Potential Theory in Classical Electrodynamics

W. Engelhardt1 , retired from:

arXiv:1209.3449v2 [physics.gen-ph] 20 Mar 2013

Max-Planck-Institut für Plasmaphysik, D-85741 Garching, Germany

Abstract
In Maxwell’s classical theory of electrodynamics the fields are frequently expressed
by potentials in order to facilitate the solution of the first order system of equations.
This method obscures, however, that there exists an inconsistency between Faraday’s
law of induction and Maxwell’s flux law. As a consequence of this internal contra-
diction there is neither gauge invariance, nor exist unique solutions in general. It is
also demonstrated that inhomogeneous wave equations cannot be solved by retarded
integrals.

PACS numbers: 03.50.-z, 03.50.De

1 Introduction
Maxwell’s first order system of equations in vacuo specifies the divergence and the curl
of the electromagnetic fields E~ andB.
~ In principle, this allows to calculate the fields
from the given sources ρ and~j, but the equations are coupled which confronts us with
certain complications. In particular, it is not guaranteed that a unique solution exists
at all, as it was questioned in [1]. The two source-free equations impose a necessary
condition on the fields, if they exist: They must be derivable from a vector and a
scalar potential in the following way [2]:
~
~ =∇×A
B ~, E~ = −∇φ − 1 ∂ A (1)
c ∂t
The potentials itself are to be determined from the inhomogeneous equations which
represent a coupled system of second order equations depending on the sources.
Equation (1) leaves the fields unchanged when a “gauge transformation” is imposed
on the potentials:
~→A
A ~ + ∇ψ , φ → φ − 1 ∂ψ (2)
c ∂t
where ψ is an arbitrary differentiable function. As a consequence the divergence of
the vector potential is arbitrary to the same extent as the Laplacian ∆ψ. This fact
has been exploited to decouple the second order system. There is, however, no proof
in the literature that the chosen procedure is viable and leads to unique solutions. In
principle, the potentials are defined by the inhomogeneous equations which depend on
the divergence of the vector potential ∇ · A.~ Whether this dependence cancels when
the solutions of the second order system are substituted into (1) is an open, non-trivial
question. In Sect. 2 this problem is investigated and it is found that ∇ · A~ cannot be
chosen arbitrarily. This explains why solutions for the fields in Lorenz gauge are at
variance with those obtained in Coulomb gauge as was found in [1] and [3].
1 Home address: Fasaneriestrasse 8, D-80636 München, Germany
Electronic address: [email protected]
 The standard procedure of solving Maxwell’s equations in Lorenz gauge leads to
decoupled inhomogeneous wave equations for the potentials which are thought to be
solved by retarded integrals. In [1], however, it was claimed that the inhomogeneous
wave equations cannot be solved in general, since they connect sources and potentials
at the same time, whereas in the retarded solutions the potentials and the sources are
to be evaluated at different times. This ambiguity is again analyzed in Sect. 3 where
it is shown for charges moving at constant velocity that the Liénard-Wiechert scalar
potential cannot be considered as a solution of the inhomogeneous wave equation.
It turns out then that the system of Maxwell’s first order equations does not permit
a solution in general. Only the homogeneous wave equations, which were exclusively
considered by Maxwell in the context of his theory of light [1], are suitable to describe
travelling electromagnetic waves which are disconnected from their sources. The prob-
lem is deeply rooted in an inconsistency of the first order system which is usually
concealed by the potential ansatz (1). Analyzing the fields inside a plate capacitor the
ambiguity is made visible in Sect. 4. Concluding remarks in Sect. 5 terminate this
study on the potential method.

2 Dependence of the fields on the divergence of the vector potential

When we substitute the potential ansatz (1) into the inhomogeneous Maxwell equa-
tions we obtain the system:
1 ∂χ
∆φ = −4π ρ − (3)
c ∂t
2~
~ − 1 ∂ A = − 4π ~j + ∇χ + 1 ∇ ∂φ
∆A (4)
c2 ∂t2 c c ∂t
where the abbreviation ∇ · A ~ = χ was used. Naturally, the gauge function ψ does
not enter into these equations, as it cancels according to (1) in the expressions for the
fields. The potentials, however, will become a function of χ according to (3) and (4).
One must now investigate whether χ will also cancel in (1), when the solutions of (3)
and (4) are substituted. To this end one can exploit the linearity of eqs. (3, 4) and
split them in the following way:
φ = φ1 + φ2 (5)
∆φ1 = −4π ρ (6)
1 ∂χ
∆φ2 = − (7)
c ∂t
This set of equations is entirely equivalent to (3). Similarly:
~=A
A ~1 + A
~2 (8)

1 ∂2A~1 4π 1 ∂φ1
~1 −
∆A = − ~j + ∇ (9)
2
c ∂t 2 c c ∂t
2~
∆A~ 2 − 1 ∂ A2 = ∇χ + 1 ∇ ∂φ2 (10)
c2 ∂t2 c ∂t
Applying Helmholtz’s theorem on the vector potential Chubykalo et al. [4] have shown
that the set of equations (6) and (9) determines the fields uniquely, when the solutions
φ1 and A~ 1 are substituted into (1). In a comment by V. Onoochin and the present
author [5] it was pointed out that Chubykalo’s procedure is equivalent to choosing
χ = 0, or adopting Coulomb gauge. It follows then by insertion of (5) and (8) into (1)
~ ~
~ =∇×A
B ~1 + ∇ × A
~2 , E~ = −∇φ1 − 1 ∂ A1 − ∇φ2 − 1 ∂ A2 (11)
c ∂t c ∂t
that the terms containing χ must vanish separately
~2 = 0
∇×A (12)
~2
1 ∂A
∇φ2 + =0 (13)
c ∂t
in order to render the fields independent of the chosen gauge χ.
Let us check whether the solutions of (7) and (10) for an arbitrary choice of χ
~ 2 must satisfy the necessary
satisfy the conditions (12) and (13). First we notice that A
condition
A~ 2 = ∇U (14)
because of (12). Inserting this into (13) yields
1 ∂U
φ2 + =0 (15)
c ∂t
and equation (10) becomes
1 ∂2U 1 ∂φ2
∆U −
2 2
=χ+ (16)
c ∂t c ∂t
The retarded solution of this wave equation – subject to the boundary condition
U (∞)
~ = 0 – is:
d3 x′ 1 ∂φ2 (~x ′ , t′ )
" #
−1
ZZZ
′ ′
U= χ (~
x , t ) + (17)
4π V |~x − ~x ′ | c ∂t′ t′ =t−|~
x−~
x ′ |/c

The instantaneous solution of the Poisson equation (7) under the boundary condition
φ2 (∞)
~ = 0 is:
1 d3 x′ ∂χ (~x ′ , t)
ZZZ
φ2 = (18)
4π c x − ~x ′ |
V |~ ∂t
where the integration has to be carried out over all space. Substituting (18) into (15)
yields an instantaneous solution for U after integration with respect to time:
−1 d3 x′
ZZZ
U= χ (~x ′ , t) (19)
4π V |~x − ~x ′ |
that is not compatible with (17) for an arbitrary function χ (~x, t). Choosing, for
example,
4 
2
. 
2
q
χ= √ exp −r d sin ω t , r = x2 + y 2 + z 2 (20)
π d3
equation (18) yields:
ω erf (r/d)
φ2 = cos ω t (21)
c r
and (19) results in:
erf (r/d)
U =− sin ω t (22)
r
On the other hand, one has from (17) and (21) the result
d3 x′ exp (−r 2 /d2 ) ω 2 erf (r/d)
ZZZ " #
U =− ′ 3 − 2
sin ω (t − |~x − ~x ′ |/c) (23)
V |~x − ~x | π2 d 3 4π c r
The first term may be integrated analytically, but this is not possible for the second
one. Obviously, there is a discrepancy between (22) and (23) which proves that the
necessary and sufficient condition (15) cannot be met by the solutions (17) and (18).
Consequently, the electric field expressed by the potentials is a function of χ in general,
as the divergence of the vector potential does not cancel in (1).
For the magnetic field one can draw a similar conclusion by writing the inhomoge-
neous flux equation in integral form:
 
I ~
1 ZZ  ~ ∂ E
~ ~
B · dl = 4π j + ~
 · dS (24)
c ∂t

~ depends on χ, this holds also for B

If E ~ due to the connection in (24). On the other
hand, if one takes the curl of (10), one obtains a homogeneous wave equation for
∇×A ~ 2 which has only the solution ∇ × A~ 2 = 0 assuming A
~ 2 (∞)
~ = 0. This implies
that the magnetic field does not depend on χ in agreement with (12), but in contrast
to (24). In Sect. 4 this ambiguity will be investigated in order to clarify whether
Maxwell’s equations have unique solutions at all. Before, however, let us analyze an
inhomogeneous wave equation of type (4) and demonstrate in the next Section that it
cannot be solved by a retarded integral in general.

3 Attempt to solve an inhomogeneous wave equation

Although the Coulomb gauge χ = 0 is the natural gauge, since it follows also from
Helmholtz’s theorem applied on the vector potential [4], most textbooks make use of
the Lorenz gauge
1 ∂φL
χ=− (25)
c ∂t
which results in an inhomogeneous wave equation for the scalar Lorenz potential by
substitution into (3):
1 ∂ 2 φL (~x , t)
∆φL (~x , t) − = −4π ρ (~x , t) (26)
c2 ∂t2
We may also consider the Poisson equation in Coulomb gauge

∆φC (~x , t) = −4π ρ (~x , t) (27)

and subtract it from (26)
1 ∂ 2 φL (~x , t)
∆ (φL (~x , t) − φC (~x , t)) = (28)
c2 ∂t2
Note that the charge density cancels, since it is taken both in the instantaneous equa-
tion (27) and in the wave equation (26) at the same time t when the potentials are
evaluated. The formal unique solution for the difference of the potentials is the inte-
gral:
1 ∂ 2 φL (~x ′ , t)
" #
ZZZ
3 ′ −1
φL (~x , t) − φC (~x , t) = dx (29)
|~x − ~x ′ | 4π c2 ∂t2
assuming φC (∞)~ = φL (∞)~ = 0. This equation must be satisfied when the individual
solutions of (26) and (27) are substituted. In order to check on this let us consider a
point charge e moving along the x-axis with constant velocity v. The instantaneous
Coulomb potential resulting from (27) is the well known expression
e
φC (x, t) = q (30)
(x − x0 − v t)2 + y 2 + z 2
where x0 ist the position of the charge at t = 0. Liénard-Wiechert have calculated the
retarded integral for this situation obtaining [1]:
e
φL (x, t) = q (31)
2
(x − x0 − v t) + (1 − v 2 /c2 ) (y 2 + z 2 )
On the x-axis expressions (30) and (31) are identical so that the integral on the r.h.s. of
(29) must vanish there. This is, however, not the case, if one substitutes solution (31)
into (29) and carries out the integration over all space. One obtains (see Appendix):
e v2
φC (x, t) − φL (x, t) = (32)
x (c2 − v 2 )
This proves that the retarded integrals are not proper solutions of the inhomogeneous
wave equations which appear not to have a solution at all. The same conclusion was
reached in [1] and [6].

4 An inconsistency in determining the magnetic field

In order to facilitate the analysis of the flux law (24) let us consider an axisymmetric
case where a plate capacitor is charged up by a variable current (Fig. 1). In cylindrical
coordinates one has for the Z- component of (24):
1 ∂ (R Bϕ ) 4π 1 ∂EZ
= jZ + (33)
R ∂R c c ∂t
In the region between the plates, where the conduction current vanishes, one may
integrate (33) and obtain for the circular magnetic field component
ZR
1 ∂EZ ′ ′
Bϕ = R dR (34)
cR ∂t
0

The quasi-static electric gradient field, which is created by the surface charges on
the capacitor plates according to (6), is easily obtained from the global equations
describing a capacitor. One has
Q=CV , I = dQ/dt , EZ = V /d (35)
where Q is the total charge, C the capacitance, V the voltage, I the current, and d
the distance between the plates. Inserting this into (34) one obtains
ZR
1 I IR
Bϕ = R′ dR′ = (36)
cR dC 2c d C
0

for the magnetic field between the plates. In fact, a measurement of this field created
by the “displacement” current was reported in [7] in agreement with Stokes’ law (36).
The slope of this magnetic field was constant according to the results in Fig. 4 of Ref.
[7]. Taking the curl of this field one obtains a spatially constant displacement current
between the plates that is proportional to the time derivative of EZ in agreement with
(35).
producing a magnetic field.
G
I G
 B
G
E G
B


Fig. 1 Magnetic field created by a quasi-stationary electric field in a plate capacitor

Figure 1: Magnetic field created by a quasi-stationary electric field in a plate capacitor

On the other hand, Faraday’s law predicts for the same field component the ex-
pression
1 ∂Bϕ ∂Ez ∂ER
= − =0 (37)
c ∂t ∂R ∂z
as ER = 0 , EZ = V /d. Equation (37) yields upon integration over time a mag-
netic field exclusively created by the solenoidal part of the electric field, whereas the
irrotational part in (35) produced by the charges on the plates does not contribute.
According to (37) the magnetic field between the plates could never change, but in
agreement with (36) and the measurement as reported in [7] the changing electric gra-
dient field is well capable of producing a magnetic field in contrast to the prediction
of (37).
There is apparently an intrinsic inconsistency between Faraday’s law of induction
and Maxwell’s flux law. The discrepancy is not obvious as long as the potential ansatz
(1) is adopted. It guarantees that Faraday’s law is satisfied automatically once the
vector potential is determined from the flux law. A temporal evolution of the magnetic
~ s according
field, however, is only possible in the presence of a solenoidal electric field E
to Faraday’s law: Z
~
B = −c dt ∇ × E ~s (38)
whereas by spatial integration one finds that the magnetic field is also a function of
~ i produced by charge separation:
the changing irrotational electric field E
 
~s + E
~i

∂ E ~x − ~x ′
~ =1
ZZZ
B d3 x′ 4π ~j + × (39)
c V ∂t |~x − ~x ′ |3
In general, equations (38) and (39) are incompatible as demonstrated by the discrep-
ancy between (36) and (37).

5 Concluding remarks
The analysis presented in this paper forces us to recognize that Maxwell’s system of
first order equations cannot be solved consistently as it contains an internal contra-
diction. The potential method – which was already adopted by Maxwell himself –
conceals this fact, but it allows deriving a second order wave equation for the vector
potential modelling electromagnetic waves successfully. Maxwell considered the homo-
geneous wave equation in a region far away from the sources and formulated boundary
conditions for its solution [1]. Possibly, he was aware that an inhomogeneous wave
equation is not solvable as shown in Sect. 3. This is also true for the vector wave
equation (4) regardless which gauge is chosen.
Classical electrodynamics requires apparently a thorough revision with special at-
tention to the interaction of waves with matter. At this point a concrete proposal is
not available, but it is likely that Planck’s constant must be built into the system of
equations, since it plays a major role in the quantum theory of light that has replaced
Maxwell’s theory of light.

Appendix
In Sect. 2 the volume integral (29) had to be evaluated:
∂ 2 φL (~x ′ , t)
" #
1 1
ZZZ
3 ′
I=− dx
4π c2 |~x − ~x ′ | ∂t2
 
2
1 1 ∂  e
ZZZ
=− d3 x′ q 
4π c2 ′
|~x − ~x | ∂t2 ′ 2 2 ′2 ′2
(x − x0 − v t) + (1 − β ) (y + z )
 
2 ′ 2 2 ′2 ′2
eβ 1  −2 (x − x0 − v t) + (1 − β ) (y + z ) 
ZZZ
= d3 x′  
i5 
4π |~x − ~x ′ |  h ′ 2 2 ′2 ′2 2
(x − x0 − v t) + (1 − β ) (y + z )

where β = v/c. Changing variables one may write

~
~x − ~x ′ = R
Evaluation at the time t = −x0 /v when the charge has reached the origin yields on
the x-axis:
  
2 2
eβ 2 ZZZ −2 (x + Rx ) + (1 − β )
3 ′
d x  + Ry2 Rz2 
I=  
i 5 
4π R h 
(x + Rx )2 + (1 − β 2 ) Ry2 + Rz2

2
In spherical coordinates one has
Rx = R cos θ , Ry = R cos ϕ sin θ , Rz = R sin ϕ sin θ
and the volume element becomes
d3 x′ = R2 sin θ dϕ dθ dR
This results in:
 
Z∞ Zπ Z2π 2
e β2  −2 (x + R cos θ) + (1 − β 2 ) R2 sin2 θ 
I= R dR sin θ dθ dϕ  
i5 
4π h
2 2 2
0 0 0 (x + R cos θ) + (1 − β 2 ) R2 sin θ
Integration over the angles yields
q q
Z∞
(R − x)2 (R + x) − (R + x)2 (R − x)
I = −e β 2 x R dR q q
0 (R − x)2 (R + x)2 (x2 − R2 β 2 )2
For R > x the integrand vanishes, and for R ≤ x the integral becomes
Zx
2 xR − e β2
I = −2 e β dR =
0
(x2 − R2 β 2 )2 x (1 − β 2 )

Acknowledgments
It is a pleasure to thank Vladimir Onoochin for his extremely valuable contributions
to the discussion of a very difficult subject. Peter Enders suggested introducing the
potential U in Sect. 2 which led to a substantial simplification of the analysis.

References
[1] Engelhardt W 2005 Gauge invariance in classical electrodynamics Ann. Fond.
Louis de Broglie 30 157-78
[2] J. D. Jackson, Classical Electrodynamics, Second Edition, John Wiley & Sons,
Inc., New York (1975), Sect. 6.4
[3] Onoochin V 2002 On non-equivalence of Lorentz and Coulomb gauges within
classical electrodynamics Ann. Fond. Louis de Broglie 27 163-84
[4] Chubykalo A, Espinoza A, Alvarado Flores R 2011 Electromagnetic potentials
without gauge transformations Phys. Scr. 84 015009
[5] Engelhardt W, Onoochin V 2012 Comment on ’Electromagnetic potentials with-
out gauge transformations’ Phys. Scr. 85 047001
[6] Engelhardt W 2012 On the solvability of Maxwell’s equations Ann. Fond. Louis
de Broglie 37 3-14
[7] Bartlett D F, Corle T R 1985 Measuring Maxwell’s Displacement Current Inside
a Capacitor Phys. Rev. Lett. 55 59-62