a

r
X
i
v
:
0
7
0
8
.
2
1
7
4
v
1


[
p
h
y
s
i
c
s
.
c
l
a
s
s
-
p
h
]


1
6

A
u
g

2
0
0
7
A geometric analysis of
the Maxwell eld in a vicinity
of a multipole particle
and new special functions
Jerzy Kijowski
a
and Piotr Podles
b
a
Center for Theoretical Physics, Polish Academy of Sciences,
Aleja Lotnikow 32/46, 02668 Warszawa, Poland
b
Department of Mathematical Methods in Physics,
Faculty of Physics, Warsaw University,
Ho za 74, 00682 Warszawa, Poland
Abstract
A method of solving Maxwell equations in a vicinity of a multi-
pole particle (moving along an arbitrary trajectory) is proposed. The
method is based on a geometric construction of a trajectory-adapted
coordinate system, which simplies considerably the equations. The
solution is given in terms of a series, where a new family of special
functions arises in a natural way. Singular behaviour of the eld near
to the particle may be analyzed this way up to an arbitrary order. Ap-
plication to the selfinteraction problems in classical electrodynamics
is discussed.
1 Introduction
Classical approach to the problem of motion in classical electrodynamics is
due to Dirac (cf. [3], [6]). In this approach, equations of motion of charged
point particles, interacting via Maxwell eld, are derived from eld equations
1
by the following procedure: a generic state of the composed particles + eld
system is treated as a deformation of the ground state of the eld, uniquely
determined by the positions of the particles. The ground state is dened
via a (non-local in time) decomposition of the actual eld into the retarded
(or advanced) eld and the remaining radiation eld. Unfortunately, to
decide what is the retarded or the advanced eld, the entire trajectory of the
particle must be known in advance and, whence, the causality of the theory
is violated.
Recently, it was shown that such a non-causal procedure may be avoided.
In this approach, equations of motion are a simple consequence of the conser-
vation laws imposed on an (appropriately dened, already renormalized)
total four-momentum of the particles + eld system (see e.g. [7], [5], [11]).
Here, the ground state of the system may be dened via a conditional min-
imization of the energy, with the positions and the charges of the particles
being xed.
Mathematically, this leads to a simple (elliptic) variational problem for
the behaviour of the eld in a topologically non-trivial region of R
3
(exterior
of the particles), where the charges of the particles provide the necessary
boundary conditions. Such a reduction of electrodynamics to electrostatics
provides basic information about the behaviour of the eld in the vicinity
of the particle (e. g. the leading r
2
term for the monopole particle or the
r
3
behaviour for the dipole) but fails to capture its more subtle features,
which are necessary to describe such phenomena like polarization, which is
necessary for the analysis of the dynamical stability of the classical elec-
trodynamics (see e. g. discussion of these issues in [10] and the references
therein).
A substantial improvement of description of the eld singularity near
the trajectory is obtained if we reduce the dynamics with respect to a one-
parameter group of boosts (see [11]), instead of the one-parameter group of
time translations, corresponding to the standard, at (3+1)-decomposition
of the Minkowski spacetime. As a result, we obtain the electrostatics in
a uniformly accelerated reference frame as a proper tool to construct the
ground state of the eld. Here, eld equations are again elliptic, but the
at Eucledean geometry is replaced by the Lobaczewski geometry. Within
this framework, the Born solution describing Maxwell eld attached to a
uniformly accelerated particle (see e. g. [13]) arises as a fundamental solution
of the Laplace-Beltrami operator
L
in the Lobaczewski space:
L
=
0
,
where the right hand side is the Dirac delta distribution (a nice analysis of
2
this structure was proposed by Turakulov in [14]).
In the present paper we show how to use this (naturally arising) Loba-
czewski three-geometry to the analysis of the full dynamical problem: analy-
sis of the Maxwell eld generated by a (moving along an arbitrary trajectory)
multipole particle. Our method consists in splitting the portion of spacetime
in the vicinity of a generic trajectory of a particle into a collection of 3-
spaces orthogonal to the trajectory. Each of these spaces carries a natural
Lobaczewski geometry, uniquely implied by the instantaneous acceleration
of the particle. As a result of this construction, the at spacetime geome-
try is replaced by a curved (general relativistic) geometry, whose parameters
are uniquely dened by the parameters of the trajectory in question. We
show that the Maxwell equations may be rewritten in terms of this geome-
try. This method provides a new, powerful tool, which enables us to solve
Maxwell equations by a successive approximation method.
The results presented here give us an important improvement and gen-
eralization of earlier methods (see [8] or [9]), based on the use of the Fermi
coordinates, dened by the Fermi tetrad attached to the trajectory. The
geometric construction proposed in the present paper will be referred to as
a modied Fermi system.
Naively speaking, the standard (3+1)-approach consists in approximating
the generic trajectory by a straight line, matching only the instantaneous
velocity of the particle. In the approach based on Fermi coordinates, we
use much better approximation given by hyperbolic (uniformly accelerated)
trajectories, which match not only the velocity but also the acceleration of
the particle. The value of acceleration is encoded in the parameter of the
Lobaczewski 3-space arising in this construction. In our new approach (which
we refer to as the modied Fermi), proposed in the present paper, we use
still better approximation: instantaneous acceleration a(t) of the trajectory
implies the parameter of each of the (Lobaczewski) 3-space folia {t =const. }
of the constructed geometry; there is, however, a non vanishing shift vector
whose value encodes the derivatives a
j
(t).
The basic technical tool of our method consists in deriving an explicit
formula for the operator inverse to the Lobaczewski curl operator (which
we denote by curl

) on a certain space of divergence-free vector elds.

The inversion requires a successive solution of Legendre dierential equation
with non trivial right hand side. This, in principle, could lead very quickly
to non-elementary functions (see formula (33) in Section 4). To our great
surprise, we were able to express the results (up to the order +1 in the
3
radius r) in terms of rational functions of two universal combinations of the
radial coordinate, namely u = ar/2 and R = log(
1u
1+u
), where a is the (scalar)
acceleration of the particle (see Section 6). This result (together with the
extremely simple and natural physical context) leads us to a conjecture that
the functions obtained this way form a new family of special functions, which
we discuss at the end of the paper.
Physically, our results form a basis for the theory of motion of a po-
larizable particle: a particle which carries a dynamical electric or magnetic
moment, i. e. a moment which is not frozen within the particle but depends
dynamically upon the surrounding electromagnetic eld. There is a serious
hope that such a theory would be free of the standard non-stability prob-
lems accompanying the Abraham-Dirac theory (see [10] and the discussion
therein).
The paper is organized as follows. In Section 2 we dene the modied
Fermi system. In Section 3 we derive Maxwell equations in the modied
Fermi system and interpret them in terms of the Lobaczewski geometry. In
Section 4 we dene basic elds X and Y and show that the operations div,
curl

and # act on X, Y in a simple way. These results give a generalization

of the construction proposed in [9]. In Section 5 we solve Maxwell equations
in a vicinity of the multipole particle, in terms of a series consisting of the X
and Y elds. The procedure is analogous to that used in [9]. In case of the
monopole and the dipole particle, the resulting series up to the order r
1
is
explicitely given in Section 6, where r is the radial coordinate in the modied
Fermi system. In Section 7 we propose a conjecture concerning the special
functions arising naturally in our construction. Several specic computations
and proofs are shifted to the Appendices A-D. The results of Sections 6 and
7 were obtained partially with help of the symbolic computations program
MAPLE 8.
Thorough the paper we use greek indices (running from 0 to 3) to label
space-time coordinates, and latin indices (runing from 1 to 3) to label space
coordinates. We always use Einstein convention: summation over repeated
indices (in both space-time and space cases). The components of an n-tensor
T are denoted by T
i
1
i
2
...in
.
4
2 The Fermipropagated and the modied
Fermipropagated systems
In this section we recall properties of the Fermipropagated system of coor-
dinates, in which the particle remains at rest at each instant of time. Next
we modify that system in a way which is similar to transition from cartesian
to bispherical (bipolar) coordinates in R
3
, with the particles position being
one of its centers (poles).
Let y

, = 0, 1, 2, 3, denote the (Minkowski) spacetime coordinates in

a xed inertial (laboratory) system, corresponding to the metric tensor
= diag(, +, +, +). Consider an arbitrary particles trajectory q

(t) =
(t, q
k
(t)). Let = (t) denote a particles proper time along the trajectory.
The normalized four-velocity is given by u = dq/d and the particles accel-
eration by a = du/d = d
2
q/d
2
. We dene (see e.g. [11]) the restframe
space
(t)
as the hyperplane orthogonal to the trajectory (i.e. to the vector
e
(0)
:= u) at the point q(t). We choose an orthonormal basis e
(l)
, l = 1, 2, 3,
in

, such that e
()
are positively oriented. Thus (e
()
|e
()
) =

. Since
(u|a) = 0, one gets a = a
l
e
(l)
for some a
l
.
At each instant of time , the above system may be subject to the SO(3)
group of rotations of the Dreibein (e
(l)
()), playing role of the group of gauge
transformations. The Fermi-propagated system (see e.g. [11]) is dened by
the following condition:
de
(l)
/d = a
l
u .
Once the above condition is satised, there remains a single, global (time-
independent) SO(3) gauge transformation, corresponding to an arbitrary
choice of initial conditions: e
(l)
(
0
). Moreover, we obtain in this case:
du/d = a
l
e
(l)
,
which enables us to interpret a
l
, l = 1, 2, 3; (with a
0
= 0) as components of
the acceleration vector. Next, we dene in a neighbourhood of the trajectory
the Fermi-propagated (local) coordinates

:= (
0
,
l
) putting
0
:= and
y

= q

() +
l
e

(l)
(). (1)
Hence,
l
are cartesian coordinates on

, related to e
(l)
and centered at the
particles position (
l
= 0 for y = q(t)). The Minkowski metric tensor in
that system is given by

kl
=
kl
,
0l
= 0,
00
= N
2
, (2)
5
where k, l = 1, 2, 3, N = 1 + a
l

l
(see e.g. [11], p. 373, but with denoted
there by x).
Now, we change variables on each

and dene the modied Fermi-

propagated system x

= (x
0
, x
l
) by putting x
0
:= =
0
and:
x
l
=

l
+
1
2
a
l

2
1 + a
i

i
+
1
4
a
2

2
, (3)
where a = (
kl
a
k
a
l
)
1/2
, = (
kl

l
)
1/2
. It is easy to check that the inverse
transformation is given by the following formula:

l
=
x
l

1
2
a
l
r
2
1 a
i
x
i
+
1
4
a
2
r
2
, (4)
where r = (
kl
x
k
x
l
)
1/2
(a
i
= a
i
is the same as before because the particle is
situated at x = = 0, where x
k
/
l
=
k
l
and (2) holds).
A simple computation shows that the at (Minkowski) metric tensor has
now the following components:
g
kl
=
N
2

2

kl
, g
0l
=

k

k
x
l
, g
00
= N
2
+

k

, (5)
with
= 1 (ar/2)
2
(6)
Of course, we have: g
00
= g(d ; d) = N
2
, and thus the lapse function N
is the same as for the previous Fermi system, because the time variable has
not been changed (cf. [9]). In terms of the new coordinates, its value may
be written as follows: N = /M, where we denote M = 1 a
i
x
i
+
1
4
a
2
r
2
=
(1 + a
k

k
+
1
4
a
2

2
)
1
. Moreover,

det g
kl
= (N/)
3
= M
3
. Denoting by
g the 3-dimensional inverse of (g
kl
), k, l = 1, 2, 3, the shift vector may be
calculated as follows:(cf. [9])
N
k
= g
kl
g
0l
=
x
k

= x
k
a
j
x
j

1
2
r
2
a
k
, (7)
where dot denotes the derivative w.r.t. , e.g. a
j
= da
j
/d.
To illustrate the geometric structure of the above coordinate system, let
us apply to the above picture such a rotation O, which positions the third axis
e
(3)
in the direction of the acceleration a, i. e. such that we have: a = a e
(3)
().
6
Apply now rotation O to coordinates (x
1
, x
2
, x
3
) and denote by (z
1
, z
2
, z
3
)
new coordinates on

, obtained via such a rotation. Next, using z

k
as Carte-
sian coordinates, construct the corresponding spherical coordinates (r, , ).
Finally, dene the variable by
r =
2
a
exp().
Then, it is easy to show that (, , ) form the bispherical system of coor-
dinates on the Euclidean space

, with the particles position being one

of its centers: r = 0 (i.e. +). The vector connecting this center
with the other center of the bispherical system is parallel but opposite to the
acceleration and its length equals
2
a
.
3 Maxwell equations
Maxwell equations: df = 0 and df = J, with denoting the Hodge
star operator, can be written in an arbitrary system of coordinates, in an
arbitrarily curved spacetime, as follows (cf. e.g. [9]):

[
f
]
= 0, (8)

= J

, (9)
F

=
_
det g g

, (10)
where the brackets [ ] denote the complete anti-symmetrization. More-
over, we denote: J

det gJ

, where J

is the fourcurrent vector

and J

is the current density (an odd three-form, see e. g. [2]), sat-

isfying the continuity equation

= 0. The two vectordensities: D

(electric induction) and B (magnetic induction), are dened (cf. [9]) as the
following components of these tensors:
1
2
f
kl

klm
= B
m
, F
0k
= D
k
, where

klm
is the standard LeviCivitta symbol (totally antisymmetric and nor-
malized:
123
= 1). Due to (5), the covariant components of the mag-
netic eld, calculated in our modied-Fermi coordinates (3), are equal to
(cf. [9]): B
m
= g
mk
B
k
= g
mk
(det g
kl
)
1/2
B
k
= (/N)B
m
. Hence, we have:
NB
m
= B
m
and, similarly, ND
m
= D
m
. These formulae show that a sub-
stantial simplication of the eld equations is obtained, if we use a ctitious,
at metric
kl
to identify upper and lower indices, i. e. to x an isomor-
phism between vectors and covectors. From now on, we strictly observe this
7
convention. Consequently, we can rewrite Maxwell equations (8)(10) (cf.
(9)-(12) of [9]) as equations for the two quantities (D, B), treated as two
vectors in this ctitious at geometry:

k
D
k
= J
0
, (11)

k
B
k
= 0, (12)

D
k

l
(N
l
D
k
N
k
D
l
) =
kil

i
(
lj
B
j
) J
k
, (13)

B
k

l
(N
l
B
k
N
k
B
l
) =
kil

i
(
lj
D
j
), (14)
where the vector N
k
is uniquely implied by the derivative of the trajectory up
to the third order, according to formula (7), whereas the conformal factor
is implied by the second derivative, according to formula (6). We introduce
the following short-hand notation for the dierential operators appearing
here:
W
#
=

W +

W, (15)
where

W
k
=
l
(N
k
W
l
N
l
W
k
) , (16)
and
curl

W := curl(W) .
This enables us to rewrite Maxwell equations as follows:
divD = J
0
, (17)
divB = 0, (18)
D
#
= curl

B J, (19)
B
#
= curl

D . (20)
Now, we are going to show that the dierential operator curl

can be
nicely interpreted as a genuine curl operator in the Lobaczewski geometry.
For this purpose consider the covectors which are subject to dierentiation
in the denition of curl

, see formulae (13), (14):

d
l
:=
lj
D
j
, b
l
:=
lj
B
j
.
We want to nd an appropriate (yet another) 3-metric on each leaf

,
such that the above covectors (dierential one-forms) are related with the
8
vector densities D and B (dierential three-forms) via the standard Hodge
formula:
D
k
=

kl
d
l
, B
k
=

kl
b
l
. (21)
It is easily seen that must be conformally similar to the at metric:
kl
=

kl
. Taken into account that

kl
=
1
2

kl
, we conclude that to full (21)
we must take
1
2
= 1, i. e. =
2
. It is a matter of easy calculations to
check that, indeed, the resulting metric tensor:

kl
:=
2

kl
=
1
1 (ar/2)
2

kl
, (22)
dened on the ball K := {x R
3
: r = x <
2
a
} is equal to the Lobaczewski
metric with constant negative curvature R = 6a
2
.
In the particular case of a uniformly accelerated trajectory, i. e. when
a = 0, the shift vector vanishes: N = 0. As noticed by several authors,
the electrostatics (with respect to such a uniformly accelerated reference
system!) implies: curl

D = 0, equivalent to d = grad . Then, Gauss

equation (17) reduces such a theory to the standard potential theory in the
Lobaczewski space: div grad =
L
= J
0
. The fundamental solution of
this equation, corresponding to the point particle: J
0
=
0
(Dirac-delta-like
charge density cf. [14]), describes the electromagnetic eld accompanying
the uniformly accelerated particle and is equal to the Born solution (cf. [1]
or [13])
4 The calculus of X and Y elds
In this section we dene elementary elds X and Y, which enable us to
reduce the asymptotic analysis of Maxwell equations in a neighbourhood of
a moving, polarized particle, to a relatively simple algebra of these elds.
Given a pair (n, m) of natural numbers such that n = 1, 2, . . ., m =
0, 1, 2, . . ., consider the vector space L
n,m
of Laurent series of a single variable
r, of the form

k=0
c
k
r
n+2m+2k
,
convergent in an annular neighbourhood 0 < r < , where the coecients c
k
are real. Observe that we have L
n,m+1
L
n,m
. By the order of the series we
9
understand the order of its rst non-vanishing term (i. e. f is of order l if it
behaves at r = 0 like r
l
).
For any pair (f, Q), where f L
n0
, and Q = (Q
i
1
in
) is a completely
symmetric, traceless tensor of rank n (i. e. satisfying:
i
1
i
2
Q
i
1
i
2
in
= 0),
consider the following vector elds:
X
k
(n)
(f, Q) = [(n +1)f rf
,r
]
1
r
n+3
x
k
Q
i
1
in
x
i
1
x
in
+f
,r
1
r
n
Q
k
i
2
in
x
i
2
x
in
,
(23)
Y
k
(n)
(f, Q) =
f

1
r
n+1

kjl
x
j
Q
li
2
in
x
i
2
x
in
. (24)
For an exceptional case n = 0, every Q is a scalar and, therefore, the eld
Y
(0)
is not dened. Moreover, we assume in this case that f is constant.
This is an appropriate generalization of the elds X and Y , which were
dened in [9] for purposes of the analysis of the Maxwell equations in terms of
the standard Fermi system. As will be seen in the sequel, this generalization
enables us to describe the asymptotic behaviour of the Maxwell eld in a
neighbourhood of a polarized, point particle in a much ecient way.
It can be easily shown (cf. Appendix A) that, outside of the center r = 0,
the elds X and Y are divergence-free:
divX
(n)
(f, Q) = divY
(n)
(f, Q) = 0, (25)
and the operator curl

acts in a simple way on these elds:

curl

Y
(n)
(f, Q) = X
(n)
(f, Q) , (26)
curl

X
(n)
(f, Q) = a
2
Y
(n)
(h
(n)
(f), Q), (27)
where
h
(n)
(f) =

2
a
2
f
,rr
+
r
2
f
,r
+
n(n + 1)
2
a
2
r
2
f . (28)
(The full action of these operators understood in the sense of distributions
contains also Dirac--like terms at r = 0, cf. Appendix D). A much simpler
form of the operator h
(n)
is obtained if we introduce the following new vari-
ables: u = ar/2, z =
1
2
(u
1
+ u) =
1
2
(
2
ar
+
ar
2
) = cosh . Then z 1 and as-
suming u < 1 (i. e. near to the particle) we have: r = 2u/a =
2
a
(z

z
2
1).
Now, h
(n)
takes the form:
h
(n)
[g(z)] = (z
2
1){(1 z
2
)
d
2
g
dz
2
2z
dg
dz
+ n(n + 1)g}. (29)
10
Equation h
(n)
[g(z)] = 0 is equivalent to the Legendre equation and, therefore,
has two independent solutions: the Legendre polynomial of order n: P
n
(z) =
1
2
n
n!
d
n
dz
n
(z
2
1)
n
= A
n
(r) and the Legendre function
1
2
P
n
(z) log(
z+1
z1
)+v
n1
(z) =
B
n
(r), where v
n1
is a polynomial of order n1, v
1
= 0, P
n
and v
n
are even
(odd) for n even (odd) (cf. [15], pp. 302 and 319). Thus A
n
(r) and B
n
(r)
are solutions of equation h
(n)
(f) = 0. A
n
has the same parity as n, B
n
the
opposite parity (one can formally dene them for |u| =
a|r|
2
< 1).
It can be proved (see Appendix B) that A
n
is a Laurent polynomial of
r with the lowest power r
n
, whereas B
n
is regular at r = 0 and its Taylor
series starts from r
n+1
.
Proposition. There exists a unique operator (the index m is suppressed
for simplicity) s
(n)
: L
nm
L
n,m+1
, which is inverse to h
(n)
: L
n,m+1
L
nm
.
Proof. Because solutions of the homogeneous problem are known, equa-
tion
h
(n)
(f) = l (30)
can be solved by the variation of constants method:
f(r) = A
n
(r)
_
a
2
l(r)B
n
(r)dr

2
(A
n
B
n

B
n
A
n

)
B
n
(r)
_
a
2
l(r)A
n
(r)dr

2
(A
n
B
n

B
n
A
n

)
(31)
where the Wronskian A
n
B
n

B
n
A
n

is proportional to
1
, so the denom-
inators under integrals are proportional to . Assume now that l L
nm
.
We dene s
(n)
(l) as the right hand side of (31) with the integrals performed
term by term, using the formula
_
r

dr =
r
+1
+1
( = 1). Then s
(n)
is linear.
Moreover, the Laurent series of the function under the rst integral is odd
and starts at least from r
1+2m
, so the series for the integral is even and starts
at least from r
2+2m
. The Laurent series under the second integral is even and
starts at least from r
2n+2m
. Its integral is odd and starts from r
2n+2m+1
.
(This proves that, indeed, the case = 1 never occurs.) Thus, s
(n)
(l) is
even or odd (depending upon n) and starts at least from r
n+2(m+1)
, i. e. we
have s
(n)
(l) L
n,m+1
. Any other solution f of (30) diers from s
(n)
(l) by a
combination of A
n
and B
n
. But the condition f L
n,m+1
excludes both A
n
(its series starts from lower power r
n
) and B
n
(its parity is dierent from
n). This proves the uniqueness of s
(n)
. Q.E.D.
We conclude that, restricted to the space of those X
(n)
(f, Q) and Y
(n)
(f, Q),
for which f L
nm
, the operator curl

has a right inverse curl

1

(i.e.
curl

curl
1

= id) given by the following formulae:

curl
1

X
(n)
(f, Q) = Y
(n)
(f, Q), (32)
11
curl
1

Y
(n)
(f, Q) = a
2
X
(n)
(s
(n)
(f), Q) . (33)
Let us notice that for f L
nm
being of minimal order (n + 2m), one has
that:
X
(n)
(f, Q) is at least of order r
n+2m2
,
Y
(n)
(f, Q) is at least of order r
n+2m1
.
Those orders are called the generic orders. Observe that (generically) curl
1

increases the order (in r) by 1.

Now, suppose that both f and Q are time-dependent. It turns out that
also time-depending operator # (cf. (15)(16)) acts in a simple way on the
elds X and Y. Indeed, we have:
Theorem 1. For r = 0
X
(n)
(f, Q)
#
=
1
2
X
(n+1)
(r
2
f
,r
+ (n + 1)rf, b Q) +Y
(n)
(g
1
, b Q) +
+
n
2
1
2n(2n + 1)
X
(n1)
(r
2
f
,r
+ nrf, bQ) +X
(n)
(

f, Q) + X
(n)
(f,

Q),(34)
Y
(n)
(f, Q)
#
= Y
(n+1)
(g
2
, b Q)
1
2(n + 1)
X
(n)
(
fr
2

, b Q) +
+Y
(n1)
(g
3
, bQ) +Y
(n)
(

f, Q) +Y
(n)
(f,

Q) +Y
(n)
(
fr
2
2
, a aQ), (35)
where b
k
= (a
k
), a = da/d = a
k
b
k
/a, whereas the symmetric traceless
tensors: b Q, b Q and bQ of ranks (respectively) n + 1, n and n 1,
are built from the n-tensor Q and the vector Q according to the following
formulae (the letters placed below dots denote missing indices):
(b Q)
i
0
in
=
1
n + 1
n

s=0
b
is
Q
i
0

..
s
in

2
(n + 1)(2n + 1)

k<l

i
k
i
l
b
m
Q
mi
0

..

l
in
, (36)
(b Q)
i
1
in
=
1
n
n

k=1

i
k
lm
b
l
Q
mi
1

..

k
in
, (37)
12
(bQ)
i
2
in
= b
m
Q
mi
2
in
, (38)
g
1
=

2(n + 1)
[r
2
f
,rr
+ 2rf
,r
n(n + 1)f],
g
2
=
1
2
(n + 3)rf
1
2
r
2
f
,r

a
2
fr
3
4
,
g
3
=
n
2
1
n(4n + 2)
_
(n 2)rf r
2
f
,r

a
2
fr
3
2
_
,
= 1 (ar/2)
2
.
Remark. For the properties of bQ, bQ, bQ (computed in a simpler
situation, but valid universally) see [9], pp. 304-305 (Warning: Normalization
of the tensors b Q and b Q used in (36)(37) diers slightly from that
used in (23)-(24) of [9]). Let us notice that for W being X or Y eld, W
#
is at least of the same generic order of r as W. If, moreover, f, Q do not
depend upon , the generic order of W
#
increases by at least 1 with respect
to the generic order of W.
We shall use these formulae in a special case, when f depends upon time
only via the acceleration a = a() contained in the combination u = ar/2.
Assuming, therefore, that f(r, ) = f(u), one can rewrite (34)(35) as follows:
X
(n)
(f, Q)
#
= X
(n+1)
_
u
2
f
,u
+ (n + 1)uf,
1
a
b Q
_
+
1
2(n + 1)
Y
(n)
[(u
2
f
,uu
+ 2uf
,u
n(n + 1)f), b Q] +
n
2
1
n(2n + 1)
X
(n1)
_
u
2
f
,u
+ nuf,
1
a
bQ
_
+X
(n)
[uf
,u
,
a
a
Q] +X
(n)
[f,

Q], (39)
Y
(n)
(f, Q)
#
= Y
(n+1)
_
(n + 3)uf + u
2
f
,u
+
2u
3
f

,
1
a
b Q
_
+

2
n + 1
X
(n)
(
fu
2

,
1
a
2
b Q) +
n
2
1
n(2n + 1)
Y
(n1)
_
(n 2)uf u
2
f
,u

2u
3
f

,
1
a
bQ
_
+
Y
(n)
[uf
,u
,
a
a
Q] +Y
(n)
[f,

Q] +Y
(n)
[
fu
2

,
2 a
a
Q]. (40)
The proof of Theorem 1 is given in Appendix A.
13
5 Expansion of the Maxwell eld
In this section we nd an asymptotic solution for the Maxwell eld in a vicin-
ity of a time-dependent, electric 2
n
pole, moving along a given trajectory.
Given a timelike trajectory, we construct its modied Fermi frame like in
Section 2 and consider Maxwell equations (17)-(20) with electric charge:
J
0
= c
n
Q
i
1
...in

i
1
. . .
in

(3)
(41)
and the corresponding electric current
J
k
= c
n

Q
ki
2
in

i
2
. . .
in

(3)
, (42)
at the right-hand side. The above four-current is conserved:

= 0, cf.
[9]. Here, c
n
= (1)
n
4/(2n 1)!! is the normalization factor. For n = 0
we assume J
k
= 0 and Q = const. (See Appendix C for relation of these
2
n
poles, dened in the modied Fermi frame, with the standard 2
n
-poles in
the Fermi frame.)
Theorem 2. Maxwell equations (17)(20) with the sources (41)-(42) are
solved by the following (formal) series:
D = D
d
+D
(n2)
+D
(n)
+D
(n+2)
+ . . . , (43)
B = B
(n1)
+B
(n+1)
+B
(n+3)
+ . . . , (44)
where
D
i
d
= (1)
n
4n
(2n + 1)!!
Q
ii
2
...in

i
2
. . .
in

(3)
, (45)
D
(n2)
= X
(n)
(k
n
P
n
(z), Q), (46)
B
(k+1)
= curl
1

[D
(k)
#
], k = n 2, n, n + 2, . . . , (47)
D
(k+1)
= curl
1

[B
(k)
#
], k = n 1, n + 1, n + 3, . . . , (48)
i.e. B
(n1)
= curl
1

[D
(n2)
#
], D
(n)
= curl
1

[B
(n1)
#
],
B
(n+1)
= curl
1

[D
(n)
#
], . . . , k
n
= (2a)
n
/(
2n
n
). Moreover, B
(l)
and D
(l)
are
at least of order r
l
.
Remark 1. The elds B, D and the operations div, curl

, # are consid-
ered here in the sense of distributions. On the other hand, curl
1

acts on
14
the elds X and Y via (32)(33). One can add to (17) (20) any regular
solution (D, B) of homogeneous Maxwell equations.
Remark 2. One has k
n
P
n
(z) = r
n
+ f
n
, where lim
r0
f
n
(r)r
n
= 0.
Thus the leading term in D equals D
d
+X
(n)
(r
n
, Q) = grad
n
where

n
=
Q
i
1
...in
x
i
1
. . . x
in
r
2n+1
is the standard potential for the classical 2
n
-pole (41) with a 0, Q = const.
(cf. Appendix D).
Proof of Theorem 2 is given in Appendix D.
6 Application to the monopole and the dipole
cases
In this section we apply the above formulae to the case of a monopole and a
dipole particle. We nd explicitly asymptotic expansion of the Maxwell eld
(D, B) up to the order r
1
.
In the monopole case we start from D
(2)
= X
(0)
(1, Q) which is of order
r
2
, in the dipole case we start fromD
(3)
= X
(1)
(k
1
P
1
(z), Q) = X
(1)
(
1
u
,
a
2
Q)+
X
(1)
(u,
a
2
Q). The last two components are, respectively, of orders r
3
and
r
1
. To compute the next terms we use operations curl
1

and #. While
curl
1

generically increases the order by one, (39)(40) show that # applied

to X
(n)
(f, Q) or Y
(n)
(f, Q) leads to terms of dierent orders: in generic case,
rst three terms of (39) or (40) have order increased by one, the next two
terms there have order unchanged and the last term in (40) has order in-
creased by two. We present the results of computation in the monopole and
in the dipole cases in terms of u = ar/2 and R = log(
1u
1+u
) = 2u + . . .,
obtained using the symbolic calculus provided by the program MAPLE 8.
The terms of order higher than r (or u) are denoted by o(u). We collect all
the terms of generic order r
l
from dierent parts of (43) or (44) and denote
them, respectively, by D
(l)
or B
(l)
. In the monopole case we assume that the
charge Q = const.
The monopole case
D = D
(2)
+ D
(1)
+ o(u), B = B
(0)
+ o(u), (49)
15
where
D
(2)
= X
(0)
(1, Q), D
(1)
= X
(1)
(f
0
,
1
a
3
Q

b), B
(0)
= Y
(1)
(u,
1
a
Qb), (50)
with
f
0
=
1
4
(u + u
1
)R
2
R u = u
3
+ . . . . (51)
The dipole case
D =
4
3
Q
(3)
+D
(3)
+D
(1)
+D
(0)
+D
(1)
+ o(u), (52)
B = B
(2)
+B
(1)
+B
(0)
+B
(1)
+ o(u), (53)
where
D
(3)
= X
(1)
(u
1
,
a
2
Q), (54)
D
(1)
= X
(1)
(f
1
,
1
2a

Q) +X
(1)
(u,
a
2
Q), (55)
D
(0)
= X
(2)
(f
2
,
1
2a
2
(b Q) ) +Y
(1)
(u,
1
2a
(b Q) ) +
+Y
(1)
(
u
1 u
2
,
1
2a
b

Q) +X
(2)
(f
3
,
1
2a
2
b

Q), (56)
D
(1)
= Y
(2)
(
u
2
1 u
2
,
1
3a
2
b (b Q)) X
(3)
(f
4
,
1
2a
3
b (b Q)) +
Y
(2)
(u
2
,
3
2a
2
b (b Q)) X
(1)
(f
5
,
a
a
2

Q) +
X
(1)
(f
0
,
a
a
3
( aQ+
a
2

Q)) X
(1)
(f
0
,
1
a
2
( aQ +
a
2

Q) ) +X
(1)
(f
8
,
a
2
2a
5

Q) +
X
(1)
_
f
7
,
a
a
3
(

Q
2a
) +
1
a
2
(
a

Q
2a
2
)
_
+X
(1)
(f
6
,
1
a
2
(

Q
2a
)),(57)
B
(2)
= Y
(1)
(u
1
,
a
2

Q), (58)
B
(1)
=
1
2
Y
(2)
(1, b Q) +X
(1)
(u,
1
2a
b Q), (59)
16
B
(0)
= Y
(1)
(u, aQ +
a
2

Q) +Y
(1)
(f
1
, (

Q
2a
) ) +Y
(1)
(f
9
,
a
2a
2

Q), (60)
B
(1)
=
3
2
Y
(2)
(u
2
, b Q) Y
(2)
(f
10
,
a
2a
3
(b Q) )
Y
(2)
(f
2
, [
1
2a
2
(b Q) ] ) X
(1)
(f
0
,
a
2a
4
(b Q) )
X
(1)
(f
0
,
1
a
2
[
1
2a
(b Q) ] ) Y
(2)
(f
11
,
a
2a
3
b

Q)
Y
(2)
(f
3
, (
1
2a
2
b

Q) ) X
(1)
(f
12
,
a
2a
4
b

Q)
X
(1)
(f
5
,
1
a
2
(
1
2a
b

Q) ) Y
(2)
(f
13
,
1
2a
2
b

Q) +
X
(1)
(f
14
,
1
8a
3
b

Q), (61)
where f
0
is given by (51),
f
1
=
1
4
(u
1
+ u)R
2
R + u = 2u + . . . ,
f
2
=
_

3
16
(u
1
+ u)
2
+
1
4
_
R
2

3
4
(u
1
+ u)R
3
4
= u
2
+ . . . ,
f
3
=
1
8
(3u
2
+ 2 + 3u
2
)R
2

3
2
(u
1
+ u)R
3 5u
2
+ 6u
4
2(1 u
2
)
= 3u
2
+ . . . ,
f
4
=
3
32
(5u
3
+ 3u
1
+ 3u + 5u
3
)R
2

1
8
(15u
2
+ 14 + 15u
2
)R

15u
1
+ 4u 23u
3
+ 20u
5
8(1 u
2
)
=
10
3
u
3
+ . . . ,
f
5
=
u
3
1 u
2
= u
3
+ . . . ,
f
6
=
1
48
(u
1
+ u)R
4
+
1
6
R
3

1
4
(u
1
+ u)R
2
R u = 2u
3
+ . . . ,
f
7
=
1
2u
R
2

2
1 u
2
R
2u + 4u
3
1 u
2
= 2u
3
+ . . . ,
f
8
=
1
48
(u
1
+ u)R
4
+
1
6
R
3

3
8
(u
1
+ u)R
2

17
3 4u
2
+ 3u
4
2(1 u
2
)
2
R
3u 3u
3
+ 2u
5
2(1 u
2
)
2
= 2u
3
+ . . . ,
f
9
=
1
4
(u
1
u)R
2
+
1 + u
2
1 u
2
R +
3u u
3
1 u
2
= 2u + . . . ,
f
10
=
3
8
(u
2
u
2
)R
2
+
3u
1
2u + 3u
3
2(1 u
2
)
R +
3(1 +u
2
)
2(1 u
2
)
= 2u
2
+ . . . ,
f
11
=
3
4
(u
2
u
2
)R
2
+
3u
1
2u + 3u
3
1 u
2
R+
3 + 2u
2
15u
4
+ 6u
6
(1 u
2
)
2
= 6u
2
+. . . ,
f
12
=
1
16
(u
1
+ u)R
2

1
4
R
u + u
3
4(1 u
2
)
2
= u
3
+ . . . ,
f
13
=
1
4
(1 + 3u
2
)R
2

u 3u
3
1 u
2
R +
5u
2
3u
4
1 u
2
= 6u
2
+ . . . ,
f
14
=
1
u
R
2

4
1 u
2
R
4(u + u
3
)
1 u
2
=
20
9
u
5
+ . . . .
7 Description of the functions appearing in
the Maxwell elds (a conjecture)
The method of analyzing Maxwell equations in a neighbourhood of a freely
moving electric multipole leads, in case of a monopole and a dipole par-
ticle, to an interesting family of relatively simple functions of the variable
r. These functions are obtained from A
n
(r) = P
n
(z) contained in (46)
by consecutive application of the dierential operators contained in # and
the integral operator contained in curl
1

. The latter is based on solving

the dierential equation (30) and could a priori lead very quickly to non-
elementary functions. An unexpected result of our analysis consists in the
fact, that the functions arising here are polynomials of the universal quantity
R = log(
1u
1+u
) = log(
2ar
2+ar
), with coecients being rational combinations of
u = ar/2 itself. This fact leads us to a conjecture, that a new family of
special functions arises here, which makes the physical picture of the eld
relatively simple.
In this Section we present a conjecture about the smallest algebra of func-
tions f appearing in X
(k)
(f, Q), Y
(k)
(f, Q), when we solve Maxwell equations
with multipole particles, using the method proposed in this paper.
18
We set R
s
=
R
s
2
s
s!
. First we notice that only constant functions appear in
X
(0)
(remember that Y
(0)
= 0). Indeed, the leading term for the monopole
particle contains only P
0
(z) = 1 and the terms with X
(0)
produced by (39)
(40) vanish.
We dene H
k
as the linear span of functions f entering X
(k)
(f, Q), Y
(k)
(f, Q),
which appear when we consecutively use formulae (32) (33) or (39) (40),
starting from multipole elds (46). More precisely, we take the following
Denition. We set H
k
, k = 1, 2, . . ., as the smallest vector spaces such
that
u
k
, u
k+2
, . . . u
k
H
k
, (62)
and such that for any f H
k
one has s
(k)
(f) H
k
,
u
2
f
1 u
2
H
k
, uf
,u
H
k
, (1 u
2
)[u
2
f
,uu
+ 2uf
,u
k(k + 1)f] H
k
, (63)
u
2
f
,u
+ (k + 1)uf H
k+1
, u
2
f
,u
+ (k + 3)uf +
2u
3
f
1 u
2
H
k+1
, (64)
u
2
f
,u
+ kuf H
k1
, u
2
f
,u
(k 2)uf +
2u
3
f
1 u
2
H
k1
(k = 1). (65)
Remark. Due to condition (62) we have:
P
n
(z) = P
n
(
1
2
(u
1
+ u)) span{u
n
, u
n+2
, . . . , u
n
} ,
and, whence, functions f contained in D
(n2)
= X
(n)
(k
n
P
n
(z), Q) are in H
n
.
Also for n = 0 and f appearing in B
(1)
= Y
(1)
(u,
1
a
Qb) we have f = u H
1
.
It remains to consider operations curl
1

and # which yield B

(l)
and D
(l)
with
higher l. Due to (32)(33) and (39)(40), conditions (63)(65) imply that
also f appearing in those B
(l)
, D
(l)
belong to the corresponding H
k
.
It is easy to see that conditions (63)(65) can be reformulated in a simpler,
equivalent form:
f
1 u
2
H
k
, uf
,u
H
k
, (1 u
2
)[u
2
f
,uu
+ 2uf
,u
k(k + 1)f] H
k
, (66)
uf
1 u
2
H
k+1
, u
2
f
,u
+ (k + 1)uf H
k+1
, (67)
uf
1 u
2
H
k1
, u
2
f
,u
kuf H
k1
(k = 1). (68)
19
Conjecture: The space H
k
can be represented as
H
k
=
_
m>k
H
km
(increasing sequence of vector spaces), (69)
where H
km
are the direct sums:
H
km
=

j=0
H
kjm
, (70)
H
k0m
= span
_
u
k
(u
2
1)
m
,
u
k+2
(u
2
1)
m
, . . . ,
u
k+2m
(u
2
1)
m
_
, m = 0, 1, 2, . . . ,
(71)
H
k,2l1,m
= span
_
R
2l1
u
k+1
(u
2
1)
m
,
R
2l1
u
k+3
(u
2
1)
m
, . . . ,
R
2l1
u
2mk1
(u
2
1)
m
_
, (m > k),
(72)
H
k,2l,m
= span
_
w
l
k
(R, u)
u
k
(u
2
1)
m
,
w
l
k1
(R, u)
u
k2
(u
2
1)
m
, . . . ,
w
l
1
(R, u)
u
k+2
(u
2
1)
m
,
R
2l
u
k
(u
2
1)
m
,
R
2l
u
k+2
(u
2
1)
m
, . . . ,
R
2l
u
2mk
(u
2
1)
m
,
w
l
1
(R, u
1
)u
2mk+2
(u
2
1)
m
,
w
l
2
(R, u
1
)u
2mk+4
(u
2
1)
m
, . . . ,
w
l
k
(R, u
1
)u
2m+k
(u
2
1)
m
_
, (m > k), (73)
l = 1, 2, . . ., with w
l
k
given recursively by w
l
0
(R, u) = R
2l
,
w
l
k
(R, u) =
_
_
min(k,2l1)

j=1
A
kj
w
l[j/2]
kj
(R, u)u
2j2
_
_
+(2k 3)!!R
2l1
u
2k1
, (74)
k, l = 1, 2, . . . (here, we put (1)!! = 1, so that (1)!! 1 = 1!!), for some
real numbers A
kj
, k = 1, 2, . . ., j = 1, 2, . . . , k. Moreover, we conjecture that
A
k1
= 2k 1, A
k2
= 1, A
kk
= (2k 3)!!. Unfortunately, we were not able to
nd a general formula for A
kj
.
Remark 1. We know only a recursive procedure to nd A
kj
: Suppose
we have found all A
rj
, r < k. Then we can compute w
l
r
, r < k (cf. (74)).
We set
f =
w
l
k1
(R, u
1
)u
2m+k2
(u
2
1)
m
,
20
i.e. f is the last but one element of (73). Then we compute g H
k
by means
of the last formula of (66). Next we decompose g into the bases of H
k,j,m+1
,
j 2l, where w
l
k
is given by (74). This gives us an equation which may
be used to nd coecients A
kj
, j = 1, 2, . . . , k. That kind of procedure was
implemented by us for k 10, using the symbolic calculus provided by the
program MAPLE 8. In particular, we have obtained the following results for
k 6:
A
11
= 1, (75)
A
21
= 3, A
22
= 1, (76)
A
31
= 5, A
32
= 1, A
33
= 3, (77)
A
41
= 7, A
42
= 1, A
43
= 50, A
44
= 15, (78)
A
51
= 9, A
52
= 1, A
53
= 1273/9, A
54
= 485/9, A
55
= 105, (79)
A
61
= 11, A
62
= 1, A
63
= 62564/225, A
64
= 2703/25, A
65
= 9985/3, A
66
= 945.
(80)
Using (74), one gets that
w
l
k
(R, u) =
min(k,2l1)

s=0
g
ks
(u)R
2ls
, (81)
where g
ks
are polynomials. In particular, we have:
w
l
0
(R, u) = R
2l
, (82)
w
l
1
(R, u) = R
2l
+ uR
2l1
, (83)
w
l
2
(R, u) = 3R
2l
+ (3u + u
3
)R
2l1
+ u
2
R
2l2
, (84)
w
l
3
(R, u) = 15R
2l
+(15u+5u
3
+3u
5
)R
2l1
+(6u
2
+3u
4
)R
2l2
+u
3
R
2l3
. (85)
(Observe that, due to (81), for l = 1 we omit the last term in w
l
2
and the last
two terms in w
l
3
.)
Remark 2. One can prove that for A
k1
= 2k 1 and any value of
the remaining coecients A
kj
, the elements (71)(73), spanning each vector
space H
kjm
are linearly independent and the direct sum condition is obvious.
Moreover, using (74), H
km
form an increasing sequence of vector spaces and
therefore H
k
are well dened. The elements (62) belong to H
k00
H
k0m

H
k
(m > k). Using once again (74), it can be also checked that the rst
operations of (66)(68) provide elements of H
k
, H
k+1
and H
k1
, respectively.
21
It remains to nd the formula for A
kj
such that the remaining four relations
of (66)(68) are fullled, to prove that s
(k)
(f) H
k
for f H
k
and to show
the minimality of H
k
.
Remark 3. Elements of H
k
are rational functions of u and R. However,
the generators of these spaces, proposed in formulae (71)(73) are quite com-
plicated. But splitting the nominators into monomials doesnt simplify the
situation, because application of s
(k)
(which must preserve H
k
) to the com-
ponents obtained in that way leads to functions which seem to be in general
not rational in u and R. Thus the form (71)(73) of the generators seems to
be the simplest one.
Appendices
A Properties of elds X and Y.
We assume r = 0. Using (23)(24), after some computations we get

l
X
k
= [r
2
f
,rr
+ (2n + 3)rf
,r
(n + 1)(n + 3)f]
x
l
x
k
r
n+5
H+
[(n + 1)f rf
,r
]
1
r
n+3

kl
H + n[(n + 1)f rf
,r
]
x
k
Q
l
r
n+3
+
(rf
,rr
nf
,r
)
x
l
Q
k
r
n+2
+
f
,r
r
n
(n 1)Q
kl
,

l
Y
k
= [
rf
,r

+
a
2
r
2
f
2
2

(n + 1)f

]
x
l
E
k
r
n+3
+
f
r
n+1

kl
m
Q
m
+
(n 1)f
r
n+1
F
kl
,
where H = Q
i
1
...in
x
i
1
x
i
2
. . . x
in
, Q
k
= Q
ki
2
...in
x
i
2
. . . x
in
, Q
kl
= Q
kli
3
...in
x
i
3
. . . x
in
,
E
k
=
k
jm
x
j
Q
m
, F
kl
=
k
jm
x
j
Q
ml
. That implies (25), (26)(27),
x
l
X
l
=
(n + 1)f
r
n+1
H, x
l
Y
l
= 0 (86)
and (cf. (16))

X
k
=
1
2
[r
2
f
,rr
+ rf
,r
(n + 1)
2
f]
x
k
(b
l
x
l
)H
r
n+3
+
22

1
2
[rf
,r
+ (n + 1)f]
b
k
H
r
n+1

1
2
[(n 2)rf
,r
n(n + 1)f]
(b
m
Q
m
)x
k
r
n+1
+

1
2
[rf
,rr
+ (n + 2)f
,r
]
(b
l
x
l
)Q
k
r
n
+
n 1
2
f
,r
r
n2
b
l
Q
kl
, (87)

Y
k
=
1
2
[
rf
,r

+
a
2
r
2
f
2
2
+
(n + 3)f

]
E
k
(b
l
x
l
)
r
n+1
+
f
2r
n1
G
k
+
(n 1)f
2r
n1
b
l
F
kl
+
fx
k
r
n+1
(b
l
E
l
), (88)
where G
k
=
k
lm
b
l
Q
m
. Let us notice that Q
m
m
= 0 (Q is traceless) implies
0 = 2
ms
Q
ms

al
k
x
a
b
l
= D
ijs

ijs

al
k
x
a
b
l
,
where D
ijs
=
ij
m
Q
ms
. That and

ijs

alk
=
ia
(
jl

sk

jk

sl
) +
il
(
jk

sa

ja

sk
) +
ik
(
ja

sl

jl

sa
)
give

alm
x
a
b
l
Q
mk
= b
l
F
kl
+ G
k
. (89)
Using (36)(38) and (89), we obtain
X
k(n+1)
(, b Q) = [(n + 2) r
,r
]
x
k
r
n+4
[(b
l
x
l
)H
n
2n + 1
r
2
(b
m
Q
m
)] +

,r
r
n+1
{
1
n + 1
[b
k
H + nQ
k
(b
l
x
l
)]
2
(n + 1)(2n + 1)
[nx
k
(b
m
Q
m
) +
n(n 1)
2
r
2
b
m
Q
mk
]},(90)
Y
k(n)
(, bQ) =

nr
n+1
[nb
k
H(n1)x
k
(Q
m
b
m
)n(b
l
x
l
)Q
k
+(n1)r
2
b
l
Q
kl
],
(91)
X
k(n1)
(, bQ) = (n r
,r
)
x
k
(b
m
Q
m
)
r
n+2
+

,r
r
n1
b
m
Q
mk
(n 2), (92)
Y
k(n+1)
(, bQ) =

r
n+2
[
1
n + 1
(
k
jl
x
j
b
l
)H+
n
n + 1
E
k
(b
l
x
l
)
n(n 1)
(n + 1)(2n + 1)
r
2
b
l
F
kl
],
(93)
X
k(n)
(, b Q) =

,r
r
n
[G
k

n 1
n
b
l
F
kl
] [(n + 1) r
,r
]
x
k
(b
l
E
l
)
r
n+3
, (94)
23
Y
k(n1)
(, bQ) =

r
n
b
l
F
kl
(n 2). (95)
Computing
l
km

m
ij

j
ac
x
k
x
i
Q
a
b
c
by means of
lk
m

mij
=
li

kj

lj

ki
or

mi
j

jac
=
ma

ic

mc

ia
, one obtains
(
k
jl
x
j
b
l
)H = (b
l
x
l
)E
k
x
k
(b
l
E
l
) r
2
G
k
. (96)
Proof of Theorem 1. Comparing (87)(88) with (90)(95) and using (96),
we get

X
k
= X
k(n+1)
(, b Q) + Y
k(n)
(, b Q) + X
k(n1)
(, bQ), (97)

Y
k
= Y
k(n+1)
(, b Q) + X
k(n)
(, b Q) + Y
k(n1)
(, bQ), (98)
for =
1
2
[r
2
f
,r
+ (n + 1)rf], = g
1
, =
n
2
1
2n(2n+1)
(r
2
f
,r
+ nrf), = g
2
,
=
1
2(n+1)
fr
2

, = g
3
.
Using (23)(24), we obtain that

X
k
gives the last two terms in (34) while

Y
k
gives the last three terms in (35). That, (15) and (97)(98) prove (34)
(35). Q.E.D.
B Proof of the properties of A
n
and B
n
The quantity
z
k
=
_
1
2
_
2
ar
+
r
2a
__
k
= 2
k1
a
k
r
k
+ . . .
is of order r
k
. Therefore P
n
(z) (with the highest term of order z
n
) is a
Laurent polynomial of order r
n
and analogously v
n1
is of order r
n+1
.
Moreover,
1
2
log
z + 1
z 1
= log
1 + u
1 u
= ar + . . . (99)
and, therefore, B
n
(r) can be written as a Laurent series in r, starting from (at
least) r
n+1
. Inserting this series into h
(n)
(B) = 0, using (28) and denoting
the order of B
n
(r) by l n + 1, we get that vanishing of the r
l2
term in
h
(n)
(B) implies l = n (which is impossible) or l = n + 1. Hence, B
n
(r) is
of order r
n+1
.
24
C The transformation of 2
n
poles (n = 0, 1, 2)
Let I

be the current density in the Fermi frame (coordinates ) and J

be
the corresponding current density in the modied Fermi frame (coordinates
x). Then
J

=
x

det
_

x
_

.
Integrating with the test function f(x), we obtain
S

_
J

(x)f(x)d
3
x =
_
x

| det(

x
)|f(x)d
3
x =
_
x

()

f()d
3
,
(100)
where

f() = f(x()). Dierentiating (3), one obtains
x
k

l
= (
k
l
+ a
k

l
)M (
k
+
1
2
a
k

2
)(a
l
+
1
2
a
2

l
)M
2
, (101)
x
k

0
=
1
2
a
k

2
M (
k
+
1
2
a
k

2
)( a
m

m
+
1
2
a
m
a
m

2
)M
2
, (102)
where M = (1 +a
k

k
+
1
4
a
2

2
)
1
. Moreover,
x
0

l
= 0,
x
0

0
= 1. Thus
S
0
=
_
I
0
()

f()d
3
, (103)
S
k
=
_
L
k
()

f()d
3
, (104)
where
L
k
=
x
k

l
I
l
+
x
k

0
I
0
. (105)
Suppose that I

= I

(n)
(, Q) is the standard 2
n
-pole, given by the right
hand sides of the formulae (41)(42) with respect to the coordinates

. We
shall calculate the corresponding value of J

= J

(n)
by means of (100) in
the case of a monopole, dipole and quadrupole, i. e. for n = 0, 1, 2 (for n = 0
we assume Q = const.).
For n = 0 I
0
(0)
(, Q) = 4Q
(3)
, I
k
(0)
(, Q) = 0, hence S
0
= 4Q

f(0) =
4Qf(0), L
k
= 0, S
k
= 0,
J

(0)
= I

(0)
(x, Q), (106)
25
where I

(n)
(x, Q) is given by (41)(42) with respect to the coordinates x

.
For n = 1 I
0
(1)
(, Q) = 4Q
k

(3)
, I
k
(1)
(, Q) = 4

Q
k

(3)
, hence S
0
=
4Q
k
(
k

f)(0) = 4Q
k
(
k
f)(0), L
k
= 4

Q
k

(3)
, S
k
= 4

Q
k

f(0) = 4

Q
k
f(0),
J

(1)
= I

(1)
(x, Q). (107)
For n = 2 I
0
(2)
(, Q) =
4
3
Q
kl

(3)
, I
k
(2)
(, Q) =
4
3

Q
kl

(3)
, hence
S
0
=
4
3
Q
kl
(
k

l

f)(0) =
4
3
Q
kl
(
k

l
f)(0)
8
3
Q
kl
a
l
(
k
f)(0), L
k
=

8
3
Q
km
a
m

(3)

8
3

Q
km
a
m

(3)

4
3

Q
kl

(3)
, S
k
=
8
3
(Q
kl
a
l
) f(0) +
4
3

Q
kl
(
l
f)(0),
J

(2)
= I

(2)
(x, Q) +I

(1)
(x, P), (108)
where the dipole charge P
k
=
2
3
Q
kl
a
l
. Therefore the Fermi frame quadrupole
has a dipole component in the modied Fermi frame.
D Integrals, distributions and the proof of
Theorem 2.
First we compute the following integral over twosphere:
S
klm
=
_
S
2
(r)
_
x
r
_
2k
_
y
r
_
2l
_
z
r
_
2m
d, k, l, m = 0, 1, 2, . . . , (109)
where r = (x
2
+ y
2
+ z
2
)
1/2
is the radius of the twosphere, d = sin dd,
r, , are the spherical coordinates.
Proposition 1. One has
S
klm
=
(2k 1)!!(2l 1)!!(2m1)!!
(2k + 2l + 2m + 1)!!
4.
Proof. Let A > 0. We use
_
R
3
x
2k
e
Ax
2
y
2l
e
Ay
2
z
2m
e
Az
2
dxdydz =
_

0
r
2(k+l+m)
e
Ar
2
r
2
S
klm
dr,
_
R
x
2k
e
Ax
2
dx = (d/dA)
k
_
R
e
Ax
2
dx =
(2k 1)!!
2
k

1/2
A
k1/2
and the analogous formulae for y, z and r. Q.E.D.
26
Corollary.
_
S
2
(r)
x
i
1
x
i
2
. . . x
i
2n
d =
4r
2n
(2n + 1)!

2n

i
(1)
i
(2)
. . .
i
(2n1)
i
(2n)
. (110)
Proof. We get nonzero results only if i
s
= 1 in 2k cases, i
s
= 2 in 2l cases,
i
s
= 3 in 2m cases, k + l + m = n. Then the left hand side gives S
klm
r
2n
and the sum gives the factor (2k 1)!!(2l 1)!!(2m1)!!2
n
n!. Next we use
(109). Q.E.D.
Proposition 2. Let Q
i
1
...in
be a symmetric traceless tensor, S
j
1
...jm
be
any tensor. We set Qx x = Q
i
1
...in
x
i
1
x
in
and similarly for S. Then for
m = n
_
S
2
(r)
(Qx x)(Sx x) d = Q
i
1
...in
S
i
1
...in
4(n!)
2
2
n
(2n + 1)!
r
2n
, (111)
while for m < n the left hand side of (111) equals zero.
Proof. Let m = n. Then the left hand side of (111) is equal to Q
i
1
...in
S
i
n+1
...i
2n
multiplied by (110) and summed over all i
k
. We may rst sum over all i
k
and then over
2n
. In such a case, due to the traceless condition for Q,
nonzero terms in the sum over are obtained if for each k = 1, . . . , n one of
elements (2k 1), (2k) belongs to {1, . . . , n} and the other one belongs to
{n + 1, . . . , 2n}. We get
4r
2n
(2n + 1)!
2
n
n!

i,j
Q
i
1
...in
S
j
1
...jn

i
(1)
j
1
. . .
i
(n)
jn
,
which due to the symmetry of Q gives the right hand side of (111). For
m < n a similar arguments show that all terms vanish (we use the traceless
condition for Q if m+n is even and the antisymmetry of the expression under
the integral if m + n is odd). Q.E.D.
Let n = 0, 1, 2, . . ., Q be a tracelless symmetric tensor of rank n, f be an
even analytic function of r near 0 (f is a constant for n = 0). We set
D
nQf
= f
Qx x
r
2n+1
and dene F as the linear span of all D
nQf
. The elements F F become
distributions if for any test function we set
< F, >=
_

0
dr
_
S
2
(r)
r
2
Fd = lim
R0
+
_
R
3
\K(0,R)
Fd
3
x. (112)
27
That is well dened because setting F = D
nQf
and using
=
n1

k=0
1
k!

i
1
...i
k
x
i
1
. . . x
i
k
+ O(r
n
), (113)

i
1
...i
k
= (
i
1
. . .
i
k
)(0), (114)
we obtain
_
S
2
(r)
r
2
Fd =
_
n1

k=0
1
k!
f
r
2n1
_
S
2
(r)
(Qx x)(x x)d
_
+ O(r) = O(r),
since all the integrals over S
2
(r) vanish due to k < n and Proposition 2.
For F F let
i
denote the partial derivative in the sense of distributions,

C
i
- the partial derivative as function,

R
i
F =
i
F
C
i
F. (115)
We dene l
i
as tensor of rank 1 such that (l
i
)
j
=
ij
.
Proposition 3.
x
i
D
nQf
= D
n+1,l
i
Q,r
2
f
+
n
2n + 1
D
n1,l
i
Q,f
,

C
i
D
nQf
= D
n+1,l
i
Q,(2n+1)f+rf,r
+
n
2n + 1
D
n1,l
i
Q,f,r/r
.
Proof. We use x
i
(Qx x) = (l
i
Q)(x x) +
n
2n+1
r
2
Q
i
, where Q
i
=
Q
ii
2
...in
x
i
2
. . . x
in
= (l
i
Q)(x. . . x). Q.E.D.
Thus for F F,
C
i
F and
R
i
F are distributions.
Proposition 4. For F F and a test function
<
R
i
F, >= lim
R0
+
_
S
2
(R)
Fx
i
Rd.
Proof. Due to (112)
<
i
F, >= < F,
i
>= lim
R0
+
_
R
3
\K(0,R)
(
C
i
F)d
3
x lim
R0
+
_
R
3
\K(0,R)

C
i
(F)d
3
x,
where the rst term equals <
C
i
F, >. Setting (G
i
)
j
= F
ij
and using
(115), one gets
<
R
i
F, >= lim
R0
+
_
R
3
\K(0,R)
divG
i
d
3
x
28
= lim
R0
+
_
S
2
(R)
(G
i
)
j
x
j
R
R
2
d = lim
R0
+
_
S
2
(R)
Fx
i
Rd. Q.E.D.
Proposition 5.

R
i
D
nQf
= [lim
r0
f(r)](1)
n1
4n
(2n + 1)!!
Q
iI

(3)
,
where I = (i
2
, . . . i
n
),
I
=
i
2
. . .
in
.
Proof. Using Proposition 4 for F = D
nQf
, (113), setting c = lim
r0
f(r),
(
i
)
i
1
...i
k
i
k+1
=
i
1
...i
k

i
k+1
i
and using Proposition 2, we get

i
(x x) =
i
1
...i
k
x
i
1
x
i
k
x
i
,
<
R
i
D
nQf
, >=
n1

k=0
1
k!
lim
R0
+
_
S
2
(R)
f(R)

i
(x x)Q(x x)
R
2n
d =
c
(n 1)!
Q
ii
2
...in

i
2
...in
4(n!)
2
2
n
(2n + 1)!
.
Moreover, we use <
I

(3)
, >= (1)
n1

i
2
...in
(cf. (114)) and
(2n+1)!
n!2
n
=
(2n + 1)!! Q.E.D.
Proposition 6. Assuming f L
n0
(cf. Section 4) and setting g = fr
n
(which is regular at 0), we obtain
X
k(n)
(f, Q) = D
n+1,l
k
Q,(2n+1)grg,r
+
n + 1
2n + 1
D
n1,l
k
Q,g,r/r
,
Y
k(n)
(f, Q) = D
n,l
k
Q,g/
.
Proof. By a direct computation. Q.E.D.
Proposition 7.

R
i
X
k(n)
(r
n
, Q) = (1)
n
4(2n + 1)
(2n + 3)!!

[
ik
Q
J

(3)
+ nQ
i
I

kI

(3)

n(n 1)
2n + 1
Q
ik
L

m
mL

(3)

2n
2n + 1
Q
k
I

iI

(3)
],

R
i
X
k(n)
(r
(n2)
, Q) = (1)
n
4(n 1)
(2n 1)!!
Q
ik
L

(3)
,

R
i
X
k(n)
(f, Q) = 0 if lim
r0
(fr
n2
) = 0,
29

R
i
Y
k(n)
(r
n
, Q) = (1)
n
4
(2n + 1)!!
[(n 1)
k
ma
Q
mi
L

aL

(3)
+
ik
m
Q
m
I

(3)
],

R
i
Y
k(n)
(f, Q) = 0 if lim
r0
(fr
n
) = 0.
Proof. We use Proposition 6 and Proposition 5. Q.E.D.
Remark. In the case of n = 0, 1 some equations related to the rst
equation of Proposition 7 were presented in (6)(7) of [4].
Next, we obtain that in addition to (25), (26)(27), (34)(35), the op-
erations div, curl

and # acting on the elds X and Y yield the following

distribution parts:
Proposition 8.
div
R
X
(n)
(r
n
, Q) = (1)
n
4(n + 1)
(2n + 1)!!
Q
J

(3)
,
[curl
R

X
(n)
(r
n
, Q)]
k
= (1)
n
4n
(2n + 1)!!

ki
j
Q
iI

jI

(3)
,
[X
(n)
(r
n
, Q)
#R
]
k
= (1)
n
4n(n
2
1)
(2n + 1)!!
Q
ki
L
b
i

(3)
,
_
curl
R

Y
(n)
(r
n
, Q)

k
= (1)
n
4(n + 1)
(2n + 1)!!
Q
k
I

(3)
,
div
R
Y
(n)
(r
n
, Q) = Y
(n)
(r
n
, Q)
#R
= 0,
while for f such that lim
r0
(fr
n
) = 0 the distribution parts vanish.
Proof. We use Proposition 7 and notice that the operator d/d appearing
in # (cf. (15)) gives no additional terms. Q.E.D.
Let us notice that
0
= D
011
=
1
r
is the classical monopole potential,

0
= 4
(3)
. By induction one has Q
I

0
= (1)
n
(2n 1)!!D
nQ1
(no
distribution parts! - cf. Proposition 5) and therefore
n
= D
nQ1
satises

n
= (1)
n
4
(2n 1)!!
Q
I

(3)
(cf. (41)) and
n
is the classical 2
n
pole potential. Using (115) and Propo-
sitions 3,5 and 6, we obtain that the corresponding electric eld is given
by
D
cl
n
= grad
n
, (D
cl
n
)
k
= X
k(n)
(r
n
, Q) + (D
d
)
k
, (116)
30
where
(D
d
)
k
= (1)
n
4n
(2n + 1)!!
Q
k
I

(3)
(cf. (45) and Remark 2 in Section 5). After some computations one obtains
divD
d
= (1)
n
4n
(2n + 1)!!
Q
J

(3)
, (117)
(curl

D
d
)
k
= (1)
n+1
4n
(2n + 1)!!

ki
j
Q
iI

jI

(3)
, (118)
(D
#
d
)
k
= (1)
n
4n
(2n + 1)!!

Q
k
I

(3)
. (119)
Proof of Theorem 2. Let us rst omit the terms containing
(3)
and its
derivatives. Then D
(l)
and B
(l)
consist of X and Y elds and therefore (cf.
(25)) divD
(l)
= divB
(l)
= 0, which proves (17)(18). Moreover, applying
curl

to (47)(48), we obtain
curl

B
(k+1)
= D
(k)
#
, k = n 2, n, n + 2, . . . , (120)
curl

D
(k+1)
= B
(k)
#
, k = n 1, n + 1, n + 3, . . . . (121)
But one has also
curl

D
(n2)
= 0 (122)
(we use (27) and h
(n)
[P
n
(z)] = 0). Inserting (43)(44) into the both sides of
(19)(20) and using (120)(122), one proves that (19)(20) are fullled.
Let us notice that
(3)
and its derivatives appear in the sources (41)(42),
in D
d
and also when acting by div, curl

or # on X
(k)
(r
k
, P), Y
(k)
(r
k
, P).
Using (32)(35), one shows that X
(k)
(r
k
, P), Y
(k)
(r
k
, P) appear only in
D
(n2)
which contains X
(n)
(r
n
, Q) and in B
(n1)
which contains

n
2
1
2n + 1
Y
(n1)
(r
(n1)
, bQ) Y
(n)
(r
n
,

Q).
Therefore divD contains additionally
divD
d
+ div
R
X
n
(r
n
, Q) = (1)
n
4
(2n 1)!!
Q
J

(3)
= J
0
,
31
divB has no additional terms, D
#
curl

B contains additionally (cf. Propo-

sition 8, (119) and (42))
D
d
#
+X
(n)
(r
n
, Q)
#R
+
n
2
1
2n + 1
curl
R

Y
(n1)
(r
(n1)
, bQ)
+curl
R

Y
(n)
(r
n
,

Q) = J,
B
#
+ curl

D contains additionally curl

D
d
+ curl
R

X
(n)
(r
n
, Q) = 0, hence
the Maxwell equations are satised also in the distributional sense. The last
statement of Theorem 2 follows from the properties of #, curl
1

and X and
Y elds. Q.E.D.
Acknowledgements
We thank Dr Marcin Koscielecki for fruitful discussions.
References
[1] M. Born, Ann. Phys. (Leipzig), 30 (1909), 1.
[2] P. Chrusciel, J. Jezierski and J. Kijowski, Hamiltonian Field Theory in
the Radiating Regime, volume 70 of the series: Springer Lecture Notes in
Physics, Monographs (2001).
[3] P. A. M. Dirac, Classical theory of radiating electrons, Proc. Roy. Soc. A
167 (1938), 148169.
[4] C. P. Frahm, Some novel deltafunction identities, Am. J. Phys. 51
(1983), 826829.
[5] H. P. Gittel, J. Kijowski, E. Zeidler, The relativistic dynamics of the
combined particle-eld system in renormalized classical electrodynamics,
Commun. Math. Phys. 198 (1998), 711736.
[6] R. Haag, Die Selbstwechselwirkung des Elektrons, Naturforsch. 10 a
(1955), 752761.
[7] J. Kijowski, Electrodynamics of moving particles, Gen. Relat. Grav.
26 (1994), 167201. See also J. Kijowski, On electrodynamical self
interaction, Acta Phys. Pol. A 85 (1994), 771787.
32
[8] J. Kijowski, M.Koscielecki, Asymptotic expansion of the Maxwell eld in
a neighbourhood of a multipole particle, Acta Phys. Pol. B 31 (2000),
1691 1707.
[9] J. Kijowski, M.Koscielecki, Algebraic description of the Maxwell eld sin-
gularity in a neighbourhood of a multipole particle, Rep. Math. Phys. 47
(2001), 301311.
[10] J. Kijowski, M.Koscielecki, On stability of renormalized classical elec-
trodynamics, physics/0305123, Acta Physica Polonica B 36 (2005) p. 75
107.
[11] J. Kijowski, P. Podles, Born renormalization in classical Maxwell elec-
trodynamics, J. Geom. Phys. 48 (2003), 369384.
[12] F. Rohrlich, Classical Charged Particles. Foundations of Their Theory,
AddisonWesley, Reading 1965.
[13] W. Thirring, A course in mathematical physics, Vol. 2 Classical eld
theory, SpringerVerlag, New York Wien 1979.
[14] Z. Ya. Turakulov, Electromagnetic eld of a charge moving with constant
acceleration, J. Geom. Phys. 14 (1994), 305308.
[15] E. T. Whittaker, G. N. Watson, A course of modern analysis, Cambridge
University Press, Cambridge 1946.
33