Jackson Notes part 2 January 2019

1 Energy in the Electromagnetic field

1.1 Electrostatic case
We put a charge q in a region where there is an electrostatic field E. Then the
force acting on the charge is F = q E and the work done by the fields on the
charge as it moves from A to B along path C is
B
W = q E ·d
A
B
= −q ∇Φ · d
A
B
= −q dΦ = q (ΦA − ΦB )
A

independent of the path C. Since the work done by the fields must decrease
the stored energy, we may interpret the term qΦ as the potential energy of the
system comprising the charge q and the fields E.

W = Ui − Uf = qΦA − qΦB

For a system of point charges qi at positions xi , we may write the potential

as
N
1 qi
Φ (x) =
4πε0 i=1
|x − xi |
When we compute the potential at the position of a charge qj we do not include
the potential due to charge qj (charges do not exert forces on themselves, or
equivalently, we ignore the infinite self-energy of the charge). Thus the energy
of charge qj in the presence of all the other charges is
N N
1 qi qj
Uj = qj Φ (xj ) = = Uij (1)
4πε0 |x − xi |
i=1, i=j i=1, i=j

Thus we see that the energy Uj is the sum of the energies Uij of all the pairs of
charges we can form with charge j. The energy of the total distribution appears
to be
N N N
U= Uj = Uij
j=1 j=1 i=1, i=j

However, we have to be careful not to double count pairs of charges. (Uij = Uji ,
but we only want one of them.) (See Lea and Burke Ch 25 for more on this).

1
So the correct expression is
N N N
1 qi qj 1 1 qi qj
U= = (2)
4πε0 j>i i=1, i=j
|x − xi | 2 4πε0 j=1 i=1, i=j
|x − xi |

Next we look at what happens when the charge distribution is continuous.

We replace qi with the charge element dq = ρ (x) d3 x and qj with ρ (x ) d3 x and
replace each sum with an integral to obtain
1 1 ρ (x) ρ (x ) 3 3 1
U= d xd x = ρ (x) Φ (x) d3 x (3)
2 4πε0 |x − x | 2 all space

where we used eqn (29) in Notes 1 for Φ (x) . Notice, though, that we have no
way to express the constraint i = j, and the self-energy is now included. Using
Poisson’s equation (Notes 1 eqn 32), we may eliminate the charge density and
obtain the energy in terms of the fields alone:
1
U = −ε0 ∇2 Φ Φ (x) d3 x
2
ε0
= − ∇ · Φ∇Φ dV − ∇Φ · ∇Φ dV
2
ε0
= − Φ∇Φ · n̂ dA − E · E dV
2 S∞

If the sources of the fields are localized, then the potential decreases at least as
fast as 1/r as r → ∞, and so the surface integral is zero. (It goes as 1r r12 r2 = 1r
as r → ∞.) Thus
ε0
U= E 2 dV
2 all space
and the energy density is
ε0 2
E uE = (4)
2
Extension to fields in media (see Jackson §4.7— we’ll do more with this
later). We replace ε0 E with D
1
uE = E · D (5)
2
and this result is true in media with ε = ε0 as well.
{Digression that shows the equivalence of (5) and (3).
1 1 1
U = E · D dV = − ∇Φ · D dV = − ∇ · ΦD − Φ∇ · D dV
2 2 2
 
1 1
= −  ΦD · n̂dA − Φρ dV  = ρ (x) Φ (x) dV
2 2
S∞

2
 Example: energy of a pair of point charges.
Two charges, q1 and q2, are separated by the displacement x2 . To simplify
the calculation we can put the origin at the position of one of the charges, q1 .

Then the electric field at point P is

x x − x2
E (x) = E1 + E2 = kq1 3 + kq2
|x| |x − x2 |3
and the energy density (4) is

k2 ε0 x x − x2 x x − x2
u (x) = q1 + q2 · q1 + q2
2 |x|3 |x − x2 |3 |x|3 |x − x2 |3
ε0 q12 q22 x (x − x2 )
= k2 4 + 4 + 2q1 q2 3 · 3
2 |x| |x − x2 | |x| |x − x2 |

The first two terms are the self-energy terms, and the volume integral of each is
infinite if the charges are truly points.1 The third term is the interaction energy.
It is the only term that changes as the charges move, and thus (classically) it is
the only energy we can extract from the system.
ε0 x (x − x2 ) 3
Uint = k2 2q1 q2 3 · d x
2 |x| |x − x2 |3
1 1
= k2 ε0 q1 q2 ∇ ·∇ d3 x
|x| |x − x2 |
1 1 1 1
= k2 ε0 q1 q2 ∇· ∇ − ∇·∇ d3 x
|x − x2 | |x| |x − x2 | |x|
1 1 1 1
= k2 ε0 q1 q2 ∇ · n̂ dA − ∇2 d 3 x
S∞ |x − x 2 | |x| |x − x2 | |x|
U∞ ∞
1 Note that the infinity arises at r = 0 : 1
4πr2 dr = − 4π 
0 r4 r 0

3
The surface integral is zero (integrand goes as 1/r3 while dA goes as r2 as
r → ∞), so, using Lea eqn 6.26, we have
1
Uint = k2 4πε0 q1 q2 δ (x) d3 x
|x − x2 |
q1 q2
= k
|x2 |
This is the expected result. Compare with eqn (1).

1.2 Magnetic energy

The work done by the fields on moving charges (currents), per unit time, is
δW
P = = j · E dV
δt
When the currents are confined to wires, the expression simplifies
δW
P= = IE · d
δt Ci
loops

We may use Stokes theorem to re-express the line integral:

δW
= I ∇ × E · n̂ dA
δt Si
loops

where Si is a surface spanning loop i, and then from Faraday’s law

δW ∂B
= I − · n̂ dA
δt Si ∂t
loops

So the work done by the fields is

∂B
δW = − I · n̂ dAδt
Si ∂t
loops

= − I δΦB,i
loops

where ΦB,i is the magnetic flux through surface Si . Work done by the fields
decreases the stored energy, δW = −δU. Thus

δU = I δΦB,i
loops

∂ρ
Now we let the charge density build up infinitely slowly ∂t 0 and thus ∇ · j 0 .
This means the "field lines" of j are closed loops to which we can apply the result

4
we have just derived. The current I = j∆σ where ∆σ is a small area element
perpendicular to j. Thus

δU = j∆σ δ B · n̂ dS
loops

= j∆σ ∇ × δ A · n̂ dS
loops

= j∆σ δA · d = j · δ A d ∆σ
loops loop i loops loop i

= j · δ A dV

where we used the fact that j is parallel to d . Next we use Ampere’s law to
write j in terms of the fields. Remember: we are changing things infinitely
slowly so ∂∂tE 0.
δU = ∇ × H · δ A dV

Now, from the cover,

∇· b×a = a· ∇×b −b· ∇×a

Thus, with a = δ A and b = H, we have

δU = ∇ · H × δA + H · ∇ × δA dV

We use the usual trick of converting the first integral to a surface integral, and
arguing that it is zero. (Note that A → 0 at least as fast as 1/r2 for localized
j, while H goes as 1/r3 .) Then

δU = H · δ B dV

For a linear medium, H ∝ B, and so

1
δU = δ H · B dV
2

5
and thus
1
U= H · B dV
2
and the energy density is
1
uB = H · B
2
In vacuum
1 2 µ0 2
uB = B = H
2µ0 2
Combining with the electric energy, we get
1 B2
uem = ε0 E 2 + (6)
2 µ0

1.3 Conservation of energy, general case

When there exists a current distribution j and field E, the fields do work per
unit volume to maintain the current at a rate

P =j ·E

Using Ampere’s law, we may eliminate j. This time we allow arbitrary time
variation.
∂D
j · E dV = ∇×H − · E dV
V V ∂t

Substitute for E · ∇ × H using

∇· E ×H =H · ∇×E −E· ∇×H

and use Faraday’s law and the divergence theorem to get:

∂D
j · E dV = H · ∇×E −∇· E×H − ·E dV
V V ∂t

∂B ∂D
= H· − − ·E dV − E × H · n̂ dA
V ∂t ∂t
S

In a linear medium, we can simplify the term in square brackets:

1 ∂
j · E dV = − H ·B+E·D dV − S · n̂ dA
V 2 V ∂t
S

or
∂ 1
H ·B+E ·D dV = − S · n̂ dA − j · E dV (7)
V ∂t 2 V
S

6
where S is the Poynting vector E × H. Remembering that n̂ is the outward
normal, we may interpret this result as:

rate of change of stored energy in the volume = flow of energy into the volume − work done by fields

Applying the divergence theorem to the first term on the RHS, we may write
the energy conservation equation in differential form:
∂uem
+ ∇ · S = −j · E (8)
∂t
Compare with the equation of charge conservation (Notes 1 eqn 7). Here the
non-zero term on the right hand side shows that energy may be converted to
non-electromagnetic forms by the currents.

1.4 Systems of conductors

With the usual convention that Φ → 0 at infinity, a spherical conductor of
radius a carrying charge Q has potential V = kQ/a. Similarly, the surface of
any conductor is an equipotential and the value of the potential is proportional
to the charge the conductor carries. The constant of proportionality depends
on the size and shape of the conductor:

V = Q/C (9)

where C is the capacitance. Note that C equals the charge on the conductor
when it is raised to unit potential.
Now suppose we have a collection of charged conductors. The fields pro-
duced by all of the conductors affect the potential on each of them. Thus we
may write
N
Vi = pij Qj (10)
j=1

The coefficients pij depend on the geometry. We can invert the matrix pij to
obtain the relation
N
Qi = Cij Vj (11)
j=1

If all the conductors except one are grounded, and the remaining one has Vi = 1,
then
Qi = Cii Vi = Cii
Compare this relation with equation (9). The coefficients Cii are called capac-
itances while the Cij with i = j are called coefficients of induction.

The capacitance Cii is the charge required to raise conductor i

to unit potential with all the others grounded.

7
 Note that this definition is not quite the same as the capacitance of a system
of two conductors defined in elementary treatments (e.g. Lea and Burke Ch 27
and example below). Since the name is the same, you must use context to
decide which is meant.
We may also express the energy of the system in terms of the Cij . Since
charge exists only on the surface of the conductors (J Problem 1.1), and each
surface is an equipotential, equation (3) becomes:
N
1
U = ρ (x) Φ (x) d3 x = ρ (x) Φ (x) d3 x
2 all space i=1 Vi
N N
1
= σ (x) Vi dA = Vi σ (x) dA
i=1 Si 2 i=1 Si
N
1
= Qi Vi
2 i=1
N N N N
1 1
U = Cij Vj Vi = Qi pij Qj (12)
2 i=1 j=1
2 i=1 j=1

As an example of finding coefficents of capacitance, consider a simple system

composed of two concentric, spherical conducting shells of radii a and b > a.
Let the potential be zero at infinity, and let the charges on the two conductors
be Qa and Qb respectively. Then we can use the spherical symmetry and Gauss’
law to find the electric fields, and hence the potentials on the two surfaces.

For r < a, E=0

Qa
For a < r < b, E=k r̂
r2
Qa + Qb
For r > b, r̂E=k
r2
To find the potentials, use the line integral (Notes 1 eqn 30):
2
Φ1 − Φ2 = E ·d
1

We start from infinity where we chose Φ = 0. Then:

∞ ∞
Qa + Qb Qa + Qb
Φb − Φ∞ = Φb = k 2
dr = −k
b r r b
Qa + Qb
= k
b
and then:
b b
Qa Qa 1 1
Φa − Φb = k dr = −k = kQa − (13)
a r2 r a a b

8
and so
Qa + Qb 1 1 Qa Qb
Φa = k + kQa − =k +
b a b a b
Using superposition, we can think of this as the constant potential inside the
charged sphere of radius b due to its charge Qb , plus the potential at the surface
of the charged sphere of radius a due to its own charge Qa .
Now we want to put all this into the language of capacitances. First:

Vi = pij Qj
j

or, equivalently:
Va p11 p12 Qa
=
Vb p21 p22 Qb
For our example, we can see that

p11 p12 1/a 1/b

=k
p21 p22 1/b 1/b

Then we can also write:

Qi = Cij Vj
j
or
Qa C11 C12 Va
=
Qb C21 C22 Vb
where
−1 −1
C11 C12 p11 p12 1 1/a 1/b 1 b a −a
= = =
C21 C22 p21 p22 k 1/b 1/b k b−a −a b
(14)
We may put in the expressions for the potentials to check this result:

Qa 1 b a −a Va 1 b a (Va − Vb )
= =
Qb kb−a −a b Vb kb−a −aVa + bVb

aQb +Qa b
Qa 1 ab Va − Vb 1 ab ab − Qa +Q
b
b
= = k
Qb kb−a −Va + bVb /a kb−a − aQbab
+Qa b b Qa +Qb
+a b
ab Qa b−a
ab Qa
= =
b−a Qb b−a
ab
Qb

as expected.
The capacitances of the two conductors (from (14)) are

1 ab 1 b2
C11 = and C22 =
kb−a kb−a

9
.
The “capacitance” of this system, as defined in elementary treatments for
Qa = −Qb = Q, is (equation 13)
Q Q 4πε0 ab
Cpair = = = = C11
V kQ (b − a) /ab (b − a)
Is this relation true in general? No.
In general, if Qb = −Qa , we have:
Va p11 p12 Q p11 Q − p12 Q
= =
Vb p21 p22 −Q p21 Q − p22 Q
and thus
Q 1
Cpair = =
∆V p11 − p12 − p21 + p22
Using Green’s reciprocation theorem (J problem 1.12), we can show that p12 =
p21 , and thus
1
Cpair =
p11 − 2p12 + p22
whereas
−1
C11 C12 p11 p12 1 p22 −p12
= =
C21 C22 p21 p22 p11 p22 − p12 p21 −p21 p11
So
p22 p22 1
C11 = = =
p11 p22 − p12 p21 p11 p22 − p212 p11 − p212 /p22
which is not equal to Cpair , in general.
For this system of two spheres, p22 = p12 (you might want to investigate the
conditions necessary for this relation to hold), and we have:
1
C11 = = Cpair
p11 − p22
The energy of the system is:
1
U = Q·P ·Q
2
1 1/a 1/b Qa
= Qa Qb k
2 1/b 1/b Qb
2
k Qa Qa Q2
= + 2 Qb + b
2 a b b
Note that we could also get this result from Qa Φa + Qb Φb . In the special case
that Qa = −Qb = Q, we get
k Q2 Q2 Q2 kQ2 1 1 1 Q2
U= −2 + = − = (15)
2 a b b 2 a b 2 Cpair
as expected.

10
1.5 Poynting’s theorem for harmonic fields
The above discussion does not include any dissipative effects that convert en-
ergy to non-electromagnetic forms. The easiest way to include these effects is
by working with the Fourier transforms of the Maxwell equations. This im-
mediately gives us the results for AC circuits, for example, by considering a
single frequency component. The general results for time dependent fields with
dissipation are in Jackson §6.8.
For a single frequency component
1
E (x, t) = Re E (x, ω) e−iωt = E (x, ω) e−iωt + E ∗ (x, ω) eiωt
2
When multiplying two such terms, we have to be careful. For example, the rate
at which the fields do work is

P (x, t) = Re j (x, t) · Re E (x, t)

1 1
= j (x, ω) e−iωt + j ∗ (x, ω) eiωt E (x, ω) e−iωt + E ∗ (x, ω) eiωt
2 2
1
= j (x, ω) · E ∗ (x, ω) + j ∗ (x, ω) · E (x, ω) + j (x, ω) · E (x, ω) e−2iωt
4
+j ∗ (x, ω) · E ∗ (x, ω) e2iωt
1
= Re j ∗ (x) · E (x) + j (x) · E (x) e−2iωt
2
Taking the time average, the second term vanishes, and we get
1
< P (x) > = Re j ∗ (x) · E (x)
2
In what follows, the "Re" will be implicit. Then with these conventions, the
transformed Maxwell’s equations become

∇·D = ρf
∇·B = 0
∇×E = iωB
∇ × H + iωD = jf

and the time-averaged rate at which the fields do work in a volume V is

1
< P > = j ∗ (x) · E (x) dV
2
1
= ∇ × H ∗ − iω D∗ · E (x) dV
2
But
∇ · E × H∗ = H∗ · ∇ × E − E · ∇ × H∗

11
So
1
< P > = −∇ · E × H ∗ + H ∗ · ∇ × E − iω D∗ · E (x) dV
2
1
= −∇ · E × H ∗ + H ∗ · iωB − iω D∗ · E (x) dV
2
The time averaged rate of energy flow is
1 1
S= E × H∗ = E∗ × H
2 2
making

ω
< P > = − S · n dA − i D∗ · E − H ∗ · B dV
2
S

1
j ∗ (x) · E (x) dV = − S · n̂ dA − 2iω (uE − uM ) dV (16)
2
S

Note that the last term has a real part only if

1 ε∗ (ω) 2
uE = D∗ · E = E
4 4
has an imaginary part, that is, ε has an imaginary part, indicating losses.
Re (uE ) is the time-averaged electric energy density at frequency ω, and simlarly
for uM . Equation (16) is the analog of equation (7) for harmonic fields.

Now consider a system with all the fields confined inside a volume V, as
shown. The Poynting flux S is zero on the bounding surface S. (This means
we have no radiation resistance. See J §6.9 for a fuller discussion that includes
radiation resistance). Instead power is provided to the system by an input

12
current Ii . Vi is the potential difference across the system, input terminal to
output terminal, as shown in the diagram. Then we modify equation (16) as
follows:
1 ∗ 1
Ii Vi = j (x)∗ · E (x) dV + 2iω (uE − uM ) dV
2 2
We can interpret this in terms of the impedance of the system (see, eg, Lea Ch
2 Prob 14, Jackson page 266):

Vi = Ii Z = Ii (R − iX)

where R is the resistance and X is the reactance. (The minus sign is a result
of choosing the time dependence to be e−iωt .) Then, if the conductivity is real,

2 ∗
|Ii | Z = j (x) · E (x) dV + 4iω (uE − uM ) dV

1 2
R − iX = σ E dV + 4iω (uE − uM ) dV
|Ii |2

Thus, equating real and imaginary parts, we have

1 2
R= 2 σ E dV
|Ii |

and
4ω
X =− (uE − uM ) dV (17)
|Ii |2
At low frequencies, the time-averaged stored energies may be expressed in terms
of capacitance (eqn 15) and inductance as

1 |Q|2 1 |Ii |2 1
UE = = and UM = L |Ii |2
4 C 4 ω2 C 4
dQ
where Ii = dt = −iωQ, so that (17) reduces to

1 1
X =ω L− = ωL −
ω2C ωC

as expected.

13