Method of Images for Magnetostatics

William Talmadge

Fig. 1:
Cross section of a truncated cone with magnetization M ~ real embeded in a re-
gion of low permeability µ1 . The cone is a distance d from a region of highly
permeable material µ2 . Both regions are separated by a plane defined by the
normal vector ~n. The associated image magnetization M~ image is also displayed.
~
(Note the pseudovector nature of M .)

Introduction
In electrostatics you may have encountered a method for solving field equations
known as the method of images. In electrostatics this method is often exploited
due to a physical property of conducting surfaces. A conducting surface is nec-
essarily an equipotential of an electric potential field. Because of this property a
conducting surface can be replaced with a boundary having properties inherited
from the equipotential that coincides with it. This equipotential (and hence the
boundary) is well defined enough to be used as a boundary condition for electric
field equations.
In electrostatics this method involves creating a mirror image of the electric
charge distribution about a symmetry plane and then reversing the polarity of

1
 2

this image. The virtual field created by the virtual charge distribution is actually
the solution to the field generated by conducting surface when restricted to the
problem domain. Thus the full electric field solution can be found by adding
the virtual field within the problem domain and the real field.
This method can be used in magnetostatics despite some difficulties to over-
come. First there is no such thing as an isolated “magnetic charge.” Also there
is typically not any actual case in magnetostatics where an exact equipotential
surface coincides with a material boundary. However, in the case of highly per-
meable materials the material boundaries can approximate equipotentials. This
allows the method of images to be used in some cases as a good approximation.
Before we can conduct such an analysis we first have to characterize the nature of
the potential field implied by stating the surface of a highly permeable material
is approximately an “equipotential.” After establishing the potential field we
must investigate the properties of the potential field to ensure that it is useful
in the method of images.

Method of Images

Magnetizing Field Potential

The potential field we will investigate is the potential field associated with the
~ as it is often expressed in Maxwell’s equations. We do not
magnetizing field, H
consider the magnetic field (B-field) as a candidate for our potential field as in
general it does not have a scalar potential field.
The magnetizing field is defined in terms of the magnetic field B ~ and magneti-
~
zation field M as,
~ = 1B
H ~ −M ~ (1)
µ0
In the case of magnetostatics, Maxwell’s equation,
~
∂D
~ = J~f +
∇×H (2)
∂t
becomes, in the absence of free currents,
~ =0
∇×H (3)

This implies from the Helmholtz decomposition theorem,

F~ = −∇ϕ + ∇ × A
~ (4)

that the magnetizing field H~ is irrotational, thus we can define it entirely in

terms of the gradient of some scalar potential as,
~ = −∇ϕ
H (5)
 3

~ as the “magnetizing scalar potential,”

We’ll define this potential ϕH for H
−∇ϕH = H ~ (6)
we’ll regard it as a mathematical tool and not consider its physical interpreta-
tion.

Magnetizing Field Boundary Conditions

As stated earlier our objective is to replace the material boundary with an
equipotential that we can use as a symmetry plane and boundary surface. To
do this we will construct a model for an idealized, highly permeable and linear
ferromagnetic medium boundary. The space outside of the material boundary
has a permeability µ1 . The permeable material has a permeability µ2 such that
µ2  µ1 (See figure 1). In linear media the magnetic field and magnetizing field
are related as,
~ = µH
B ~ (7)
We can determine the boundary condition at the equipotential surface that will
replace the idealized material boundary by considering the limiting behavior of
the boundary conditions as µ2 → ∞.
The magnetizing field boundary conditions for magnetostatics are,
~ 1 · ~n = µ2 H
µ1 H ~ 2 · ~n (8)
~ 1 × ~n = H
H ~ 2 × ~n (9)
Where ~n is a vector normal to the boundary between the two materials, directed
from µ2 to µ1 . We want to know the angles between the normal and the mag-
netizing field on each side of the boundary. To do this we will first expand the
vector products in eq. 8 and eq. 9 to their trigonometric forms,
µ1 H1 cos θ1 = µ2 H2 cos θ2
H1 sin θ1 = H2 sin θ2
Dividing these equations gives,
µ1 cot θ1 = µ2 cot θ2
We now only need to examine the limiting behavior of this expression. In the
limiting case µ1 is a very small, non-vanishing number and µ2 → ∞. This
implies,
cot θ1 → ∞
which is true when θ1 → 0 (the angle between H ~ 1 and ~n is 0). This implies
that in our idealized model of an interface between a highly permeable, very low
permeability linear materials; it is a good approximation to assume that H ~ 1 is
everywhere normal to the boundary such that,
H~ 1 · ~n = H1 (10)
 4

Magnetizing Field Boundary as an Equipotential

With the property of of the magnetizing field at the boundary expressed by eq.
10 we can show that the boundary is an equipotential. By eq. 3 it follows that
we have an irrotational and conservative field. Thus we can use the gradient
theorem, ˆ
ϕ(~b) − ϕ(~a) = ∇ϕ · d~r (11)
L
to investigate the difference in potential between arbitrary points on the material
boundary.
We can express the gradient theorem for our particular boundary model by first
introducing two arbitrary coordinates restricted to the boundary plane; ~q1 and
~q2 . The path ~r(t) is a path restricted to the boundary plane over which the line
integral in eq. 11 is integrated.
ˆ
ϕH (~q2 ) − ϕH (~q1 ) = ~ · d~r
−H
L

considering the property expressed in eq. 10, this becomes

ˆ
ϕH (~q2 ) − ϕH (~q1 ) = −H~n · d~r
L

The symmetry plane normal ~n is everywhere orthogonal to ~r so the line integral

goes to zero and we are left with,

ϕH (~q2 ) = ϕH (~q1 )

it follows that the entire boundary is an equipotential. To use this equipotential

as a boundary for the problem domain we need an exact value for this equipo-
tential. If assign a potential of zero to points at infinity then all points on the
symmetry plane boundary must also be zero since this plane extends to infinity
as well.

Uniqueness of the Magnetizing Potential Field

The uniqueness theorem for the Poisson equation allows us to establish that a
particular solution is unique if a specific set of boundary conditions are satisfied.
In other words the uniqueness theorem allows us to determine how much infor-
mation we need to know about the boundary conditions of a problem domain
to know for certain a field solution has only one solution.
Before determining the uniqueness conditions for our field we need a Poisson
field equation in terms of quantities that are known. Specifically we would like
to know the magnetizing scalar potential as it relates to a given magnetization
distribution.
 5

Poisson’s equation is a partial differential equation of the form,

∇2 ϕ = f

where ϕ and f are real or complex valued functions. We will begin by deriving
the form of the Poisson equation for the magnetizing scalar potential. We begin
by taking the divergence of eq. 1 and noting that,
~ =0
∇·B

We are then left with the relation,

~ = −∇ · M
∇·H ~ (12)

When we substitute the definition for the magnetizing field potential into eq.
12 we obtain,
∇ · (−∇ϕH ) = −∇ · M~
~
∇2 ϕH = ∇ · M (13)
Eq. 13 is the Poisson equation we must find uniqueness conditions for.
If we assume that a solution to eq. 13 is not unique then there must be at least
two solutions, ϕ1 and ϕ2 . If there is no difference between these solutions then
we have established conditions for uniqueness. We begin by defining a potential
that is the difference between solutions,

φ = ϕ2 − ϕ1

We then apply the Laplacian operator to this potential,

∇2 φ = ∇2 ϕ2 − ∇2 ϕ1

We can write two magnetizing Poisson equations for our candidate solutions. It
is important to note that although the solutions are assumed to be unique they
are both associated with the same magnetization distribution. Therefore these
equations can be expressed as,
~
∇2 ϕ 1 = ∇ · M
~
∇2 ϕ 2 = ∇ · M

Thus,
~ −∇·M
∇2 φ = ∇ · M ~

Giving us a field equation for the solution difference potential,

∇2 φ = 0 (14)
 6

We now will derive another useful identity that can be used in the divergence
theorem. We begin with the following product rule,
~ = f (∇ · A)
∇ · (f A) ~ +A
~ · (∇f )

~ = ∇φ. Expanding we have

then substitute f = φ and A

∇ · (φ∇φ) = φ(∇ · ∇φ) + ∇φ · ∇φ

∇ · (φ∇φ) = φ(∇2 φ) + (∇φ)2

Note that from eq. 14 the term φ(∇2 φ) → 0 thus this expression simplifies to

∇ · (φ∇φ) = (∇φ)2 (15)

Note that the term ∇ · (φ∇φ) is in the form that the divergence theorem
ˆ ˆ
~
(∇ · A)dV = ~ · d~a
A
V ∂V

The divergence theorem can be applied with A~ = φ∇φ, V is the problem domain
volume (the area above the symmetry plane in figure 1), d~a is an element of the
boundary surface enclosing the domain, ∂V .
ˆ ˆ
(∇ · (φ∇φ))dV = (φ∇φ) · d~a
V ∂V

From eq. 15 the left term simplifies this expression to,

ˆ ˆ
(∇φ)2 dV = (φ∇φ) · d~a
V ∂V

Our first condition is that φ exist, be well defined and real valued everywhere.
It follows that the the integrand of the volume integral must have the property,

(∇φ)2 ≥ 0

If the volume integral is non-vanishing then we do not have conditions for a

unique solution. We are only interested situations where the volume integral is
vanishing. Thus our boundary counstraints are characterized by,
ˆ
(φ∇φ) · d~a = 0 (16)
∂V

If φ = 0 then the boundary constraint integral in eq. 16 is satisfied. This implies

ϕ2 − ϕ1 = 0
 7

ϕ2 = ϕ1
It follows that the solution to the magnetizing field equation is unique if the
values of the field are defined on the entire boundary of the field. This is known
as a Dirichlet boundary condition.
Furthermore it is also clear by inspection that if ∇φ = 0 in eq. 16 then the
boundary constraint integral again vanishes. It follows that

∇ϕ2 − ∇ϕ1 = 0

∇ϕ2 = ∇ϕ1

It follows that the solution is unique if the first derivative of the field is known
everywhere on the boundary. This is known as a Neumann boundary condition.
In a mixed boundary condition both Dirichlet and Neumann boundary condi-
tions can be specified. They will both cause the boundary constraint integral
to vanish and thus produce a unique solution.
We now have a sufficient battery of tests to verify whether a proposed image
solution is unique.

Uniqueness of the Image Solution

The proposed image solution is to take the magnetizing potential and apply a
reflection operator. We will see that the magnetizing potential is, in fact, a
pseudo-scalar thus it will require a polarity change after undergoing a reflec-
tion operation (a form of improper rotation where at least one coordinate axis
inversion occurs).
To begin we will define some fields.
Let ϕreal (~r) be the magnetizing potential field associated with the real magne-
tization distribution M~ real (~r).

Let ϕimage (~r) be the magnetizing potential field associated with the image mag-
netization distribution M~ image (~r)

The magnetizing field ΦH = ϕreal + ϕimage is the solution for the magnetizing
field within our problem domain. It is the solution to test for uniqueness.
The real and image magnetizing potential fields both will go to zero at infinity
so if we sum them this property will hold, this satisfies one boundary condition
for ΦH . The second and final boundary condition is that ΦH = 0 everywhere
on the symmetry plane. To satisfy this we can define the image magnetizing
potential as,
ϕimage = −ϕreal (R~r)
where R is a mirror transformation operator that mirrors a vector about the
symmetry plane defined by ~n. We have to also change the polarity of the field
 8

when we reflect it. We can see why this is necessary by checking the solution
on the boundary by first defining a vector constrained to the symmetry plane ~s
and noting that,
~
ϕreal (R~s) = ϕreal (s)
In other words coordinates on the symmetry plane are not transformed under
a reflection. We then have,

ΦH (~s) = ϕreal (~s) + [−ϕreal (~s)]

ΦH (~s) = 0 (17)

With all boundary conditions satisfied we have confirmed that the solution in
the problem domain is,

ΦH (~r) = ϕreal (~r) − ϕreal (R~r)

Taking the gradient of this solution,

∇ΦH (~r) = ∇ϕreal (~r) − ∇ϕreal (R~r)

recalling that the chain rule still applies to a gradient,

∇ϕreal (R~r) = R∇ϕreal (~u)

with ~u = R~r. Also substituting the definition for the magnetizing field gradient
in eq. 6 we get,
H~ solution = H
~ real (~r) − RH
~ real (R~r) (18)

Because of the uniqueness theorem this solution is identical to the field solu-
tion if we had not used the image method and instead left the highly permeable
material where it was. This also implies that the field “felt” by the real magneti-
zation distribution from a highly permeable material arises from a magnetization
distribution,
M~ image = −RM ~ real (R~r) (19)

Eq. 19 would prove most useful in calculating the force exerted on a permanent
magnet by a large ferrous plate of high permeability. The plate can be replaced
by the image of the magnetization distribution of the magnet.

Sample Problem

A bar magnet of radius R = 0.03m and length L = 0.15m has a magnetization

of 1 × 106 A/m. One of its flat ends is placed in contact with an infinitely
permeable plate of iron of side A  L. Find the force needed to separate the
magnet from the sheet.
 9

Bound Current Approach

With the image method in mind; one approach to solving this problem is to
determine the magnetic field resulting from the bound currents in the magnet.
Once those are determined we can create a virtual magnetization distribution
according to eq. 19. We then calculate the total force on these dipoles due to
the magnetic field created by the bound current distribution. The total force
on these virtual dipoles will be the same as the force on the metal plate by the
magnet.
We’ll begin by defining the magnetization of the real magnet as a function of
the position using cylindrical coordinates. In fact this entire problem will be
done in cylindrical coordinates because this will exploit the symmetry of the
problem.

~ = CM H(R − r)[H(−z) − H(L − z)]ẑ

M (20)

Applying the image transformation in eq. 19 with the plate surface normal
aligned to the z-axis gives the magnetization distribution image,

~ i = CM H(R − r)[H(−zi ) − H(−zi − L)]ẑ

M (21)

Where H is the Heaviside step distribution,

(
0 x<0
H(x) =
1 otherwise

whose derivative is the Dirac Delta distribution

(
∞ x=0
δ(x) = (22)
0 otherwise

which has the property,

ˆb
f (x)δ(x − c) dx = f (c) where a ≤ c ≤ b . (23)
a

Bound Current

We can determine the bound current by starting with the definition of the
magnetic field in terms of the magnetizing field and magnetization field,

~ = µ0 H
B ~ + µ0 M
~ (24)
 10

We take the curl of eq. 24,

~ = µ0 ∇ × H
∇×B ~ + µ0 ∇ × M
~ (25)

Because this is a magnetostatics problem there are no free currents so the mag-
netizing field has no curl,

~ = 0.
∇×H

Substituting into eq. 25 gives the curl of the magnetic field inside of the domain
of the magnet in terms of the magnetization distribution,
~ = µ0 ∇ × M
∇×B ~ (26)

With the following Maxwell equation,

~
∂E
~ = µ0 J~ + µ0 0
∇×B
∂t
in the case of magnetostatics, eq. 26 becomes,

µ0 J~ = µ0 ∇ × M
~

Note that the total current can be separated into two terms as follows,

J~ = J~b + J~f
There is no free current so the definition of the bound current density is,

J~b = ∇ × M
~ (27)

Using the definition for curl in cylindrical coordinates,

     
1 ∂vz ∂vθ ∂vr ∂vz 1 ∂ ∂vr
∇ × ~v = − r̂ + − θ̂ + (rvθ ) − ẑ
r ∂θ ∂z ∂z ∂r r ∂r ∂θ

eq. 27 becomes,
∂
J~b = − [CM H(R − r)[H(−z) − H(L − z)]] θ̂
∂r

J~b = CM δ(R − r)[H(−z) − H(L − z)]θˆ. (28)

 11

Magnetic Field

With the bound current density known we can use it to calculate the magnetic
field from the Biot-Savart Law in MKS units.
ˆ ~
~ x) = µ0 (JdV ) ×~r
B(~ (29)
4π |r|3

Where ~r is the displacement vector from a volume element dV to ~x. In this

specific case we will define ~s as the coordinate of dV which is being integrated
over. We will define ~si as the coordinate in the image region we wish to know
the magnetic field at. With this definition

~r = ~si − ~s

where
ri cos θi
* +
~si = ri sin θi ,
zi
* r cos θ +
~s = r sin θ
z
are cartesian vectors in terms of their polar coordinate parameters.
Since we are trying to find the force on a volume of dipoles we use the dipole
force law,
F~ = ∇(m ~ .
~ · B) (30)

We substitute an arbitrary, infinitesimal dipole element in the image region,

dm ~ i dVi
~i = M

into eq. 30 and get,

dF~ = ∇(M
~ i dVi ) · B
~

We substitute eq. 29 and only determine the z component of dF~ since we are
only interested in the force between the plate and the magnet which will be
aligned to the z-axis.
ˆ ~ ) ×~r
µ0 (JdV
dF~ = ∇(M
~ dVi ) ·
4π |r|3
ˆ ~
∂ µ0 [JdV ×~r]z
dFz = (Miz dVi )
∂zi 4π |r|3

~
[JdV ×~r]z = (Jx ry − Jy rx )dV
 12

ˆ
∂ µ0 (Jx ry − Jy rx )rdrdθdz
dFz = (Miz ri dri dθi dzi )
∂zi 4π |r|3

* − sin θ +
θ̂ = cos θ
0
ˆ
∂ µ0 J[− sin θ(ri sin θi − r sin θ) − cos θ(ri cos θi − r cos θ)]rdrdθdz
dFz = (Miz ri dri dθi dzi ) 3
∂zi 4π [(ri cos θi − r cos θ)2 + (ri sin θ − r sin θ)2 + (zi − z)2 ] 2

Applying trigonometric identities this greatly simplifies to,

ˆ
∂ µ0 J[r − ri cos(θ − θi )]rdrdθdz
dFz = (Miz ri dri dθi dzi ) 3
∂zi 4π [r + ri2 − 2rri cos(θ − θi ) + (zi − z)2 ] 2
2

Expanding J and Miz ,

∂ 2
dFz = C [H(−zi ) − H(−zi − L)]dri dθi dzi . . .
∂z M
ˆ i
µ0 δ(R − r)H(R − r)[H(−z) − H(L − z)][r − ri cos(θ − θi )]ri rdrdθdz
3
4π [r2 + r2 − 2rri cos(θ − θi ) + (zi − z)2 ] 2
i

factoring the delta function out and integrating over r while applying the defi-
nition of the delta distribution.
ˆ ˆ R
∂ 2 µ0
dFz = CM dri dθi dzi δ(R − r)u(r)drdθdz
∂zi 4π 0
ˆ
∂ 2 µ0
dFz = C dri dθi dzi u(R)dθdz
∂zi M 4π

Expanding u
ˆ
∂ 2 µ0 H(R − R)[H(−z) − H(L − z)][R − ri cos(θ − θi )]ri Rdθdz
dFz = C dri dθi dzi
∂zi M 4π [R2 + ri2 − 2Rri cos(θ − θi ) + (zi − z)2 ] 2
3

Placing the appropriate limits on the integration allows us to assume the heav-
iside terms are constant within the range of integration and thus reduce to 1.
Also note by definition H(R − R) = 1. I am pairing the differentials with their
associated definite integral to make it clear which variable is associated with
which range of integration.
ˆ 2π ˆ L
∂ 2 µ0 [R − ri cos(θ − θi )]ri R
dFz = CM dri dθi dzi dθ dz 3
∂zi 4π 0 0 [R2 + ri2 − 2Rri cos(θ − θi ) + (zi − z)2 ] 2
 13

Magnetic Force

So far we have only focused on the magnetic field integral. Of course we actually
want the force integral, that’s why we’ve been dragging the differentials along
so far. We have put this off because the partial derivative operator arising from
the gradient operator in the magnetic dipole force law requires some special
consideration. It is nothing complex; but, the partial derivative does not simply
“cancel” the associated differential out. First we will construct the full force
integral and isolate zi ,
ˆ ˆ R ˆ 2π ˆ L "ˆ −L #
2
CM µ0 2π ∂ [R − ri cos(θ − θi )]ri R
Fz = dθi dri dθ dz dzi
4π 0 0 0 0 0 ∂zi [R2 + r2 − 2Rri cos(θ − θi ) + (zi − z)2 ] 32
i

Evaluating the brackets involves nothing less than the fundamental theorem of
calculus,
ˆ b
d
f (x)dx = f (b) − f (a)
a dx

"ˆ ˆ ˆ ˆ #zi =−L

2π R 2π L
C 2 µ0 [R − ri cos(θ − θi )]ri R
Fz = M dθi dri dθ dz 3
4π 0 0 0 0 [R2 + ri2 − 2Rri cos(θ − θi ) + (zi − z)2 ] 2 zi =0

To further simplify this integral it is possible to set θi = 0 from the cos ar-
guments. As a result it simplifies to a three variable integral that must be
evaluated numerically.
"ˆ ˆ 2π ˆ L #zi =−L
2 R
CM µ0 [R − ri cos θ]ri R
Fz = dri dθ dz 3
2 0 0 0 [R2 + r2 − 2Rri cos θ + (zi − z)2 ] 2
i zi =0

Evaluating this integral with

CM = 1 × 106 A/m

µ0 = 4π × 10−7 N/A2
L = 0.15m
R = 0.03m

Gives an attraction force between the magnet and the plate of,

Fz = 1718 N

Depending on the method of integration this triple integral may take consider-
able time to evaluate.
 14

Pole Strength Approach

The method of using pole strength involves treating a magnetostatics problem

as though it consists of magnetic charges. This isn’t a physically accurate model
as magnetic fields are created by current loops and thus magnetic poles cannot
be isolated. However, it is possible to use this model without violating Maxwell’s
equations globally and obtain a good approximation for magnetostatics prob-
lems. Since I do not wish to imply the actual existence of isolated magnetic
charges I will refer to a region that has a non-zero divergence of the magnetiza-
tion distribution as a “pole” the magnitude of a pole will be referred to as “pole
strength.”
The method of images can easily be applied to the pole strength approach. Since
we are using an analog to electric charges it is reasonable to assume that applying
the usual image transformation will work. Throughout this problem we will refer
to the image magnet as though it is a real magnet. For all practical purposes
an image problem reduces to calculating the force between two magnets.

Magnetic Poles

We can define pole density distribution by observing that eq. 13 is very similar
to the Poisson equation for electrostatics. By analog we can express this as,

∇2 ϕH = ρg

where,
~
ρg = ∇ · M

Additionally since,
~ = −∇ · M
∇·H ~

we can express the divergence of the magnetizing field in terms of the pole
density distribution as,
∇·H ~ = −ρg

Force Law

First we’ll begin by introducing a force law by drawing an analogy between

magnetic and electric dipoles. We will use the Gilbert model of treating a
magnetic dipole as two poles of magnitude g separated by a distance d. We will
now use empirically verified torque laws to infer a force law that involves the
magnetic field and pole strength,

~τm = d~ × F~m
 15

~τm = m~ ×B~
p~ = q d~ =⇒ m
~ = g d~
~ ×B
m ~ = d~ × F~m
g d~ × B
~ = d~ × F~m

The cross product is a linear operator and thus the former statement is equiva-
lent to
d~ × g B
~ = d~ × F~m =⇒ F~m = g B
~.

Pole Distribution

Before we can apply our force law we need to know the distribution of poles
both real and imaged. The real pole distribution will be the divergence of eq.
20. Using the definition for the divergence in cylindrical coordinates,
1 ∂ 1 ∂vθ ∂vz
∇ · ~v = (rvr ) + +
r ∂r r ∂θ ∂z
The magnetization distribution,
~ = CM H(R − r)[H(−z) − H(L − z)]ẑ
M
becomes the pole density distribution,
ρreal = CM H(R − r)[−δ(−z) + δ(L − z)] (31)

The image pole density can be found with eq. 19,

~ (R~r))
ρimage (~r) = ∇ · (−RM
In this case R serves to invert the z-axis.
ρimage (~r) = ∇ · (CM H(R − r)[H(z) − H(L + z)]ẑ)
ρimage (~r) = CM H(R − r)[δ(z) − δ(L + z)] (32)

We should stop to examine how both of our pole density distributions create
“poles.” Note that the distributional derivative of the magnetization distribution
results in two delta functions. Those both completely localize the pole density
strength to two separate locations. Each of the poles could actually be regarded
as entirely separate disks at both ends of the magnet. However, it really isn’t
even necessary to worry about this due to the property of the delta functions
in eq. 23. The delta distribution gives us a convenient way of expressing a
discontinuous distribution in a way that we can simply evaluate volume integrals
over the distributions and get the same solutions as if we had explicitly treated
them as some kind of surface object.
With the distribution of poles known we can now calculate the forces between
the poles and determine the net force between the real magnet and virtual
magnet and hence the force between the magnet and the iron plate.
 16

Symmetry Plane Poles

The poles near the symmetry plane can be regarded as being a very small
distance  from the symmetry plane. This small gap represents the real physical
gap that would exist between two magnets in physical contact. Additionally
if the poles are regarded as coincident then there is no free space available for
a field to exist between them that can exert any force. Because the distance
between these two poles is so small we can ignore effects of fringing.
We calculate the magnetic field in the air gap as follows,

∇·H ~ = −∇ · M
~
ˆ ˆ
~ dV =
−∇ · M ~ · d~a
H
V ∂V
In the following boundary integral we use a very thin, axis aligned cylinder of
radius R, ˆ ˆ
−CM H(R − r)[δ(z) − δ(z − L)]dV = ~ · d~a
H
∂V
dV = rdrdθdz
ˆ R ˆ 2π ˆ 
dr dθ dz(−CM H(R − r)[δ(z) − δ(z − L)]r)
0 0 −
ˆ R ˆ 2π ˆ  ˛
−CM rdr dθ δ(z) − δ(L − z)dz = H da
0 0 −

1
−CM R2 2π = H2πR2
2
1
H = − CM
2
µ0
B = − CM
2
With the magnetic field and the force law we calculate the force between these
poles as follows,

dg = ρimage dV = ρimage rdrdθdz

dF = dgB
 µ 
0
dF = CM H(R − r)[δ(z) − δ(L + z)] − CM rdrdθdz
2
ˆ R ˆ 2π ˆ   µ 
0
F = dr dθ dzCM [δ(z) − δ(L + z)] − CM r
0 0 − 2
1
F = − µ0 R2 CM
2
π (33)
2
 17

Distant Poles

The rest of the poles are far enough apart that we cannot disregard the distance
between them. We will have to compute the force between the poles a distance
2L apart and the force from the poles a distance L apart. Because all the poles
are the same shape we need only construct one integral and evaluate it for the
two distances we are interested in. We will be somewhat less rigorous than the
bound current approach in our evaluation of the integral as the finer details of
using Heaviside and delta distributions are covered sufficiently above.
We are going to be integrating the force between pole elements so we need an
equation that gives us the magnetic field due to a point pole. We will use

~ = µ0 g r̂
B (34)
4πr2

Substituting this into the force on a pole we have the force between two point
poles,
g1 g2
F~ = µ0 r̂
4πr2

dg1 dg2 %
~
dF~ = µ0
4π|%|3
The displacement vector between two pole elements is,

~ = ~r2 − ~r1
%

The pole elements will be restricted to the pole surfaces separated by a distance
L,
* r cos θ +
2 2
~r2 = r2 sin θ2
L
r1 cos θ1
* +
~r1 = r1 sin θ1
0

We only need the force along the z-axis so we express the differential force as,
2
CM Lr1 dr1 dθ1 r2 dr2 dθ2
dFz = µ0 3/2
4π [r12 + r22 − 2r1 r2 cos(θ1 − θ2 ) + L2 ]

Letting θ2 = 0, changing L to a parameter h and integrating,

ˆ R ˆ R ˆ 2π
1 2 r1 r2
Fz (h) = µ0 CM h dr1 dr2 dθ1 3/2
2 0 0 0 [r12 + r22 − 2r1 r2 cos(θ1 ) + h2 ]
 18

This integral must be evaluated numerically. The total force will be,
1
F = − µ0 R2 CM
2
π + F (L) − F (2L)
2

Evaluating this with the parameters defined in the problem gives,

−1698.6N

Where the negative sign merely indicates this is the force of the plate on the
magnet and not vice versa as is done in the bound current approach earlier. The
choice of symmetry is really irrelevant as long as the polarity of the charges and
dipoles are kept straight in both methods. Both methods are in close agreement.