See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/299424379

Lecture Notes in Plasma Physics. Short and Simple.

Book · September 2012

CITATIONS READS
0 2,066

1 author:

Mikhail Malkov
University of California, San Diego
165 PUBLICATIONS   2,298 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cosmic Ray Acceleration in shocks View project

Magnetic Fusion Energy Studies View project

All content following this page was uploaded by Mikhail Malkov on 25 March 2016.

The user has requested enhancement of the downloaded file.

Plasma Physi s 218A
 Part I.

Unmagnetized Plasmas

2
 Physics 218A

3
Contents

I. Unmagnetized Plasmas 2
1. Introdu tion to basi plasma on epts 6

1.1. Fourth state of matter, some examples . . . . . . . . . . . . . . . . 6

1.2. Long-range collective interactions between particles. Self-consistent fields in plasmas 7
1.3. Quasineutrality of plasmas, Debye shielding . . . . . . . . . . . . . 7
1.4. Non-equilibrium character of plasmas, collisional relaxation, drag force on particles, run-away electrons
1.5. Basic mathematical description: Kinetic equation with self-consistent field. Vlasov plasma 13
Appendix: BBGKY hierarchy[1] . . . . . . . . . . . . . . . . . . . . . . 14
1.6. Collisionless vs collisional hydrodynamics. One vs two-fluid model. Basic nonlinearities 16
1.6.1. Transport coefficients . . . . . . . . . . . . . . . . . . . . . 19
1.7. Simple equilibrium solutions, Maxwell-Boltzmann distribution function 21

2. Plasma Wave Motion 23

2.1. Langmuir (plasma) waves . . . . . . . . . . . . . . . . . . . . . . . 23

2.2. Electromagnetic waves in plasmas. Laser plasmas. Mode conversion. 27
2.2.1. Penetration of EM waves (laser light) into plasma . . . . . . 31
2.3. Collisionless (Landau) damping. Landau rule for singularity in plasma dispersion function 36
2.3.1. Ion Langmuir and ion-acoustic waves . . . . . . . . . . . . 42
2.4. Plasma surface waves . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5. Nonlinear waves, solitons and BGK- waves . . . . . . . . . . . . . 45
2.5.1. Langmuir waves . . . . . . . . . . . . . . . . . . . . . . . 45
2.5.2. Ion acoustic nonlinear waves and solitons . . . . . . . . . . 48
2.5.3. BGK waves . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3. Plasma Instabilities and wave-parti le intera tions 52

3.1. Beam-plasma instability . . . . . . . . . . . . . . . . . . . . . . . 52

3.2. Particle trapping and saturation of instability . . . . . . . . . . . . . 53
3.3. Quasilinear theory. Resonant and nonresonant wave-particle interactions. Energy and momentum conser
3.4. “Quasi-linear” integral, relaxation of wave packets and unstable beams 57
3.5. Parametric instability, Weak turbulence and wave cascading [2] . . . 59
3.6. Nonlinear Landau damping (induced scattering) . . . . . . . . . . . 65
3.7. Wave Energy Pathways. Plasmon Condensate . . . . . . . . . . . . 67
3.8. Modulational instability and Langmuir wave collapse . . . . . . . . 69
3.9. Anomalous resistivity . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.9.1. Double Layers . . . . . . . . . . . . . . . . . . . . . . . . 75

4
 Physics 218A

II. Magnetized Plasmas 77

4. Parti le Orbits in Magneti Field [3, 4, 5, 6℄ 78

4.1. Guiding center (drift) theory . . . . . . . . . . . . . . . . . . . . . 78

4.1.1. ExB drift . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1.2. Gradient drift . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.3. Curvature drift . . . . . . . . . . . . . . . . . . . . . . . . 81
4.1.4. Adiabatic invariants and particle mirroring . . . . . . . . . 81

5. Fluid des ription of magnetized plasmas 83

5.1. Two-fluid description . . . . . . . . . . . . . . . . . . . . . . . . . 83

5.2. Magnetohydrodynamics (MHD) . . . . . . . . . . . . . . . . . . . 83
5.2.1. Flux Freezing . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2.2. Plasma Equilibrium in Magnetic Field . . . . . . . . . . . . 85
5.3. Linear waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.4. Stability of Plasma Equilibria . . . . . . . . . . . . . . . . . . . . . 90
5.4.1. Energy principle of stability . . . . . . . . . . . . . . . . . 94

5
1. Introdu tion to basi plasma on epts

1.1. Fourth state of matter, some examples

Simple definition: ionized gas

Electrons, ions (single or multiply ionized) and, in some cases, neutrals

Universe

Most of the visible matter in the Universe is in plasma state

⊲ the Sun and other stars are huge self-
gravitating spherical plasmas
⊲ on the top of the Earth’s atmosphere is
the ionosphere, then plasmasphere and finally
magnetosphere, all filled with plasmas
⊲ radiation belts, solar wind and the entire
Heliosphere
⊲ further out, the interstellar medium
(ISM)
⊲ intergalactic medium (IGM)

Te hnology: Nuclear Fusion Energy, Mag-

netic + Inertial (laser plasmas). Largely
driven by the following reaction

D + T →2 He4 + n + 17MeV

Everyday life: Neon lights, flames, electric sparks...

6
 Physics 218A

1.2. Long-range olle tive intera tions between parti les.

Self- onsistent elds in plasmas

Consider a gas of charged particles. Compare it to the ideal gas (which is characterized by weak
interaction between particles): Particle interaction energy to kinetic energy in plasmas (average
distance between particles r = n−1/3 ):

e2 /r e2 n 2 1 r2
= ·r = ≡ ε /4π ≪ 1
T T 4π λD2
(replacing kT → T , we use energy units for T , e.g. eV rather than ◦ K). Characteristic length λD
naturally occurred in the above energy ratio which is known as
q
Definition. Debye Length (radius) λD ≡ T /4π e2 n

ε ≪ 1 sometimes called the plasma parameter. Refers to “ideal plasma” , “gaseous plasma”.
ε & 1 strongly coupled plasma
Other definition of plasma parameter ([large] number of particles in the Debye sphere):

4π 3 4π −3/2
Definition. Λ ≡ λD n = ε
3 3
Again, an ideal plasma is the one with Λ ≫ 1.

1.3. Quasineutrality of plasmas, Debye shielding

Debye scale is an important scale of plasmas. How does this scale emerge in formal calculations?
Consider the following situation

• assume ne = ni = n0 , on average
p
• in most cases ions move slowly VTi = m/MVTe ≪ VTe

• electrons are distributed in the electric potential φ according to the Boltzmann rule ne =
n0 eeφ /Te

• Poisson equation for a plasma with a point charge q at the origin

 
∆φ = 4π en0 eeφ /Te − 1 − 4π qδ (r) (1.3.1)

Naturally, assume a spherical symmetry, eφ /Te ≪ 1, so for r 6= 0 we have

1 ∂ 2∂ 1
r φ− 2φ =0 (1.3.2)
r ∂r ∂r
2 λD
with r
Te VT
λD = = e
4π e n0 ω pe
2

7
 Physics 218A

rewrite (1.3.2):

∂2
 
1 1
rφ − 2 rφ =0 (1.3.3)
r ∂r 2 λD

• general solution for r > 0

rφ = Ce−r/λD

• to obtain constant C, submit this solution to eq.(1.3.1) and take the limit r → 0:

1
C∆ = −4π qδ (r)
r
• as ∆ (1/r) = −4πδ (r) (the Green function of Poisson equation), we obtain C = q and
finally

q
φ = e−r/λD
r

Con lusion: each point charge q (potential in vacuoφ0 = q/r) is screened by a cloud of elec-
trons attracted by this charge (repelled, if q is negative).

Note: Size of the plasma is important: L ≫ λD

Quasineutrality is an important concept:
Take 1cm3 of air under normal condition, ionize and separate to a ∼ 0.5cm
Force:
F = N 2 e2 /a2 ∼ 1016 lb (!)
As N ∼ 1019 , ionization potential I ≃ 13.6 eV, total ionization energy only ~100 Joules.
Note Instead of assuming static ions to pro-
vide the constant charge density that neu-
tralize (on average electrons), one may as-
sume Boltzmann distribution
√ for both species
which yields λD → λD / 2. Validity of either
assumption depends on the problem consid-
ered (time-scale).
Plasma frequency. While λD is a spatial
scale of charge separation, there is also a fre-
quency (time scale) characterizing plasma re-
sponse to charge perturbations.
Consider a plasma cube with electrons displaced by ξ with respect to the ions. The dis-
placement will create negatively and positively charged layers on the opposite faces of the cube.
These, in turn, will induce electric field inside the volume

E = 4π enξ (1.3.4)

8
 Physics 218A

The E-field will return electrons to the equilibrium according to the equation

d2ξ
m = −4π e2 nξ (1.3.5)
dt 2
So electrons will oscillate with respect to the ions at a frequency called the electron plasma
frequency
q
ω pe = 4π e2 n/m

Note Ions oscillate at the amplitude ∼ m/M of that of electrons. However, if electrons are
ultra-relativistic, they may be considered as a fixed background charge, not affected by
the oscillatory electric field.
p In this case p
ions will respond to the field oscillating at the ion
plasma frequency ω pi = 4π e n/M = m/M ω pe . Electron plasma frequency becomes
2
∗ = ω /√γ , where γ = 1 − v2 /c2 −1/2 . Such situation occurs, e.g., in laser plasmas,
ω pe


pe
where electrons are often energized to ultra-relativistic temperatures and become much
less responsive to electric fields than ions. The roles of the species are thus largely ex-
changed.

Take home lesson: Quasineutrality may only be violated either at scales . λD or during a time
. ω p−1 .

Degree of Ionization
No need for a plasma to be completely ionized in the above treatment. Neutrals can be added.
May or may not play a role depending on the phenomenon studied and degree of ionization.
Often, ion-neutral collisions become important.
Typically, if T . 0.2eV –neutral gases, solid states.

1.4. Non-equilibrium hara ter of plasmas, ollisional

relaxation, drag for e on parti les, run-away ele trons

High mobility of electrons in plasmas makes them extremely responsive to, e.g., electric fields,
both external and self-generated. How does plasma relax? Simplest way:

Collisions

May be characterized either by the mean free path, collision frequency, or cross section.

Relevant Length is also λD ≈ 800 T /n

p

(T/eV, n cc), beyond which the scattering cen-

ter has no effect (screened).
Close encounter:

9
 Physics 218A

Bohr radius is clearly too small a0 =

h̄2 /me2 ≈ 5 · 10−9 cm
More important may be the de Broglie
wavelength h̄/mv at close collisions
Scattering is strong (χ ∼ 1) when the im-
pact parameter b is such as to make potential
energy of interaction of the order of kinetic
energy

mv2 ∼ e2 /b
(if Z 6= 1, e2 → Ze2 , etc.)
Cross section σ = π b2 ∝ v−4 ∝ T −2 . Impact factor required for χ ∼ 1

b ∼ b∗ = e2 /T = r̄3 /4πλD2

b∗ /r̄ ∼ r̄2 /λD2 (= ε ) ≪ 1

Large-angle scatterings are rare. Mostly small angle scattering.
The mean free path

λ = 1/nσ
Collision frequency

ν = VT /λ

Therefore, typical particle trajectory is a smooth wavy curve rather than a zigzag trajectory, as
in ordinary gases. Reason: long range Coulomb interactions.

Perpendicular momentum

dv⊥ e2 e2 b
m = −∇⊥ =
dt r (b2 + v2t 2 )3/2
e2 ∞ e2 1 ∞ d sinh u 2e2
ˆ ˆ
bdt
v⊥ = = 3
=
m −∞ (b2 + v2t 2 )3/2 m vb −∞ cosh u mvb
Here we substituted sinh u = vt/b.

v⊥ 2e2
χ≈ =
v mbv2
decrease in velocity parallel to the initial direction

δ v = v (1 − cos χ ) ≈ vχ 2 /2

10
 Physics 218A

ˆ
1
∆v = v χ 2 dN
2
After traveling distance dx, particle encounters

dN = ndxds = 2π ndxbdb
scattering centers.
Define the m.f.p. λ as dv¯ = −vdx/λ

e4
ˆ
1 bmax
dv = − v χ 2 (b) dN = −4π 2 3 ndx ln
¯
2 m v bmin
Integral diverges at both limits. Upper limit

bmax = λD .
The simplest approach to bmin : consistency requirement, χ . 1, i.e. bmin = b∗ . Finally for the
mfp (e → Ze)

1 m2 v4 1
λ=
4π n Z 2 e4 3 ln (λD /r̄)

Note more accurate bmin - de Broglie wavelength. Only weak dependence on b’s. 3 ln (·) ∼
10 − 20. In most practical cases leave the ln (bmax /bmin ) ≡ LC as a parameter in your
answer varying in the range ∼ 10 − 20. (Other names for LC : Λ, ln Λ, λ , etc.)

Collision frequency ν = VT /λ , Collision time τ = 1/ν .

Four types of collisions: ei, ie, ee, ii (ie − ie are not really different)

• mass ratio M/m = 1836

• therefore, in the ei collisions the energy exchange is much slower than in the ee and ei

• electrons can be heated (because of they high mobility) and stay hot for the above reason

• above treatment of collisions, formally done for e on an i at rest, generalizes straightfor-

wardly to other collisions by transforming to the CM frame, replacing m with the reduced
mass µ = ma mb /(ma + mb ). Note, that the CM velocity V = ∑ ma Va / ∑ ma is conserved

Equilibration

p
First, between electrons, then between ions ( M/m times slower, since their speed is less by
this factor for the same temperature), finally ei. When an electron hits an ion, only m/M fraction
of energy is transferred. Thus, equilibration times:
p
tee : tii : tei = 1 : M/m : M/m

11
 Physics 218A

Pra ti al formula
n 1eV
νei = 3 · 10−6 LC
cm−3 T 3/2
Other collision rates

λii ≃ λ (Ti /Te )2 = λei (Ti /Te )2

νee ∼ νei

νii /νee ≃ m/M (Te /Ti )3/2

p

Appli ation: Run-away electrons [7]. Consider, e.g., a thunderstorm. There are some free
electrons (produced by cosmic ray ionization) in the air and electric field that may accel-
erate them.

Drag force on electrons, while they collide with neutrals F = −νen mv :

4π NZ 2 e4
 2
mv
F= ln
mv 2 zεi
where N is the density of air molecules, z is the mean nuclear charge of the nitrogen and oxygen
atoms, Z = 2z ≈ 14.5, εi ≈ 15eV - ionization energy. Electron energy ε ≪ mc2 . In the opposite
case

4π NZ 2 e4
 2 
mc
F= ln γ
mc2 ε1

−1/2
where γ = ε /mc2 = 1 − v2 /c2 , ε1 ≈ 270eV. Force F has a minimum at γ = 3 − 4,

4π NZ 2 e4
Fmin = Ln
mc2
with Ln ≃ 10.
Equation:

dv v
m = eE − F
dt v
Projection onto v direction

dv
m = eE µ −F (v) , where µ = E · v/Ev = cosϑ
dt
(1.4.1)
Projection onto E direction

12
 Physics 218A

dµ e
E 1 − µ2


= (1.4.2)
dt mv
Run-away boundary: µ = µc (v) ≡ F (v) /eE.
Qualitatively, for the run-away process to be
macroscopic the eE > F (VT ) ≡ eED , where ED is the so called Dreicer field. One can also
introduce a critical velocity Vc , according to eE = F (Vc ), so that particles with V < Vc do not
run-away, contrary those with V > Vc and µ > µc .

1.5. Basi mathemati al des ription: Kineti equation with

self- onsistent eld. Vlasov plasma

Our treatment of plasma deviation from quasineutrality has been applied to either static or ho-
mogeneous plasmas leading to Debye shielding and plasma oscillations, respectively. Clearly,
those are very limited results, but it is impossible to follow individual particle dynamics, and
not necessary, in fact. A more promising approach – kinetic equation for particle distribution
function f (r, v,t) that has a meaning of the phase space density of particles, or the number of
particles in drdv volume at a point r, v:

dN (r, v,t) = f (r, v,t) drdv

so that the total number of particles
ˆ
N= f (r, v,t) drdv

Then, the number density at r,t:

ˆ
n (r,t) = f (r, v,t) dv

As f /n is the probability density, other moments, such as the average speed can be defined
ˆ
1
u (r,t) = v f dv (1.5.1)
n
If particles do not collide with each other, their number in the infinitesimal phase space volume
drdv does not change, so the number density stays constant while the volume moves in the (r, v)
phase space under the action of forces and inertia (Lioville’s theorem, see Appendix below for
a formal derivation). Note that the form of the phase space cell drdv may change very strongly
and become extremely irregular as the initially neighboring particles separate from each other in
course of time. The equation for f manifests the conservation of the phase space density along
characteristics (particle trajectories)

df ∂ f dr ∂ f dv ∂ f
≡ + + =0 (1.5.2)
dt ∂t dt ∂ r dt ∂ v
Or, using the equations of motion

13
 Physics 218A

∂f ∂f e ∂f
+v + E =0 (1.5.3)
∂t ∂r m ∂v
Formally, this coincides with a Lioville’s equation (or collisionless Boltzmann)

∂f
+ {H , f } = 0
∂t
where the Hamiltonian

p2
H = + eφ (1.5.4)
2m
with E = −∇φ and

∂H ∂ f ∂H ∂ f
{H , f } = −
∂p ∂r ∂r ∂p
being the Poisson brackets.

Note Any f = f (H ) is a steady state solution if H does not depend on time.

However, the field E is now associated with the plasma motion itself. This is often called a
Vlasov approach, Vlasov plasma, self-consistent field description etc. Eq.1.5.3 thus needs to be
supplemented with an equation for E which is the Poisson’s equation

∇ · E = 4π e (ni − ne ) (1.5.5)
with
ˆ
ne,i (r,t) = fe,i dv

Other forces may be included in addition to eE/m, such as magnetic (and/or gravitational) force
by adding the respective terms to the Hamiltonian (1.5.4). The equations for magnetic field
(Maxwell) should be added to the system with the current calculated from the respective mo-
ments of fe,i (see 1.5.1).

Appendix: BBGKY hierar hy[1℄

BBGKY stands for Bogolubov, Born, Green, Kirkwood and Yvon.

Eq.(1.5.3) can be obtained from an N- particle Lioville equation which is equivalent to a
complete description of dynamics of N identical particles interacting via electromagnetic fields.
Denote xi ≡ (ri , pi ) the particle coordinates (i = 1, . . . , N) in the 6N-dimensional phase space.
Let’s D (x1 , . . . , xN ,t) be a probability density such that all N particles are at the respective points
in the intervals dx1 , . . . , dxN with a probability

dw = Ddx1 dx2 . . . dxN .

14
 Physics 218A

D is normalized to unity:
ˆ
Ddx1 . . . dxN = 1

The description of particle dynamics using D is equivalent to an individual particle approach and
is clearly not productive for N ≫ 1. Of interest is not D but a less detailed function f (x1 ,t) which
provides information about the probability of finding one particle at the point x1 regardless of
positions of other particles (averaged over all possible such positions). In some cases, however,
simultaneous coordinates of 2-3 particles are of interest. We write the one-particle distribution
as
ˆ
f1 (x1 ) = N Ddx2 . . . dxN

Similarly, for s- particle distribution

ˆ
s
fs (x1 , . . . , xs ) = N Ddxs+1 . . . dxN (1.5.6)

The Lioville equation can be written as follows1

∂
D + {H , D} = 0 (1.5.7)
∂t
where the Hamiltonian
N
1 e2
H = ∑ Hi +
i=1 2 i6∑
=i′ |ri − ri |
′

with

p2i
Hi = + eφ ext (ri ,t)
2m
where φ ext denotes an external electric field potential (not to be confused with the self-consistent
field, represented by the second term in H ).
The main idea behind the transition from the full description given by N-particle D in eq.(1.5.7)
to the one particle eq.(1.5.3) is in truncation of the coupled system of equations for fs . Indeed, in-
tegrating (1.5.7) according to (1.5.6), the equation for f1 ≡ f (r, v,t) (our main interest) will con-
tain f2 (x1 , x2 ) and so on. Two major assumptions are made here: (i) N → ∞ but n = N/V < ∞,
(ii) the plasma parameter 1/nλD3 → 0. Then, the pair correlations C2 in the following equation
can be neglected: f2 (x1 , x2 ) = f1 (x1 ) f1 (x2 ) +C2 (x1 , x2 ) (see, e.g., [1] for details). The equation
for the one-particle f (r, v,t) (with the f2 split according to the above rule) takes the following
form
1 See, e.g. [8]. This equation can be understood by considering a continuity equation for a 6N-dimensional phase
space “fluid” of the density D and velocity components ẋi . The continuity equation is ∂ D/∂ t +(∂ /∂ xi ) (ẋi D) = 0,
where repeated indices imply summation, as usual. As the velocity components satisfy the Hamiltonian equations
ṙi = ∂ H /pi , pi = −∂˙H /∂ ri , the flow is incompressible ∂ ẋi /∂ xi = 0.

15
 Physics 218A

∂f ∂f
∂
+v· + e Eext + Esc f =0 (1.5.8)
∂t ∂r ∂p
where the self-consistent field component results from the above splitting of f2 and can be writ-
ten as

∂ f (x′ ) dx′
ˆ
sc
E (r,t) = −e
∂r |r − r′ |
It has a meaning of electrostatic field generated by all plasma particles. It is called self-consistent,
self-generated etc. field. With Eext = 0 eq.(1.5.8) coincides with (1.5.3) and the self-consistent
field is to be determined from eq.(1.5.5).

1.6. Collisionless vs ollisional hydrodynami s. One vs

two-uid model. Basi nonlinearities

Strictly speaking, hydrodynamic description requires frequent collisions to hold fluid (plasma)
particles in the fluid element. Vlasov collisionless plasma seems to be an opposite case, where
particles traverse any fluid element. But, one can attach certain characteristics to it, such as the
number density of particles, their average velocity (vector) and its dispersion (tensor). Besides,
the lhs of eq.(1.5.3) is the same as that of a kinetic equation for strongly collisional plasma
with the particle collision term on the rhs. Finally, this collision term should conserve certain
moments of f , such as n, so that, after taking moments, the respective equation will be the same.
First, let’s take a zero moment. Rewrite eq.(1.5.3) as (with added collision integral that makes
a Boltzmann equation)

∂f e ∂
+∇·vf + E f = St { f } (1.6.1)
∂t m ∂v
Note ”St” stands for Stoss, collision (German, historic notation). Other notations C { f }, (∂ f /∂ t)c ,
St f . In its most accurate representations (Landau, Lenard-Balescu), St- term is a compli-
cated integro-differential operator. It can be derived systematically (but difficult) by con-
tinuing calculations briefly outlined in Appendix to Sec.1.5 with correlations included.
There exists a convenient simple model-form of the collision term, the so-called τ - ap-
proximation, or Bhatnagar–Gross–Krook operator (BGK, for short). It assumes that a
particle distribution relaxes under collisions to an equilibrium solution f0 , on which the
St-term is known to vanish. So, one simply writes St{ f } = − ( f − f0 ) /τ , with τ = ν −1
being the characteristic time between collision.

For collisions between like particles, apart from their number2 , also the momentum and energy
are conserved

mv2
ˆ ˆ ˆ
St { f } dv = 0, mvSt { f } dv = 0, St { f } dv = 0 (1.6.2)
2
2 It is assumed that no new particles are created in collisions.

16
 Physics 218A

Now integrate eq.(1.6.1) over v (third term vanishes as f → 0, for v → ∞):

∂n
+ ∇ · nu = 0 (1.6.3)
∂t
Then, multiply by mvi and integrate again

∂ ←→
ρ u + ∇ · Π = enE (1.6.4)
∂t
where the mass density ρ = mn and the momentum density flux tensor

Πi j = ρ ui u j + Pi j
with the pressure tensor
ˆ
Pi j = m (vi − ui ) (v j − u j ) f dv (1.6.5)

Using (1.6.3) and returning to vector notations, we transform (1.6.4) to standard form of Euler-
Navier-Stokes equation

du ←→
ρ + ∇ · P = enE
dt
←→
where P is the pressure tensor (1.6.5) and the substantial (total) time derivative

d ∂
= +u·∇
dt ∂t
As in the conventional hydrodynamics, it is defined on a particular vector field u as a derivative
along the streamline. Similarly, multiplying (1.6.1) by mv2 /2 we obtain the energy equation

∂
  2    2  
u u ←→
ρ + ε + ∇ · ρu +ε +u· P +q = j·E (1.6.6)
∂t 2 2
where
ˆ ˆ
m m
ε= 2
(v − u) f dv, q = (v − u)(v − u)2 f dv, j = enu (1.6.7)
2ρ 2
Similarly to Navier-Stokes equation, using (1.6.3,1.6.4) the energy equation can be simplified,
so that all three equations can be represented as follows

∂n
+ ∇ · nu = 0 (continuity)
∂t
du ←→
ρ + ∇ · P = enE (momentum) (1.6.8)
dt
dε ←
→ ← →
ρ + U : P +∇·q = 0 (energy)
dt
←
→
where U is the strain tensor

17
 Physics 218A

∂ ui ∂ u j
 
1
Ui j = + .
2 ∂ x j ∂ xi
Eqs.(1.6.8) are valid for collisionless plasmas. However, they are also valid for plasmas with col-
lisions between like particles, e.g. for interpenetrating electron and ion fluids, interacting with
each other only through electric field. Often, however, collisions between unlike particles are
more important as they lead to momentum and energy exchange between species. Eqs.(1.6.8)
can be generalized for this case by taking into account the additive character of St-term in (1.6.1),
e.g. St { fe } = Stee { fe } + Stei { fe } etc. Both Stoss components conserve the number of particles
but Stei does not necessarily conserve energy and momentum in the sense of (1.6.2). However,
the sum Stei + Stie does, so that the collision related terms on the r.h.s. of the momentum and en-
ergy eqs.(1.6.8) will be of the form ±F and ±Q, respectively. These terms represent the friction
force between electron and ion fluids and the heat exchange. The ± signs reflect conservation
of the total momentum and total energy in the e + i plasma. Note that, as electric field enters
the momentum equations with opposite signs, the relative motion between both fluids naturally
occur.
Eqs.(1.6.8) do not form a closed system, and a meaningful truncation requires further assump-
tions. For example, in many cases it is reasonable to assume that f (v) is isotropic in the local
fluid frame moving with velocity u. In this case the pressure tensor
ˆ
1
Pi j = δi j P, P = m (v − u)2 f dv (1.6.9)
3
and q = 0. Further simplification is achieved in true (collisional) hydrodynamic regime when
the characteristic time and spatial scales are longer than collision time and the m.f.p., ∂ /∂ t ≪ ν ,
λ ∇ ≪ 1. Then, to the leading order eq.(1.6.1) can be written simply as St { f } = 0 and its
solution is a Maxwellian
!
 m 3/2 m (v − u)2
f0 = n exp − (1.6.10)
2π T 2T
Here the three fundamental moments of f0 , that are n, u and T are to be determined from the
next order. Denoting the lhs of eq.(1.6.1) as L { f }, linearizing St { f } (it is a nonlinear operator
in f ) for small f1 = f − f0 ≪ f , i.e. St { f } ≈ Lc { f1 }, where Lc is a linear operator, the next
order will be

L { f 0 } = Lc { f 1 }
As this is an inhomogeneous linear equation for f1 , the condition for its unique solution to exist
impose certain restrictions on its inhomogeneous part (similar to the conventional orthogonality
to the solution of the homogeneous equation). This results in a requirement that n, u and T are
defined by (1.6.10) exactly, i.e. the integrals 1.6.2 vanish on f1 .
We may repeat now our derivation of eqs.(1.6.8) separately for electron and ion fluids while
the linearized collision term will provide the momentum and energy exchange between the flu-
ids. The continuity equations remain the same (as St term conserves the number of particles
individually) while the friction forces will be given by

18
 Physics 218A

ˆ ˆ
Fei = m St { fe } vdv = −Fie = −M St { fi } vdv

The equations of motion (momentum equation) for the two-fluid hydrodynamics can be written
as follows

due
ρe = −∇Pe − ene E + Fei
dt (1.6.11)
dui
ρi = −∇Pi + eni E + Fie
dt
The simplest form of the friction force can be obtained by using the BGK collision term, which
yields

Fei = −ρe ν (ue − ui) . (1.6.12)

Turning to the energy equation, for the local Maxwellian we obtain

3T 2
ε= , P = nT = ρε
2m 3
for each fluid. Submitting this to (1.6.8) we obtain

3 dT
+T∇·u = Q (1.6.13)
2 dt
For Q = 0, this is simply an adiabatic law. Using continuity equation,

d 5
P + P∇ · u = 0
dt 3
or, since ∇ · u = −ρ −1 d ρ /dt,

d P
=0
dt ρ γ
- adiabatic motion with the specific heat ratio γ = CP /CV = 5/3. Note that this for 3D motion.
Plasma often moves effectively only in 1D (e.g. along magnetic field), or in 2D (e.g., across
magnetic field). Then γ = (d + 2)/d, which can be deduced from the definition of P in (1.6.9).
Including the energy exchange between species, in τ - approximation, one gets for example
for ions the exchange term
mn
Qi = 3 (Te − Ti )
Mτ

1.6.1. Transport oe ients

Electroconductivity Here, the electron transport is determined by a balance between their ac-
celeration in electric field and the friction force with the ion-fluid. In the τ - approximation this
balance reads

19
 Physics 218A

e ∂f
− E = −νei ( f − f0 )
m ∂v
Recalling the run-away process, this is a regime in which the electric field is below the Dreicer
field. We can use (1.6.11) and (1.6.12).
m
Fei = −ρe ν (ue − ui ) = νei j
e
where j is the total current, which is almost entirely carried by electrons because by virtue of
their higher mobility. Neglecting then electron inertial and pressure forces, we obtain from the
electron momentum equation 1.6.11 the Ohm’s law

j = σE
with the elctroconductivity

e2 n 1 ω pe
2
σ= =
mνei 4π νei
Thermoconductivity Here, important are ee, ei, may be ii but not ie collisions. In a quasi-steady
state, collision term is balanced by the gradient of the distribution function (instead of E- term
above):

v · ∇ f0 = −τ −1 ( f − f0 )
Assume the axis z is directed along the local ∇T , then f0 is a Maxwellian of the form:
3/2
mv2
  
m
f0 = n exp − (1.6.14)
2π T (z) 2T (z)
Obviously, f0 does not contribute to the thermal flux q (by symmetry), so we calculate f1 =
f − f0 = −τ vz ∂ f0 /∂ z which is antisymmetric in vz and does generate qz :

τ n ∂ T T ∞ −t ∞ 2 4
ˆ ˆ  
2 2 3 2 2
−y2
qz = − √ dte y y + 2y t + t − y + t e dy
π ∂z m 0 −∞ 2
where we introduced t = mv2⊥ /2T and y2 = mv2z /2T , so that

5 T ∂T
qz = − τ n (1.6.15)
2 m ∂z
Comparing this result with (1.6.8,1.6.13) we may rewrite eq.(1.6.13) including the thermal flux
as follows

dT 2 ∂ ∂T T
+ T∇·u = κ , κ = 5τ (1.6.16)
dt 3 ∂z ∂z m

20
 Physics 218A

1.7. Simple equilibrium solutions, Maxwell-Boltzmann

distribution fun tion

The τ - approximation describes relaxation to an equilibrium which is merely postulated to be

a Maxwellian. How do we obtain the Maxwellian mathematically? A rigorous solution is pro-
vided by the Landau collision integral which is based on the Fokker-Planck approach, utilizing
probability of scattering of particles in short time intervals resulting in small velocity steps. The
process is equivalent to particle diffusion in velocity space. The derivation is somewhat long but
a simplified version may be obtained from the following consideration.
First, recall that particle collisions result in a dynamical friction Fd = −mν v that systemati-
cally decelerate particles moving too rapidly compared to the others (cf run-away in an electric
field). However, particles also diffuse in v because of the random kicks received from other
particles. This process, by contrast with the friction, increases the particle velocity on average.
Note that individual collisions resulting in diffusion may decrease or increase the particle ve-
locity v0 . The average growth of velocity is because there is more room to move in v > v0 part
of the velocity space than in 0 < v < v0 . Taken together with the friction, the process acts as
a thermostat bringing particles to an equilibrium. Particles that move too fast slow down by
the friction, but if they move too slow they are energized by diffusion. These two phenomena
manifest themselves in two terms of collision operator. Consider first a simple 1D case. The first
term comes from v̇∂ f /∂ v in eq.(1.5.2) with v̇ = −Fd /m. The second term describes diffusion in
velocity space ∂ /∂ vD (v) ∂ f /∂ v, where the diffusion is caused by the same collisions of the rate
ν . Therefore, D = ν ∆v2 with ∆v ∼ VT . Accurate calculations are lengthy, but we can attempt
at writing ∆v = VT and see whether the two terms of the collision operator balance at a valid
particle distribution. Thus we write

∂ T ∂
 
St ( f ) = ν v + f
∂v m ∂v
As it turns zero on f0 (v) eq.(1.6.14), the guessed D is correct. This result can be straightfor-
wardly generalized to the 3D case, including collisions with other species at the frequencies
νa :

∂ T vi v j ∂
 
St { f } = ∑ νa vi + f.
a ∂ vi m v2 ∂ v j
Consider, as an example, an equilibrium state when an electric field is present. On writing
E = −∇φ and assuming a steady state, from eq.(1.6.1) we have for electrons

∂ ∂ T vi v j ∂
 
e
v∇ f + ∇φ f= νa vi + f
m ∂v ∂ vi m v2 ∂ v j
The lhs vanishes for any function depending on the integrals of motion. Therefore, f = F (E )
the particle energy

1
E = mv2 − eφ
2

21
 Physics 218A

The rhs requires that f = G (r) exp −mv2 /2T . Both requirements are satisfied by the following

Maxwell-Boltzmann distribution

mv2 − 2eφ
 m 3/2  
f = n0 exp −
2π T 2T
Being integrated over velocity, this solution provides the conventional Boltzmann distribution
for electron density in the electric potential φ that we already used

n = n0 eeφ /T .
One has to be careful with cases of strong electric fields E & ED , extended over the distances
much longer than the mfp, where particles with high velocities, v > Vc , run away and the steady
solution does not exist.

22
2. Plasma Wave Motion

2.1. Langmuir (plasma) waves

There are two approaches to study wave phenomena in plasmas. One approach is to start from
the most general equations and select solutions of interest. In many cases, it is a very long
(and tedious) way. The other approach is to discard what appears to be unimportant right at
the beginning and get to the mode of interest more quickly. At the end one may still look at
the result and estimate whether the other terms were discarded legitimately. Here, we take the
second route.
Consider fast electron oscillations using one - fluid approximations. Ions are at rest. The
equations for electrons are

∂n
+ ∇nv = 0 (2.1.1)
∂t
dv e 1
= ∇φ − ∇p (2.1.2)
dt m mn

∆φ = 4π e (n − n0 ) (2.1.3)
Before considering any waves it is important to understand what is the equilibrium state of the
plasma. Here, it is the simplest state n = n0 = const, v = φ = 0, p = const. Assume n, v, φ are
perturbed as follows

n = n0 + ℜnk exp (ikr − iω t) ≡ n0 + |nk | cos (kr − ω t + α )

and similar expressions for the other variables, with v0 = φ0 = 0. Since we will linearize equa-
tions in nk ≪ n0 etc, we may omit ℜ’s and ℑ′ s. Note, that a strong restriction of possible plasma
motion is imposed by the potential character of electric field, E = −∇φ , so that Ek = −ikφk .
Here we will concern with only one mode with a specific number k, so we use the scalar index
in Fourier components, nk etc.
Substituting the perturbations of the above form and linearizing the equations, we get

n0 kvk − ω nk = 0 (2.1.4)
e 1
ω vk + kφk − kpk = 0 (2.1.5)
m mn0
k2 φk + 4π enk = 0 (2.1.6)

23
 Physics 218A

Now we need an equation of state to connect n and p, p = p (T, n) = nT . We expect the process
to be very fast so that compressive temperature perturbations are not erased by the thermocon-
ductivity and we assume an adiabatic motion, p ∝ nγ . Since v′ k k, the motion in a single wave
is clearly one-dimensional, that is γ = 3. Thus

pk = 3T nk
From these equations we obtain the following dispersion relation

ω 2 = ω pe
2
+ 3k2VTe
2
(2.1.7)

In the long wave limit, k → 0, we thus recover the result (1.3.5) for the electrons oscillating as
a whole with respect to the ions at rest. The phase velocity of these waves Vph ≡ ω /k → ∞ for
k → 0 so that they do not interact with particles resonantly. The dispersion relation simplifies in
this limit:
 
3 2 2
ω = ω pe 1 + k λD (2.1.8)
2
Rewriting eq.(2.1.7) in the form

2
ω pe
2
2
Vph = 2 + 3VTe
k
shows two different restoring forces that support plasma oscillations. The first one is associated
with the electric field caused by the electron displacement with respect to the ions (eq.[1.3.5]).
The second one is the electron pressure which is similar to the main restoring force in acoustic
waves where Vph 2 = C 2 = ∂ p/∂ ρ with p ∝ ρ γ , γ = 3 and C being the sound speed. Our choice
s s
of γ = 3 is also justified by the fact that an alternative, 3D adiabatic index γ = 5/3 would
require very strong collisions to achieve the equipartition of the pressure perturbations between
all three degrees of freedom. Note, however, that γ = 5/3 was actually the original Langmuir’s
approach, who was the first to study the plasma (Langmuir) waves. However, in situations
where collisions are indeed strong enough to ensure equipartition in a ∼ ω pe −1 time, these waves

are heavily damped by these collisions and virtually do not exist.

The dispersion relation above also shows the significance of the charge separation in plasma
oscillations. As we know, such separation is possible only at scales shorter than λD particularly
if time variations are slow, i.e. kVT & ω (Boltzmann distribution is justified). In a short wave
limit, however, Vph ∼ VTe , so that thermal particles will resonantly interact with the wave thus
absorbing the wave energy without collisions. The charge separation effects (that require short
scales L ∼ k−1 < λD ) are also strong in this regime and particles are energized by these fields,
which leads to a strong collisionless wave damping, Sec.2.3. Qualitatively, this may be under-
stood by transforming to the wave frame moving at v = Vph , so that a (sinusoidal) wave profile
(electric potential φ (x)) becomes stationary. If Vph ≫ VT , there are only a few supra-thermal
particles that move together with the wave and strongly interact with it via a Cerenkov resonance
ω = kv. For an equilibrium (e.g., Maxwellian) plasma ∂ f /∂ v < 0. Therefore, there are more

24
 Physics 218A

particles that are slower than the wave so that they receive energy from the wave and the latter
decays.
In fact, it is not difficult to calculate the wave damping rate γk approximately, using the con-
servation of total energy of the resonant particles and the wave. By definition, for a wave with
the wave number k we have

1 dWk
γk = − (2.1.9)
2Wk dt
where Wk is the wave energy (as it is quadratic in amplitude, Wk (t) = Wk (0) exp (−2γk t)). It
comprises the average electric field energy and particle oscillation energy

1 2 m 2 1 2
Wk = E + n0 v = E
8π 2 8π k
We have substituted E 2 = |Ek |2 /2, |vk |2 = (e/mω )2 |Ek |2 where ω ≃ ω pe . Note that the electric
field energy and particle oscillation energy make equal contributions to the wave energy.
Physically, the wave energy decreases at the expense of the work done by the wave on the
resonant particles. Clearly, most of the plasma particles (i.e. thermal, with |v| . VTe ) are non-
resonant in that they experience only rapidly oscillating force with the frequency |ω − kv|  ∼ω
which does not change their average energy. Only a small group of particles with v −Vph  . wk
are in resonance with the wave, where Vph = ω /k and wk is the velocity with which the resonant
particles oscillate in the wave potential φ :

2eφk /m
p
wk =
Introducing w = v −Vph , for the energy of resonant particles we have

 
ˆ 2
Vph
ˆ
m
f (v) v2 dv = m nres w f Vph + w dw + nres O w2k 



Eres = +Vph

2 2
|v−Vph |<wk |w|<wk
(2.1.10)
Here f is the distribution function averaged over the wave length, and nres is the number density
of resonant particles
ˆwk
nres = f dw
−wk

Now, for dWk /dt we clearly have dWk /dt = −dEres /dt. Assuming that the wave amplitude does
not change during the time τ ∼ 1/kwk (bounce period of resonant particles in the wave potential)
we neglect dwk /dt, dnres /dt ≈ 0 when substituting dWk /dt into (2.1.9). Next, we represent the
second integral in the brackets above as
ˆwk
2 3 ∂ f 

+ −



w f Vph + w dw = wk n − n ≈ wk (2.1.11)
3 ∂ v v=Vph
−wk

25
 Physics 218A

where we have split nres = n+ + n− with

ˆ ±wk
1
n±


k = w f Vph + w dw
wk 0
so that

dEres d  +
n (t) − n− (t)

≈ mVph wk
dt dt
We may approximate Ėres as follows

dEres Eres (τ /2) − Eres (0)

≈ (2.1.12)
dt τ /2
Clearly, after half a period τ /2, resonant particles from w > 0 and w < 0 resonance domains will
swap and expression n+ − n− above will change its sign. Then, using (2.1.10,2.1.11) we obtain

4 3 ∂ f 

1 2
γk = − mVph wk
2Wk τ 3 ∂ v v=Vph
making necessary substitutions and omitting the coefficient ∼ 1 which is meaningless because
of the approximation (2.1.12) we obtain

ω pe
3
1 ∂ f 

γk = − 2
k n0 ∂ v v=Vph
This expression differs only by a factor π /2 from the exact formula for the collisionless (Landau)
damping rate that will be obtained later from the full kinetic treatment.
A few comments are in order here. First
if ∂ f /∂ v were positive, then instead of being
damped the wave would grow because n+ >
n− initially and swapping these groups of par-
ticles decreases their total energy. Second, in
both cases after one half of the bounce period,
particles start moving back to their initial po-
sitions and damping (if ∂ f /∂ v < 0) should
turn to growing (or vice versa if ∂ f /∂ v > 0).
However, this should be the case only if the
wave does not decay strongly over the time
∼ τ /2. Moreover, even if the initial wave
amplitude is large enough so that it does not
drop significantly over that time, the restora-
tion cannot be complete since different parti-
cles have different bounce period due to the
nonlinear motion in the wave. In this case af-
ter a few revolution, the distribution function

26

Figure 2.1.1.: Phase plane of resonant particles

 Physics 218A

in the resonance area (particles trapped by the

wave, Fig.2.1.1, where the initial n− group
of particles occupies a dashed area inside the
separatrix) becomes very irregular and oscil-
latory so that an averaging would be appropri-
ate. It is easy to see that the total energy lost
by the wave will be

3 ∂f

m
∆E = Vph wk

3 ∂ v v=Vph
The wave damping must stop at this point. Clearly, ∆E < W is the requirement for this switch-off
damping.

2.2. Ele tromagneti waves in plasmas. Laser plasmas.

Mode onversion.

First, consider the plasma response to a high frequency external field. Let’s assume that the
plasma is contained in a volume V . Taking into account that the total charge of the plasma is
zero
ˆ
ρ dV = 0
V
we may write

ρ = −∇ · P
where P is the plasma polarization vector and P = 0 outside of V . Consequently, we can write
the plasma dipole moment (electric moment) as follows
ˆ ˆ
rρ dV = PdV
V V
and also
ˆ ˆ
∇ · (E + 4π P)dV = ∇ · DdV
V V
For an isotropic plasma D = ε E

1
P= (ε − 1) E
4π
Let E in x− direction,

E = ex E0x exp (−iω t) (2.2.1)

27
 Physics 218A

Then for the electron displacement we have

d2x e
2
= − E0x exp (−iω t)
dt m
and

e 1 d2x
x= E 0x exp (−iω t) = exp (iπ ) (2.2.2)
mω 2 ω 2 dt 2
The last expression shows that unlike the polarization in conventional solid dielectrics, the
polarization in plasmas lags acceleration (force) by half a period due to their inertial response
to the high-frequency field. The polarization vector is

1 ω p2
Px = −exn0 = − E0x
4π ω 2
Therefore

ω p2
ε = 1−
ω2
For the potential Langmuir waves E k k, ε = 0, so that we recover (2.1.7) in the cold plasma
limit. For the electromagnetic wave E ⊥ k, and ε = N 2 , where N is the refractive index:

k2 c2 ω p2
N2 ≡ = 1 −
ω2 ω2
Therefore, the dispersion relation for electromagnetic waves in plasma reads

ω 2 = ω pe
2
+ k2 c2 (2.2.3)
Two immediate observations can be made here. First, the phase velocity of these waves always
exceeds the speed of light, ω /k > c. Second, an electromagnetic wave with the frequency ω <
ω pe cannot propagate (penetrate) into a plasma. This can be understood by considering such
a wave propagating from a vacuum half-space (x < 0) into a plasma with gradually increasing
density. Since the wave frequency ω = const, and in vacuum ω = kc, there will be a point (say
x = x0 ) where ω = ω pe . Going deeper into the plasma, ω pe > ω and thus k2 < 0 (2.2.3). This
means that the wave amplitude will decay inside the plasma and,qin its core, where ω pe ≈ const,
we can write for the wave field E ∝ exp (−k (x − x0 )), with k = ω pe2 − ω 2 /c.

As ω /k > c, there will be no collisionless wave damping and such a wave in a plasma with
ω pe < ω is likely to be damped by e − i collisions. To include them into our dispersion relation,
we need to modify the electron displacement (include the friction, sec.1.6):

d2x e dx
2
= − E0x exp (−iω t) − νei
dt m dt
Then, the displacement x in eq.(2.2.2) is replaced as follows x → xω / (ω + iνei ) so that

28
 Physics 218A

ω pe
2
ε = 1−
ω (ω + iνie )
Usually, νie ≪ ω and instead of (2.2.3) we have
ν
ω 2 = ω pe
2
+ k2 c2 − iω pe
2
ω0
where ω02 = ω pe
2 + k2 c2 . Thus the previous solution ω = ω becomes
0

νie ω pe
!
2
ω ≈ ω0 1 − i
2ω03
that is the wave indeed decays by collisions.
We calculated the electron displacement in the wave field assuming E0x = const. At the same
time we have learned that E decays spatially where ω pe > ω . Assume that an EM wave is
propagating in z- direction. The electric and magnetic fields are in x, y- directions, and they are
related by the Maxwell equation:

1 ∂H
∇×E = −
c ∂t
Instead of (2.2.1) we write

1
E = ex E0x (z) exp (−iω t) + c.c.
2
1
H = ey H0y (z) exp (−iω t) + c.c.
2
Then, by the Maxwell equation

dE0x iω
= H0y
dz c
Since EM forces on electrons depend on z, apart from their rapid oscillations with the frequency
ω , electrons may drift in z- direction as well. Let’s represent their motion as
1 1
r (t) = r⊥ e−iω t + ez z (t) + c.c.
2 2
The motion in z is driven by the Lorentz force

d2z e dx e dx dEx
m =− Hy = i (2.2.4)
dt 2 c dt ω dt dz
contains both fast and slow components. Taking into account

dx ie
≈− E0x e−iω t + c.c.
dt 2mω
and averaging over the fast oscillations

29
 Physics 218A

d2 e2 ∂ ∂U
m hzi = − |E0x |2 ≡ − (2.2.5)
dt 2 4mω 2 ∂ z ∂z
It is seen that EM wave exerts force on electrons that have a potential U (z). The following
quantity is also called a pondermotive pressure, or high-frequency (wave) pressure

e2 n0 2 ω pe
2
|E0x |2
Pw = |E 0x | = = n0U. (2.2.6)
4mω 2 ω 2 16π
This result has been formally derived for a transverse wave of linear polarization and the longi-
tudinal force in k direction is actually the Lorentz force, eq.(2.2.4). However, the result is also
valid for high frequency fields of various origins, such as the longitudinal (potential) oscilla-
tions. Important is the spatial variation of the amplitude. Indeed, if the electric field is in the
wave-propagation direction, E = ez Ez (z), we can represent the force as

∂ Ez (z̄) −iω t
 
e
F (z,t) = − Ez (z̄) + z̃ e + c.c.
2 ∂ z̄
where the electron coordinate is decomposed into slow (averaged) motion and fast oscillations

z (t) = z̄ + z̃
whereas z̃ ≪ E (∂ E ∂ z)−1 . Separating fast and slow motions in

d2z
m = F (z)
dt 2
for z̃ we have
e
z̃ (t) = Ez (z̄) e−iω t + c.c.
2mω 2
while after averaging over the fast oscillations (effectively discarding e±2iω t terms), for z̄ one
obtains

d 2 z̄ e2 ∂ Ez∗
m = − E z
dt 2 4mω 2 ∂z
which is clearly equivalent to the result for hzi in eqs.(2.2.5,2.2.6) with the replacement E0x (hzi) →
Ez (z̄).
The pondermotive force turns to zero only if the high frequency pressure in (2.2.6) is con-
stant, as e.g. in a wave of constant amplitude. Otherwise this force expels plasma from regions
where the wave amplitude is higher. This phenomenon suggests using powerful lasers to confine
plasmas. Indeed, no EM wave can penetrate into a plasma beyond the point where ω pe = ω .
Therefore, the wave amplitude must decay automatically beyond such point thus building a pon-
dermotive pressure gradient. This gradient will push the plasma inwards at its boundary and
if the radiation illuminate the plasma volume evenly over its surface, the plasma may be con-
fined and even compressed. The techniques may be different, but there an additional bonus of
such confinement approach. Even if some fraction of the heated (superthermal) plasma escapes

30
 Physics 218A

against the wave pressure, by carrying momentum outwards it exerts additional pressure onto
the plasma (ablative pressure).
Up to now we considered only the electron motion. Ion direct response to a high frequency
force is usually weak due to their inertia. However, as often happens in plasmas, electron bulk
motion also set ions in motion via electric field that builds up in response of electron transport
to maintain quasineutrality. To describe this process we can use the two-fluid model of Sec.1.6.
Assuming that electron bulk motion creates a macroscopic electric field E (z), we can write for
them a force balance equation

∂U 1 ∂
−eE − − nTe = 0
∂z n ∂z
and for the ions

1 ∂
eE − nTi = 0
n ∂z
We have used the quasineutrality condition since we are interested in scales ≫ λD . For the
density distribution we thus have
 
U
n = n0 exp −
Te + Ti
while the electric field potential
Ti U
Φ=
e Te + Ti

2.2.1. Penetration of EM waves (laser light) into plasma

Assume that the plasma density is growing in z- direction in which also an EM wave propagates
from a vacuum. The wave has two components of wave vector ky,z , so the formal dispersion
equation is as follows

ω2 ω 2 − ω p2
ky2 + kz2 = ε ≈
c2 c2
which can also be written as

ω2
kz2 = ε − sin2 ϑ .


2
(2.2.7)
c
Here ky2 = k2 sin2 ϑ and k2 = ω 2 /c2 are fixed by the incident wave properties. Once kz2 becomes
negative, the wave cannot propagate beyond this point, called the turning point (or wave cut-off ),
that is the point where ε (z) = sin2 ϑ . However, a simple Fourier-Laplace analysis used to derive
eq.(2.2.7) breaks down near the turning point and a more general analysis of the underlying
differential equations is required. We therefore return to the Maxwell equations, in which we

31
 Physics 218A

will retain the z- dependence (since ε = ε (z)). In other words, the electric and magnetic fields
depend on coordinates and time as

E (z) exp (iky y − iω t)

ω
∇×E = i H
c
(2.2.8)
ω 4π ω
∇ × H = −i E + j = −i ε E
c c c
Here we have used the linear relation j = σ E and ε = 1 + 4π iσ /ω . Eliminating H, we obtain

ω2
∆E + ε E − ∇ (∇ · E) = 0 (2.2.9)
c2
Similarly, we may obtain the following equation for H

ω2 1
∆H + ε H + ∇ε × ∇ × H = 0 (2.2.10)
c2 ε
Consider first a simple case of an S- polarized wave which is characterized by the electric
field directed perpendicularly (S for German ’senkrecht’ - perpendicular) to the wave incidence
plane. More specifically, E = ex E (z), so that (2.2.9) simplifies to

d2E ω 2
− 2 sin2 ϑ − ε E = 0


dz2 c
Introduce a dimensionless coordinate by scaling it to the skin-depth c/ω , i.e. zω /c → z and
rewrite the above equation as follows

d2E
− q (z) E = 0 (2.2.11)
dz
where q = sin2 ϑ − ε . To the right of the turning point z = 0, where q > 0, we must retain only
the decaying solution
 z 
√
ˆ
E ∼ q−1/4 exp − qdz
0

Using the WKB transition formulae, for the region z < 0 (wave transparent medium), we obtain
s  z 
4 sin ϑ − 1 π
2
√
ˆ
E ≈ E0 cos  qdz − 
q (z) 4
0

Two last formulae clearly describe the process of wave reflection from the turning point where
q = 0 : a skin layer solution to the right and a standing wave (superposition of incident and
reflected waves) to the left.

32
 Physics 218A

Consider now P- polarized wave where E is in the plane of wave incidence (P stands for
parallel here), i.e. Ez 6= 0. This component of the electric field will result in charge separation
and thus in generation of longitudinal plasma waves. The latter can propagate into plasma
beyond the critical point where kz = 0, thus carrying away at least some part of the incident
wave energy. For a P-polarized wave H =ex H and for H (z) (dimensionless z) we have

d 2 H ε ′ dH
− − qH = 0
dz2 ε dz
√
where ε ′ ≡ d ε /dz. Transform to the canonical form by H = ε G

ε ε ′′
 ′2
d2G

− − + q G=0 (2.2.12)
dz2 ε 2 2ε
The term ε ′′ may be neglected, as ε ′ /ε ∼ c/ω L ≪ 1, where L is a scale at which the plasma
density changes. But there is a singularity (resonance point) where ε = 0. In order to the field
perturbation
 z 
ˆ p
G ∼ Q−1/4 exp − Qdz (2.2.13)
0

where
ε ′2 ε ′′
Q≡ − +q
ε 2 2ε
retain some power at the resonance point, we require
3/2
q′ zr .1
′
As zr ∼ sin2 ϑ /ε in
sin3 ϑ ≈ ϑ 3 . c/ω L
- almost normally incident waves can penetrate far enough. But, if ϑ → 0 there is no Ez com-
ponent needed to generate charge separation, Fig.2.2.1. So, there is an optimal ϑ to launch the
plasma wave beyond the resonance point most efficiently. To determine the residual power in the
wave that reaches the resonance point ε = 0, it is necessary to generalize the asymptotic
√ solution
(2.2.13) of eq.(2.2.12) for z → 0 where Q has a pole. The overall behavior of H = ε G turns
out to be fairly regular, but the analysis is complicated because of merging the turning point and
the pole in Q. As a function of parameter τ = c/ω L sin3 ϑ , the field amplitude has a maximum
at τ ∼ 1 emphasizing the above restrictions to ϑ , Fig2.2.2:

Φ = sin ϑ [Hx (zr ) /H] (2πω L/c)1/2 (2.2.14)

where H is the amplitude of the incident wave.

To understand wave dissipation mechanisms near the resonance point and a possible further
penetration of the longitudinal plasma wave, let us start with introducing some collisions into
electron dynamics. Effectively, this can be done by replacing ε → ε + iν /ω , as the equation of

33
 Physics 218A

motion of electrons in the wave electric field will contain a friction force −mν ve . Then, from
(2.2.8)

sin ϑ
Ez = Hx (zr ) (2.2.15)
ε + iν /ω
As ε ∼ ϑ 2 − z/L, the collision dominated region around the resonance point is of a size ∆zc ∼
Lν /ω . Turning to the absorption coefficient we write the energy dissipation rate due to collisions

|Ez |2
ˆ
Ẇ = ν dz
8π
or using the preceding equation for Ez we obtain

ν |Hx |2
ˆ
1
Ẇ = sin2 ϑ dz
8π ε 2 + ν 2 /ω 2
Due to the sharp resonance (ν /ω ≪ 1), the main contribution to the integral is from the vicinity
ε = 0, so that we may write
−1 ˆ
|Hx |2 2 dε ν

1
Ẇ ≃ sin ϑ d ε = Lω |Hx |2 sin2 ϑ .
8π dz ε 2 + ν 2 /ω 2 8
Note that the energy dissipation rate does not depend on ν (ν → 0). This situation is similar
to e.g., dissipation of mechanical energy in shock waves. Vanishing viscosity (collisions here)
results in an infinitely sharp gradient at the shock transition so that the dissipation rate does not
depend on viscosity.
The energy flux of the incident wave is cH 2 /4π , so that the absorption coefficient can be
represented as

Ẇ
K=
cH 2 /4π
Expressing then H through Φ, we obtain

1
K ≃ Φ2 (τ )
4
One may see that the maximum absorption is achieved at ϑ ≃ 0.5 (c/ω L)1/3 and Kmax ≃ 0.4.
A competing mechanism of field limitation near the resonance point is due to the generation
of Langmuir waves with their subsequent propagation into the plasma. To describe this process,
we have to take into account the finite group velocity of the waves. Returning to eq.(2.2.15), we
include the dispersive term in ε instead of the collisional term:

ω p2
ε = 1− 1 + 3k2 λD2


ω 2

To simplify calculations, we replace ε in eq.(2.2.15) by the above one also replacing k2 →

−∂ 2 /∂ z2 . The equation for Ez then rewrites

34
 Physics 218A

d 2 Ez
3λD2 + ε0 Ez = Hx (zr ) sin ϑ
dz2
with ε0 = 1 − ω p2 /ω 2 ≃ (zr − z) /L. Since λD is the smallest scale in the problem, we may apply
the WKB approximation or use Airy functions with varying coefficients to solve the inhomo-
geneous equation above. However, simple estimates suffice. The scale on which the field Ez
changes is

1/3
∆z ∼ LλD2
so that the field amplitude can be estimated by comparing the second term on the lhs with the
term on the rhs
Emax ∼ Hx sin ϑ (L/λD )2/3
Now we can compare the two competing mechanisms to see which one of them dominates.
Clearly, if ∆z is larger than the collisional resonance broadening ∆zc , then the second mechanism
dominates. The latter is true if

ν λD
 2/3
<
ω L
The energy flux is

Ez2 1 2 c
Vg ≃ Hx (zr ) sin2 ϑ Lω p ≃ 2 H 2 Φ2 (τ )
4π 4π 8π
As Φ ∼ 1 for τ ∼ 1, it is clear that the efficiency of the mode conversion is of order one in this
case. Indeed, on the lhs is the energy flux of plasma wave while on the rhs is that of the incident
wave times a factor ∼ 1. More accurate consideration shows that the absorption coefficient due
to collisions coincides with the mode conversion efficiency (recall ν - independence of absorption
at small ν ).

35
 Physics 218A

Figure 2.2.1.: EM-wave electric field near the

turning point and the resonance.

2.3. Collisionless (Landau) damping. Landau rule for

singularity in plasma dispersion fun tion

Consider Langmuir waves in homogeneous

plasmas using a kinetic approach. As we
learned already, the linear wave-particle res-
onance is most important for longitudinal
waves, so here we limit our consideration to
this type. Considering plane waves we may
also discuss a one dimensional case. Effec-
tively, a one dimensional distribution function
may be considered as a full distribution inte-
grated over the transverse to the direction of
wave propagation velocity components. The
results can be easily generalized to three di-
mensions (see Sec.2.1).
The kinetic equation with a self-consistent
electric field and no collisions has the form
∂f ∂f e ∂φ ∂ f
+v + =0 (2.3.1)
∂t ∂x m ∂x ∂v Figure 2.2.2.: Here Φ and τ are defined in
eq.(2.2.14).
d2φ
ˆ 
= 4π e f dv − n0 (2.3.2)
dx2

36
 Physics 218A

As usual for fast electron oscillations we consider plasma ions to be at rest whose role is limited
to provide a neutralizing background of the density n0 . The equilibrium state of the system is
φ = 0, f = f0 (v). Consider first the linear theory, i.e. small φ and linearize the equations. We
also Fourier - transform them in x:
ˆ∞ ˆ∞
1 1
φk (t) = √ φ (x,t) exp (−ikx) dx, φ (x,t) = √ φk (t) exp (ikx) dk, 0 < |ℑk| < a
2π 2π
−∞ −∞

and similarly for fk ≪ f0 , k 6= 0, i.e.

ˆ∞
1
f (x, v,t) = f0 (v) + √ fk (v,t) exp (ikx) dk
2π
−∞

∂ fk ie ∂ f0
+ ikv fk + kφk =0 (2.3.3)
∂t m ∂v
ˆ
k2 φk = −4π e fk dv (2.3.4)

As a first attempt, let fk (t) = fk,ω exp (−iω t) and φk (t) = φk,ω exp (−iω t). For fk,ω we obtain
the following “simple” equation

e ∂ f0
(ω − kv) fk,ω = k φk,ω (2.3.5)
m ∂v
with an “obvious” solution
e ∂ f0
fk,ω = k φk,ω (ω − kv)−1 . (2.3.6)
m ∂v
After being plugged into the Poisson equation, it leads to the dispersion relation between ω and
k (that is a condition required to resolve equation (2.3.4) for φk,ω 6= 0):

4π e2 dv ∂ f0
ˆ
ε (ω , k) ≡ 1 + =0 (2.3.7)
mk ω − kv ∂ v

Observe, that in the limit Vph ≡ ω /k ≫ VT , ε ≈ 1 − ω p2 /ω 2 (To obtain this result, we formally
expand the denominator and integrate by parts). The first obvious problem with this equation,
however, is a singular (resonant) point at v = ω /k. The wave-particle resonance phenomena
have been already discussed in the framework of hydrodynamic treatment of Langmuir waves.
Unless ∂ f0 /∂ v = 0 at v = ω /k, one needs a rule for the treatment of the singular integral in the
dispersion equation.
There are three possibilities here. First, one can interpret the improper integral as the Cauchi’s
principal value. This was the original Vlasov’s suggestion. His rationale was a ’micro-plateau’
at the resonance point due to the particle trapping effect, Sec.2.1 that should result in a local
flattening of their distribution function. However, as we have learned already such p a choice
would only be justified for a relatively strong wave, for which the bounce frequency k eφ /m ≫
ℑω . As we consider a linear theory, here it must handle the case φ → 0 correctly in the first place.

37
 Physics 218A

The second possibility is based on the knowledge of the result, namely the wave-particle
interaction near the resonance point makes ℑω 6= 0 (e.g. ℑω < 0 for ∂ f0 /∂ v < 0 at v = ω /k).
This seems to solve the integration problem as the contour in v- plane (ℑv = 0) does not pass
through the singular point and must be above it (for k > 0) and below it (k < 0). As we will
see, this is a wrong approach. The third possible approach is a physical one and it relies on
collisions, introduced, e.g., in a τ - approximation. This leads to the transformation ω → ω + i/τ
(cf., e − i friction, discussed earlier) and leads to a rule which is opposite to the above precisely
for a damped mode, Fig.2.3.1.
To add even more confusion to these con-
tradictory choices we observe that the solu-
tion (2.3.6) to eq.(2.3.5) is not complete. In-
deed, a complete solution is

e ∂ f0
fk,ω = k φk,ω (ω − kv)−1 +Ck δ (ω − kv) Figure 2.3.1.: Deformation of integration con-
m ∂v
(2.3.8) tour in the case of rare collisions,
with some arbitrary constant Ck . The mean- τ → ∞.
ing of the additional term is as follows. Apart
from the perturbations of the electron distri-
bution caused by the wave electric potential φ , there is a contribution of a modulated beam of
resonant particles with v = ω /k propagating at exactly the phase velocity of the wave, thus corre-
sponding to trapped particles. Indeed, in the x,t representation it corresponds to the contribution
of the mode with the wave number k of the following form f ∼ Ck exp [ik (x − vt)]. While the
first term in (2.3.8), after being submitted to the Poisson equation, leads to a discrete spectrum
ω = ω (k) (see 2.1), the second term does not impose any additional solvability condition (like
the dispersion relation) thus comprising the continuous part of the spectrum. This was first estab-
lished by Van-Kampen and the corresponding term in the above solution is called Van-Kampen
mode. Sometimes, it is also called Van-Kampen-Case mode, as the similar continuous spectrum
solution has been found by Case in hydrodynamics. Note that in contrast to the discrete modes
this part of the solution is not an eigen solution of the system given by eqs.(2.3.3,2.3.4), so that
its role in asymptotic behavior at t → ∞ is different in that this contribution decays in time even
if ℑω = 0 for the discrete spectrum. However, only the two solutions (discrete and continu-
ous) combined, form a complete system and an arbitrary initial perturbation can be decomposed
using this system and then its time evolution can be adequately described.
To prove the above statements and to establish an unambiguous rule for handling the singu-
larity in the dispersion equation for the discrete part of the spectrum, it is necessary to return to
the initial value problem posed by eqs.(2.3.3,2.3.4). The most efficient way to analyze it is to
make a Laplace transform:
ˆ∞
fkp = e−pt fk (t, v) dt
0

38
 Physics 218A

The inverse transform is

σˆ+∞
1
fk (t) = e pt fkp d p
2π i
σ −i∞

It is clear already from these relations that, as fkp must be analytic for sufficiently large ℜp ≥ p0 ,
any possible singularity of fkp in the complex p -plane may only be in the half-plane ℜp < p0
so that the integration contour in the inverse transform should be to the right of them, σ > p0 .
Recalling that ω = ip and applying this rule to the dispersion relation (2.3.7) the v− integration
contour should loop below the pole1 v = ω /k in the dispersion relation regardless of the value
of ℑω . This correspond to an analytic continuation of the singular integral from the upper ω -
plane (right p- plane) where it is a priori analytic. This procedure is known as a Landau rule,
that indeed corresponds to the collisional correction for ℑω → 0+.
The solution of the initial value problem (2.3.3,2.3.4) can be obtained as follows. After ap-
plying the Laplace transform, from eq.(2.3.3) we have

e ∂ f0
 
1
fkp = ig (v) + k φkp
ip − kv m ∂v

where gk = fk (t = 0, v). From eq.(2.3.4) we obtain

4π e gk (v) dv
ˆ
φkp = −i 2
k ε (k, ip) ip − kv

We may now return to the ω = ip- variable and write the solution e.g., for φk (t)

ˆ σ
+∞+i
2e exp (−iω t)
φk (t) = − 2 G (ω , k) d ω (2.3.9)
k ε (ω , k)
−∞+iσ

where
ˆ∞
gk (v) dv
G (ω , k) = (2.3.10)
ω − kv
−∞

and ε is given by (2.3.7). The standard strategy of performing the integration in ω in the above
expression for φk is to distort the contour to ℑω → −∞ as much as possible, thus making the
respective contribution negligible due to the exp (−iω t) factor. Let’s assume for a moment that
G (ω ) is holomorphic and ε has only isolated zeroes. Then, the part of the new contour with
ℑω = −∞ will not contribute to the integral, and only the zeroes of ε will contribute through
the respective residues. Assuming that there are only simple zeros corresponding to the eigen-
(n)
frequencies ω = ωk , we have:

4π e ∂ε
  
φk ≈ i 2 ∑ exp (−iω t) G (ω , k)

(2.3.11)
k n ∂ω ω =ω
(n)
k

1k > 0 here and below, for concreteness. The opposite case is considered similarly.

39
 Physics 218A

Clearly, time asymptotic behavior is determined by the eigen-frequency with the largest positive
imaginary part. Let us turn to the contribution from G, if it is nonholomorphic. Eq.(2.3.10) is a
singular Cauchi’s integral that has a jump
2π i  ω 
gk
k k
across the the integration path, i.e. ℑω = 0. This is easily seen from the Plemelj formulae
1 1
=P ∓ iπδ (ω − kv)
ω − kv ± i0 ω − kv
According to the Landau rule, the upper sign should be taken when the integrals in (2.3.7,2.3.10)
run along the real v- axis. However, as G is generally nonholomorphic on the integration path,
the contour on ω - plane in (2.3.9) can not be shifted to −i∞ without including the contribution
from the jump of G. The easiest way to understand what is going on is to consider a function
g (v) vanishing outside of an interval, say v1 < v < v2 . The function G (ω ) is holomorphic in
the entire ω - plane cut along the interval kv1 < ω < κ v2 across which it has a jump specified
above. Then, to the contribution from the discrete spectrum (2.3.11) the following contribution
from the cut will be added

ˆkv2
4π e exp (−iω t)
φkc (t) =i 3 gk (ω /k) d ω
k ε (ω , k)
kv1

Here φkc
stands for continuous spectrum contribution (Van-Kampen mode). It is different from
the point spectrum in that while the latter may oscillate in time indefinitely, may grow (if some
ω (n) have positive) or decay exponentially, the continuum part will always decay, though slowly.
Indeed the last integral can be evaluated for ω → ∞, using Watson’s lemma as follows
ω =kv2
1 exp (−iω t)
φkc gk (ω /k)

∼
t ε (ω , k) ω =kv1

It has a clear signature of mixing of the modulated particle beams originating form the initial
perturbation of distribution function that decay in time as 1/t.
Let us consider the solutions of (2.3.7) for the case of ℑω ≪ ℜω . Using the Landau rule with
Plemelj formula we have

ε (ω , k) = ℜε + iℑε (2.3.12)
with
4π e2 ∂ f0 /∂ v
ˆ
ℜε = 1+ P dv
mk ω − kv
4π e2 ∂ f0
ˆ
ℑε = − πδ (ω − kv) dv (2.3.13)
mk ∂v

40
 Physics 218A

If the first equation has a solution ω = ωk ≡ ℜω (k), the second one gives the damping (growth)
rate γk ≡ ℑω (k): 
∂ ℜε −1
  
γk = − ℑε (ω , k) (2.3.14)

∂ω
ω =ωk


The last relation has an important energetic interpretation. Introducing the wave amplitude
|Ek (t)| it can be represented as follows

∂ ∂ ℜε |Ek |2 ωk
 
ωk = − ℑε (ωk , k) |Ek |2 = −ℜσ |Ek |2 = h jEi (2.3.15)
∂t ∂ω ω =ωk 8π 4π

where σ = i (1 − ε )/4π is the electric conductivity. On the r.h.s is the energy dissipation rate.
Therefore the wave energy is

∂ ℜε |Ek |2
 
Wk = ωk (2.3.16)
∂ω ω =ωk 8π

In the simplest case ε = 1 − ω p2 /ω 2 , Wk = |Ek |2 /4π which corresponds to the equipartition

of the total field energy between the electric field energy (|Ek |2 /8π ) and nonresonant parti-
cle oscillations. In more interesting cases, where for example beams are present, i.e. ε =
1 − ω p2 /ω 2 − ωb2 / (ω − kVb )2 , the wave energy does not have to be positive (here ωb2 = ω p2 nb /n0
and Vb is the beam velocity). In fact, the wave energy is not an invariant of the Galilean trans-
formation x → x + V t, that is ω → ω − kV . However, γk is an invariant of such transformation
as it should be.
If a cold plasma is moving with the velocity V , then ε = 1 − ω p2 / (ω − kV )2 and from the
two roots ωk = kV ± ω p , the second one corresponds to the negative wave energy Wk < 0 for
sufficiently short waves kV > ω p . According to eq.(2.3.15) such waves will grow under a normal
resistivity ℜσ > 0. The waves with positive energy require a negative active conductivity ℜσ <
0 to grow.
Consider an eigen-mode, that is the solutions of
eq.(2.3.7) for an important case of Maxwellian
distribution f0 (v) = (m/2π T )1/2 exp −mv2 /2T . Substituting this into eq.(2.3.7) we obtain

1  √
ε = 1+ 1 + i π zW (z)

(2.3.17)
k2 λD2
√
where z = ω / 2 |k|VT and

ˆx
 
2i
W (z) = exp −z2 1 + √ exp t 2 dt 



π
0

This function can be approximated as follows

(
1 + √2iπ z + ..., |z| ≪ 1
W (z) = i 1

−z2
√
πz
1 + 2x2 + ... + e , |z| ≫ 1

41
 Physics 218A

Obviously, for |z| . 1, the Langmuir wave is strongly damped, while in the opposite case ω ≫
kVT we recover our hydrodynamic result for the real part of the frequency (2.1.8) while the
damping rate can be written down as follows

1 π ω
r  
3 1
γk = − exp − − 2 2
2 2 (kλD )3 2 2k λD

2.3.1. Ion Langmuir and ion-a ousti waves

Let us turn to the low-frequency part of the plasma spectrum where the ion contribution to the
dielectric function ε becomes important. Note, that the electron contribution to ε in (2.3.12) was
calculated without any assumption about their thermal velocity, so that the ion contribution can
be written in precisely the same form. It was neglected for the reason that ω ≫ ω pi . If this is not
the case we simply add to ε the ion contribution which is identical to that of the electrons with
the following obvious replacements m → M and Te → Ti .
It is instructive to consider a specific case of Maxwellian plasma. Thus from the above results
we have

1 √
ε (ω , k) = 1 + ∑ 1 + i π ze,iW (ze,i ) = 0
 
e,i k λDe,i
2 2

√
where ze,i = ω / 2 |k|VTe,i . To avoid a strong electron wave damping of Langmuir waves one
requires ze ≫ 1. Moving to lower frequencies will apparently violate this condition, so that to
avoid strong electron damping we require ze ≪ 1. However, the same problem may arise with
ions, so we look at the range zi ≫ 1 ≫ ze , in other words, the phase velocity of the waves should
be between electron and ion thermal velocities. Then, the dispersion equation rewrites
√ ω √
 
1 1 
ε = 1+ 2 2 1+i π √ + 2 2 1 + i π ziW (zi )

k λDe 2 |k|VTe k λDi
or
ω pi
2 
k2V 2

1
ℜε ≈ 1 + 2 2 − 2 1 + 3 2Ti
k λDe ω ω
so that for the eigen-frequency we have

ω pi
2
ωk2 = 1 + 3k2 λDi
2
1 + 1/k2 λDe
2


1 + 1/k2 λDe
2

From here we obtain that zi ≫ 1 (weak damping on the ions) requires

Ti /Te + k2 λDi
2
≪1

which actually imposes two constraints. First, the plasma must be nonisothermic, Te ≫ Ti and,
second, the waves should not be too short. In the longwave limit, k2 λDe
2 ≪ 1, we have acoustic

type waves, called ion-acoustic

ωk = kCs 1 + 3Ti /Te

p
(2.3.18)

42
 Physics 218A

where the ion-sound speed is Cs = Te /M. In the opposite limit k2 λDe

p 2 ≫ 1, this branch tends to

the ion plasma frequency, sometimes called ion Langmuir wave for its similarity with the usual
electron Langmuir wave (2.1.8)

ω 2 = ω pi
2
+ 3k2VTi2

It is important to remember, that in contrast to the electron plasma waves where there is always
a long-wave regime in which the collisionless damping can be made negligibly weak, the above
ion-branch is generally damped quite substantially. So, the ion-acoustic waves has the damping
rate "r #
π
r  3/2
m Te
γk = − |k|Cs + e−(Te /Ti +3)/2
8 M Ti
while the ion Langmuir wave is damped at the rate
"r #
π
r  3/2
1 m Te
e−(1/k λDi +3)/2
2 2
γk = − ω pi 3 3 +
8 |k| λDe M Ti
One may give a hydrodynamic interpretation of the ion-acoustic mode with some ramifica-
tions, left out by the simple kinetic description above. As in any other continuous medium,
acoustic perturbations in plasmas are supported by the pressure reaction force. If one consid-
ers plasma as a combined electron-ion pressure is simply P = Pe + Pi . The adiabatic
p fluid, the p
acoustic wave speed is simply Cs = ∂ P/∂ ρ = γ (Te + Ti ) /M, where γ = 5/3, which is not
quite the same as (2.3.18). The adiabaticity condition, however, can only be justified if the ther-
moconductivity does not spread out the pressure related temperature perturbations. This is more
difficult to do for electrons, which can be reasonably expected to have an isothermic equation
of state, Pe = nT with T = const, i.e. γ = 1. Ions can still be considered adiabatic but, as the
wave is one-dimensional, γ = 3=(d + 2) /d. Under these circumstances, one recovers the result
(2.3.18) immediately.

2.4. Plasma surfa e waves

Up to now we have been dealing primar-

ily with wave phenomena in infinite homoge-
neous plasmas. The only exception was the
interaction of electromagnetic waves with a
’diffuse’ plasma boundary, but the inhomo-
geneity scale was assumed to be larger than
the wave length. Plasma modes involved were
therefore the same as in the homogeneous
case. On the other hand, the plasma inhomo-
geneity itself may support new modes which
are localized to it. A limiting case of this

Figure 2.3.2.: Three modes in plasma without

magnetic field

43
 Physics 218A

kind is a sharp plasma boundary where sur-

face waves may propagate along it, similarly
to, e.g., the well known gravity waves on a
free water surface.
Consider a semi-infinite plasma of a density n0 = const in x ≥ 0 half-space, whereas in the
vacuum half-space, that is in x < 0, n0 = 0. A surface plasma wave is analogous to the con-
ventional Langmuir wave, except it is localized to the plasma surface and has, for that reason,
somewhat different dispersive properties. The surface wave approximation that we will use re-
quires the wave to be much longer than the thickness of the plasma-vacuum interface. Expecting
fast electron oscillations, we assume that ions are at rest, as in the Langmuir waves, but, in addi-
tion they hold the plasma surface flat. Otherwise, the surface rippling would need to be included
as well. Let us look for a wave solution in the form of an electron density perturbation

n − n0 (x) = nk (x) eiky−iω t

and similarly for the electric field potential φ , and for the two components of the electron velocity
perturbations, vx and vy . The third coordinate (z) is ignorable, i.e. the wave propagates along y,
and we will look for the solution which vanishes at both x- infinities nk → 0, |x| → ∞.
The continuity equation can be written as follows
∂
− iω nk + ikn0 vyk + n0 vxk = 0, (2.4.1)
∂x
while from the Euler equation we have

e ∂ φk
−iω vxk =
m ∂x
e
−iω vyk = i kφk
m
Finally, Poisson equation closes the system

∂ 2 φk
− k2 φk = 4π enk
∂ x2
Using the above equations, we can write the equation for the field potential as follows
∂ ∂ φk
ε (x) − ε k2 φk = 0 (2.4.2)
∂x ∂x
where
ω p2 (x)
(
1, x<0
ε = 1− =
ω2 1 − ω p0 /ω , x ≥ 0
2 2

and where we have denoted ω p0 2 = 4π e2 n . Integrating (2.4.2) across the plasma-vacuum inter-
0
face, and requiring the continuity of φk , we obtain the following boundary conditions at x = 0

∂ φk 0+

ε = φk |0+
0− = 0, (2.4.3)
∂ x 0−

44
 Physics 218A

along with the previously stated

φk → 0, |x| → ∞.
Using this and the second boundary condition (2.4.3), from (2.4.2) we obtain

φk = Ck exp (− |kx|)

However, substituting this solution into the first BC (2.4.3) we obtain

ω p0
2
1− = −1
ω2
or
1 2
ω 2 = ω p0
2
The wave frequency is thus identical to the Langmuir wave in a plasma with the density ’aver-
aged’ over the region of wave localization.
This simple dispersion relation is valid for relatively short waves kc/ω p ≫ 1, while in the
opposite limit the field perturbations becomes non-potential and the the wave frequency tends to
the usual EM wave dispersion ω ≈ kc.

2.5. Nonlinear waves, solitons and BGK- waves

2.5.1. Langmuir waves

The simplicity of the plasma wave description discussed so far is largely due to the linear char-
acter of the waves. What if the wave amplitude is high enough to nonlinear terms become
important? Generally, the treatment is complicated but one may start with the following one
dimensional simple version of eqs.(2.1.1-2.1.3)

∂n ∂
+ nv = 0 (2.5.1)
∂t ∂x
dv e ∂
= φ (2.5.2)
dt m ∂x

∂ 2φ
= 4π e (n − n0 ) (2.5.3)
∂ x2
Already the previously obtained linear wave solution suggests that by translating to the wave
frame x → x +Vph t the wave profile becomes time independent (exp (−iω t + ikx) → exp (ikx)).
One may hope that the steady state solution for the wave profile can be continued for higher
amplitude.
We explore such possibility for the Langmuir waves below. Above Galilean transformation
and the steady state requirement is equivalent that in the laboratory frame, where the plasma is
at rest, all the wave variables depend only on ξ = x −Vph t (this Ansatz is called traveling wave

45
 Physics 218A

solution). Note, that Vph may now be related to the wave amplitude. The first two equations
rewrite
dn d
−Vph + nv = 0
dξ dξ
dv dv e dφ
−Vph +v =
dξ dξ m dξ
which we integrate to obtain  √ 
v = Vph 1 − 1 + Ψ

and
n0
n= √
1+Ψ
where we have introduced a dimensionless wave potential Ψ = 2eφ /mVph 2 and have chosen the

integration constant to satisfy the conditions n → n0 and v → 0 for φ → 0. Substituting the last
result into the Poisson equation we obtain

d2Ψ ω pe
2  
1
=2 2 √ −1 (2.5.4)
dξ 2 Vph 1+Ψ
The first integral of this equation is
 2
1 dΨ
+U (Ψ) = E = const (2.5.5)
2 dξ

where E can be interpreted as the constant energy of a nonlinear oscillator, if Ψ is its coordinate
and ξ is time. The potential energy

ω pe
2 h √ i
U (Ψ) = 2 2
Ψ + 2 − 2 1 + Ψ (2.5.6)
Vph
The complete solution of (2.5.4) can be obtained in a closed form by integrating (2.5.5). On
writing
ˆ
1 dΨ
ξ (Ψ) = √ ,
E −U (ξ )
p
2
we obtain for ξ (Ψ):
" √ #
Ψ+1−1
 r √
Vph 2
ξ (Ψ) = sin −1 2
− S − Ψ+1−1 (2.5.7)
ωp S

where we used the following amplitude parameter instead of E:

r
Vph E
S=
ωp 2

46
 Physics 218A

Qualitatively, the character of the solution is better understood from the equations (2.5.5-2.5.6).
Namely, Ψ is oscillating between Ψmin and Ψmax determined by

U (Ψmin ) = U (Ψmax ) = E (2.5.8)

whereas Ψmin has a lower limit Ψmin = −1. It is clear that for small E the oscillator is close
to linear (one may expand [2.5.6] for small Ψ ≪ 1) while for growing E the left turning point
Ψmin → −1 and Ψmax → 3, making the oscillation strongly nonlinear (asymmetric).
It is not difficult to establish the dispersive properties of nonlinear waves for a general form of
potential U (ξ ). As in the case of linear waves these properties result in a certain relation between
the wave number and its frequency. In contrast to the linear treatment, however, this relation
usually also contains the wave amplitude. The wave amplitude is labeled by the integration
constant E in (2.5.5,2.5.8). To find the nonlinear dispersion relation we write the wave period as

2π
˛ ˛ ˛
dΨ 1 dΨ
= dξ = =√ (2.5.9)
dΨ/d ξ
p
k 2 E −U (Ψ)

where the integral should be taken along the closed “particle trajectory”, that is with the change
of sing at the square root. Therefore, we can write
Ψ
ˆmax
2π √ dΨ
== 2 p
k E −U (Ψ)
Ψmin

For the particular case of Langmuir waves we get from the last result
2π Vph
= 2π ,
k ωp

regardless of the wave amplitude, which may be seen from the inspection of eq.(2.5.7). This
is just the linear Langmuir wave dispersion Vph ≡ ω /k = ω p /k, as it is appropriate for the cold
plasma approximation. Therefore, the wave nonlinearity affects only the form of the wave but
not its dispersive properties. Generally, this is not so, as we will see in the next section.
The Langmuir waves, however, are special in that they can be described by linear equations
at arbitrary high amplitudes, of course, using appropriate variables. To demonstrate this, let us
transform eqs.(2.5.1-2.5.3) to a Lagrangian ’mass’ coordinate x,t → ξ ,t ′ according to
n
dξ = (dx − vdt)
n0
t′ = t

so that ∂ /∂ t + v∂ /∂ x → ∂ /∂ t and ∂ /∂ x = (n/n0 ) ∂ /∂ ξ . The system (2.5.1-2.5.3) can indeed

be easily shown to be evaluated to the following simple equation for N (ξ ,t) = (n − n0 ) /n

∂ 2N
+ ω p2 N = 0
∂ t2

47
 Physics 218A

where ω p2 = 4π e2 n0 /m. The general solution of the last equation is

N = a (ξ )cos [ω p t + α (ξ )]

where the arbitrary functions a and α may be easily obtained from the initial conditions N0 (ξ )
and v0 (ξ ), where index 0 refers to t = 0:

N0 (ξ ) = a (ξ ) cos [α (ξ )]
∂ v0
= ω p a (ξ )sin [α (ξ )]
∂ξ
Here we have used the relation ∂ N/∂ t = −∂ v/∂ ξ that follows from (2.5.1).

2.5.2. Ion a ousti nonlinear waves and solitons

Here we look for the traveling wave solution based on the ion-acoustic mode. It is convenient to
work in the wave reference frame, where all the characteristics of the wave, which are electric
field potential φ , electron and ion densities and ion velocity are time independent. Since the
ion-acoustic motions are slow compared to the electron velocity we assume that electrons obey
the Boltzmann distribution. We will also use the following normalized variables
eφ x v
→ φ, → x, →v
Te λD Cs
where
4π e2 n0 Te
λD2 = , Cs2 =
Te M
The Poisson equation then reads

∂ 2φ Vph
= exp (φ ) −
∂x 2 v
Here we have also used the conservation of the ion flux nv = n0Vph with n0 and Vph being the
unperturbed flow density and velocity. For example, for a localized perturbation, these quantities
refer to infinity. The flow velocity is obtained from the equation

∂ 1 2
 
v + φ = −ν (v −V∗ )
∂x 2
where the r.h.s. is due to the small friction force between the the ions (and some other, un-
specified group of particles, e.g. electrons, reflected ions). First, we neglect the friction and
obtain q
v = Vph 1 − 2φ /Vph 2

so that the Poisson equation can be easily integrated once to give

1 ∂φ 2
 
+U (φ ) = E (2.5.10)
2 ∂x

48
 Physics 218A

where q
U = − exp (φ ) −Vph
2
1 − 2φ /Vph
2 + 1 +V 2
ph

The form of the potential is shown in Fig.2.5.1. Depending on the value of the integration
constant E , the wave profile is changing from a linear wave, when E is close to the minimum
of U to a highly nonlinear wave profile when E → 0−. In this case the wave period (2.5.9)
diverges logarithmically k−1 ∝ ln E −1 , but the form of individual pulses in the wave train does
not change since varying E does not affect the profile of U. The pulses thus recede to ±∞ when
E → 0−, so only one remains. This single pulse is called soliton, or solitary wave. Based on the
above treatment, it can be understood mathematically as a limiting case of a cnoidal wave with
the infinite period.
The flow potential can be expanded for small φ as follows
1  1  1  
U = − 1 −Vph −2
φ2 + 3 −Vph−4
φ3 + −6
15 −Vph φ4
2 6 24
which shows that the potential well becomes very narrow when Vph → 1+. This is the linear
regime corresponding to ω = kCs dispersion according to the normalization of Vph = ω /kCs ≡
M , where M is the Mach number of the wave. Increasing M allows the nonlinear waves
including the soliton to appear in the continuum of solutions labeled by E . The limitation on Vph
or M from above is due to the requirement U (φmax , M ) > 0 which yields

exp M 2 /2 < 1 + M 2

or M < M∗ ≈ 1.6.
Clearly, the symmetry of the soliton with respect to the point of maximum potential is ensured
by the reversibility of the ion flow, that is the incoming ions decelerate by the soliton potential
and then accelerate to their initial speed. This symmetry can be broken, e.g., by reflection
of some fraction of incoming ions near the maximum due to their finite velocity dispersion.
Then, the ion density distribution and thus the soliton profile become asymmetric. In terms of
the oscillator analogy given by eq.(2.5.10), the oscillator trajectory still starts at φ = 0 (x =
∞) but after having reached the turning point φ = φmax does not return to φ = 0 as its energy
decrease monotonically until the oscillator energy approaches the bottom of the potential well.
The overall potential profile takes the form a shock wave with a trailing damped wave. The fact
that the wave train is behind the shock transition is because the ion-acoustic waves dispersion is
negative (∂ Vg /∂ k = ∂ 2 ω /∂ k2 < 0). For modes of positive dispersion, the wave train is ahead of
the shock transition.

2.5.3. BGK waves

The traveling wave Anzatz, that is the assumption φ (x,t) = φ (x −Vph t) used above can be gen-
eralized for a kinetic description of ions and electrons. This approach was used by Bernstein,
Green and Kruskal (1957), and the respective traveling wave solutions are called the BGK waves.
As the distribution functions are assumed to be stationary in the wave frame (x −Vph t → x), they
must depend only on the particle energy (see eq.2.3.1)
me,i v2
Ee,i = ∓ eφ
2

49
 Physics 218A

Figure 2.5.1.: Ion-acoustic wave ’virtual’ potential U (called Sagdeev potential) and wave pro-
files depending on the wave parameter E .

50
 Physics 218A

The Poisson equation can thus be written as

 
ˆ∞ ˆ∞
d φ
2 f (E) dE fi (E)dE 
= 4π e  p e − p

dx 2
2m (E + eφ ) 2M (E − eφ )

−eφ eφ

and integrated once to give

dφ
 2
1
+U (φ ) = E = const
2 dx

where  
ˆ∞ ˆ∞
U = 8π  2 (E + eφ ) /mdE + 2 (E − eφ ) /MdE 
p p
fe (E) fi (E)
 

−eφ eφ

One can consider this expression as an integral equation for fe,i that can be easily solved by
the Laplace transform for either sort of trapped particles. But, there remains a great deal of
arbitrariness in this solution as to allow an arbitrary form of the potential trough into which
particles are trapped.

51
3. Plasma Instabilities and wave-parti le
intera tions

3.1. Beam-plasma instability

The simplest version of the beam plasma instability can be readily described by the following
dispersion relation
ω p2 ω pb
2
ε = 1− 2 − =0 (3.1.1)
ω (ω − kVb )2
Here, one can easily recognize the contributions of two electron components, one of which is
the usual background of the density n0 , and the second one (the beam) has density nb , so that
ω pb
2 = 4π e2 n /m ≡ αω 2 and it is moving at the speed V . By writing ε = 1 − F (ω , k) , one can
b p b
determine the parameter range where the instability occurs.
The minimum of F (ω ) is at ωmin = kVb 1 + α 1/3 and if

ω p2  3
Fmin ≡ 1 + α 1/3
> 1,
k2Vb2
then eq.(3.1.1) has only two real roots. Theretofore the two remaining must be complex and,
since the coefficients of eq.(3.1.1) are real, they have to be complex conjugate, and thus one of
them should be unstable, ℑω = γ > 0. The last inequality determines the instability boundary

ωp  3/2
k < kcr = 1 + α 1/3
Vb
The growth rate can be obtained (apart from explicitly solving the quartic equation [3.1.1]) in a
simple form either near the threshold, k ≈ kcr , or in the case of a tenuous beam, α ≪ 1. Near the
threshold, we get
ω = ωmin ± i 2 (Fmin − 1) /F ′′ (ωmin ),
p

while for the second case, α ≪ 1

iω pb
ω = kVb ± q
ω p2 /k2Vb2 − 1

This approximation breaks down when k → ω p /Vb , where γ , in reality, approaches its maximum
so that for ℜω = ω p = kVb , direct from (3.1.1) we have
h  √  i
ω = ω pe 1 − 2−4/3 1 ± i 3 α 1/3 .

52
 Physics 218A

Therefore, for this instability we have

 1/3
nb
γmax ≃ ω p α 1/3
≡ ωp
n0
The validity of the monoenergetic beam approximation is limited by the conditionγ ≫ k∆Vb ,
where ∆Vb is the width of the beam, that is
∆Vb
 1/3
nb
≪
Vb n0
For broader beams, one needs to use the general kinetic formula (2.3.14):
π ω p2 ω ∂ f 

γ=
2n0 k2 ∂ v v=ω /k
Note that, even when the beam satisfies the above condition initially, upon the instability set on,
it becomes broader which may justify a switch to kinetic description.
Consider the mechanism of the beam instability. It starts from week charge fluctuations and
waves providing a seed beam modulation. Then electrons bunch in the decelerating phase of
the wave which amplifies it. Therefore, one can think of the beam instability as of the beam
self-modulation.

3.2. Parti le trapping and saturation of instability

Let us assume that the beam is monoenergetic initially, Vb = V = V0 . If the wave is growing
slowly, one can write m (V − ω /k)2 − eφ = const, so that δ V ≈ (eφ /m) (V0 − ω /k). In course
of time, while φ grows, particles with V > V0 overtake the wave and get bunched in decelerating
phase of the wave, which thus requires that initially V0 > ω /k. As a result, the beam is broken
down into clusters near the decelerating phases of the wave which further increases the wave
amplitude due to the increased beam modulation.
When the wave amplitude is high enough to trap the beam clusters into potential troughs, the
wave ceases to grow. Indeed, as it follows from the analysis of Landau damping in Sec.2.1,
when bunched particles bounce off the decelerating phase of the wave, more particles will move
slower than the wave and the latter will even decay for a while until the situation is reversed
again. The particle phase mixing will suppress the bunching effect and the wave amplitude
oscillations decay.
The condition for the beam particles to be trapped is that in the time when they cross the
trapping area (wave length) the wave amplitude should grow significantly
 ω 
k V −  ∼ γ ,

k
or
eφ /m ∼ γ 2 /k2 .
The saturation wave energy thus amounts to
 1/3
E2 nb
∼ mVb2 nb .
4π n0

53
 Physics 218A

3.3. Quasilinear theory. Resonant and nonresonant

wave-parti le intera tions. Energy and momentum
onservation laws

Quasilinear theory (QLT) applies when the spectrum of modes is broad and dense. The modes
may result from an unstable particle distribution, e.g. beam-plasma instability. Another (strong)
assumption is made about the randomness of the phases of the modes. These assumptions are
characteristic of what is known as a weak turbulence theory. In the simplest QLT the amplitudes
are also assumed to be relatively low to neglect their nonlinear interaction, but well in excess of
the thermal plasma fluctuations.
The main subject of the QL theory is the backreaction of the excited modes onto the particle
distribution function. However, it is treated self-consistently with the mode evolution. The
fundamental physical process here is the emission and absorption of waves by resonant particles
at (Landau, or Cerenkov resonance)
ω = kv
For simplicity, we consider the 1-D version of the QLT, but in many cases it can be readily gen-
eralized to 3D by replacing kv → k · v etc. To simplify notations, we include only the electrons.
For an isolated, quasi-monochromatic mode the backreaction on the particle distribution is a
trapping effect considered in the previous subsection and in Sec.2.1. Clearly, when many modes
are present and they overlap, the particle dynamics becomes different. The overlapping means
that the difference between the phase velocities of the neighbor modes
r
ω  eφ̄k
δ < (3.3.1)
k m
where φ̄k is the characteristic wave potential, associated with the scale k−1 . In terms of the
energy spectral densityqEk2 (which is the main variable of the QLT, along with f ), it can be
introduced as φ̄k = k−1 Ek2 ∆k, where ∆k is the mode spacing in the wave number space. Such
spacing always exists in a system of length L, ∆k = 2π /L.
The particle motion may become chaotic (in the sense of dynamical chaos), if only two modes
are present, naturally more so if modes are numerous and the resonance regions are overlapped.
Thus particles undergo Brownian motion in velocity space, a random walk from one resonance
region to the next one. Ultimately, particles become evenly distributed over the interval
ω  ω  ω 
∆ = −
k k max k min
provided that all the waves from the interval maintain their amplitudes to meet the condition
(3.3.1) in course of the evolution. On the other hand, the waves in the packet should not be too
strong to create their own potential well to make the particle dynamics quasi-coherent, akin to
trapping phenomenon considered earlier. This limitation means
s
 ω  r eφ e
ˆ 1/2
0
∆ ≫ = 2
Ek dk/k 2
k m m

54
 Physics 218A

Otherwise, most of the particles from this velocity interval will be involved in a collective quasi-
coherent motion in the effective potential φ0 .
The formal mathematical description of the particle diffusion in velocity space is as follows.
First we decompose the particle distribution into an averaged, slowly varying part f0 (v,t) and a
small perturbation f ′ (v, x,t), where the averaging is over the random phases of the modes that
are responsible for the perturbation f ′ . Equivalently, this averaging can be performed over rapid
oscillations of these modes either in time or in space, or both, whereas f0 changes only slowly in
time and, possibly, in space (we ignore the latter for simplicity). Substituting the decomposition
f = f0 + f ′ = h f i + f ′ into the kinetic equation (2.3.1) and averaging it, we obtain

∂ f0 ∂ f′
 
e
= E (3.3.2)
∂t m ∂v
∂ f′ ∂ f′ e ∂ f0
+v = E (3.3.3)
∂t ∂x m ∂x
These equations clearly show the ’quasilinear’ character of approximation: the equations for f ′
and for E (that will be given below) are linear. However, nonlinearity is crucial in equation for
f0 describing the backreaction of the waves on the particles. The last equation can be readily
solved for f ′ after a Fourier-Laplace transformation (see Sec.2.3), that is after representing
1
f = √ ∑ fk exp (−iωk t + ikx) + c.c.
2 k

1
E = √ ∑ Ek exp (−iωk t + ikx) + c.c.
2 k
Here ωk = ℜωk + iγk ≈ ℜωk is the solution of the appropriate dispersion equation, e.g. (2.3.7),
and we omit the sign ℜ where it does not cause confusion. From (3.3.3) we thus have

e ∂ f0
fk = i Ek (ωk − kv)−1
m ∂v
When substituting this into (3.3.2), we explicitly use the random phase approximation

hEk Ek∗′ i = |Ek |2 δ k − k′ .

Thus, eq.(3.3.2) rewrites

∂ f0 ∂ ∂ f0
= D (v) (3.3.4)
∂t ∂v ∂v
where
e2
m2 ∑
D= τk |Ek |2 (3.3.5)
k

where
γk
τk ≡ , γk ≥ 0.
(ωk − kv)2 + γk2

55
 Physics 218A

Eq.(3.3.4) is a diffusion equation in velocity space. However, the diffusion coefficient D is

somewhat unusual. To understand this, we need to make a closed system out of this equation by
adding the equation for D, that is for the wave spectral density |Ek |2 , which is simply
∂
|Ek |2 = 2γk |Ek |2 (3.3.6)
∂t
where (see 2.3.14)
ω p2
−1
∂ ℜε ∂ f0
 ˆ
γk = πδ (ωk − kv) dv (3.3.7)
∂ ωk n0 k ∂v
The growth rate of the waves is determined by the current distribution f0 (v,t) which itself
evolves in time under the wave backreaction on particles. The diffusive contributions τk have
one distinctive limit
τk → πδ (ω − kv) , γk → 0+ (3.3.8)
Physically, this means that particles are kicked by the resonant waves when they are near the
points v = ω /k.
A physically different situation occurs in the region ωk ≫ kv, e.g. bulk electrons and the
spectrum of Langmuir waves with Vph ≫ VTe . Then, we can write τk ≈ γk /ω p2 . In this part of the
velocity space we thus have
!
∂ f0 ∂ 1 ∂ |Ek |2 ∂ f0
∂ v 2m ∂ t ∑
=
∂t k 4π n0 ∂v

This equation can be easily solved by introducing a new variable instead of t

|Ek |2
T̃ = ∑ + Te (3.3.9)
k 4π n0

where Te is the electron temperature, so that the last equation has the solution
mv2
r  
m
f0 (v,t) = exp −
2π T̃ 2T̃
The choice of the arbitrary constant in effective temperature T̃ implies that electrons have a
Maxwellian distribution with the temperature Te when waves are absent. When they are ex-
cited, the only effect is a reversible increase of the temperature. In fact it is caused by elec-
tron nonresonant oscillations in the plasma wave field. Indeed, from the equation of mo-
tion of electrons in nonresonant plasma waves with ω ≈ ω p , v̇ = − (e/m) E, we may see that
mv2 = ∑k m |vk |2 = ∑k |Ek |2 /4π n0 . Therefore, the heating effect disappears together with the
waves as there is no entropy production.
Let us return to the resonant wave-particle interaction in the γk → 0 limit (3.3.8). The presence
of δ function in the QLT equations (3.3.4,3.3.6) allows one to eliminate k as an independent
variable. The final system of equations is as follows

∂ f0 ω p2 ∂ W (v) ∂ f0
=π
∂t mn0 ∂ v |v −Vg (v)| ∂ v

56
 Physics 218A

∂ ω p2 ∂ ℜε −1 ∂ f0
 
W = 2π ωk |k| v W (v)
∂t n0 ∂ ωk k=ωk /v ∂ v
Here Vg (v) = ∂ ωk /∂ k and W (v) is the wave energy density (2.3.16),´ both ´ at k =
calculated
ω /v.
´k 2 Using these equations, one can show that the total energy mv2 f dv/2 + W dk =
0 k
mv f0 dv/2 + ω p W (v) dv/v2 , and momentum mv f0 dv + Pk dk are conserved. The wave
´ ´ ´

momentum density is Pk = kWk /ωk = W (v) /v and ω∂ ℜε /∂ ε ≈ 2. Note, that the quantity
Nk = Wk /ωk may be interpreted as a number of quanta, , using a quantum-mechanical analogy.
Indeed, the energy of a quantum with the wave number k (sometimes called plasmon) can be
written as ∆Wk = h̄ωk , so that the number of such quanta is Nk = Wk /∆Wk . The momentum of
the quantum is then ∆Pk = h̄k. As the phenomena, we are interested in are purely classical, we
may use the units in which h̄ = 1.

3.4. Quasi-linear integral, relaxation of wave pa kets and

unstable beams

Let us assume Langmuir waves propagate at phase velocities ωk /k ≈ ω p /k > 0. The QL equa-
tions are simplified to
∂ f0 ω p2 ∂ W (v) ∂ f0
=π (3.4.1)
∂t mn0 ∂ v v ∂ v
∂ ω p ∂ f0
W = π v2 W (v) (3.4.2)
∂t n0 ∂ v
Apart from the velocity integrated particle phase space density, energy and momentum conser-
vation also the following ’local’ quantity is conserved

∂ ωp ∂
 
3
f0 − W /v = 0 (3.4.3)
∂t m ∂v

The latter result is the so-called QL integral, a powerful tool in solving wave-particle interaction
problems. This is because not only the initial conditions f0 (t = 0) = f00 (v), W (t = 0) = W0 (v)
is known but also the final stage of the evolution, when the system reaches a steady state, can
also be determined without calculations in many practical cases. The first such case is when
f0 (t = ∞) = f0∞ (v) = const where W (v) 6= 0 (this nulls both r.h.s. in [3.4.1,3.4.2]). The sec-
ond case is when W (v) = 0, where f0∞ (v) 6= const. Then, integrating (3.4.3) in time, we can
determine either W or f0 in the final state, or we can relate them at any moment of time.
Let us consider first the following simple problem. Assume that at t = 0, a Langmuir wave
packet having phase velocities 0 < v1 < ω p /k < v2 is launched into a plasma with ∂ f0 /∂ v < 0
for v1 < v < v2 (therefore, the waves can only be damped). From (3.4.3) we obtain
ˆv
∞ m 3
W (v,t = ∞) = W (v) = W0 (v) + f0∞ − f00 dv
 
v (3.4.4)
ωp
v1

57
 Physics 218A

where f0∞ can be taken obtained from the final ’plateau’ and particle conservation requirement

ˆv2
1
f0∞ = f00 dv,
v2 − v1
v1

provided that W ∞ > 0. The opposite case, W ∞ < 0, simply means that the initial wave energy
is insufficient to form a plateau and, in reality, the time asymptotic solution is W ∞ = 0, while
f0∞ (v) 6= const. The latter quantity can be determined from (3.4.4), Fig.3.4.1.

Next, consider a ’warm’ beam instead of

the wave packet, and assume waves to be ini-
tially weak, Fig.3.4.2 for all v. Clearly, those
waves that are resonant with the beam parti-
cles where ∂ f0 /∂ v > 0 are unstable. Their
amplification will result in diffusion of parti-
cles to lower velocities with the further wave
excitation. Finally, a plateau will form wher-
ever ∂ f0 /∂ v was positive during the beam re-
laxation. The height of the plateau is given by
the above formula, with the v1 being the inter-
section point of the f0∞ level with the thermal
core distribution (v1 ≃ VT ), and v2 - is the in-
tersection point with the right side of the beam
distribution, where ∂ f0 /∂ v < 0. The wave en-
ergy density W ∞ (v) can still be obtained from
(3.4.4) where W0 can be neglected in most in-
teresting cases.
The process of the beam relaxation can be
followed also dynamically. Since the wave
Figure 3.4.1.: Absorption of the Langmuir wave
energy to the left from the unstable region is
packet by stable particle distribu-
very small (at the level of thermal fluctuations
tion
Wth ), the relaxation occurs in the form of a
front on the beam distribution propagating to
lower velocities. The waves are rapidly excited within the front since ∂ f0 /∂ v is large there and
a plateau forms behind it, f0 = f0+ = f0 (u + 0,t) where u (t) is the front coordinate in veloc-
ity space. At v ≈ u, we may represent the time dependence of f0 as f0 (v,t) = f0 (v − u) and
similarly for W (v − u). Therefore, dividing eq.(3.4.2) by W and integrating across the front we
approximately have
W+ ωp ω p u2 nb
−u̇ ln = π u2 f0+ = π
Wth n0 n0 v2 − u
where W + = W (v = u + 0,t), v2 is the right edge of the plateau (approximately equal to v2
discussed earlier if the beam is narrow) and nb is the beam density. We also used the particle

58
 Physics 218A

conservation to obtain the plateau height f0+ . Integrating the last equation we obtain
ˆt
v2 u nb dt nb t
+ ln = π ω p + 1 ≈ π ωp + 1
u v2 n0 Λ (t) n0 Λ
0

The quantity Λ (t) = ln (W + (t) /Wth ) ≫ 1 is so large that we have neglected the slow dependence
W + (t) and took it out of the integral. The characteristic beam relaxation time, that is when u
becomes u ∼ VT ≪ v2 ≃ Vb is τrel = ω p−1 Λn0 /nb . The factor Λ is close to the Coulomb logarithm
LC , Sec.1.4.
The total energy transferred from the beam
to the waves can be obtained from (3.4.4) and
it is equal to beam energy loss:
ˆvmax
mV 2 mv2 mV 2
∆Eb = Wtot = nb b − f∞ dv ≃ nb b
2 2 3
vmin
(3.4.5)

3.5. Parametri
instability, Weak turbulen e Figure 3.4.2.: QL beam relaxation t0 < t1 < t∞ .
and wave as ading [2℄

QL equations include only the resonance ω −

kv = 0 that arises in the first order wave-
amplitude expansion of the Vlasov equation.
Obviously, for certain wave spectra ωk , this
resonance condition may not be met for many (if at all) particles, e.g., if ω /k ≫ VT (ω /k > c).
Under these circumstances, it is natural to continue the expansion of the kinetic equation which
should lead to higher order resonances with two or more waves with different ω and k involved
in the wave-particle interaction. The next resonance is

ωk1 ± ωk2 = (k1 ± k2 ) v (3.5.1)

which signifies the particle resonance with a beat wave. Another interpretation of this relation
is that a wave quantum (plasmon, phonon) in a state k1 ≡ (ω1 , k1 ) scatters to state k2 on a free
particle.
However, waves may strongly interact with each other without particles. This can happen
when the combination frequencies and wave numbers appearing in the second order amplitude
expansion are in a resonance with linear terms. This occurs when

ω1 + ω2 = ω ; k1 + k2 = k

and can be interpreted as a decay of the plasmon k into two plasmons k1 and k2 , schematically
l → l1 + l2 . Consider, for example, an acoustic type wave with the dispersion law ω 2 = C2 k2 .

59
 Physics 218A

The phase velocity C depends on the plasma characteristics such as its temperature, density etc.
Therefore, if a strong wave is present in such a plasma, the dispersive properties of other (weak,
or test-) waves are modified (modulated) by the “pump” wave. Specifically, we can expand

C2 = C02 [1 + a cos (ω0 t − k0 x)] + β (k)

Here a is the amplitude of the modulation (strong wave) and β (k) is a dispersive correction
(amplitude independent) to the phase velocity. Using this expression and the relation ω 2 =
C2 k2 , let us reconstruct the differential equation that governs the wave interactions by replacing
ω 2 → −∂ 2 /∂ t 2 and k2 → −∂ 2 /∂ x2 :

∂ 2u ∂ 2u
−C 2
[1 + a cos (ω0 t − k0 x)] + Lu = 0
∂ t2 0
∂ x2
where L is a linear differential operator corresponding to the Fourier image k2 β .
In the Fourier-Laplace representation
ˆ
uk,ω = u (x,t) eiω t−ikx

(we write uk instead of uk,ω for short) from the above equation we obtain
a h i
ω 2 − ωk2 uk = C02 (k0 − k)2 u∗k0 −k + (k0 + k)2 uk0 +k


(3.5.2)
2
where we have denoted ωk2 ≡ k2C02 + β k2 . Clearly, the uk -phonon will be excited efficiently if
ω 2 ≈ ωk2 , that is, it should belong to the eigenwave spectrum of the linear system. It is natural
to assume that the “pump” wave, k0 is also close to one of the plasma oscillation branches, e.g.,
ω02 ≡ k02C02 + β (k0 ) k02 , but, generally speaking, it may deviate from it due to a strong nonlinear
deformation of the dispersion curve or it may belong to a different than k and/or k0 − k branches.
In addition to (3.5.2), we need equations for k0 − k and k0 + k, the side bands of the pump wave.
They can be formally obtained from (3.5.2) by replacing k → k0 ± k and ω → ω0 ± ω and taking
c.c where needed. So, for u∗k0 −k we obtain
  a h i
(ω0 − ω )2 − ωk20 −k u∗k0 −k = C02 k2 uk + (2k0 − k)2 u2k0 −k (3.5.3)
2
The problem we encounter here is clear. New unknown amplitudes, such as u2k0 −k , appear when
we attempt to obtain a closed system. However, for these amplitudes to be large, the resonant
factors on the l.h.s. of the respective equations should be small. First of all, we require that it is
small in the last equation, i.e. for the mode uk0 −k which means

ω0 = ωk ± ωk0 −k
Assume that the daughter waves k, k0 − k belong to the same branch. The dispersion rela-
tion imposes an additional constraint, so that we may only satisfy the relation with one (say,
plus) sign. Similarly, the beat wave with the wave number 2k0 − k will be excited also inef-
ficiently. Thus, we neglect the term uk0 +k in eq.(3.5.2) and u2k0 −k in (3.5.3). The unknown
frequency ω we represent as ω = ωk + iγ with γ ≪ ω . Thus, from these equations we have

60
 Physics 218A

a2 2 k2 (k0 − k)2
γ2 = C (3.5.4)
16 0 ωk ωk0 −k
The decay instability requires ωk ωk0 −k > 0,
which together with the decay condition
ω0 = ωk + ωk0 −k
means that ω0 > ωk , ωk0 −k . The last condi-
tion is justified by the energy conservation
in the wave decay; the energy of the parent
wave should be larger than each of the decay
products. If all the three modes are on the
same branch it is not always possible to sat- Figure 3.5.1.: The left plot corresponds to
isfy the decay conditions. For instance, Lang- nonzero intersection of ω (q) and
muir waves do not decay within their own ω (|k + q|) surfaces, where vector
spectrum l 9 l1 + l2 , as ωk + ωk0 −k > ωk0 q originates at the coordinates of
for any triplet k0 , k0 − k and k. In certain vector k. This means that the de-
more complicated situations the possibility of cay conditions are satisfied for the
decay within a single branch of plasma os- wave vectors on the intersection
cillation can be established graphically. Con- lines. Surfaces on the right do not
sider the case when ω (k) = ω (k) for a two- intersect, that means no decay.
dimensional spectrum k = (kx , ky ) and ω (0) = 0. The requirement ω (|k + q|) = ω (k) + ω (q)
can be analyzed by considering Fig.3.5.1.
However, selecting the daughter waves on different branches greatly facilitates the decay pro-
′
cess. Consider the l → l + s process (plasmon → plasmon + phonon). The main nonlinear effect
here leading to the parametric (decay) instability is modulation of the plasma density (the fre-
quency of the parent wave ω p ) by an acoustic wave. The dispersion equation for the Langmuir
plasmon implies

ω 2 − ω p2 − 3k2VTe
2


Ek = 0
Here Ek is the Fourier-Laplace component of the electric field of the plasmon, k = (ω , k). When
the density is modulated, we replace ω p2 → ω p2 (1 + δ n/n0 ). Returning to the x,t representation
of the last equation we get

∂ 2 ∂ δn
 2 2

− 3VTe 2 + ω p E (x,t) = ω p2 E
2
(3.5.5)
∂t 2 ∂x n0
The linear equation for δ n is
ω 2 − k2Cs2 δ n = 0

where Cs2 = Te /M = ∂ P/∂ ρ , so that MCs2 δ n = δ nTe is the pressure perturbation. As we know,
when the Langmuir waves are present, the pressure needs to be supplemented by the pondermo-
tive pressure
2
E
δ nTe → δ nTe +
16π

61
 Physics 218A

where hi is high frequency (ω p ) averaging. Again, returning to the x,t- representation we have

∂ 2 ∂ ∂ 2 E2
 2

2 
−Cs 2 δ n = 2 (3.5.6)
∂ t2 ∂x ∂ x 16π M
Based on the general decay principles we represent the parent plus daughter Langmuir wave
fields as
E = E0 ei(k0 x−ω0t) + E1 ei[(k0 −k)x−(ω0 −ω )t] + c.c.
1
δ n = δ nk ei(kx−ω t) + c.c.
2
On substituting to the above equations we obtain

k2
ω 2 − ωs,k
2
δ nk = E0 E1∗


(3.5.7)
2mM ω p2
h i ω p2
(ω0 − ω )2 − ωk20 −k E1∗ = δ nk E0∗ (3.5.8)
2n0
Here ωk2 = ω p2 + 3k2VTe
2 , ω 2 = k2C 2 . As in the general case above, we look for the solution for
s,k s
ω = ωs + iγ ≈ ωs . The decay instability has the growth rate γ (see 3.5.4)

ωs ω p |E0 |2
γ2 = ≡ γd2 (3.5.9)
16 4π n0 Te
As the frequency of the daughter plasmon is lower than ω0 the wave energy is transferred longer
waves during the decay process. The condition for the decay k0 → k, k0 − k can be written as
r
3 2h 2 i m
λD k0 − (k0 − k) = kCs = kλD
2
2 M
or r
1 m
(k0 − k/2) λd =
3 M
Evidently, the wave length of the parent plasmon should not be too long, k0 λD > (m/M)1/2 /3.
If, however k0 λD ≫ (m/M)1/2 , then the phonon has k ≈ 2k0 , that means that the parent and the
daughter plasmons should propagate in opposite direction, k1 ≡ k0 − k ≈ −k0 . In this pcase, the
change of the plasmon wave number k after each decay is small, ∆k = k0 + k1 ≈ λD−1 m/M ≪
k0 , so that a differential approximation may be used in describing the wave kinetics. Apart
from the l → l ′ + s decay, also the processes with the electromagnetic (transverse) plasmons are
possible: t → t ′ + l, s; t → l + l ′ , s.
Similar processes are known from other physical disciplines. So t → t ′ + s corresponds to the
scattering of light on the acoustic lattice oscillation in solid state physics (Brillouin scattering).
The process t → l + l ′ is known as light scattering on the optical oscillations of lattice (Raman
scattering).
If more than one of the waves involved into the process is damped, the process has a threshold.
However, damping of only one daughter wave only reduces the instability growth rated but does

62
 Physics 218A

not stabilize it. Indeed, let us return to the l → l ′ + s decay with the damping νs of the sound
q by replacing ω → ω + 2νs ω . The growth
wave. This should be included into eq.(3.5.7) 2 2

rate instead of γ = γd becomes γ = −νs /2 + νs2 /4 + γd2 ≈ γd2 /νs which is characteristic of a
dissipative instability. Only if both daughter waves are damped, then the decay instability of
the parent wave has a threshold. In this case we should modify also the equation (3.5.8) for the
plasmon, ω12 → ω12 + 2iνe ω1 , where ω1 = ω0 − ω and νe are the daughter plasmon frequency
and the damping rate, respectively. The result then modifies to

γ = γd2 /νs − νe
so that the instability may be stabilized. The amplitude threshold is given by γd2 = νs νe .
The above treatment of the decay instability is limited to its initial phase in which the am-
plitudes of the decay products remain small and the amplitude of the pump wave is close to its
initial value. As the energy is transferred to the decay product, the linear stage of the instability
ceases and the equation for the parent wave amplitude needs to be added to the two used. This
can be done in exactly the same way as the transition from eq.(3.5.5) to (3.5.8) has been made.
The three equations thus obtained can be recast in a symmetric form, using, not surprisingly, the
respective number of quanta for each wave as dependent variables instead of their amplitudes

∂
 2 
+ ωl Cl = −2ωlVkl k j kn C jCn
2
(3.5.10)
∂ t2
 2
where the number of quanta is related to the amplitudes by C j  = Nk j = Wk j /ω j , the phases
of C’s are those of the respective amplitudes and l 6= j 6= n. The quantities V are the matrix
elements of the wave-wave interactions. Note that they can be made real by changing phases
in C’s. It is convenient to separate the fast oscillations in the wave amplitudes by representing
C j = A j (t) exp (−iω j t) where A (t) is assumed to be slow due to the weakness of interaction.
The equations for A’s are then the following

∂ A1 ∂ A2 ∂ A3
= V1 A2 A3 , = V2 A1 A∗3 , = V3 A1 A∗2
∂t ∂t ∂t
where we have used the decay conditions ω1 = ω2 + ω3 and denoted V1 = Vk1 k2 k3 , V2 = Vk2 k1 ,−k3 , V3 =
Vk3 k1 ,−k2 . As the total energy and momentum of the three interacting waves conserve, the matrix
elements are constrained by certain relations. These can be obtained from underlying differen-
tial equations in each particular case, e.g., (3.5.5-3.5.6), or, in general, using the conservation
laws. The latter imply that

ω1 N1 + ω2 N2 + ω3 N3 = const (3.5.11)
k1 N1 + k2 N2 + k3 N3 = const

From these relations, using ω1 = ω2 + ω3 and k1 = k2 + k3 and the above equations for A’s we
obtain
k2 (V1 +V2 ) + k3 (V1 +V3 ) = 0
where, again, we have used the 4-vector notation k = (ω , k) for short. The above relation is an
(overdetermined) system of four scalar equations for two variables V1 + V2 and V1 + V3 and the

63
 Physics 218A

only solutions are the trivial ones. Therefore, V2 = V3 = −V1 . The next symmetry comes from
the fact that ℑV = 0 and from the fact that the original Fourier transform was made from the real
physical variables. Therefore, their Fourier images should posses the quality of being complex
conjugate under the transformation k → −k. Consequently, we have Vk1 k2 k3 = V−k1 ,−k2 ,−k3 . In
addition, based on (3.5.10) we have Vk1 k2 k3 = Vk1 k3 k2 . Therefore we have the following symme-
tries
Vk1 k2 k3 = Vk1 k3 k2 = V−k1 ,−k2 ,−k3 = −Vk2 k1 ,−k3 = −Vk3 k1 ,−k2 ≡ V
Therefore, the three-wave equations simplify to
∂ A1 ∂ A2 ∂ A3
= VA2 A3 , = −VA1 A∗3 , = −VA1 A∗2
∂t ∂t ∂t
They obviously conserve |A1 |2 + |A2,3 |2 so that we obtain the following integrals of motion
N1 + N2 = const
N1 + N3 = const
These are known as Manley-Rowe relations in optics. Their meaning is fairly clear; when a new
quantum of type k2 or k3 is born, the number of parent quanta N1 must decrease accordingly.
These conservation laws of the number of quanta, along with the energy integral (3.5.11)
allow the following geometrical interpretation of the process. Assuming that the A’s are real, we
obtain
ω1 A21 + ω2 A22 + ω3 A23 = const, A21 + A22 = const
√
which are the equations of an ellipsoid with half-axes 1/ ω j in coordinates A j and that of
a cylinder with the axis A3 . As ω1 > ω2,3 (and we can assume ω2 > ω3 ), the ellipsoid axis
along a3 is the longest, so that the intersection curves are shown in Fig.3.5.2, and the respective
temporal evolutions of the wave in Fig.3.5.3.

Figure 3.5.2.: Trajectories in three-wave inter-

action

64
 Physics 218A

3.6. Nonlinear Landau damping (indu ed s attering)

Let us return to the beat-wave-particle reso-

nance (3.5.1). In continuation of the discus-
sion at the beginning of the preceding sec-
tion, we reconsider the parametric decay of a
Langmuir wave into another Langmuir and an
acoustic wave, l → l ′ + s. However, this time
we assume an isothermal plasma, Te ≃ Ti , so
that free phonons cannot propagate in such
plasma. It turns out that a nonlinear process
associated with the resonance

ωk1 − ωk2 = (k1 − k2 ) v

that we characterized earlier as an inelastic Figure 3.5.3.: Time dependent three-wave inter-
(induced) scattering of a plasmon on a ther- action in stable and unstable cases
mal particle, accompanied by a change of
ω , k, can also be interpreted as a l → l ′ + s decay. Since the acoustic mode is strongly damped,
the process is associated with emission/absorption of Langmuir waves by a thermal ion resonat-
ing with the beat wave.
To describe this process ions should be treated kinetically. We include their contribution in the
form of dielectric function εi (ω , k), while electron response is of the Boltzmann type, 1/k2 λD2 ,
so that the equation for the charge density perturbations in the Fourier-Laplace representation
(after some obvious rearrangement of terms) can be written down as follows

k λD + εi−1(ω , k) δ nk = 0.
 2 2 

To couple low-frequency sound-type motion with the Langmuir wave we add, as usual, the
pondermotive force by replacing

k2
k2 λd2 δ nk → k2 λd2 δ nk +

2
E
16π mω pe
2

In the pondermotive force we choose a beat wave between the pump wave E0 and the test wave
E1 , so that the equation for δ n rewrites

k2
k2 λD2 + εi−1 (ω , k) δ nk = − E E∗
 
2 0 1
(3.6.1)
8π mω pe

where, as in the preceding section, we use the notations ω = ω0 − ω1 , k = k0 − k1 . For the test
wave E1 , we use eq(3.5.8)
ω p2
ω1 − ωk21 E1∗ = δ nk E0∗
 2 
(3.6.2)
2n0
Coupling between E1 and E0 results in a frequency shift, so that ω1 = ωk1 + δ ω , with δ ω ≪ ω1 .
The imaginary part of δ ω will determine the growth rate of the induced scattering process. From

65
 Physics 218A

the last equation we have E1∗ ≃ ω p δ nk E0∗ /4n0 δ ω . In eq.(3.6.1) we expand εi = εi′ +iεi′′ assuming
εi′′ ≪ εi′ and taking into account the linear relation εi′ ≃ εe ≃ 1/k2 λD2 . Thus, for the instability
growth rate we have
|E0 |2
δ ω ′′ = ω pe k2 λD2 εi′′
128π n0 T
The imaginary part ε ′′ can be taken from (2.3.17), where the beat wave frequency and wave
number ω0 − ω1 , k0 − k1 should substituted for ω , k. Clearly, for ω /kVTi > 1 the imaginary
part of ε is exponentially small, so we use the opposite approximation, where ε ′′ (ω , k) ≈
π /2ω / |k|3 λDi
p 2 V . The result then reads
Ti

π |E0 |2
r r
M k02 − k12
δ ω ′′ = 3 ω pe λD
2 m |k0 − k1 | 128π n0 T
Note, that we assumed Te ≈ Ti .
The last formula shows that the decay of the parent plasmon k0 occurs only into a longer wave
k1 < k0 . Obviously, if a wave packet of plasmons is excited instead of a monochromatic wave
E0 , the instability growth rate can be generalized to this case as follows

πM |Ek0 |2
r
k2 − k12
δω = 3
′′
ω pe ∑ 0 λD (3.6.3)
2m k0 |k0 − k1 | 16π n0 T

(we have taken into account the relation E0 = 2Ek0 corresponding to the normalization of the
monochromatic wave E0 ). The k1 - plasmons will then grow according to
d
|Ek1 |2 = 2δ ω ′′ |Ek1 |2 .
dt
Therefore, an inverse wave cascade results from this process, that is the wave energy is system-
atically decreased. This is physically obvious, as the scattering occurs on a stable (Maxwellian)
ion distribution (∂ f /∂ v < 0). Therefore, more virtual (non-eigenwave) phonons are absorbed
by particles than emitted. Consequently, the daughter plasmon should have a longer wave and
lower energy than the parent plasmon. The total number of plasmons Nk is
conserved. However,
most of the plasmon energy in the expression ωk Nk = ω pe 1 + 3k2 λD2 /2 Nk is in its constant
k- independent part, so that the energy does not change significantly. It is seen from the last
equation that

d
dt ∑
|Ek1 |2 = 0
k1

The maximum growth rate is achieved at the virtual phonon ω /kVTi ≈ 1 and is equal to

|Ek0 |2
δ ωmax
′′
≡ γmax ∼ ω p
16π n0 T
Each scattering event results in a∆k wave number variation of
pa plasmon determined by the
relation ω0 − ω1 ≈ 3k∆kλd ω pe ∼ kVTi , which yields ∆kλD ∼ m/M/3. This means that the
2

66
 Physics 218A

scattering of relatively short waves 1 > kλD > m/M is a small-step (differential) process,
p

∆k ≪ k. Similarly to the parametric instability l → l ′ + s, resulting in ion-acoustic waves, the

oppositely directed plasmons are strongly coupled here, k0 + k1 ≈ ∆k ≪ k0 . One may also
assume that ∆k is much smaller than the total spectral width, k. Then, according to (3.6.3)
and the differential spectral transformation, the total time for the spectral transformation will
be increased by a factor (k/∆k)2 . Indeed, the equation for the spectral transformation can be
written as follows

∂ Nk k2 − k2
=∑ 0 Nk Nk0 H (|k0 − k|)
∂t k0 |k0 − k|

where the kernel H (κ ) vanishes at κ & ∆k due to the frequency constraint imposed on the two
interacting plasmons discussed above (see ε ′′ ). Assuming ∆k ≪ kmax − kmin ∼ k, we expand Nk0
near |k0 | = |k|:
∂ Nk ∂ Nk
− α Nk =0 (3.6.4)
∂t ∂k
where, k > 0 (for simplicity, we assume N−k = Nk , as the spectrum quickly symmetrized in k
according to the strong coupling of opposite plasmons) and α = ∑κ κ 2 H (|κ |) ∼ (∆k/k)2 . It is
seen from the last equation, that the nonlinear transformation rate of the spectrum is

∆k2 m |Ek0 |2
γNL ≃
k2
γmax ≃ ω pe ∑
M k0 16π n0 T k2 λD2
.

In view of this result, it is worthwhile to reconsider the problem of quasi-linear beam relax-
ation in plasmas. In particular, we need to substitute the wave energy transferred from the beam
obtained earlier (3.4.5) into the above spectrum transformation rate. For the QL picture to be
valid, it is required that the linear instability and beam relaxation proceed faster than the waves
are nonlinearly cascaded away from the resonant region, γNL < γb /Λ < γb . In this case, one can
separate these processes and the nonlinear spectrum transformation occurs after the QL stage is
finished:
γb
 2
M T
∼ ≫1
γNL m mVb2
Here we have taken into account the following estimates for the beam growth rate and char-
acteristic wave number, γb ∼ ω p nb /n0 and k ∼ ω p /Vb . For energetic beams, however, when
γb /γNL = ε ≪ 1, a self-induced transparency of the beam is a viable possibility. In this case,
the energy of beam generated waves grows to the level of ε W ∼ ε mnbVb2 . Then, the waves are
continuously removed from the resonant region by the induced scattering process and the beam
relaxes only at a slower, εγb ≪ γb rate.

3.7. Wave Energy Pathways. Plasmon Condensate

The wave energy pumped into a plasma by e.g. beam instability must be ultimately dissipated.
However, nonlinear, weakly turbulent processes, such as the parametric decay and induced scat-
tering (NL Landau damping) result in a red-shifted plasmon spectra which cannot dissipate on

67
 Physics 218A

thermal particles. This problem is known as the Langmuir wave “condensate” problem. Again,
it is convenient to use the analogy between the gas of plasmons or other waves (ion-acoustic, EM
plasmons) and conventional gases. We have already introduced the the distribution of number
of wave quanta in wave numbers
Nk = Wk /ωk
by treating the total wave energy density as a sum of Nk quanta, each of an energy h̄ωk (using
h̄ = 1 units, however, in view of the classical character of the phenomena).
The equation for Nk that emerged while considering the above mentioned weakly turbulent
processes of wave dynamics can be cast in the following general form

d
Nk = St {Nk } + 2γk Nk , (3.7.1)
dt
where, in a general case, d/dt = ∂ /∂ t + (∂ ωk /∂ k) ∂ /∂ x − (∂ ωk /∂ x) ∂ /∂ k is the full derivative
along the plasmon trajectory with the second term describing the plasmon propagation at the
group velocity and the third term being responsible for wave refraction. This form of the wave-
kinetic equation makes the analogy between the plasmon gas and the conventional gases even
more complete by underlying the common Hamiltonian structure (Poisson bracket) with ω (k, x)
(quantum energy) interpreted as a Hamiltonian of a quantum and x, k as a canonical coordinate
and momentum. For the purposes of this section it is sufficient to consider the homogeneous
plasma with ∂ Nk /∂ x = 0, so that d/dt = ∂ /∂ t and we will return to the more general case in the
next section.
The first term on the r.h.s. is the so-called wave-wave collision integral, again analogous to
the particle collision integral in Boltzmann equation. Its specific form depends on the dominant
NL process (there-wave interaction, decay, induced scattering). It can be derived by re-writing
the wave amplitude equations, e.g. eqs. (3.5.5,3.5.6) in a quadratic form (Nk ) and averaging over
the wave random phases. The second term on the r.h.s. of eq.(3.7.1) is the energy input by an
instability.
It is instructive to make contact with the corresponding problem of the dissipation of tur-
bulence in ordinary fluids. Contrary to the plasma, the turbulent motion of a fluid cannot be
represented as an ensemble of (weakly interacting eigen-) modes. Instead, we have a picture of
continuous breakdown of larger vortices into smaller ones until they reach the dissipation scale.
The most reliable simple approach to the resulting spectrum is based on a dimensional analysis
and on the assumption that the energy flux across the scales is constant in the so-called inertial
interval. This is the range of wave numbers where neither the energy input (long wave, outer
scale of turbulence) nor the dissipation process (viscosity) play any role. The energy flux for a
given scale k−1 can be written down as the wave energy density Wk multiplied by the cascade
speed k/τk so that
k
Wk = const (3.7.2)
τk
As the breakdown of the vortices is generated by the term v · ∇v, we can write 1/τk ∼ kvk .
√ of pulsations kWk associated with the scale k may be represented as ρ vk , so that
The energy −1 2

1/τk ∼ k kWk . Substituting this in the above condition, we obtain the famous Kolmogorov law

68
 Physics 218A

Wk ∝ k−5/3
Thanks to the presence of weakly interacting modes in plasmas with a mathematically rigorous
description is the framework of the weak turbulence theory, one may go beyond the dimensional
analysis. The difficulty, however, is in the multiple possibilities of wave-wave interactions be-
longing to different branches so there is no universal Kolmogorov type spectrum. However,
when only one particular mode is known to be dominant, such spectrum can be found. The
simplest such example is the spectrum of ion-acoustic turbulence.
Let us assume the source is, as usual, at long scales. In this range, the wave dispersion is
not important and the wave decay is admitted. The most strongly interacting waves have almost
parallel wave vectors, as wave momentum and energy should be conserved simultaneously (and
ω = kCs ):

ωk = ωk1 + ωk2 ; k = k1 + k2
The wave interaction process is described by the St- term of the following form
ˆ
St {Nk } = Vkk1 k2 Nk1 Nk2 dk1 dk2

which describes here the merging of two long wave phonons into one short wave phonon ac-
cording to the above conservation laws. The mode interaction time we have from eq.(3.7.1)
1
∼ ωk kWk /n0 T
τk
The only quantity that have the dimension of frequency is ωk = kCs . Assuming a constant energy
flux (3.7.2) we obtain for the ion-acoustic turbulence

Wk ∝ k−3/2

The same spectrum has been obtained by an accurate solution of eq.(3.7.1).

The problem of the Langmuir wave spectrum is considerably more difficult. As we know,
both l → l ′ + s and l → l ′ + s̃ processes (where the second implies that a virtual photon is born
and absorbed by a thermal ion) lead to the spectrum reddening. This does not remove much of
the Langmuir wave energy due to the big k−independent term (ω p ) in the plasmon frequency,
leading to the formation of a condensate, similar to the Bose condensate phenomenon.

3.8. Modulational instability and Langmuir wave ollapse

It turns out that the long-wave condensate, treated as a plasmon gas, is itself unstable with respect
to modulations of its density. The key here is the pondermotive pressure that builds up in the
regions of increased plasmon density. This pressure expels electrons which pull ions so that a
plasma cavity is formed. More plasmons get trapped into this cavity and the process continues.

69
 Physics 218A

As we know, plasmons can be treated as gas particles with an energy ωk , that also depends on
coordinate through the perturbed plasma density

δn
 
3
ωk (x) = ω p0 1 + + ω p0 k2 λD2 (3.8.1)
2n0 2

so that (x, k) are coordinate and momentum of the plasmon. The equations of motion of plas-
mons can be written, considering, their energy to be a function of coordinate and momentum in
the following Hamiltonian form (see eq.3.7.1):

dk ∂ω ∂ δn
=− = −ω p0
dt ∂x ∂ x 2n0
dx ∂ ω
= = 3ω p0 λD2 k
dt ∂k
The function ω (x, k) plays the role of a Hamiltonian for the plasmon motion. Using the
analogy with a particle motion (of momentum ∝ k and coordinate x) we see that the density
perturbationsδ n play the role of a force potential for the plasmon and so, plasmons will cluster
near density troughs. Their increased pondermotive pressure will expel more electrons from the
troughs making them deeper, thus resulting in a density self-modulation instability. Note, that
the density wells will trap preferentially the long wave (low energy) plasmons, i.e. precisely the
plasmons from the Langmuir condensate.
Similarly to the particle trapping in a Langmuir wave, which cannot be treated perturbatively
(i.e., quasilinearly), it is impossible to treat the plasmon trapping within the weak turbulence
theory. The trapped plasmons do not propagate freely. Instead, they form a bound state with the
density well they are trapped in. However, as in the case of, e.g., beam particles trapped into a
plasma wave, where the wave potential is de facto created by the same beam particles through
the beam-plasma instability, a qualitative analysis is possible and can be performed as follows.
We begin this analysis by continuing the analogy between the plasmon gas distribution in
the δ n (x) modulation and the ordinary plasma particles in the electrostatic potential φ (x). The
Maxwell-Boltzmann distribution of particles depending on their energy E = mv2 /2 − eφ reads

mv2 eφ
 
n0
f (v) = exp − +
(2π )3/2 VT3 2VT2 T
Using this analogy we can write the plasmon distribution as follows

1 δn
 2 
N0 k
Nk = 3/2 3 exp − 2 − 2 2
π k̄ k̄ 3k̄ λD n0

From the plasmon energy we have subtracted the constant “ground state” energy ω p0 , eq.(3.8.1)
and normalized Nk similarly to the particle distribution. The plasmon spatial density, again
analogously to the Boltzmann distribution, may by obtained by integrating the last result in k

1 δn
 
N (x) = N0 exp − 2 2
3k̄ λD n0

70
 Physics 218A

which clearly show plasmon condensation near the density wells, δ n < 0. For sufficiently small
density variations, δ N ≈ −N0 δ n/3k̄2 λD2 n0 . These variations in turn perturb the pondermotive
pressure
Pw0 δ n
δ Pw = − 2 2
3k̄ λD n0
where Pw0 = E /8π is the equilibrium plasmon pressure. The plasmon equation of state

2

(EOS) is evidently anomalous: pressure is increasing with decreasing density. This usually
leads to an instability of the medium (in this case plasma with a plasmon gas). However, there is
also the gas contribution to the EOS which may stabilize the medium. To obtain the quantitative
results, we turn to the equations of slow, quasi-neutral (acoustic type) plasma motions in which
the total pressure variation is
 
Pw0
δ P = δ Pw + T δ n = T δ n 1 − 2 2
3k̄ λD n0 T

The square of the wave propagation Cs2 = ω 2 /k2 = ∂ P/∂ ρ becomes negative (instability) for
Pw0
> 3k̄2 λD2
n0 T
When the wave energy strongly exceeds this threshold, the growth rate is close to
r
m Pw0
γ = k ∂ Pw0 /∂ ρ ≃ ω p
p
3M n0 T
Finally, we need to check that the adiabatic approximation used here is valid. For that, we require
γ < kVg = 3kλD2 ω p , which, together with the above strong instability regime means
Pw0 M
k̄2 λD2 ≪ ≪ k̄4 λD4
n0 T m
Next, we turn to the question about the fate of the plasmons trapped into the density cavities.
Their kinetic energy there is of the order of potential energy, that is

k2 λD2 ∼ δ n/n0 . (3.8.2)

When the cavity becomes deeper and narrower because of its compression by the surround-
ing gas pressure, the wave length of the trapped plasmon approaches the size of the, k−1 ∼ l.
Therefore, by the above relation, the growth of the density modulation δ n is accompanied by a
collapse of the cavity, l ∝ |δ n|−1/2 . Since the number of trapped plasmons does not change, their
energy density grows and the plasma expulsion rate increases. The process is thus explosive in
nature. What is the collapse termination mechanism? The answer depends on the dimension-
ality d of the collapsing cavity. Assuming the density modulation to be moderate, δ n < n, the
conservation of plasmon number may be represented simply as
ˆ
|E|2 dx = const

71
 Physics 218A

so that the pondermotive pressure grows as |E|2 ∝ l −d . At the same time, one needs to overcome
the internal plasma pressure while compressing the cavity. This pressure scales, by virtue of
(3.8.2), as δ nT ∝ l −2 . In one dimension, the plasmon pressure grows slower (only as ∝ l −1 )
and the plasma pressure will ultimately balance the compression and the collapse will stop. As
a result, the Langmuir soliton will form, which, however, turns out to be unstable with respect
to its modulations along the remaining dimensions.
The two-dimensional case, d = 2, is marginal, the pressure imbalance imposed at the begin-
ning is sustainable and the collapse may continue. In the 3D case, the high frequency pressure
grows faster than the gas-kinetic pressure, so that the speed of the collapse increases in time. In
these two latter cases the collapse proceeds to such a small cavity size that l ∼ λD . Under these
circumstances the plasmon velocity ω /k ∼ VTe and strong Landau damping will dissipate the
Langmuir wave energy.
The collapse dynamics may be described by eqs.(3.5.5-3.5.6), generalized to the required
collapse dimensionality, i.e. bu replacing ∂ 2 /∂ x2 → ∆. The explosive character of the process
may be revealed as follows. Assuming that the cavity compression is supersonic, in the l.h.s of
eq.(3.5.6) we retain only the first term. Using also the condition for the plasmon trapping (3.8.2)
δn 1
∼ 2 (3.8.3)
λD n0 l
2

to estimate the operator∆ on the r.h.s of eq.(3.5.6) as ∆ ∼ l −2 , we obtain the following dynamic
equation for δ n
∂2 |E|2
δ n ≃ δ n
∂ t2 16π n0 M λD2
It follows that δ n grows faster than exp (at), in time, since |E|2 ∝ l −d grows as well. Assuming
that the “explosion” of δ n obeys the law δ n ∝ (t0 − t)−α , where t0 is the moment of “explosion”,
we deduce that |E|2 ∝ (t0 − t)−2 . The index α can be obtained from the above condition of
conservation of plasmon number in the cavity, |E|2 ∝ l −d . From here and from (3.8.3) we find
that

δ n ∝ |E|4/d ∝ (t0 − t)−4/d

while the cavity itself collapses as l ∝ (t0 − t)2/d .
Now we can overview the entire Langmuir turbulence scenario. At the initial stage the wave
energy accumulates at long scales by the weak turbulence processes. Then, when the threshold
of the modulational instability is exceeded, the latter deposits the wave energy into a number
of randomly distributed cavities, where the energy is cascaded to short scales while the cavities
collapse.

−1/2
The inertial range of the turbulence is between the modulational instability scale l0 ∼ 2πλD E 2 /16π nT
and the Landau damping scale lLD ∼ 2πλD . The wave energy is carried across this interval by
collapsing cavities. The spectrum in this interval can be found, as usual, from the constant
energy flux condition (3.7.2). For convenience, it can be rewritten here as follows
dt (k)
|Ek |2 ∝ Wk ∝
dk

72
 Physics 218A

where dt is the cascading time from k to k + dk. We know that k (t) ∝ l −1 (t) ∝ (t0 − t)−2/3 , from
which we we obtain
Wk ∝ k−5/2

3.9. Anomalous resistivity

Among a plethora of plasma instabilities, particularly interesting from the theoretical standpoint
are instabilities that do not depend on the specific plasma setting, i.e., boundary conditions, or
type of the plasma equilibrium. Very important such instabilities are those driven by the plasma
current. They set on whenever the current density exceeds a critical value.
The lowest threshold for unmagnetized plasmas 1 has the ion-acoustic instability. p It starts
when the electron-ion relative (drift) velocity exceeds the ion-sound speed, Vd & Cs = Te /M.
The next one is the Buneman instability which operates when Vd & VTe , but this one is too strong
for the weak turbulence theory by which the ion-acoustic instability can be studied. Regardless
of the instability, its development should result in eliminating or significantly weakening its
cause. As this is the current, the instability results in an increased plasma resistivity.
Already the first experiments in mirror machines confirmed the idea. Early attempts to esti-
mate the effect of active resistivity were based on the conventional expression for the collisional
plasma conductance σ = ω p2 /4πν , where the frequency of binary collisions ν , was suggested to
be replaced by a characteristic frequency of the instability, ν → νe f f ≫ ν . For instance, already
using the Buneman instability growth rate γB ∼ (m/M)1/3 ω pe = (M/m)1/6 ω pi for this purpose,
results in a reasonable order of magnitude agreement with the data. However, the ion-acoustic
instability with the growth rate γs ∼ ω piVd /VTe gradually transitions into the Buneman instability
for Vd ∼ VTe , where the quantitative analytical results are really difficult to obtain.
At the same time, certain characteristics of anomalous resistivity may be established indepen-
dently of the instability at work. If the drifting electrons unstably drive waves in a plasma, it
automatically result in their loss of momentum, that is transferred to the waves, in fact to the
collective oscillations of ions. The growth of the total wave momentum is

∂
ˆ ˆ
k dk k dk
Wk = γk Wk
∂ t ωk (2π ) 3 ωk (2π )3

which should be equal to the loss rate of electron momentum, i.e. to the friction force Fd =
−νe f f mn0 Vd . Therefore, for νe f f we find

k · Vd
ˆ
1 dk
νe f f = γk Wk .
mn0Vd 2 ω k (2π )3

Note that the problem is thus reduced to the determination of the turbulence spectrum Wk . But
the growth rate should be understood quasilinearly, i.e. including a possible back-reaction of the
waves on the electron distribution.
1 Many of results obtained for B = 0 are also pertinent to plasma flows parallel to an external magnetic field.
However, other instabilities, specific to the B 6= 0 case need to be considered, as some of them have lower
thresholds.

73
 Physics 218A

From the above consideration, one can obtain the ratio of the electron to ion heating rates
independently of the type of instability at work. Indeed, the electrons are heated by the work of
the friction force, so their thermal energy increases as

dTe
n0 = n0Vd Fd = νe f f mn0Vd2
dt
Ions, on the other hand, receive the wave energy, that is generated at the rate γk and, in a steady
state, must be dissipated on ions (via induced scattering)
ˆ
dTi dk
n0 = γkWk
dt (2π )3

The ratio of the heating rates thus yields

 
dTe k · Vd Vd
= ∼
dTi ωk Vph

Where hi means averaging with the weight function γkWk and Vph = ω /k. In the most cases,
the instability requires Vd > Vph , which automatically results in a predominant electron heating.
This heating imbalance makes the ion-acoustic turbulence particularly viable for the anomalous
resistivity, even if the current velocity exceeds the threshold of Buneman instability initially.
The ion-acoustic current instability has a growth rate maximumγ ∼ ωkVd /VTe that is achieved
at kλD ∼ 1. Thus, we may estimate the effective collision frequency as
W
νe f f ∼ ω pe
n0 Te
In this part of the ion-acoustic dispersion curve, the decay is prohibited and we turn to the
induced scattering of phonons on thermal ions as the primary ion heating mechanism. We may
write the steady state wave energy budget as follows

∂W
 
W
0= = 2γk − Aωk Wk
∂t n0 Te

Here A is an operator, which in a simple case of differential wave cascading may take a form
similar to (3.6.4). We may estimate this nonlinear term associated with the induced scattering
on thermal ions as follows. As thermal motion of the ions is a key to the interaction, A ∝ Ti /Te
which is a small parameter in the case of ion-acoustic turbulence. As A, is dimensionless, it is
reasonable to assume only a numerical correction may result from a rigorous treatment, which
indeed yields
W Te γk Te Vd
≈ 10−2 ≈ 10−2
nTe Ti ωk Ti VTe
from which we find the effective collision frequency
Te Vd
νe f f ≈ 10−2 ω pi
Ti Cs

74
 Physics 218A

When νe f f ≫ νei , the current-voltage characteristics, that start from linear dependence for small
currents, when Vd is below the instability threshold, turn to almost a plateau above it. For higher
Vd , however, when the weakly turbulent picture becomes invalid the current resumes growing
again, but slower than linearly, roughly ∝ E 1/2 .
Problem with the above scenario: The resonance condition for the wave-particle interaction
ω = kv is not met for all the electrons. An electron loss cone is formed. The resolution of this
problem depends essentially on the ambient magnetic field which introduces additional modes
to scatter electrons and/or removes electrons from the loss cone.

3.9.1. Double Layers

In the above analysis the electric field was implied to be constant in space so the plasma resis-
tivity increases simultaneously in the entire volume when the current exceeds a critical value.
It is more realistic that the instability threshold is exceeded first only locally and the result-
ing anomalous resistivity increases the electric field (to preserve the total current in a steady
state) in this area which, in turn, should drive the system further above the instability thresh-
old providing a positive feedback. Thus a domain with dramatically increased electric field
may form, which means a sharp drop in electric potential (so called double layer), Fig.3.9.1.

The equations for such structure may be

obtained by considering cold electron and ion
flows accelerated in opposite directions by the
double layer. The density of each component,
obtained as a function of local potential, may
then be substituted into the Poisson equation
and the electrostatic potential may thus be ob-
tained. Assuming that the total potential drop
eφ0 ≫ Te,i , and that electrons and ions enter Figure 3.9.1.: Potential distribution and particle
the double layer from opposite sides with ve- motion within and near a dou-
locities Ve,i0 , for the bulk velocities inside the ble layer. Reflected components,
double layer we have needed to compensate for charge
q imbalance are shown with dashed
Ve = Ve0 + 2eφ /m,
2 lines.
q
Vi = − Vi02 + 2e (φ0 − φ ) /M

As in a steady state the fluxes je,i = ±ne,iVe,i = const, we can write the Poisson equation as
follows  
d2φ je ji
= 4π e  q −q  (3.9.1)
dx2 Ve0 + 2eφ /m
2 Vi0 + 2e (φ0 − φ ) /M
2

75
 Physics 218A

After integrating this equation once, we have

dφ
 2  q q q 
= 8π je m Ve0 + 2eφ /m − je mVe0 + ji M Vi0 + 2e (φ0 − φ ) /M − ji M Vi0 + 2eφ0 /M
2 2 2
dx
(3.9.2)
The integration constant is chosen as to make the electric field vanish where φ → 0. To satisfy
the same condition on the opposite side of the double layer, where φ → φ0 , and assuming eφ0 ≫
mVe02 , MV 2 , we arrive at the so called Langmuir relation
i0
p
je = ji M/m.

The φ (x) profile between 0 < φ < φ0 can be formally solved for by quadrature, but the problem
is that φ approaches its end points φ = 0, φ0 at finite x, say at x = 0, L and not at ±∞ as it should
in a localized structure. Indeed, on inspection of eq.(3.9.1), we see that at φ = 0, φ0 there is no
charge neutrality and the solution cannot be extended to the entire interval (−∞, ∞). However,
this problem can be corrected by adding background plasma components, ions at x < 0 and
electrons at x > L. The double layer will then reflect either of these components and ensure
quasineutrality outside of the domain where E 6= 0, Fig.3.9.1.
Finally, it should be noted that the condition eφ0 ≫ Te,i means that the electron drift veloc-
ity exceeds the threshold of Buneman instability. The resulting plasma oscillations near ω pi
will lead to anomalous resistivity in and nearby double layer. The quantitative theory of this
phenomenon is yet to be build.

76
 Part II.

Magnetized Plasmas

77
4. Parti le Orbits in Magneti Field
[3, 4, 5, 6℄

When an external magnetic field is applied, plasmas become anisotropic. While particles motion
along the field remains largely unaffected (with an important exception for the field strength
varying along the field line), particle motion across the field is constrained by the Lorentz force
(e/c) v × B = (e/c) v⊥ × B, where v⊥ is the perpendicular to the local magnetic field velocity
component. The general equation for the particle motion thus reads

dv e
m = f + v × B. (4.0.1)
dt c
Here f denotes the sum of all the forces other than Lorenz’s force, and the above system may be
split into the following two

dv⊥ e
m = f⊥ + v⊥ × B (4.0.2)
dt c
dvk
m = fk , (4.0.3)
dt
where vk and fk are the respective components of the particle velocity and the force along the
local field direction. While the last two equations appear to be independent from each other, they
are generally note. If f and B depend on r, the above equations may still be coupled through the
equation ṙ = v, that closes the above system and eq(4.0.1).

4.1. Guiding enter (drift) theory

We start with a simple case in which B does not change in its own direction, let’s say z, so locally
B = Bez . The equations (4.0.2) then rewrite

dvx 1
= Ωvy + fx (4.1.1)
dt m
dvy 1
= −Ωvx + fy (4.1.2)
dt m
where Ω = eB/mc being the gyro-frequency. Introducing the following complex variables in
place of the perpendicular vector components, V = vx + ivy and F = fx + i fy , the last equations
rewrite

78
 Physics 218A

1
V̇ + iΩV = F (4.1.3)
m
If Ω and F are constant, the solution is
i
V (t) = v⊥ e−iΩt − F (4.1.4)
mΩ
The resulting velocity in eq.(4.1.3) thus consists of an oscillating part, or particle rotation (gen-
eral solution of eq.(4.1.3) with an arbitrary constant v⊥ ) around a guiding center (see below), and
a drift with the velocity directed perpendicularly to the force f (particular solution of eq.4.1.3).
The particle coordinate R = x + iy evolves accordingly:

R (t) = iρ e−iΩt + RGC (t) (4.1.5)

where
i
RGC = − Ft + R0
mΩ
and ρ = v⊥ /Ω, so that |ρ | is the particle gyroradius (or Larmor radius). We will let ρ and Ω to
carry the sign of the e/B ratio where it does not cause a confusion. Summarizing the last results,
a particle orbit across the field consists of the drift motion of the particle guiding center (GC)
with a superimposed rotation around it. The velocity component along the field is controlled
by eq.(4.0.3). The drift velocity vd of the guiding center can thus be obtained from (4.1.4) or,
directly from eq.(4.0.1), by crossing its both sides with B and averaging over the particle rotation
around the GC. Note, that the lhs will vanish as a derivative of purely oscillating function and
because v̇d = 0. Expanding then the double cross product on the rhs we obtain
c
vd = f×B (4.1.6)
eB2
Note that electrons and ions drift in opposite direction if f is charge-independent.

4.1.1. ExB drift

An important implementation of the last formula is when f is due to the electric field, and the
result is known as E × B drift
c
vE = E×B
B2
Note that electrons and ions drift together here, they do not generate a net electric current across
magnetic field and, in many respects, may be considered as a single fluid. Moreover, on rewriting
this formula as

1
vE × B + E = 0
c
we recognize the expression on the lhs as E′ - the electric field in the co-moving (at speed vE )
plasma frame. This field should thus be zero in a perfectly conducting fluid.

79
 Physics 218A

4.1.2. Gradient drift

Other forces, such as gravity, or a centrifugal force associated with the curvature of the magnetic
field line along which the guiding center moves according to eq.(4.0.3), may also be substituted
in the formula (4.1.6) for vd to calculate the respective drift. Let us first consider, however, a
case in which the magnetic field strength changes across its own direction, say B = B (x) but
f = 0 (which is not essential now). Eq.(4.1.2) conserves the quantity
ˆ
vy + Ωdx

which is, of course, nothing but a component of the particle canonical momentum along an ig-
norable coordinate mvy + eAy /c, where A is a vector potential, defined so that B = ∇ × A. Using
(4.1.4-4.1.5) we represent the current particle position x (t) as the guiding center position x0 plus
an oscillatory part: x = x0 − vy /Ω and assuming that Ω (x) changes slowly on the gyroradius
scale v⊥ /Ω, from (4.1.1) we obtain
v̇x
vy ≈ ,
Ω − Ω′ vy /Ω
where Ω = Ω (x0 ). On averaging both sides over the fast oscillations, expanding in small Ω′ and
using (4.1.1) to the same accuracy, we obtain

Ω′ 2 Ω′ v2⊥
vy = v = .
Ω2 y Ω2 2
Comparing this with the general expression (4.1.6) for the drift velocity associated with a force
f, we identify the force behind the last result, not unexpectedly, with the diamagnetic force
f = −µ ∇B. Indeed, a particle of charge e in a magnetized plasma carries a ring current and thus
the magnetic moment µ = (e/2c) ρ v⊥ = mv2⊥ /2B and therefore it is subjected to diamagnetic
force. Hence, in the vector form the particle gradient drift can be written as

1 B × ∇B
vB = v⊥ ρ (4.1.7)
2 B2
If a particle is moving in a current-free region where ∇ × B = 0, the gradient drift may be
rewritten as follows. The magnetic field line equation is dl × B = 0, where dl is the line element.
In particular, dxB = dzBx (note that B ≃ Bz here), so that if the curvature radius is Rc , then
d 2 x/dz2 along the field line is d 2 x/dz2 = 1/Rc = (d/dz) Bx /B ≃ B−1 ∂ Bx /∂ z. The last expression
is, of course, just an x-component of the general formula for the curvature radius R of a line
defined by a unit tangent vector τ , that is

n/Rc = τ · ∇τ (4.1.8)
so that in our case τ = B/B and n is the unit normal to the local field line. Using the ∇ × B = 0
condition, we have 1/Rc = B−1 ∂ B/∂ x (or n/Rc = ∇B/B) and, finally

1 ρ v2
vB = v⊥ τ × n = ⊥ τ × n
2 Rc 2ΩRc

80
 Physics 218A

4.1.3. Curvature drift

When a particle moves along a curved field line with the velocity vk , its co-moving frame in
non-inertial and that particle perceives the centrifugal force fc f = −nmv2k /Rc . The drift velocity
is therefore

v2k
vC = τ × n.
ΩRc
In the case of a curl-free field the curvature drift may be combined with the gradient drift into
the following simple formula

v2⊥ /2 + v2k v2⊥ /2 + v2k B × ∇B

vBC = τ ×n = (4.1.9)
ΩRc Ω B2
In contrast to the E × B drift, the last result does depend on the particle charge and the electric
current is generated by the relative motion of electrons and ions. This imposes limitations on the
curl-free magnetic field approximation.

4.1.4. Adiabati invariants and parti le mirroring

Since

dvk
m = −µ ∇k B − e∇k φ (4.1.10)
dt
where φ is the electrostatic potential, we can write
2
!
d mvk mv2 dB
+ eφ + ⊥ = 0,
dt 2 2B dt
or, as the magnetic field does not do any work on particles and the total energy should be con-
served we rewrite this relation as

d mv2⊥ mv2⊥ dB d mv2⊥ dµ

− ≡B ≡B = 0,
dt 2 2B dt dt 2B dt
so that the magnetic moment is conserved, µ̇ = 0 to the accuracy of eq.(4.1.10). As always in the
GC theory, fields should vary slowly and smoothly on the scales of gyroperiod and gyroradius.
The total particle energy may be written as

mv2k
E = + eφ + µ B = const (4.1.11)
2
It is clear from this expression that not only the electrostatic potential may create a barrier for
particles but also the magnetic field which varies along the particle orbit. To study the magnetic
barrier, let φ = 0 and
r
B
vk = ±v 1 − sin2 ϑ0
B0

81
 Physics 218A

where ϑ0 is the particle pitch angle (sin ϑ = v⊥ /v) to the field line at a point where B = B0 . If
B = Bmin , say at the middle of a magnetic mirror machine where the field has a minimum making
thus a potential well for the particles, eq.[4.1.11], any particle that starts moving from the center
of the trap at a pitch angle ϑ0 will be mirrored at the point where sin ϑ0 = Bmin /B > Bmin /Bmax
where Bmax /Bmin is called mirror ratio. If, however, sin ϑ0 < Bmin /Bmax , such particles will pass
through the mirror (loss cone).
The conservation of magnetic moment clearly corresponds to an adiabatic invariant associated
with the quasi-periodic rotation across the field

˛ ˆ2π
1 1 2c
I1 = p⊥ dq⊥ = mv⊥ ρ d φ = µ.
2π 2π me
0

If the particle motion is also constrained in the longitudinal direction due to the magnetic mirror
effect, then also the second adiabatic invariant is approximately conserved

1√
˛ ˆ p
1
I2 = pk dl = 2m E − µ Bdl
2π π
where the integral is taken between the reflection points. If the field satisfies such condition
that the GC motion across the field is also quasi-periodic (e.g. axially symmetric magnetic trap
where particles exercise gradient drift around the device axis), also the third adiabatic invariant
may be introduced, similarly to the first one.

82
5. Fluid des ription of magnetized
plasmas

Proceeding from individual particle motion in magnetic field to the hydrodynamic approach one
follows the same steps as in Sec.1.6 but with the Lorentz force added to the l.h.s. of eq.(1.6.1):

∂f e ∂
 
1
+∇·vf + E + v × B f = St { f } (5.0.1)
∂t m ∂v c
The formal requirements, being the short mfp and collision time compared with the characteristic
spatial and time scales under consideration remain the same, including the “collisionless hydro-
dynamics”. It should be noted, however, that plasma transport across magnetic field is generally
strongly suppressed namely in the collisionless situation (apart from the drift motion, discussed
in the preceding chapter). Indeed each particle collision leads to a “jump” of its guiding center
(which is otherwise and integral of motion) across the field by an amount of gyroradius.

5.1. Two-uid des ription

By taking the first moments of eq.(5.0.1) separately for electrons and ions we obtain the two
fluid magnetohydrodynamics. Obviously, the continuity equations (zero moments) will remain
the same as the forces do not contribute due to their divergent form and the usual assumptions
regarding the St-term (conservation of particles). The Euler equations (first moments) read:

dui e
nmi = −∇pi + enE + nui × B − meνei n (ui − ue) (5.1.1)
dt c
due e
nme = −∇pe − enE − nue × B − meνei n (ue − ui) (5.1.2)
dt c
Plasma is considered to be quasi-neutral, ne = ni = n as the treatment is applied to scales
≫ λD , ω p−1 .

5.2. Magnetohydrodynami s (MHD)

The transition from the two to one-fluid description of plasma motion (which is usually under-
stood under the term magnetohydrodynamics) can be performed by simply adding eqs.(5.1.1)
and (5.1.2). The result is as follows

du 1
Mn = −∇(pe + pi ) + j × B (5.2.1)
dt c

83
 Physics 218A

where we have introduced the center mass speed

mi ui + me ue
u= ≈ ui
mi + me
M = mi + me ≈ mi , and the current j = en (ui − ue ). Using Maxwell equation c∇ × B = 4π j, one
may express the Ampere’s force c−1 j × B in eq.(5.2.1) through the magnetic field

B2
 
du 1
Mn = −∇ pe + pi + + B · ∇B (5.2.2)
dt 8π 4π
As may be seen from this equation, the magnetic force on the fluid consists of the gradient
of magnetic pressure B2 /8π and magnetic line tension B · ∇B/4π , associated with bending of
magnetic lines (cf. with eq.[4.1.8] for the magnetic line curvature).
MHD generally operates on two vector fields u and B and two scalars, n (or ρ = Mn) and
p = pe + pi . So far we have obtained eq.(5.2.2) in addition to the continuity equation. Therefore,
we need another vector equation which may be obtained by eliminating electric field from the
Maxwell equation c∇ × E = ∂ B/∂ t. This can be done using the Ohm’s law, and here different
levels of consideration are possible. At a basic level of perfectly conducting fluid, one assumes
that the electric field should vanish in a frame comoving with the local fluid envelope, that is

1
E′ = E + u × B = 0 (5.2.3)
c
Substituting E from this relation into the Maxwell equation provides the required equation for
B. However, finite conductivity and thermal pressure may need to be included. Neglecting the
inertia (l.h.s.) in eq.(5.1.2), to the same accuracy it can be rewritten as follows
e e en
enE + nui ×B + ∇pe + n (ue − ui ) × B − j = 0
c c σ
where the plasma conductivity σ = e n/me νei and we may replace ui by u in the second term,
2

so that the generalized Ohm’s law takes the following form

1 1
σ −1 j = E′ − j × B + ∇pe (5.2.4)
enc en
One sees that even in the case of infinite conductivity, the co-moving electric field E′ is generated
in the flow unless two last terms on the r.h.s. cancel.
However, substituting the E′ ≈ j/σ and eq.(5.2.3) into the Maxwell equation we obtain the
induction equation of MHD

∂B c2 2
= ∇×u×B+ ∇ B (5.2.5)
∂t 4πσ
Equations (5.2.1), or (5.2.2) together with eq.(5.2.5) and the continuity equation for n along with
the equation of state p (n) comprise the basic system of MHD equations. Note that fluid viscosity
is neglected in eqs.(5.2.1-5.2.2) while the second term on the r.h.s. of eq.(5.2.5) represents the
so called magnetic viscosity νm = c2 /4πσ , or magnetic field diffusion both associated with the
finite electric resistivity of the conducting fluid, σ < ∞. The meaning of the term diffusion is
clarified in the next section.

84
 Physics 218A

5.2.1. Flux Freezing

One important property of eq.(5.2.5) is revealed by inspecting the case B = B (x, y,t) ez in which
the equation rewrites

∂B
+ ∇⊥ · Bu = νm ∇2⊥ B (5.2.6)
∂t
where B · ∇⊥ = 0. For νm → 0 this equation for B coincides with the continuity equation for
n so that in this planar symmetry case B/n = const in that sense that the both quantities are
convected with the flow. Note that in other flow symmetries the different combinations of B and
n are conserved. The r.h.s. of the above equation for B describes its diffusion.
With νm = 0 eq.(5.2.5) is identical to the Euler equation written for the fluid vorticity Ω =
∇ × u : ∂ Ω/∂ t = ∇ × u × Ω and the Kelvin’s vorticity conservation theorem applies to eq.(5.2.5).
Indeed, consider a fluid contour, that is a closed line Γ each point of which moves with a fluid
element attached to it. The magnetic flux through the surface S enclosed by Γ is
ˆ
Φ = B · dS
S
This quantity may change in time for two reasons. First, because B does and second, because
the contour moves with the fluid. Therefore

∂B
ˆ ˆ
dΦ d
= · dS + B · dS
dt ∂t dt
S S

Since ∇·B = 0, the contribution to the second term will come from the contour displacement, not
from the rest of the surface S, that may also move with the fluid. This part of dS is dS = udt × dl,
where dl is the line element of Γ. Therefore

∂B
ˆ ˆ
dΦ
= · dS + (B × u) · dl
dt ∂t
S Γ
Replacing the integral along Γ through the flux of the curl of the integrand by Stokes’ theorem,
and using eq.(5.2.5) with σ = ∞ we find that

dΦ
= 0.
dt
As the contour Γ was chosen arbitrarily, this result means that magnetic field evolves during the
fluid motion in such a way as its lines were moving together with the local fluid elements, that
is they are frozen into the fluid.

5.2.2. Plasma Equilibrium in Magneti Field

As plasma motion across magnetic field is constrained, one may attempt to confine it in this
direction. The basic properties of various types of equilibria can be obtained from eq.(5.2.1).
First, in a steady state (equilibrium) eq.(5.2.1) requires

85
 Physics 218A

1
∇p = j × B (5.2.7)
c
This, in turn, constrains both the magnetic field lines and the current. They both should line the
surfaces of constant pressure. Furthermore, using eq.(5.2.2) we obtain

B2 1
∇(p + )= B · ∇B
8π 4π
When the field changes only across its direction, the r.h.s. vanishes and the last equation is just
a pressure balance (magnetic plus gas pressure). For the curved field, however, the force on the
left is balanced with magnetic tension. Indeed, using τ = B/B (see eq.[4.1.8]), the last relation
can be written as

B2 B2 B2 n
∇p + ∇⊥ = τ · ∇τ = (5.2.8)
8π 4π 4π Rc
In this form the equilibrium equation suggests that the imbalance of the total pressure across the
field direction should be compensated by the curvature of magnetic surfaces.

ϑ - Pin h

In a cylindrically symmetric configuration with the straight magnetic field lines B = Bz ez , the
equilibrium condition is

B2z (r) B2
p+ = 0
8π 8π
where B0 is the field outside the plasma column. Plasma is clearly diamagnetic as it makes field
weaker inside the column, say r < a. The degree of field weakening depends on the so called
plasma beta parameter: β = 8π p/B20 . If we assume that p is constant inside, r < a and zero
outside, then Bz (r) jumps at r = a. The equilibrium is then ensured by the jϑ being the surface
current jϑ = j0 δ (r − a) and

dBz 4π
=− jϑ
dr c

z- pin h

This is an equilibrium in which the current is flowing in z direction along the plasma column
supporting thus Bϑ component that confines the plasma. Maxwell equation provides

d 4π
rBϑ = jz r (5.2.9)
dr c
so rBϑ → 2I/c as r → ∞, where I = 2π jz rdr is the current flowing through the plasma. Mul-
´

tiplying the equilibrium equation d p/dr = −c−1 jz Bϑ by r2 and integrating using eq.(5.2.9), one
obtains

86
 Physics 218A

ˆ∞
1 2 2  I2
2 prdr = r Bϑ r=∞ =
8π 2π c2
0

As the integral with p can be written as N (Te + Ti ) /2π where N is the plasma density integrated
across the plasma column (number of particles per unit length of the column), the equilibrium
condition rewrites

I 2 = 2c2 N (Ti + Te)

(Bennett condition).

For e-free equilibrium

Even if plasma pressure is small compared to magnetic pressure (β ≪ 1) and it can be neglected
in eq.(5.2.7), certain constraints on the magnetic configuration remain. Namely, j × B =0 and
by virtue of Maxwell equation

∇ × B = µB (5.2.10)
This is the force-free configuration as no pressure is involved. It should be noted that this plasma
state may be obtained as a result of dynamical evolution. This may be done using a variational
principle minimizing the magnetic energy under the fixed magnetic helicity
ˆ
H = A · Bdr

where B = ∇ × A.The conservation of H can be demonstrated´ applying the same method as we

used in Sec.5.2.1. One may expect that since the energy W = B2 dr/8π should evolve towards
its minimum because of dissipation, H would remain approximately constant as it contains lower
spatial derivatives (Taylor hypothesis).
One simple cylindrical symmetry solution can be obtained from eq.(5.2.10) by writing

1 d dBz
rBϑ = µ Bz = − µ Bϑ
r dr dr
so that Bz = B0 J0 (µ r) and Bϑ = B0 J1 (µ r), where J’s denote the Bessel functions. The solution
thus predicts the reversal of Bz at µ r ≈ 2.4 where J0 has its first zero. Experiments indeed
support this finding.

5.3. Linear waves

Just as the long range electric field perturbations support collective interactions between plasma
particles in charge density waves (e.g. Langmuir and ion-acoustic), magnetic field perturbations
interact with plasma currents and can propagate in form of Alfven and magneto-acoustic waves.
These can be studied using the linear version of MHD equations (5.2.2,1.6.3,5.2.5):

87
 Physics 218A

∂u B0 B′ 1
ρ0 = −∇(p + ) + B0 · ∇B′ (5.3.1)
∂t 4π 4π
∂ ρ′
+ ρ0 ∇ · u = 0 (5.3.2)
∂t
∂ B′
= ∇ × u × B0 (5.3.3)
∂t
Here the primed variables are considered to be small compared to the (constant) zero order
variables, ρ ′ = ρ − ρ0 ≪ ρ0 , B′ = B − B0 , whileu0 = 0. It is convenient to use the plasma
displacement ξ instead of u:

∂ξ
=u
∂t
The continuity and induction equations can be readily integrated in time

ρ′
= −∇ · ξ , B′ = ∇ × ξ × B0 ≡ B0 · ∇ξ − B0 ∇ · ξ
ρ0
Substituting these relations into eq.(5.3.1) and using the usual approach to the pressure pertur-
bation as p′ = Cs2 ρ ′ = (∂ p/∂ ρ ) ρ ′ we obtain

∂ 2ξ  
= C 2
∇∇ · ξ +C 2
∇∇ · ξ − τ · ∇∇ ξ τ + (τ ∇)2
ξ − τ ∇∇ · ξ
∂ t2 s A

where τ = B0 /B0 and the Alfven velocity

B20
CA2 =
4πρ0
Rearranging the terms and decomposing ∇ = ∇⊥ + τ∂ /∂ z and ξ = ξ ⊥ + τξk , the last equation
rewrites

∂ 2ξ 2 ∂ ξ⊥
2
= C 2
∇∇ · ξ +C 2
∇ ⊥ ∇ ⊥ · ξ +C (5.3.4)
∂ t2 s A ⊥ A
∂ z2

Alfven Waves

Consider a special type of motion characterized by ξk = 0 and is incompressible in perpendicular

direction ∇⊥ · ξ ⊥ = 0. The last equation simplifies to

∂ 2ξ ⊥ 2 ∂ ξ⊥
2
= C
∂ t2 A
∂ z2
so that the dispersion relation is ω 2 = kz2CA2 . The above equation describes propagation of mag-
netic line bending and torsion perturbations along the field. As seen from this solution, the group
velocity across magnetic field is zero and initially localized disturbances of this type propagate
inside magnetic tubes without lateral spreading.

88
 Physics 218A

Magneto-a ousti Waves

The remaining two types of perturbations are associated with the longitudinal and transverse
compressions ξz 6= 0, ∇ · ξ ⊥ 6= 0. As opposed to the preceding case of the Alfven wave they can
not be separated from each other and the two remaining equations are coupled

∂ 2 ξz 2 ∂ ∂ ξz
 
= CS + ∇·ξ⊥ (5.3.5)
∂ t2 ∂z ∂z
∂2 ∂ ξz
 
∇ · ξ ⊥ = Cs ∇⊥ ∇⊥ · ξ ⊥ +
2 2
+CA2 ∇2 ∇⊥ · ξ ⊥ (5.3.6)
∂ t2 ∂z
However, if Cs2 /CA2 ∼ β ≪ 1, the quantity ∇ · ξ ⊥ ≡ ψ can be readily decoupled

∂ 2ψ
= CA2 ∇2 ψ
∂ t2
This equation is similar to that describing acoustic perturbations in a gas except the role of the
sound speed is played by the Alfven speed, as the restoring force is the magnetic rather than gas
kinetic pressure. Indeed CA2 = 2B2 /8πρ which is the “sound speed” in a gas with p = B2 /8π
and adiabatic index γ = 2. The “two dimensional” equation of state is because of the magnetic
flux freezing, cf. eq.(5.2.6), so B2 ∝ ρ 2 .
As ξz enters eq.(5.3.6) with a small parameter ∼ β compared to ξ⊥ in the second of the two
coupled equations, it should be dominant in eq.(5.3.5) as there is no small parameter. This can
be shown by writing eq.(5.3.6) as Aψ = Cs2 Bξz , where A and B are linear differential operators.
Applying then A to both sides of eq.(5.3.5) (that is projecting it onto the ψ -orthogonal sub-space)
eliminates ψ and ξz appears in its place with an extra Cs2 compared to the first term on its rhs.
Therefore, we indeed can discard the ψ -term in eq.(5.3.5) and formally applying A−1 we arrive
at the following equation for ξz :

∂ 2 ξz 2 ∂ ξz
2
= C
∂ t2 S
∂ z2
This equation describes the familiar ion-acoustic waves that are, however, constrained to propa-
gate along the field.
The separation of the latter two modes is possible only for β ≪ 1. A general consideration of
eqs.(5.3.5-5.3.6) is easiest within a plane wave analysis in which all the disturbed variables are
∝ exp (ikr − iω t). From those equations we obtain

ω 2 − kz2Cs2 ξz − kzCs2 (k⊥ · ξ ⊥ ) = 0

Cs ξz + ω 2 − k2CA2 − k⊥
2 2
Cs (k⊥ · ξ ⊥ ) = 0
2 2


−kz k⊥
which yields the following dispersion relation

ω 4 − CA2 +Cs2 k2 ω 2 + k2 kz2CA2Cs2 = 0

The two solutions for the phase velocities are

89
 Physics 218A

s

2
ω2 CA2 +Cs2 CA2 −Cs2 2
k⊥
= ± +CA2Cs2 (5.3.7)
k2 2 4 k2
These are often called fast and slow MHD waves. The former becomes the MS wave obtained
earlier in the β ≪ 1 limit while the latter tends to the ion-acoustic mode propagating along the
field. The third MHD mode, which is the Alfven wave found earlier has the phase velocity
between the two in eq.(5.3.7). It is therefore also called intermediate MHD wave, or shear
Alfven wave, as the fluid displacement there is perpendicular to both k and B0 . The three modes
are illustrated by polar (Friedrichs) diagrams showing showing the value of phase velocity (a)
and group velocity (b) as a function of angle ϑ between k and B0 , i.e. k⊥ = k sin ϑ .

5.4. Stability of Plasma Equilibria

While magnetized plasmas are easier to confine than free plasmas it is by no means guaranteed
that any equilibrium magnetic configuration will remain stable with growing plasma pressure.
For instance, the z-pinch equilibrium where the current along the plasma cord supports the az-
imuthal field that in turn confines plasma can quite easily become unstable. Indeed, suppose
that the column radius is locally decreased. Then the current density increases and the enhanced
magnetic field compresses the plasma column stronger precisely at this position reenforcing
column narrowing. This is the so called “sausage” instability and other instabilities, such as
“kink” and “screw” are associated with the column bending, where one can see how a sideway
displacement further increases due to the imbalance of magnetic forces (see Figure).

90
 Physics 218A

Two basic ideas significantly improved the stability and energy confinement in z- pinches.
First, strong magnetic field along the column (akin to the ϑ -pinch) appear as rigid walls to the
plasma and stabilizes its confinement. The second idea is to fold up the cylinder into a torus
which eliminates the energy and plasma losses through the ends of the column. This system is
known as “tokamak”.
The poloidal φ -field in Figure (more frequently denoted Bϑ ) can be supported by inductively
driven toroidal current by making use of external coils. But, to achieve the equilibrium also the
B⊥ component is needed to compensate the radial forces of expanding plasma by the Ampere
force IB⊥ /c. It is sufficient to have B⊥ ≪ Bϑ , so the magnetic field lines are still close to spirals
running around the toroidal direction. Most of them are not closed thus densely covering toroidal
surfaces. As the plasma with small Larmor radii moves primarily along the field lines (apart from
drifting across them and across the field gradient and other possible forces) the understanding
of magnetic surfaces provides a good handle on the plasma distribution inside the tokamak. In
particular, the surfaces should be isobaric, as required by the equilibrium, eq.(5.2.7).

91
 Physics 218A

Clearly, plasma confinement time is limited by diffusion, eq.(5.2.6) but much faster processes
associated with instabilities turn out to be more important. The most dangerous are the MHD
instabilities which can be studied using the methods already applied in the above studies of MHD
equilibria and waves. The instability of a plasma boundary should be of a primary concern since
it may throw the plasma on to the chamber walls.
Consider a plasma supported against gravity by an external magnetic field (see Figure). The
word “gravity” here is a synonym for other possible forces such as centrifugal force due to
the particle motion along the curved magnetic field. Let the plasma sit on the top of a vac-
uum with sufficiently strong magnetic field to withstand the plasma pressure in x-y plane while
B is along z (ignorable coordinate here). Assume that the motion is slow, so it is incom-
pressible, ∇⊥ · ξ = 0. Assuming also the hydrostatic equilibrium d p0 /dx=ρ g, with x point-
ing across the plasma boundary, a surface-wave type perturbation for the plasma displacement
ξ = ξ (x) exp (iky − iω t), and the pressure perturbation p′ = p − p0 = p′ (x) exp (iky − iω t), we
have the following equations for ξ and p′ :

d ξx
+ ikξy = 0
dx

d p′
ρω 2 ξx = (5.4.1)
dx

ρω 2 ξy = ikp′ .
Eliminating ξy , ξx we obtain for p′

d 2 p′
− k2 p′ = 0
dx2
Clearly, we have to choose the exp (−|k|x) solution decaying away from the boundary on the
plasma side, that is

92
 Physics 218A

d p′
= − |k| p′ (5.4.2)
dx
In a situation still close to the equilibrium, the gravity force should be compensated by the
pressure gradient (hydrostatic approximation)

p′ = ρ gξx (5.4.3)
Substituting the last two equations into (5.4.1) we obtain the dispersion relation

ω 2 = − |k| g
This result differs from the well known expression for the long gravitational water waves by
only its sign, thus making the equilibrium aperiodically unstable rather than neutrally stable.
This is not surprising as the heavy fluid (plasma) is on the top of the light fluid (magnetic field),
precisely opposite to the situation with the gravitational waves (one may switch between these
cases by replacing g → −g).
There is no magnetic stabilization in the above example since the field lines move with the
plasma independent of z (field direction) by the flux freezing condition. These are the flute-type
perturbations that are most dangerous for the confinement as they result in a simple interchange
of magnetic tubes with and without plasma. To understand in what type of configuration the
magnetic field has a stabilizing effect, consider the case when magnetic field lines are limited
by the conducting walls (see Figure). Even more interesting is the closed magnetic field con-
figuration that can also be modeled by this simple setting. In a toroidal geometry, a plasma
displacement should result in magnetic line stretching so that magnetic tension will come into
play.
If the ends of magnetic lines are fixed at the walls (or lines are closed), the field bending results
from the plasma displacement and creates the force B2 /4π R, eq.(5.2.8). Here R ∼ L2 /2δ x, L
is the length of the line and δ x is its midpoint lateral displacement, which we may replace
here by ξx . According to eq.(5.4.1,5.4.2,5.4.3), the destabilizing force is kρ gξx . Therefore, the
stabilization is achieved if

B2i + B2e
> gρ k
2π L2
where Bi,e are the internal and external fields and we have added the respective magnetic forces
as they act in the same direction. The bad news about this result is that for sufficiently short
scale perturbations the equilibrium unstable. However, the above sharp boundary approximation
is strictly applied when kδ ≪ 1, so that k cannot be chosen larger than δ −1 where δ is the width
of the plasma boundary. Therefore, magnetic pressure B2 /8π & gρ L2 /δ should stabilize the
plasma boundary against Raleigh-Taylor instability considered above.
As we mentioned the role of g may be played by other forces so that a convex plasma surface
may become unstable due to the curvature and gradient drifts. To derive the relevant stability
criterion we substitute

93
 Physics 218A

 
1 2 1 2
g→ v + v ,
R k 2 ⊥
(see, eq.[4.1.9]). Bearing also the previous result in mind we obtain the following stabilization
criterion

1 B2i L2
= >
β 8π n0 T δR
where L is the effective length of the field line, R is its curvature. Note that other phenomena
can also be modeled by an effective gravity, e.g. when a plasma moves with acceleration under
a rapid compression in inertial confinement experiments or in ϑ - pinches.

5.4.1. Energy prin iple of stability

As we have already seen the most stable should be configurations where magnetic field is in-
creasing outside the plasma. However, in the promising toroidal systems it is impossible to
create a field that would grow in all directions outside a given toroidal surface. In a tokamak,
for example, the field increases on the inner side of the torus and decreases on the outer. An
important parameter is β , so in a high β case, plasma may locally leak from the torus meeting
progressively weaker field so that magnetic tension and magnetic pressure forces are insufficient
to stop the plasma diamagnetic expansion. In a β ≪ 1 case such expansion is much easier to
stabilize as the strong magnetic field cannot be deformed significantly.
To predict whether a magnetic tube with the enclosed dense plasma can be displaced outwards
thus setting on the interchange instability, let us write the volume of a tube with the crossection
∆S as follows
ˆ ˆ
V = ∆Sdl = Φ dl/B

where the flux δ Φ = B∆S = const. The plasma in the tube will do a positive pdV work on its
environment if V increases while the tube moves. Note´ that a gas normally tends to increase
its volume. Furthermore, we may consider pU = −p dl/B as the tube potential energy so the
tube will move only if the energy is not at its minimum. Therefore, the stability condition can
be written as
ˆ
δ dl/B < 0.

The above variation of the functional is to be taken between adjacent field lines
¸ near the plasma
surface. In the case of the closed field line configuration we define U = − dl/B. The above
criterion is consistent with the plasma occupying the minimum field domain.
By analogy with the stratified fluid under gravity, we deduce that the necessary condition for
the equilibrium is p = p (U ) (the surfaces of constant U coincides with isobars). The convective
stability of a tube may then be examined using the same analogy (see e.g. [9]). If the tube rises
from U to U + δ U without change in magnetic field then δ V /V = δ U /U , whereas the adiabatic
pressure change is d p/p = −γδ U /U (recall p/ρ γ = const). The pressure of the environment is

94
 Physics 218A

p (U + δ U ) = p + (d p/dU ) δ U . If the pressure inside the tube at its new position is lower than
that of the environment, the tube will continue to rise. Therefore, the stability condition is

dp γp
<
dU U
Again, similarly to the atmosphere stability condition, where the air temperature may decrease
with height, if not faster than the adiabatic lapse rate g/c p , the plasma pressure may also rise
with U , but not too fast.
A more accurate treatment of an equilibrium stability is based on a variational principle. It
is very useful when a complicated geometry and boundary conditions make the normal mode
treatment of Sec.5.3 inapplicable and when only the stability of a given configuration needs to
be examined regardless of the growth rate or mode structure. We start with an equation for
small displacements, quite similar to eq.(5.3.4) in which the variations of p0 (r) and B0 (r) are
included

∂ 2ξ 1 1
ρ0 = ∇ (ξ ∇p0 + γ p0 ∇ξ ) + (∇ × B0 ) × ∇ × (ξ × B0 ) + ∇ × ∇ × (ξ × B0 ) . (5.4.4)
∂t 2 4π 4π
Rewriting this symbolically as

∂ 2ξ
ρ0 = −K ξ (5.4.5)
∂ t2
and ignoring for a moment the differential character of operator K, we may consider K as the
elasticity constant in the restoring force −K ξ for the motion of a point mass ρ0 . The above equa-
tion should be supplemented by boundary conditions which is essential for the self-adjointness
of K in a realistic geometry. ´ The self-adjointness, in turn, allows one to derive eq.(5.4.5) from
the variational principle δ Ldt = 0, where Lagrangian L = T − W with T being the kinetic
energy
ˆ
1 2
T= ρ0 ξ̇ dr
2
Vi

and
ˆ 
1 1
W= γ p0 (∇ · ξ )2 + ξ · ∇p0 ∇ · ξ + [∇ × (ξ × B0 )]2
2 4π
Vi

1
− ξ × (∇ × B0 ) · ∇ × (ξ × B0 ) dr +We (5.4.6)
4π
being the potential energy. The last term We denotes the contribution of the vacuum field part
and boundary conditions that actually ensure the self-adjointness of K, but the exact form of We
is not essential for what follows. Variation of the kinetic part of Lagrangian
ˆ ˆ
δ T dt = − ρ0 ξ̈ δ ξ drdt

95
 Physics 218A

while the variation of its potential potential part yields

ˆ ˆ
δ W dt = K ξ δ ξ drdt

so that eq.(5.4.4) indeed follows, as usual by virtue of the δ ξ arbitrariness. More importantly,
using the mechanics analogy we conclude, that if the potential energy of the system (5.4.6)
is positive for an arbitrary ξ (r), (W > 0) the system is stable. Furthermore, on letting ξ ∝
exp (−iω t), from eq.(5.4.5) we obtain

ξ K ξ dr
´
ω =´
2
ρ0 ξ 2 dr
so that ω 2 is real and it is simply related to the norm of K. In particular, if K is a positive operator,
the system is stable.

∗∗∗

96
 Bibliography

[1] A. I. Akhiezer, Plasma electrodynamics - Vol.1: Linear theory; Vol.2: Non-linear theory
and fluctuations, 1975. I, 1.5, 1.5

[2] R. Z. Sagdeev and A. A. Galeev, Nonlinear Plasma Theory, W.A. Benjamin Inc. New York,
New York, 1969. I, 3.5

[3] L. D. Landau and E. M. Lifshits, Mechanics, Butterworth-Heinemann, Oxford, 1996. I, 4

[4] R. Goldston and P. Rutherford, Introduction to Plasma Physics, Plasma Physics Series,
Taylor & Francis, 1995. I, 4

[5] L. D. Landau and E. M. Lifsic, Electrodynamics of continuous media, Elsevier [u.a.],

Amsterdam [u.a.], 2009. I, 4

[6] R. M. Kulsrud, Plasma physics for astrophysics, 2005. I, 4

[7] L. Pitaevskii and E. Lifshitz, Physical Kinetics, Number v. 10 in Course of theoretical

physics, Elsevier Science, 1981. 1.4

[8] V. I. Arnold, Mathematical methods of classical mechanics, 1978. 1

[9] L. D. Landau and E. M. Lifshitz, Fluid Mechanics, Pergamon Press, 1987. 5.4.1

97

View publication stats