Canonical ensemble is a statistical ensebmel, where the energy is not know exactly,
but the number of particles is fixed. The canonical ensemble is suitable for de-
scription of a system, which has a heat exchange with a surrounding environment,
however there is not particle exchange with the surroundings. Each member of the
ensemble has the same number of particles, their energies are not defined, how-
ever there is a constraint on a mean energy (usually expressed by a temperature),
which is given by the temperature of the surrounding bath.

Grandcanonical ensemble is a statistical ensemble, where neither the energe nor the
particle number is fixed. Similarly to the previous case, instead of the total en-
ergy the temperature is defined, and in place of the total number of particles the
chemical potential is defined. The grandcanonical ensemble is appropriate for de-
scription of open systems, which has a thermal contact with the surroundings and
exchanges particles with a reservoir. For members of the grandcanonical ensemble
1.6. THERMODYNAMICAL EQUILIBRIA 24

the temperature and the chemical potentical of the reservoir places constraints to
to the mean energy of each member and mean number of particles.

In the following, we will use the properties of both the canonical and grandcanonical
ensembles in order to derive the equilibrium particle distribution function.

1.6.2 Maxwell-Boltzmann distribution

We will derive the particle distribution function for an open system of particles. We
will use the calculation for a grandcanonical ensemble. We remind for clarity that the
usage of such a statistical ensemble means that we do not know exact total energy of
the particles in the volume nor the total number of particles. However the contraints
are given to mean values of energy and mean number of particles, these quantities are
further denoted by a temperature T and a chemical potential . Elaborating Eq. (1.69)
for this case the distribution function is given by

1
D= exp [H + N ] , (1.72)
ZG

where ZG is a partition function of the grandcannonical ensemble and the two opera-
tors connected with the two known constraints are H, Hamiltonian as an operator of
energy and N is a formal operator of number of particles. These two operators use La-
grange multipliers , which represent the temperature and representing the chemical
potential. Note that the argument of the exponential is a hybrid function, because the
hamiltonian H is continuous, whereas the operator of number of particles N is discrete.
The grandcanonical partition function may be written using its canonical counter-
part ZC

ZG = eN ZC (, N ), (1.73)
N =0

which basically say that the partition function (also termed as sum over states) simply
is a sum of partition functions of the canonical ensembles with fixed number of particles
N with a given temperature hidden in .
For a hamiltonian HN of N noninteracting particles in the external potential V (r)
given by
HN = H1 (r i , pi ), (1.74)
i

where
H1 (r, p) = p2 /(2m) + V (r) (1.75)
is a hamiltionian of one particle in the external potential, we may write the canonical
partition function as

+

ZC = exp [HN ] dN , (1.76)

25 1.6. THERMODYNAMICAL EQUILIBRIA

where dN has a usual meaning

d3 r 1 d3 p1 . . . d3 r N d3 pN
dN = . (1.77)
3N N !
Turning the sum over N particles into a multiplication (it is possible because the
functions H1 are the same) we obtain

+ N
( )
p2i /(2m) V (r i ) 3 3 1
ZC = e e d r i d pi . (1.78)
3N N !
i=1

Thanks to the unification of the description of the energy we may perform inter-
grations both over the positions and momenta.
+ The integration over the momenta is

simpler, as it is in fact an Euler integral exp (x2 ) = . The intergration in the
spatial domain requires a bit of the justification.
In order to be able to perform the integration in the spatial domain, we must de-
fine the external potential V . To keep things simple, we would require an integrable
potential, which is zero outside the region of interest (outside our volume , where we
study the problem), and which has some value inside . It is easy to show that the
potential well will have exactly those properties. Then the integration of r leads to a
scaled volume of our region of interest. Hence we have

1 3/2 N N Z1N
ZC = (2mK B T ) . (1.79)
3N N ! 3N N !
By choosing the potential V in a form of a infinite potential well (i.e. V (r) = 0 for
r and V (r) = for r / ) we set = 1 and the integration of the spatial part
in (1.78) exactly leads to the volume of the region of interest. Beware of a conflicting
notation, when means both the regions of interest and the volume of this region.
Further note that we defined a particle partition function Z1 () ZC (, N = 1).
In the following, we will apply the above discussed approximation of the infinite
potential well. By combining (1.73) and (1.79) we have

[ ]N [ ]
1
ZG = e 3 (2mKB T ) 3N/2
= exp e 3 (2mKB T ) 3/2
exp [e Z1 ] ,
N!
N =0
(1.80)
where we used the Taylor expansion of the exponential function.
Now we search for a probability of finding a particle in spatial point r with a mo-
mentum of p. The procedure is somewhat complicated, because we dont know the
exact number of particles in our open system. The sought probability f 2 is however
2
Let us take a short excursion to the past chapters. The notation f (r, p) was already used when we
discussed the reduced distribution functions of N particles see eq. (1.15). The question is what is the
relation of (1.15) to (1.81). We need to keep in mind that in the past chapters our ensemble consisted
of fixed number of particles. Therefore strictly speaking, for the previous case P(N ) = 1 and the first
sum disappears. Then the two f s have the same meaning.
1.6. THERMODYNAMICAL EQUILIBRIA 26

given by [ ]


N
f (r, p) P(N ) PN (i, r, p) , (1.81)
N =0 i=1

where P(N ) evaluates the probability that there is exactly N particles in the ensemble
and PN (i, r, p) indicates the probability that i-th particle (from fixed number of N
particles) can be found around coordinates (r, p) in the phase space. P(N ) may be
intuitively written as a trace of the full-system distribution function D from (1.72):

+
eN ZC (, N )
P(N ) = DdN = . (1.82)
ZG (, )

Further PN is simply defined for 1-particle probability (probability of finding particle

carrying flag 1 on phase coordinates (r, p)) by

+
+
dr 2 dp2 . . . dr N dpN 1
PN (1, r, p) = DN = DN dN 1 (1.83)
3N N ! N 3

where DN = eHN /ZC (, N ) is the distribution function for the canonical ensemble.
We further use (1.79) to have

+
+
1 eHN 1 3N N (N 1)!eHN 3N eH1
PN (1, r, p) = = = 3 .
N 3 1
Z
3N N ! 1
N () N 3 N
Z1 () Z1

(1.84)
The 1-particle hamiltonian H1 is defined by (1.75).
Then (1.81) takes a form of

eN ZC (, N ) 3N eH1
f (r, p) = N 3 , (1.85)
ZG (, ) Z1
N =0

where the second sum over i in (1.81) was turned into a multiplication by N , as we
have N undistinguishable particles. We further elaborate (1.85) by using (1.79) and by
moving terms not depending on N to the front of the sum to obtain:


e(N 1) Z1N 1

(N 1)!
1 eN Z1N 3N eH1 e eH1 N =1
f (r, p) = N = , (1.86)
3 Z G 3N N ! Z1 3 exp [e Z1 ]
N =0

where used (1.80) and realised that the sumation index N must start from 1 instead of
the original 0, because the the expression (N 1)! is not defined for N = 0. Now we
realise that the sum in the numerator is the Taylor expansion of the denominator and
hence finally
eH1 (r,p)
f (r, p) = . (1.87)
3
27 1.6. THERMODYNAMICAL EQUILIBRIA

Now we use the fact that the mean values of physical quantities may be expressed
from the partition function (1.70):
ln ZG
N = = e Z 1 = e Z 1 (1.88)

+
and our previous finding that exp[V (r)] = to finally have

+
+
1 [ ]
f (p) = f (r, p)d3 r = d3 r 3 e exp p2 /(2m) + V (r) =


[ ]
p2
= 3 exp = N g(p). (1.89)
2m
Function g(p) is what is our final solution to the problem and is usually termed the
Maxwell-Boltzmann distribution function. Explicitly written by using (1.88) and (1.79)
we have
ep /2m
2
p2 1
g(p) = 3 e 2m = . (1.90)
Z1 (2KB mT )3/2

1.6.3 Chemical equilibrium, Sahas equation

Lets consider a system of three dierent kinds of particles free electrons and protons
and neutral hydrogen atoms. Lets define the system by a simple hamiltonian

N e Np NH

H= H1e (r i pi ) + H1p (r i pi ) + H1H (r i pi ) (1.91)

i=1 i=1 i=1

with
p2
H1a (r, p) = + V (r) + ha1 , a = {e, p, H} . (1.92)
2ma
Hamiltonians ha1 correspond to internal degrees of freedom and we will asusume them
to have a discrete spectrum. Partition function of this grandcanonical enssemble can
be written in the form

ZG (, e , p , H ) = {Tr [exp(a N a )] Tr [exp(H a )]} . (1.93)
a

Partition function of a canonical enssemble of Na particles of the species a reads:

( [ ( 2 )] 3 3 )Na
1 p d rd p ( { })Na
a a
ZC (, N ) = a exp + V (r) Tr exp (ha1 )
N ! 2m a 3
[ ]Na
1 1 1 a
= a (2ma KB T )3/2 a (T ) = a [Z1a a (T )]N , (1.94)
N ! 3 N !
where
exp [ V (r)]d3 r , (1.95)
1.6. THERMODYNAMICAL EQUILIBRIA 28

is the fixed volume of the configuration space accessible to the particles (i.e. we assume
V (r) to represent an infinitely deep potential well) and
{ } a
a (T ) Tr exp (ha1 ) = di exp [Eia ] , (1.96)

where dai is degeneracy of the energy level Eia . Finally, let us assume for simplicity that
each type of particles has only one internal energy level,

e = de exp [E0e ]
p = dp exp [E0p ] (1.97)
H
= d0 exp [(E0e + E0p Ui )] ,

where de = dp = 2 due to spin degeneracy of free electrons and protons and d0 denotes
the degeneracy of the ground state of the hydrogen atom with (positive) ionisation
energy Ui . Inserting (1.94) into (1.93) and replacing trace in the space of N a by sum,
we obtain
1 [ a ]Na [ a a a ]
ZG (, e , p , H ) = e
Z1
a a
(T ) = exp e Z 1 (T ) . (1.98)
a a
Na ! a
N

This allows us to evaluate mean number of particles of species a,

( a )
1 ZG (2ma KB T )3/2
N =
a
= a
(T ) exp . (1.99)
a 3 KB T
Chemical equilibrium means that the rate of the reaction e + p H is equal in both
directions, which can be expressed in terms of chemical potentials,

e + p = H . (1.100)

Equality (1.100) can be rewritten in more complicated way:

[ ( a )] a

exp = 1 , with e = p = 1, H = 1 . (1.101)
a
K B T

Inserting (1.99) into (1.101) and denoting na Na / we obtain:

[ ] a
3 1
na a
=1 (1.102)
a (2ma KB T )3/2 (T )

which may be finally rewritten with use of (1.98), mH mp and ne np to the so-called
Sahas equation for equilibrium of ionised and neutral atoms,
( )
ne 4 1 (2ma KB T )3/2 Ui
= exp . (1.103)
nH d0 ne 3 KB T

(There is a dierence of factor 2 in my hand-notes; check the derivation once more.)

Chapter 2

Plasma

Plasma is being defined as a quasi-neutral gas of mixture of charged and neutral particles
showing a collective behaviour. It is characterised by its density n, temperature T
(which is usually bound to the width of the velocity distribution, the Maxwell velocity
distribution (1.90) for the equilibrium states), and the ionisation degree (usually well
described by the Saha equation (1.103)). The collective behaviour indicates that the
electromagnetic interaction is long-range one, hence the behaviour of each particle of the
ensemble influence the behaviour of all other particles1 . The quasineutrality indicates
that on the large scales, the possible charge concentrations are eciently shielded and
thus not observable from the outside. The shielding length scale is termed a Debye
length D .
An illustrative conception of the shielding process may be descibed in the following
model (see Fig. 2.1). In the system of charged particles, suddenly a concentration of
one charge appears. There are random mechanisms that can drive such clumping. The
particles of the oposite charge then cluster around this clump to remove the charge
disballance. We need to mention explicitly that the cluster of the oposite charge is not
stationary. Thermal motions of particles are very large, so particles propagate through
shielding region almost freely, only the electric potential of the initial charge concetration
slows them down. The propagating particles spend longer time in the shielding region
creating an apparent oposite charge clump.
The electric potential must follow the Poisson equation. Thus:

= , (2.1)
0
where is a charge density and 0 is a permitivity of the vacuum. Note that at the
1
The Coulomb electric interaction decreases as 1/r2 , where r is the distance, whereas for the given
solid angle (where r/r = const) the volume of the plasma increases as r3 . Hence altogether the electric
force does not vanish to large distances within the plasma.

(figure here)

Figure 2.1: The concept of Debye shielding.

29
 30

point of the initial charge concentration the potential may diverge. The approach we
use here is fully justified for scales larger than the microscopic scale, but small enough
to feel the perturbation in the potential. In one-dimensional case we have

2 e(ni ne )
2
= = , (2.2)
x 0 0
where e is an electron charge and ni and ne are ion and electron number densities re-
spectivelly. The charge introduces an electric field V = e. Assuming the equilibrium,
the charges organise according to the Boltzmann distribution
[ ]
V
n = n0 exp , (2.3)
KB T
where KB is the Boltzmann constant and T is the temperature of the gas. We will
solve2 the density distribution for electrons only, assuming that the other charges are
caused by much heavier ions. Hence
[ ]
e
ne = n0 exp , (2.7)
KB Te
which is termed Boltzmann relation and we will encounter it once more in Section 4.5.2.
By using a Taylor expansion we have
( )
e e
ne = n0 1 + ne n0 = n0 . (2.8)
KB Te KB Te
Putting ni = n0 and combining (2.2) and (2.8) we have

2 e2
2
= n0 , (2.9)
x 0 K B T e
which might be solved as
x
= 0 e D , (2.10)
2
We may start from assuming that the electrons are in a equilibrium state and their motion is
influence by an external potential q. Then we integrate the distribution function
[ ]
1
mv 2 + q
f (v) = A exp 2
, (2.4)
KB Te

where A is a normalisation constant, over all velocities v. The result is a total particle density depending
on the external potential. We have:
[ ] [ ]
q
1
mv 2
ne () = f (v)dv = exp A exp 2
. (2.5)
K B Te K B Te
0 0

The integral is a definition of the background density n0 , when there is no external potential. Hence
we have [ ]
q
ne () = n0 exp . (2.6)
KB Te
31

where
0 KB Te
D (2.11)
n 0 e2
is the Debye length. Thus in such a system with artificially inserted ion charge the
electrons will concentrate so that the inserted charge is exponentially attenuated.
Note that Debye length increases with temperature, when the thermal motions of
electrons are larger and the shielding is thus less eective (the electrons are running out
of the shielding region) and decreases with increasing density (the more electrons, the
more is the shielding eective).
A few additional notes.

In physics, the temperature is a measure of the width of the particle velocity

distribution and usually is being expressed in Kelvins. In plasma physics, a quite
often used unit expressed by the kinetic energy (in electronvolts eV) of the particle
having the given temperature. From
1
v 2 = KB T
Ek = m (2.12)
2
where v is the root-mean-square velocity, we find an usefull working relation that
1eV corresponds to 11 600K.

Plasma may have many temperatures at once. Electrons and ions may have dif-
ferent velocity distributions and thus dierent temperatures. This is due to the
fact that the frequencies of ion-ion and electron-electron collisions are much higher
than the frequency of ion-electron collisions. The velocity distribution of a single
particle species does not have to be isotropic, e.g., when a magnetic field is present.
Than the temperature of the plasma is usually higher along the field than across
the field.

Plasma densities vary over many orders in the universe. The density of 106 m3
is the typical plasma density in the interplanetary space around the Earth, it
might be many orders less in the interstellar or intergalactic space. It is around
108 m3 in the giant molecular clouds, from which the stars are born. The plasma
density in the solar photosphere is in the order of 1022 m3 , in the solar core than
1031 m3 . In the laboratory plasmas, the densities between 1020 1028 m3 are
considered for the experiments with the termonuclear fussion.

Various approaches can be practically used to describe plasmas and their dynam-
ics. The most detailed is the kinetic description using the Boltzmann equation.
The dynamics of plasmas is also being described using the fluid and magnetohydro-
dynamic approaches. The motion of individual or testing particles is investigated
using the drift description. Using some of these approaches, we may state the
conditions for the plasma equilibrium, to study instabilities and conditions for the
propagation, excitation and damping of waves and other oscillatory motions. We
will study all these aspects in the following chapters.
 32

The ionisation degree may be estimated using the Saha equation (1.103). It
strongly depends on the temperature. For example, for the air at the room temper-
ature (assumed that the air is composed from nitrogen only), one get the estimate
(assuming n0 3 1025 m3 , T 300 K, and Ui 14.5 eV)) of the ionisation
degree to be ni /n0 10122 . The air in the room certainly is not a plasma.

In practice, there are three criteria, which help to decide whether the studied
system may be considered plasma:

1. The system is much larger than the Debye length, hence the quasineutrality
holds. L D , where L is the characteristic size of the system.
2. Shielding is eective. The number of particles in the Debye sphere is much
larger than one, ND 1. In case it is not, the studied system is probably a
bound system.
3. The particle interactions are dominated by electromagnetic forces and not
by collisions. > 1, where is a characteristic oscillation frequency of
plasma and is a characterictic time between particle collisions. For example
an exhaust of the jet engine cant be considered plasma despite the huge
temperature and high ionisation, because the particle interactions are purely
collisional and electromagnetic forces play negligible role.
Chapter 3

Charged particle motion in

electromagnetic fields

In plasma, we sometimes have to consider motions of each single particle to capture

some of the eects. The densities of plasmas are not large enough to be allowed to
completely ignore the motions of individual particles. The plasmas usually are not
dominated by collisions. Hence motions of individual particles play an important role.
In this approach we study the motion of test particle in the prescribed electric and
magnetic fields. In general, the trajectory of such motion may be unambigously found
by solving the equation of motion

dv(r, t)
m = q [E(r, t) + v(r, t) B(r, t)] , (3.1)
dt

In the following sections we will discuss some examples by giving perticular analytical
forms to background fields E(r, t) and B(r, t). We will derive the relations for some
drifts, thus systematic motion of charged particles which are not captured when we use
other approaches to study plasmas. It is important to study drifts of particles because
these drifts often drive instabilities, which is crutical e.g. for laboratory plasma in
plasma confinement and thermonuclear fusion.

3.1 Homogeneous magnetic field

Let us define the system with constant homogeneous magnetic field and no electric
field, where we define Cartesian coordinate system so that the z axis is parallel to the
direction of magnetic induction. Then we have the vector equation of motion in a form
of

v vx 0
d x
m vy = q vy 0 , (3.2)
dt
vz vz B

33
3.1. HOMOGENEOUS MAGNETIC FIELD 34

8
x 10
3 x 10

1
1

0.5 0.8

0.6

z [m]
z [m]

0.4

0.5
0.2

1 0
5
5
0
5 0.2
0 0
4 10 0.4
x 10 0 x 10
0.6
5 4
x 10
9
x 10
5 0.8
5 1
10 10
y [m] y [m]
x [m] x [m]

9 10 8
x 10 x 10 x 10
0 6 1.2

0.2 4 1

0.4 2 0.8
x [m]

y [m]

z [m]

0.6 0 0.6

0.8 2 0.4

1 4 0.2

1.2 6 0
0 1 2 0 1 2 0 1 2
t [s] x 10
8 t [s] x 10
8 t [s] x 10
8

10 8 8
x 10 x 10 x 10
1.2 1.2
6
1 1
4
0.8 0.8
2
y [m]

z [m]

z [m]

0 0.6 0.6

2 0.4 0.4
4
0.2 0.2
6
0 0
1 0.5 0 1 0 1 1.5 1 0.5 0
x [m] 9
x 10 y [m] x 10
9 x [m] x 10
9

Figure 3.1: Motion of an electron in homogenenous magnetic field. (A) The field con-
figuration, blue marking B and red marking E, (B) the 3-D view of the motion, (C)
trajectories in the coordinatetime and coordinatecoordinate spaces.
35 3.1. HOMOGENEOUS MAGNETIC FIELD

where m is the mass of the particle charged with electrical charge q. This equation
becomes a system of scalar dierential equations:

dvx
m = qBvy , (3.3)
dt
dvy
m = qBvx , (3.4)
dt
dvz
m = 0. (3.5)
dt
We immediately see that the parallel component of velocity has a trivial solution, which
is the constant motion along the magnetic field. By dierentiating the perpendicular
components with respect to time we have

d2 vx q dvy d2 vy q dvx
2
= B and 2
= B . (3.6)
dt m dt dt m dt
By combining the two we have

d2 vx ( q )2 d2 vy ( q )2
= B vx and = B vy , (3.7)
dt2 m dt2 m
which can trivially be solved and the solution to it has a form of harmonic functions.
We define
qB sgn q |q|B
c = = (3.8)
m m
to be a cyclotron frequency. Lets take the solution of the second order ordinary dier-
ential equations are
vx,y = v exp (ic t + ix,y ) , (3.9)
where v is the amplitude of the velocity and obviously1 is identical for both perpendic-
ular components and is usually initially driven by thermal motion. Both perpendicular
components are allowed to have dierent phase shifts x,y . The trajectory in the phase
space obviously is a circle. Lets set the spacetime coordinate system so that x = 0.
Then
vx = v exp ic t, (3.10)
and we consider only the real part to have a physical meaning. Then

m dvx m m qB sgn q
vy = = v ic exp ic t = v i exp ic t = v i sgn q exp ic t.
qB dt qB qB m
(3.11)
We see that the sense of rotation in the phase space depends on the charge of the
particle, thus is opposite for electrons and ions.
2 2
1 dvx dvy
It might be shown that if we multiply (3.3) by vx , (3.4) by vy and sum the two, we obtain dt
+ dt
=
dv 2
dt
= 0, hence the amplitude of the velocity is constant in time.
3.2. HOMOGENEOUS ELECTRIC FIELD 36

To obtain the trajectory, one needs to integrate (3.10) and (3.11):

[ ]
dx v v
= [vx ] = [v exp ic t] x x0 = i exp ic t = sin c t (3.12)
dt c c

and
[ ]
dy v v
= [vy ] = [v i exp ic t] y y0 = sgn q exp ic t = sgn q cos c t
dt c c
(3.13)
Of course, we should not forget about the parallel component

dz
= [vz ] = 0 z z0 = v t. (3.14)
dt
The the total trajectory is a helix. The ratio v /c = rL is the Larmor radius, the
typical radius of gyration. Note that
v m
rL = (3.15)
qB sgn q

depends on the perpendicular velocity and the mass of the particle. More massive
particle has a larger Larmor radius and thus smaller cyclotron frequency. Also, stronger
magnetic field causes the particles to gyrate using tighter orbits.
Electrons and ions circulate the magnetic field lines in the oposite directions (see
Fig. 3.1), but they both follow the field lines (we will generalise this case further). The
charge travelling along the helical trajectory induces its own magnetic field (according
to the Maxwells equations), which has the sign oposite to the background magnetic
field. Thus the plasma weakens the background magnetic field, it is diamagnetic.

3.2 Homogeneous electric field

Now we add a homogeneous constant electric field to the problem, and thus we choose
the coordinate system so that the magnetic field is still along the z axis, and the electric
field has non-trivial components only in x and z directions (see Fig. 3.2. Then the
equation of motion is

vx Ex vx 0
d
m vy = q 0 + vy 0 . (3.16)
dt
vz Ez vz B
The parallel component can be solve immediately:

dvz q q
= Ez z z0 = Ez t2 + vz,0 t. (3.17)
dt m 2m
There is an acceleration of the particle along the magnetic field caused by the parallel
component of the electric field. By using the definition of the cyclotron frequency, the
37 3.2. HOMOGENEOUS ELECTRIC FIELD

9
3 x 10
x 10

1
5

0.5 4

z [m]
z [m]

0.5
1

1 0
5
5 0
0 0.2
0 5 0.4
4 10
x 10 0 x 10 5 0.6
5 4 9
x 10 0.8 x 10
5
10 1
10 10
y [m] y [m]
x [m] x [m]

9 9 9
x 10 x 10 x 10
0 1 6

0.2 0.5 5

0.4 4
0
x [m]

y [m]

z [m]

0.6 3
0.5
0.8 2

1 1 1

1.2 1.5 0
0 1 2 0 1 2 0 1 2
t [s] x 10
8 t [s] x 10
8 t [s] x 10
8

10 9 9
x 10 x 10 x 10
5 6 6

5 5

0 4 4
y [m]

z [m]

z [m]

3 3

5 2 2

1 1
10
0 0
10 5 0 2 1 0 1 1.5 1 0.5 0
x [m] x 10
10 y [m] x 10
9 x [m] x 10
9

Figure 3.2: Motion of an electron in the crossed electric and magnetic fields.
3.2. HOMOGENEOUS ELECTRIC FIELD 38

remaining two components of the equation of motion are

dvx q
m = Ex + c sgn qvy , (3.18)
dt m
dvy
m = c sgn qvx . (3.19)
dt
Similarly to the previous case, we take the time derivative of (3.18) and (3.19) and
combine the two. We have
d2 vx dvy
2
= c sgn q = sgn 2 qc2 vx = c2 vx , (3.20)
dt dt
where we assumed that q = 0, and
(q ) ( )
d 2 vy dvx Ex
= sgn q c = sgn q c E x + c sgn qvy = c2 + vy . (3.21)
dt2 dt m B

We solved (3.20) already in the previous case and obtained (3.10):

vx = v exp ic t. (3.22)

For (3.20) we note that the term Ex /B does not depend on time, thus we may simply
substitute vy + Ey /B vy and solve
( ) ( )
d2 Ex Ex
vy + = c vy +
2
, (3.23)
dt2 B B
which is trivial again. Then we have solution in a form of
Ex
vy = v i sgn q exp ic t . (3.24)
B
We see that in the solution and additional term, drift appeared. Lets find a general
solution in a vector form. The equation of motion is
dv
m = q(E + v B) (3.25)
dt
If we substitute v = v EB
B2
(we transform to a dierent coordinate system, which
moves with a speed of EB
B2
with respect to the original one). As we will find out later,
in this moving coordinate system the parallel components of the electric field disappear.
Thus (3.25) becomes:
( )
dv dv EB
m =m =q E+v B+ B =
dt dt B2
[ ]
1   B(E B)
=q  + v B 
E EB +2
=

B2 B2
[ ]
= q v B + E , (3.26)
39 3.2. HOMOGENEOUS ELECTRIC FIELD

where we defined the parallel component of the electric field

B(E B)
E = . (3.27)
B2
We transformed the new problem to the previous case, as in this moving coordinate
system the eects of the electric field (causing the acceleration along the magnetic field
lines) and the magnetic field (causing the gyration around the magnetic field lines) are
separable. Thus we solve the problem in the moving coordinate system and transform
back to the original coordinate system.
EB
v = v + . (3.28)
B2
We see that a new component of the particle velocity appeared,
EB
vE = , (3.29)
B2
which is the expression for the drift of the gyration centre, the E-B drift. Note that the
sign of this drift does neither depend on the particle mass nor on the particle charge.
This drift does not cause the separation of the charge.

3.2.1 Relativistic E-B drift

It is interesting to show that the E-B drift naturally shows up as a consequence of the
Lorentz transform of the relativistic electromagnetic tensor. ((Do a complete derivation
in Appendix.)) That is we transform F ,

0 Ex /c Ey /c Ez /c
Ex /c 0 Bz By
F = Ey /c Bz
(3.30)
0 Bx
Ez /c By Bx 0
using the tensor of Lorentz transform

0 0
0 1 0 0
=
0
, (3.31)
0 1 0
0 0

where = v/c and = (1 v 2 /c2 ) 2 . Note that the velocity v is not known yet and
1

we will obtain it as a solution to the problem. The transformed F is

F = F =

0 0 0 Ex /c Ey /c Ez /c 0 0
0 1 0
0 Ex /c 0 Bz By 0 1 0 0
= 0
.
0 1 0 Ey /c Bz 0 Bx 0 0 1 0
0 0 Ez /c By Bx 0 0 0
(3.32)
3.2. HOMOGENEOUS ELECTRIC FIELD 40

Now we need to properly multiply these three matrices to obtain

0 (Ex /c By ) (Ey /c + Bx ) Ez /c
(By Ex /c) 0 Bz (Ez /c By )

F = (Ex /c + Bx )
=

Bz 0 (Ey /c + Bx )
Ez /c (By Ez /c) (Ey /c + Bx 0

0 Ex /c Ey /c Ez /c
Ex /c 0 Bz By
= Ey /c Bz
. (3.33)
0 Bx
Ez /c By Bx 0
Hence

Ex /c = (Ex /c By ), (3.34)
Ey /c = (Ey /c + Bx ), (3.35)
Ez = Ez , (3.36)
Bx = (Ey /c + Bx ), (3.37)
By = (By Ex /c), (3.38)
Bz = Bz (3.39)

are the transformation equations for all the components of the electromagnetic tensor.
The target mutual velocity of both coordinate systems is the one when the eect of the
magnetic and electric fields will be separable, thus when E B, which is an equivalent
of E B = 0. For reasons which will show up later, lets compute the square of the
cross product amplitude, which will certainly be zero for parallel vectors. Then by using
the rules for mixed product we have

|E B|2 = (E B) (E B) = E [B (E B)] = E 2 B 2 (E B)2 = 0 (3.40)

Lets compute this relation for transformed vectors Ec and B and lets deal only with
perpendicular components, because the parallel componets are already separated. Doing
this algebra we have
E 2 2 [ 2 ]
| | |B | = (Ex /c By )2 + 2 (Ey /c + Bx )2
c [ 2 ]
(Ey /c + Bx )2 + 2 (By Ex /c)2 =
= 4 (Ex /c By )2 (Ey /c + Bx )2 + 4 (Ex /c By )2 (By Ex /c)2 +
+ 4 (Ey /c + Bx )2 (Ey /c + Bx )2 + 4 (Ey /c + Bx )2 (By Ex /c)2
(3.41)
and
E
| B |2 = 4 (Ex /c By )2 (Ey /c + Bx )2 + 4 (Ey /c + Bx )2 (By Ex /c)2 +
c
+ 2 4 (Ex /c By )(Ey /c + By )(Ey /c + Bx )(By Ex /c). (3.42)
41 3.2. HOMOGENEOUS ELECTRIC FIELD

Hence [we immediately see that when subtracting (3.42) from (3.41), first terms cancel
with each other, as do also the fourth term from (3.41) and second term from (3.42)]:
E E
| B |2 | |2 |B |2 = 4 (Ex /c By )2 (By Ex /c)2 +
c c
+ 4 (Ey /c + Bx )2 (Ey /c + Bx )2
2 4 (Ex /c By )(Ey /c + By )(Ey /c + Bx )(By Ex /c) =
[ ]2
Ex Ex Ey Ey
4
= ( By ) (By ) (
2 2
+ Bx ) ( + Bx ) =
c c c c
[ ( ) ( )]
2 E 2
E y E y E x
= 4 x
+ 2 + Bx2 + By2 + (1 + 2 ) Bx By =
c2 c c c
[ ( 2 ( ) )]
E E
4
= + B + (1 + ) B
2 2
= 0, (3.43)
c2 c z
where we used
( ) ( )
E Ey Ex
(1 + ) B
2 2
= (1 + ) Bx By . (3.44)
c z c c
Equation (3.43) may be true if and only if we have

E
c B
= E2
(3.45)
1 + 2 + B2
c2

which implicitly defines the target speed vE of the coordinate system, in which the
eects of magnetic and electric fields are separable. This speed is hidden in = vE /c.
We will not solve this equation properly but rather take a approximation of < 1, when
we may use the 2 as a correction term and obtain
EB
v E = (1 + 2 ) E 2 . (3.46)
c2
+ B2

Note that this is the relativistic version of (3.29). When assuming non-relativistic case,
thus when is small and |E| c, the two expressions equal exactly.

3.2.2 Drift in the field of general force

Lets make a generalisation, which we will use further. In the case of E-B drift, we have
an additional force (the force of the electric field) acting on the motion of the particle.
This force (the Lorentz force) is simply given by

F E = qE. (3.47)

After substitution into (3.29) we have

1 FE B
vE = . (3.48)
q B2
3.3. INHOMOGENEOUS MAGNETIC FIELD 42

Could this expression be generalised for any force F other than Lorentz? The answer
is yes, if and only if the force in question acts homogeneously, in particular, when the
vector of the force does not depend on speed and position. Then we have a drift of
gyration centre v gs
1F B
v gs = . (3.49)
q B2
An example? The action of gravity field may be considered homogeneous in laboratory
or at least when the small spatial scales (compared to the curvature of the gravity field
lines) are in question. Then we have the gravity drift v g
mgB
vg = , (3.50)
q B2
mv2
where g is the vector of gravity acceleration. How about the centrifugal force F = R2
R
acting in the curved magnetic field? Is the expression for the curvature drift v c
mv2
vc = RB (3.51)
qR2 B 2
correct? It is not, because we violated one of the assumptions written above: the
vector of the acting force does depend on position in this case. Magnetic field is not
homogeneous anymore and we must find another correction.

3.3 Inhomogeneous magnetic field

Let us study the eects of the non-homogeneous magnetic field. We will keep our
coordinate system, thus having magnetic field in the z-direction, however we will allow
for a smooth change in the y-direction. What does the word smooth mean? Lets
remind that there is a characteristic scale of the particle motion even in the case of
homogeneous field: the Larmor radius rL = qBmvsgn
q . By smooth we further mean that
the characterictic spatial change of the magnetic field occurs on much larger scales than
the Larmor gyration. The characterictic scale of the spatial change of the magnetic field
L may be estimated by comparing the gradient of the field to its magnitude, thus
|B|
L . (3.52)
|B|
We will study the case when L rL . The ratio of the two is a small number, thus can
be used as a small parameter to expand the physical functions to the Taylor series and
neglect the higher order of this small parameter. Hence in our case of the inhomogeneous
magnetic field we have to the first order
B = B 0 + (r )B (3.53)
and in our coordinate system we have
Bz
Bz = B0 + y . (3.54)
y
43 3.3. INHOMOGENEOUS MAGNETIC FIELD

What eect do we expect? By waving our hands we can say that when we study the
gyration of the testing particle, we see that the particles still gyrate along the magnetic
field lines. However, in that part of the orbit, where the magnetic field is stronger,
the Larmor radius is smaller and vice versa. We should see a drift in the direction
perpendicular to both B and |B|. We would expect the eect to increase with the
gradient of the field, when the dierence between the eective radii in the upper and
lower part of the orbit is larger.
In our case, only the y component of the Lorentz force is perturbed
( )
Bz
Fy = qvx Bz = qvx B0 + y . (3.55)
y

The first term on the right-hand side is responsible for the normal Larmor rotation,
the second term is a correction caused by the inhomogeneous magnetic field and thus
causes the expected drift. The approach we will take is that we will separate the eects
of the B0 causing the Larmor rotation and the eect of the small correction from |B|.
To investigate the correction, we will average the particle motion over one Larmor
rotation. Lets remind again that this is possible if and only if the Larmor rotation is
the dominant motion of the studied particle and thus when |Bz /y| |B0 |. Then we
use the solution for vx from (3.10) and for y from (3.13). We have
( )
Bz
Fy = qv cos c t B0 + rL sgn q cos c t . (3.56)
y

The averaged (over Larmor orbit) Lorentz force is thus

2 2 ( )
1 1 Bz
Fy = Fy d(c t) = qv cos c t B0 + rL sgn q cos c t d(c t) =
2 2 y
0 0
2 2
1 Bz 1
= qv B0 cos c td(c t) qv rL sgn q cos2 c t d(c t) =
2 y 2
0 0
1 Bz
= 0 v rL q sgn q . (3.57)
2 y

We leave it to the reader to prove that the integral over one period of the sine or cosine
function vanishes and that the integral over one period of the sine squared or cosine
squared equals to .
We found a correction term. Since the averages of the remaining components of the
Lorentz force vanish, we may generalise
1
F = v rL q sgn q|B|, (3.58)
2
and use it in the expression for the drift in the field of general force (3.49). Note that
under approximations we took the additional force defined by (3.58) is homogeneous
3.3. INHOMOGENEOUS MAGNETIC FIELD 44

and does not depend on velocity. Hence we have a drift of the gyration centre, the
grad-B drift v B :
1 B |B|
v B = sgn qv rL . (3.59)
2 B2
Note that the sign does depend on the sign of the particle charge and thus the grad-B
drift may cause the separation of charges and to drive additional electric field. See
Fig. 3.3.
At the beginning we used the assumption of small spatial changes of the magnetic
field to derive this relation and we should check that the assumption holds and thus the
solution is consistent. We simply compute how much does the particle move due to the
drift after one Larmor rotation, hence after one period = 2/c . We will further use
(3.52) and v /c = rL :

2 1 |B| rL
= |v B | = v rL = rL rL (3.60)
c 2 |B| L

under the assumption L rL . The solution is consistent with the assumptions, the
dominant motion still is the Larmor rotation.
Now we may continue to investigate the motion of particles in the curved magnetic
field. Obviously, additionally to the curvature drift (3.51) we need to account for inher-
ited gradient of the field. Lets make a qualitative estimate. The Cartesian coordinates
are no longer convenient to investigate this case, cylindrical coordinates (R, , z) are
more appropriate (they incorporate naturally the curvature of the field). In any case,
we assumed the changes of the magnetic field to be smooth, thus we may always in-
troduce local cylindrical coordinates to approximate the real configuration of the field
lines. In vacuum we have
B = 0 j = 0, (3.61)

hence when we assume that the dominant direction of the magnetic field lines is along
the azimuthal direction, |B| must have only the component in R direction. Hence

1
B = (RB ) = 0 (3.62)
R R
and therefore
C
B = , (3.63)
R
where C is an integration constant. We further obtain

C 1 R
B = = C 2 (3.64)
R R R R
and when expressing C from (3.63) we finaly have

|B|
B = R. (3.65)
R2
45 3.3. INHOMOGENEOUS MAGNETIC FIELD

8 8
x 10 x 10

2
1
1.5

1 0.8

0.5
0.6

z [m]
z [m]

0.5 0.4

1
0.2
1.5

2 0
5
1
1.5
1 0
0 0.5 0 2
8 10
x 10 0 x 10 4
1 0.5 8 6 10
1 x 10 x 10
8
1.5 5 10
2 2
y [m] y [m]
x [m] x [m]

10 10 8
x 10 x 10 x 10
2 6 1.2

0 4 1
2
2 0.8
4
x [m]

y [m]

z [m]

0 0.6
6
2 0.4
8

10 4 0.2

12 6 0
0 1 2 0 1 2 0 1 2
t [s] x 10
8 t [s] x 10
8 t [s] x 10
8

10 8 8
x 10 x 10 x 10
1.2 1.2
6
1 1
4
0.8 0.8
2
y [m]

z [m]

z [m]

0 0.6 0.6

2 0.4 0.4
4
0.2 0.2
6
0 0
10 5 0 1 0 1 2 1 0 1
x [m] x 10
10 y [m] 9 x [m] 9
x 10 x 10

Figure 3.3: Drift of an electron in an inhomogeneous magnetic field.

3.4. INHOMOGENEOUS ELECTRIC FIELD 46

Now we can ingest this expression for grad-B into (3.59) and (when also explicitly
writting the expression for rL ) have
2
1 v rL R 1 mv
v B = sgn q 2 B |B| = R B. (3.66)
2 B R 2 qB 2 R2

After adding the curvature drift (3.51) we finally have an expression for the drift in the
curved magnetic field:
( )
mRB 2 1 2
v gs = v c + v B = v + v . (3.67)
q R2 B 2 2

This has important consequences, especially for laboratory plasmas. It shows that
the plasma cannot be confined in the simply curved magnetic field. Both important
drifts, the curvature drift and the grad-B drift, sum up and thus act to displace the
plasma. Both these drifts are drivers for kink instability. Lets imagine that the plasma
flows along the cylindrical tube of the magnetic field (the flux tube) with a prevalently
poloidal configuration of the magnetic field, which is generated by the current. When
this flux tube is displaced (e.g. by some random motions), the tube gets curved with
the gradient of the magnetic field pointing against the displacement. The drifts in the
curved field cause the charge separation in the direction perpendicular to the plane of
the kink. Separated changed introduce a small electric field and the E-B drift cause the
instability to grow.
When the configuration of the magnetic field in the flux tube is rather helical than
cylindrical, E-B drift may cancel the drift in the curved magnetic field and the plasma
may remain confined within this tube. This is the famous tokamak configuration.

3.4 Inhomogeneous electric field

Also the inhomogenous electric field requires the correction to the normal E-B drift of
particles. Let us investigate one example only, when the electric field has a harmonic
character. Such configuration is not uncommon, later on we will study the plasma
oscillations and waves propagating through plasmas, where we will see that there are way
too many harmonic motions in plasmas, that may be approximated as a superposition
of simple harmonic functions. Let us assume that we have a periodic electric field in
the x direction (see Fig. 3.4),
E = E0 cos kyex , (3.68)
where k is the wave number, which is related to the wavelength ( = 2/k), which is
the characteristic spatial scale of the perturbation. Otherwise we still keep the same
configuration of the coordinate system as we did in the previous cases. Hence the
perpendicular components of equation of motion are

dvx qB q dvy qB
= vy + Ex (y) and = vx . (3.69)
dt m m dt m
47 3.4. INHOMOGENEOUS ELECTRIC FIELD

8
x 10
8 x 10

2
1
1.5

1 0.8

0.5
0.6

z [m]
z [m]

0.5 0.4

1
0.2
1.5

2 0
0
1
2 2
1
0 1
8 9 4 0
x 10 0 x 10
1 1
8
x 10 6 1 9
x 10
2 2
2 8
y [m] y [m]
x [m] x [m]

9 9 8
x 10 x 10 x 10
2 2 1.2

1 0 1

2 0.8
0
x [m]

y [m]

z [m]

4 0.6
1
6 0.4
2 8 0.2

3 10 0
0 1 2 0 1 2 0 1 2
t [s] x 10
8 t [s] x 10
8 t [s] x 10
8

9 8 8
x 10 x 10 x 10
1.2 1.2
0
1 1
2
0.8 0.8
y [m]

z [m]

z [m]

4 0.6 0.6

0.4 0.4
6
0.2 0.2
8
0 0
4 2 0 2 1 0 1 4 2 0 2
x [m] x 10
9
y [m] 8 x [m] 9
x 10 x 10

Figure 3.4: Drift of electrons in the inhomogeneous electric field.

3.4. INHOMOGENEOUS ELECTRIC FIELD 48

The standard way to solve this system is to take the time derivative of these equations
and combine the two, hence

d2 vx sgn q dEx d2 vy Ex (y)

2
= c2 vx + c and 2
= c2 vy c2 . (3.70)
dt B dt dt B
This is a system of implicit ordinary dierential equations, which cannot be trivially
solved. We would need to proceed iteratively, since in order to find the solution, we
need to know E, which is the function of the solution. Lets focus to the case of small
electric fields and apply the averaging formalism again in order to find the correction.
In this approach we simply use the unperturbed solution and average over one Larmor
rotation.
By substituting (3.13) into (3.70) we have

d2 vy E0
2
= c2 vy c2 cos [k(y0 + rL sgn qrL sin c t)] , (3.71)
dt B
which we average over one Larmor orbit to have

d2 vy E0
2
= c2 vy c2 cos [k(y0 + rL sgn qrL sin c t)]. (3.72)
dt B
The action of the additional force oscilates, therefore when averaged over one Larmor
orbit, it vanishes, thus
d2 vy
2 = 0. (3.73)
dt
We approximate the cosine term by expansion into Taylor series in the parameter krL ,
which we assume to be small2 . Essentially again the electric field is perturbed on scales
much larger than the Larmor radius

cos [k(y0 +rL sgn qrL sin c t)] =

= cos ky0 cos (krL sgn q sin c t) sin ky0 sin (krL sgn q sin c t) =
= cos ky0 cos (krL sin c t) sgn q sin ky0 sin (krL sin c t)
[ ]
1 2 2 2
= cos ky0 1 k rL sin c t sgn q sin ky0 (krL sin c t) (3.74)
2
Hence after averaging the term with the sine vanishes, while the term with the sine
squared becomes 1/2. Finally
( ) ( )
E0 1 2 2 Ex (y) 1 2 2
vy = cos ky0 1 k rL = 1 k rL . (3.75)
B 4 B 4
We may generalise this expression to
( )
EB 1 2 2
vE = 1 k rL (3.76)
B2 4
2
We remind that cos 1 12 2 and sin for 0.
49 3.5. DRIFT CURRENTS

and we keep it to the reader to prove that in our case it is the correct generalisation. In
the section dealing with waves in plasmas we will often use the solution as a superpo-
sition of various modes, when for each mode we may conviniently solve the equations
in the Fourier space. The Fourier representation is convenient because dierential op-
erators become multiplications. We ask the reader to see (5.13) for a short moment
and apply it to the equation we just obtained. Then we have We may generalise this
expression to ( )
B 1 2
v E = 2 1 + rL E. (3.77)
B 4
The correction term captures the eect of the finite Larmor radius. Since rL is larger
for ions, ions feel the changes in the background electric field stronger than electrons.
When an electric field appears in plasma, the drift instability causes this electric field to
grow. Also note that the eect of the inhomogeneous electric field is larger ( (krL )2 )
than the eect of the inhomogeneous magnetic field ( krL ).

3.5 Drift currents

Some of the drifts discussed above depend on the mass or charge of the particles, thus
lead to the separation of charges and give rise to the drift currents. In plasmas we have
to consider k dierent particle species, each having particle density nk , mass mk and
qk . From the general equation for the drift velocity (3.49) we derive the relation for the
drift current

nk F k
j drift = nk qk v drift,k = k 2 B. (3.78)
B
k
For instance the drift current due to the grad-B drift is

1 mk nk v,k
2
j B = (B B). (3.79)
2 B2
k

This current leads to the charge separation. An example of the natural drift current is
the ring current in the Earths magnetosphere.
7

3.6 Guiding centre motion

Let us show that it is possible to derive all the above discussed drifts and much more
in a very elegant and general way3 . We will correctly solve the equations of motion
and will find the corrections to the normal gyration under the approximation of slowly
varying fields. That means that we limit ourselves to the case when
rL v 1
= 1 and 1, (3.80)
L c L c
3
Following de Blank 2012: Guiding center motion, Fusion Science and Technology 61(2T), 6168
3.6. GUIDING CENTRE MOTION 50

Figure 3.5: Charged particle drifts in a homogeneous magnetic field]. (A) No disturbing
force (B) With an electric field, E (C) With an independent force, F (e.g. gravity)
(D) In an inhomogeneous magnetic field, grad B. Based on Hannes Alfvns, Cosmical
Electrodynamics (1950); (cc) Stannered, redesigned by Ian Tresman.
51 3.6. GUIDING CENTRE MOTION

where L and are the characteristic spatial and temporal changes of the fields E and
B respectively. We see that c1 m/q, which is small. Thus we define m/q to be
a small parameter of the system and we will expand the equations to the Taylor series
using this parameter. Then rL L and 1 c . We solve a system of ordinary
dierential equations

dx
=v (3.81)
dt
dv q
= (E + v B) . (3.82)
dt m

We will expand these equations in . But first, we introduce the coordinate transforma-
tion (x, v) (R, u). Coordinates (R, u) are functions of time.

R(t) = x(t) (t), (3.83)

u = v vE. (3.84)

Here R is the instanteneous position of the gyration center, x is the instantenous po-
sition of the given particle (with respect to the fixed reference coordinate system),
v E E B/B 2 is the velocity of the E-B drift [see (3.29)], and is the instanteneous
position of the particle with respect to the gyration centre. From the previous section
we may see that

b u, (3.85)
B
where b(x, t) = B(x, t)/|B(x, t)| is the unity vector along the field lines. Vectors b
and v E are functions of both space (x) and time t and their values are measured with
respect to the particle. Note that

B(x, t)
c (t) = (3.86)

is an instanteneous cyclotron frequency.

We further introduce the local cylindrical coordinates in the plane perpendicular to
the magnetic field with two direction vectors (e1 , e2 ), so that

u = v b + u e , where e e1 cos + e2 sin , and = 0 c t. (3.87)

Note again that the base vectors of this local coordinate system (e1 , e2 , b) vary in both
the space and time. Unit vectors e and

e e2 cos e1 sin (3.88)

are projections of the particles velocity u and relative position , respectively, to the
plane perpendicular to b.
3.6. GUIDING CENTRE MOTION 52

Figure 3.6: Sketch of the coordinate system of the guiding centre variables.

Hence we derived useful relations defining the guiding centre variables (see also
Fig. 3.6).

x = R + , (3.89)
m
= 2 v B = sgn qe , (3.90)
qB
d
v = = u e , and (3.91)
dt
v = v b + u e + v E . (3.92)

Now we transform (3.81) and (3.82) to the newly introduced coordinates. For (3.84) we
have:

dR dx d d ( )
= =v bu =
dt dt dt dt B ( )
d b du
= v b + u e + v E + u b =
dt B B dt
( ) ( )
d b dv dv E
= v b + u e + v E + u b =
dt B B dt dt
( )
d b dv E
= v b + v E + u + b , (3.93)
dt B B dt

where we used the equation of motion to evaluate the term

Bb dv
dt :

[ ]
dv 1
b = b (E + v B) =
B dt B
1 1 1 BB B
= b E (vb B Bb v) = b E v 2
+ v =
B B B B B
= v E v + v b = u e . (3.94)
53 3.6. GUIDING CENTRE MOTION

For the equations of motion, we expand v using (3.92), then we have

d(v b) d(u e ) dv E 1[ ]
+ + = E + (v b + u e + v E ) B =
dt dt dt
1[ ]
= E + v b B + u e B + v E B =
[ ]
1 E  E B
=  + u e B 

E BB+ B =
B2 B2
1[ ]
= E b + u e B , (3.95)

where we defined E EB
|B| .
Now we project this equation to the directions e , b, and e . For the first one we
have
[ ]
1[ ] d(v b) d(u e ) dv E
E b e + u (e B) e = + + e . (3.96)
dt dt dt
The left-hand side is zero since b e and (e B) e = (e e ) B = 0. Thus
we have
db dv du de dv E
0 = e v + e b + e e + u e + e . (3.97)
dt dt dt dt dt
The second term vanishes, because e b and the fourth term is
de 1d
e = (e e ) = 0, (3.98)
dt 2 dt
because e e = 1. Finally, we have
( )
du db dv E
= e v + . (3.99)
dt dt dt
For the projection into the b direction we have:
[ ]
1[ ] d(v b) d(u e ) dv E
E b b + u (e B) b = + + b. (3.100)
dt dt dt
The second term on the right-hand side of (3.100) is zero because (e B) b =
(b B) e = 0. Writing out the terms on the right-hand side explicitely, we obtain
E dv db du de dv E b db
=bb + v b + b e + u b + vE . (3.101)
dt dt dt dt dt dt
The second term on the right-hand side of (3.101) is zero from the same arguments as
in (3.98). The third term vanishes as b e and the fifth term also vanishes, because
v E b. The fourth term may be rewritten as
du de b db db
b e = u u e = u e . (3.102)
dt dt dt dt
3.6. GUIDING CENTRE MOTION 54

Finally, we have
dv E db
= + (u e + v E ) . (3.103)
dt dt
The most intense algebra concerns the projection into the e direction, where we
will evaluate the individual terms one by one.

[ ] 3 4 5
1 1
2
d(v b) d(u e ) dv E
E b e + u (e B) e =e + e + e . (3.104)
dt dt dt

1 :
b e b e = 0 (3.105)

2 :


sin cos
u (e B) e = u (e e ) B = u cos sin B =
0 0

0 0
= u 0 0 = u B (3.106)
1 B

3 :
d(v b) db dv

e = e v b
+ e (3.107)
dt dt  dt

4 :

d(u e ) d(e1 cos + e2 sin ) du

e = u (e2 cos e1 sin ) + e e
 =
dt [ dt  dt ]
de1 d de2 d
= u (e2 cos e1 sin ) (cos e1 sin + sin + e2 cos ) =
dt dt dt dt
[   

de1 de1 d  d
= u cos2 e2 
sincos e 1 + sincos   e1 e2 + sin2 e1 e1 +
dt  dt  dt dt
   ]

 de2 de2 2 d 
d

+sincos e sin 2
e + cos e e sin 
cos  e e 2 =

2 1 2 2 1
 dt dt dt dt
[ ( ) ( )]
d de1 de2 de1
2
= u (sin + cos ) 2
+ e2 sin e1
2
e2 =
dt dt dt dt
[ ( )]
d de1 d(e1  e2) de
 de

= u + e2 sin
2  e2
1
+ e2
1
, (3.108)
dt dt dt  dt  dt

where in the last-but-one step we added and subtracted the term sin2 e2 de1
dt .

5 remains unchanged
55 3.6. GUIDING CENTRE MOTION

Hence, together we have

( )
d B de1 1 db dv E
= e2 e v + . (3.109)
dt dt u dt dt
Let us repeat that now we have again six dierential equations for modified variables
the position of the gyration center R and for variables in the plane perpendicular to
the field lines. Thus we have to solve the following set of equations:
( )
dR d b dv E
= v b + v E + u + b , (3.110)
dt dt B B dt
( )
du db dv E
= e v + , (3.111)
dt dt dt
dv E b db
= + (u e + v E ) , (3.112)
dt dt ( )
d B de1 1 db dv E
= e2 e v + . (3.113)
dt dt u dt dt
(3.114)
To proceed further, we need to replace the total time derivative by the derivative
along the trajectory
d ( )
= + v b + v E + u e (3.115)
dt t
Let us do the proper derivation for (3.110). Lets first decompose u according to
formula (3.87) and use relation (3.115) to express d/dt in (3.110):
[ ]
dR ( ) ( )
= v b + v E + v b + u e + v b + v E + u e b+
dt B t
[ ]( )
( ) ( ) 1
v b + u e b + v b + v E + u e +
B t B
[ ]
( )
b + v b + v E v E + b (u e ) v E . (3.116)
B t B
Next, we add a zero term
[ ]
( )
bb + v b + v E v = 0 (3.117)
B t
to the right hand side of (3.116). Then, we drop several terms containing b b and
rearrange the equation to obtain
dR ( )
= v b + v E + b Dt v b + v E + v b [u e ] b+
dt [ B B]
u ( )
e + v b + v E + u e b
B t
[ ]
u ( )
e b + v b + v E + u e B + b (u e ) v E , (3.118)
B2 t B
3.6. GUIDING CENTRE MOTION 56

where we denote
( )
Dt + v b + v E . (3.119)
t
Finally, we have to average (3.118) over .
The averaging has a mathematically correct basis. The set of equations (3.110) to
(3.113) can be symbolically written in a form
dz
= fz (z) = fz + fz , (3.120)
dt
where z {R, v , u , }. The angled brackets denote the average over the gyroangle ,

hence . = (2)1 d(.). The function fz is periodic in the fast-oscillating gyroangle
. Such separation may always be done. We will show that the oscillating component
can be removed to the leading order by means of a small parameters by redefinition of
the variables z to

z = z + fz ( )d . (3.121)
B 0
From (3.113) we have:
( )
d B de1 1 db dv E B
= e2 e v + = + O(0 ). (3.122)
dt dt u dt dt

The total time derivative (3.115) in terms of the guiding centre variables sounds

d dR dv du d
= + + + + . (3.123)
dt t dt R dt v dt u dt
d
The last term is the leading order term, where for dt we use (3.122). Note that actually
dv E
the third term on the right-hand side is a leading order term ( dt O(1 )),
however we silently assume here that E O(), othewise the assumption of small
changes get violated. Hence
d B
= + O(0 ). (3.124)
dt
Note that such an approximation is far from being general and is valid only for the
system we investigate.
By applying the leading order time derivative (3.124) to the transformed variables
(3.121) we have

d
z dz B
= fz ( )d =
dt dt B
dz
= fz = fz (z) + fz fz = fz (z) = fz (
z ) + O(), Q.E.D. (3.125)
dt
Let us go back to averaging (3.118) over . For the terms independent of e , this
step does not lead to any change. On the other hand, terms linear in e are zero as
57 3.6. GUIDING CENTRE MOTION

sin = cos = 0. The terms quadratic in e are non zero in general and have to
be treated carefully. Dropping for better readability notation of the mean over , i.e
., on the right hand side, we write

dR ( ) u2 u2
= v b + v E + b Dt v b + v E + e (e ) b e b (e ) B .
dt B B B2
(3.126)
Now, let us consider the last term in (3.126). Using a notation 1 e1 and
2 e2 we may write:

(e b) (e ) B = (e1 sin e2 cos ) (cos 1 + sin 2 ) B (3.127)

Considering sin cos = 0 and sin2 = cos2 = 21 , averaging of (3.127) over

gives:

(e b) (e ) B = 1
2 (e1 2 B e2 1 B) = 21 (b B) . (3.128)

Note that an alternative derivation of this relation can be found in Appendix A.1.
Similar approach will be applied to the fourth term of the right hand side of
eq. (3.126). In this case, we have to keep in mind that the vector field b has com-
ponents (0, 0, 1) at the point R though, but derivatives of all of its components are
non-zero in general. Hence,

e (e ) b = (e1 cos + e2 sin ) (cos 1 + sin 2 ) b

= (e1 cos + e2 sin ) [e1 (cos 1 + sin 2 ) b1 +
e2 (cos 1 + sin 2 ) b2 + e3 (cos 1 + sin 2 ) b3 ]
= e1 [sin (cos 1 + sin 2 ) b3 ] e2 [cos (cos 1 + sin 2 ) b3 ] +
e3 [cos (cos 1 + sin 2 ) b2 sin (cos 1 + sin 2 )] b1(3.129)
.

Averaging over then leads to

e (e ) b = 1
2 (e1 2 e2 1 ) b3 + 12 e3 (1 b2 2 b1 ) . (3.130)

First term on the right hand side of eq. (3.130) may be neglected as i b3 = O((i b[1,2] )2 ).4

The fact that i b3 = O((i b[1,2] )2 ) is a consequence of normalisation |b| = 1. Vector b at a small
4

displacement R from R can be expressed by means of Taylor series:

b(R + R) = b(R) + R b(R) + O(R2 ) . (3.131)

Denoting b b(R + R) and bi R bi (R), we write b = b + b1 e1 + b2 e2 + b3 e3 . We further

consider b = (0, 0, 1) and both b and b be normalised. Hence |b |2 = 1 which implies
 2
 0 b1 
 
|b | = 1 =  0 + b2  = b21 + b22 + (1 + b3 )2 b21 + b22 + b23 + 2b3 + 1.
2
(3.132)
 1 b3 

and, therefore, 2b3 b21 b22 b23 O(b2[1,2] ), Q.E.D.

3.6. GUIDING CENTRE MOTION 58

The second term on the right hand side of eq. (3.130) can be rewritten in the vector
notation:
e (e ) b = 21 e3 (e3 b) = 12 b (b b) , (3.133)

where we used the identity b e3 . Note that an alternative derivation of this relation
can be found in Appendix A.1. Inserting (3.128) and (3.133) gives

4 5

dR 1
2
3
( ) u2 u2
=v b + v E + b Dt v b + v E + (b b) b + b B .
dt B 2B 2B 2
(3.134)
This is the equation for the guiding centre motion. It contains all drift we discussed in
the previous chapters and much more.

1 This term represents the movement along the field lines. Local vector b is curved in
(x, t) space, nevertheless the motion of the particle is always dominantly parallel
to the local vector of the magnetic induction.

2 This is the E-B drift.

3 This term contains the generalised curvature drift (term B b (v b )(v b)5 ),
the polarisation drift6 and the eects of the non-homogeneous electric field.

4 This is the correction to the parallel current: b b = 0 j /B.

5 This is the grad-B drift.

Similar equations might be derived for mean values of time derivatives of velocity
5
Deriving it more explicitly, lets assume the parallel velocity to be constant (or weakly variable),
then

qv2 qv2
vc = [b (b )b] = B [b (e1 b1 + e2 b2 + bb3 )] =
mB mB 2
2 2
qv qv
= 2
B b3 = B R, (3.135)
mB mB 2 R2
where we noted that in the local cylindrical coordinates, b3 is along the constant radius R and in order
to fulfill B = 0 in a vaccuum the field must decrease as 1/R and hence the gradient of the field must
decrease as 1/R2 R. This relation is exactly the same as (3.51).
6
Let us handle the third term of 3.134 a little more in detail, in each step always by focusing to one
particular term, while leaving the remaining terms unevaluated.
( )
( ) dv E m d EB m E
b Dt v b + v E = b +A= B + A = B B +B =
B B dt qB 2 dt B 2 qB 4 t
( )
m E E m E sgn q E
= B B B B +B = +C = + C, (3.136)
qB 4 t t qB 2 t c B t

where the first term is the polarisation drift

59 3.7. MAGNETIC MIRRORS

components (see Appendix A.2), namely:

dv E u2
= B + v E Dt b (3.137)
dt 2B
du v u u
= B ( v E b v E ) (3.138)
dt 2B 2
d B v
= e2 Dt e1 b (v b + v E ) (3.139)
dt 2
The equations we obtained are still expressed in the phase variables (x, v). In order
to obtain the closed set of equations, equations (3.134) to (3.139) need to be expressed
in guiding centre variables (R, v , u , ). It can be done only approximately. First, the
fields E and B (and consequently vectors b, e1 , e2 , and v E ) at the particle position are
expanded in a Taylor series around the position of the guiding center. For instance:
u
b(x, t) = b(R, t) + b(R, t) + = b(R, t) + e b(R, t) + . . . (3.140)
B
For the averaged variables even the first order terms vanish, because e = 0 and thus
the system of equations keeps to the leading order the same shape also for the guiding
centre variables.

3.7 Magnetic mirrors

We deliberately did not study the motion of the testing particle in the case of inhomo-
geneous magnetic field with |B| B. We will derive some useful properties of this
configuration of the magnetic field now. Let us consider the axisymmetric configuration
in cylindrical coordinates, where B = (Br , B , Bz ) and B = 0 and / = 0. From
the Gausss law for magnetism
1 1 B Bz
B = rBr + + =0 (3.141)
r r r z
we immediately see that
r
Bz
rBr = r dr . (3.142)
z
0
Bz
When assuming that we know on the axis and consider only particle motions close
z
enough to the axis, we may neglect the radial dependence of this term and obtain
[ ] r [ ]
Bz 1 Bz
rBr r dr = r2 , (3.143)
z r=0 2 z r=0
0

and hence [ ]
1 Bz
Br r . (3.144)
2 z r=0
3.7. MAGNETIC MIRRORS 60

Thus we have an additional component to the Lorentz force, hence we may expect a
drift. It will be a grad-B drift, but not the radial one, since B/ = 0. The Lorentz
force is

vr Br v Bz vz B
F = qv B = q v B = q vz Br vr Bz =
vz Bz vr B v Br
1
v Bz

= q vz Br vr Bz .
2 3
(3.145)

v Br
4

Terms 1 and 3 are not interesting, as they are responsible for the Larmor rotation.

Term 2 vanishes around the axis, however even when we had relaxed the assumptions,

this term would only cause the normal grad-B drift. Term 4 is something new.

Hence we have an additional force

1 Bz
Fz = qv Br = qv r , (3.146)
2 z
which we (in agreement with the procedure already used before) average over one Larmor
rotation. By taking into account that v = sgn qv , r = rL = v /c and c =
q sgn qB/m we have
2
1 v m Bz 1 mv Bz Bz
Fz = qv sgn q = = , (3.147)
2 q sgn qB z 2 B z z

where = 1/2mv 2 /B is the magnetic moment. This equation may be generalised,

when we identify z direction with the parallel direction, hence

B
F = = B, (3.148)
s
where ds is the trajectory element.
We will further show that during the particles motion, its rL changes, but its
magnetic moment remains invariant. First we handle the parallel component of the
equation of motion
dv B
m = (3.149)
dt s
by multiplying it by v :
( )
dv d 1 B B ds dB
mv = mv 2 = v = = . (3.150)
dt dt 2 s s dt dt

The energy conservation law gives

( ) ( )
d 1 2 1 2 d 1 2
mv + mv = mv + B = 0, (3.151)
dt 2 2 dt 2
61 3.7. MAGNETIC MIRRORS

where we used 12 mv
2 = B. Finally we subtract (3.150) and (3.151) to have

dB d dB dB d
0 = + B = + +B (3.152)
dt dt dt dt dt
and hence
d
= 0. (3.153)
dt
The magnetic moment remains invariant. Unfortunately, this conclusion cannot be
considered general, as we took very restrictive assumptions at the beginning. It is
interesting to study to what extent is the magnetic moment conservation valid. Here
we will go back to the guiding centre equations we derived in the previous section. Lets
work with (3.138) in a form

du v u u A
B
= b B ( v E b v E ) + O(). (3.154)
dt 2B 2
Lets trim the terms one-by-one (there are some errors in the following)

EB 1
vE = = 2 [B ( E) E ( B)] =
B2 B
1 B 1
= 2B 2 E ( Bb) =
B [ t B ]
1 B b 1 B
= b b +B + O() = + O(), (3.155)
B t t B t

where we used the assumption that E O() and the already proven relation
that b b
t = 0. Let us remind that the small parameter = m/q was used for
expansion to Taylor series.

b v E = b (b v E ) = b [(b v E ) v E b] =
( ) ( )
B 1 1
= b v E = v E (b B) b v E B =
B B B
1 B
= v E B + v E , (3.156)
2B B
where the first term vanishes, because v E B.

Hence all together (3.154) reads

du v u u 1 B u B
= b B + + vE + O() =
dt 2B[ 2 B t ] 2 B
u u
= + (v b + v E ) B + O() = Dt B + O(), (3.157)
2 t 2B
3.7. MAGNETIC MIRRORS 62

where again Dt is the full time derivative along the guiding centre trajectory.
Hence by taking the time derivative of the magnetic moment = u2 /2B and by
using (3.157) we have

d u du
u2 d 1 u2 u2 dB
= dt
+ = 2 Dt B + O() 2 = O(). (3.158)
dt B 2B dt B 2B 2B dt

Hence the magnetic moment is invariant when the changes of the fields are small (the
assumption of E O().
A measureable consequence of the magnetic moment conservation is the eect of
magnetic mirrors. Lets imagine that a charged particle moves (due to the thermal
motion for example) from the weak into the stronger magnetic field. It sees the
stronger field, hence its Larmor radius decreases and perpendicular velocity increases,
so that remains constant. But also the total kinetic energy must be conserved, hence
the parallel velocity decreases as the particle enters the stronger field. For B large
enough the parallel velocity drops to zero and the particle suers from the reflection
back to the weaker field.
The reflection is not perfect. For example particles with originally v = 0 do not
have magnetic moments and thus penetrate the magnetic trap. They do not feel
the Lorentz force along the magnetic field lines. Similar behaviour may be expected for
particles with small ration v /v . Is there a limit to this ratio?
Lets assume that in the region of the weak field B0 , the particle has the following
values of velocity components: v = v,0 and v = v,0 . In the place of reflection we
have v = v and v = 0 and the field has the induction B . Since

1 1
2
= const = mv,0 /B0 = mv /B and 2
v 2
= v,0 2
+ v,0 v02 , (3.159)
2 2

we have
2
v,0 2
v,0
B0
= 2 = 2 = sin2 , (3.160)
B v v0

where is a pitch angle in the region of the weak field. For small the particle does
not reflect. Hence for some B > Bmax the particle does not escape and reflect. This
allows to set the condition for the minimal pitch angle for the particle to reflect. From
the equation above it becomes

B0
sin2 max = . (3.161)
Bmax

For > max the particle reflects. If we plot all three components of velocity for the
particles, we find that max creates a cone in the phase space. Particles located inside
this cone will escape the reflection. Hence the name loss cone. Note that the shape of
the loss cone depends neither on charge of the particle nor on its mass.
63 3.8. ADIABATIC INVARIANTS

3.8 Adiabatic invariants

Lets have a hamiltonian of the system H = H(p, q) and a system of equations of
motion
q H p H
= and = . (3.162)
t p t q
Should any of the coordinates qi by cyclic, thus
qi (t) = qi (t + T ), (3.163)
where T is the period, then I
pi dqi = const (3.164)

is the integral of motion. For example, in case of the mathematical pendulum, the
deviation (t) = 0 cos t is cyclic with period of T = 2/. Lets define the canonical
pair q = l and p = ml, where m and l are the mass and the length of the pendulum,
respectively. Then we have
T T
pdq = (ml0 sin t)(l0 sin t)dt =
0 0
T
= ml2 20 sin2 td(t) = ml2 0 = const (3.165)
0

When the integral of motion contains a parameter, which is slowly variable (slowly
means that the time scale of the change of this parameter is much larger than period T
of the cyclic coordinate), it turns into the adiabatic invariant. In case of the pendulum,
the invariant is conserved when we slowly change the length of the pendulum. Clearly,
when l increases, must decrease in order for the invariant to remain constant. That
is true only if the time scale, with which the length of the pendulum changes, is much
longer that the period T of the cyclic coordinate. Should the time scale of the length
changes be too short, the system behaves in a completely dierent regime.

3.8.1 First adiabatic invariant

Charged particles moving in the background electric and magnetic field have coordi-
nate , which is obviously cyclic and connected to Larmor rotation, see (3.87). The
canonically conjugated momentum is an angular momentum p = mv rL . Hence
I 2
mv 4m
J1 = pi dqi = mv rL d = 2v = = const. (3.166)
q sgn qB q sgn q
0

It basically proves what we have derived in the previous chapters: for small changes is
the magnetic moment an adiabatic invariant.
There are some important situations in which the magnetic moment is not invariant:
3.8. ADIABATIC INVARIANTS 64

(Figure here)

Figure 3.7: To the second adiabatic invariant.

Magnetic pumping: If the collision frequency is larger than the pump frequency,
is no longer conserved. In particular, collisions allow net heating by transferring
some of the perpendicular energy to parallel energy.
Cyclotron heating: If B is oscillated at the cyclotron frequency, the condition
for adiabatic invariance is violated and heating is possible. In particular, the
induced electric field rotates in phase with some of the particles and continuously
accelerates them.
Magnetic cusps: The magnetic field at the center of a cusp vanishes, so the cy-
clotron frequency is automatically smaller than the rate of any changes. Thus
the magnetic moment is not conserved and particles are scattered relatively easily
into the loss cone.

3.8.2 Second adiabatic invariant

Let us consider a system with two magnetic mirrors and the particles out of the loss
cone, thus cycling from one mirror to another. This motionHis cyclic, thus this motion
b
would in the normal system have an motion integral I = a mv ds. In plasmas the
motions of particles are subjects of drifts, hence the absolute periodicity is violated.
We would assume that the motion integral becomes an adiabatic invariant, when the
eects of the drifts are much smaller than the dominant motion of the system (reflection
between two mirrors).
We will derive an longitudinal invariant J, which is defined only between two con-
secutive reflections, hence
b
J2 = v ds (3.167)
a
and we will show that J is invariant in the stationary magnetic field. Further, we will
generalise this case when Bt is small. When the distance between the two mirrors
shortens, the reflecting particles gain kinetic energy, which is the basic of the Fermi
acceleration.
Lets study the system of an asymmetrical magnetosphere, when due to the drifts the
particle occupies dierent geometries (a dierent field line). The particle drifts from
one field line to another, which has a dierent curvature Rk in time t (see Fig. 3.7).
Then for small angles we have
s s
= . (3.168)
Rk Rk
The fractional change of the trajectory is
s s R Rk Rk 1 s 1 ds
= k = v gc 2 = = , (3.169)
ts tRk Rk s t s dt
65 3.8. ADIABATIC INVARIANTS

where we used the radial component of the curvature drift to estimate the speed of jump
between the field lines  
 Rk Rk  Rk
 
 t  = v gc Rk . (3.170)

We studied the drift speed in the curved magnetic field earlier, thus we only apply
results from (3.67) here:
2
1 B B mv Rk B
v gs = sgn qv rL + . (3.171)
2 B2 q Rk2 B 2
The second term does not have any component along Rk , thus we do not have to
consider it further. Hence
2
1 ds 1 m v Rk
= 3
(B B) , (3.172)
s dt 2 q B Rk
which represents the fractional change of s from the particles point of view. Further,
we will investigate the change of v . Lets use the total energy of the particle W ,
1 1 1
W = mv2 + mv 2
= mv2 + B = W + W . (3.173)
2 2 2

The parallel velocity is then v = m2
(W B) and its time derivative

dv 1 1 dB 2
= ( ) (3.174)
dt 2 2 (W B) dt m
m

and the fractional change of v is

dv
1 dB dB
dt
= dt
= dt2 , (3.175)
v 2 W B mv

where we used (3.173) to express W B. The field B is stationary, but its derivative
(seen by the moving particle) is generally not vanishing due to the drift :

dB dB dr mv2 Rk B
= = v gc B = B. (3.176)
dt dR dt q Rk2 B 2

We insert this useful relation into (3.175) and have

dv
(Rk B) B 2
1 mv (B B) Rk
dt
= 2 = . (3.177)
v q Rk B 2 2 Bq Rk2 B 2
The goal is to evaluate the fractional change of v s during the exchange of trajectory:

1 d(v s) 1 ds 1 dv
= + . (3.178)
v s dt s dt v dt
3.8. ADIABATIC INVARIANTS 66

The first term we already evaluated in (3.172) and the second in (3.177). The both
equal except for the sign, hence
1 d(v s)
=0 (3.179)
v s dt
and v s remains constant when the particle exchanges the field line. This is not the
b
claim we wanted to proof, which was that J = a v s is constant. However, we may
split this integral into three terms,
a b b
J2 = v s + v s + v s, (3.180)
a a b

representing the dierent reflection points for the dierent field lines. When a and a
are close and b and b are also close, then the contributions of the first and third terms
are negligible and J remains approximately invariant. The smallness of these boundary
contributions are emphasised by the fact that v is small near the reflection points
(where the total velocity amplitude is dominated by the perpendicular component).
By replacing the local longitudinal velocity by its average v over the distance
between the two reflection points L, we have
J2 = v L, (3.181)
which is invariant. When L decreases, v increases, which is the basic of the Fermi
acceleration. High-energetic particles of the kosmic radiation of the galactic origin were
mostly accelerated by this mechanism, e.g. by repeated reflections between magnetic
field embedded in the intelstellar gaseous clouds, which moved towards each other.
In an axially nonsymmetric magnetosphere such as the magnetosphere of the Earth
the gravitation drift drives the ring current, which drives the jumping of the particles
between dierent field lines. The second adiabatic invariant is conserved on both sides
of the magnetosphere.

3.8.3 Third adiabatic invariant

This adiabatic invariant is defined for axisymmetrical magnetic field, when the curvature
drift is also cyclical. Then the total magnetic flux enclosed by a drift surface

J3 = = B dS (3.182)

is invariant. It is associated with the periodic motion of mirror-trapped particles drifting

around the axis of the system. The particle moves on the surface which adjust itself
to variantions in the magnetic field so that the flux enclosed by this surface remains
constant.
When the waves in the Earths ionosphere (usually having long periods) are in
resonance with the drifting particles, the energy of the drift converts to the energy of
these waves. The drift is relatively slow, thus in practice, is not conserved in real
systems.
67 3.9. VAN ALLEN RADIATION BELTS

3.9 van Allen radiation belts

We will apply what we have just learned to a realistic problem, which is the particle
trapping in the axially symmetric magnetosphere. This is the reason why there are
radiation (known as van Allen) belts in the Earths magnetosphere and also in the
magnetospheres of other planets. We do not aim to solve the full problem, we aim
more to show the basic principles using a rather simplistic approach. We will search for
allowed trajectories of the particles.
Let us assume the axially symmetric magnetic field with components

B = (BR (R, z), 0, Bz (R, z)) , (3.183)

where Br and Bz are the radial and vertical components of the magnetic field induction
vector. The Gauss law must be fulfilled also in this case, hence
1 Bz
B =0= RBR + . (3.184)
R R z
Such condition is fulfilled when the radial and vertical components of the magnetic
field are given as
1 F 1 F
Bz = and BR = . (3.185)
R R R z
The function F is called the flux function and they can be explicitly computed for the
given configuration of the magnetic field. E.g., for a dipole

R2
F =M , (3.186)
(R2 + z 2 )3/2

where M is the magnetic moment of a dipole7 . For now we can take F as a given
function, by taking the spatial derivatives of which we obtain the magnetic induction
components. It may easily be shown that definition (3.185) fulfils the Gauss law.
Now let us deal with the particle equation of motion in the cylindrical coordinates
and we will be interested in the azimuthal component of it:
( ) ( )
dv q F F q dF
m = q (vz BR vR Bz ) = vZ + vR = , (3.187)
dt R z R R dt

where we used vz = dz dR F F
dt and vR = dt and assumed that t = = 0.
Let us look at the coordinate system we have. We defined two linear coordinates R
(radial coordinate, distance from the axis) and z (height) and one angular (the azimuthal
angle) . The studied particle has a position vector r. Obviously, the unity vector which
is tangential to the azimuth at the given point is parallel to the cross-product of the
position vector r and the unity vector in the z-direction, hence

e ez r. (3.188)
7
For instance, the dipole magnetic moment of the Earth is ME = 8 1015 T m3 .
3.9. VAN ALLEN RADIATION BELTS 68

But what is it amplitude? This we easily determine from the amplitude of the cross
product, which is
|z||r| sin = R, (3.189)
where is an angle between ez and r (also termed as co-latitude). Altogether we have

Re = ez r. (3.190)

Then
Rv = Re v = (ez r) v. (3.191)
Now we compute the time derivative of this expression. We have
d dv dr
Rv = (ez r) + (ez ) v. (3.192)
dt dt dt
The second term vanishes as dr
dt = v and the result of the cross product is perpendicular
to v. By using (3.190) we have
( )
d dv dv
Rv = Re =R . (3.193)
dt dt dt

Finally let us investigate the expression

( )
d qF dv qF
Rv =R = 0, (3.194)
dt m dt m

where we used (3.193) and (3.187). Hence

qF
Rv = I1 (3.195)
m
is a motion integral.
Another motion integral is the total kinetic energy (we assumed that no dissipation
takes place), which may be written in a simplistic form as

v 2 = I22 (3.196)

Certainly, v2 v 2 , as v is only one component of v.

Then
qF
R|v | |v|R |I1 + | RI2 . (3.197)
m
We will not solve this inequation explicitly but will rather point out some general
properties of it by drawing a graph (see Fig. 3.8). On the horizontal axis let us represent
the radial coordinate R, on the vertical axis then the value of the expression qF m (R, ),
where again represents the co-latitude (how precisely did we get to the spherical
coordinates derive it from stating the F for the dipole). The solutions for a given
combination of q, F , and m will show as the hyperbola-like curves for varying in the
canvas. The degenerate limiting = 0 curve coincides with the coordinate axes.
69 3.9. VAN ALLEN RADIATION BELTS

Figure 3.8: A graphical solution of the trapped particles in the axisymmetrical mag-
netosphere. In the shadowed regions, the particles are allowed due to the unequality
(3.197). For a given , the particle may move along the given field line. Hence the total
allowed region is given by the intersection of the two (denoted by the solid thick line).
Such a region obviously depends on the co-latitude and the energy of the particle.

The remaining terms in the expression will create a sector on the canvas, apex of
which will be at a coordinate (0, I1 ) and the opening angle will correspond to I2 (the
larger the more is the sector opened). Particles with a given kinetic energy and the
value of I1 are now allowed to move in this diagram freely along their corresponding
qF
m -line for a given , but only in the limits determined by the sector.
This simplistic approach gave us some important physical predicitons. Particles with
larger speeds are allowed to more within the larger range of radial distances. At lower
co-latitudes (hence near the poles) the particles are in general located nearer the surface
(lower R) and the extend of their allowed radial coordinate is lower (the belt is thinner)
than near the equator. These predictions are fully consistent with the observations of
the real van Allen radiation belts.
3.9. VAN ALLEN RADIATION BELTS 70
Chapter 4

Plasma as fluid

In Section 1.5 we derived the moments of the Boltzmann equation by neglecting the
particle distribution function and by considering the physical parameters averaged over
the ensemble of the particles. Plasma may be treated the same way, however, we have to
bear in mind that the individual particles are charged and thus must follow additional
set of physical laws the Maxwell equations.

4.1 Double-fluid approximation in physics of plasmas

When we consider electrons and ions non-interacting or at least weakly interacting, we
may separate the particle distribution function to two terms, describing independently
electrons and ions. Hence we have two independent fluids in the plasma, the electron
fluid (described by ne , ue , pe ,. . . ) and the ion fluid (fully described by ni , ui , pi ,. . . ).
For each type of particles , where {e, i} we have the equation of continuity, the
Euler equation, and the equation of state, hence

n
+ (n u ) = 0 (4.1)
[ t ]
u
m n + (u )u = q n (E + u B) p (4.2)
t
p = C(m n ) , (4.3)

where is the ratio of specific heats and C is a constant. This set of equations must
be accompanied by the Maxwell equations

0 E = n i qi + n e qe (4.4)
B
E = (4.5)
t
B =0 (4.6)
1 E
B = ni qi ui + ne qe ue + 0 . (4.7)
0 t

71
4.2. HIGHLY IONISED PLASMAS 72

That is all together 18 equations (by considering the vector equations as a triplet) for
16 variables (for each electrons and ions we have n, ux , uy , uz , and p and we must
add components of the magnetic and electric fields). The disharmony is only apparent.
Equations (4.4) and (4.6) play the role of boundary conditions and are not independent
from (4.5) and (4.7). For instance, when we apply the operator of divergence to (4.5),
we have
E
( E) = = B. (4.8)
t t
The left-hand side is a vanishing vector identity, hence

B = const, (4.9)

and our universe is set up so that this constant equals to zero. The physical meaning
of this boundary condition is that there are no magnetic charges.
The equation of state (4.3) requires some commentary. Note that when dropping
the index , we define the mass density as = mn. The polytropic equation of state in
the new variable reads
p
p = C or = . (4.10)
p
The value of varies with the type of the process in question. E.g. for the isothermic
compression, we have = 1. For the adiabatic expansion, = (N + 2)/N , where N is
the number of degrees of freedom. Note that = 5/3 for the adiabatic process of the
ideal gas.

4.1.1 Plasma approximation

In solving the problems using the fluid approximation, we often use the plasma approx-
imation. That essentially means that we neglect the possible concentration of charges
occuring on small scales, and consider large scales only. Due to the requirement on
the bulk neutrality of plasma, we must assume that the particle density of both fluids
equals, hence ne = ni = n. In the process of deriving useful relations we then do not
use the Poisson equation (4.4), which would otherwise lead to a discrepancy, because in
details, E = 0, which is not true due to the collective behaviour. We will show later
that the plasma approximation is valid only when scales larger than Debye shielding
length are considered.

4.2 Highly ionised plasmas

The need to capture the particle interactions on scales smaller than the Debye length
leads naturally to additional term in the Euler equation. Such term is termed collisional
term, eventhough strictly speaking it is not the same as the collisional term in the
Boltzmann equation we derived earlier. The Euler equation for electrons then reads

due
me ne = ene (E + ue B) pe + Pei . (4.11)
dt
73 4.2. HIGHLY IONISED PLASMAS

(figure here)

Figure 4.1: The geometry to estimate the specific rezistivity

The term Pei represents the momentum change due to the collisions with ions. Due to
the momentum conservation law, it is clear that

Pei = Pie . (4.12)

Heuristically, the collisional term may be written as

e2 ne ni (ui ue ) e2 n2 (ui ue ),
Pei = |{z} (4.13)
|{z} |{z} | {z }
5 1 2 3
| {z }
4

where we may notice the construction of the expression from the following

1 The interaction is coulombic, hence proportional to the multiplication of charges of

both species.

2 The rate of collisions is proportional to the electron flux.

3 Relative speed of the fluids.

4 Describes how often do the electrons meat ions.

5 The scaling parameter (describing the eciency of the collisions), it has a physical
meaning of the specific rezistivity.

Alternatively,
Pei = me n(ui ue )ei , (4.14)
where ei is the collisional frequency, with which the electrons hit the ions. Such expres-
sion may also be viewed phenomenologically: the expression me n(ui ue ) physically
means the momentum exchanged due to the collisions, which is normalised by the
collisional frequency (as the Euler equation evaluates the change of the momentum
in time). By comparing the two formulations of Pei we obtain the expression for the
collisional frequency as
ne2
ei = = p2 0 , (4.15)
me
where p is a plasma frequency (to be derived in Section 5.4).
Let us estimate the specific rezistivity from a model of the electron squeezing
around the ion and hence losing some of its momentum. See Fig. 4.1. The interaction
is driven by the electric force, hence the force is

e2
F = , (4.16)
40 r2
4.2. HIGHLY IONISED PLASMAS 74

where r is the distance between the ion (which is considered fixed in this coordinate
system) and the passing electron. We roughly estimate that the interaction takes the
total time T , where
r0
T = , (4.17)
v
where r0 is the impact parameter and v is the mutual speed of the electron and ion.
During the interaction, the electrons momentum changes by
e2
(me v) |F T | = . (4.18)
40 r0 v
Let us consider that during the interaction, the electron looses all of its momentum,
hence (me v) = me v. This is the case when the electron is scattered to the direction
perpendicular to the direction of the original motion.
e2
(me v) = me v = , (4.19)
40 r0 v
which may be used to express the corresponding value of the impact parameter r0 . For
this special case of the perpendicular scattering, we will denote this special value of the
impact parameter as r90 :
e2
r90 = . (4.20)
40 me v 2
Lets define the eective cross-section of the interaction as

2 e4
= r90 = . (4.21)
1620 m2e v 4
Then the collisional frequency (how many collisions in a unit of time) may be expressed
as
ne4
ei = nv = . (4.22)
1620 m2e v 3
If we further relate the typical speed of the electrons and their temperature by me ve2
KB Te , we obtain for the specific rezistivity
1/2
ei me e2 e2 m e
= = = . (4.23)
ne2 1620 me v 3 1620 (KB Te )3/2
So far we have considered only the perpendicular scattering. To account for all

impact parameters, we must integrate over all of them, 0 1/r dr. Unfortunately, such
function diverges, and the workaround is to limit the integration by the lower and
upper bounds from the physical view. The lower bound is obviously the perpendicular
scattering. The upper bound is the Debye length, since on larger scales, the plasma is
neutral. Then we have
D
1 D
dr = [ln r]r90
D
= ln = ln , (4.24)
r r90
r90
75 4.3. SINGLE-FLUID MODEL

where is the coulombic logarithm. This brings a correction to what we have previously
derived, hence the full expression of the specific rezistivity reads
1/2
e2 me
= ln . (4.25)
1620 (KB Te )3/2

Note that

does not depend on density. The larger density of the charge carriers is ballanced
by the higher rate of collisions.

T 3/2 , hence for large temperatures the specific rezistivity is very small. As a
consequence, if we want to heat the plasma via the Joule heating from the electric
current flowing through the plasma, the processes get very inecient when large
temperatures are reached, hence one cannot obtain the temperatures necessary
for thermonuclear fussion using such energy pumping. Dierent mechanism must
be used.

For KB Te 100 eV the plasma specific rezistivity is 5 107 m, which is of the

same order as the rezistivity of metals such as copper.

Also note that

Pei = e2 n2 (ui ue ) = ene E, (4.26)
when
E = en(ui ue ) = j, (4.27)
which is the Ohms law.

4.3 Single-fluid model

Even a rougher approximation exists, when we stop considering two charged species in
the plasma a single-fluid approximation. The two species are electrons (having the
particle mass m) and ions (having mass M ). The governing equations of the single-fluid
approximation are obtained by linearly combining the equations for double-fluid approx-
imation. These equations are replenished with the collision terms Pei = e2 n2 (|ui ue |),
where is a specific resistivity, and an ad-hoc term of the gravity force. We further ne-
glect the convective derivative (u)u by assuming the the speeds of fluids are generally
small and hence this is the quadratic term. We further define a the bulk density

ni M + ne m n(M + m) nM, (4.28)

the bulk velocity

1 n(M ui + mue ) M ui + mue

u (ni M ui + ne mue ) = , (4.29)
n(M + m) M +m
4.3. SINGLE-FLUID MODEL 76

the bulk current density

j e(ni ui ne ue ) ne(ui ue ), (4.30)

and a bulk pressure

p = pi + pe . (4.31)
Lets consider the sum of Euler equations for both fluids

ui
Mn = en(E + ui B) pi + M ng + Pie , (4.32)
t
ue
mn = en(E + ue B) pe + mng + Pei (4.33)
t
by using the symmetry Pie = Pei and definitions above we have
( )
M ui + mue
n(M + m) = n (M ui + mue ) = en(ui ue ) B p + n(M + m)g.
t M +m t
(4.34)
which is
u
= j B p + g, (4.35)
t
which is the equation of motion for sigle-fluid plasma.
Lets consider a dierent combination of the same equations. Lets multiply (4.32)
by m and (4.33) by M

ui
mM n = emn(E + ui B) mpi + mM ng + mPie , (4.36)
t
ue
mM n = eM n(E + ue B) M pe + mM ng + M Pei . (4.37)
t
Now subtract the two. We have

nmM (ui ue ) = en(M + m)E + en(mui + M ue ) B
t
mpi + M pe (M + m)Pei . (4.38)

Now ui ue may be expressed using the bulk electric current and

j
mui + M ue = M ui + mue + M (ue ui ) + m(ui ue ) = u (M m) . (4.39)
n ne
and also
Pei = e2 n2 (ui ue ) = enj (4.40)
By further dividing (4.38) by en(M + m) = e we have
[ ]
1 M mn j
E + u B j = + (M m)j B + mpi M pe . (4.41)
e e t n
77 4.4. APPROXIMATIONS TO PLASMA FLUID

For m/M 0 and when

M mn j m j j
= M nB = M nBc1 (M m)j B, (4.42)
e t n eB t n t n

which is fullfilled when the time variations of the current density are much slower than
the cyclotron frequency, we obtain

1
E + u B j = (j B pe ) , (4.43)
en

which is the generalised Ohms law. The first term on the right-hand side is the ex-
pression for the Halls current. In various applications, the right-hand side is often
completely neglected.
Similarly (and we leave the derivation to the reader) one may obtain the continuity
equation for density and charge density e = e(ni ne ) for single-fluid plasma by
summing and subtracting the continuity equations for both fluids:

+ u = 0 (4.44)
t

and
e
+ j = 0. (4.45)
t
Hence the full set of equations for a sigle-fluid plasma reads

u
= j B p + g, (4.46)
t
E + u B = j, (4.47)

+ u = 0, (4.48)
t
e
+ j = 0. (4.49)
t

This system must be accompanied by Maxwell equations to fully describe the plasma in
equilibrium, where the definitions for e and j are simply used in Equations (4.4)-(4.7).
This approximation is used usually then the plasma resistivity shows in the problem
and is often used also in astrophysics. The system of single-fluid equations is most often
understood under the term magnetohydrodynamic equations or MHD equations.

4.4 Approximations to plasma fluid

Additional approximations to plasma fluid approximation are used in some cases.
4.4. APPROXIMATIONS TO PLASMA FLUID 78

4.4.1 Force-free and ideal MHD

The force-free approximation is valid for systems, where locally the plasma does not
feel the Lorentz force, i.e.
fLorentz = e E + j B = 0. (4.50)
This approximation is also used in the case when the Lorentz force is negligible compared
to other considered forces. Other view to this approximation may also be formulated in
such a way that the electromagnetic field induced by the motion of plasma fluid exactly
and immediately cancels out the external forces.
The ideal MHD approximation is somewhat dierent. It uses the fact that in the
comoving frame the electric field disappears. I.e.
E = E + u B = 0. (4.51)
By using the generalised Ohms law it turns out that the ideal MHD approximation
equals to the case when the plasma is ideally conductive, therefore when the resistivity
is zero. This may be immediately seen from
E = j, (4.52)
where the left-hand side is zero because of (4.51) and this relation may be valid in a
trivial case when j = 0 of when = 0 (which represents an infinite conductivity).
The two approximations mentioned in this subsection are usable in the case when
the electromagnetic field is degenerated, i.e. when E B = 0. The two coincide when
j = e u.

Frozen field
The consequence of the ideal MHD approximation is something called frozen field. It
is a description of an observed behaviour of highly-conductive plasmas, where a strong
bound between the plasma and the field exists. It was found responsible for strenght-
ening the magnetic field in the stars or temporal changes of the Earths magnetosphere
under the influence of the varying solar wind.
Let us start from taking the curl of the Amp`eres law:
( )
E
( B) = B B = 0 j + 0 . (4.53)
t
The term B vanishes due to the Gausss law for magnetism. By using the ideal
MHD approximation (4.51), Eq. (4.52), and the definition of 1/ we have
E
0 (E + u B) + 0 = B. (4.54)
t
For must (E + u B) = 0 in order to B be finite. By additionally
using the Farradays law we then have
B B
+ (u B) = 0 (u B) = . (4.55)
t t
79 4.4. APPROXIMATIONS TO PLASMA FLUID

Drawing to a frozen field approximation

Figure 4.2: caption

And important note here. Stating that (E + u B) = 0 does not automatically

mean that E + u B = 0. A mathematical object being zero does not necessary have
to have vanishing derivatives!
The frozen-field approximation may also be understood as a condition, when the
magnetic flux through a closed curve bound to the plasma does not change with time.
Let us prove this statement. Lets start from defining the system as in Fig. 4.2. By
zooming in it is obvious that the change of the surface vector element is given by a
cross-product of the plasma velocity and the elementary length along the closed curve,
d
i.e., dt S = u dl.
We will now derive the change of the magnetic flux with time:

d d dS B
= B dS = B + dS. (4.56)
dt dt dt t
We apply the relation between the surface and length elements from above and after
using the cyclic exchange in a mixed product we have
I I
B B
B (u dl) + S = (u B) dl + S. (4.57)
l t l t
We further use the Stokes mathematical theorem and obtain
[ ]
B B
(u B) S + dS = (u B) dS = 0, (4.58)
t t

where we used (4.55). Q.E.D.

4.4.2 Axisymmetrical systems, stream functions

Another class of approximations comes with the symmetries and/or degenerations of
the system. Let us consider the axially symmetric stationary system represented in the

cylindrical coordinates (R, , z). I.e. = 0 and t = 0. The curl of E that has
components
( )
1 Ez ER ER Ez 1 1 ER
E = , , RE = 0. (4.59)
R z z R R R R
From the third component it turns out that E = 0.
By using the ideal MHD approximation we further have

E = (u B) = uR Bz + uz BR = 0, (4.60)

and therefore
uR BR
= . (4.61)
uz Bz
4.4. APPROXIMATIONS TO PLASMA FLUID 80

Thus the poloidal components are bound with an unknown scalar function

up = (R, z)B p , (4.62)

which is a consequence of the axial symmetry, stationarity, and the used ideal MHD
approximation. Such result is therefore not valid generally!
Let us only remind that the poloidal component of a vector a is simply a combination
of the radial a height components, i.e., ap = aR eR + az ez .
Because B = A, where A is a magnetic vector-potential, we may write

(R, z) = RA , (4.63)

where is a scalar magnetic flux function. Then simply

1 1
BR = and Bz = . (4.64)
R z R R

By using the continuity equation

v
+ (up ) + = (up ) = 0 (4.65)
t

we have
(up ) = (B p ) = B p + B p = 0. (4.66)

The last term vanishes due to the Gausss law for magnetism. By decomposing the
divergence operator into the non-vanishing components and using (4.64)

1 1
(B p ) = R ()BR + z ()Bz = R ()z + z ()R = 0 (4.67)
R R

we see that this is the component of an expression R1 () . Due to the

assumptions the arguments of the gradient operators do not depend on . The gradient
therefore has only the component perpendicular to both eR and ez non-trivial. That is
the -component, which also vanished as we proved in (4.67). Thus

() = 0. (4.68)

It may be interpretted as () and therefore the scalar function may be related

to as
4 = F1 (). (4.69)

Function F1 is termed the stream function, which is obviously constant on a con-

stant surface of . Its particular mathematical expression may be determined from the
boundary conditions. The use of the streamfunction is the reduction of variables to be
solved for the given system as it introduces a bound between u and B.
81 4.5. DRIFTS IN THE PLASMA FLUID

Similarly, using the poloidal and azimuthal (toroidal) decomponsiton and Equations
(4.62) and (4.64) we have

u B = u e B p + up B e = u e B p + B p B e
( )
1
= (u B )(Bz BR )ez = (u B ) ez
R R z
u B
= . (4.70)
R

In a stationary frozen-field approximation we have

B
(u B) = = 0, (4.71)
t

and therefore
( ) ( )
u B u B
= = 0, (4.72)
R R

where we proceeded similarly to the previous case. Again, the surfaces of the argu-
ments of the gradient operators must coalign, and hence we may define a second stream
function as
u B
= F2 (). (4.73)
R
By taking the projection of the Euler equation to e we can show that

B e (RB F1 Rv ) = 0, (4.74)

and hence define the third streamfunction as

RB F1 ()Rv = F3 (). (4.75)

And finally by taking the projection of the Euler equation onto B p we obtain a
definition for a fourth streamfunction as

1 2 dP
u + + Ru F3 () = F4 (). (4.76)
2 =const

Alltogether the steamfunctions provide a technology for seaking the solution of

the system in a given geometry.

4.5 Drifts in the plasma fluid

Lets discuss the drifts in the plasma fluid.
4.5. DRIFTS IN THE PLASMA FLUID 82

4.5.1 Drifts perpendicular to B

When investigating the drifts in the direction perpendicular to B, we solve the perpen-
dicular component of the Euler equation, which for collision-less plasma reads:
( )
u
mn + u u = qn(E + u B) p (4.77)
t

Let us consider systems, where the left-hand side is negligible to the right-hand side.1
Hence we solve a much simpler equation

0 = qn(E + u B) p (4.78)

by applying the vector multiplication by B:

0 = qn[E B + (u B) B] p B =
= qn(E B uB 2 + Bu B) p B. (4.79)

The perpendicular component of this equation is

qn(E B u B 2 ) p B = 0 (4.80)

and thus
E B p B
u = (4.81)
B2 qnB 2
We found the expression for the perpendicular drift in the plasma fluid. The first term
resembles the already known E-B drift, the second term is new and represents the
diamagnetic drift

4.5.2 Drifts parallel to B

Now lets look at the remaining component of the drifts in a plasma fluid. Lets consider
the coordinate system so that the axis ez is parallel to the magnetic field. Then we
study the z-component of the equation of motion
( )
uz p
mn + (u )uz = qnEz (4.82)
t z

with equation of state in a form

p n
= , (4.83)
p n
1
To estimate this, lets use the Fourier transform of Euler equation, for now blindly applying the rules
(5.11) to (5.13). By simply waving the hands we see that the left-hand side is proportional to i + i L1 ,
where is a mode frequency and L is a characterictic scale of spatial change, while the right-hand side
is proportional to ic r1L . Usually, c and rL L, hence the left-hand side is negligible.
83 4.5. DRIFTS IN THE PLASMA FLUID

where we further use = 1, because we will solve the equation only for electrons that are
fast and may be considered isothermic. Additional assumptions apply: (u )uz 0,
uz homogeneous, and p = KB T n. Then
uz n
mn = qnEz KB T (4.84)
t z
For electrons we have q = e, m 0. By using the Poisson equation Ez = /z we
have
KB T n
e = (4.85)
z n z
and hence
e = KB T ln n + C (4.86)
and finally [ ]
e
n = n0 exp , (4.87)
KB T
which is the Boltzmann relation. The Boltzmann relation describes the spatial distri-
bution of electrons around the random concentration of ions. In the local concentration
of ions, the pressure gradient pushes electrons away from this region. The consecutive
charge separation gives rise to the local electric field, which acts against the pressure
gradient. In the equilibrium, both the pressure and electrostatic forces are ballanced.
The Boltzmann relation is a very useful relation when we further solve the system of
equations for ions. Due to the mass dierence between the two species the spatial
distribution of electrons may be approximated by the Boltzmann relation.
4.5. DRIFTS IN THE PLASMA FLUID 84
Chapter 5

Waves in plasmas

5.1 Linear waves

It is not trivial to comprehensively describe solutions supporting waves in plasmas. We
will investigate a few examples only of linear waves. In practice, we will decompose the
variables describing both plasma and magnetic and electric fields into two components:
a background component (having index 0) and a fluctuating component (having index
1), e.g., = 0 + 1 . In (silently) applying this decomposition, we will assume that:

1. The background components automatically fullfil the full set of magnetohydrody-

namic equations. The background is the zeroth order of smallness variable.

2. The fluctuating components are small compared to the background (e.g., 1 /0

1). When averaging the fluctuating component (both in space and/or time), the
averaged value is zero (e.g., 1 = 0). Implicitly, we assume the same about both
temporal and spatial derivatives of the fluctuating component. The fluctuating
part is the first order of smallness variable.

We will solve the set (or subset) of magnetohydrodynamic equations to the first order.
Lets stop for a while and make this decomposition crystal clear for ever. It might get
confusing especially in the case of velocity. When we computed moments of Boltzmann
equation, we performed also a decomposition having a fluctuating part. Lets compare
the two:

v = u + w, (5.1)
u = u0 + u1 . (5.2)

These two are crucially dierent. While in Eq. (5.1) a particle velocity v is decomposed
into a mean (or bulk) velocity u and the component w fluctuating about this mean
value, where the amplitude of the fluctuations may be (and typically is) much larger
than the amplitude of the bulk motion, in Eq. (5.2) the bulk velocity u is decomposed
into the background fullfilling the MHD equations u0 and a small correction to this
background value. Hence u from Eq. (5.2) indeed equals to u from Eq. (5.1).

85
5.2. EXAMPLE: ACOUSTIC WAVES IN FLUIDS 86

5.2 Example: Acoustic waves in fluids

To illustate the linearisation approach, let us derive the dispersion relation for waves in
a fluid, fully described by the Euler equation in the form
[ ]
u p
+ (u )u = p = , (5.3)
t

where on the right-hand side we assumed the adiabatic equation of state p = const,
and a continuity equation

+ (u) = 0. (5.4)
t
We will assume the following:

Pressure waves in fluids

Strictly hydrodynamical case (no fields)

No viscosity

No bulk motion of the background, u0 = 0

The background is homogeneous

By introducing the fluctuating part to the velocity and density, i.e., u = u0 + u1 ,

p = p0 + p1 , and = 0 + 1 Equations (5.3) and (5.4) take the form of
{ }
u0 u1 (p0 + p1 )
(0 + 1 ) + + [(u0 + u1 ) ] (u0 + u1 ) = (0 + 1 ) (5.5)
t t (0 + 1 )
(0 + 1 )
+ [(0 + 1 )(u0 + u1 )] = 0. (5.6)
t
Now, lets take the continuity equation as an example. It can be explicitly written
as
1 2
0 1 3 4 5 6 7
+ + (0 u0 ) + 0 u1 + u1 0 + (1 u0 )+ (1 u1 )= 0. (5.7)
t t
Lets discuss the terms in the above equation:

Terms 1 and 3 automatically fulfill the continuity equation for the background
0
variables: t + (0 u0 ) = 0.

Term
5 is zero because we assumed homogeneous background, i.e. 0 = 0.

Term
6 is zero because we assumed u0 = 0.
87 5.2. EXAMPLE: ACOUSTIC WAVES IN FLUIDS

Term
7 is a second order therm, thus can be neglected.

Thus only relation

1
+ 0 u 1 = 0 (5.8)
t
remains. This is the linearised continuity equation.
Similarly1 , we obtain the linearised Euler equation:
u1 p0
0 = 1 . (5.9)
t 0
Equations (5.8) and (5.9) need to be solved. We are looking for solution supporting
waves, thus we may search for a solution in a form
A = A exp [i(k r t)], (5.10)
where k is a wave vector in the Fourier space and is a frequency. There are two
possibilities how to see this approach:
1. We solve these equations in the Fourier space. At this step, a carefull reader
should open the mathematical textbook and repeat all the assumptions of the
Fourier transform.
2. We view the solution as a superposition of modes, each mode2 is described by a
unique combination of k and . Thus we solve the equations for one general mode
and a complete solution is a superposition of all modes with various amplitudes.
Either way, the beauty of the formulation of the equations in the Fourier space is that the
dierential operators turn into the multiplication. One can easily prove from Eq. (5.10)
that

i (5.11)
t
ik (5.12)
k 2 (5.13)
Our goal is to find a dispersion relation, thus a relation which couples k and .
Dispersion relation D(k, ) = 0 fully describes the given mode of the wave. Equations
(5.8) and (5.9) written in the Fourier space take the following form:
i1 + i0 k u1 = 0, (5.14)
p0
i0 u1 + i k1 = 0. (5.15)
0
1
For clarity, lets handle the fraction on the right-hand side explicitly:
p0 +p1 p0 p1
p0 + p1 0 0
+ 0 p0
= 0 +1 = 1 = ,
0 + 1 0
1+ 0
0

because p01 and 01 are negligible terms.

2
A mode is defined either as any of the wave-like and/or oscillation-like plasma motions, or as one
possible configuration of the travelling or steady wave. Both definitions are equivalent.
5.2. EXAMPLE: ACOUSTIC WAVES IN FLUIDS 88

For simplicity lets define the coordinate system so that the wave propagates in the main
direction, thus wave vector k may be replaced by its scalar equivalent wave number k.
Then by solving the equations above one obtains a dispersion relation

p0
= cs , (5.16)
k 0

where cs is a speed of sound.

A few remarks to what we just obtained:
The dispersion relation exists, thus the waves propagate. The propagation is fully
described by the wave vector k. Possible combinations of the wave vector and
frequency of the mode are bound by the dispersion relation. The waves propagate
in a form of pressure fluctuations.

The wave number of the wave is inversely proportional to its wavelength =

(2)/k.

The phase speed v is a speed with which places with the same phase propagate
through the medium. The phase speed is given by v = /k. In our case, v = cs .

On the other hand, the group speed vg is a speed with which the modulation
information propagates. It also relates the wave number and the frequency of the
mode by vg = d/dk. In our case, again, vg = cs and vg = v . As we will show
later, this does not have to always be the case. It can be illustrated on a very
simple example: we have one mode only with a constant amplitude in time. A
constant beep does not carry any information. Its dispersion relation reads:
= const. Thus its phase speed is not defined (it might be viewed as infinite
as well), but the group speed is zero. We just showed an additional important
property: phase speed does not have to be limited (again, it does not carry any
information), while the group speed is limited (from the top) by the speed of light
c.

The term dispersion relation indicates that the wave disperses, changes the wave
form during the propagation. That indeed is the case, when the wave is dispersive,
i.e., when the propagation phase speed depends on wave frequency or wave num-
ber. Then the amplitude spectrum of the wave packet changes as the packet passes
through the medium individual frequencies exhibit various phase shifts. Then
the wave-packet propagation is not a simple translation parallel with the direction
of propagation, but is more complex and additionally to the movement of the wave
packet the packet itself changes its shape. Dispersive waves usually have group
speed dierent from the phase speed. Sound waves in fluids are non-dispersive.

Note that the approach we will use is equivalent to approximation by the geo-
metrical optics a.k.a. ray approximation, which ignores the eects of the finite
89 5.3. TYPES OF PLASMA WAVES

wavelength. Such approach is justified when the wavelength is much shorter than
the typical spatial scale of the changes in the medium. In the other case, eect
may be observed, which are not captured by the ray approximation. Finite wave-
lenght eect are captured by the scattering theory, which are not easy to be solved
both analytically and numerically. Approximations to the scattering theory, such
as the Born approximation, capture some of the most important finite wavelength
eects quite well.
Let us see what happens when we relax an assuption of u0 = 0. Then the Fourier image
of the continuity equation will read:

i1 + i1 k u0 + i0 k u1 = 0, (5.17)

which may be written as

i1 ( k u0 ) + i0 k u1 = 0. (5.18)

Hence by introducing = ku0 we change the equation to the previous case. Similar
operation may be done to the Euler equation3 . Hence the final dispersion relation will
be
k u0 = kcs , (5.21)
and the motion of the background fluid with respect to the observers frame introduces
a Doppler shift of the frequencies.

5.3 Types of plasma waves

To study the waves in plasma, we will use the fluid approximation. It is fine enough to
capture all essential properties to investigate the possible propagation of waves, but is
also simple enough to be able to obtain some analytical solutions. There is the whole
spectrum of waves propagating through plasma, which may be decompones into several
modes with dierent dispersion relations. We will study only a handfull of the possible
modes, some of which may have important physical consequences.
Plasma waves are usually excited by means of instabilities (this is not surprising, e.g.
sound waves propagating through a convective envelope of Sun-like stars are excited by
the convective instability below the photosphere of the star) or they may be excited
artificially. The dispersion relation does not say anything about the excitation process.
Basic types of plasma waves can be distinguished by given one of the two values to the
following three properties:
Due to u0 = 0 we must keep the convective derivative term in the equation, leading to the
3

linearised form
u1 p0
0 + 0 (u0 u1 ) = 1 , (5.19)
t 0
which has a Fourier form of
p0
i0 ( k u0 )u1 + ik 1 , (5.20)
0
where a possible substitution = k u0 is immediately visible.
5.3. TYPES OF PLASMA WAVES 90

1. In the wave, either electrons or ions oscillate. Thus we have electron or ion wave.
Usually, when we solve for an electron wave, the ions are approximated as a non-
movable background, because ions are at least four order havier than electrons.

2. The wave can be longitudinal or electrostatic when k E 1 or tranversal or elec-

tromagnetic, when k E 1 , where E 1 is a fluctuating component of the electric
field.

3. The wave is parallel when k B 0 or perpendicular when k B 0 , where B 0 is a

background magnetic field.

Plasma waves we will discuss in the following Sections will always be distinguishable by
the three above described tags.
The approach we will take is that we will prescribe expected geometry of the wave (in
terms of the type, direction of propagation and which of the components it contains)
and search for the solution using the linearised fluid and Maxwell equations. If the
solution in a form of a dispersion relation exists, the plasma supports this kind of wave.
Our solutions will not give the answer if this particular wave really gets excited under
given conditions and what is its amplitude (for this one would need to state the initial
or boundary conditions), not even saying anything about whether this particular mode
is dominant or completely marginal at given conditions.
The linearisation approach explicitly states that we introduce some plasma back-
ground, which fulfills the fluid equations and remains unchanged by the wave propa-
gation. It is the perturbations in density, velocity, pressure, electric or magnetic field,
which oscillates and possibly propagates from place to place as a wave. We will see that
the type of the propagating wave usually depends on the direction with respect to the
background magnetic field.
Lets make a simplified example, which we will in details investigate in Section 5.8.3.
Let us consider a stationary and homogeneous plasma cloud in an background magnetic
field. Somewhere inside this cloud we shortly blick with a light-source, a bulb if you
want. In normal transparent neutral gas such impuls would cause the spherical light-
wave to propagate from the point of initiation with a speed of light in all directions. In
plasma it is more dicult, as the light wave are known to be variations of the electric
and magnetic field. In plasma these small electric and magnetic field oscillating with
the wave interact with the charges particles by displacing them slightly, which causes
the modification of the electric and magnetic field in the wave. The motion of particles
and the electric and magnetic fields are bound via the set of (linearised) fluid equations,
which in general depends on the direction to the background magnetic field. Hence
the waveform gets modified depending on the angle of the propagation with respect to
the background magnetic field. We will see that along the magnetic field two circularly
polarised modes propagating with dierent speeds appear, accompanied with incoherent
local oscillations of plasma. In the direction perpendicular to the magnetic field two
linearly polarised modes, again with dierent propagation speeds, appear. In a general
direction of propagation we will find a combination of all five modes. Electromagnetic
waves do not cause the variations in plasma density, hence only the electric and magnetic
91 5.4. PLASMA OSCILLATIONS

fields are aected, and inheritedly also the motions of individual particles. Another
kind of impuls, such as a sudden local increase of plasma density, will excite other
type of waves, but not the complex of electromagnetic ones, only if such impuls is not
accompanied with some emission mechanism.

5.4 Plasma oscillations

The very basic oscillatory movement registered in plasmas takes a form of plasma os-
cillations. Lets derive the dispersion relation first. We search for a solution under the
following assumptions:

Plasma oscillations

No background fields, i.e., E 0 = B 0 = 0, we do not consider induced perturbation

to the magnetic field, i.e., B 1 = 0. Thus from the set of Maxwell equations only
the Poisson equation must be retained

No thermal motion, plasma is cold

Ions are fixed in space and have a uniform distribution

Background density of ions and electrons is the same, i.e., ne0 = ni0 = n0

Background is stationary and homogeneous, ue0 = 0

Thus we solve the following set of linearised equations:

ue1
mn0 = en0 E 1 , (5.22)
t
ne1
+ [n0 ue1 ] = 0, (5.23)
t
E 1 = 1 , (5.24)
0 E 1 = ene1 . (5.25)

Now we express E 1 from (5.24) and insert it into appropriate terms in (5.22) and
(5.25), and apply a divergence operator to (5.22) to obtain:

0 1 = ne1 e, (5.26)
ue1
mn0 = en0 1 , (5.27)
t
ne1
+ n0 ue1 = 0. (5.28)
t
5.4. PLASMA OSCILLATIONS 92

Now we express term 1 from (5.26) and insert it into (5.27). We do a simple
algebraic adjustment to (5.28) by leaving only ue1 on the left-hand side of equation:
ue1 e2 n0
mn0 = ne1 , (5.29)
t 0
1 ne1
ue1 = . (5.30)
n0 t
By combining these two equations we finally obtain
2 ne1 e2 n 0
m = ne1 . (5.31)
t2 0
Now we take the Fourier transform of this equation, practically by applying rules (5.11)
and (5.12):
e2 n 0
2 ne1 = ne1 . (5.32)
0 m
Thus we obtained a dispersion relation for plasma oscillations

n 0 e2
p2 = . (5.33)
0 m

This result requires a little discussion

The dispersion relation shows that there exist oscillations with this characteristic
plasma frequency p , but there is no wave propagating, since there is no wave
vector k in the dispersion relation. The physical interpretation is the following:
electrons may be displaced from their equilibrium positions determined by station-
ary background ions by some random fluctuation. The induced electric field acts
as a restoring force to return them back to their original positions. As they move
back, they have some inertia and overshoot their equilibrium positions. They are
displaced again and the electric field again acts as a restoring force. This random
movement is periodic. The motion of individual electrons is independent, their
mutual phase shifts are random.
The phase speed of this motion, v = /k, is not defined. This is not surprising in
the frame what we just said. Since the mutual phase shifts of individual electrons
are random, a formal phase speed takes any value.
The group speed, vg = d/dk = 0. No signal propagates.
The plasma frequency is essentially the function of the background density only.
Debye length relates to plasma frequency as 2D = 0nK0 Be2Te = Km B Te 1
p2
12 vT
2 1 ,
p2

where we approximated the thermal speed vT 2KmB Te . Obviously, the fluctua-
tions induced by plasma oscillations are shielded on the length scales comparable to
the Debye length, thus plasma oscillations do not induce any macroscopic charge.
93 5.5. ELECTRON PLASMA WAVE

5.5 Electron plasma wave

In the previous section we showed that plasma supports oscillatory motion, however we
did not derive the dispersion relation of the wave. We need to keep additional physical
eect. Thermal motion may cause the oscillations to propagate.

Longitudinal electron wave

No background fields, i.e., E 0 = B 0 = 0, we do not consider induced perturbation

to the magnetic field, i.e., B 1 = 0. Thus from the set of Maxwell equations only
the Poisson equation must be retained

Ions are cold, electron are subjects of thermal motion

Ions are fixed in space and have a uniform distribution

Background density of ions and electrons is the same, i.e., ne0 = ni0 = n0

Background is stationary and homogeneous, ue0 = 0

To include the eect of the thermal motion, we need to keep the pressure term in
the equation of motion. The gradient of pressure is coupled with the thermal motion
through the equation of state
pe = (3KB Te ne ) = 3KB Te (n0 + ne1 ), (5.34)
where number 3 indicates that there is only one degree of freedom, thus we (silently) as-
sume the propagation in one direction only. The modified system of linearised equations
to be solved then reads:
ue1
mn0 = en0 E 1 3KB Te ne1 , (5.35)
t
ne1
+ n0 ue1 = 0, (5.36)
t
E 1 = 1 , (5.37)
0 E 1 = ene1 . (5.38)
The sequence of algebraic steps performed is identical to those in Section 5.4, thus we
state only the final step here
2 ne1 e2 n 0
m = ne1 3KB Te ne1 . (5.39)
t2 0
By taking the Fourier transform we obtain
KB Te
2 ne1 p2 ne1 3k 2 ne1 = 0 (5.40)
m
and finally a dispersion relation for longitudinal electron waves in plasmas
5.5. ELECTRON PLASMA WAVE 94

3
2 = p2 + k 2 vT
2
, (5.41)
2

where vT = 2KmB Te is the thermal speed of electrons.
The dispersion relation is only slightly dierent from the dispersion relation for
electron oscillations, however, it has important consequences:

This relation describes a propagating wave (it contains wave vector k). It propa-
gates only for frequencies larger than the plasma frequency, for frequencies smaller,
the resulting wave vector takes imaginary values, thus the spatial part causes the
exponential in (5.10) to decay. Excited waves with frequencies smaller than plasma
frequency are rapidly attenuated. The wave propagates only in plasmas, the dis-
persion relation degenerates to = 0 in vacuum.
2 2
p +3/2k2 vT
The phase speed is given by v = k . It depends on the wave number,
thus the waves are dispersive (dierent modes travel with a dierent phase speed).
2 2k
3 vT k vT
The group speed vg = 2 = 32 2
is not zero, thus the waves are able to
p2 +3/2k2 vT
v k 2
carry information. Assymptotically for large k, lim vg = 32 lim 2 T 2 2
p +3/2k vT
2
k k
3 vT k
2 2 2
= 32 vT , the waves propagate essentially with the thermal speed.
3/2k vT

We solved the dispersion relation for perturbation in density of electrons ne1 . The
definition of the Fourier transform (5.10) allows to introduce a phase shift, so that the
variable fullfill the system of equations. Lets assume that the full solution for ne1 has
a form of
ne1 = ne1 cos (k r t) (5.42)
and we allow for a phase shift E for E 1 :

E 1 = E 1 cos (k r t + E ). (5.43)

Both variables E 1 and ne1 are coupled via Poisson equation 0 E 1 = ene1 .
Thus
0 k E 1 sin (k r t + E ) = ene1 cos (k r t). (5.44)
By expressing the left-hand side we obtain

e k ne1
sin (k r t) cos E + cos (k r t) sin E = cos (k r t). (5.45)
0 E 1

This equation has a solution when we take for example E = 2 .

The perturbation to electron density and the electric field induced by the wave are
shifted in phase by quarter of period. How about the direction of the vector variables?
95 5.6. ION WAVES

The Fourier image of the Lorentz force is

ue1 = e(E 1 + ue1 B 0 ). (5.46)
Since in our case B 0 = 0, the relation simplifies to
e
ue1 = E 1 , (5.47)

thus ue1 E 1 . Similarly Farradays law written in the Fourier space is
k E 1 = B 1 , (5.48)
but B 1 = 0, hence k E 1 . The waves are longitudinal.
Relation B 1 = 0 was among the assumptions, however it is not clear that the
assumptions are fulfilled. A correct way is to verify the assumptions to make sure that
the derivation we just did is selfconsistent. To prove that B 1 = 0 we take a curl of the
linearised Ohms law
k E 1 + k (ue1 B 0 ) = k j e1 . (5.49)
The current j e1 is caused by fluctuating electrons, hence proportional to electron velocity
ue1 . However, ue1 is parallel to the direction of propagation k, thus the right-hand side
of the equation is identically zero. The second term on the left-hand side also vanishes,
since B 0 = 0. By expressing k E 1 from the Farradays law we obtain
0 = B 1 . (5.50)
which has the only solution when B 1 = 0, Q.E.D.

5.6 Ion waves

When it comes to ion waves, the situation becomes more complicated, since we cant
consider ions an immovable background anymore. In section 5.2 we derived the dis-
persion relation for the acoustic waves in the non-magnetised neutral fluid. Will the
solution change, when the fluid is composed from charged particles? We search for the
wave, assuming:

Ion acoustic wave

No background fields, i.e., E 0 = B 0 = 0, we do not consider induced perturbation
to the magnetic field, i.e., B 1 = 0. Thus from the set of Maxwell equations only
the Poisson equation must be retained

Both ions and electrons are subjects of thermal motion, their temperatures may
be dierent

Background density of ions and electrons is the same, i.e., ne0 = ni0 = n0

Background is stationary and homogeneous, ue0 = ui0 = 0

5.6. ION WAVES 96

Then we solve a system of linearised equations

ue1
mn0 = en0 E 1 3KB Te ne1 , (5.51)
t
ne1
+ n0 ue1 = 0, (5.52)
t
ui1
M n0 = en0 E 1 3KB Ti ni1 , (5.53)
t
ni1
+ n0 ui1 = 0, (5.54)
t
E1 = 1 , (5.55)
0 E 1 = e(ni1 ne1 ). (5.56)
Ions are much havier than electrons, thus we may assume that electrons will simply
follow the ions and organize themselves so that they shield possible charge concentra-
tions. Therefore, the perturbation to the electron density ne1 may be estimated by the
Boltzmann relation (4.87), thus
( )
e1 e1
ne1 = ne ne0 = n0 exp 1 n0 (5.57)
KB Te KB Te
and we may consider (5.51) and (5.52) solved.
Now we take E 1 from (5.55), insert it into (5.53) and use (5.57) in (5.56). Addition-
ally, we apply a divergence operator to (5.53). We have
ui1
M n0 = en0 1 3KB Ti ni1 , (5.58)
t
ni1
+ n0 ui1 = 0, (5.59)
t
e1
0 1 = eni1 en0 . (5.60)
KB Te
Now we express ui1 from (5.59), insert into (5.58) and take a Fourier transform of
the two remaining equations. We have
2 M ni1 = en0 k 2 1 + 3KB Ti k 2 ni1 , (5.61)
1
0 k 2 1 = eni1 e2 n0 . (5.62)
KB Te
Finally, we take 1 from (5.62), insert into (5.61) and solve for ni1 to get
e2 n 0 k 2 3KB Ti k 2
2 = ( )+ . (5.63)
M 0 k 2 + e2 n0 M
KB Te

e2 n0
Let p = be an ion plasma frequency. If we remind Debye length D = 0nK0 Be2Te
M 0

2KB Ti
and ion thermal speed vTi = M , we finally obtain a disperion relation of ion
acoustic waves
97 5.7. ELECTROSTATIC WAVES IN BACKGROUND MAGNETIC FIELD

2p 2D k 2 3
2 = + v2 k2 . (5.64)
2D k 2 + 1 2 Ti

For further discussion, lets take one more approximation, the plasma approxima-
tion. In practice we focus to large-scale eects, thus we set kD 1. That is essentially
equivalent of setting ni = ne .4 Under this approximation, the dispersion relation sim-
plifies to: ( )
KB Te KB Ti
2 = k2 +3 . (5.65)
M M

Phase speed v = KB Te +3K M
B Ti
cs is a speed of sound in plasma. Under the
plasma approximation these waves are not dispersive. Dispersivity shows up as
the eect of Debye shielding, which is mode-dependent.

Group speed equals to the phase speed, thus vg = v = cs . Ion waves are waves
with constant speed of propagation.

In laboratory plasmas, Ti Te , thus cs KB Te
M . The propagation speed is
essentially set by the mass of ions and temperature of electrons.

The waves are longitudinal. We keep the proof to the reader.

The ion acoustic wave exists only if the thermal motion of at least electrons is pos-
sible. Electrons are pulled by ions to ballance the charge concentrations, hence
vg = v . However, the shielding is not perfect, there exist microscopic accumula-
tions of ions, which propagate as the wave by through a E 1 field.

By assuming that Ti Te , the assymptotical dispersion relations read: = 0 for

k 0 and = p for k . The assymptotical solution for small scales (large
k) is fundamentally dierent from longitudinal electron waves.

5.7 Electrostatic waves in background magnetic field

5.7.1 Electron

Electron electrostatic perpendicual waves

Background magnetic field, no background electric field, i.e., B 0 = 0, E 0 = 0

Ions stationary, thermal motion of electrons is being neglected

4
Note that by assuming the plasma approximation we consider the bulk densities of ions and electrons
to be equal. This does not imply either ni1 = ne1 or ni0 = ne0 !
5.7. ELECTROSTATIC WAVES IN BACKGROUND MAGNETIC FIELD 98

Background density of ions and electrons is the same, i.e., ne0 = ni0 = n0

Background is stationary and homogeneous, ue0 = 0

k E1

Lets write the Fourier images of fluid equations for electrons:

imue1 = e(E 1 + ue1 B 0 ), (5.66)

ine1 + in0 k ue1 = 0, (5.67)
e
ik E 1 = ne1 , (5.68)
0

We express E 1 from (5.66) and combine together (5.67) and (5.68) through ne1 .

im
E1 = ue1 ue1 B 0 , (5.69)
e
ien0
k E1 = k ue1 . (5.70)
0

Since we are considering longitudinal waves, k E 1 , hence k E 1 = kE1 . Thus equation

(5.70) allows to express the amplitude E1 = |E 1 | of electric field:

ien0
E1 = k ue1 . (5.71)
0 k

Let us consider the Cartesian coordinate system, so that ez B 0 and E 1 lies in

the x z plane. Hence B 0 = (0, 0, B0 ) and E 1 = (E1 sin , 0, E1 cos ), and similarly
k = (k sin , 0, k cos ), where is the angle between B 0 and k. We may then write
(5.69) in components

E1 sin ue1,x sin2 ue1,x ue1,y
0 = ien0 0 = im ue1,y B0 ue1,x . (5.72)
0 e
E1 cos ue1,z cos2 ue1,z 0
2
Then using (5.71), multiplying the whole equation by 1i me
and using p2 = e0nm0 and
c = eB
m , and reorganising everything to the left-hand side of the equation we have

ue1,x p2 ue1,x sin2 ue1,y
ue1,y 0 + i c ue1,x = 0. (5.73)
2
ue1,z ue1,z cos2 0

Such equation may be written as

Mu ue1 = 0, (5.74)
99 5.7. ELECTROSTATIC WAVES IN BACKGROUND MAGNETIC FIELD

with matrix Mu being

p2

1 sin2 i c 0
2
Mu = i
c
1 0 (5.75)
p2
0 0 1 2
cos2

Equation (5.74) has only the nontrivial solution if det Mu = 0. To do so, we compute
det Mu using the Laplace expansion using the element in the third row and third column.
We have ( )( )
p2 p2 c2
det Mu = 1 2 cos 2
1 2 sin 2 = 0.
2
(5.76)

which is the dispersion relation for electrostatic waves in presence of the background
magnetic field. Note that the dispersion relation no wave propagates, only oscillations
appear in plasma.
Lets discuss two extreme cases. For = 0 hence for oscillatory motions along the
magnetic field, we have ( )( )
p2 c2
1 2 1 2 = 0, (5.77)

which has two physical solutions

2 = p2 , and 2 = c2 . (5.78)

Therefore in the direction of the magnetic field, we have ordinary plasma oscillations
and the Larmor rotation.
For = /2, oscillatory motions perpendicular to the background magnetic field,
we have
p2 2
1 2 c2 = 0, (5.79)

which may be written as

2 = p2 + c2 h2 . (5.80)

The symbol h indicates the upper hybrid frequency, the only possible frequency for
electron oscillations perpendicular to B 0 .

5.7.2 Ion
5.7. ELECTROSTATIC WAVES IN BACKGROUND MAGNETIC FIELD 100

Ion electrostatic perpendicular waves

Background magnetic field, no background electric field, i.e., B 0 = 0, E 0 = 0

Electrons are subject of thermal motion if they are allowed to, ions are cold

Background density of ions and electrons is the same, i.e., ne0 = ni0 = n0

Background is stationary and homogeneous, ue0 = ui0 = 0

k E1

In case of the ion wawes perpendicular to the background magnetic field, we need to
strictly distinguish the two cases: when the waves are nearly perpendicular, and when
they are strictly perpendicular. The dierence is fundamental. In case of the nearly
perpendicular configuration, there is some, however small, component of the background
magnetic field parallel to the wave vector. It is imporant since in this approximation
the electrons follow ions and in such case, they are still able to shield the concentrations
of charges via the mechanism of Debye shielding. They are free to move along the
magnetic field lines and the shielding is eective. In the case of a strictly perpendicular
magnetic field, there is no option for electrons to shield anymore, since they cant move
around. Lets investigate the two cases separately.

k nearly perpendicular to B 0
In this case we consider the eect of the Debye shielding and thus we can use the
Boltzmann relation as a solution to the problem of electrons. We further solve the
Euler equation for ions having the form of

iM ui0 = e1 + eui1 B 0 (5.81)

in the Fourier space. In components we have (again, assuming the Cartesian coordinate
system, where k = kex and B 0 = B0 ez )

iM ui1,x = eik1 + eui1,y B0 , (5.82)

iM ui1,y = eui1,x B0 , (5.83)

which gives the x-component of the perturbed velocity

( )1
ek1 2c
ui1,x = 1 2 . (5.84)
M

The continuity equation gives as usually

k
ni1 = n0 ui1,x , (5.85)

101 5.7. ELECTROSTATIC WAVES IN BACKGROUND MAGNETIC FIELD

and the electron density perturbation is solved by the Boltzmann relation

e1
ne1 = n0 . (5.86)
KB Te

By assuming the plasma approximation ni = ne we have ni1 = ne1 and hence

e1
ui1,x = , (5.87)
KB Te k

which may be compared to (5.84) to obtain the dispersion relation

KB Te 2
2 2c = k . (5.88)
M
Let us remind that the definition of the speed of sound in the plasma reads
( )
2 KB Te 3KB Ti
cs = + (5.89)
M M

which has a simpler form due to the assumption of Ti 0 (or equivalently, Ti Te ).

Hence the dispersion relation (5.88) may be rewritten in a form

2 = 2c + k 2 c2s . (5.90)

This case is possible only if the limiting angle = 2 , where is the angle
between k and B 0 , is larger than the ratio of ion and electron speed in the direction
of B 0 , hence roughly for > (m/M )1/2 . In this case, the electrons have kinetic energy
(from the thermal motion) large enough to chase the ions and also move around them
to shield the originating concentrations of charges. For angles smaller when electrons
do not have thermal speed large enough to shield the ion, we need to use a dierent
formalism.

k B0
For k exactly perpendicular to B 0 there is no component of k parallel to B 0 , thus
electrons are not allowed to move along the B 0 and thus they cannot keep the neutrality.
Hence we must solve fluid equations for both electrons and ions. The system of equations
is closed by using the plasma approximation. Hence we have simply use (5.84) for
describing the motions of ions in the direction of wave vector, and equivalently construct
the corresponding equation for electrons just by replacing e e and M m. Thus
( )1
ek1 c2
ue1,x = 1 2 . (5.91)
m
5.8. ELECTROMAGNETIC WAVES 102

Similarly for the continuity equation,

k
ni1 = n0 ui1,x , (5.92)

k
ne1 = n0 ue1,x . (5.93)

By applying the plasma approximation approach we have ne1 = ni1 and hence
ue1,x = ui1,x . Thus ( ) ( )
2c c2
M 1 2 = m 1 2 , (5.94)

and further
( ) ( )
2 1 1 m+M
(M + m) = mc2 + M 2c =e 2
B02 + =e 2
B02 (5.95)
m M mM

and thus

eB0 eB0
2 = = c c d2 . (5.96)
m M

In this case, the oscillations are allowed to have the only one possible frequency, the
lower hybrid frequency.

5.8 Electromagnetic waves

When discussing the electromagnetic waves, we automatically assume that the resulting
solution will fulfill the assumption k E 1 . This assumption should always be verified
at the end. In this text, we will keep the verification to the reader. Since they are
electromagnetic, we cannot neglect the magnetic component of the wave, thus B 1 = 0.

5.8.1 In vacuum: Light waves

Electromagnetic waves in vacuum degenerate to light waves. Thus we need only the set
of Maxwell equations.

Electromagnetic waves in vacuum

No background magnetic or electric field, i.e., B 0 = E 0 = 0

No plasma, thus ne = ni = n0 = 0, ue = ui = 0

k E1
103 5.8. ELECTROMAGNETIC WAVES

B 1
E1 = , (5.97)
t
1 E 1
B 1 = 0 , (5.98)
0 t
B 1 = 0. (5.99)

Now we take a time derivative of (5.97) and apply the curl operator to (5.98). We have
E 1 2B1
= , (5.100)
t t2
1 E 1
B1 = , (5.101)
0 0 t
B 1 = 0. (5.102)

Both (5.100) and (5.101) contain the term E

t . Using this term we combine these
1

two equations and take a Fourier transform of the system of remaining two equations.
We have

c2 k (k B 1 ) = 2 B 1 , (5.103)
k B 1 = 0, (5.104)

where we used the definition of the speed of light c = (0 0 )1 . The double vector
multiplication may be written using two terms. We have
[ ]
c2 (k B 1 )k k 2 B 1 = 2 B 1 , (5.105)
k B 1 = 0. (5.106)

Now we finally use (5.106) in (5.105) and have

c2 k 2 B 1 = 2 B 1 , (5.107)

which simplifies to

2 = c2 k 2 . (5.108)

This is the dispersion relation for electromagnetic waves in vacuum, describing the
propagation of light. Obviously, the light waves are non-dispersive and both phase and
group speeds equal to the speed of light in vacuum.

5.8.2 In plasma
In plasma, the dispersion relation for electromagnetic waves gets modified due to the
charged particles.
5.8. ELECTROMAGNETIC WAVES 104

Electromagnetic waves in plasma

No background magnetic or electric field, i.e., B 0 = E 0 = 0

Background stationary and homogeneous

We except large frequencies ( p ), thus heavy ions remain stationary, only

electrons respond

Both ions and electrons are cold (no thermal motion)

k E1

We proceed similarly to the case of electromagnetic waves in vacuum, we only need

to retain the electric current j 1 in the Amperes law.
B 1
E1 = , (5.109)
t
1 E 1
B 1 = 0 + j 1. (5.110)
0 t
Similarly to the previous section, we take a time derivative of (5.110) and apply a curl
operator to (5.109) and combine the two using the term B t , which is contained
1

in both equations. We have

1 2E1 1 j 1
E1 = 2 2
. (5.111)
c t 0 c2 t
Again, we take a Fourier transform of the above given equation and split the double
vector multiplication into two terms:
i 2
(k E 1 )k + k 2 E 1 = j 1 + E1. (5.112)
0 c2 c2
From the beginning we search for the dispersion relation of electromagnetic waves, thus
we ourselves to the case when k E 1 , hence k E 1 = 0. Thus (5.112) simplifies to
i
( 2 c2 k 2 )E 1 = j . (5.113)
0 1
In the solution we except frequencies much larger than ion plasma frequency, thus the
perturbed component of the electric current is composed from electrons only, ions are
considered as fixed in space in this approximation. Thus
j 1 = n0 eue1 . (5.114)
The fluctuating component of electron velocity ue1 may be obtained from Euler equation
for electrons, which for cold plasma with no background magnetic field simplifies to
ue1
m = eE 1 . (5.115)
t
105 5.8. ELECTROMAGNETIC WAVES

Hence, the current density j 1 is

n 0 e2
j1 = E1, (5.116)
im
where we expressed the Euler equation for electron in the Fourier space. Now we
continue to modify (5.113):

i n0 e2 n 0 e2
( 2 c2 k 2 )E 1 = E1 = E 1 = p2 E 1 . (5.117)
0 im 0 m
Hence the dispersion relation for electromagnetic waves in plasma reads

2 = p2 + c2 k 2 . (5.118)

Lets make a few comments

The dispersion relation describes propagating non-dispersive waves.

( )2 2
Phase velocity: v2 = k = c2 + k2p > c2 . Phase velocity of these waves is larger
than the speed of light. However, it is the phase velocity, thus it does not carry
any signal, and the principles of the special theory of relativity are not violated.

Group velocity: vg = d/dk = 2c2

2k c2
= v < c for v > c. The group speed
2 p +c2 k2
is smaller than the speed of light.

The waves propagate only for large frequencies, when > p . The electron
plasma frequency p serves as a cut-o frequency. For frequencies smaller than
p the waves do not propagate
(the wave vector k is imaginary). Then we have
ck = ( 2 p2 )1/2 = i p2 2 . When taking the spatial part of the Fourier
[ (5.10),] we find the physical variables to vary as exp ik r exp ikx =
transform
2 2 [ ]
exp x pc = exp x , where = 2c 2 is the attenuation length.
p

The electromagnetic waves propagating through plasma may serve as a primitive mean
to estimate the plasma density. From measuring the spectrum we find the cut-o fre-
quency, which depends on the plasma density n0 only. This is being used in astrophysics
for example in estimating the plasma density of the solar atmosphere. Imagine that we
have a ionised plasma cloud hurled during the solar flare by the magnetic reconnec-
tion, which rises in the atmosphere, and acts as a source for dierence kinds of plasma
waves, including the electromagnetic ones, usually in the band of radio waves. As the
cloud rises, the density of ambient plasma decreases, therefore also the cut-o frequency
decreases and thus waves at lower frequencies are allowed to escape the region and prop-
agate toward the observer. By measuring the cut-o frequency as a function of time
5.8. ELECTROMAGNETIC WAVES 106

and estimating the rising speed of the plasma blob, we perform the density scanning of
the solar atmosphere.
(Add a story of measuring the density of the interstellar matter from Voyager 1,
must confirm what kind of waves were involved and perform the calculation)

5.8.3 Complex of electromagnetic waves in presence of B 0

Electromagnetic waves in plasmas interact mostly with electrons, the heavy ions cant
adjust to the large frequencies of the electromagnetic waves. Hence we will solve the
system of equations

imue1 = e(E 1 + ue1 B 0 ), (5.119)

ik E 1 = iB 1 , (5.120)
i
k B 1 = 0 iE 1 en0 ue1 . (5.121)
0
The equations are already written in the form of Fourier images. Note that we did not
consider the continuity equation, which over-determines the system. That is because
the density perturbation ne1 does appear in the continuity equation only. Now we
express ue1 from (5.119) and combine (5.120) and (5.121) via B 1 , where in the combined
equation we keep solely E 1 on the left-hand side:
e eB0
ue1 = i E1 i ue1 b0 , (5.122)
m m
c2 en0
E 1 = 2 k (k E 1 ) + i ue1 , (5.123)
0
where we used the definition of the speed of light c2 = (0 0 )1 and defined the unity
vector b0 in the direction of the magnetic field, so that B 0 = B0 b0 . Now we express
ue1 from (5.123) as
0 c2 0
ue1 = i E1 i 2 k (k E 1 ) (5.124)
en0 en0
and insert it into (5.122). One needs to be carefull here, since we find ue1 in (5.122) on
both sides of the equation. We further use c = eB0 /m to obtain
0 c2 0
i E1 i k (k E 1 ) =
en0 en0
[ ]
e c 0 c2 0
i E1 i i E1 i k (k E 1 ) b0 . (5.125)
m en0 en0
Now we express the double cross products and multiply the whole equation by eni00 .
We have

2 E 1 + c2 (k E 1 )k c2 k 2 E 1 =
e2 n 0 c c
= E 1 ic E 1 b0 i c2 (k E 1 )(k b0 ) + i c2 k 2 (E 1 b0 ). (5.126)
0 m
107 5.8. ELECTROMAGNETIC WAVES

e2 n0
By reordering and using p2 = 0 m we have
c 2 c
( 2 p2 c2 k 2 )E 1 + i ( c2 k 2 )(E 1 b0 ) + c2 (k E 1 )k + i c2 (k E 1 )(k b0 ) = 0.

(5.127)
To solve this equation, we introduce the Cartesian coordinate system, so that B 0 ez
and k lies in the xz plane. Therefore b0 = (0, 0, 1) and k = (k sin , 0, k cos ), where
is the angle between B 0 and k. The electric intensity vector has all components
E 1 = (E1x , E1y , E1z ). In this particular representation (5.127) turns into

E1x E1y
c
( 2 p2 c2 k 2 ) E1y + i ( 2 c2 k 2 ) E1x +

E1z 0

k sin
+ c2 (kE1x sin + kE1z cos ) 0 +
k cos

0
c 2
+ i c (E1x k sin + kE1z cos ) k sin = 0. (5.128)

0

This is a set of three equations for three components of perturbation of electric

intensity, which may we written in the matrix form as

ME 1 E 1 = 0 (5.129)

with system matrix

2
p2 c2 k 2 cos2 i c ( 2 c2 k 2 ) c2 k 2 sin cos
ME 1 = i c ( 2 c2 k 2 cos2 ) 2 p2 c2 k 2 i c c2 k 2 cos sin . (5.130)
c2 k 2 sin cos 0 2 p2 c2 k 2 sin2

Equation (5.129) has the nontrivial solution when det ME 1 = 0. We expand det ME 1
around the third row to have

[ ]
c2 2
det ME 1 = c k sin cos 2 ( c k ) ( p c k ) +
4 4 2 2 2 2 2 2 2 2

[
+ ( p c k sin ) ( 2 p2 c2 k 2 cos2 )( 2 p2 c2 k 2 )
2 2 2 2 2

]
c2 2
2 ( c k )( c k cos ) = 0.
2 2 2 2 2 2
(5.131)

This is the dispersion relation for a complex of electromagnetic waves in plasmas. We

will not study two special cases.
5.8. ELECTROMAGNETIC WAVES 108

Electromagnetic waves parallel to B 0

Waves parallel to B 0 have k B 0 and thus = 0. The dispersion relation (5.131)
simplifies to [ ]
c2 2
( p ) ( p c k ) 2 ( c k ) = 0.
2 2 2 2 2 2 2 2 2 2
(5.132)

This equation has four roots, one of which is not physical. By setting the first parenthesis
equal to zero one obtains

2 = p2 , (5.133)

therefore along the magnetic field, the first mode of oscillatory motions is composed of
ordinary plasma oscillations.
By setting the second parenthesis equal to zero we obtain other two roots:
c 2
2 p2 c2 k 2 = ( c2 k 2 ) (5.134)

which may be written as

p2
2 c2 k 2 = (5.135)
1 c
and
p2
2 c2 k 2 = . (5.136)
1 + c

The wave described by (5.135) carries the name R-wave, while the other one de-
scribed by (5.136) is named the L-wave. Their names have the origin in their circular
polarisation, which we will study in a moment.

Both waves are dispersive

We keep the derivation of phase and group speed to the reader, however already
from the dispersion relation we see that at the same frequency, the propagation
speed of the R-wave is larger.

Lets investigate the polarisation of these two waves. We start from (5.129) with
= 0. We have
2
p2 c2 k 2 i c ( 2 c2 k 2 ) 0 E1x
i c ( 2 c2 k 2 ) 2 p2 c2 k 2 0 E1y = 0, (5.137)

0 0 p
2 2 E1z
109 5.8. ELECTROMAGNETIC WAVES

from which we immediately see that E1z = 0. The relation between the two other
components is
c 2
( 2 p2 c2 k 2 )E1x + i ( c2 k 2 )E1y = 0. (5.138)

( )
c 1
We further use the dispersion relation in a form c2 k 2 = 2 p2 1 to
finally obtain
E1y = iE1x . (5.139)
Thus we have Ex = iEy for the R-wave (the upper sign is valid) and Ex = iEy for
the L-wave. We immediately see that the waves are circularly polarised (the ampli-
tudes of both components are equal and both components are shifted in phase by
quarter of circle). By explicitly expressing the spatio-temporal dependence of the
electric intenzity, E 1 = E 1 exp [i(k r t)] = E 1 [cos (k r t) + i sin (k r t)].
If we introduce a phase shift E between the components and define Ex = (E 1 ) =
E 1 cos (k r t) and Ey = (E 1 ) = E 1 sin (k r t + E ) we see that for the
R-wave, E = /2 and thus Ey preceeds Ex , the waves polarisation is right-
handed. Similarly, in the case of the L-wave, Ey is retarded by E = /2 and
the waves polarisation is left-handed. Note that the polarisation does not depend
on the propagation direction, only on the direction of the background magnetic
field.
[ ]
The R-wave has a cut-o at R = 12 c + (c2 + 4p2 )1/2 (hence the designation
of[ this frequency) and ]a resonance at c . The L wave has a cut-o at L =
2 c + (c + 4p )
1 2 2 1/2 and no resonance.

Application: Farraday rotation. The detection of R- and L-waves in the radio

spectrum of astrophysical objects allows to estimate either the distance of these objects
or the plasma frequency of the interstellar environment. Assume that we have a plane
wave composed of both R- and L-waves, which propagates along interstellar B 0 . As we
pointed out, the R-wave propagates faster.
Let us write the vector of electric fields of both considered R- and L-wave in com-
ponent with respect to some Cartesian coordinate system

E R = E0 exp[i(kR r t)] [ex + iey ] , (5.140)

E L = E0 exp[i(kL r t)] [ex iey ] . (5.141)

The components are written in agreement with (5.139) and ex and ey are unity vectors
in x and y directions.
The two waves superpose to form one detected wave:
1 1 [ ]
E = (E R + E L) = E0 eit eikR r (ex + iey ) + eikL r (ex iey ) . (5.142)
2 2
Should kR = kL = k then
1 [ ]
E = E0 eit 2eikr ex = E0 exp[i(k r t)]ex , (5.143)
2
5.8. ELECTROMAGNETIC WAVES 110

which is the linearly polarised wave in the x-direction.

In our case, however, kR = kL ,

p2 / 2
kR,L = 1 . (5.144)
c 1 c /

In the high-frequency regime, for c , p by using Taylor series x 1 x2 and
(1 + x)1 1 x we obtain
( ) [ ]
1 p2 1 1 p2 ( c )
kR,L 1 1 1 (5.145)
c 2 2 1 c / c 2 2

Hence when we use a decomposition of both wave numbers as

kR,L = k k, (5.146)

we have [ ]
1 p2
k= 1 (5.147)
c 2 2
and
1 p2 c
k = . (5.148)
2 c 2
By inserting (5.146) into (5.142) we have
1 [ ]
E = E0 ei(krt) eikr (ex + iey ) + eikr (ex iey )
2
= E0 ei(krt) [cos(k r)ex + sin(k r)ey ] . (5.149)

Obviously, the dierence in the phase speeds of the R- and L-wave causes the rotation
of the polarisation plane of the resulting linearly polarised wave. When adjusting the
coordinate system so that k r = kz the angle of the polarisation (it equals to
the argument of the goniometric functions) plane fullfills

d
= k. (5.150)
dz
To obtain the total rotation of the polarisation plane for the waves travelling the
distance d we must integrate (5.150):

d d d
d 1
= dz = kdz = 0 + p2 (z)c (z)dz =
dz 2c 2
0 0 0
d
e3
= 0 + n0 (z)B0 (z)dz. (5.151)
2m2 0 c 2
0
111 5.8. ELECTROMAGNETIC WAVES

When we approximate both the density and magnetic field by constants, we simply
obtain
e3 1
= 0 + n0 B0 d. (5.152)
2m 0 c 2
2

By measuring the angle of the polarisation plane for various frequencies one may fit this
relation and obtain the information about either the properties of the interstellar space,
or about the distance of the radiating object.

Electromagnetic waves perpendicular to B 0

Let us go back to the general dispersion relation of the complex of electromagnetic waves
in plasmas (5.131) and discuss the other special case, when = /2. The dispersion
relation simplifies to
[ ]
( 2 p2 c2 k 2 ) ( 2 p2 )( 2 p2 c2 k 2 ) c2 ( 2 c2 k 2 ) = 0. (5.153)

This equation again has two solutions. By zeroing the first parenthesis we obtain

2 = p2 + c2 k 2 , (5.154)

which is the dispersion relation identical to the electromagnetic waves in non-magnetised

plasmas. Due to this correspondence, this waves is called ordinary or the O-mode. It
linearly polarized with E 1 B 0 . It has a cut-o at the plasma frequency.
By zeroing the second parenthesis we have

( 2 p2 )( 2 p2 c2 k 2 ) = c2 ( 2 c2 k 2 ), (5.155)

which by expressing 2 c2 k 2 from both sides and rearranging becomes

2 p2
c k =
2 2 2
p2 , (5.156)
2 h2

where we used the definition of the upper hybrid frequency h2 = p2 + c2 . Such wave
is termed the extraordinary wave or the X-mode. It is partly transversal and partly
longitudinal, it propagates perpendicularly to B 0 . As the density is increased, the
phase velocity rises from c until the cut-o at R is reached. As the density is further
increased, the wave is evanescent until the resonance at the upper hybrid frequency h .
Then it can propagate
[ again until the] second cut-o[ at L . The cut-o frequencies
] are
1 2
given by: R = 2 c + (c + 4p ) 2 1/2 and L = 2 c + (c + 4p )
1 2 2 1/2 .
For the O- and X-modes the plasma acts as birefringent environment.
5.9. MHD WAVES 112

5.9 MHD waves

By MHD waves (A.K.A. hydromagnetic waves) we understand low frequency ion oscil-
lations in the presence of the magnetic field. A special case is the Alfven wave.

5.10 Alfv
en wave

Alfv
en waves

Background magnetic field, no background electric field, i.e., B 0 = 0, E 0 = 0

Background stationary and homogeneous

We solve equations of motion for both ions and electrons

Both ions and electrons are cold (no thermal motion)

We introduce the Cartesian coordinate system (ex , ey , ez ), so that:

B 0 = B 0 ez ,

E 1 j 1 ex , E 1 B 0 ,

B 1 u1 ey ,

k B0, B1 E1.

Again, we use the solution of the system of Maxwell equations (5.109) and (5.110)
from the previous section by directly using (5.113):
i
( 2 c2 k 2 )E 1 = j . (5.157)
0 1
We do not assume that the ion create a stationary background, thus the current density
j 1 is proportional to the dierence between fluctuation speed of ions and electrons.
Hence
j 1 = j1 ex = n0 e(ui1x ue1x )ex . (5.158)
The background is assumed static, thus we further drop the index 1 to denote that the
fluctuating part of the particle velocity is discussed. Hereafter, uix = ui1x and uex = ue1x
and similarly for the y-components. Expression for both velocity fluctuations must be
obtained from the Euler equations. For ions we have
ui1
M = e(E 1 + ui1 B 0 ), (5.159)
t
113 WAVE
5.10. ALFVEN

which in components and in the Fourier space gives

iM uix = eE1 + euiy B0 , (5.160)

iM uiy = euix B0 . (5.161)

By combining these two equations we have

eB0
iM uix = eE1 + eB0 uix (5.162)
iM
and after some algebra we obtain
( )1
ie 2c
uix = 1 2 E1 . (5.163)
M

Similarly [e.g. by inserting (5.163) into (5.161)], we obtain for uiy :

( )1
e c 2c
uiy = 1 2 E1 . (5.164)
M

We solve the Eulers equation for electrons by analogy (formally by application of the
following substitutions: M m, e e, and c c ), to obtain:
( )1
ie c2
uex = 1 2 E1 , (5.165)
m
( )1
e c c2
uey = 1 2 E1 . (5.166)
m

We expect low-frequency waves. Thus we are allowed to use the approximation c .

Under this approximation (5.165) and (5.166) becomes:
( )1
ie 2 c2 ie 2
uex = E1 E1 0 (5.167)
m 2 m c2

and
( )1
e c 2 c2 e c 2 e E1
uey = E1 E1 = E1 = . (5.168)
m 2 m c 2 c m B0

Now we use the obtained estimates for the fluctuating speeds (5.163) and (5.167)
and use them in (5.158), which we use in (5.157).
[ ( )1 ]
i ie 2c
( c k )E1 = n0 e
2 2 2
1 2 E1 0 =
0 M
( )1 ( )1
n 0 e2 2c 2c
= 1 2 E1 = p 1 2
2
E1 . (5.169)
0 M
 WAVE
5.10. ALFVEN 114

Hence the dispersion relation reads

( )1
2c
c k =
2 2 2
2p 1 2 . (5.170)

If we limit ourselve to the case, when c , which is a very strong assumptions, we
basically say that we are interested only in the very low-frequency waves, the dispersion
relation simplifies to
( 2 )1
2c 2
2 p n0 M
c k = p
2 2 2 2
= 2 2 = 2 . (5.171)
2 c2 0 B0 0 B02
Hence
2 c2
= . (5.172)
k2 1 + B 2
0 0

In plasma the expression 1+ B 2 has a meaning of the square-root of relative permitivity

0 0

r , which in plasmas is usually much larger than unity. Hence 0 B02
1 and we have

2 c2 B02
= 0 c2
= . (5.173)
k2 0
B02

Thus the dispersion relation for Alven waves reads

B0
= cA . (5.174)
k 0

cA is the Alfven speed.

Notes:
Figure

Magnetic component of the wave, By , is in the y direction and looks like sinusoid
stringing the field lines of the background magnetic field

We have a drift vy = E1 ex B 0 /B02 , which is the same for both ions and electrons

The phase speed of the perturbations to the field lines equals to the speed of
both ions and electrons. To prove it, lets take the Fourier image of the induction
equation without the dissipative term B1 = kvy,B B0 and the Farradays law
kE1 = B1 . Combining the two we obtain vy,B = k k B E1
0
= BE1
0
, which is the
expression for the speed of particles. Therefore, both the fluid and the perturbed
magnetic field oscilate together. The plasma is frozen in the magnetic field.

The results we obtained are valid only if there is no background electric field
parallel to the magnetic.
115 5.11. GENERAL MHD WAVES

Alfven speed is a characterictic speed of propagation of perturbations to the mag-

netic field. This may yield important estimates. For example, sunspots are be-
lieve to be caused by the flux tube, which was perturbed at the base of the
solar convection zone and rises up to the photosphere. The characterictic time of
rise may be roughly estimated when using the Alfven speed as the estimate for
the rising speed. For the Sun, assuming the depth of the convection zone to be
200 Mm, the average bulk density in the convection zone 200 kg m3 , and the
initial field strength in the perturbed flux tube to be 10 kG=1 T. Then we have
6
cdA 20010 1 s 36 days. This estimate is not far from the more ap-
4107 200
propriate modelling, which includes a detailed ballance of the rising flux tube and
the ambient solar plasma. The characteristic rising time is larger than the solar
rotation period, thus one should expect the deflection of the rising flux tube due
to the Corriolis force.

5.11 General MHD waves

As the final case, let us study the general case of magnetohydrodynamic waves. Hence,
we will solve the linearised version of the following system of equations:

+ u = 0, (5.175)
t
du
= p + ( B) B/0 , (5.176)
( dt)
d p
= 0, (5.177)
dt
B
= (u B), (5.178)
t
B = 0. (5.179)

Note that:

We assume a homogeneous and stationary background and no background motion

(u0 = 0)

We collapse the set of Maxwell equations into the induction equation with no
dissipative terms, to the Gausss law for magnetism and explicitly use Amperes
law in the expression for the Lorentz force

We assume the adiabatic approximation to the equation of state.

5.11. GENERAL MHD WAVES 116

Except for the equation of state, the linearisation is trivial and can be done from the
top of anyones head:

1
+ 0 ( u1 ) = 0, (5.180)
t
u1
0 = p1 + ( B 1 ) B 0 /0 , (5.181)
t
B 1
= (u1 B 0 ), (5.182)
t
B 1 = 0. (5.183)

The equation of state needs to be done more carefully. Let us do a slow derivation, in
each step we immediately neglect the second and higher order terms:
{ }[ ]
p0 + p1
0= + u1 =
t (0 + 1 )
[ ] { [ ]}
1 p1 1
= + (u 1 )p 0 ) + (p 0 + p 1 ) + (u 1 ) 0 =
(0 + 1 ) t (0 + 1 )+1 t
[ ]
1 p1 p0 + p1 1 p0 + p1
= + (u1 )p0 (u1 )0 =
(0 + 1 ) t 0 + 1 t 0 + 1
{ [ ]}
1 p1 2 1
= + (u1 )p0 cs + (u1 )0 , (5.184)
(0 + 1 ) t t

where we used the definition of the speed of sound c2s = p0 /0 and we evaluated term
p0 +p1 1
0 +1 already in Section 5.2. Since the term (0 +1 ) is some non-zero number, the
expression in curly parentheses must be zero, which is our linearised equation of state
(by further taking into account that the background is homogeneous):

p1 1
c2s = 0. (5.185)
t t
Thus we combine (5.180), (5.181), (5.182), (5.183), and (5.185) to obtain the dispersion
relation. We start from taking a time derivative of (5.181):

2 u1 1 p1 B 1 B0
2
= + ( ) . (5.186)
t 0 t t 0 0

Now we use (5.185) and have

( )
2 u1 1 2 1 B 1 B0
= cs + ( ) (5.187)
t2 0 t t 0 0

and further by using (5.180) and (5.182) we have

2 u1 B0
2
= c2s ( u1 ) + { [ (u1 B 0 )]} . (5.188)
t 0 0
117 5.11. GENERAL MHD WAVES

The equation is being solved in the Fourier space again:

B0
2 u1 = c2s k(k u1 ) + {k [k (u1 B 0 )]}
0 0
= cs k(k u1 ) + {k [k (u1 b0 )]} b0 c2A ,
2
(5.189)

where we used b0 = B 0 /|B 0 | and c2A = B02 /(0 0 ) is the Alfven speed. The complicated
multiplication of the right-hand side is

{k [k (u0 b0 )]} b0 = {k [u1 (k b0 ) b0 (k u1 )]} b0 =

= (k b0 ) [(k u1 ) b0 ] (k u1 ) [(k b0 ) b0 ] =
[ ]
= (k b0 ) [k(u1 b0 ) u1 (k b0 )] + (k u1 ) kb20 b0 (k b0 ) =
= (k b0 )2 u1 (k u1 )(k b0 )b0 + k [k u1 (k b0 )(u1 b0 )] .
(5.190)

No we use k b0 = k cos , where is the field inclination. We continue evaluating

(5.189):
{ }
2 u1 = c2s k(k u1 ) + c2A (k b0 )2 u1 (k u1 )(k b0 )b0 + k [k u1 (k b0 )(u1 b0 )] ,
(5.191)
thus
[( ) ]
2 u1 c2s
= k cos u1 (k u1 )k cos b0 + k 1 + 2 (k u1 ) k cos (u1 b0 ) .
2 2
c2A cA
(5.192)
We take the projections of (5.192) first into the direction of k

2 (k u1 ) = k 2 c2A cos2 (u1 k) c2A k 2 cos2 (k u1 )+

+ c2A (k u1 )k 2 + c2s (k u1 )k 2 c2A k 3 cos (u1 b0 ). (5.193)

Hence
[ 2 ]
+ c2A k 2 + c2s k 2 (u1 k) = c2A k 3 cos (b0 u1 ). (5.194)
Then we project (5.192) into the direction of b0 .

2 (b0 u1 ) = c2A k 2 cos2 (u1 b0 ) c2A (k u1 )k cos +

+ c2A (k u1 )k cos + c2s (k u1 )k cos c2A k 2 cos2 (u1 b0 ), (5.195)

hence
2 (b0 u1 ) = c2s (k u1 )k cos . (5.196)
Equation (5.194) may further be played with:

b0 u1 2 + c2A k 2 + c2s k 2
= (5.197)
u1 k c2A k 3 cos
5.11. GENERAL MHD WAVES 118

and similarly (5.196) becomes

b 0 u1 c2 k cos
= s 2 . (5.198)
u1 k

These two equations equal, thus we have

2 + c2A k 2 + c2s k 2 c2 k cos

2 3
= s 2 (5.199)
cA k cos

and therefore
4 2 k 2 (c2A + c2s ) + c2s c2A k 4 cos2 = 0. (5.200)

When assuming /k > 0 for the wave propagating outwards, we have a solution:

[ ]1
1 2 1 2
= (cs + cA )
2
cs + cA 4cs cA cos + 2cA cs .
4 4 2 2 2 2 2 (5.201)
k 2 2

This is the general dispersion relation for the magnetoacoustic wave. The one with
the plus sign is called a fast wave, while the one with the minus sign is called the slow
wave. Obviously, the speed of propagation depends on the direction of propagation with
respect to the vector of the background magnetic field.
For waves travelling along the magnetic field = 0 we have:
( )2 {
1 1 1 1 c2s
= (c2s + c2A ) c4s + c4A 2c2A c2s = (c2s + c2A ) (c2s c2A ) = , (5.202)
k 2 2 2 2 c2A

for cs > cA and opositely for cA > cs . In summary we may write that the fast mode has
a dispersion relation /k = max (cs , cA ), whereas the slow mode disperses according to
/k = min (cs , cA ). Note that as a special case of waves propagating along the field,
we obtained the dispersion relation for the Alfven wave. Also the slow wave, which is
purely acoustic, propagates. For waves travelling across the field for = /2 we have
( )2 {
1 1 1 1c2s + c2A
= (c2s + c2A ) c4s + c4A + 2c2A c2s = (c2s + c2A ) (c2s + c2A ) =
,
k 2 2 2 2 0
(5.203)
therefore only the fast wave propagates. In a special case of no background field, i.e.
b0 = 0 we simply have cA = 0 and thus
( )2 {
0
= , (5.204)
k c2s

and the wave just degenerates to the ordinary acoustic wave.

119 5.11. GENERAL MHD WAVES

5.11.1 Mode conversion

In the stratified medium, we should supply additional terms in our equations (namely
the gravitation term) and drop the requirement of the homogeneous background. In
some cases we may find surfaces, where the sound speed equals the Alfven speed, the
equipartition surface (prove that at this surface, the plasma equals one). Here all
the dispersion relations collapse and the waves are resonant. The use of the simple ray
approximation (which we were using all the time) is not appropriate in these regions,
finite-wavelength eects play an important role. Using the higher-order theory one may
find5 that a mode conversion may occur. Essential in the equipartition layers the waves
convert types, purely acoustic wave may become magnetoacoustic fast wave etc. The
ecienty of the conversion depends on an incident angle to the magnetic field.
The mode conversion is an eect which essentially prohibits reliable helioseismic
tomography of the Sun beneath sunspots. The sunpots are regions of the strong field,
thuse there definitely is the equipartition layer below the surface of the Sun. The (purely
acoustic) seismic waves excited by the vigorous surface convection propagate through
the convection zone (in the stratified medium, thus they do not propagate straight but
rather bend as the local sound speed varies), go through the equipartition layer and
partly convert to the magnetiacoustic modes. The undergo one more conversion after
total reflection from the surface. As a consequence the total power of the acoustic waves,
which is measured and analysed by helioseismology, is lower in the regions of the strong
field, leading to unknown large systematic errors in analyses.
Similarly, the acoustic waves in the solar atmosphere are totaly reflected at the tem-
perature minimum region (lower chromosphere, T 4200 K), where the acoustic cut-o
frequency is around 5.2 mHz (the cut-o frequency is a function of the local plasma state
g
parameters, cuto = 2c s
, where g is the local gravity acceleration. However, in the
magnetised regions, the magnetoacoustic waves are allowed to propagate along the mag-
netic field. Thus when the conversion takes place, the originally acoustic waves may leak
to the higher atmosphere as converted magnetoacoustic and Alfven waves. Numerical
simulations revealed that the inclined magnetic field acts as a factor lowering the cut-o
frequency, which is then modified to cuto = g2c cos
s
, where is the magnetic field
inclination to the local vertical. In the magnetised regions with inclined magnetic field
the sound waves excited by the solar convection may carry some non-negligible acoustic
flux to the chromosphere and thus contributing to its heating6 .

5
e.g. Cally, P. 2006: Dispersion relations, rays and ray splitting in magnetohelioseismology, Royal
Society of London Transactions Series A 364, p.333349
6
e.g. Sobotka, M., Svanda, k, J. et al. 2013: Dynamics of the solar atmosphere above a
M., Jurca
pore with a light bridge, Astronomy&Astrophysics, in press, ArXiv:1309.7790S
5.11. GENERAL MHD WAVES 120
Chapter 6

Diusion

6.1 Diusion in weakly ionised plasmas

In Section 4.2 we introduced an additional term to the Euler equation accounting for
collisions between particles. We defined the particle flux

= nu, (6.1)

where n was the particle density and u their speed.

The formalism used here is similar. For simplicity, let us assume that the particles
flow along the x-axis of the Cartesian coordinate system. By collisions occuring along
the spatial interval dx, the fraction of the flux d is lost. We may write

d = nn dx, (6.2)

where nn is the density of neutral particles, because we expect the collisions to be

dominanted by neutral particles. is the eective cross-section of the collisions. From
the eective cross-section we may define useful quantities s = 1/nn to be the mean
free path between the consecutive collisions and similarly the mean collisional frequency
= nn u, where the overbar indicates the averaging over the velocity space. The
averaging is performed because the quantum mechanics teaches us that the eective
cross-section may be a function of velocity. The mean period between the consecutive
collisions is than = 1/.
By taking the time derivative of (6.1) and (6.2) when for simplicity neglecting the
possible time dependence of other variables but the coordinates and velocity and using
the definition of we have
d du dx
=n = nn nu = nn nuu = nu, (6.3)
dt dt dt
Hence
du
= u, (6.4)
dt
which is the (average) velocity change due to the collisions.

121
6.1. DIFFUSION IN WEAKLY IONISED PLASMAS 122

Hence the Euler equation with magnetic field set to zero and with collisional term
added reads [ ]
u
mn + (u )u = qnE p mnu (6.5)
t
d
Lets investigate the system where dt u = 0 (in such system we find only a constant
motion due to the diusion). Then
qE KB T n
u= = , (6.6)
m mn n
q sgn q
where we used the definition of the particle flux. By introducing mobility = m
and diusion coecient D = KmBT
we have for the particle species j

j = sgn q nj E Dj n. (6.7)

For E = 0 we have
= Dn, (6.8)
which is the form of Fick law for diusion. The simple interpretation of this useful
relation is that the particles flow from locations with the higher density to lower density
regions, which is to be expected. In plasmas the Fick law is generalised to (6.7) noting
the eect of the electric field. One should also note that the coecients j and Dj may
be generally dierent for various particle species (most often electrons and ions) and
hence the diusion in plasmas may cause the separation of charges and thus drive the
electric field. However, to fulfill the quasineutrality of plasma, i = e = on scales
L D , which allows us to derive the necessary electric field which will keep the plasma
quasineutral.
By assuming the plasma approximation ne = ni = n we have

ni E Di n = ne E De n (6.9)

and hence
Di De n
E= (6.10)
i + e n
is the induced electric field triggered by the separation of charges. This is the repulsive
electric field, which stops the charges from further separation due to the diusion and
hence keeps plasma neutral on large scales. This electric field causes the electrons to
slow down and accelerate ions simultaneously, so that the bulk of plasma (with charges
separated on scales smaller than D ) diuses according to modifies Fick law. This is
the basic picture of ambipolar diusion.
Then
Di De n
= i = ni Di n =
i + e n
i Di i De i Di e Di i De + e Di
= n = n =
i + e i + e
Da n, (6.11)
123 6.1. DIFFUSION IN WEAKLY IONISED PLASMAS

where Da is the coecient of the ambipolar diusion. For e i and using the
definition of mobility coecient we have
( )
i De + e Di i Te
Da = Di + De = 1 + Di . (6.12)
i + e e Ti

Thus the coecient of the ambipolar diusion is mostly determined by the diusion
coecients for ions with some correction.
As an example, lets take the continuum equation

n
+ nu = 0, (6.13)
t
and hence
n
= nu = = Da n. (6.14)
t
This equation is similar to the heat conduction equation. It may be solved by assuming
the 1-D solution separating the variables n(x, t) = T (t)S(x), where T and S are some
functions. Then
dT
S = Da T S, (6.15)
dt
and hence
1 dT Da 1
= S = const = . (6.16)
T dt S
Both sides of the equations must equal the same constant because the left-hand side
depends purely on time, thus the right-hand side cant depend on time and vice versa
for the spatial dependency of the right-hand side. From the dimension reasons this
constant must have the dimension of the reciprocal time. The equation may then by
solved by parts
dT T
T = T0 e
t
= (6.17)
dt
and
d2 S 1
2
= S, (6.18)
dx Da
which has the solution in the series of harmonic functions.
This equation must be supplied with the boundary conditions. Lets assume a simple
model of plasma confined between two infinite planparallel walls, where the density of
the plasma vanishes outside interval x (L, L). Then the solution to the problem is
{ [ ] }
t (l + 12 )x

[ mx ]
t
n = n0 ae,l e e,l cos + ao,m e o,m sin (6.19)
L L
l=0 m=1

The coecients e,l and o,m may be determined from inserting the solution above into
(6.15) and comparing the terms one by one. For the first term (l = 0) in the cosine
series we have
6.1. DIFFUSION IN WEAKLY IONISED PLASMAS 124

t 1 x t
( )2 x
ae,0 e e,0 cos = Da ae,0 e e,0 cos (6.20)
e,0 2L 2L 2L
and hence
4L2
e,0 = . (6.21)
2 Da
For a general mode l we would similarly find

L2
e,l = . (6.22)
(l + 12 )2 2 Da

For the first term (m = 1) in the sine series we have

t 1 x t
( )2 x
ao,1 e o,1 sin = Da ao,1 e o,1 sin (6.23)
o,1 L L L

and hence
L2
o,1 = . (6.24)
2 Da
For the other modes m we similarly obtain

L2
o,m = . (6.25)
m2 2 Da
The lifetime of various modes e,l and o,m shows that the higher modes are attenuated
faster. Therefore, if there is a wild density profile at the beginning (in time t = 0), then
the profile is being simplyfied with time.
The question is whether it is possible to achieve the stationarity in the presence of
diusion. It is not if the continuity equation holds. It is possible in presence of the
source term,
nu = Q. (6.26)
In case of the collisional ionisation Q = Zn, where Z is the ionisation function, we
have
Z
nu = Da n = Zn n = n, (6.27)
Da
solution to which is again the series of harmonic functions.
The recombination on the other hand takes a dierent form, because it does change
the number of charged particles.
n
= n2 , (6.28)
t
where n2 dependence is the consequence of the recombination being proportional to
both the density of electrons and ions and the use of plasma approximation. Then
1 1 1
dn = dt = t. (6.29)
n2 n n0
125 6.2. DIFFUSION IN HIGHLY IONISED PLASMAS

For t we have n t 1
, which says that due to the recombination in plasmas, the
density of the charged particles drops as 1/t. The spatial part of the continuity equation
is not relevant to the problem, thus we neglect it from the solution.
This solution is important for understanding which process drives the diusion of
charged particles in a of left-alone cloud of plasma in time. Obviously, in initial moment,
the recombination will be the dominant process, as its time dependence is steeper for
small t. Contrary, when the density of the plasma of gets smaller, the diusion with a
exponecial time dependence prevails and further drives the diusion.

6.2 Diusion in highly ionised plasmas

We will show that in highly ionised plasmas, the diusion resembles the process of
recombination rather than ambipolar diusion. Lets take the fluid approximation to
deal with the problem.

u
= j B p (6.30)
t
E + u B = j (6.31)

Note that generally the specific resistivity of plasma is not a scalar value but rather
a tensor and thus may depend on direction. In plasma, it usually is the case. When
assuming the stationarity, by vector-multiplying (6.31) by B and using (6.30) we have

E B + (u B) B = E B B 2 u + B 2 u = p. (6.32)

Since u = u + u , then the two last terms on the left-hand side give together B 2 u
and thus
E B p
u = . (6.33)
B2 B2
The first term on the right-hand side is the same for both (electron and ion) fluids.
The second one on the other hand may drive the diusion. Lets study the particle flux
triggered by this term. The perpendicular particle flux is

n(KB Te + KB Ti )
= nu = n = D n. (6.34)
B2
Note that the diusion coecient is a function of particle density an may be for simpler
further computations decomposed as

D = 2An (6.35)

The factor of two is introduced so that (2Ann) = 2A (nn) = A n2 .

Then the continuity equation

n
+ (nu) = 0 (6.36)
t
6.3. DIFFUSION IN THE MAGNETIC FIELD 126

transforms to
n
= A n2 . (6.37)
t
Such equation may be solved by separation of variables, hence n(r, t) = T (t)S(r). Then
dt = AT S and
S dT 2 2

1 dT A 1
2
= S 2 = const = 2 (6.38)
T dt S
and in the case we solved previously. Then the temporal part has a solution
1 t 1
= 2+ (6.39)
T T0
and the form of depends on the spatial part S. For t
2
T , (6.40)
t
which reminds the solution of the change of the particle density caused by recombina-
tions. The dierence between the ambipolar diusion and the result we just obtained
is in the diusion coecient depending on particle density. Note that the result we
obtained is not consistent with the assumptions of the derivation of the Ficks law, as
we allow for non-stationarity here.

6.3 Diusion in the magnetic field

In presence of the magnetic field, the Euler equation retains an additional term. Let
us see what happens in the presence of the magnetic field. We must go back to the
double-fluid approximation, since collisions are essential in the process of diusion.
[ ]
u
mn + (u )u = sgn q en(E + u B) KB T n mnu (6.41)
t
Lets assume that the collisions are much faster than the overall evolution of the system
in time, or in another words dtd
c , , where is the collision frequency. Then the
left-hand side may be neglected. Thus in components in the Cartesian system with z
axis oriented along the magnetic field (B = Bez ) we have
n
mnux = sgn q enEx KB T + sgn q enuy B, (6.42)
x
n
mnuy = sgn q enEy KB T sgn q enux B. (6.43)
y
Along the magnetic field a classic diusion as derived above occurs.
By introducing = me
, D = Km BT
= c KeB
BT
we have
D n c
ux = sgn q Ex + sgn q uy , (6.44)
n x
D n c
uy = sgn q Ey sgn q ux , (6.45)
n y
127 6.3. DIFFUSION IN THE MAGNETIC FIELD

and
D n Ey KB T n
ux (1 + c2 2 ) = sgn q Ex + c2 2 sgn q c2 2 , (6.46)
n x B neB y
D n E x K BT n
uy (1 + c2 2 ) = sgn q Ey c2 2 + sgn q c2 2 . (6.47)
n y B neB x
The last two terms on the right-hand side obviously indicate the already known E-B
and diamagnetic drifts. Hence
vE + vD D n
u = 2 2
+ sgn q 2 2
E 2 2
, (6.48)
1 + 1/(c ) 1 + c 1 + c n

where v E = E B/B 2 and v D = p B/(qnB 2 ). Hence the generalised Ficks law

transforms to
n vE + vD
u = sgn q E D + . (6.49)
n 1 + 1/(c2 2 )
The last term represents the already familiar drifts in the plasma fluid, only with the
correction term for the diusion. Obviously, the correcting term is important for > c
hence in the weak field regime.
The other correcting terms have reciprocal signs, hence in the strong field the diu-
sion is suppresed. For c2 2 1 we have

KB T KB T
D and D = , (6.50)
mc2 m

hence the decay caused by diusion in plasmas have dierent rates in dierent directions.
In case the collisional frequency is proportional to m1/2 (and when T mv 2 m 2 )
then the diusion depends also on the species of the particles. Then

D m1/2 and D m1/2 . (6.51)

In the perpendicular direction the diusion is faster for ions than electrons, while in the
parallel direction the electrons difuse faster. It has to be noted than generally speaking,
the collisions are important for the particles to be able to penetrate across the magnetic
field.
6.3. DIFFUSION IN THE MAGNETIC FIELD 128
Chapter 7

Equilibrium and stability

7.1 Equilibrium
The state of equilibrium indicates the forces acting on plasma are in ballance. In the
investigation of the possible equilibrium of plasma we will use the fluid approximation.
However, it should be noted that this approximation might be too rough. In plasmas,
the drifting motions are usually the sources of instabilities driving the plasma from equi-
librium, and the drifting motions are largely not captured by the fluid approximation.
In investigating the stability, we assume that the forces are in ballance, hence the
solution is stationary. Hence the necessary condition of equilibrium is t 0.
We recongnise a stable equilibrium, where small perturbations deviating the system
from an exact ballance are attenuated, and an unstable equilibrium, when even a small
perturbation of the system grows and brings the system out of equilibrium. The equi-
librium generally is a non-linear problem. We will limit ourselves to discuss the linear
problems only by considering the small perturbations. To describe the equilibrium, the
MHD description suces, however it is usually not sucient to discuss the stability of
such equilibrium. Let us remind that the mechanical stability equals to the work that is
needed to bring the object from the stable equilibrium position to another (investigated)
equilibrium position. Should this work be negative, the position at which it is located
is not stable.
Let us describe a general concept of investigation of equilibrium on an example of
hydromagnetic equilibrium. Lets consider a plasma cylinder, where the magnetic field
aligns along the cylinder axis. The simplified equation of motion reads

u
= j B p. (7.1)
t

We search for a stationary state, hence we set u

t = 0 and the Euler equation degenerates
to
p = j B (7.2)

(Figures of various equilibria)

129
7.1. EQUILIBRIUM 130

It is clear now that there is an obvious ballance between the Lorentz force which supports
the pressure gradient. Such ballance may be achieved when an azimuthal current flows
through plasma cylinder. By taking the vector multiplication by B we have
B p = B (j B) = B 2 j B jB = B 2 (j j ). (7.3)
By assuming that j = j +j we obtain the expression for a required azimuthal current
B p B n
j = 2
= (KB Ti + KB Te ) , (7.4)
B B2
which is an expression for a diamagnetic current. Such situation may be viewed as if
a plasma cylinder is to be in an equilibrium, there must be a current induced, which
takes a form of a diamagnetic current. One should bear in mind that this expression
was derived when neglecting all other terms in the Euler equation.
One should also note that both j and B are perpendicular to p, hence both j and
B lie in the isosurface of constant pressure. The field lines and the current lines may
be curved in this surface.
By considering Amperes law
B = 0 j (7.5)
we have
1
p = j B = ( B) B =
0
[ ]
1 1 1
= [(B) B + (B )B] = (B )B B 2
(7.6)
0 0 2
By rearranging we have
( )
1 2 1
p + B /2 = (B )B. (7.7)
0 0
The right-hand side indicates the action of forces along field lines, the magnetic tension.
If we assume that there are negligible changes of the magnetic field along the field lines,
the right-hand side vanishes. Then
B2
p+ = const. (7.8)
20
The second term has a physical meaning of the magnetic field pressure. By comparing
the two terms we define the plasma beta,

n KB Ta
p
= a B2 , (7.9)
pmag
20

where a identificates the particles by type.

For > 1 the gas pressure prevails. This high-beta regime is common to interior
of stars and most of the laboratory experiments. For < 1 the magnetic field pressure
prevails. This low-beta regime is common to solar/stellar atmosphere(s) and also to
interstellar matter.
131 7.1. EQUILIBRIUM

7.1.1 Diusion of the magnetic field into plasmas

Let us discuss the system of two limites half-spaces, one of them filled with plasmas
only, and the other one containing the magnetic field in a vaccuum. In case there is
no resistivity in plasmas, the two regions will remain disconnected. The electromotoric
force induced by any possible motion of the field would drive an infinite electric current
on the boundary separating these two regions, but the field would never penetrate the
plasma. However, the shape of the boundary may change significantly. If a pressure
balance (7.8) is to be hold, then it must be that if we somehow force the magnetic field
to penetrate the plasma region, the gas must leave such a region.
Since (7.8) holds, automatically = 1 on the boundary. For the plasma half-space
the second term vanishes and vice versa for the magnetic field half-space. However,
by considering the finite resistivity of the plasma we practically allow the two distinct
fluids to mix. In the following, we want to estimate the time scale of this process. Let
us start from a Farradays law of induction, Amperes law, and an generalised Ohms
law:
B
E = (7.10)
t
B = 0 j (7.11)
E + u B = j (7.12)

Let us assume that the plasma is stationary, hence u = 0 and thus the second term
on the left-hand side of the Ohms law vanishes. Note that even when we assume no
bulk motion of the plasma, there still might flow electric currents through plasma, as
we have two distinct charged species in the plasma. Remind that the velocity of the
bulk motion is defined as
1 M ui + mue
u= (ni M ui + ne mue ) , (7.13)
M +m

hence as a weighted mean of the velocities of the two species, while the current is defined
as a dierence of the velocities of the two species:

j = e(ni ui ne ue l) ne(ui ue ). (7.14)

By combining equations (7.10), (7.11), and (7.12) we have

B
= j = [( B) B] = B, (7.15)
t 0 0

where we used B = 0.
Such equation may be solved using the separation of variables. For simplicity, we
approximate the operator by division over the length-scale L. Then we solve the
equation
B
= B, (7.16)
t 0 L2
7.2. INSTABILITIES 132

which has an solution in a form of exponential

[
]
t
B = B 0 exp , (7.17)

from which we take only the negative solution as physical, since we solve for a dissipation
of the magnetic field into plasma.
0 L2
= (7.18)

is a characteristic time-scale of the penetration of the magnetic field into plasma.
The solution might be also understood as the magnetic field dissipating due to the
induction of the electric currents in plasma, which dissipate via the Ohmic heating. Let
us estimate how much of the energy of the magnetic field dissipates due to the Ohmic
heating processes.
The loss output of the Ohmic heating is given by

P = j 2 . (7.19)

Hence the energy drained from the magnetic field over the time is

W = j 2 . (7.20)

The electric current is given by the Amperes law (7.11). Hence the energy loss (again,
when approximating the operator) we have
( )2
B 0 L2 B2
W = j 2 = = . (7.21)
0 L 0

It turned out that all the total energy of the magnetic field dissipates and heats up the
plasma. Actually it is by factor two larger, which is due to our rough estimates.

7.2 Instabilities
The systems out of the thermodynamical equilibrium contain some of the free energy,
which may drive the rise of random fluctuations. In this case, for instance the self-
excitation of plasma waves is possible. There is no other mean to self-excite plasma
waves. Of course, they might be driven from the outside. In this case, we speak about
the unstable equilibrium.
We recognise basically four types of plasma instabilities and in the further sections
we will study some of the examples.

Stream instabilities A stream of plasma or and electric current propagates through

plasma. A spectrum of waves is excited that lead to an instability of the system.
133 7.2. INSTABILITIES

Rayleigh-Taylor instability Should there be a sharp density gradient in plasmas

or the plasma be confined and there is an external force (for instance, gravity),
the external force drives the growth of instabilities. A classical example of non-
plasmatic Rayleigh-Taylor instability is a glass of water turned upside down, when
the atmospheric pressure supports the water in the glass. The system is allowed
to be in an unstable equilibrium, however once perturbed, the instability grows.

Universal instability Is called universal because it is always present in the confined

plasmas (and the plasmas are always confined somehow). The confined plasma
is not in the total thermodynamical ballance even when there are no important
external forces. The pressure (expanding the plasma) drives the instabilities.

Kinetic instability The kinetic instability is driven by the non-equilibrium distribu-

tion function (usually non-Maxwellian). An anisotropy in the distribution function
may drive the instabilities.

7.2.1 Two-stream instability

Let us study first one example of stream instabilities, the two-stream instability. Let us
assume a homogeneous plasma in the coordinate system, where ions are stationary. Let
us allow electrons to have a systematical speed u0 with respect to the ion background.
We assume u0 = 0. The plasma is cold (Ti = Te = 0) and there is no magnetic field
(hence B 0 = 0).
Then the linearised Euler equations read:
ui1
M n0 = en0 E 1 , (7.22)
[ t ]
ue1
mn0 + (u0 )ue1 = en0 E 1 . (7.23)
t

We search for electrostatic waves propagating in the x direction, where E 1 ex

k u0 . Let us use the approach of solving the equations in the Fourier space (using the
Fourier transform (5.10) ) similarly to the problem of deriving the dispersion relation
of plasma waves. Hence

iM ui1 = eE1 , (7.24)

m(i + iku0 )ue1 = eE1 . (7.25)

The continuity equation for ions reads (with assumptions applied: n0 = 0 and ui0 0)
ni1
+ n0 ui1 = 0, (7.26)
t
where we again consider only the x-component of the dierential operator. We express
ni1 and further use (7.24) to get
k ien0 k
ni1 = n0 ui1 = E1 . (7.27)
M 2
7.2. INSTABILITIES 134

Similarly, the continuity equation for electrons reads

ne1
+ n0 ue1 + (u0 )ne1 = 0, (7.28)
t
hence the parallel component in the Fourier space is

(i + iku0 )ne1 + ikn0 ue1 = 0, (7.29)

from which we express ne1 by using (7.25):

kn0 ue1 iekn0
ne1 = = E1 . (7.30)
ku0 m( ku0 )2
We do not assume the plasma approximation to be valid in this case, thus we further
use Poisson equation to finally derive the dispersion relation
e
E1 = (ni1 ne1 ), (7.31)
0
where the density perturbations of the two species are given by (7.27) and (7.30). Then
[ ]
e 1 1
ikE1 = iekn0 + E1 (7.32)
0 M 2 m( ku0 )2
and thus
[ ] [ ]
e2 n0 m/M 1 2 m/M 1
1= + = p + . (7.33)
0 m 2 ( ku0 )2 2 ( ku0 )2
This is the dispersion relation of the two-stream instability. Please note that for u0 = 0
and m/M 0 this relation degenerates to 2 = p2 , hence to normal plasma oscillations.
This is the equation of the forth order in . The equation may either have four real
roots, each of them will correspond to the individual oscillatory mode, hence it will not
describe an instability. The other option is that the equation will have two real roots
and the remaining two will be complex, one conjugated to the other. Let us discuss
these two. They might be symbolically written as

j = j + ij , where j = [j ] and j = [j ]. (7.34)

In the case of the complex roots, the solution for e.g. the electric field reads

E1 = Eex exp [i(kx t)] ej t . (7.35)

Since the two complex roots are complex conjugates, one of them will have j > 0 and
at least one of the modes will depict an exponential growth.
To study the dispersion relation at least qualitatively, we introduce new variables
= /p and 0 = ku0 /p . Then the dispersion relation reads

m/M 1
1= + = F (; 0 ). (7.36)
2 ( 0 )2
135 7.2. INSTABILITIES

(figure here)

Figure 7.1: Two-stream instability

We draw F (; 0 ) as a function of for the given 0 . Two possible plots are displayed
in Fig. 7.1.
It turns out that for small 0 the plasma is unstable (function F (; 0 ) has two
conjugated complex roots). In another words, for a given u0 , the plasma is unstable
with respect to perturbations with small k (and hence large wavelength). In this case
an internal inconsistency appears. The linearisation approach to solve the system of
initial equations will no longer be valid, the assumptions of small perturbations will
break. Our solution is therefore no longer correct. We just derived that such solution
qualitatively is unstable (that is a correct result), however quantitatively (deriving the
exponential growth) is not correct.

7.2.2 Gravitational instability

Let us consider another example: the space is divided into two half-spaces, one empty
and one occupied by plasma. The Cartesian coordinate system is set so that the ex
is normal to the boundary between the two half-spaces and ey and ez lie within the
boundary, perpendicular to each other. The backround magnetic field is parallel to axis
z. The plasma has the density gradient parallel to axis x. The plasma is cold and the
magnetic field is homogeneous. We allow for a plasma bulk velocity parallel to axis y.
Then in the state of equilibrium the Euler equation reads

M ni0 [(ui0 )ui0 ] = eni0 ui0 B 0 + M ni0 g. (7.37)

Note that we added an ad hoc term of the gravity force. For g constant also ui0 is
constant and hence (ui0 )ui0 0. Then

0 = eni0 ui0 B 0 + M ni0 g (7.38)

We multiply (7.38) by B 0 to obtain

0 = eB 0 (ui0 B 0 ) + M B 0 g = eB02 ui0 eB 0 ui0 B 0 + M B 0 g (7.39)

and hence
M B0 g g
ui0 = 2 = ey , (7.40)
e B0 c
where ui0 is the velocity of the ion drift in the field of gravity. The drift of electrons is
negligible, since they have m/M -times smaller eect. Regarding the other drifts, there
is no E-B drift, since E 0 = 0, there is also no diamagnetic drift, because we assumed
the cold plasma. However, the conclusions might be dierent, when the boundary is
perturbed and corrugated (see Fig. 7.2). Then the following eects occur:
1. The ions drift and will cause the accumulation of the charge. The electrons react
with some delay.
7.2. INSTABILITIES 136

(figure here)

Figure 7.2: The origin of the gravitational instability

2. The accumulated charges cause the additional electric field E 1 .

3. Now the E-B drift appears with drift velocity pointing parallel to the normal
direction to the boundary.
Let us search for the increment of the growth. Let us assume that there is a wave
propagating parallel to ey . For ions we introduce the perturbations to the density and
velocity, similarly as we did in the case of the plasma waves. The the Euler equation
reads:
{ }

M (ni0 + ni1 ) (ui0 + ui1 ) + [(ui0 + ui1 ) ](ui0 + ui1 ) =
t
e(ni0 + ni1 ) [E 1 + (ui0 + ui1 ) B 0 ] + M (ni0 + ni1 )g (7.41)

We further proceed slightly dierently from the procedure we took when studying the
waves. We append (7.41) with (7.37) multiplied by 1 + ni1 /ni0 to obtain the system of
two equations:
{ }

M (ni0 + ni1 ) (ui0 + ui1 ) + (ui0 )ui0 + (ui0 )ui1 + (ui1 )ui0 =
t
e(ni0 + ni1 ) [E 1 + (ui0 + ui1 ) B 0 ] + M (ni0 + ni1 )g (7.42)
M (ni0 + ni1 )(ui0 )ui0 = (ni0 + ni1 )eui0 B 0 + M (ni0 + ni1 )g. (7.43)
ui0
By subtracting the two some of the terms cancel, and when we further use t = 0 and
(ui1 )ui0 = 0 we have to the first order
[ ]
ui1
M ni0 + (ui0 )ui1 = eni0 (E 1 + ui1 B 0 ). (7.44)
t
It seems that apparently the term with the gravity acceleration disappeared. An explicit
term containing g indeed disappeared, however we have to keep in mind that the gravity
acceleration is still hidden in ui0 !
Equation (7.44) in the Fourier space transforms to

M ( k ui0 )ui1 = ie(E 1 + ui1 B 0 ). (7.45)

Let us write this equation in components by assuming E1x = 0, since we search for
wave propagating in the y-direction. We further use k = kex and ui0 = ui0 ex . For ions
we have
ie
ui1x = ui1y B0 , (7.46)
M ( kui0 )
ie
ui1y = (E1y ui1x B0 ). (7.47)
M ( kui0 )
137 7.2. INSTABILITIES

By combining the two we obtain for uiy

( )
ie ieB02
ui1y = E1y + ui1y (7.48)
M ( kui0 ) M ( kui0 )
and hence
[ ]1 [ ]1
ieE1y e2 B02 /M 2 ieE1y 2c
ui1y = 1 = 1 . (7.49)
M ( kui0 ) ( kui0 )2 M ( kui0 ) ( kui0 )2

To obtain a relation for ui1x we simply insert (7.49) into (7.46) and have
[ ]1
ec E1y 2c
ui1x = 1 . (7.50)
M ( kui0 )2 ( kui0 )2

By further assuming low-frequency waves, hence using approximation 2c (

kui0 )2 , we have
E1y kui0 E1y
ui1x = , ui1y = i (7.51)
B0 c B0
and analogously for electrons
E1y kue0 E1y
ue1x = , ue1y = i , (7.52)
B0 c B0
where c c and thus ue1y 0.
The continuity equation for ions is
ni1
+ (ni0 ui0 ) + (ui0 )ni1 + ni1 ui0 + (ui1 )ni0 + ni0 ui1 + (ni1 ui1 ) = 0.
t
(7.53)
The second term vanishes, because by assumptions ni0 ui0 , forth term vanishes
because we assumed that ui0 = 0 and the last term on the left-hand side also
vanishes, because it is the second order term. Hence in the Fourier space we have
ni0
ini1 + ikui0 ni1 + ui1x + ikni0 ui1y = 0. (7.54)
x
Similarly we construct the continuity equation for electrons, however it is simpler, be-
cause we assumed that ue0 ui0 and we showed that ue1y 0. We will also expect
the low-frequency waves, hence the use of plasma approximation n1 = ne1 = ni1 and
n0 = ne0 = ni0 is justified. In the end, we should show the consistency of the results
with the assumptions. Thus for electrons we have
n0
in1 + ue1x = 0. (7.55)
x
By combining (7.54) and (7.51) we have
E1y n0 kui0 E1y
in1 + ikui0 n1 + i2 kn0 = 0, (7.56)
B0 x c B0
7.2. INSTABILITIES 138

which trivially turns into

E1y n0 kui0 E1y
( kui0 )n1 + i + ikn0 = 0. (7.57)
B0 x c B0

Similarly we combine (7.55) and (7.52) to get the equivalent of (7.57):

E1y n0
in1 + = 0, (7.58)
B0 x

which allows us to express the term E1y /B0 and insert it into (7.57). Then

n1 n0 kui0 n1
( kui0 )n1 n0 x
kn0 n0
= 0. (7.59)
x
c x

After some algebra we obtain

ui0 c n0
( kui0 ) = . (7.60)
n0 x

Now we use (7.40) and have

g n0
( kui0 ) = , (7.61)
n0 x
or ( )
n0
2 kui0 g x
= 0. (7.62)
n0
This is the dispersion relation of the wave we searched for. The solution is
[ ( )] 1
n0 2
1 1 2 2
= kui0 k ui0 + g x . (7.63)
2 4 n0

We will have instability (and exponential growth) in case is complex. That will be
fullfilled for ( )
n0
1
g x > k 2 u2i0 , (7.64)
n0 4
thus g and n0 must have oposite signs. By waving the hands, the lighter fluid
supports the heavier one. Similar situation we had in case of the glass of water turned
upside down.
Similar instability appears in the column of plasma, when the gravity force is re-
placed by the action of the centrifugal force. The conclusions are the same. The
boundary of the plasma column is then corrugated and the instability is termed flute
instability. The onset of this instability is observed in the experiments of the laser-
induced fusion. The idea of the laser-induced fusion is that the fusion begins faster
than the flute instability sets on and desintegrates the plasma confinement.
139 7.2. INSTABILITIES

(figure here)

Figure 7.3: Universal instability

7.2.3 Universal instability

We will proceed similarly to the previous section, however, the g B drift will be
replaced by the diamagnetic drift. Since in this drift there is no scaling with mass of
the particles, we have to solve equations for both ions and electrons.
For the background velocities we have relations:

KB Ti 1 n0
ui0 = uDi = ey (7.65)
eB0 n0 x
KB Te 1 n0
ue0 = uDe = ey , (7.66)
eB0 n0 x
We expect the non-zero component kz of the wavevector, the motion of electrons
along the background magnetic field is important to shield the electrostatic field induced
by ions. We will further assume that the Boltzmann relation (4.87) fully describes the
density perturbation of the electrons.

ne1 e1
= . (7.67)
n0 KB Te
The geometry of the problem is sketched in Fig. 7.3.
At point A, the density is larger than in equilibrium, which is denoted by the solid
line, hence ne1 > 0 and thus 1 > 0. At point B, the density is smaller than in
equilibrium, hence ne1 < 0 and 1 < 0. Thus between points A and B there must be an
electric fields E 1 , which drives the E-B drift with the drift velocity
E1 B0
u1 = . (7.68)
B02

There will be a wave in direction ey , hence both ne1 and 1 oscilate in time. Therefore
also ue1 oscillates in time and in fact the oscillations in the drift velocity are the cause
for the oscillations of density. As a result, there will be a motion of the plasma fluid
in x direction. Let us support this physical scenario by some equations. The Fourier
image of (7.68) is
Ey iky 1
u1x = = , (7.69)
B0 B0
where we used the Poisson equation E 1 = 1 = ik1 . The speed of the drift is
the same for both ions and electrons. Let us further assume for simplicity, that the
fluid of plasmas is incompressible and oscillates only in the x direction. Mathematically
speaking: u1x = u1x (x) and kz ky . Than the continuity equation for divergence
centres reads:
n1 n0
= u1x , (7.70)
t x
7.2. INSTABILITIES 140

other components are being neglected. We express (7.70) in the Fourier space and use
Boltzmann relation to handle n1 for electrons and (7.69) to handle u1x . We have:
n0 iky 1 n0 e1
ine1 = ue1x = = in0 , (7.71)
x B0 x KB Te
which solves as
KB Te 1 n0
= , (7.72)
k eB0 n0 x
which is exactly the expression (7.66). We confirmed the physical picture sketched in
words a few paragraphs above. The perturbation waves propagate with the speed of
the diamagnetic drift, hence termed drift waves.
However, the dispersion relation shows that not instability occurs, it does not have
any term assuring the growth. We did not prove that this configuration is unstable. Let
us only comment that the approach we took is not correct enough and that we neglected
two important ingredient: the polarisation drift and the drift in the inhomogeneous
electric field. The polarisation drift shows up in the case we have a variable electric
field. Then an additional drift with a speed of vp = sgn q c1B dE
dt t needs to be taken into
account. We will further only draw a physical picture in words: the corrections by the
additional two drifts causes the phase shift between 1 , which depicts a delay between
u1 and n1 . Hence, in places, where the plasma is already skewed (n1 > 0), u1 directs
out of the plasma and vice versa. The perturbation on the boundary grows.

7.2.4 Landau damping

The highlight of these lecture notes consists of an example of the kinetic instability. We
started this text with an introduction to the statistical physics to describe important
properties of the ensemble particles. We will end the text by investigating the oscillatory
motions in plasmas, which occur in the case of the particle distribution function not
being in equilibrium. We need to leave the fluid approximation and need to go one
step deeper, to the single-particle distribution function. Let us remind that for the
single-particle distribution function f the Boltzmann equation may be derived,
( )
f h f h f f
+ = , (7.73)
t pi ri ri pi t coll
p 2
where h = 2m + V (r) is a Hamiltonian of one particle. We showed that this equation
may be rewritten as ( )
f F f f
+ v f + = , (7.74)
t m v t coll
where F are the external forces comming from the external potential V (r). By playing
explicitly with the term on the right-hand side to extract the term of the self-potential,
which has a similar shape as the action of the external potential, we got
( )
f F f f
+ v f + = , (7.75)
t m v t coll
141 7.2. INSTABILITIES

where F is the averaged action of both external forces and the self potential com-
bined. Than the Vlasov approximation consists of neglecting the right-hand side, which
is equivalent of neglecting the mutual correlation of distribution function of dierent
particles. For the further case, we substitute F by the Lorentz force (as an eect of
the sum of both external and internal electric and magnetic fields). Hence finally we
have a Vlasov equation to solve:
f q f
+ v f + (E + v B) = 0, (7.76)
t m v
where E and B represent the joint action of both external and internal electric and
magnetic fields. In the following, we will neglect the external fields and in the case
of the magnetic field, we will neglect also the internal one. Alltogether, we use the
following approximations:
1. B = 0,

2. E 0 = 0, but E 1 = 0,

3. f = f0 + f1 , f0 = f0 (v), f1 = f1 (t, r, v),

4. all perturbations are expressed by means of the Fourier series as A1 = A1 exp [i(kx t)],

5. and A1 A0 and A a a . The perturbations are much smaller than the

1 A0

background values and also the derivatives of the perturbations are much smaller
than the derivatives of the background.
We solve the problem for electrons. Thus the Vlasov equation (7.76) under the assump-
tions listed above reduces to
e f0
if1 + ikf1 vx = E1x , (7.77)
m vx
hence
ieE1x f0 /vx
f1 = . (7.78)
m kvx
we add the Poisson equation

+
en1 e
E 1 = ikE1x = = f1 d3 v. (7.79)
0 0

Let us rescale the distribution

function so that f0 = n0 f0 (v), where the normalisation of
the newly introduced f0 is f0 d3 v = 1. This definition has an advantage of n0 , which
may be put outside of the integrals, because it is not a function of v. Let us integrate
(7.78) over the whole velocity space

+
+
ieE1x f0 /vx 3 ikE1x 0
3
f1 d v = n0 d v= , (7.80)
m kvx e

7.2. INSTABILITIES 142

Figure 7.4: A sketch of the integration path bypassing the pole in /k.

where we also used (7.79). From that we obtain

+
+
e2 n 0 f0 /vx 3 p2 f0 /vx 3
1= d v= 2 d v. (7.81)
0 mk kvx k k vx

Let us assume that f0 is a background equilibrium distribution, essentially having the

shape of Maxwell-Boltzmann distribution. Than we may decompose f0 as a product of
f0 (v) = f0 (vx )f0 (vy )f0 (vz ), (7.82)

with a normalisation f0 (v)dv = 1. Than (7.81) simplifies to

+
+
+
+
p2 f0 (vx )/vx p2 f0 /vx
1= 2 dvx dvy f0 (vy ) dvz f0 (vz ) = 2 dvx . (7.83)
k vx k
k vx k

We will let be complex. The function (7.83) has a pole in /k. By using the residuum
theorem for the integration of the complex functions, we must integrate along the closed
curve, but bypass a pole, which results in an additional contribution of the residuum of
this pole. According the the integration path displayed in Fig. 7.4 we have
R
I + Iv.p. + u1 + u2 + + u3 + u4 = 0, (7.84)
2
where I is the integral we want to evaluate (note the negative sign as we integrate
[ against ]
f0
the direction of the x axis), Iv.p. is the principal value of this integral, R = 2i v x
k
is a residuum in the pole and the segments u1 to u2 may be made infinitezimally small.
(7.83) may thus be written as
[ ]

p2 f0
+

f0 /vx
1 = 2 v.p. dv x + i . (7.85)
k vx k vx
k

The integral may be solved by parts

+ [ ] +
f0 dvx f0 f0 dvx
= + . (7.86)
vx vx /k vx /k (vx /k)2
+
143 7.2. INSTABILITIES

The first term vanishes due to the fast decay of the distribution function. Note that
the phase speed v = /k and let v be positive, i.e., we will deal with the positive
velocities only. Lets approximate using the Taylor expansion
( )
2 2 2vx 3vx2
(vx v ) v 1 + + 2 . (7.87)
v v

Then (7.86) turns into

+

+

+
+
2
f0 dvx 2vx 3vx
v2 f0 dvx + f0 dvx + f0 dvx =
(vx v )2 v v2

[ ] ( )
2 3v
2 3v 2
= v2 1 + vx + 2x = v2 1 + 2x , (7.88)
v v v

where we used vx = 0, which certainly is true for the symmetrical velocity distribution,
such as the Maxwell-Boltzmann one.
Lets continue to deal with (7.85):
[ ]
p2 k 2 3v 2 f0
2
1 = 2 2 1 + 2x + i . (7.89)
k v k 2 vx
k

Let us make a step to the side. If we were to neglect the residuum in the pole, we
would neglect the last term. By further using 21 mvx2 = 12 KB Te we have
( )
p2 k 2 KB Te
1= 2 1+3 2 . (7.90)
m

If we further assume that p in the second order (in another words 4 2 p2 ) we

have
KB Te
2 = p2 + 3k 2 , (7.91)
m
which is the dispersion relation for electron electrostatic waves. By neglecting the
contribution in the pole we obtained the classical result.
The new physics shows up when we properly treat the residuum in the pole. For
simplicity, lets omit the term with the thermal motion, which (as we just showed)
results in the electrostatic electron waves. The dispersion relation then reads
[ ]
f0
2
2 = p2 1 + i 2 . (7.92)
k vx
k

Using the expansion to the Taylor series to the first order we have
[ ]
1 f0
2
p 1 + i , (7.93)
2 k 2 vx
k
7.2. INSTABILITIES 144

which for p gives

[ ]
1 p f0
2
p 1 + i . (7.94)
2 k 2 vx
k

This relation has a suitable form of the complex number with separated purely real and
purely imaginary parts.
By going back to the definition of the Fourier series A1 = A1 exp [i(kx t)], the
f0
new results will come obvious. For v x
> 0 () > 0 and thus we have an instability.
f0
For vx< 0 () < 0 the external perturbations (such as waves) are damped.
So if we plot the distribution function and the perturbation appears at the given
phase speed, the derivative of the distribution function decides whether this perturbation
f0
will be damped or will grow. Note that for the Maxwell-Boltzmann distribution, v x
<0
for all speeds, thus in for plasma in a perfect equilibrium the perturbations are always
damped. This eects is termed the Laundau damping.
For a Maxwell-Boltzmann distribution, we may obtain an explicit term of the imag-
inary part, which is
( ) ( )3 [ ]
p 1
0.22 exp 2 , (7.95)
p kvT 2k D

which shows that the waves with wave numbers comparable to the Debye shielding
length are the most eectively damped.
A dierent situation appears when the stream of particles having a dierent typical
speed perturbs the distribution function and creates a bump. Part of this bump will
f0
have v x
> 0. If the system is perturbed e.g. by the waves with the phase speeds in the
susceptible interval, the instability will occur. The energy exchange between the wave
and the particles will occur (see Fig. 7.5). A mixture of two Doppler-shifted maxwellian
distributions is unstable.
Laundau damping eect play an important role in the heating of the solar corona
(and also coronae of Sun-like stars), where the outer layers of the atmosphere (coronae)
are much hotter (millions of degrees) than the below located photospheres (thousands
of degrees). The heating is therefore due to the non-thermodynamical eect, such as the
dissipation of the coronal currents flowing along the loops of the magnetic fields, thermal
energy from the all-scale reconnection processes and the heating due to the MHD waves
propagating to the coronae. The waves exchange the energy with the coronal plasma
by means of the Landau damping.
145 7.2. INSTABILITIES

(Figure here)

Figure 7.5: The system in the frame moving with phase speed v . At point A the speed
of the particle is a little smaller than v , hence the particle falls behind into the point
x , where the electric potential is higher and thus pushes the particle forward. Particles
having v < v are being accelerated. Contrary, particles with v > v are decelerated. If
in total there is less faster-than-the-wave particles than slower-than-the-wave particles,
the energy streams from the wave to the particles and the perturbation is damped. This
f0
is the case of Landau damping, when v x
< 0. The other case is obvious.
7.2. INSTABILITIES 146
Appendices

147
Appendix A

Guiding centre motion

A.1 Alternative derivation of (3.133)

cos 0 cos B
1
(e b)(e )B = sin 0 sin 2 B =
0 b 0 3 B

b sin
= b cos (cos 1 B + sin 2 B) =
0
b sin cos B + b sin2 B
1 2
= b cos2 1 B b sin cos 2 B =
0

b2 B 0 1 B
1 1
= b1 B = 0 2 B =
2 2
0 b 3 B
1
= b B (A.1)
2

149
A.2. DERIVATION OF EQUATIONS (3.137)(3.139) 150

cos
cos

1 b

(e [e ] b = sin sin 2 b =
0 0 3 b

cos
1 b1 1 b2 1 b3
= sin (cos , sin , 0) 2 b1 2 b2 2 b3 =
0 3 b1 3 b2 3 b3
cos
cos 1 b1 + sin 2 b1
= sin cos 1 b2 + sin 2 b2 =
0 cos 1 b3 + sin 2 b3
sin cos 1 b3 + sin2 2 b3

= cos2 1 b3 cos sin 2 b3 =

sin cos 1 b1 sin 2 b1 + cos 1 b2 + cos sin 2 b2
2 2

2 2 b3
1

= 12 1 b3 = 1 (e1 2 e2 1 )b3 + 1 b(1 b2 2 b1 ) =

2 2
12 2 b1 + 12 1 b2
1 [ ] 1
= b( b)3 + O (b[1,2] ))2 = b[b ( b)]. (A.2)
2 2
Note that in the second step we expressed the dot product by means of matrix multi-
plication respecting the rule a b = aT b = abT , where a is the matrix representation of
vector a.

A.2 Derivation of equations (3.137)(3.139)

Derive equations for variables in the perpendicular plane

dv E u2
= B + v E Dt b (A.3)
dt 2B
du v u u
= B ( v E b v E ) (A.4)
dt 2B 2
d B v
= e2 Dt e1 b (v b + v E ) (A.5)
dt 2