Electromagnetic waves in plasmas and

conductors

Andrew M. Steane
Exeter College and Department of Atomic and Laser Physics, University of Oxford.

November 11, 2024

Abstract

This note describes electromagnetic waves in plasmas and conductors, at second year under-
graduate level. The treatment of waves in a dielectric medium also follows (it is simpler).

Contents

1 Preliminaries 2

1.1 Mathematical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Wave equation 5

3 The ohmic conductor 6

3.1 Good conductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.2 Any conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.3 The magnetic field in a conductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.4 Reflection from a good conductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.5 Permittivity of a conductor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.6 Field energy and Poynting vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.7 Anisotropic conductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1
4 The neutral plasma 17

4.1 Transverse waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.2 Longitudinal waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.3 Further remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.4 Postscript: treatment via polarization and permittivity . . . . . . . . . . . . . . . . . . 24

4.5 Energy flow in a plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.5.1 Energy treatment via polarization . . . . . . . . . . . . . . . . . . . . . . . . . 27

5 Waves in a dielectric 28

1 Preliminaries

Our starting point is Maxwell’s equations:1

∇ · E = ρ/0 M1
∇·B=0 M2
∂B
∇×E = − M3
∂t
∂E
∇×B = µ0 j + 0 µ0 M4
∂t
where ρ is the total charge density at any given point, j is the total current density and E, B are
the electric and magnetic fields. In classical electromagnetism (which is the setting of the whole our
discussion) these equations always hold everywhere.

In order to complete the theory of classical electromagnetism, we also need the Lorentz force equation,
telling how the fields influence the motion of charged particles:

f = q(E = v × B). (1)

Current and current density. The quantity ρ is a charge per unit volume; the quantity j is a current
per unit area. A current density is not quite the same as a current (it has different physical dimensions
for example) but obviously they are closely related. It is common practice in field theory to use the
short name ‘current’ when referring to j. We shall adopt this practice. The physicist reader is expected
to understand that this is a short-hand and j is a current per unit area in fact.
1 We adopt SI units throughout.

2
There is a related set of equations which we will not need, but we shall mention them briefly at the
end:
∇ · D = ρf M1a
∇ · B̄ = 0 M2a
∂ B̄
∇×Ē = − M3a
∂t
∂D
∇×H = jc + M4a
∂t
where
D = 0 Ē + P̄ (2)
B̄
H = − M̄. (3)
µ0
In these equations, P is polarization (defined as electric dipole moment per unit volume) and M is
magnetization (defined as magnetic dipole moment per unit volume), ρf is free charge density and jc
is conduction current density, and the bar indicates a spatial average over a region large compared
to the atomic structure of the local material medium such as a solid or a gas, but small compared to
the wavelength of any electromagnetic waves under discussion. It practice one often takes this spatial
average for granted and then the bar is not used; we shall adopt this practice.

The derivation of (M1a)–(M4a) from (M1)–(M4) is not included here. The essence of the derivation
is to work out how ∇ · P relates to the charge, and how ∇×M relates to the current. We have
ρ = ρf + ρb , j = jc + jb (4)
where ρb and jb (the ‘bound’ charge and current) are those parts of the charge and current that are
associated with dipoles as opposed to charges that can move over longer distances.

The relationship between electric field and polarization is sometimes expressed in terms of a dimen-
sionless quantity called susceptibility, and similarly for the relationship between magnetization of
magnetic field. The standard definitions are (in SI units)
P = 0 χe E, M = χm H. (5)
where χe is the electric susceptibility and χm is the magnetic susceptibility. Note: these susceptibilities
are introduced merely for convenience in gaining physical insight and simplifying some mathematical
expressions. They do not introduce any physical concept or quantity that is not already expressible
in terms of E, B, P, M.

A final piece of notation is the notion of relative permittivity r and relative permeability µr . These
are not always applicable, but they are applicable when the polarization and magnetization are in the
same direction as the relevant fields, such that D is in the same direction as E and H is in the same
direction as B (note, we are dropping the bars now). In this situation we define r and µr through
D = 0 r E, B = µ0 µr H. (6)

3
One then finds

r = 1 + χe , µr = 1 + χm . (7)

One may also invoke tensor versions of these quantities in order to treat anisotropic media.

1.1 Mathematical results

The following are useful mathematical results which the reader is encouraged to first prove (e.g. by
writing out components) and then commit to memory:

If

F = F0 ei(k·r−ωt) (8)

where F is any field, and F0 , k and ω are all constants (with no dependence on either position or
time), then

∂F
= −iωF (9)
∂t
∇·F = ik · F (10)
∇×F = ik × F (11)
2 2
∇ F = −k F. (12)

We will also use that, for any vector field A,

∇×∇×A = ∇(∇ · A) − ∇2 A (13)

(this can be memorized by committing to memory the phrase ‘curl curl equals grad div minus del-
squared’).

The reader should note that whereas ∇ · A yields a scalar field, the quantity ∇2 A is a vector field
whose components are
  ∂ 2 Ax ∂ 2 Ax ∂ 2 Ax


∇ 2 Ax ∂x2 + ∂y 2 + ∂z 2
 ∇ 2 Ay  =  ∂ 2 Ay ∂ 2 Ay ∂ 2 Ay
+ + . (14)

 ∂x2 ∂y 2 ∂z 2
∇ 2 Az ∂ 2 Az
+ ∂ 2 Az
+ ∂ 2 Az
∂x2 ∂y 2 ∂z 2

Written out in full like this, it seems a bit of a mouthful, but in practice we often don’t need to resort
to component form.

4
2 Wave equation

Let’s take the curl of (M4), and immediately use (13) on the left hand side:
∂E
∇(∇ · B) − ∇2 B = µ0 ∇×j + 0 µ0 ∇× . (15)
∂t
The div B term is zero by (M2), and we can reverse the order of partial differentiation in the final
term, giving (∂/∂t)(∇×E) which is −∂ 2 B/∂t2 using (M3). Hence we have

∂2B
0 µ0 − ∇2 B = µ0 ∇×j. (16)
∂t2

We will comment on this in a moment. Before doing so, let’s take a similar approach to (M3). We
take the curl, and reverse the order of partial differentiation on the right, yielding
∂
∇(∇ · E) − ∇2 E = − ∇×B (17)
∂t
so, by bringing in (M1) and (M4):

∂2E ∂j
0 µ0 − ∇2 E = −∇(ρ/0 ) − µ0 . (18)
∂t2 ∂t

Equations (16) and (18) are important for several reasons:

1. They always apply: we have made no special assumptions about the conditions (amounts of
charge and current etc.)
2. They have the form of wave equations in which B or E is the field under discussion, and the
right hand side serves as source.
3. In the special case of a current that is either uniform or zero, (16) is the ordinary (i.e. source-free)
wave equation for B.
4. In the special case of a charge that is uniform or zero and a current that is constant in time (or
zero), (18) is the ordinary (i.e. source-free) wave equation for E.

The source-free wave equation has plane wave solutions. These are solutions of the form

E = E0 ei(k·r−ωt) (19)
i(k·r−ωt)
B = B0 e (20)

where E0 , B0 , k and ω are constants. E0 is the amplitude of the wave of electric field, k is its wave-
vector and ω is its angular frequency. B0 is the amplitude of the wave of magnetic field, k is its

5
wave-vector and ω is its angular frequency. By substituting the solutions into the wave equation, one
finds that the phase velocity (ω/k) is given by
ω 1
=√ . (21)
k 0 µ0

This is the speed of light in free space, c.

In the rest of this note we discuss two situations where there are sources (currents and/or charges)
related to the waves in such a way that plane-wave solutions can still be found.

3 The ohmic conductor

A conductor such as a metal is said to be ohmic when it gives rise to behaviour satisfying Ohm’s
law: V = IR. If we apply this law to a small region of such a material we shall find that the current
density is proportional to the electric field:

j = σE. (22)

The proportionality constant σ is called the conductivity. (Proof: consider a small cylinder of length L
and cross-section A, then when an electric field E is in the directed along the cylinder, the resistance
is L/(σA) and the voltage is V = EL, so Ohm’s law gives I = (EL)(σA/L) = EσA so the current
density is j = I/A = σE.)

We shall now consider electromagnetic waves in the case where

j = σE, ∇ρ = 0. (23)

In this case (18) gives

∂2E ∂E
0 µ0 − ∇2 E = −µ0 σ . (24)
∂t2 ∂t
We now propose a trial solution in the form of a plane wave (19) and we make free use of the
mathematical results (9)–(12), thus obtaining

−0 µ0 ω 2 E + k 2 E = iωµ0 σE. (25)

Now notice that owing to the simple behaviour of planes waves, the field itself divides out of the
equation, leaving just a relationship between k and ω. This is extremely useful and therefore important!
Gathering terms related to ω on the right, and using c2 = 1/0 µ0 , we have

Dispersion relation for a conductor

ω2
k2 = + iωµ0 σ (26)
c2

6
In wave theory in general, the term dispersion relation refers to the relationship between k and ω.
It is one of the most useful pieces of information in order to understand what the waves are doing.
The above dispersion relation applies whenever the situation is an ohmic medium (j = σE) with no
or uniform charge (∇ρ = 0), and we assumed that the conductivity σ is constant in time. We did not
need to assume that σ is uniform in space, and it can have any dependence on ω. The mathematical
method also involves that we allow that k can be a complex number.

3.1 Good conductor

The interaction between electromagnetic waves and metals can be treated to good approximation by
taking σ to be a real number, as long as the wavelength is large compared to the interatomic spacing
and the frequency lies below that of atomic resonances. By this method we can treat radio waves
accurately and even visible light roughly, but not ultra-violet radiation or X rays.

For simplicity we first examine two limiting cases, where the dispersion relation simplifies:

insulator ωµ0 σ  ω 2 /c2 , k = ω/c (27)

2 2
p
good conductor ωµ0 σ  ω /c , k = iωµ0 σ (28)

These are called ‘poor’ and ‘good’ conductor because the condition is on the conductivity: the ‘good’
conductor limit is
ω
σ = 0 ω. (29)
µ0 c2
From the simplified dispersion relations (27), (28) we deduce that the poor conductor leads to a
behaviour just like waves in free space (but see comments after (49)), and the good conductor gives

k = α(1 + i) (30)

where
p
α= ωµ0 σ/2 (31)
√ √
and we used that i = (1 + i)/ 2. (Notice that we are here allowing k to be a complex number, not
just a real number; you should confirm that all the equations and methods we have employed do not
require any assumption that k is real.) Substituting this solution into the plane wave (19) and taking
the z axis in the direction of k, we have

E = E0 ei(α(1+i)z−ωt)
= E0 e−αz ei(αz−ωt) (32)

The physical electric field will be given by the real part of this solution:

e.m. wave in a good conductor

Ephys = E0 e−αz cos(αz − ωt) (33)

7
 1

0.5
E

0 1 2 3 4 5 6
z/δ

Figure 1: Left: the function E = e−z/δ cos(z/δ), describing a wave propagating into a good conductor.
The dashed curves show ±e−z/δ . The wave amplitude decreases by a factor exp(−π/2) ' 0.208 each
quarter-wavelength. Right: the animation shows a wave incident from the left and mostly reflected,
for the case σ = 1000 ω. Blue full line = Ex , red dashed line = By .

where we assumed E0 is real. The result is plotted in Fig. 1.

One should note that (33) is not the general solution of the differential equation. It is one of a whole
class of solutions, and because we have a linear differential equation, sums of such solutions also solve
the equation. A wave propagating in the negative z direction, for example, is expressed by

Ephys = E0 eαz cos(−αz − ωt). (34)

The presence of eαz here has the appearance of the solution which is exponentially ‘blowing up’, but
the way to read the expression is to see that this wave is propagating in the other direction (see the
cos(−αz − ωt) term) so it is decaying as it goes, just like (33). Neither solution can be valid at all z,
but either or both may be valid in some spatial region; an example will follow.

Let’s return to the solution (33) and Fig. 1. We find that as it ‘tries’ to propagate into the conductor,
the wave is rapidly suppressed. Its amplitude decays on the length scale 1/α; this is called the

skin depth
r
1 2
δ= = (35)
α ωµ0 σ

This is the formula in the case of a good conductor; a more general formula is derived in the next
section.

Now let’s recall that our conductor is ohmic, so the behaviour of the electric field also tells us the

8
behaviour of the current, through j = σE. In a typical case the electric field is directed along a wire,
and therefore so is j. But we see that if δ is small compared to the radius of the wire then E is
confined to a region close to the surface of the wire, and therefore so is j. Since δ falls with frequency
(assuming σ does not depend strongly on frequency) this is liable to happen at higher frequencies,
and it leads to an increase in resistance since the current is confined to only part of the cross-section
of the wire, rather than all of it. This is why wires are often made multi-stranded, in order to increase
the surface area and thus reduce the resistance. The skin effect is an important consideration in r.f.
circuit design.

Example.
The conductivity of aluminium is 3.8 × 107 (Ωm)−1 and may be taken to be independent
of frequency. Find the resistance of an aluminium wire of length ` = 1 m and radius
a = 0.2 mm, at 50 Hz and at 50 MHz.
Solution.
First we find the ratio σ/0 ω. At 50 Hz it is about 1016 and at 50 MHz it is about 1010 .
Therefore in both cases we may employ the good conductor approximation. Next we obtain
the skin depth using (35). At 50 Hz it is 1.2 cm and at 50 MHz it is 11.5 µm. Therefore at
50 Hz the wire conducts throughout its width and the resistance is R = `/(πr2 σ) = 0.21 Ω.
At 50 MHz, on the other hand, the skin depth is small so the current only flows in the
region near the surface. Since δ  a we can ignore the effect of the curvature of the
surface, treating it as a plane of width 2πa (and length `). The current is
Z a
I = 2πa σjdz
Z0 ∞
' 2πa σE0 e−αz ei(αz−ωt) dz
0
σE0
= 2πa (1 + i)e−iωt . (36)
2α
Taking the voltage to be `E0 we find the complex impedance Z = V /I is

`
Z= (1 − i) (37)
σ2πaδ
where we used δ = 1/α. The impedance therefore has a resistive and an inductive part.2
They are in parallel physically, but the result is equivalent to an arrangement in series
with resistive part R = `/(σ2πaδ) = 1.8 Ω.

A final remark. The alert reader will realise that we started with a quadratic dispersion relation but
we only discussed one solution. For a quadratic equation there should be two solutions, and indeed
there are. The full story is

k = ±α(1 + i). (38)

2 When the voltage goes as exp(−iωt) a capacitor has impedance i/ωC and an inductor −iωL.

9
The second solution yields

E = E0 eαz ei(−αz−ωt) . (39)

We already mentioned the fact that there are solutions of this form too, by bringing in the linearity
of the differential equation and the fact that k may be in the opposite direction to z.

3.2 Any conductivity

Next we treat a general conductor, retaining all terms in the dispersion relation (26). The convenient
way to handle the mathematics is to express k in terms of its real and imaginary parts:

k = α + iβ (40)

Then we have

k 2 = α2 − β 2 + 2iαβ. (41)

Substituting this into the dispersion relation gives the simultaneous equations

ω2
α2 − β 2 = (42)
c2
2αβ = ωµ0 σ (43)

Taking the second equation (β = ωµ0 σ/2α) and substituting it into the first one obtains

α4 − α2 ω 2 /c2 − (ωµ0 σ/2)2 = 0. (44)

This is a quadratic equation for α2 whose solution is

 s 
2
 2
ω σ
α 2 = 2 1 + 1 +  (45)
2c 0 ω

where we must take the + sign because α2 must be positive (we have assumed α to be real by
definition). Substituting this into (42) gives
 s 
2
 2
ω σ
β 2 = 2 −1 + 1 + . (46)
2c 0 ω

Also, substituting α into (43) yields

µ0 σc
β=√ p √ (47)
2 1 + 1 + s2
where s = σ/0 ω. This version is more useful for obtaining the skin depth, as we now show.

10
 freq. aluminium copper silver sea drinking deionised
water water pure water
σ (Ω−1 m−1 ) 3.78 × 107 5.97 × 107 6.29 × 107 5 0.01 5.6 × 10−6
50 Hz 0.0116 0.0092 0.0090 32 710 30200
1 kHz 0.0026 0.0021 0.0020 7 160 6800
1 MHz 82 × 10−6 65 × 10−6 63 × 10−6 0.23 5 960
1 GHz 2.6 × 10−6 2.1 × 10−6 2.0 × 10−6 0.007 0.5 960
6 × 1014 Hz 3.3 × 10−9 2.7 × 10−9 2.6 × 10−9

Table 1: Skin depth in metres for various materials and frequencies. The values for water are rough,
because the conductivity is a function of temperature, frequency and concentration of impurities. The
last row applies to visible light of wavelength 500 nm.

The plane wave form is

E = E0 ei(kz−ωt) = E0 e−βz ei(αz−ωt) (48)
so we deduce that the general formula for the skin depth is

skin depth
√ q
1 2 p
δ= = 1 + 1 + (σ/0 ω)2 (49)
β µ0 σc

The good conductor limit of this formula reproduces our previous result (35). The poor conductor
limit (σ  0 ω) gives
2
δ→ . (50)
µ0 σc
This seems to negate our previous observation from (27) that the situation is ‘just like in free space’.
That observation was not entirely wrong, however, since as σ → 0 we get δ → ∞. The interesting
further information to be seen in (50) is that even before σ reaches zero, the skin depth becomes
independent of frequency in poor conductors. This is observed at high frequencies in especially poor
conductors such as semi-conductors.

Some example values of skin depth are displayed in table 1. For metals the depth is about 1 cm at the
mains frequency of 50 Hz, and a few microns at 1 GHz. For visible light the skin depth in a metal is a
few nanometres, which is a few tens of atomic spacings. This implies that the continuous model of the
material remains reasonably reliable, but at optical frequencies the conductivity of metals depends on
frequency and becomes complex.

The conductivity of water depends on frequency above a few Hz and also on temperature; the table
shows the order of magnitude of the values. One implication is that radio communication with
submarines is difficult. (The behaviour of water at optical frequencies is not adequately modelled by
simple conductor theory. In fact water transmits quite well at optical frequencies, and considerably
less well at both infra-red and ultra-violet.)

11
3.3 The magnetic field in a conductor

So far we discussed the electric field. To treat the magnetic field, recall the wave equation (16) and
adopt the plane wave (20) as a trial solution. Using also j = σE on the right hand side, one finds

−0 µ0 ω 2 B + k 2 B = iµ0 σk × E. (51)

The fields also satisfy (M3) which gives

ik × E = iωB (52)

(obviously the factor i divides out of this equation but I left it in at this stage in order to make it
clear how the result was obtained.) After substituting this into (51) we obtain the same dispersion
relation as before, (26), therefore at any given ω the two fields have the same k (as we should expect).

Eqn. (52) shows that B is perpendicular to both k and E and we already know k is perpendicular to
E. It follows that the (possibly complex) amplitudes of the fields are related by

k
B0 = E0 . (53)
ω
We shall now adopt the good conductor approximation (the reader will be able to go back and deduce
the more general case should they wish to). In the good conductor limit we have k = (1 + i)α (this is
(52)) where α is given by (31), so we find
r r
1 + i µ0 σ µ0 σ
B0 = √ E0 = eiπ/4 E0 . (54)
2 ω ω

Two observations follow: first, the magnetic field is out of phase with the electric field by π/4 (B lags
E); secondly: the magnetic field is ‘large’. That is, the ratio B0 /E0 is much larger than it is for waves
in free space. To express this, examine the dimensionless ratio
r
c|B0 | σ
=  1. (55)
|E0 | 0 ω

(In free space the value would be 1.)

3.4 Reflection from a good conductor

We will find the coefficient of reflectivity for electromagnetic waves incident on a good conductor. For
simplicity we just treat the case of normal incidence on a flat boundary. The calculation is most easily
done by treating the fields E and H. This is because the fields are transverse to the wave vector, so
they are directed along the surface of the boundary in the case of normal incidence, and it is Ek and

12
Hk (as opposed to D and B) that are continuous at the boundary in the absence of surface charge or
current.

In view of the fact that we have been discussing the current j, the reader may wonder whether there
is a surface current. For the moment we shall simply assert that there is not; we shall return to this
point at the end.

Under the assumed conditions, then, the standard reflection calculation applies and one finds that the
amplitude reflection coefficient, for waves passing from medium 1 to medium 2
Z2 − Z1
r= (56)
Z2 + Z1
for the electric field amplitude, where Z1 , Z2 are the wave impedances. For a plane electromagnetic
wave of amplitude E0 propagating in the positive z direction the impedance is defined through

E0 = ZH0 (57)

In free space we have E = cB and B = µ0 H, hence one finds Z = µ0 c. Thus

p
Z1 = µ0 c = µ0 /0 . (58)

In a good conductor (54) gives

p
Z2 = µ0 ω/σ e−iπ/4 (59)
p
(we took the permeability µ = µ0 ; for remarks on this see section 3.6). Note that |Z2 /Z1 | = 0 ω/σ 
1 (consistent with the observation that ‘the magnetic field is large’ as we said above.)

After substituting Z2 /Z1 into (56) we find

−1 + s p
r= where s = e−iπ/4 0 ω/σ (60)
1+s
Using that |s|  1 we can simplify this to

r = −1 + 2s + O(s2 ). (61)

The −1 indicates there is a phase change on reflection. The intensity reflection coefficient is
p
r2 ' (−1 + 2s)(−1 + 2s∗ ) = 1 − 2(s + s∗ ) ' 1 − 80 ω/σ. (62)

For example, for silver at optical frequencies one finds |r2 | ' 0.93. This is consistent with the measured
value. More generally, metals make good reflectors. Together with its small chemical reactivity, this
explains why silver is commonly used as a coating on glass to make mirrors.

But is there a surface current or isn’t there? Now let’s return to the point about surface current which
was postponed above. The more general version of the continuity equation for H is

Hk,2 − Hk,1 = K (63)

13
where K is the current per unit length flowing along the surface (such that the two H fields make a
loop around K). When there is a volume current j the surface current in a thickness δz is jδz and this
tends to zero in the limit δz → 0. This is why we correctly took K = 0 in the above theory of reflection
at a conducting surface. It is all about what we mean by the symbol H2 . We used that symbol to
refer to the magnetic field at a distance within the surface of just one or a few atomic spacings, a
distance small compared to the skin depth. We accounted for the current j = σE by including it in
the calculation of the fields.

3.5 Permittivity of a conductor?

Our discussion of the conductor invoked Maxwell’s equations (M1)–(M4) and we only ever needed
to mention the fields E and B and the current j. There was no mention of either polarization P or
magnetization M and therefore the conductor is correctly treated by taking

r = 1, µr = 1. (64)

That is not to say you cannot also have a medium which both conducts and is also polarizable or
magnetizable, but here we are just dealing with a non-dielectric conductor.

However you often see quoted a different result:

iσ
r = 1 + . (65)
0 ω
This is a very different answer! It is complex and has a magnitude large compared to 1 for a good
conductor! So what is going on?

The two results represent two ways of interpreting the motion of the charges. In the first way we
say the oscillating charges are making a current, which we have accounted for correctly. It is the
whole current and there is nothing further to say; in particular there is no polarization, so P = 0 and
therefore r = 1.

In the second approach, we treat the very same motion of the charges, but we interpret it as an
oscillating polarization. Suppose that particles of charge q are oscillating with displacement x =
x0 e−iωt . This will produce a current density j = nq ẋ = −iωnqx where n is the number density of
the charges. And if for each moving particle there is an equal charge −q that does not move, then
the displacement of any given particle gives rise to an electric dipole moment qx, and therefore the
overall polarization (electric dipole moment per unit volume) is

P = nqx = ij/ω. (66)

Using now j = σE we find

P = nqx = iσE/ω (67)

which implies the susceptibility is χe = iσ/0 ω and eqn (65) follows. In this second approach we must
take jc = 0 because we already accounted for the motion of the charges by including it in P.

14
The lesson is you can use either of (64) or (65) as long as you know what you are doing and define P
in a way consistent with your choice.

3.6 Field energy and Poynting vector

Electromagnetic waves in a conductor provide a nice example of the way energy conservation is treated
in classical electromagnetism. There are three fundamental concepts:
1
field energy density u= (E · D + B · H) (68)
2
Poynting vector (field energy flux) S=E×H (69)
power density p=E·j (70)

For comments on the related quantities (0 E 2 + B 2 /µ0 )/2 and E × B/µ0 see the note Energy in
electromagnetism. For the ordinary conductor there is no magnetisation so µr = 1, and if we treat
the current as entirely conduction (not polarization) current then r = 1 as explained in section 3.5.

The power density p is the rate at which energy is transferred from the fields to the charges, per unit
volume.3 The conservation of energy is expressed by4

Conservation of energy
∂u
+ ∇ · S + E · j = 0. (71)
∂t

To apply these ideas to waves, the first step is to beware of the complex notation! The use of complex
numbers to describe physical quantities is useful when we are concerned with linear combinations such
as a sum of fields, but less so then we are concerned with non-linear terms such as E 2 or EB. This
is because, for any complex numbers z1 and z2 ,

Re (z1 ) Re (z1 ) 6= Re (z1 z2 ) (72)

It follows that if E and H are complex then the Poynting vector is not E × H. Rather, it is

S = Re (E) × Re (H) (73)

Is this a contradiction of (69)? No, but it is a hidden change of notation. What happened is that
in (69) I quoted the standard form of the definition where it is understood that all the quantities in
play are real-number-valued, but when dealing with waves we often use symbols such as E and H to
refer to complex numbers whose real part yield the physical fields, so in that case we must use (73).
Similar considerations apply to power and energy density.
3 To derive this, observe that the work done per unit time on a charge q moving at velocity v is f ·v = q(E+v×B)·v =

qE · v and therefore the rate per unit volume, if there are n such particles per unit volume, is nqE · v = E · j by using
j = nqv.
4 Derived in the above-mentioned note, Energy in electromagnetism.

15
The equation for energy conservation (71) applies at all times and places and correctly tracks energy
movements in detail. In the case of wave motion typically all the quantities u, S and p oscillate at the
wave frequency, and we can in principle calculate the detailed movements of energy within each cycle
and each wavelength (for an example see section 4.5). However we are often mainly interested in the
average over time, and that is all we shall consider here. The following mathematical ‘trick’ will be
useful. In case of oscillating quantities of the form z = z0 e−iωt+φ (such as our plane waves (19),(20))
the time average (here indicated by the notation h· · ·i) can be found from
1
hSi = hRe (E) × Re (H)i = Re (E∗ × H) . (74)
2
1
hui = Re ((E∗ · D + B∗ · H)) (75)
4
1
hpi = Re (E∗ · j) (76)
2
Note, this mathematical ‘trick’ is not self-evident and the reader should either prove it for themselves
or else consult Energy in electromagnetism.

We shall now apply the above equations to the case of waves in a conductor, using the solution for E
and B (eqs (48), (53)) which we repeat here for convenience:
E = E0 e−βz ei(αz−ωt) ,
k α + iβ
B = E = E0 e−βz ei(αz−ωt) . (77)
ω ω
By substituting these into (75), and using (45), (46), we obtain
1 p 
hui = 1 + 1 + (σ/0 ω)2 0 E02 e−2βz . (78)
4
Notice that for a good conductor the field energy is dominated by the magnetic part (and in an
insulator the electric and magnetic parts contribute equally).

Turning now to the Poynting vector and power density, (74),(76) and (77) give
α
hSi = E 2 e−2βz ẑ, (79)
2µ0 ω 0
1
hpi = σE 2 e−2βz . (80)
2 0
By using (43) it is now straightforward to confirm that
h∇ · Si = ∇ · hSi = −hpi. (81)
Therefore the conservation of energy is respected (since the time average of u is constant here). A
good way to grasp the physical interpretation is to write the result in the form
S(z + dz) = S(z) − p (82)
which asserts that “the energy flowing to the plane at z + dz is equal to the energy arriving at z, less
the energy given up to drive the current against the resistance of the matter in the conductor.” By
integrating hpi between z and ∞ you can also confirm that all the field energy arriving at any given
z is eventually transferred to the matter.

16
3.7 Anisotropic conductor

Throughout our discussion of a conductor we have taken Ohm’s law in its simplest form j = σE
where σ is a scalar quantity. This is the form it takes in an isotropic medium (one in the which the
conductivity is the same in all directions). In an anisotropic medium such as a crystalline solid, the
relationship may be
¯E
j = σ̄ (83)

where the notation indicates that σ̄¯ is a tensor: that is, a quantity which can be expressed by a 3 × 3
matrix (and which obeys the mathematical rules for tensors). The treatment of this situation is a more
advanced exercise. If the fields are directed along the principle axes of the tensor, the mathematics is
not much more elaborate, but we shall not investigate this here.

4 The neutral plasma

A neutral plasma is a fluid medium composed of equal densities of positive and negative charge, such
as positive ions and electrons. Such a plasma can be regarded as a fourth state (also called phase) of
matter, in addition to solid, liquid, gas. This state is of especially high significance in astrophysics,
because stars are plasmas and so is much of the interstellar medium.

In the presence of an electromagnetic wave the charges in a plasma will be driven to oscillate at the
frequency of the wave, and the mathematical treatment is almost identical to that for a conductor,
but with a pure-imaginary conductivity, as we now show.

Let’s suppose a plane wave is propagating in a plasma. We treat a small region of the plasma, and
examine the motion of a single charged particle in the first instance. It obeys Newton’s second law:
dv
m = q(E + v × B) (84)
dt
where q is the charge and m the mass of the particle. We assume the magnetic contribution to the
force is negligible in comparison to the electric. Then
dv q
= E0 ei(k·x−ωt) (85)
dt m
For transverse waves (E perpendicular to k) the particle is displaced in a direction orthogonal to the
wave vector and therefore experiences the same field, at any given time, no matter how far displaced
it is. For longitudinal waves (E parallel to k) the displacement is along the direction of k and this
complicates matters. However, if the size of the particle’s motion remains small compared to the
wavelength then this complication goes away: we place the origin of coordinates near the particle and
then k · x  1 and we have
dv q
= E0 e−iωt . (86)
dt m

17
This equation is straightforward; it can be integrated immediately, giving
q
v = E0 e−iωt
−iωm
iq
= E (87)
mω
(and the constant of integration is zero since we assume the particle has no drift velocity in addition
to this oscillating velocity).

Now let’s use the above to make deductions about the plasma as a whole. As long as the plasma is
not too dense, we can apply (87) to each charged particle in it. If there are n particles per unit volume
then we have a net flux of electric charge given by
iq 2
j = nqv = n E. (88)
mω
If there are several species of particle with different masses and charges (as is usually the case) then
we must sum over them. However in a plasma of electrons and ions the current is dominated by the
electrons (since their mass is about 2000 times smaller than that of the ions) so we shall ignore the
other contributions. The most important fact is that (88) has the form j = σE with

Conductivity of plasma

q2 n
σ=i (89)
mω

Now we can treat the propagation of waves in the plasma. We take as starting-point equation (18),
and substitute into it a plane wave solution. This yields
∂j
−0 µ0 ω 2 E + k 2 E = −∇(ρ/0 ) − µ0 (90)
∂t
Now use j = σE and observe that for the plane wave the charge density term can itself be written in
terms of k and E:

∇(ρ/0 ) = ∇(∇ · E) = −k(k · E). (91)

Hence we obtain
ω2
k2 E = E + iωµ0 σE + (k · E)k (92)
c2
This is almost a dispersion relation. It would yield a dispersion relation if we could divide out a factor
of E. The overall form of the equation implies that k must be along the same direction as E unless
k · E = 0. In other words, there are two cases:
ω2
transverse k · E = 0, k2 = + iωµσ (93)
c2
ω2
longitudinal k k E, k 2 = 2 + iωµσ + k 2 (94)
c

18
where in the second case we used that when k is parallel to E we have k · E = kE and k̂ = Ê so
(k · E)k = (kE)k Ê = k 2 E. The E factor then divides out of the whole equation. Upon substituting
σ from (89) we find

plasma dispersion relation

transverse ω2 = k 2 c2 + ωp2 (95)

longitudinal ω = ωp (96)

where

Plasma frequency
s
nq 2
ωp = (97)
0 m

In the following we shall say a bit more about these solutions. The transverse waves are like the
more familiar waves in free space: they have electric and magnetic parts, they carry energy, etc. The
longitudinal waves are a direct property of the plasma: they amount to the fact that the plasma has
a natural oscillation frequency and can oscillate without being driven. The longitudinal waves have
the same frequency irrespective of their wave vector (and therefore zero group velocity).

4.1 Transverse waves

The dispersion relation for transverse waves can usefully be displayed in the form
r
ω ωp2
k= 1− 2 (98)
c ω
This form makes it clear that k is real when ω > ωp and k is pure imaginary when ω < ωp . Hence:
high frequency waves propagate, low frequency waves decay. The phase velocity is
ω c
= (99)
k nr
(this defines the refractive index nr ), so we find the refractive index is
r
ωp2
nr = 1 − 2 . (100)
ω
For the case of propagating waves the refractive index is less than 1 and the phase velocity is greater
than c. This is allowed in special relativity because the individual wavefronts do not themselves convey
information (for a monochromatic wave): their arrival is ‘expected’.

To obtain the group velocity it is easiest to employ (93) which gives

dω
2ω = 2kc2 , (101)
dk

19
 x
E, j

B
k̂
y

transverse

longitudinal

Figure 2: Waves in a plasma. The top diagram shows a plot of the E and B fields and the current j
(at some instant of time) for a transverse wave propagating in the z direction and linearly polarized
along x. The middle and bottom diagrams show the plasma with a transverse and longitudinal wave,
respectively. Shading represents the charge density, arrows indicate the current density. In all cases
the electric field is along the same direction as j, oscillating π/2 radians out of phase with it.

20
hence
dω ω
= c2 (102)
dk k
which is
vgr vph = c2 . (103)
Hence vgr < c when vph > c, and

1/2
vgr = nr c = 1 − ωp2 /ω 2 c. (104)

The transverse waves have k · E = 0 and therefore ∇ · E = 0, which implies ρ = 0. This shows that
they do not disturb the neutrality of the plasma: in these waves the charges move from side to side
without ‘bunching up’ (c.f. Fig. 2). Using (M3) we have also that
k × E = ωB (105)
so we deduce there must be a magnetic wave accompanying the electric wave, with amplitude kE/ω =
nr E/c. It is transverse too (says (M2)).

The solutions at high frequency behave very much like ordinary electromagnetic waves in free space, in
that they are transverse electromagnetic waves which propagate over long distances without decaying.
What do the solutions at low frequency mean? They do not propagate at all, since when k is pure
imaginary we have k = iα for real α, and the form is
E = E0 e−αz e−iωt (106)
(for a z axis aligned with k). This type of behaviour is called an evanescent wave. A good physical
insight is obtained by examining the impedance Z, defined by
E0 = ZH0 = ZB0 /µ0 . (107)

21
By combining this with (105) we find

Z = µ0 ω/k. (108)

Thus when k is pure imaginary, so is Z. Now consider waves incident on a boundary between two
media, one with real Z, the other with imaginary Z. At normal incidence the intensity reflection
coefficient is
2
Z2 − Z1
R = (109)
Z2 + Z1
(Z2 − Z1 ) (Z2∗ − Z1∗ )
= (110)
(Z2 + Z1 ) (Z2∗ + Z1∗ )

But if Z1 = Z1∗ and Z2 = −Z2∗ (that is, Z1 is real and Z2 is imaginary) then this yields R = 1.
In other words, there is perfect reflection. The lesson is that low-frequency electromagnetic waves,
incident from some other medium, are reflected by a plasma.

In the case ω > ωp , when waves can in principle propagate in the plasma, one has n < 1 and therefore
there will also be total reflection for waves incident at angles to the normal above a critical angle
which can be determined from Snell’s law:

ni sin θi = nr sin θr . (111)

sin θr cannot exceed 1 so this equation has no solution (i.e. no refracted wave, therefore complete
reflection) when sin θi > (nr /ni ).

There is a plasma over our heads, in the region of Earth’s atmosphere called the ionosphere—see Fig.
3. Radio signals can be bounced off the ionosphere, and this is used for long-distance communication
without the need for wires or satellites. An easy way to remember the fact that it is the high frequencies
that can be transmitted is to recall that when we look up at night, we can see the stars. Their visible
radiation can propagate through the ionosphere, whereas radio signals (of frequency below about 20
MHz) can not.

Another system that can be treated by plasma theory to good approximation is a metal at very high
frequencies (the X ray region of the spectrum). At the highest frequencies the response of the metal is
not ohmic; one can ignore the scattering processes associated with ohmic conduction and just assume
the electrons respond directly to the fields by undamped driven oscillations, just as for a plasma.

4.2 Longitudinal waves

The longitudinal waves are also known as plasma oscillations or Langmuir waves. They are not like
electromagnetic waves in various respects. For one thing, they are not magnetic, since when k is
parallel to E we have k × E = 0 so (M3) informs us that ∂B/∂t = 0: the magnetic field (if there is
one) is not oscillating. For this reason these waves are sometimes called ‘electrostatic waves’ (a name
I do not like because they are not static). The phase velocity ω/k can have any value and the group

22
 1,000
ionosphere
800
height / km

Earth 600
day
night
R 400

200

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
n/(1012 m−3 ) ·1012

Figure 3: Left: long-distance radio communication using reflection from the ionosphere. A signal sent
horizontally will hit the layer at height 400 km at an angle of 70◦ to the normal, and therefore will
be reflected if its frequency is below about 3ωp . Right: the electron density as a function of height
(there are considerable variations with weather, season and especially sunspot activity). The plasma
frequency for n = 1.5 × 1012 m−3 is (ωp /2π) = 11 MHz.

23
velocity is zero. The Poynting vector is also zero: there is no propagation of energy (there is just
exchange of between the current and the field at any given location.)

The longitudinal waves are associated with non-zero values of the charge density. In other words,
they disturb the charges in the plasma in such a way that the net charge density is positive in some
places and negative in others (c.f. Fig. 2). This case did not arise in our treatment of conductors
because there we assumed ∇ρ = 0 throughout. If we now allow for the possibility of non-zero charge
variations in a conductor we shall obtain (92)–(94) but for a medium in which σ is real-valued. The
only solution in the longitudinal case is then ω = 0; in other words longitudinal waves do not arise in
an ohmic conductor.

4.3 Further remarks

Our treatment of a plasma assumed that the particles in the plasma were not moving significantly apart
from the oscillations in response to the oscillating field. We also ignored the magnetic contribution to
the force. In practice the motion of the plasma will introduce pressure variations, and these will add
to the restoring forces. The approximation of neglecting the pressure forces is called the ‘cold plasma’
approximation. For a ‘hot’ plasma, where pressure gradients are non-negligible, the transverse waves
are unchanged and the longitudinal waves have a dispersion relation

ω 2 = ωp2 + 3k 2 vth
2
(112)

where the thermal velocity vth is defined as

q
vth = kB T /m (113)

The plasma is ‘hot’ when kvth is not negligible compared with ωp .

For the transverse waves there is a magnetic field but we ignored its contribution to the force. Let’s
confirm that this is legitimate. The ratio of magnetic to electric force is at most vB/E for a given
charged particle moving at speed v and (105) gives B = nr E/c so we find that the ratio of magnetic
to electric force is at most nr v/c. This is small compared to 1 for non-relativistic motion.

If there is an applied static magnetic field whose size is sufficient to produce non-negligible forces
then the situation is significantly more complex. In this case there are several types of wave, having
different dispersion relations. Plasmas in strong fields occur in tokamak fusion reactors. Their full
description is a highly complex, and highly significant, area of physics in its own right.

4.4 Postscript: treatment via polarization and permittivity

We treated the plasma using the concept of conductivity, and in consequence we never needed the
concepts of polarization, susceptibility and relative permittivity. An alternative approach is to note
that the charged particles form dipoles when they move. There is an equal charge density of either

24
sign in the undisturbed plasma, so if a given electron is displaced by x, with the nearby positive ions
unmoved, then an electric dipole of size qx is produced. The polarization (= dipole moment per unit
volume) of the plasma is therefore

P = nqx . (114)

Starting from the equation of motion (86) we integrate twice, obtaining x = −q/(mω 2 )E (with
constants of integration equal to zero) and therefore

nq 2
P=− E. (115)
mω 2
Hence the susceptibility is χe = −nq 2 /(0 mω 2 ) and the relative permittivity is

q2 n ωp2
r = 1 − = 1 − . (116)
0 mω 2 ω2
Meanwhile there is no magnetization. This is because the currents in the plasma are in straight lines:
they do not form loops and hence they do not make magnetic dipoles. Therefore µr = 1.

We can now obtain the refractive index from

c 1
vph = =√ (117)
nr 0 r µ0 µr
so
√ q
nr = r µr = 1 − ωp2 /ω 2 (118)

as before.

This method via polarization is equally good as the one via conductivity, but one should be careful
not to muddle them. When using conductivity the current j = σE is the total current: the quantity
that appears in (M4). It is not the quantity jc that appears in (M4a). Indeed we never invoked any
of (M1a)–(M4a) when we did the calculation the first way.

In the treatment via polarization, by mentioning quantities such as relative permittivity we are im-
plicitly invoking the equations (M1a)–(M4a). The whole of the current in the plasma is now ascribed
to oscillations of dipoles so we must take jc = 0 in (M4a).

When ω = ωp we have r = 0. In the wave equation for E it is this vanishing of the permittivity that
makes it possible for there to be longitudinal as well as transverse wave solutions. A zero permittivity
does not occur in dielectrics or ohmic conductors, so they don’t exhibit longitudinal waves.

4.5 Energy flow in a plasma

We will treat energy flow in a plasma in the case of transverse waves.

25
In the case ω > ωp the field energy density is
 
1 1
u = 0 (Re (E))2 + (Re (B))2
2 µ0
0
= (2 − ωp /ω )E0 cos2 (kz − ωt)
2 2 2
(119)
2
where we used B = kE/ω = (1 − ωp2 /ω 2 )1/2 E/c. The Poynting vector is

S = Re (E) × Re (H)
q
= 0 c 1 − ωp2 /ω 2 E02 cos2 (kz − ωt)ẑ (120)

The power density is

p = Re (E) · Re (j)
ωp2 0 2
= − E0 cos(kz − ωt) sin(kz − ωt). (121)
ω
Therefore
∂u
= 0 ω(2 − ωp2 /ω 2 )E02 cos(kz − ωt) sin(kz − ωt )
∂t
∇ · S = −20 ω(1 − ωp2 /ω 2 )E02 cos(kz − ωt) sin(kz − ωt ) (122)

and one finds (∂u/∂t)+∇·S+p = 0 as required for energy conversation (recall the continuity equation
(71)).

Let’s consider now the time averages of the above quantities. For the power density the time average
is hpi = 0. This shows that whereas there is an oscillation in which energy passes between field and
matter, there is no net energy transfer after averaging over a cycle, with the result that the waves
propagate without losing energy. This is confirmed by
1 q
hSi = 1 − ωp2 /ω 2 E02 ẑ (123)
2µ0 c
which is non-zero and independent of z.

The above applies to propagating waves, with ω > ωp . Next we treat the case of evanescent waves,
with ω < ωp . We have B = kE/ω (from (M3) as usual) which gives

B = i(α/ω)E0 e−αz e−iωt (124)

where
q
α = (ω/c) ωp2 /ω 2 − 1. (125)

26
Therefore
1
0 E02 e−2αz cos2 (ωt) + (αc/ω)2 sin2 (ωt) ,


u = (126)
2
1 αc2 2 −2αz
S = 0 E e sin(2ωt), (127)
2 ω 0
1 ωp2 2 −2αz
p = 0 E e sin(2ωt). (128)
2 ω 0
In this case we find hpi = 0 as before, but now hSi = 0 as well. The evanescent waves do not transport
energy on average; in this respect they are like standing waves. This is connected to the fact that the
expressions for E and B factorize into a part dependent on position and a part dependent on time.

4.5.1 Energy treatment via polarization

In the discussion of energy presented in (119)–(128) we have treated the current in the plasma in
terms of conductivity. But we can also treat it in terms of polarization, as presented in section 4.4.
In this case the physical behaviour is unchanged but it is expressed in terms of different quantities.

We have D = 0 r E with, r = 1 − ωp2 /ω 2 (eqn (116)). Hence

u = (1/2) (Re (E) · Re (D) + Re (B) · Re (H))

2
= 0 r Re (E)
= 0 r E02 cos2 (kz − ωt). (129)

Both magnetic and electric energy have been included; they contribute equally.

The Poynting vector is unchanged.

For the power density, since we have attributed the whole of the current to the oscillating polarization,
we must take the conduction current to be jc = 0 and therefore the power density is

p = 0. (130)

We now appear to have a discrepancy between (119) and (129), and also between (121) and (130).
The reason is that the quantity called u in (119) only includes electric and magnetic field energy,
whereas the quantity called u in (129) includes those and also polarization energy as well. Also the
quantity called p in (130) includes all transfer of energy from field to matter, whereas the quantity
called p in (130) only includes the part not involving polarization or magnetization. One finds
∂u2nd ∂u1st
= + p1st (131)
∂t ∂t
where the subscripts 1st and 2nd refer to the first and second methods of calculation. It follows that
everything is consistent, but one must keeps ones wits about one when invoking either method of
calculation.

27
5 Waves in a dielectric

For a pure dielectric medium—that is, one with no free charge and zero conductivity so no conduction
current—the equations (M1a)–(M4a) simplify to

∇·D = 0, ∇ · B = 0, (132)
∂B ∂D
∇×E = − , ∇×H = . (133)
∂t ∂t
In the case of a linear isotropic medium, where D = E and B = µH, these further simplify to
1 ∂H ∂E
∇×E = − , ∇×H =  (134)
µ ∂t ∂t
where  = 0 r is the permittivity and µ = µ0 µr is the permeability. By taking the curl of either of
these equations one finds a wave equation in which the wave speed (phase velocity) is
ω 1 c
vph = =√ =√ . (135)
k 0 r µ0 µr r µr

From (M1a) and (M2a) one deduces that the waves are transverse, and from (M3a) one deduces that
the amplitudes are related by B = (k/ω)E as usual.

28