Fernow RC Principles of Magnetostatics

PRINCIPLES OF MAGNETOSTATICS
The subject of magnetostatics—the mathematical theory that describes the forces

and fields resulting from the steady flow of electrical currents—has a long history.
By capturing the basic concepts, and building toward the computation of mag-
netic fields, this book is a self-contained discussion of the major subjects in
magnetostatics.
Overviews of Maxwell’s equations, the Poisson equation, and boundary value
problems pave the way for dealing with fields from transverse, axial, and periodic
magnetic arrangements and assemblies of permanent magnets. Examples from
accelerator and beam physics give up-to-date context to the theory. Furthermore,
both complex contour integration and numerical techniques (including finite dif-
ference, finite element, and integral equation methods) for calculating magnetic
fields are discussed in detail with plentiful examples.
Both theoretical and practical information on carefully selected topics make this
a one-stop reference for magnet designers, as well as for physics and electrical
engineering students. This title, first published in 2017, has been reissued as
an Open Access publication on Cambridge Core.
richard c. fernow received his PhD at Syracuse University for work on

particle physics and worked at Brookhaven National Laboratory. He contributed
to the optimization of the coil design for collider magnets and made calculations of
magnetic fields in solenoid channels. He is a member of the American Physical
Society.
PRINCIPLES OF MAGNETOSTATICS
RICHARD C. FERNOW
Formerly Brookhaven National Laboratory
Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467
Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781009291149
DOI: 10.1017/9781009291156
© Richard C. Fernow 2022
This work is in copyright. It is subject to statutory exceptions and to the provisions
of relevant licensing agreements; with the exception of the Creative Commons version the
link for which is provided below, no reproduction of any part of this work may take
place without the written permission of Cambridge University Press.
An online version of this work is published at doi.org/10.1017/9781009291156 under a
Creative Commons Open Access license CC-BY-NC-ND 4.0 which permits re-use,
distribution and reproduction in any medium for non-commercial purposes providing
appropriate credit to the original work is given. You may not distribute derivative works
without permission. To view a copy of this license, visit
https://creativecommons.org/licenses/by-nc-nd/4.0
All versions of this work may contain content reproduced under license from third parties.
Permission to reproduce this third-party content must be obtained from these third-parties directly.
When citing this work, please include a reference to the DOI 10.1017/9781009291156
First published 2017
Reissued as OA 2022
A catalogue record for this publication is available from the British Library.
ISBN 978-1-009-29114-9 Hardback
ISBN 978-1-009-29116-3 Paperback
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents
Preface page ix
1 Basic concepts 1
1.1 Current 2
1.2 Magnetic forces 3
1.3 The Biot-Savart law 6
1.4 Divergence of the magnetic field 11
1.5 Circulation of the magnetic field 12
1.6 The Ampère law 14
1.7 Boundary conditions at a current sheet 17
1.8 Inductance 19
1.9 Energy stored in the magnetic field 20
2 Magnetic materials 23
2.1 Magnetization 23
2.2 Magnetic field intensity 26
2.3 Permeability and susceptibility 27
2.4 Types of magnetism 27
2.5 Magnetic circuits 31
2.6 Boundary conditions between regions with different μ 33
2.7 Method of images 34
3 Potential theory 39
3.1 Vector potential 39
3.2 Vector potential in two dimensions 41
3.3 Boundary conditions on A 44
3.4 Vector potential for a localized current distribution 45
3.5 Force on a localized current distribution 49
3.6 Magnetic scalar potential 50
v
vi Contents
3.7 Scalar potential for a magnetic body 52

3.8 General solutions to the Laplace equation 53
3.9 Boundary value problems 59
3.10 Green’s theorem 67
4 Conductor-dominant transverse fields 71
4.1 General solution to the Laplace equation in
two dimensions 71
4.2 Harmonic expansion for a line current 76
4.3 Field for a current sheet 80
4.4 Ideal multipole current sheet 84
4.5 Multipole dependence on the current distribution 88
4.6 Approximate multipole configurations 90
4.7 Field for a block conductor 93
4.8 Ideal multipole current block 100
4.9 Field from a magnetized body 102
4.10 Superconductors 104
4.11 End fields 105
5 Complex analysis of transverse fields 108
5.1 Complex representation of potentials and fields 108
5.2 Maxwell’s equations in complex conjugate
coordinates 111
5.3 Field from a line current 113
5.4 Field from a current sheet 115
5.5 cos ϕ current sheets 120
5.6 Green’s theorems in the complex plane 123
5.7 Field from a block conductor 124
5.8 Block conductor examples 126
5.9 Field from image currents 132
5.10 Multipole expansion 135
5.11 Field due to a magnetized body 138
5.12 Force 140
5.13 Conformal mapping 141
5.14 Integrated potentials 147
6 Iron-dominant transverse fields 150
6.1 Ideal multipole magnets 150
6.3 Dipole configurations 154
6.4 Quadrupole configurations 159
Contents vii
7 Axial field configurations 165

7.1 Circular current loop 165
7.2 Radial expansion of the on-axis magnetic field 170
7.3 Zonal harmonic expansions 172
7.4 Multiple coil configurations 178
7.5 Sheet model for the solenoid 182
7.6 Block model for the solenoid 189
7.7 Bent solenoid 190
7.8 Toroid 193
8 Periodic magnetic channels 197
8.1 Field from a helical conductor 197
8.2 Planar transverse field 204
8.3 Helical transverse field 206
8.4 Axial fields 209
9 Permanent magnets 211
9.1 Bar magnets 211
9.2 Magnetic circuit energized by a permanent magnet 214
9.3 Material properties 216
9.4 Model for rare earth materials 217
9.5 Rare earth model in two dimensions 219
9.6 Multipole expansion for continuously
distributed material 221
9.7 Segmented multipole magnet assemblies 224
10 Time-varying fields 229
10.1 Faraday’s law 229
10.2 Energy in the magnetic field 232
10.3 Energy loss in hysteresis cycles 234
10.4 Eddy currents 235
10.5 Skin effect 239
10.6 Displacement current 240
10.7 Rotating coil measurements 241
11 Numerical methods 245
11.1 Finite difference method 245
11.2 Example solution using finite differences 251
11.3 Finite element method 254
11.4 Integral equation method 258
11.5 The POISSON code 265
11.6 Inverse problems and optimization 267
viii Contents
Appendices 274
A Symbols and SI units 274
B Vector analysis 275
C Bessel functions 279
D Legendre functions 283
E Complex variable analysis 287
F Complete elliptic integrals 293
Index 297
Preface
My career at Brookhaven National Laboratory began in 1978, when I was hired to

work on a particle physics experiment at the alternating gradient synchrotron
(AGS). At that time, one of the lab’s major projects was a high-energy proton
collider known as ISABELLE. Unfortunately, development of the superconducting
magnets proposed for the project ran into serious technical difficulties. As a result,
a member of the physics department, Dr. Robert Palmer, proposed a radically
different design for the collider magnets. He began recruiting a small group from
the physics and accelerator departments to work on the alternate magnet design.
I was one of the staff members who joined his group.
Because I came from a background in particle physics, the work on the design of
high-field magnets was a revelation to me. One of my main responsibilities was to
work on the optimization of the dipole conductor cross-section and end designs
needed to achieve the demanding field quality requirements for the accelerator.
I quickly discovered that the methods needed for practical magnetic field design
went far beyond my academic training in electricity and magnetism. Much of the
work required frequent feedback with the engineers working on the project.
Although I moved on from the magnet division after about four years, my interest
in calculating magnetic fields remained with me throughout the rest of my career.
A significant part of the contents of this book are based on my notes from that
period.
The subject of magnetostatics has a long history. There is a vast literature, so
a book of this size has to make difficult choices about which topics to include.
My primary objective was to produce a self-contained discussion of the major
subjects in magnetostatics with an emphasis on the computation of magnetic fields.
For that reason, I have included brief treatments of most standard background
material, such as the magnetostatic Maxwell’s equations and potential theory.
However, the choice of example topics relies heavily on my background and
interests. Many of the examples come from the fields of accelerator and beam
ix
x Preface
physics. I also felt it was important to expose a wider audience to a series of very
insightful papers by the late Klaus Halbach of Lawrence Berkeley Laboratory. For
the discussion of numerical methods, I decided to concentrate on a small number of
subjects while including sufficient details to enable readers to begin writing their
own computer codes, if they so desired.
The first three chapters are mostly a survey of basic material. Chapter 1 treats the
theory of magnetic fields from conductors in free space, and Chapter 2 discusses
fields from magnetic materials. Chapter 3 introduces the vector and scalar poten-
tials. It includes general solutions to the Laplace equation and the solution of
boundary value problems.
Chapters 4–6 discuss transverse fields in two dimensions. Chapter 4 looks at
fields from line currents, current sheets, and current blocks. Field quality is
introduced in terms of multipole expansions, and the effects of approximating
ideal current distributions are described. Chapter 5 looks at transverse fields using
complex variable methods. The powerful techniques for computing the fields of
block conductors using contour integration are discussed in detail. Contour inte-
grals are also derived for the fields from image currents and magnetized bodies.
Chapter 6 looks at transverse fields that are determined by the shape of the iron.
The discussion here mainly concerns the iron surfaces used in dipole and quadru-
pole magnets.
Chapters 7–9 discuss some other field configurations. Chapter 7 looks at axial
field arrangements. This includes the fields from current loops, solenoids, and
systems of coils. The solution of the solenoid field using the sheet model is treated
in detail. Chapter 8 considers periodic magnetic channels. First the field from
a helical conductor winding is discussed. Then inverse problems are introduced,
and some of the field configurations used for magnetic wigglers and particle beam-
focusing channels are examined. Chapter 9 begins with a standard treatment of the
properties of permanent magnets. This is followed by a discussion of Halbach’s
model for rare-earth cobalt magnets and his analysis of assemblies of permanent
magnets.
In Chapter 10, we relax the strict conditions of magnetostatics and allow for the
case of slowly varying currents. This leads to brief discussions of some standard
subjects such as Faraday’s law, but also some more engineering-related topics such
as eddy currents and the skin effect. This also seemed an appropriate place to
include a brief discussion of magnetic field measurements using rotating coils.
Chapter 11 discusses numerical methods. No attempt is made here to survey the
thousands of papers devoted to numerical solutions of magnetic field problems.
Instead, three methods for solving the Poisson equation are presented with
a significant amount of detail. This chapter also includes a discussion of the
POISSON code, which is freely available and extremely useful for investigating
Preface xi
2D problems. The chapter ends by returning to the inverse problem and presenting
several examples of using optimization methods.
The appendices collect some important details about mathematical techniques
and special functions used in the book.
The level of the treatment of background magnetostatic topics in the book is
typical of those encountered by undergraduate physics majors. Some of the mate-
rial in Chapters 4–11 will likely be unfamiliar to many readers. However, an
attempt has been made to include sufficient details and references, so interested
physics and engineering majors should be able to follow the discussions.
At a number of places in the text, I have indicated the source of a mathematical
relation in footnotes using the following notation.
AS M. Abramowitz & I. Stegun (eds.), Handbook of Mathematical

Functions, Dover Publications, 1972.
CRC S. Selby (ed.), Standard Mathematical Tables, 14th ed., The Chemical
Rubber Company, 1965.
GR I. Gradshteyn & I. Ryzhik, Table of Integrals, Series and Products,
Academic Press, 1980.
I would like to thank several of my former colleagues, especially Gerry Morgan,

Steve Kahn, Bob Palmer, Juan Gallardo, and Scott Berg, for many interesting
discussions concerning magnetic fields and the methods used for calculating them.
I would like to thank Peter Wanderer and Animesh Jain of the Superconducting
Magnet Division at Brookhaven National Laboratory for providing the image of
field lines in a RHIC dipole, which has been used on the cover of this book. I would
also like to thank Simon Capelin and the staff at Cambridge University Press for
their collaboration on this project. Finally, I would like to thank my wife Ruth for
her constant support and encouragement.
1
Basic concepts
Magnetic phenomena have been known since antiquity when a natural ore later
called lodestone was discovered to attract bits of iron. The scientific study of
magnetism dates from around 1600, when William Gilbert summarized experi-
ments on the subject in his treatise De Magnete.[1] However, interest in the
subject greatly increased after 1820, when Hans Christian Øersted reported that
electrical currents could deflect magnetic needles, thereby establishing a con-
nection between the subjects of electricity and magnetism.[2] Almost immedi-
ately, André-Marie Ampère, Jean-Baptiste Biot, and Félix Savart performed a
series of seminal experiments that determined the forces acting between current
loops. Experimental work and theoretical developments continued throughout
the first half of the nineteenth century. A long program of experimental inves-
tigations by Michael Faraday lead him to the conception that the force between
current loops occurred through the action of an intermediary field that existed in
the space around the loops. Faraday’s field concept was developed mathemati-
cally by William Thomson (later Lord Kelvin). This work culminated in a
synthesis of knowledge about electrical and magnetic phenomena by James
Clerk Maxwell in his famous treatise of 1873. Many clarifications of
Maxwell’s ideas and studies of their implications were carried out over the
next twenty years by a small group of followers. Of particular note was the
work of Oliver Heaviside who introduced the use of vector analysis and
reworked the set of equations in Maxwell’s treatise to the four equations we
use today.[3] The resulting Maxwell equations are now accepted as the theore-
tical description underlying electromagnetic phenomena.
Magnetostatics is the study of the fields, forces, and energy associated with
steady currents and magnetic materials. In this chapter, we will review some
basic concepts underlying magnetic effects due to conductor currents in free
space.
1
2 Basic concepts
1.1 Current
Experiments have shown that there exist two kinds of electrical charge q, which are
denoted as positive and negative. A current I exists when there is a net temporal
flow of charge across some arbitrary plane in space.
dq
I¼ : (1.1)
dt
If the current is flowing through a conductor with length L and cross sectional area
A, we can write the current as
ρLA
I¼ ¼ ρv A;
L=v
where ρ is the charge density and v is the velocity of the charges. The current
density J along some direction n is a vector given by
! I
J ¼ n^ ¼ ρ!
v; (1.2)
A
where ^
n is the unit vector perpendicular to A.
If we consider a volume of space V enclosed by a surface S, the conservation of
charge requires that any change in the charge density inside V must be compensated
by a flow of current through the surface or
ð ð
∂ρ !
dV ¼ J ·^n dS:
∂t
Using the Gauss divergence theorem,1 the right-hand side can be written as
ð ð
! !
n dS ¼ r· J dV:
J ·^
Then, since V is arbitrary, we can remove the integrands from the volume integrals
on both sides of the equation and obtain the continuity equation
∂ρ !
þ r· J ¼ 0: (1.3)
∂t
In magnetostatics, we have by definition ∂ρ=∂t ¼ 0, which leads to the relation
!
r· J ¼ 0: (1.4)
1
Readers unfamiliar with vector analysis should review Appendix B.
Often we are interested in the current flow in a “central” region far from the ends
of a magnet. If the current and the geometry are uniform along z in this region, we
can simplify the analysis by examining problems in two dimensions. If we consider
a conductor whose thickness is small compared with the distance to the observation
point, we can approximate the conductor as a current sheet.
In addition, we frequently consider line currents or “filaments,” where we
ignore the transverse dimensions of the conductor altogether and use the equiva-
lent current
ð
!
I ¼ J ·^n dS:
1.2 Magnetic forces

Experiments have shown that test currents and charges in the vicinity of a current-
carrying conductor experience a force. We assume that this force takes place
through the actions of an intermediary magnetic field. The mathematical descrip-
tion of a field is a continuous function that is defined for all points in space and for
all times. However, the magnetic field also has physical properties associated
with it, such as stored energy. The force experiments can be explained by
assuming that a current produces a vector field B, and then this field produces a
force on other currents and charges. The vector field B is called the magnetic flux
density2 or magnetic field for short. The magnetic flux through some surface S is
defined as
ð
! !
ΦB ¼ B · dS : (1.5)
The direction of the magnetic field is often represented using Faraday’s concept
of lines of induction.3[4] The lines of induction are defined to be tangent to the
magnetic field at every point in space. It follows that corresponding components of
the lines of induction and the magnetic field are always proportional to each other.
If ds is a small displacement along the line of induction, we have
dx dy dz ds
¼ ¼ ¼ :
Bx By Bz B
In two Cartesian dimensions, the lines can be plotted, for example, by integrating
the equations
2
The vector B is also known as the magnetic induction.
3
Historically, these curves have been referred to as lines of force.
4 Basic concepts
Bx ðx; yÞ
dx ¼ ds
Bðx; yÞ
By ðx; yÞ
dy ¼ ds:
Bðx; yÞ
The magnitude of the magnetic field can be represented by the density of lines in a
given region. The lines of induction do not have to form closed loops.[5, 6] In
particular, the lines become undefined at locations where B = 0.
Now consider two circuits carrying currents Ia and Ib . The force exerted by
circuit a on circuit b is found experimentally to be
! !
! μ0 ! dla R
F ab ¼ Ia Ib dlb ; (1.6)
4π ∯ R3
where the constant μ0 ¼ 4π 107 is known as the permeability of free space,4 dl is a

displacement along the circuit in the direction of the current, and R is the distance
vector from dla to dlb. Note that the force is proportional to the product of the
currents times a geometric factor that depends on the shape and orientations of the
two circuits. It is possible to rewrite this equation in a form that manifestly obeys
Newton’s Third Law of motion. Using the vector triple product identity from
Equation B.1 in Appendix B, we have
! ! ! ! ! ! ! ! !
dlb ð dla R Þ ¼ dla ð dlb · R Þ R ð dla · dlb Þ:
The double integral of the first term on the right-hand side is then
! ! ! þ þ ! ! þ þ
dl a ðdl b · R Þ ! ðdl b · R Þ ! dR
¼ dl a ¼ dl a :
∯ R3 R3 R2
The last integral vanishes because the scalar integrand is taken over a closed path.
Thus we can express the force as
!! !
! μ0 R ðdl a ·dl b Þ
F ab ¼ Ia Ib : (1.7)
4π ∯ R3
In this form, we see that Newton’s law Fab ¼ Fba is obeyed since R changes
direction for the two cases.
Returning to Equation 1.6, we rewrite the force on circuit b in a form that
explicitly depends on the current in circuit b and on an integration of the elemental
4
We will use SI units exclusively in this book. For more details, see Appendix A.
interactions taking place around that circuit. We collect the other factors in
Equation 1.6 into a new vector Ba, which we define as the magnetic field due to
circuit a. Then the force on the circuit can be written as
þ
! ! !
F ab ¼ Ib dl b B a : (1.8)
The force acts at right angles to the direction of Ba. Dropping the subscripts, we see
that the force on a charge q moving with velocity v can be written as
ð
! dq ! ! !
F ¼ dl B ¼ q! v B : (1.9)
dt
Note that the force only acts on moving charges.

Now consider a rectangular current loop with length L and width w in a constant
magnetic field B, as shown in Figure 1.1. The forces on each pair of opposite sides
cancel, so there is no net force on the loop. However, there are moment arms
between sides 1 and 3 and the axis of the loop. This creates a torque given by
!
τ ¼!
! r F:
Figure 1.1 Rectangular current loop in an external field.

6 Basic concepts
For the example here,

w
τ¼2 NILB sin θ;
2
where N is the number of turns in the loop. We define the magnetic moment m of a
planar loop to lie along the normal n to the loop, so that
!
m ¼ NIA ^n ; (1.10)
where A is the area of the loop. Then the torque acting on the loop can be expressed as
!
τ ¼!
! m B: (1.11)
1.3 The Biot-Savart law

Comparing Equations 1.6 and 1.8, we see that the force experiments require that the
magnetic field can be expressed in the form
þ! !
! μ0 dl R
B ¼ I ; (1.12)
4π R3
where we have dropped the subscripts referring to circuit a. The vector R points
from the current element source to the observation (or field) point where the
magnetic field is determined. This relation, known as the Biot-Savart Law, is an
important tool for finding analytic and numerical solutions for the magnetic field
produced by known current distributions. For a surface distribution of current, the
total current in the Biot-Savart law can be generalized to give
ð! !
! μ0 K R
B ¼ dS; (1.13)
4π R3
where K is the surface current density. Likewise, for a volume distribution of

current, we have
ð! !
! μ0 J R
B ¼ dV: (1.14)
4π R3
It is important to keep in mind that the Biot-Savart law and many of the other
mathematical laws that we will subsequently develop ultimately depend on the
validity of the experimental results on magnetic forces.
We consider next several elementary applications of the Biot-Savart law that we
will need to refer to later in this book.
Figure 1.2 Current in a long straight wire.
Example 1.1: field from an infinitely long straight wire

Consider an infinitely long straight wire lying along the z axis, as shown in Figure 1.2.
Because of the symmetry, we use cylindrical coordinates. Since the wire is infinitely
long, we can chose an observation point P in the plane with z = 0 without loss of
generality. Since
!
dl ¼ dz ^z
!
R ¼pρ ^ρffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ z ^z
R ¼ ρ þ z2 ;
2
the field at point P due to the current in the wire is
! μ0 ^
B ¼ ρ ϕ 2 I;
4π
where5
ð∞
dz 1
I¼ ¼ :
0 fρ2 þ z2 g3=2 ρ2
Thus the magnetic field due to the current in the wire is

! μI ^
B ¼ 0 ϕ: (1.15)
2π ρ
The field is directed azimuthally around the wire and falls off with distance like 1/ρ.
5
GR 2.271.5.
8 Basic concepts
Figure 1.3 Force between two parallel wires.
Example 1.2: force between two parallel wires

Consider two infinitely long parallel wires, as shown in Figure 1.3. From Equation
1.8, the incremental force between the two wires is
! ! !
dF b ¼ Ib dl b B a
and from the previous example, the field at P due to the current in wire a is
! μ Ia
B a ¼ 0 ^z :
2π ρ
If the current direction in wire b can be either parallel or antiparallel to the current in
wire a, we find that the force per unit length of the wire is
!
dF b μ
¼ 0 Ia Ib ^x : (1.16)
dy 2π ρ
The force between the wires is attractive when the currents are in the same direction
and repulsive when they are antiparallel.
Example 1.3: field above an infinite current sheet

Consider an infinite current sheet with current flowing uniformly in the y direction.
We calculate the magnetic field at point P, shown in Figure 1.4. We have
! ¼ K ^y
K y
!
R ¼ x ^x þ y ^y þ zo ^z :
Figure 1.4 Field above an infinite current sheet.
The field is given by

ð∞ ð∞
! ¼ μ0 K y zo ^x x ^z
dx dy
B
4π ∞ ∞ fx2 þ y2 þ z2o g3=2
μ0 K y
¼ ðzo ^x I1 ^z I2 Þ;
4π
where
ð∞ ð∞
1 2π
I1 ¼ dx dy ¼
∞ ∞ fx2 þ y2 þ z2o g3=2 zo
and
ð∞ ð∞
x
I2 ¼ dx dy ¼ 0:
∞ ∞ fx2 þ y2 þ z2o g3=2
The integral I2 vanishes because the integrand is an odd function and the integration
extends over an even interval. The magnetic field above the sheet is
! μ0
B ¼ Ky ^x :
2
The direction of the field is parallel to the sheet and perpendicular to the current
density. The magnitude of the field is constant and independent of the distance from
the sheet. In the general case, the field above the sheet can be written as
! μ0 !
B ¼ K ^n ; (1.17)
2
10 Basic concepts
Figure 1.5 Field along the axis of a current loop.
where n is the normal to the sheet pointing to the side where B is computed. Note that
the direction of B follows the right-hand rule with respect to the current filaments in
the sheet.
Example 1.4: on-axis field due to a circular current loop

We look for the field at a point P that is along the axis of the loop and a distance zo
above the plane of the current loop, as shown in Figure 1.5. In cylindrical coordinates,
we have
! ^
dl ¼ a dϕ ϕ
!
R ¼ a ^r þ zo ^z :
The contributions of the current elements to the field at P lie in a cone surrounding P.
By symmetry, the net field must be in the z direction and
! !
ðdl R Þz ¼ a2 dϕ:
Thus we have ð 2π
μ0 I a2
Bz ¼ dϕ
4π 0 fa2 þ z2o g3=2
(1.18)
μ0 I a 2
¼ :
2fa2 þ z2o g3=2
Note that Bz is proportional to the area of the current loop and falls off at large
distances like z3
o : The field is largest at the center of the loop where the value is
μ0 I
Bz0 ¼ :
2a
1.4 Divergence of the magnetic field 11
1.4 Divergence of the magnetic field

Starting from the Biot-Savart law,
ð! !
! μ0 J R
B ¼ dV
4π R3
and taking the divergence of both sides, we have

ð ! !
! μ0 r·ðJ 0 R Þ 0
r· B ¼ dV ;
4π R3
where we use primes to indicate the use of source coordinates. The operator r is
defined in terms of field coordinates, while the distance R is a function of both
source and field coordinates. Using the vector identity B.4, we can write
! ! ! ! ! !
r·ðJ 0 R Þ ¼ R ·ðr J 0 Þ J 0 ·ðr R Þ:
!
The first term on the right-hand side vanishes because r J 0 ¼ 0. When the
second term is written in terms of a determinant, we obtain

^x ^y ^z
!
r R ¼ ∂x ∂y ∂z ¼ 0:

x x0 y y0 z z0
Thus we find that the divergence of the magnetic field vanishes.

!
r· B ¼ 0: (1.19)
This vector relation is one of the fundamental properties of magnetic fields. Its
validity depends on the fact that isolated magnetic charges (monopoles) do not
appear to exist.
If we integrate Equation 1.19 over some volume of space V that is enclosed by a
surface S, we get
ð
!
r· B dV ¼ 0:
Then using the divergence theorem, we find Gauss’s Law for magnetism
ð
!
n dS ¼ 0:
B ·^ (1.20)
12 Basic concepts
1.5 Circulation of the magnetic field

Returning again to the Biot-Savart law, consider the integral
! !
Ð J0 R
I¼ dV 0
R3
Ð 1 !0
¼ r J dV 0 :
R
Using the vector identity B.6, we can write this as
ð !0 ð
J 0 1 !
I ¼ r dV r J 0 dV 0 :
R R
!
The quantity r J 0 ¼ 0 in the second integral. We can bring the r operator in the
first term outside the integral sign because it operates on the observation point
coordinates, while the integral is over the source point coordinates. Thus
ð !0
J
I¼r dV 0
R
and an alternate expression for the magnetic field is
ð !0
! μ0 J
B ¼ r dV 0 : (1.21)
4π R
Taking the curl of both sides of this equation, we find

ð !0
! μ0 J
r B ¼ rr dV 0 :
4π R
Using the vector relation B.7, we can write this in the form
" ð !0 ! ð !0 ! #
! μ0 J J
r B ¼ r r· dV 0 r2 dV 0 :
4π R R
Then since r does not operate on J 0 , we have

ð ð
! μ0 !0 1 0
!0 2 1 0
r B ¼ r J ·r dV J r dV : (1.22)
4π R R
Consider for the moment the relation

1 0 1
r ¼ r :
R R
1.5 Circulation of the magnetic field 13
In the second integral in Equation 1.22 when R ≠ 0, we have

2 1 1 1
r ¼ 2 ∂r r ∂r
2
R r R
!
in spherical coordinates. Then since R ¼ j! r r0 j, we find that r2 R1 ¼ 0:

Although r2 R1 is undetermined when R = 0, the integral of this expression is

still defined. Performing the integral on a small sphere surrounding R = 0, we find
ð ð ð
2 1 1 1 !
r dV ¼ r·r dV ¼ r · dS
R R R
using the divergence theorem. Evaluating the last integral on the surface of the
small sphere, we find that
ð
1 ! 1
r · dS ¼ 2 4πR2 ¼ 4π:
R R
We can summarize these results by writing the expression in terms of the Dirac
delta function δ.

2 1
r ¼ 4πδðRÞ: (1.23)
R
Now we can do the first integral in Equation 1.22 using Equation B.3 to give
ð ð !0 ð
!0 0 1 0 0 J 0 1 !0 0
J ·r dV ¼ r · dV r·J dV :
R R R
The first term on the right-hand side can be converted to a surface integral using the
divergence theorem. It vanishes if the surface enclosing the volume in the
!integrals
is sufficiently large. The second integral also vanishes because r0 ·J 0 ¼ 0 for
magnetostatics. Thus we are only left with the second integral in Equation 1.22,
which because of the delta function from Equation 1.23, gives
! !
r B ¼ μ0 J : (1.24)
Thus we have shown that a steady current creates a magnetic field that circulates
around the current. This is a second fundamental vector relation for magnetic
fields.6
6
We will find in Chapter 10 that this relation requires an additional term if the current varies with time.
14 Basic concepts
1.6 The Ampère law

If we integrate both sides of Equation 1.24 over an arbitrary surface S, we find
ð ð
! ! ! !
ðr B Þ· dS ¼ μ0 J · dS :
On the right side, the current density integrated over the surface gives the total
current I. On the left side, we can use Stokes’s theorem from Appendix B to give
ð þ
! ! !!
ðr B Þ· dS ¼ B ·dl ;
where the contour on the right-hand side extends along the perimeter of the surface
S. Thus we have the result
þ
!!
B ·dl ¼ μ0 I: (1.25)
This equation is known as the Ampère law.7 It can be most usefully applied in
highly symmetric cases where, for example, the magnitude of the field is constant
along the integration path.
Again let us consider several elementary examples of using the Ampère law to
derive results that we use later in the book.
Example 1.5: a long cylindrical conductor

Consider a long cylindrical conductor with constant current density J inside the radius
a, as shown in Figure 1.6. Since by symmetry the magnitude of the field must be
independent of ϕ, we choose a circular path of integration. When the path is outside
the conductor, all of the current is enclosed by the path and the Ampère law gives
Bϕ 2πρ ¼ μ0 I. Thus the field outside the conductor is
μ0 I
Bϕ ¼ ; (1.26)
2πρ
which falls off like 1/ρ. Since this result is independent of the radius of the conductor,
it also applies to the field from a current filament, which we previously derived using
the Biot-Savart equation.
When the path of integration is inside the conductor, only part of current is
enclosed by the path and the Ampère law gives
πρ2
Bϕ 2πρ ¼ μ0 I:
πa2
7
According to O. Darrigol,[2] this equation was first given by Maxwell. In that case, we agree with him that it’s
really not appropriate to call it Ampère’s law.
1.6 The Ampère law 15
Figure 1.6 Cylindrical conductor of radius a. The two integration paths are shown
with dotted lines.
Thus the field inside the conductor is
μ0 I ρ μ0 J
Bϕ ¼ ¼ ρ; (1.27)
2πa2 2
which increases linearly with ρ.
Example 1.6: ideal solenoid

We define an ideal solenoid as an infinitely long system of parallel circular current
loops with radius a, as shown in Figure 1.7. This is an approximation to a real
solenoid when the observation points are far from the ends of the solenoid and the
conductor is tightly wound, such that we can ignore any gaps or the helical nature of
the windings. First consider a cylinder containing the points achj. From Gauss’s law,
Equation 1.20, we know that the flux passing through the surface must be 0. From
symmetry, the flux through the top and through the bottom faces of the cylinder have
to cancel. The contribution through the side of the cylinder then gives
2π ρ L Bρ ¼ 0:
Since the same argument applies for a cylinder of any radius, we must have Bρ ¼ 0
everywhere for the ideal solenoid.
Now consider the Ampère law applied to the path bdgi. The contributions to the
integral vanish along bd and gi since Bρ ¼ 0. Then we have
Bz ð0ÞL Bz ðr1 ÞL ¼ μ0 n I L;
where n is the number of conductor turns per unit length, or that
Bz ðr1 Þ ¼ Bz ð0Þ μ0 n I:
16 Basic concepts
Figure 1.7 Cross-section of an ideal solenoid. The dashed line is the axis of the
solenoid. The dots and crosses refer to the direction of the current.
If we apply the same argument over the path befi, we find that
Bz ðr2 Þ ¼ Bz ð0Þ μ0 n I:
Both of these equations for Bz outside the solenoid have the same right-hand side.
Thus the value of Bz outside the solenoid is constant, independent of radius. However,
if we consider the total flux outside the solenoid, we find
ð 2π ð ∞
ΦB ¼ z r dr d ϕ:
Bout
0 a
The total flux would be infinite if Bz outside the solenoid is any constant other than 0.
This is clearly non-physical, so we must have Bz ¼ 0 everywhere outside the
solenoid.
Since the field vanishes outside the solenoid, applying the Ampère law to the path
bdgi gives
Bz ð0Þ ¼ μ0 nI:
Similarly, on the path cdgh we find
Bz ðrÞ ¼ μ0 nI; (1.28)
where we write r for the length bc. Thus the field of the ideal solenoid is constant and
along the axis of the solenoid on the inside and it vanishes outside.
1.7 Boundary conditions at a current sheet 17
Figure 1.8 Gaussian pillbox across a current sheet.
1.7 Boundary conditions at a current sheet

We now consider how the magnetic field is influenced by the presence of a current
sheet. Assume we have a current sheet, as shown in Figure 1.8. We construct a
cylindrical pillbox across the sheet with an infinitesimal height along the normal to
the surface. Then applying Equation 1.20, we find that
B1n S B2n S ¼ 0:
Since the surface S is arbitrary, it follows that B1n ¼ B2n or that
! !
ð B2 B1 Þ·^n ¼ 0: (1.29)
Thus the normal component of B must be continuous across a current sheet.

Next construct a closed path across the current sheet, as shown in Figure 1.9.
Assume the path length perpendicular to the surface is infinitesimally small. The
path encloses any current present in the sheet. Applying the Ampère law, we find
B1t L þ B2t L ¼ μ0 K L:
Thus the change in the field across the sheet is
B2t B1t ¼ μ0 K (1.30)
or in general
! ! !
ð B2 B1 Þ ^n ¼ μ0 K: (1.31)
18 Basic concepts
Figure 1.9 Fields near a current sheet.
Figure 1.10 Refraction of the magnetic field crossing a current sheet.
The tangential component of B changes by an amount proportional to the current

density when crossing the sheet.
Finally, let us consider the angles between the magnetic field vectors and
the normal to the current sheet, as shown in Figure 1.10. In region (2), the
magnetic field vector makes an angle with the normal to the current sheet
given by
B2t
tan θ2 ¼ :
Bn
1.8 Inductance 19
The corresponding angle with the normal in region (1) is given by
B1t
tan θ1 ¼
Bn
B2t μ0 K
¼
Bn
μ K
¼ tan θ2 0 :
Bn
We see that in crossing the current sheet, the vector B is refracted in the direction of
the field from the current sheet, i.e., toward the normal for the positive current
density K shown here.
1.8 Inductance
Consider a coil with N turns. We define the flux linkage to be the product of the
magnetic flux going through the coil multiplied by the number of turns. The flux
linkage is proportional to the current flowing through the coils. We define the
coefficient of proportionality to be the self-inductance L of the coil. Thus we have
NΦB
L¼ : (1.32)
I
Example 1.7: self-inductance of an ideal solenoid

For an ideal solenoid with N turns in a length d and radius R, the field from Equation
1.28 is
μ0 N I
Bz ≃ :
d
The flux in the solenoid is

μ0 N I 2
ΦB ¼ πR ;
d
so the self-inductance is
μ0 N 2 2
L¼ πR : (1.33)
d
This result ignores any end effects present in a real solenoid.
If we now consider two coils, the mutual inductance M is defined as the flux
linkage in the second coil due to the current in the first coil. Thus we have
20 Basic concepts
N2 Φ2;1
M¼ ; (1.34)
I1
where Φ2;1 is the flux in coil 2 due to the current in coil 1. In general, M can be
defined using the Neumann equation [7]
ðð ! !
μ0 dl1 · dl2
M¼ ; (1.35)
4π r12
where r12 is the distance from the current element in the first coil to the current
element in the second coil. Note that this shows that M is a constant times a geometric
factor. The symmetry of this equation between the two coils shows that M of coil 2
due to current in coil 1 is the same as M for coil 1 due to current in coil 2.
Example 1.8: mutual inductance of two coaxial solenoids

Assume we have two coaxial solenoids. The first solenoid has length d1 and both
solenoids have approximately the same radius R. Then
μ0 N1 I1 2
Φ2;1 ¼ πR
d1
and the mutual inductance is

μ0 N 1 N 2 2
M¼ πR : (1.36)
d1
The force between two coaxial coils can be expressed in terms of the derivative
of their mutual inductance.[8]
∂M
Fz ¼ I 1 I 2 : (1.37)
∂z
1.9 Energy stored in the magnetic field

We can obtain a rough estimate for the energy stored in a magnetic field8 by
considering a simple LR circuit, as shown in Figure 1.11. From Kirchhoff’s circuit
laws,[9] we have
dI
IV ¼ I 2 R þ LI :
dt
8
We will reexamine this question more rigorously in Chapter 10.
1.9 Energy stored in the magnetic field 21
Figure 1.11 LR circuit.
The power P = IV generated by the battery is distributed between the power lost to
resistive heating in R and the power in the magnetic field associated with the
inductor L. The energy in the magnetic field is then
ð
Ð Ð dI
WB ¼ P dt ¼ LI dt ¼ LI dI
dt (1.38)
1 2
¼ LI :
2
If we consider the inductor to be a long solenoid, then using Equations 1.28 and 1.33,
1 μ0 N 2 B2 d 2 1 B2
WB ¼ πR2 2 2 ¼ πR2 d
2 d μ0 N 2 μ0
and the energy density in the magnetic field is
B2
wB ¼ : (1.39)
2 μ0
References
[1] F. Bonaudi, Magnets in particle physics, in S. Turner (ed.), CERN Accelerator School:
Magnetic Measurements and Alignment, Montreux, 1992, p. 2–5.
[2] O. Darrigol, Electrodynamics from Ampère to Einstein, Oxford University Press,
2000, p. 142.
[3] B. Hunt, The Maxwellians, Cornell University Press, 1991, p. 70–71, 245–247.
[4] D. Halliday & R. Resnick, Physics for Students of Science and Engineering, 2nd ed.,
Wiley, 1962, p. 759–760.
[5] K.L. McDonald, Topology of steady current magnetic fields, Am. J. Phys. 22:586,
1954.
[6] M. Lieberherr, The magnetic field lines of a helical coil are not simple loops, Am. J.
Phys. 78:1117, 2010.
22 Basic concepts
[7] G. Harnwell, Principles of Electricity and Magnetism, 2nd ed., McGraw-Hill, 1949,
p. 322.
[8] P. Lorrain & D. Corson, Electromagnetic Fields and Waves, 2nd ed., Freeman, 1970,
p. 360.
[9] D. Tomboulian, Electric and Magnetic Fields, Harcourt, Brace & World, 1965,
p. 267–269.
2
Magnetic materials
We saw in the previous chapter that magnetic fields are produced in a vacuum by
currents in conductors. In this chapter, we will consider magnetic effects asso-
ciated with matter. All materials have spin and orbital motion of charges at the
atomic scale. For most materials, the random orientations of these internal
currents tend to cancel out significant magnetic effects. However, in certain
magnetic materials, such as iron or permanent magnets, these internal currents
do not cancel, and there is a net external effect that also produces or enhances
magnetic fields. In the field theory we have been discussing, the magnetic field
B must be a continuous function of position. Thus, in magnetic materials, the
macroscopic field B must be an average of the rapidly varying local fields
surrounding the atoms in the material.[1]
2.1 Magnetization
When a magnetic material is placed in an external magnetic field, magnetic dipoles
in the material set up internal fields that modify the applied field. We saw in
Equation 1.10 that a current loop has an associated magnetic moment m.
We define the magnetization vector M as the average magnetic moment per unit
volume
! X! m i Ni Ii Ai
M ¼ ¼ ^n i ; (2.1)
i
V V
where Ai is the area of loop i. The volume V must be large enough so the sum is
statistically significant, yet small enough so that we can treat the variation of M as
approximately continuous. For uniform magnetization, all the internal current
loops cancel. However, as shown in Figure 2.1, there is still a net current around
the surface of the material.
23
24 Magnetic materials
Figure 2.1 Magnetic moment loops in a uniformly magnetized material.
It follows that
X
Im A
i i Im
M¼ ¼ ¼ Km ;
AL L
where Im is the Amperian loop current and Km is the magnetization surface current
density. In vector terms,
! !
K m ¼ M ^n : (2.2)
Now consider the case when there is a nonuniform distribution of magnetization

inside the volume, as shown in Figure 2.2. For the current loop on the left we have
I0
Mz0 ¼ ;
Δz
while the loop on the right gives

∂Mz I 00
Mz00 ¼ Mz0 þ Δy ¼ :
∂y Δz
Along the line AB, there is a net current
Im I 00 I 0 ∂Mz
¼ ¼ Δy: (2.3)
Δz Δz ∂y
2.1 Magnetization 25
Figure 2.2 Magnetic moment loops in a nonuniform magnetized material.
If we integrate around the front face (CDEF), we find

þ
!! 0 ∂Mz ∂Mz
M ·dl ¼ Mz þ Δy Δz Mz0 Δz ¼ Δy Δz:
∂y ∂y
Therefore, using Equation 2.3 we find, in analogy with the Ampère law, that
þ
!!
M ·dl ¼ Im ; (2.4)
where Im is the effective number of internal amp-turns through the loop of integra-
tion. Applying Stokes’s theorem, we can rewrite Equation 2.4 as
ð ð
! ! ! !
ðr M Þ · dS ¼ J m · dS :
The Amperian volume current density is then

! !
J m ¼ r M: (2.5)
This implies that the volume current density vanishes for homogeneous materials.
When magnetic materials are present, the Biot-Savart law must be generalized to
ð ! ! ! ð ! ! !
! μ0 ð K þ K m Þ R μ0 ð J þ J m Þ R
B ¼ dS þ dV;
4π R3 4π R3
where J and K without a subscript refer to the conduction current densities of free
charges.
2.2 Magnetic field intensity

Returning to the Ampère law, we take into account the effect of Amperian currents
by writing
þ
!!
B ·dl ¼ μ0 ðI þ Im Þ:
Using Equation 2.4, we have

þ þ
! ! ! !
B · dl ¼ μ0 I þ μ0 M · dl :
Combining the line integrals gives

þ
! ! !
ð B μ0 M Þ · dl ¼ μ0 I:
We can define an auxiliary vector H, known as the magnetic intensity, as

!
! B !
H ¼ M; (2.6)
μ0
so that the magnetic flux density is
! ! !
B ¼ μ0 ðH þ M Þ: (2.7)
In free space, this reduces to

! !
B ¼ μ0 H : (2.8)
The vectors B and H describe different aspects of the same magnetic field.
The advantage of working with this new vector is that the Ampère law for H
þ
!!
H ·dl ¼ I (2.9)
only depends on the true conduction current that crosses the path of integration,
despite the presence of any magnetic materials. Applying Stokes’s theorem to
Equation 2.9, we get
ð ð
! ! ! !
ðr H Þ · dS ¼ J · dS :
Since the surface of integration is arbitrary, we find that H satisfies the differential
equation
! !
r H ¼ J ; (2.10)
where J is the conduction current density.
We have introduced the vector H here as a useful mathematical artifact to take

into account the averaged behavior of atomic currents in matter. However, there is
a long history of trying to understand the physical meaning of H and of examining
the distinction between vectors B and H.[2, 3]
2.3 Permeability and susceptibility

In homogeneous and isotropic materials, the magnetization is usually found to be
proportional to the magnetic intensity, or
! !
M ¼ χ H; (2.11)
where the dimensionless coefficient χ is known as the susceptibility. Using this, we

find that Equation 2.7 can be rewritten
! ! ! !
B ¼ μ0 H þ μ0 χH ¼ μ0 ð1 þ χÞ H :
It is useful to define the permeability for magnetic materials as

μ ¼ μr μ0 ; (2.12)
where μr is a dimensionless quantity known as the relative permeability.

The susceptibility and the relative permeability are related by
μr ¼ 1 þ χ; (2.13)
and the general relation between B and H can be written as

! !
B ¼ μ H: (2.14)
Materials where the directions of B and H are parallel are called linear materials.
2.4 Types of magnetism

Normally the random orientations of atomic orbits and particle spins cause
the associated magnetic moments in a material to cancel, so there is no net
magnetization. However, when an external magnetic field is applied to the
material, the electron orbital velocity increases for one direction of circula-
tion and decreases for the opposite direction. This results in a small net
magnetic moment that is present in all materials and is known as diamagnet-
ism. The difference in frequency between the two orbital directions can be
shown to be
eB
Δω ¼ ;
2me
where e is the electron charge and me is its mass.[4] The resulting net magnetic
moment Δm is
e2 r2 B
Δm ≃ ; (2.15)
4me
where r is the radius of the atomic orbit. Note that the induced moment is opposite
to the direction of the applied magnetic field. This effect is very small and is
masked by larger effects in paramagnetic and ferromagnetic materials.
Diamagnetic materials have χ < 0 and μr < 1, independent of temperature.
In paramagnetic materials, the magnetic moments from orbital motion and spins
do not cancel, resulting in a small permanent moment. In an external magnetic
field, torques tend to align the magnetic moments with the direction of the field.
In these materials, the induced fields act to increase the magnitude of the applied
field and μr > 1. The degree of alignment is decreased by internal collisions,
vibrations and thermal agitation inside the material. The resulting magnetization
is a spin effect given by the Langevin equation

mH kT
MðHÞ ¼ Nm coth ; (2.16)
kT mH
where N is the number of atoms per unit volume, k is Boltzmann’s constant, and T is
the temperature.[5] Note that the dependence of M on the magnetic intensity H is
temperature dependent and nonlinear in general. However, in the case where
kT 1, the paramagnetic susceptibility is given by Curie’s law
mH
M N m2
χ¼ ¼ : (2.17)
H 3kT
In certain crystalline materials where one of the electron shells is not filled, it is
possible for one or more electrons to have unbalanced electron spins. In these
ferromagnetic materials, it is possible to achieve a very high degree of magnetic
alignment. Below a characteristic temperature known as the Curie temperature,
coupling is possible between neighboring atoms, which can act together in regions
known as domains. In an external magnetic field, the size of favorably oriented
domains can grow. In addition, the magnetization directions in each domain tend to
align with the external field. The dependence of B or M on H is very nonlinear in
ferromagnetic materials. For example, we show a BH curve for a low-carbon steel
alloy in Figure 2.3.[6] The flux density increases extremely rapidly for small values
Figure 2.3 A BH curve for SAE 1020 low-carbon steel.
Figure 2.4 A μH curve for SAE 1020 low-carbon steel.
of H. Then, as H increases further, the magnetization domains begin to saturate, and

the curve starts to level out. Finally, for very large applied fields, the magnetization
domains become completely saturated, and the growth in B is only due to the
increase in the conduction current.
The relative permeability in a ferromagnetic material can be much larger than 1,
as shown in Figure 2.4. The permeability in this example quickly reaches
a maximum value ~1,525 for an excitation of 365 A/m. For larger excitations,
the relative permeability decreases steadily.
If the current and thus H is cycled up and down in a ferromagnetic material, we find
that a plot of B versus H has the characteristic shape illustrated in Figure 2.5.
Figure 2.5 Hysteresis loop for a ferromagnetic material.
The material exhibits hysteresis because B is not a unique function of H. The value of
B depends on the previous history of how H was varied. The dotted line shows the
increase in B as H is increased, starting from an unmagnetized sample. After the initial
excitation, BðHÞ follows the arrows around the hysteresis loop. This effect arises
because the domain boundaries don’t completely return to their previous locations
when H is reversed. The remanence or remanent field BR is the value of B when H is
returned to 0. The nonzero value for the remanence shows that the material can remain
magnetic when the external driving current is turned off. This leads to the possibility
of making permanent magnets,1 so long as the temperature remains below the Curie
temperature. The coercivity HC is defined2 as the value of H in the negative direction
that is required to get B = 0. The intrinsic coercivity HCi is the reverse field required to
remove the magnetization in a plot of M versus H.[7] The hysteresis loop is symmetric
around the origin. Heat is generated for each cycle around the hysteresis loop.3
Magnetic materials with low values of coercivity are designated as “soft.” In this
case, it is easy for the magnetization to change direction as the external current
changes, so these materials are suitable for ac operation.[8] A number of soft
magnetic materials that have high permeabilities at low values of B are listed in
Table 2.1. Also listed are the values of B corresponding to the maximum perme-
ability, the coercivity, and the saturation value of B. The electrical resistivity ρe of
the material is important for considerations of eddy current4 losses in time-varying
operations. Values for pure iron are also listed for comparison.
1
Permanent magnets are discussed in more detail in Chapter 9.
2
Historically, this quantity is known as the coercive force.
3
The energy loss in the hysteresis loop is discussed in Chapter 10.
4
Eddy currents are discussed in more detail in Chapter 10.
2.5 Magnetic circuits 31
Table 2.1 Selected soft magnetic alloys [9]
Alloy Initial μr Max μr B at max μr [T] Hc [Oe]* Bsat [T] ρe [μΩ-cm]

Sinimax 2,200 50,000 0.54 0.06 1.10 90
Monimax 3,000 60,000 0.62 0.06 1.45 80
16 Alfenol 4,000 80,000 0.35 0.044 0.80 153
Mumetal 20,000 100,000 0.20 0.30 0.65 60
1040 alloy 20,000 100,000 0.20 0.20 0.60 56
Supermalloy 55,000 300,000 0.40 0.006 0.68–0.78 65
Iron 150 5,000 0.80 1.00 2.14 10
* 1 Oe = 1 10−4 T/μ0
2.5 Magnetic circuits

Ferromagnetic materials tend to concentrate magnetic flux. The permeability of
iron can be thousands of times larger than that of free space. When analyzing
“circuits” made up of coils and pieces of ferromagnetic materials, it is common to
assume that all the flux goes through and is uniformly distributed inside the
ferromagnetic material. A coil sets up Amperian currents in the material near the
coil that continue to initiate further currents along the material.[10] Assume that
we have an iron ring that is energized with a coil, as shown in Figure 2.6. From
the Ampère law for H (Equation 2.9), we find that B inside the ring is
μN I
Bϕ ¼ ;
2πR
and the magnetic flux inside the ring is
2
μπr
ΦB ¼ N I: (2.18)
2πR
This expression is only approximate because it ignores any leakage of the flux from
the ring. Equation 2.18 resembles Ohm’s law for circuits
V
I¼ ;
Re
where NI corresponds to the driving voltage and the resulting flux corresponds to
the current. The term “analogous to the electrical resistance” Re is called the
reluctance, which we see can be expressed as
L
R¼ ; (2.19)
μA
where L is the path length in the material, and A is its cross-sectional area.
Figure 2.6 Magnetic circuit (Rowland ring).
Figure 2.7 Magnetic circuit for a C-dipole magnet.
Example 2.1: C-shaped electromagnet

Consider an electromagnet with an air gap, as shown in Figure 2.7. From the
Ampère law
N I ¼ Hi Li þ Hg Lg ; (2.20)
where L is the mean length through the region, and the subscripts i and g refer to the
iron and the gap. On the assumption there is no leakage flux, we have Bi Ai ¼ Bg Ag .
Substituting into Equation 2.20, we get
Bi Bg Bg Ag L i Bg
NI¼ Li þ Lg ¼ þ Lg ;
μ μ0 Ai μ μ 0
so that

Li Lg
N I ¼ B g Ag þ :
μAi μ0 Ag
2.6 Boundary conditions between regions with different μ 33
Since μ μ0 , we can neglect the reluctance in the iron for reasonable values of L and
A and find that the field in the gap is
μ0 N I
Bg ≃ : (2.21)
Lg
2.6 Boundary conditions between regions with different μ

We now consider the constraints on the magnetic field at the boundary between two
regions having different permeabilities. The pillbox construction in Figure 1.8 still
applies in this case, so
B1n ¼ B2n : (2.22)
However, when dealing with permeable materials, Equation 1.30 for the tangential
component of B is no longer accurate. In the present case, we can use the Ampère
law for H for a path that encompasses both sides of the boundary to find that
H2t H1t ¼ K; (2.23)
where we recall that K is the surface current density. If there is no surface current at
the boundary, then the tangential component of H is continuous.
Consider a linear material with K = 0 and let θ be the angle between the magnetic
field and the normal to the surface, as shown in Figure 2.8. Then we have from
Equations 2.22 and 2.23
μ1 H1 cos θ1 ¼ μ2 H2 cos θ2
H1 sin θ1 ¼ H2 sin θ2 :
Figure 2.8 Refraction of H at the boundary between two permeable materials.

Dividing these two equations, we find that

tan θ1 μ1
¼ : (2.24)
tan θ2 μ2
If the field is incident from region 1 and μ is much larger in region 2 than region 1,
then θ1 will be smaller than θ2. Thus, the field will make a larger angle with respect
to the normal in the region with larger μ.
As a special case, consider the boundary between vacuum in region 1 and
iron in region 2 when K = 0. Then, using Equation 2.23, we find
Bt2 μ2
¼ :
Bt1 μ1
It is often useful to make the approximation that μ2 in the iron is infinite. Then
the right-hand side of this equation is infinite, which demands that Bt1 ¼ 0.
In this case, the field in the vacuum region would be perpendicular to the iron
surface.
2.7 Method of images

Some magnetostatic problems with planar or spherical boundaries can be solved
using the method of images.[11] In this method, the presence of an iron boundary is
replaced with virtual currents, which, together with the currents from true con-
ductors, reproduce the correct boundary conditions. Consider a current I in a
region 1 with permeability μ1 a distance d from the planar boundary with
a region 2 of material with permeability μ2, as shown in Figure 2.9. Let us designate
case (a) to be the situation when the observation point P is in region 1. Then, to
satisfy the boundary conditions, we assume there is a virtual current I 0 in region 2.
Figure 2.9 Image currents near a plane boundary.

We designate case (b) to be the situation when the observation point is inside the
iron in region 2. In this case, there is no conduction current in the observation
region. We assume the resultant field is due to a virtual current I 00 in region 1.
By symmetry, the original current and the two image currents lie on a line perpen-
dicular to the boundary. We assume that the currents all flow in the same direction
and that the distances of the currents from the boundary are equal and look for
a solution for the magnitude of the currents. Consider a point P along the boundary
at a distance y from the line connecting the currents. In case (a), the boundary
conditions are
ðaÞ d
Ht ¼ ðI I 0 Þ
þ y2 Þ
2πðd2
ðaÞ μ1 y
Bn ¼ ðI þ I 0 Þ;
2πðd2 þ y2 Þ
whereas for case (b) we have
ðbÞ d
Ht ¼ I 00
2πðd2 þ y2 Þ
ðbÞ μ2 y
Bn ¼ I 00 :
2πðd2 þ y2 Þ
Both cases must give the same solution for any position y along the boundary. Thus,
we obtain the two equations
I I 0 ¼ I 00
μ1 I þ μ1 I 0 ¼ μ2 I 00 :
Solving these two equations for the unknown magnitudes of I 0 and I 00 , we find the
solution [12]
μ2 μ1
I0 ¼ I (2.25)
μ2 þ μ1
and
2μ1
I 00 ¼ I: (2.26)
μ1 þ μ2
In the special case (a) when region 1 is vacuum and region 2 is infinitely permeable
iron, Equation 2.25 reduces to I 0 = I.
Figure 2.10 Line current near a planar iron slab.
Example 2.2: Field enhancement in a planar iron slab

Consider a filamentary conductor a distance d away from an infinite slab of iron.
We use the method of images and replace the iron slab with a virtual filament, as
shown in Figure 2.10.
Let us examine the field for locations P along the line perpendicular to the boundary.
Assume the observation point is at x and the filament is located at a. Then the boundary
is located at a + d, and the image current is at a + 2d. The field at P is
μ0 I μ0 I 0
By ðxÞ ¼ :
2πða xÞ 2πða þ 2d xÞ
Assuming μ1 ¼ μ0 and μ2 ¼ μr μ0 and using Equation 2.25, we find that

μ0 I 1 μr 1 1
By ðxÞ ¼ þ :
2π a x μr þ 1 a þ 2d x
In the limit when μr →∞ we find that By ða þ dÞ ¼ 0, as it should at the surface of the

iron. We define the iron enhancement factor E(x) to be the ratio of the field at x with
the iron present to the field at x from the conductor by itself. In the limit μr →∞, the
enhancement factor is
ax
EðxÞ ¼ 1 þ :
a þ 2d x
We show the dependence of the enhancement factor on x in Figure 2.11.
The enhancement is greater than 1 in the region to the left of the filament and then
becomes less than 1 in the region between the filament and the iron boundary.
It is also possible to use the method of images to solve problems with more
complicated arrangements of planar surfaces. For example, a line current between
two parallel iron boundaries can be solved using an infinite series of image
Figure 2.11 Planar iron enhancement factor for a = 8 and d = 2.
Figure 2.12 Images for a line current near a circular boundary.
currents,[13] and a line current near the corner of two perpendicular iron planes can
be solved using three image currents.5
The method of images is also useful when considering a line current near
a circular boundary surface. Consider a line current I0 at radius a in a region with
permeability μ1 near a circular boundary of radius R of a region with permeability
μ2 , as shown in Figure 2.12.
In case (a), the magnetic field at a point P inside the aperture of the magnet (r < a)
can be written as the sum of the fields due to the conduction current I0 at r = a and
the image current I1 at r ¼ r1 inside region 2, where
5
This is discussed in Chapter 5, Section 5.13.
μ2 μ1
I1 ¼ I0 (2.27)
μ2 þ μ1
R2
r1 ¼ : (2.28)
a
In case (b), the magnetic field at a point Q inside region 2 (r > R) can be written as
the sum of the field from two image currents in region 1, I2 at r = 0 and I3 at r = a,
where
μ2 μ1
I2 ¼ I0 (2.29)
μ2 þ μ1
2μ1
I3 ¼ I0 : (2.30)
μ2 þ μ1
The justifications for these statements come from the solution of the corresponding
boundary value problem, which we will discuss in Chapter 4, Section 4.2.
References
[1] L. Eyges, The Classical Electromagnetic Field, Dover, 1980, p. 142–145.
[2] W. Panofsky & M. Phillips, Classical Electricity and Magnetism, 2nd ed., Addison-
Wesley, 1962, p. 143–144.
[3] J. Roche, B and H, the intensity vectors of magnetism: a new approach to resolving
a century old controversy, Am. J. Phys. 68:438–449, 2000.
[4] D. Tomboulian, Electric and Magnetic Fields, Harcourt, Brace & World, 1965,
p. 213–217.
[5] C. Kittel, Introduction to Solid State Physics, 3rd ed., Wiley, 1968, p. 432.
[6] http://magweb.us/free-bh-curves
[7] C. Rudowicz & H. Sung, Textbook treatments of the hysteresis loop for ferromagnets:
survey of misconceptions and misinterpretations, Am. J. Phys. 71:1080–1083, 2003.
p. 403–406.
[9] Magnetically soft materials, American Society for Metals (ASM) Handbook, vol. 2,
1990, p. 761–781.
p. 405.
[11] P. Hammond, Electric and magnetic images, Inst. Elec. Eng. Monograph 379:306,
1960.
[12] P. Silvester, Modern Electromagnetic Fields, Prentice-Hall, 1968, p. 178–179.
[13] K. Binns & P. Lawrenson, Analysis and Computation of Electric and Magnetic Field
Problems, Pergamon Press, 2nd ed., 1973, p. 38–45.
3
Potential theory
The use of potential functions represents another approach to solving problems in

magnetostatics. We first treat the vector potential A. The physical significance of
the vector potential has been debated for many years and it plays an important role
in time-dependent phenomena.[1] However in classical magnetostatics, the poten-
tials are usually treated as auxiliary mathematical quantities that are used to
simplify the calculation of the magnetic fields. A scalar potential Vm can also be
defined that satisfies Laplace’s equation in current-free regions.
3.1 Vector potential

In our discussion of curl B in Chapter 1, we saw that the magnetic field could be
expressed in the form given by Equation 1.21.
ð!
! μ0 J
B ¼ r dV: (3.1)
4π R
We can rewrite this equation as
! !
B ¼r A; (3.2)
where we define the vector potential A as

ð!
! μ0 J
A ¼ dV: (3.3)
4π R
This equation is valid in rectangular coordinates,[2] where the direction of A is the

same as that of J.
Since B is defined as the curl of a vector by Equation 3.2, the divergence equation
!
for B, r· B ¼ 0, is automatically satisfied because of the vector identity
!
r·r V ¼ 0. Using Stokes’s theorem, we have
39
40 Potential theory
þ ð ð
!! ! ! ! !
A ·dl ¼ ðr A Þ· dS ¼ B · dS :
Thus the magnetic flux through some surface S is given by the contour integral of
A around the perimeter of S.
þ
!!
ΦB ¼ A ·dl : (3.4)
Any solution for the vector potential is unique, provided that the sources are
confined to a finite region of space.[3]
There is an uncertainty in the defining relation for A in Equation 3.2. Any vector
whose curl vanishes can be added to A without affecting the value of B. For
example, the gradient of a scalar function ψ can be added because
r rψðrÞ ¼ 0:
This freedom can be used to fix the divergence of A. Starting with Equation 3.3, the
divergence of the vector potential can be written as
! μ
r· A ¼ 0 I;
4π
where
ð !
! 1
I ¼ J ðr0 Þr· dV 0 :
j!
r !
0
r j
Primes are used to indicate source coordinates. Note that the r operator is defined
in terms of the observation point (field) coordinates, while the integration and
current density depend on source coordinates. Replacing the r operator with one
defined in terms of source coordinates, we find
ð !
! 0 0 1
I ¼ J ðr Þr · ! !0 dV 0 :
jr r j
Now we can integrate using Equation B.3 and find that

0 ! 0 1
ð J ! r ð
1 ! 0 0
I ¼ r · ! !0 dV þ ! !0 r0 · J !
0 @ A 0
r dV :
jr r j jr r j
!
The second term on the right-hand side vanishes because r0 ·J 0 ¼ 0 from
Equation 1.4. We use the divergence theorem to transform the other integral and get
ð ! !0 !
Jðr Þ !
I¼ ! ! 0 · dS :
jr r j
We choose to evaluate this integral over a large radius sphere where all the current
sources vanish. Then I ¼ 0 and we conclude that
!
r· A ¼ 0: (3.5)
This result is equivalent to setting the arbitrary scalar function ψ ¼ 0 and is known
as the Coulomb gauge. Equation 3.5 represents a constraint on the components of
the vector A.
Using Equation 1.24, we can relate the vector potential to the conduction current
density J.
! !
r ðr A Þ ¼ μ0 J ; (3.6)
Then, using the vector identity B.7, we obtain

! ! !
rðr· A Þ r2 A ¼ μ0 J :
Using Equation 3.5, we can eliminate the gradient term and find that
! !
r2 A ¼ μ0 J : (3.7)
This is the vector Poisson equation for A. The solution is valid both inside and
outside of the conductor. Equation 3.3 is a particular solution of this equation.
When J = 0, Equation 3.7 is called the Laplace equation. Solutions of the Laplace
equation are known as harmonic functions. In Cartesian coordinates, the quantity
!
r2 A has the three scalar components r2 Aα , where α corresponds to x, y, and z. For
example,
r2 Ax ¼ μ0 Jx :
!
In non-Cartesian coordinate systems, the components of r2 A must be found from
Equation B.7.
3.2 Vector potential in two dimensions

In cases where the current density is constant in one dimension, it is possible to
develop a two-dimensional version of the theory that is significantly simpler than
the general three-dimensional case. Of course this is only an approximation to the
real world, but the approximation may be quite good, for example in the central
42 Potential theory
region of a long magnet, far from its ends. The great power of the two-dimensional
approximation will be most apparent in Chapter 5, where we develop the theory
using complex analysis.
Let us consider an infinitely long current filament along the z axis. Direct
integration using Equation 3.3 to ±∞ diverges. It is possible to perform the
integration over a long, but finite length and then to develop an expression for Az
in a power series in ρ/L.[4] However, we will adopt another approach using the
differential Equation 3.2. The field from the filament was given in Equation 1.15
and so we have
! μI ^
r A ¼ 0 ϕ:
2π ρ
The ρ component of the curl in cylindrical coordinates gives
∂Aρ ∂Az μI
¼ 0 :
∂z ∂ρ 2π ρ
The vector potential is constant in the z direction, so the derivative with respect to
z vanishes and we have
μ0 I dρ
dAz ¼ :
2π ρ
Integrating this equation, we find

μ0 I
Az ¼ lnðρÞ þ c:
2π
If write the constant of integration c in terms of the value of Az at some reference

radius ρ0 , then
μ0 I
c¼ lnðρo Þ
2π
and the vector potential for the infinite current filament is

μ0 I ρ
Az ðρÞ ¼ ln : (3.8)
2π ρo
The presence of any constant terms in Az is not important since they will be
removed when taking derivatives to find B.
Figure 3.1 Bifilar conductors.
Example 3.1: bifilar conductors

One method of handling the issue of the reference radius is to argue that a single
filament of current is not physical because any real current must be part of a closed
circuit and therefore must have a return filament somewhere with the current flowing
in the opposite direction. Consider the bifilar configuration shown in Figure 3.1.
Using Equation 3.8 for both the positive (p) and negative (n) filaments, we find
μ0 I
Az ¼ lnðRp Þ lnðRpo Þ lnðRn Þ þ lnðRno Þ ;
2π
where R is the distance from the current element to the observation point P. If both
filaments return at the same reference position, the dependence on R0 drops out and
we find

μ0 I Rp
Az ðx; yÞ ¼ ln : (3.9)
2π Rn
The sign of Az depends on the relative magnitudes of Rp and Rn .
We can find the two-dimensional vector potential for current distributions by

superposition of the vector potential for a filament. The vector potential for a two-
dimensional current sheet is given by
ð
μ0 0 R
Az ðx; yÞ ¼ Kz ðs Þ ln ds0 ; (3.10)
2π Ro
where Kz ¼ dI=ds is the sheet current density. The vector potential for a block
conductor with finite cross-section is
ðð
μ0 0 0 R
Az ðx; yÞ ¼ Jz ðx ; y Þln dx0 dy0 : (3.11)
2π Ro
44 Potential theory
The magnetic field has two components given by

∂Az
Bx ¼
∂y (3.12)
∂Az
By ¼
∂x
and, except inside a conductor, Az satisfies the scalar Laplace equation
∂2 Az ∂2 Az
r2 Az ¼ þ 2 ¼ 0: (3.13)
∂x2 ∂y
3.3 Boundary conditions on A

The boundary conditions on the vector potential A can be determined from the
boundary conditions on the magnetic field. Consider a boundary surface S located
at the intersection of two regions of space. We know from Equation 2.22 that the
normal component of B must be conserved across S. Thus the magnetic flux
crossing S is conserved and from Equation 3.4
þ
!!
A ·dl ¼ ΦB ;
we see that the line integral of A around the perimeter of S must also be conserved.
Therefore the tangential component of A must be conserved on the boundary.
!ð1Þ !ð2Þ
At ¼ At : (3.14)
The boundary condition on the tangential component of H given in Equation 2.23

can be written in the form
1 ! 1 ! ð1Þ
ð2Þ
ðr A Þð2Þ
t ð1Þ ðr A Þt ¼ K; (3.15)
μ μ
where K is the surface current on S, if applicable. These two vector relations
provide four constraints on A at the boundary.[3]
The boundary conditions can be considerably simpler in two dimensions.
Assume, for example, that the problem is uniform in the z direction and that we
have a boundary between two regions, as shown in Figure 3.2. All the current is
along z, so A only has the component Az . In this case, the boundary surface is
parallel to the x-z plane and the tangential component of A is along z. From
Equation 3.14, we have
Azð1Þ ¼ Að2Þ
z ;
Figure 3.2 Boundary in two dimensions.
so the vector potential is continuous across the boundary. Since

!
r A ¼ ^x ∂y Az ^y ∂x Az ;
the tangential component of the curl is along x. Thus Equation 3.15 gives
1 1
ð2Þ
∂y Að2Þ ð1Þ
z ð1Þ ∂y Az ¼ Kz :
μ μ
In this case, we have two constraints that must be satisfied at the boundary.
3.4 Vector potential for a localized current distribution

Consider a localized distribution of current, as shown in Figure 3.3. We pick some
origin O inside the distribution and examine the potential at a field location P. Since
the distance between the source point at r0 and the field point at r is
1=2
R ¼ fr2 þ r0 2rr0 cos θg
2
;
we can express the inverse distance as

1 !
r ·!
0
1 r
≃ þ 3 þ :
R r r
Then the first two terms in the multipole expansion for A are
ð ! ð
!! μ 1 ! !0 r 0!
J ð r Þ dV 0 þ 3 · !r J ð!
0
Að r Þ≃ 0 r Þ dV 0 þ : (3.16)
4π r r
The first integral vanishes because the current distribution consists of closed
loops. In the second integral, J corresponds to one of the components of A,
46 Potential theory
Figure 3.3 Localized current distribution.
which we specify with the index i, while r0 is part of the scalar product with r,
which we specify with the index j. Thus the integral has the form
ð
I ¼ r0j Ji0 dV 0 :
Using the vector identity B.3, we have

! ! !
r0 ·ðr0i J 0 Þ ¼ r0i r0 ·J 0 þ J 0 ·r0 r0i :
The first term vanishes because of Equation 1.4 and in the second term, we have
r0 r0i ¼ î. Thus
!
Ji0 ¼ r0 ·ðr0i J 0 Þ
and so we have
ð
!
I ¼ r0 ·ðr0i J 0 Þr0j dV 0 :
We can do the integral by parts with

u ¼ r0j
!
dv ¼ r0 · r0i J 0 :
This gives
ð ð
!0 0 !
I¼ r0i r0j J dS r0i J 0 ·r0 r0j dV 0 :
The first term vanishes for a surface outside the charge distribution. In the second
term, the gradient in the integrand vanishes except for the j component. Thus
ð
I ¼ r0i Jj0 dV 0 :
Comparing with Equation 3.16, this implies that

ð ð
0 ! !
! !
ð r · r Þ J dV ¼ ð! r ·J Þ!
0 0
r dV 0 : (3.17)
Now consider the triple vector product from Equation B.1.

! ! ! !
r ð! r J0 Þ ¼ ! r ð!r ·J 0 Þ J 0 ð!
r ·!
0 0 0
r Þ:
Substituting Equation 3.17, we have
ð ð
0 ! !
2 ð r · r ÞJ dV ¼ !
! ! r ð!
0 0 0
r J 0 ÞdV 0 : (3.18)
Substituting this back into Equation 3.16, we find that the vector potential is
ð
!! 1 μ0 ! r !
!
0
Að r Þ≃ 3
r J 0 dV 0 :
2 4π r
We define the magnetic moment of the current distribution as
ð
! ! 0
m ¼½ ! r J ð!
0
r ÞdV 0 : (3.19)
Then the vector potential of the current distribution can be written as [5]
!! μ !m !r
Að r Þ¼ 0 3
: (3.20)
4π r
Thus we find that the elementary form of magnetic matter is a magnetic dipole.
In spherical coordinates, the vector potential for a magnetic dipole is directed in
the azimuthal direction. We can find the magnetic field from the dipole by taking
the curl of A.
1
Br ¼ ∂θ ðAϕ sin θÞ
r sin θ
(3.21)
μ 2m
¼ 0 3 cos θ
4π r
and
1
Bθ ¼ ∂r ðrAϕ Þ
r (3.22)
μ m
¼ 0 3 sin θ:
4π r
48 Potential theory
The remaining component Bϕ ¼ 0. We see that the field of a magnetic dipole falls
off like 1/r3.
We can relate this definition of the magnetic moment, Equation 3.19, with our
discussion in Chapter 1 of the magnetic moment of a planar current loop. We let
! 0 !
J dV →I dl0 :
Then Equation 3.19 gives
þ
! ! 0 !
m ¼½I r dl0 :
!
The magnitude of the quantity ½!
0
r dl0 is the area of a triangular region inside
the current loop. The closed integral then gives the total area A enclosed by the
loop. Thus
!
m ¼ I A ^n ;
which agrees with the result in Equation 1.10.

Now consider a volume of magnetic material that contains many magnetic
dipoles. The contribution to the vector potential from one small part of the overall
volume at r0 can be written in terms of the magnetization vector M as
! !
! μ0 M 0 R
dA ¼ dV 0 :
4π R3
Using the relation

1 1 0 1
r ¼ 2 ^r ¼ r ;
R R R
the vector potential for the whole volume is

ð
! μ0 !0 0 1
A ¼ M r dV 0 :
4π R
Using the vector identity B.6, we can write A as the two terms
ð !0 ! ð !
! μ0 0 M 0 μ 0 r0 M 0
A ¼ r dV þ dV 0 :
4π R 4π R
Then using vector identity

ð ð
! !
r W dV ¼ W ^n dS
3.5 Force on a localized current distribution 49
in the first term and dropping the primes, we get

ð! ð !
! μ0 M n^ μ0 r M
A ¼ dS þ dV: (3.23)
4π R 4π R
! !
We have seen previously from Equation 2.2 that K m ¼ M ^n is the surface
! !
current density and from Equation 2.5 that J ¼ r M is the volume current
density.
3.5 Force on a localized current distribution

The force on a localized current in an applied magnetic field is
ð
! ! !
F ¼ J 0 B dV 0 :
If the field is non-uniform, each component of B can be expanded in a Taylor’s

series.
Bi ð!
r Þ ≃ Bi ð0Þ þ !
r ·rBi ð0Þ þ :
Then
ð ð h
! ! ! !0 ! 0 ! i
F ≃ B ð0Þ J ð r ÞdV þ J ð!
r Þ ð!
0 0
r ·rÞ B ð0Þ dV 0 :
The first integral vanishes since J consists of closed loops. Using Equation B.2, we
have
0 ! ! ! ! !
rð!
r ·BÞ ¼ !
r ðr B Þ þ B ðr !
r Þ þ ð B ·rÞ !
r þ ð!
0 0 0 0
r ·rÞ B :
The first term on the right-hand side vanishes outside the current distribution, while
the second and third terms vanish because the r operator refers to field coordi-
nates. Thus the force can be written as
ð
! ! 0 !
F ¼ J 0 rð! r · B ÞdV 0 :
From Equation B.6, we have

! 0! 0 ! ! ! 0 !
r · B Þ ¼ ð!
r ½J 0 ð! r · B Þr J 0 J 0 rð!
r · B Þ:
The first term on the right side vanishes because r operates on field coordi-
nates, so
50 Potential theory
ð
! ! 0!
F ¼ r J 0 ð!
r · B ÞdV 0 :
Using the relation in Equation 3.18 with r replaced by B, we have

ð ð
! !0 !0 ! !
ð B · r ÞJ dV ¼ ½ B !
0 0
r J 0 dV 0 :
Thus
ðh
! ! !i
F ¼ r ½ B ð!
0
r J 0 Þ dV 0 :
Using Equation 3.19, we can write this in terms of the magnetic moment as
! !
F ¼ rðB !
0
m Þ:
Finally, using Equation B.9
! ! ! ! !
m Þ!
F ¼ B ðr·! m ðr· B Þ þ ð!
m ·rÞ B ð B ·rÞ!
0 0 0 0
m:
The first and fourth terms vanish because m is independent of the field coordinates.
Thus dropping the primes, we find that the force on a magnetic dipole in an
inhomogeneous magnetic field is given by
! !
F ¼ ð!
m ·rÞ B : (3.24)
For a continuous magnetization distribution, the force is given by [6]
ð
! ! !
F ¼ ðM ·rÞ B dV: (3.25)
3.6 Magnetic scalar potential

In regions of space where there are no conduction currents present, the magneto-
static field equations become
!
r H ¼0
!
r· B ¼ 0:
From the curl equation, we know that the magnetic field can be expressed as the
gradient of a scalar function
!
H ¼ rVm ; (3.26)
3.6 Magnetic scalar potential 51
which we call the magnetic scalar potential Vm . Multiplying this equation by μ0

and substituting the expression for B into the divergence equation, we find that
r2 Vm ¼ 0: (3.27)
Thus Vm satisfies the Laplace equation in regions where the conduction current J = 0.
The boundary condition on the transverse component of the field Ht in
Equation 2.23 gives
ð2Þ ð1Þ
Ht Ht ¼K
ð2Þ ð1Þ
(3.28)
∂t Vm þ ∂ t Vm ¼ K:
In the case where no surface current is present, the gradients in the two regions must
be the same. This implies that the scalar potentials in the two regions can only differ
at most by a constant factor c.
Vmð2Þ ¼ Vmð1Þ þ c
From the boundary condition on Bn in Equation 2.22, we have

ð2Þ ð1Þ
Bn ¼ Bn
(3.29)
ð2Þ ð1Þ
μð2Þ ∂n Vm ¼ μð1Þ ∂n Vm :
Thus the product of the permeability and the normal derivative of Vm is continuous
across the boundary.
Consider a field point P nearby a current loop, as shown in Figure 3.4. Assign
a normal vector n to the loop according to the right hand rule. Now displace the
loop by the vector du. The cross product of dl with du gives the area of the shaded
area in the figure. In this case, the solid angle at P subtended by the differential area
dS changes by an amount
! ! ! ! !
d S ·^r ð du dl Þ·^r ðdl ^r Þ· du
dΩ ¼ 2 ¼ ¼ :
r r2 r2
The last equation makes use of the fact that the terms in the triple vector product
permute. To get the total change in solid angle, we sum over all the parts of the loop.
þ ! !
! dl r
dΩ ¼ du ·
r3
!
¼ du ·rΩ
Comparing these two equations, we find that the gradient of the solid angle is
given by
52 Potential theory
Figure 3.4 Solid angle of a displaced current loop.
þ! !
dl r
rΩ ¼ : (3.30)
r3
Now we want to relate this to the scalar potential

!
B
rVm ¼ :
μ
Expressing B using the Biot-Savart law and using Equation 3.30, we find
þ!
I dl !r
rVm ¼
4π r3
I
¼ rΩ:
4π
The gradients on the two sides of the equation are proportional to each other, so Vm
must depend linearly on Ω.
I
Vm ¼ Ω: (3.31)
4π
This indicates that the magnetic scalar potential is directly related to the solid angle
that a current loop subtends at a field observation point. We can ignore any constant
term since it drops out when calculating the field. Note that Equation 3.31 implies
that the scalar potential is not a single-valued function. If P moves from just in front
of the loop to just behind it, the solid angle changes from 2π to –2π since ^n ·^r
changes sign.
3.7 Scalar potential for a magnetic body

In analogy with the potential due to dielectric polarization in electrostatics,[7] the
scalar magnetic potential associated with the magnetization vector is given by
ð! !
1 M· R
Vm ¼ dV; (3.32)
4π R3
which we can express as

ð
1 !0 0 1
Vm ¼ M ·r dV 0 :
4π R
Using the vector identity B.3, we can write this as

"ð !0 ð 0 !0 #
1 0 M 0 r ·M 0
Vm ¼ r· dV dV :
4π R R
Using the divergence theorem for the first term and dropping the primes gives
"ð ! ð ! #
1 M ·^n r·M
Vm ¼ dS dV : (3.33)
4π R R
We can identify the first term on the right-hand side as coming from fictitious
magnetic charges on the surface of the magnetized body with the surface current
density[8]
!
Km ¼ M ·^n (3.34)
and the second term can be described as due to the volume charge density
!
ρm ¼ r·M : (3.35)
This shows that for the purpose of finding the magnetic field outside a magnetized
body, we can replace the body with equivalent magnetic surface and volume
charges.
3.8 General solutions to the Laplace equation

We have found that single components of the vector potential and the scalar potential
both satisfy the scalar Laplace equation r2 F ¼ 0 outside conductor regions.
In general, this is a three-dimensional partial differential equation. Solutions of
the scalar Laplace equation depend on the coordinate system that is used.
A common technique for solving the Laplace equation is to use the method of
separation of variables. This method assumes that the solution is the product of
three terms, each of which only depends on one of the coordinates. As a result, the
partial differential equation in three variables is converted into three ordinary
54 Potential theory
differential equations, each of which depends on a single variable. The method is

most useful when a boundary surface in the problem lies along one of the coordi-
nate directions. The Laplace equation is known to be separable in eleven coordinate
systems.[9] We summarize results here for the three most common systems.
Rectangular coordinates [10]

The Laplace equation in rectangular coordinates is
∂2 F ∂2 F ∂2 F
þ þ ¼ 0: (3.36)
∂x2 ∂y2 ∂z2
We assume the solutions can be written in the form

Fðx; y; zÞ ¼ XðxÞYðyÞZðzÞ:
When this is substituted into Equation 3.36, we obtain
1 d2 X 1 d2 Y 1 d2 Z
þ þ ¼ 0:
X dx2 Y dy2 Z dz2
This equation can only be valid for all values of x, y, or z if each of the three terms is
equal to a constant. Thus we have
1 d2 X
¼ a2
X dx2
1 d2 Y
¼ b2 (3.37)
Y dy2
1 d2 Z
¼ c2 ;
Z dz2
where the three constants have to satisfy the constraint
a2 þ b2 þ c2 ¼ 0: (3.38)
In order to satisfy this equation, the set of constants {a, b, c} must contain both real
and imaginary members. The choice of which constants are real and which are
imaginary depends on the specific conditions for the problem under consideration.
As an example, let us suppose that the constants a and b are imaginary and c is
real. Then we can define a new set of real constants {α, β, γ} such that
a2 ¼ ðiαÞ2 ¼ α2
b2 ¼ ðiβÞ2 ¼ β2
c2 ¼ γ2 :
In the general case, there will be a set of constants that satisfy Equation 3.37.
αn ¼ fα1 ; α2 ; . . .g
βm ¼ fβ1 ; β2 ; . . .g
The constraint Equation 3.38 then becomes
α2n β2m þ γ2nm ¼ 0
and the separated differential equations are
1 d 2 Xn
¼ α2n
Xn dx2
1 d2 Ym
¼ β2m
Ym dy2
1 d2 Znm
¼ γ2nm :
Znm dz2
Then the general solution in Cartesian coordinates has the form

X
∞
Fðx; y; zÞ ¼ Cn eiαn x þ Dn eiαn x Em eiβm y þ Fm eiβm y Gnm eγnm z þ Hnm eγnm z
n;m¼1
þ I0 þ I1 x þ I2 y þ I3 z:
The oscillatory terms could also be written in terms of sines and cosines and the
nonoscillatory terms could be written using hyperbolic sines and cosines. The last
four terms are also a solution of the Laplace equation that allow for continuity of
the potential and the presence of external fields.
Cylindrical coordinates [11]

The Laplace equation in cylindrical coordinates is
1 1
∂ρ ðρ∂ρ FÞ þ 2 ∂2ϕ F þ ∂2z F ¼ 0: (3.39)
ρ ρ
Using the method of separation of variables, we assume that
Fðρ; ϕ; zÞ ¼ RðρÞΦðϕÞZðzÞ:
This leads to the three ordinary differential equations

d dR
ρ ρ þ ðk2 ρ2 n2 ÞR ¼ 0 (3.40)
dρ dρ
56 Potential theory
d2 Φ
þ n2 Φ ¼ 0 (3.41)
dϕ2
d2 Z
k2 Z ¼ 0; (3.42)
dz2
where k and n are constants that may be real or imaginary.
In order for the azimuthal dependence to be single-valued, the parameter n in
Equation 3.41 must be a real integer. Solutions for the function Φ have the general form
Φn ðϕÞ ¼ Cn cos ðnϕÞ þ Dn sin ðnϕÞ (3.43)
when n ≠ 0 and
Φ0 ðϕÞ ¼ C0 ϕ þ D0 (3.44)
when n = 0.
The equations for R and Z have different solutions, depending on the values for
n and k.
(1) real k ≠ 0
The general form of the z dependence in Equation 3.42 is
Zk ðzÞ ¼ Ek ekz þ Fk ekz :

The solution for the radial dependence in Equation 3.40 has the form
Rn ðkρÞ ¼ Gn Jn ðkρÞ þ Hn Nn ðkρÞ;
where Jn and Nn are integer Bessel functions of the first and second kind.[12]
(2) imaginary k ≠ 0
If k = i κ where κ is real, then the z solution is oscillatory.
Zk ðzÞ ¼ Ek eiκz þ Fk eiκz :

In this case, Equation 3.40 becomes

d dR
ρ ρ ðκ2 ρ2 þ n2 ÞR ¼ 0 (3.45)
dρ dρ
and the radial solution is

Rn ðκρÞ ¼ Gn In ðκρÞ þ Hn Kn ðκρÞ;
where In and Kn are modified Bessel functions.[13]

(3) k = 0
The z solution has the form
Z0 ðzÞ ¼ E0 z þ F0 :
The general radial solution for n ≠ 0 is

Rn ð ρÞ ¼ Gn ρn þ Hn ρn : (3.46)
If n = 0, the radial solution is

R0 ð ρÞ ¼ G0 lnr þ H0 : (3.47)
The general form of the solution of the Laplace equation in cylindrical coordi-
nates can then be written in the form
X
Fðρ; ϕ; zÞ ¼ C R ðkρÞΦn ðϕÞZk ðzÞ:
k;n kn n
Some additional information concerning Bessel functions can be found in

Appendix C.
Spherical coordinates [14]

The Laplace equation in spherical coordinates is
1 1 1
∂r ðr2 ∂r FÞ þ 2 ∂θ ðsin θ ∂θ FÞ þ 2 2 ∂2ϕ F ¼ 0: (3.48)
r2 r sin θ r sin θ
The radial and angular parts of this equation can be separated first in the form
Fðr; θ; ϕÞ ¼ RðrÞYðθ; ϕÞ:
This leads to the radial equation

d 2 dR
r nðn þ 1ÞR ¼ 0; (3.49)
dr dr
which has a general solution of the form

Rn ðrÞ ¼ Gn rn þ Hn rn :
The angular equation is
1 1
∂θ ðsin θ ∂θ YÞ þ ∂2ϕ Y þ nðn þ 1Þ Y ¼ 0: (3.50)
sin θ sin 2 θ
58 Potential theory
The solutions for this equation are known as spherical harmonics.[15] The constant
n must be an integer to avoid singularities in Yðθ; ϕÞ at θ = 0 and θ = π. The two
angle coordinates can in turn be separated as
Yðθ; ϕÞ ¼ ΘðθÞΦðϕÞ:
This leads to the two ordinary differential equations

d 2 dΘ m2
ð1 x Þ þ nðn þ 1Þ Θ¼0 (3.51)
dx dx 1 x2
and
d2 Φ
þ m2 Φ ¼ 0; (3.52)
dϕ2
where x ¼ cos θ. In order for the azimuthal dependence to be single-valued in

Equation 3.52, m must be an integer, and Φ has the solution
Φm ðϕÞ ¼ Cm cosðmϕÞ þ Dm sinðmϕÞ
when m ≠ 0 and
Φ0 ðϕÞ ¼ C0 ϕ þ D0
when m = 0. The solution of Equation 3.51 has the form
n ðθÞ ¼ Emn Pn ðcos θÞ þ Fmn Qn ðcos θÞ;

Θm m m
where Pmn and Qn are associated Legendre functions of the first and second
m
kind.[16] In problems with azimuthal symmetry, we have m = 0 and the asso-

ciated Legendre functions Pm
n reduce to the ordinary Legendre polynomials.
P0n ðcos θÞ ¼ Pn ðcos θÞ:
The general form of the solution of the Laplace equation in spherical coordinates
can then be written in the form
X
Fðr; θ; ϕÞ ¼ C R ðrÞΘm
n;m nm n n ðθÞΦm ðϕÞ:
Some additional information concerning Legendre functions is given in

Appendix D.
3.9 Boundary value problems

Unique solutions for boundary value problems for the Laplace equation require that
either the potential F or its normal derivative be specified on the boundary.[17]
Problems where F is specified on the boundary are known as Dirichlet boundary
value problems, while problems where ∂F=∂n are specified are known as Neumann
boundary value problems. Both of these types of problem give unique solutions for
the Laplace equation. Solutions do not exist when both F and ∂F=∂n are arbitrarily
specified because the derivatives of F have to be constrained to satisfy the Laplace
equation.
The solution of boundary value problems begins by separating the problem
space into regions with unique values for the current density and permeability.
For each region, a potential function is written in the most general form
possible. This introduces a set of unknown coefficients in the potential func-
tions. Constraints on these coefficients are determined by demanding that the
potential functions satisfy the boundary conditions at all the interfaces between
different regions.
Fourier analysis is particularly useful in problems involving rectangular
conductors in a space with rectangular boundaries.[18] If the boundaries
are infinitely-permeable iron surfaces, they can be replaced with a set of
image currents. Then the current density can be expressed as a Fourier series
and the fields can be determined from the solution to a boundary value
problem.
Example 3.2: rectangular conductor in an infinite slot

We first consider an example using the vector potential. Assume we have
a rectangular-shaped conductor near the bottom of an infinitely deep slot with
infinitely permeable walls, as shown in Figure 3.5. The current flows in the
z direction. The current density in the conductor is given by
Figure 3.5 Rectangular conductor in a slot.

60 Potential theory
Figure 3.6 Periodic current distribution.
8
<0 s=2 ≤ x ≤ w=2
JðxÞ ¼ J0 w=2 ≤ x ≤ w=2
:
0 w=2 ≤ x ≤ s=2:
We replace the parallel side walls of the slot with an infinite set of image conductors,
whose current density is shown in Figure 3.6. The Fourier series representing the
current distribution is

w 2J0 X
∞
1 nkw
JðxÞ ¼ J0 þ sin cosðnkxÞ; (3.53)
s π n¼1 n 2
where k ¼ 2π=s.
Divide the problem space vertically into three regions, as shown in Figure 3.5.
In region 1, there are no currents so the vector potential A1 satisfies the Laplace
equation. To satisfy the boundary conditions, we know that the x dependence has to
correspond with the x dependence of J(x). Thus we have
X
∞
A1 ¼ ðCn enky þ Dn enky Þ cosðnkxÞ;
n¼1
where Cn and Dn are unknown coefficients. The solution A3 for region 3 also has to
satisfy the Laplace equation. Since region 3 extends to infinite values of y, the term
proportional to enky must vanish. Far from the conductor, the field must be uniform
along the x direction, so the potential must contain a term proportional to y. It must
also contain a constant term to guarantee continuity of A. Thus the general form of the
potential in region 3 is
X
∞
A3 ¼ E0 þ E1 y þ Fn enky cosðnkxÞ:
n¼1
Since region 2 contains the conductor, the vector potential A2 has to satisfy the
Poisson equation. The total potential A2 has a general (or homogeneous) part plus
a particular solution to the Poisson equation.
A2 ¼ A2h þ A2p :
To match the potential at the boundary with region 3, the general part of the potential
has to include a constant term and a term linear in y. Thus we have
X
∞
A2h ¼ G0 þ G1 y þ ðHn enky þ Mn enky Þ cosðnkxÞ:
n¼1
Since the current density J has a constant term and a periodic term, we look for
a particular potential of the form
X
∞
A2p ¼ G2p y2 þ Ln cosðnkxÞ:
n¼1
Substitute this expression into the Poisson equation, together with Equation 3.53 for
the current density. Since the expansion makes use of orthogonal functions, we can
equate the constant term and each term in the series independently. We find the
coefficients for the particular solution are
μ0 J 0 w
G2p ¼
2s
and

2μ0 J0 nkw
Ln ¼ 3 2 sin :
πn k 2
The three equations for A contain a total of nine unknown coefficients. We find the
values of these coefficients by imposing the boundary conditions.
Case 1: y ¼ 0:
The bottom boundary is an infinite permeability surface, so B ¼ By must be
perpendicular to this surface. Thus
∂A1
Bx ¼ ¼ 0;
∂y
which gives
Cn Dn ¼ 0: (3.54)
Case 2: y = g
The nonperiodic and periodic parts of the equations for the continuity of A and ∂y A
give the four equations
μ0 J 0 w 2
0 ¼ G0 þ G1 g g (3.55)
2s
62 Potential theory

2μ0 J0 nkw
Cn enkg þ Dn enkg ¼ Hn enkg þ Mn enkg þ sin (3.56)
πn3 k2 2
μ0 J 0 w
0 ¼ G1 g (3.57)
s
Cn enkg Dn enkg ¼ Hn enkg Mn enkg : (3.58)
Case 3: y ¼ g þ h
The nonperiodic and periodic parts of the equations for the continuity of A and ∂y A
give the four equations
μ0 J 0 w
G0 þ G1 ðg þ hÞ ðg þ hÞ2 ¼ E0 þ E1 ðg þ hÞ (3.59)
2s

nkðgþhÞ 2μ0 J0 nkw
Hn e nkðgþhÞ
þ Mn e þ 3 2 sin ¼ Fn enkðgþhÞ (3.60)
πn k 2
μ0 J 0 w
G1 ðg þ hÞ ¼ E1 (3.61)
s
Hn enkðgþhÞ Mn enkðgþhÞ ¼ Fn enkðgþhÞ : (3.62)
Case 4: y→∞
At large y, far from the conductor, the field must be uniform, so we have
∂A3
Bx ¼ ¼ E1 :
∂y
From far above, the conductor looks like an infinite current sheet with current density
J0 h, whose strength is reduced by the filling factor w/s and is enhanced by a factor 2 due
to the presence of the bottom permeable surface. Then using Equation 1.17, we find
w
E1 ¼ Bx ¼ ½ μ0 J0 h 2
s (3.63)
μ0 J0 hw
¼ :
s
Equations 3.54–3.63 give ten constraints on the nine unknown coefficients.
However, Equation 3.61 is redundant since it is equivalent to Equations 3.57 and
3.63. Thus we have nine equations in nine unknowns. After solving this system of
equations, the resulting vector potentials are:
X
∞
A1 ¼ α ðenky þ enky ÞcosðnkxÞ; (3.64)
n¼1
where

μ0 J0 nkw nkg nkh
α ¼ 3 2 sin e ðe 1Þ; (3.65)
πn k 2
μ0 J 0 w X∞
A2 ¼ ðy gÞ2 þ ðβ1 enky þ β2 enky þ β3 ÞcosðnkxÞ; (3.66)
2s n¼1
where

μ0 J 0 nkw nkðgþhÞ
β1 ¼ 3 2 sin e (3.67)
πn k 2

β2 ¼ 3 2 sin e enkg þ enkg (3.68)
πn k 2

nkw
β3 ¼ 2μ0 J0 sin ; (3.69)
2
μ0 J0 hwð2g þ hÞ μ0 J0 hw X∞
A3 ¼ yþ γenky cosðnkxÞ; (3.70)
2s s n¼1
where

γ ¼ 3 2 sin e þ enkðgþhÞ þ enkg enkg : (3.71)
πn k 2
The series for the vector potential converge rapidly because of the n3 factor in the
denominators of the coefficients.
We find B by taking the curl of the vector potential. The resulting field in the slot is
shown in Figure 3.7. The dotted lines show the location of the conductor.
Example 3.3: permeable sphere in external magnetic field

We next consider an example of using the scalar potential. Assume we have a sphere
of some magnetic material located in an external magnetic field, as shown in
Figure 3.8. We choose the z coordinate of a spherical coordinate system to lie along
the direction of the external field B0 ¼ μ0 H0 . The problem has azimuthal symmetry,
so the results cannot depend on ϕ. The magnetic scalar potential for the external field
can be written as
V0 ¼ H0 z ¼ H0 r cos θ

¼ H0 r P1 ðcos θÞ;
64 Potential theory
Figure 3.7 Magnetic flux density inside the slot.
Figure 3.8 Permeable sphere in an external magnetic field.
where P1 is a Legendre polynomial. The potential for the problem including the sphere
must remain finite as r → ∞. Therefore the potential outside the sphere is given by
X
∞
Vmext ¼ V0 þ cn rn1 Pn ðcos θÞ:
n¼0
The potential must also remain finite at r = 0, so the potential inside the sphere has
the form
X
∞
Vmint ¼ V0 þ dn rn Pn ðcos θÞ:
n¼0
At the boundary surface r = a, Ht ¼ Hθ must be continuous, so
X
∞
dPn X∞
dPn
cn an2 ¼ dn an1 :
n¼0
dθ n¼0
dθ
Since this must true for any θ, the coefficients must satisfy the relation
cn ¼ dn a2nþ1 : (3.72)
At the boundary surface r = a, Bn ¼ Br must also be continuous. After simplifying

we have
" # " #
X
∞ X
∞
μ0 H 0 P 1 þ cn ðn þ 1Þan2 Pn ¼ μ H0 P1 dn nan1 Pn :
n¼0 n¼0
We require that this relation hold for any value of n. For n = 0, we obtain
μ0 c0 a2 ¼ 0:
This shows that c0 ¼ 0 and from Equation 3.72 we find that d0 ¼ 0. For n = 1, we
find that
μ0 ðH0 þ 2c1 a3 Þ ¼ μðH0 d1 Þ: (3.73)
Substituting Equation 3.72 for c1, we obtain

μ μ0
d1 ¼ H0 :
μ þ 2 μ0
Substituting this back into Equation 3.72, we find

μ μ0 3
c1 ¼ a H0 :
μ þ 2 μ0
For n > 1, we have
μ0 cn ðn þ 1Þan2 ¼ μndn an1 : (3.74)
Using Equation 3.72 for cn, we obtain dn ¼ 0. Using this in Equation 3.72, we find
cn ¼ 0. Thus the only nonvanishing coefficients are c1 and d1 . The solution for the
potential outside the sphere is
μ μ0 a3
Vmext ¼ H0 r cos θ þ H0 cos θ
μ þ 2 μ0 r 2
66 Potential theory
and the field components are

μr 1 a 3
Br ¼ B0 cos θ 1 þ 2
μr þ 2 r2

μr 1 a 3
Bθ ¼ B0 sin θ 1 Þ :
μr þ 2 r2
The potential inside the sphere is
μ μ0
Vmint ¼ H0 r cos θ þ H0 cos θ
μ þ 2 μ0
and the field components are

μr 1
Br ¼ μH0 cos θ 1
μr þ 2

μr 1
Bθ ¼ μH0 sin θ 1 Þ :
μr þ 2
The magnetic flux density in the vicinity of a sphere with μr ¼ 20 is shown in

Figure 3.9. The lines of B are pulled into the sphere and approach the boundary
approximately along a normal. The field inside the sphere is parallel to the external
field.
Figure 3.9 Cross-section of the permeable sphere in an external magnetic

field.
3.10 Green’s theorem 67
3.10 Green’s theorem

Let V be a region of space enclosed by the surface S. Let ψ and ϕ be scalar functions
of position that have continuous first and second derivatives. Applying the diver-
gence theorem to the vector ψrϕ, we get
ð ð
r·ðψrϕÞdV ¼ ðψrϕÞ·^n dS: (3.75)
Since
r·ðψrϕÞ ¼ rψ·rϕ þ ψr2 ϕ
and
∂ϕ
rϕ·^
n¼ ;
∂n
Equation 3.75 becomes
ð ð ð
∂ϕ
rψ·rϕ dV þ ψr ϕ dV ¼ ψ dS:
2
(3.76)
∂n
If we repeat this calculation, interchanging the role of ϕ and ψ, we obtain

ð ð ð
∂ψ
rϕ·rψ dV þ ϕr2 ψ dV ¼ ϕ dS: (3.77)
∂n
Subtracting Equation 3.77 from Equation 3.76, we obtain Green’s second identity
or Green’s theorem [19]
ð ð
∂ϕ ∂ψ
ðψr ϕ ϕr ψÞdV ¼
2 2
ψ ϕ dS: (3.78)
∂n ∂n
For two-dimensional problems, there is a corresponding Green’s theorem on the

plane given by [20]
ðð þ
∂Q ∂P
dx dy ¼ ðP dx þ Q dyÞ; (3.79)
∂x ∂y
where P and Q are continuous functions of x and y and have continuous partial
derivatives.
To apply Green’s theorem to magnetostatics, let us choose the function ψ to be
proportional to the inverse distance between an element of current at r0 and a field
observation point at r.
68 Potential theory
1 1
ψ¼ ¼ 0 : (3.80)
4πR 4πj!
r !r j
In the case where currents are present inside V, we choose ϕ to be one of the
components of the vector potential, e.g., Ax . Then we have
r2 Ax ¼ μ0 Jx
and from Equation 1.23

1
¼ δð!
r !
0
r 2
r Þ:
4πR
Substituting this into Green’s theorem, we get

ð ð
1 ! ! 0 0 1 ∂Ax ∂ 1
ðμ0 Jx Þ Ax δð r r Þ dV ¼ Ax dS0 :
4πR 4πR ∂n ∂n 4πR
Because of the delta function, we can solve this equation for Ax ðrÞ:
ð ð
! μ0 J x 0 1 1 ∂Ax ∂ 1
Ax ð r Þ ¼ dV þ Ax dS0 : (3.81)
4π R 4π R ∂n ∂n R
The volume integral represents a particular solution of the Poisson equation and only
includes the effects of currents inside V. Any additional currents outside V influence
the value of the surface integral. If we let r be a set of points on the surface S, then this
represents an integral equation for the unknown vector potential.
We can use Green’s theorem to develop integral equation solutions for Dirichlet
and Neumann boundary value problems. We generalize Equation 3.80 used in the
previous derivation by defining the Green’s function [21]
1
Gð!
r ;! ! !0
0
r Þ¼ ! ! 0 þ Lð r ; r Þ;
4πj r r j
where L is an arbitrary solution of the Laplace equation inside V. Substituting this

into Green’s theorem, we obtain an equation similar to Equation 3.81 with
1/R replaced with G. In the case where there are no currents inside V, we have
ð
! 1 ! ! 0 ∂ϕ ∂ ! ! 0
ϕð r Þ ¼ Gð r ; r Þ ϕ Gð r ; r Þ dS0 ; (3.82)
4π ∂n ∂n
where ϕ represents either the magnetic scalar potential or one of the components
of A.
References 69
For Dirichlet boundary value problems, if we choose L such that the Green’s
function
GD ¼ 0 (3.83)
when r0 is on the surface, we obtain the integral equation

ð
! 1 0 ∂
ϕð!r Þ GD ð! r ;!
0
ϕð r Þ ¼ r ÞdS0 : (3.84)
4π ∂n
For Neumann boundary value problems for the exterior of the volume V, if we
instead choose L such that
∂GN
¼0 (3.85)
∂n
when r0 is on the surface, we obtain the integral equation
ð !0
! 1 ! ! 0 ∂ϕð r Þ
ϕð r Þ ¼ GN ð r ; r Þ dS0 : (3.86)
4π ∂n
Thus once we have an appropriate Green’s function for a given geometry, the
potential at some field point r can be found by integration over the boundary
surface. The Green’s function is the solution of the Poisson equation for a delta
function source.
References
[1] D. Iencinella & G. Matteucci, An introduction to the vector potential, Eur. J. Phys.
25:249–256, 2004.
[2] J. D. Jackson, Classical Electrodynamics, Wiley, 1962, p. 141.
Wesley, 1962, p. 147–151.
p. 305.
[5] J. D. Jackson, op. cit., p. 145–146.
[6] L. Eyges, The Classical Electromagetic Field, Dover, 1980, p. 166–167.
[7] P. Lorrain & D. Corson, op. cit., p. 92–94.
[8] L. Eyges, op. cit., p. 138–139.
[9] J. D. Jackson, op. cit., p. 47.
[10] J. B. Marion, Classical Electromagnetic Radiation, Academic Press, 1965, p. 58–60.
[11] W. Panofsky & M. Phillips, op. cit., p. 88–89.
[12] W. Smythe, Static and Dynamic Electricity, 2nd ed., McGraw-Hill, 1950, p. 170–182.
[13] Ibid., p. 187–198.
[15] G. Arfken, Mathematical Methods for Physicists, 3rd ed., Academic Press, 1985,
p. 680–683.
70 Potential theory
[16] W. Smythe, op. cit., p. 147–153.

[17] J. D. Jackson, op. cit., p. 15–17.
Problems, 2nd ed., Pergamon Press, 1973, p. 95–103.
[19] J. Stratton, Electromagnetic Theory, McGraw-Hill, 1941, p. 165.
[20] W. Kaplan, Advanced Calculus, Addison-Wesley, 1952, p. 239–241.
[21] L. Eyges, op. cit., p. 85–91.
4
Conductor-dominant transverse fields
In this chapter, we consider transverse field configurations that can be approxi-

mated as uniform along the z axis. We begin by considering the general form
of the solutions to the magnetostatic equations in two dimensions. We next treat
the fields produced by line currents, current sheets, and current blocks. We
find that multipole errors are introduced when we approximate ideal current
distributions with practical conductor configurations. The shapes of the fields
discussed here are primarily determined by the location of the conductors.
Any iron that may be present only acts to enhance the strength of the field in
the magnet aperture. A high-field accelerator magnet is one example of
a magnet that produces fields of this type. We conclude with a brief discussion
of superconductors and 3D conductor configurations used at the end of these
magnets.
4.1 General solution to the Laplace equation in two dimensions

The two-dimensional Laplace’s equation in the polar coordinates r and θ is
1 1
∂r ðr ∂r VÞ þ 2 ∂2θ V ¼ 0: (4.1)
r r
This is identical with the first two terms in Equation 3.39, so the solutions for the
radial and azimuthal dependence of the potential follow from Equations 3.43, 3.44,
3.46, and 3.47. Thus the general solution for Laplace’s equation in polar coordi-
nates has the form
X
∞
Vðr; θÞ ¼ ðCn rn þ Dn rn Þ ðEn cos nθ þ Fn sin nθÞ þ ðC0 lnr þ D0 Þ ðE0 þ F0 Þ:
n¼1
(4.2)
71
72 Conductor-dominant transverse fields
Example 4.1: vector potential inside a magnet aperture

The aperture of a magnet is the open area enclosed by the coils. We consider here the
general form for the magnetic field in a magnet aperture. We use a polar coordinate
system with the origin inside the aperture and let the variable V in Equation 4.2 refer
to the z component of the vector potential. In addition, we assume that there are no
external fields present outside the magnet and no current filaments inside the aperture.
In this case, we can ignore the n = 0 terms in Equation 4.2. Since the potential must be
finite at r = 0, we must have all the coefficients Dn ¼ 0. Thus the vector potential
inside the aperture has the form
X
∞
Az ðr; θÞ ¼ Cn rn ðEn cos nθ þ Fn sin nθÞ: (4.3)
n¼1
The magnetic field is given by
! ! 1
B ðr; θÞ ¼ r A ¼ ^r ∂θ Az ^θ ∂r Az : (4.4)
r
Evaluating the radial field component, we find

X
∞
Br ðr; θÞ ¼ Cn rn1 ½En nsin nθ þ Fn ncos nθ: (4.5)
n¼1
Defining the new coefficients
An ¼ n Cn Fn
(4.6)
Bn ¼ n Cn En ;
the radial field inside the magnet aperture can be written as

X
∞
Br ðr; θÞ ¼ rn1 ðAn cos nθ þ Bn sin nθÞ: (4.7)
n¼1
The coefficients An and Bn describe the multipole field content of the transverse field.1
Returning to Equation 4.4, the azimuthal field component is
X
∞
Bθ ðr; θÞ ¼ Cn n rn1 ½En cos nθ þ Fn sin nθ:
n¼1
1
Caveat emptor. The reader should be aware that a number of different definitions are used in the literature to
describe the multipole content of a transverse field.
4.1 General solution to the Laplace equation in two dimensions 73
Using Equation 4.6, the azimuthal field in the magnet aperture is
X
∞
Bθ ðr; θÞ ¼ rn1 ðAn sin nθ þ Bn cos nθÞ: (4.8)
n¼1
On the midplane of the magnet (θ = 0, r = x), Bx and By are given by power series in x
Bx ðxÞ ¼ Br ðx; 0Þ ¼ A1 þ A2 x þ A3 x2 þ

(4.9)
By ðxÞ ¼ Bθ ðx; 0Þ ¼ B1 þ B2 x þ B3 x2 þ :
Example 4.2: scalar potential inside a magnet aperture

We again begin by considering the general form of the solution of Laplace’s equation
in polar coordinates given in Equation 4.2. We then specialize to the case for a region
containing r = 0 and obtain an equation analogous to Equation 4.3.
X
∞
μ0 Vm ðr; θÞ ¼ Gn rn ðHn cos nθ þ In sin nθÞ: (4.10)
n¼1
The magnetic field is given by
! ∂Vm ^ μ0 ∂Vm
B ðr; θÞ ¼ μ0 rVm ¼ ^r μ0 θ :
∂r r ∂θ
Thus the radial field is

X
∞
Br ¼ n Gn rn1 ðHn cos nθ þ In sin nθÞ: (4.11)
n¼1
The field components calculated from this potential must equal the same quantities
calculated from the vector potential. We can make Br have the same form as
Equation 4.5 if we demand that
n Gn Hn ¼ n Cn Fn ¼ An
n Gn In ¼ n Cn En ¼ Bn :
Thus we identify
G n ¼ Cn
Hn ¼ Fn
I n ¼ En :
The scalar potential inside the aperture is then given by2

X
∞
μ0 Vm ðr; θÞ ¼ Cn rn ðEn sin nθ Fn cos nθÞ: (4.12)
n¼1
The resulting magnetic field components are still given by Equations 4.7 and 4.8.
If the field in some region is known, for example through calculations or

measurements, then the multipole field coefficients can be determined using
Fourier analysis. Multiplying both sides of Equation 4.8 by cos mθ and integrating
around a circular path, we have
ð 2π X
∞
Bθ ðr; θÞ cos mθ dθ ¼ rn1 ½An I1 þ Bn I2 ;
0 n¼1
where for m ≥ 1 and n ≥ 1 the integrals have the values3

ð 2π
I1 ¼ sin nθ cos mθ dθ ¼ 0 (4.13)
0
and
ð 2π
I2 ¼ cos nθ cos mθ dθ ¼ π δmn
0
ð 2π (4.14)
¼ sin nθ sin mθ dθ:
0
Thus we find one set of multipole field components is given by

ð 2π
1
Bn ¼ n1 Bθ ðr; θÞ cos nθ dθ: (4.15)
πr 0
Likewise, we can multiply Equation 4.8 with sin mθ and find the other set of
multipole components is
ð 2π
1
An ¼ n1 Bθ ðr; θÞ sin nθ dθ: (4.16)
πr 0
2
We will see in the next chapter that this relationship between the vector and scalar potentials follows directly
from the Cauchy-Riemann equations.
3
CRC 497, 502.
4.1 General solution to the Laplace equation in two dimensions 75
A similar analysis using Equation 4.7 for the radial component of the field gives
ð 2π
1
Bn ¼ n1 Br ðr; θÞ sin nθ dθ (4.17)
πr 0
and
ð 2π
1
An ¼ n1 Br ðr; θÞ cos nθ dθ: (4.18)
πr 0
The strength of the multipole fields provides a measure of the field quality.
Limits on the field uniformity are imposed by the application that needs the
magnetic field. The presence of harmonics of the desired field limits the size of
the useful magnet aperture. Sometimes, when examining the field quality of
a magnet, it is more useful to examine the relative magnitude of the multipole
coefficients with respect to the coefficient for the desired multipole. Thus for
a dipole design, for example, one could calculate the dimensionless quantities
Bn r0n1
bn ¼ ;
B1
where r0 is a reference radius, typically ~2/3 of the magnet aperture.
The boundary conditions for the vector potential in polar coordinates at some
radius rb can be determined from the boundary conditions on the magnetic field.
From the condition on the normal component of B, we have
ð1Þ ð2Þ
Br ¼ Br
1 ∂Að1Þ 1 ∂Að2Þ
z
¼ z
:
rb ∂θ rb ∂θ
From this relation, we know that Að2Þ can differ from Að1Þ by at most a constant,
which we can ignore since constants are removed when we take derivatives to
obtain the field. Thus we have
Að2Þ ð1Þ
z ðrb ; θÞ ¼ Az ðrb ; θÞ: (4.19)
From the boundary condition on Ht ,

ð2Þ ð1Þ
Ht Ht ¼ K;
we have
1 ∂Að2Þ
z ðrb ; θÞ 1 ∂Að1Þ
z ðrb ; θÞ
þ ¼ K: (4.20)
μ ð2Þ ∂r μ ð1Þ ∂r
Figure 4.1 Geometry of a line current.
4.2 Harmonic expansion for a line current

Consider a line current perpendicular to the x-y plane, as shown in Figure 4.1.
The vector potential for a line current was given in Equation 3.8

μI R
Az ðr; θÞ ¼ ln ; (4.21)
2π r0
where the distance from the line current at (a, ϕ) to the observation point P located
at (r, θ) is given by
n o1=2
R ¼ ðr cos θ a cos ϕÞ2 þ ðr sin θ a sin ϕÞ2 (4.22)
and r0 is some constant reference radius for the two-dimensional potential. We look
for a harmonic expansion for Az. When r > a we extract a factor of r2 from the
logarithm and find that

a2 a
lnðRÞ ¼ ½ lnðr Þ þ ½ ln 1 þ 2 2 cos ðθ ϕÞ :
2
r r
Writing the cosine in terms of complex exponentials gives

a2 a iðθϕÞ a iðθϕÞ
lnðRÞ ¼ lnðrÞ þ ½ ln 1 þ 2 e e
r r r
h a a i
¼ lnðrÞ þ ½ ln 1 eiðθϕÞ þ ln 1 eiðθϕÞ :
r r
Using the series expansion for lnð1 þ xÞ for x < 1, we find

a 1 a2
lnðRÞ ¼ lnðrÞ cos ðθ ϕÞ cos ½2ðθ ϕÞ þ
r 2 r
X ∞
1 an
¼ lnðrÞ cos ½nðθ ϕÞ:
n¼1
n r
Thus the vector potential for the line current when r > a is [1]
μI μI X∞
1 an
Az ðr; θÞ ¼ lnðrÞ þ cos ½nðθ ϕÞ (4.23)
2π 2π n¼1 n r
plus a constant term involving r0. The expansion of lnR for the case r < a can be
done in a similar manner by first extracting a factor of a2 from the argument of the
logarithm. This results in the vector potential
μI μI X∞
1 r n
Az ðr; θÞ ¼ lnðaÞ þ cos ½nðθ ϕÞ: (4.24)
2π 2π n¼1 n a
Example 4.3: line current in circular iron cavity

Consider a line current at radius a inside the circular aperture of a piece of iron with radius
R, as shown in Figure 4.2. We know from Equations 4.23 and 4.24 that the contribution to
the total vector potential from the line current has different expansions depending on
whether r is greater than or lesser than the radius a of the line current. Similarly, the field
induced in the iron has different expressions depending on whether r is greater than or
smaller than the radius R of the opening in the iron.[2, 3] The induced vector potential
must have the general form given in Equation 4.2. To match the field from the line current
at the boundaries, the induced field must have the same angular dependence as the line
current. The induced field must also be finite at r = 0 and at r → ∞. Let k ¼ I=2π and
ω ¼ θ ϕ. Then the total vector potential in the region (1) with r < a is
Figure 4.2 Quarter section of a line current in a circular iron cavity.

X∞
1 r n X∞
Az1 ¼ μ0 k lnðaÞ þ μ0 k cos nω þ μ0 Cn rn cos nω:
n¼1
n a n¼1
In the region (2) with a < r < R between the line current and the iron boundary, the
vector potential is
X∞
1 an X∞
Az2 ¼ μ0 k lnðrÞ þ μ0 k cos nω þ μ0 Cn rn cos nω:
n¼1
n r n¼1
Lastly, in the region (3) with r > R inside the iron, the vector potential is
X∞
1 an X∞
Az3 ¼ μ k lnðrÞ þ μ k cos nω þ μ Dn rn cos nω;
n¼1
n r n¼1
where μ ¼ μr μ0 is the assumed constant permeability of the iron. We determine the

unknown coefficients Cn and Dn by demanding continuity of
1
Br ¼ ∂ θ Az
r
1
Hθ ¼ ∂ r Az
μ
at the surface r = R of the iron.
μ0 k n n1 μ k n n1
aR þ μ0 Cn Rn1 ¼ a R þ μ Dn Rn1
n n
Cn Rn1 ¼ Dn Rn1 :
This gives two equations in two unknowns, which can be solved to give,
k a n μ μ0
Cn ¼
n R2n μ þ μ0
k n μ μ0
Dn ¼ a :
n μ þ μ0
In the region inside the radius a of the line current, the vector potential can be
expressed as
a 2n
μ0 I μ0 I X
∞
1 r n
Az1 ¼ lnðaÞ þ 1þα cos nω; (4.25)
2π 2π n¼1 n a R
where
μ μ0
α¼ :
μ þ μ0
The term in brackets shows the enhancement factor due to the iron. For points inside
the iron, the solution is
" #
μI X∞
1 an
Az3 ¼ lnðrÞ þ ð1 αÞ cos nω : (4.26)
2π n¼1
n r
We are now in a position to relate the solution of the previous example with the
method of images for a circular boundary that we discussed in Section 2.7. For the
region interior to the filament, r < a, the vector potential can be written in the form
μ0 I μ IX ∞
1 r n μ IX ∞
1 nα
Az1 ¼ lnðaÞ þ 0 cos nω þ 0 r n cos nω; (4.27)
2π 2π n¼1 n a 2π n¼1 n r1
where
R2
r1 ¼ :
a
Comparing with Equation 4.24, the first two terms give the potential for the true
line current at r = a. Since in magnetostatics constant terms in the potential have no
physical effects, we can arbitrarily add to the potential a constant term
μ0 I
lnðr1 Þ:
2π
Then this term plus the last term in Equation 4.27 give the potential for a line current
at r1 with current α I. Thus the vector potential in region (1) can be written as
Az1 ¼ ALC ðaÞ þ α ALC ðr1 Þ;
where ALC is the vector potential for a line current. For the region inside the iron,
r > R, let us define
2μ0
β¼1α¼ :
μ þ μ0
Then Equation 4.26 can be rewritten in the form

" #
μI X∞
1 an
Az3 ¼ ln r þ β ln r β ln r þ β cos nω :
2π n¼1
n r
Using Equation 4.23 for the vector potential of a line current, we find
" !#
μI X∞
1 an
Az3 ¼ ð1 βÞ ln r þ β ln r þ cos nω
2π n¼1
n r
¼ α ALC ð0Þ þ β ALC ðaÞ:
We see that the coefficients α and β are the same as those for the image currents in
Equations 2.27–2.30.
4.3 Field for a current sheet

Consider a conductor in the form of an infinitely thin sheet that is uniform in the
z direction, as shown in Figure 4.3. Assume here that the current also flows in the
z direction. From the Biot-Savart law for a current sheet, Equation 1.13, we have
ð! !
! μ0 K R
B ¼ dS
4π R3
Assume the observation point P is in the x-y plane, as shown in Figure 4.4.
We have
! ¼ dI ^z
K
ds
ρ ¼ r2 þ a2 2 r a cos ðθ ϕÞ
2
R2 ¼ ρ2 þ z2
dS ¼ ds dz;
Figure 4.3 Current sheet with cross-section along the curve C and extending
infinitely along the z direction.
Figure 4.4 Sheet geometry.
where a ¼ aðsÞ, ϕ ¼ ϕðsÞ and s is the arclength around the sheet. The vector
ρ is the distance between the current element and the field point in the x-y plane.
Thus
ð ð∞
! ðr; θÞ ¼ μ0 KðsÞ ^z ð!ρ þ z ^z Þ
B 3=2
dz ds
4π ∞ fρ2 þ z2 g
ð
μ0
¼ KðsÞ IðρÞ ^z !ρ ds;
4π
where4
ð∞
dz 2
IðρÞ ¼ ¼ : (4.28)
∞ fρ2 þ z2 g3=2 ρ2
Thus we find that the field from the current sheet is given by
ð
! μ0 ^z ! ρ
B ðr; θÞ ¼ KðsÞ ds: (4.29)
2π ρ 2
It will also be useful to have an expression for the vector potential of a current
sheet. Assuming the sheet is composed of parallel line currents and using
ð
μ0 ρ
Az ðr; θÞ ¼ KðsÞln ds: (4.30)
2π ρ0
4
GR 2.271.5.
For the case of a circular sheet with constant radius a,
ds ¼ a dϕ
dI dϕ 1 dI
K¼ ¼ :
dϕ ds a dϕ
The field due to the circular sheet is

ð ϕ2
! μ dI ^z !ρ
B ðr; θÞ ¼ 0 dϕ (4.31)
2π ϕ1 dϕ ρ2
and the vector potential is

ð ϕ2
μ0 dI ρ
Az ðr; θÞ ¼ ln dϕ: (4.32)
2π ϕ1 dϕ ρ0
Example 4.4: field between two parallel, straight current sheets

Consider the two parallel current sheets shown in Figure 4.5. The current, which is
uniform along y, flows into the page on the sheet on the right and returns back out
of the page on the sheet on the left. The field observation point P is at ðxo ; yo Þ.
We have
!
ρ ¼ ðxo xÞ ^x þ ðyo yÞ ^y
dI
K¼ :
dy
Figure 4.5 Parallel current sheets.

Applying Equation 4.29, the field can be written as

ð∞ " #
! μ ðxo aÞ^y ðyo yÞ^x ðxo þ aÞ^y ðyo yÞ^x
B ðxo ; yo Þ ¼ 0 K þ dy:
2π ∞ ðxo aÞ2 þ ðyo yÞ2 ðxo þ aÞ2 þ ðyo yÞ2
All the current elements on both sheets give positive By because the magnitude of xo
is smaller than a. Thus we can write
μ0
By ¼ K ½ða xo Þ I1 þ ða þ xo Þ I2 ;
2π
where5
ð∞
dy π
I1 ¼ ¼
∞
2
ðxo aÞ þ ðyo yÞ 2 a xo
ð∞
dy π
I2 ¼ ¼ :
∞
2
ðxo þ aÞ þ ðyo yÞ 2 a þ xo
Thus the vertical field between the sheets is

dI
By ¼ μ 0 : (4.33)
dy
This is, as expected, twice the field we found for a single sheet in Equation 1.17 and
independent of the location of the observation point. In a similar manner, we find
μ0
Bx ¼ K½yo I1 I3 yo I2 þ I4 ;
2π
where
ð∞
y
I3 ¼ dy
∞ ðxo aÞ þ ðyo yÞ2
2
and I4 is a similar integral with xo þ a in the denominator. If we let D represent the

denominator, then6
I3 ¼ ½ ln½D∞∞ þ yo I1 :
Using l’Hopital’s rule, we can show that the first term vanishes, and find that
I3 ¼ yo I1
I4 ¼ yo I2 :
5 6
GR 2.172. GR 2.175.1.
Substituting back in, we find that the horizontal field between the sheets
vanishes.
Bx ¼ 0:
Thus there is a pure dipole field in the region between the current sheets.
4.4 Ideal multipole current sheet

Assume we have a circular current sheet with radius a and with azimuthal current
density in the z direction given by
dI
¼ I0 cos mϕ;
dϕ
where I0 is the current flowing at the midplane (ϕ = 0). We obtain the vector
potential for the sheet by integrating the weighted distribution of the vector
potential for a line current. Let us first consider the case where the observation
point (r, θ) has r < a. Using Equation 4.24 for the line current and ignoring the
constant term, the vector potential for the multipole sheet is
ð
μI0 X
∞
1 r n 2π
Az ¼ cos mϕ cos ½nðθ ϕÞ dϕ:
2π n¼1 n a 0
Expanding the integrand and using Equations 4.13 and 4.14, the integral has the
value
ð 2π
0 if m ≠ n
cos mϕ cos ½nðθ ϕÞ dϕ ¼ : (4.34)
0 π cos mθ if m ¼ n
Therefore the vector potential for r < a is [4]
μI0 r m
Az ðr; θÞ ¼ cos mθ (4.35)
2m a
and the components of the magnetic field are
μI0 r m1
Br ðr; θÞ ¼ sin mθ
2a a
(4.36)
μI0 r m1
Br ðr; θÞ ¼ cos mθ:
2a a
In the case of a dipole distribution (m = 1),

μI0
By ¼ Br sin θ þ Bθ cos θ ¼
2a
Bx ¼ Br cos θ Bθ sin θ ¼ 0
showing that the cos θ angular distribution also produces a pure vertical field in the
magnet aperture.
For the case when r > a, we use Equation 4.23 for the vector potential of the line
current and obtain
ð ð
μI0 2π μI0 X∞
1 an 2π
Az ¼ cos mϕ ln r dϕþ cos mϕ cos ½nðθ ϕÞ dϕ:
2π 0 2π n¼1 n r 0
The first integral vanishes over a complete circle and the second integral can again
be evaluated using Equation 4.34. Thus the vector potential of the multipole sheet
for the region r > a is
μI0 am
Az ðr; θÞ ¼ cos mθ (4.37)
2m r
and the magnetic field is
μI0 am
Br ðr; θÞ ¼ sin mθ
2 rmþ1
(4.38)
μI0 am
Br ðr; θÞ ¼ cos mθ:
2 rmþ1
Example 4.5: cos mθ sheet in circular iron cavity

Consider a circular cos mθ sheet with radius a inside a symmetric circular iron cavity
with radius R. We first solve the boundary value problem using vector potentials. There
are two vector potentials for the sheet, depending on whether r is smaller than or greater
than a, and two vector potentials for the image effects in the iron, depending on whether
r is smaller than or greater than R. The angular dependence of the image effects must also
use the cosine term in Equation 4.2 and only the mth term in the summation in order to
match the boundary conditions. This introduces two unknown coefficients, C and D, and
requires that the total vector potential for the three regions be given as
μ0 I0 r m
Az1 ¼ cos mθ þ C rm cos mθ
2m a
μ I0 am
Az2 ¼ 0 cos mθ þ C rm cos mθ
2m r
μI0 am
Az3 ¼ cos mθ þ D rm cos mθ:
2m r
To determine the values for C and D, we demand that Br and Hθ are continuous across
the iron boundary.
μ0 I0 a m μI0 a m
þ C m Rm ¼ þ D m Rm
2 R 2 R
Rm Rm
C ¼ D :
μ0 μ
Solving these two equations, we find that the unknown coefficients are
μ0 I0 a m μ μ0
C¼
2 m Rm R μ þ μ 0
μI0 a m μ μ0
D¼ :
2 m Rm R μ þ μ0
Using these values, the vector potential is now known in the three regions. The effect
of the iron can be summarized by defining the iron enhancement factor
μr 1 a 2m
αm ¼ 1 þ ; (4.39)
μr þ 1 R
which agrees with the enhancement factor from Equation 4.25. We can write the
vector potential inside the magnet aperture as
μ0 I0 r m
Az1 ðr; θÞ ¼ αm cos mθ: (4.40)
2m a
The corresponding field components inside the aperture are
μ0 I0 r m1
Br ðr; θÞ ¼ αm sin mθ
2a a
(4.41)
μ I0 r m1
Br ðr; θÞ ¼ 0 αm cos mθ:
2a a
The iron enhancement factor in Equation 4.39 ignores any saturation effects
in the iron. When saturation becomes significant, the enhancement factor for the
dipole field is decreased. In addition, the saturation of the iron does not occur
uniformly. This causes changes in the field from the azimuthal distribution of
the enhanced currents, leading to sextupole and higher multipole errors in the
field in the magnet aperture. Fortunately, there are techniques, such as modify-
ing the iron shape, which can be used to adjust the value of B3 at a fixed
operating current.[5]
Example 4.6: cos mθ sheet in circular iron cavity using the scalar potential
It is instructive to use an alternative method of solving the preceding boundary
value problem. In this case, we will use the scalar potential and take into account
the presence of the current sheet through the addition of another pair of boundary
conditions. Unlike the previous example, the unknown coefficients here take into
account both the field from the sheet and the field from the images in the iron.
We know from the boundary condition across a current sheet, Equation 2.23, that
the angular dependence of the current must match the angular dependence of
the fields on either side of the sheet. Since the fields are given by the derivative
of the potential and the current goes like cos mθ, this implies that we must use
the sine term in Equation 4.2 for the potential. Thus the scalar potentials for the three
regions are
Vm1 ¼ A rm sin mθ
Vm2 ¼ ðB rm þ C rm Þ sin mθ
Vm3 ¼ D rm sin mθ:
The boundary conditions at the sheet are
A am ¼ B am C am
mðB am þ C am Þ þ m A am ¼ I0 ;
while the boundary conditions at the iron surface are
μ0 ðB Rm C Rm Þ ¼ μ D Rm

B Rm þ C Rm ¼ D Rm :
Solving the four equations for the four unknowns, we find
I0
A¼ αm
2mam
I0 am μ μ0
B¼
2mR2m μ þ μ0
I 0 am
C¼
2m
μ0 I0 am
D¼ :
mðμ þ μ0 Þ
Evaluating the field components inside the magnet aperture, we again obtain
Equation 4.41.
4.5 Multipole dependence on the current distribution

We next seek to determine how the multipole fields in a magnet aperture are related
to the current distribution on a circular current sheet.
Recall that the vector potential for a line current at (a, ϕ) is
μI X∞
1 r m
Az ðr; θÞ ¼ cos ½mðθ ϕÞ:
2π m¼1 m a
The corresponding azimuthal field component is
μI X∞ m1
r
Bθ ðr; θÞ ¼ cos ½mðθ ϕÞ : (4.42)
2π m¼1 am
From Equation 4.15, the contribution to the normal multipole is

ð 2π
1
Bn ¼ n1 Bθ ðr; θÞ cos nθ dθ
πr 0
ð
μI X
∞
1 2π
¼ 2 cos ½mðθ ϕÞ cos nθ dθ :
2π m¼1 am 0
Expanding the cosine and integrating, we find
μI X∞
1
Bn ¼ π cos mϕ δmn
2π m¼1 am
2
(4.43)
μI
¼ cos nϕ :
2πan
For a current sheet, we can generalize this by integrating over the current
distribution.
ð
μ 2π dI
Bn ¼ cos nϕ dϕ: (4.44)
2πan 0 dϕ
We can find the skew multipoles in a similar manner using Equation 4.16.
ð
μ 2π dI
An ¼ sin nϕ dϕ: (4.45)
2πan 0 dϕ
As the name suggests, a multipole field is characterized by the number of poles it

has around the circumference of the sheet. Every magnet has an equal number of
positive and negative poles. We use the index N to refer to the design number of pole
4.5 Multipole dependence on the current distribution 89
Figure 4.6 Multipole symmetries for dipoles (m = 1) and quadrupoles (m = 2).

N and S indicate the north and south poles of the magnet. Plus and minus signs
indicate the direction of the current.
pairs in the multipole magnet. Thus N = 1 refers to a dipole magnet because it has one
pair of poles, N = 2 to a quadrupole, etc. The current in the magnet goes in opposite
directions on the adjacent sides of a pole, as shown in Figure 4.6. This corresponds to
the way these coils are usually wound, with the cable bending around the pole and
returning in the opposite direction. As we can see from Equations 4.44 and 4.45, the
multipole coefficients are weighted sums of the current distribution. The Bn coeffi-
cients are referred to as the normal multipoles. They are largest when the magnitude
of the current reaches a maximum on the midplane. The An coefficients are referred
to as the skew multipoles, which are largest when the current changes sign at the
midplane. The field in a skew multipole of order N has the same pattern as the normal
multipole rotated by π/2N. For example, a normal dipole has a vertical field, while
a skew dipole has a horizontal field.
Consider the current distribution for an ideal multipole of order N given by
dI
¼ I0 cos Nϕ:
dϕ
The normal multipole coefficient for a complete angular distribution is

ð
μ 0 I 0 ϕ2
Bn ¼ cos Nϕ cos nϕ dϕ:
2π an ϕ1
From Equation 4.14, this vanishes unless n = N, in which case

μ0 I0
BN ¼ :
2aN
Similarly, Equation 4.13 shows that all the An coefficients vanish for this current
distribution. The multipole coefficients are constructed such that they uniquely
identify the symmetry of the current distribution. Likewise, an ideal current dis-
tribution of the form
dI
¼ I0 sin Nϕ
dϕ
is uniquely associated with the skew multipole AN .
Example 4.7: quadrupole field

The coefficient for a quadrupole field corresponds to N = 2. The current density is
given by
dI
¼ I0 cos 2ϕ:
dϕ
Thus
ð ϕ2
μ I0
B2 ¼ 0 2 cos2 ð2ϕÞ dϕ
2π a ϕ1
μ0 I 0
¼ :
2 a2
4.6 Approximate multipole configurations

The idealized multipole current configurations discussed previously require
a continuously varying distribution of current around the entire circumference.
On the other hand, actual magnets are typically constructed from multiple layers of
cables with uniform current density. Thus it is important to develop methods for
approximating the desired multipole distribution, such that it produces the max-
imum amount of the desired multipole and still meets the required field quality for
the magnet. To illustrate this, we consider here the design of a normal dipole
magnet using circular current sheet sectors with constant current density.
We know that the idealized current distribution for a dipole goes like cos θ.
The simplest approximation to a cos θ distribution is to put constant current sectors
in each of the four quadrants, as shown in Figure 4.7. The multipole fields result
from the sum of the contributions of the four sheets.
Figure 4.7 Dipole approximation.
(ð ð πϕ1 ð πþϕ2 ð ϕ1 )

ϕ2
μI
Bn ¼ 0 n cos nϕ dϕ þ
2πa ϕ1 πϕ2 πþϕ1 ϕ2
Let S( ) = sin( ). Then Bn can be written as the sum of eight sine terms.
μ0 I
Bn ¼ ½ Sðnϕ2 Þ Sðnϕ1 Þ Sðnπ nϕ1 Þ þ Sðnπ nϕ2 Þ
2πnan
Sðnπ þ nϕ2 Þ þ Sðnπ þ nϕ1 Þ Sðnϕ1 Þ þ Sðnϕ2 Þ :
If n is even, the terms in the brackets cancel, so Bn = 0. For odd values of n, we get
2μ0 I
Bn ¼ ½sin nϕ2 sin nϕ1 : (4.46)
πnan
A similar calculation shows that An = 0 for both odd and even values of n. Thus the
allowed harmonics for this current configuration are just the Bn , for odd values of n.
In addition to the desired harmonic B1 that represents the dipole, the approxima-
tion of the ideal multipole distribution above introduces other harmonics, which
represent errors to the desired field. The dominant allowed error here is the
sextupole term B3 . The angle ϕ1 is typically set as close to 0 as possible in order
to maximize the dipole strength. The angle ϕ2 can then be used to remove the
sextupole component from the field. Since
2μ0 I
B3 ≃ sin 3ϕ2 ;
3πa3
we can eliminate the sextupole term by setting ϕ2 ¼ π=3. Removing other allowed
higher harmonics from the field requires additional degrees of freedom. A number
of coil configurations have been proposed to approximate a cos θ distribution.[6]

For example, a second current sector can be added in each quadrant that is
separated from the first sector by a non-conducting spacer, or sectors could be
added at different radii.[7] In principle, these additional sectors could also have
independent currents. A two-layer design with spacers is described in Section 11.6.
The allowed harmonics for any multipole of order N follow from the requirement
that the direction of the current is in opposite directions on either side of a pole.
The first pole is located at
π
ϕ¼ :
2N
Referring to Figure 4.7, the current distributions on the opposite sides of the
pole are
dI dI
ðβÞ ¼ ðϕÞ;
dϕ dϕ
where
π π
β¼2 ϕ ¼ ϕ:
2N N
To get a net contribution to a normal multipole of order n
dI
Bn ∝ ðϕÞ cos nϕ
dϕ
from the current on both sides of the pole, the cosine function must also change sign.
h π i
cos n ϕ ¼ cos nϕ
N
nπ nπ
cos cos nϕ þ sin sin nϕ ¼ cos nϕ:
N N
This requires that

nπ
cos ¼ 1
N
nπ
sin ¼ 0:
N
The sine equation requires that n/N is an integer, while the more restrictive cosine
relation demands that n/N is an odd integer. Thus the allowed Bn must have
n ¼ Nð2m þ 1Þ; m ¼ 0; 1; 2; (4.47)
Table 4.1 Multipole symmetries [8]
Symmetry Normal multipoles Skew multipoles Example

Up-down symmetric all An = 0 Normal dipole
Up-down antisymmetric all Bn = 0 Skew dipole
Left-right symmetric Bn = 0 for odd n An = 0 for even n Normal quadrupole
Left-right antisymmetric Bn = 0 for even n An = 0 for odd n Normal dipole
A similar argument shows that the allowed skew multipoles also satisfy
Equation 4.47.
The symmetry of the current distribution is directly related to the allowed
harmonics. Let KðϕÞ ¼ dI=dϕ, for example, and assume the current distribution
is up-down symmetric, so that
KðϕÞ ¼ KðϕÞ:
Then the skew multipoles are

ð
μ0 π
An ¼ ½KðϕÞ sin nϕ þ KðϕÞ sin ðnϕÞ dϕ
2πan 0
ð
μ0 π
¼ KðϕÞ½sin nϕ sin nϕ dϕ ¼ 0:
2πan 0
Thus the fact that the dipole approximation had An = 0 follows from the up-down
symmetry of the current distribution that we used. The consequences of some other
symmetries for current distributions are listed in Table 4.1. These symmetries are
inevitably violated to some extent in building an actual magnet and this leads to the
presence of “nonallowed” multipoles in the fields. Random errors in the construc-
tion of the magnet can introduce values for any multipole.[9] If these nondesired
multipoles exceed their tolerances, they must be removed by modifications in the
manufacturing process or by introducing correction coils.
4.7 Field for a block conductor

Block conductors, which have a finite area in the x-y plane, are the most realistic
approximation to actual conductors in two dimensions. We again consider the case
where the conductor is infinitely long in the z direction and where the currents only
flow along z. From the Biot-Savart law Equation 1.14, we have
ð! !
! μ0 J R
B ¼ dV:
4π R3
Figure 4.8 Current block geometry.
Assume an observation point P is located at (r, θ) in the x-y plane and the current
element is at (a, ϕ), as shown in Figure 4.8. Let the vector ρ be the distance between
the current element and the field point in the x-y plane and let σ be the current
density in the conductor. Then we have
! ¼ σ ^z
J
! !
R ¼ ρ þ z ^z
dV ¼ dS dz:
Thus B can be written as

ð ð∞
! ðr; θÞ ¼ μ0 σ ^z ½!
ρ þ z ^z
B 3=2
dz dS
4π ∞ fρ2 þ z2 g
ð
μ
¼ 0 σ IðρÞ ^z ! ρ dS;
4π
where IðρÞ is given by Equation 4.28. We find that the field from the current block is
given by
ð
! μ0 ^z ! ρ
B ðr; θÞ ¼ σ dS: (4.48)
2π ρ 2
The vector potential for a current block can be found by integrating the vector
potential for the line current, Equation 4.21, over the area of the block.
ð
μ0 ρ
Az ðr; θÞ ¼ σ ln dS: (4.49)
2π ro
Figure 4.9 Overlapping circular cylindrical conductors.
Example 4.8: overlapping circular conductors

Imagine we have two circular cylindrical conductors with constant current density
flowing in opposite directions. We know from Equation 1.27 that the field inside the
conductor is
μ0 Jρ
Bϕ ¼ :
2
Suppose we overlap the two conductors with the centers displaced along the x axis at
c and –c, as shown in Figure 4.9. The field at some arbitrary point P in the overlap
region is the sum of the fields from the two conductors. From the geometry in the
figure, we see that [10]
μ0 J
Bx ¼ ½r2 sin θ2 þ r1 sin θ1 ¼ 0
2
μJ μJ
By ¼ 0 ½r1 cos θ1 r2 cos θ2 ¼ 0 2c;
2 2
where 2c is the separation between the centers of the two circles. Thus the field in the
overlap region is a pure dipole. The strength of the field is proportional to the separation
between the circles. In the region where the two coils overlap, the net current is zero.
Thus the conductor in the overlap region can be removed without affecting the field there.
Next we examine the coil thickness t as a function of the azimuthal angle ϕ, as
shown in Figure 4.10. The coil thickness at some angle ϕ is the distance between the
points P1 and P2 . Point P1 is determined by the intersection of circle 1
ðx þ cÞ2 þ y2 ¼ a2
Figure 4.10 Conductor thickness for overlapping circle configuration.
with the straight line

y ¼ x tan ϕ ¼ mx:
We find that
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c þ a2 m2 c2 þ m2 a2
x1 ¼ :
1 þ m2
Point P2 is determined by the intersection of circle 2
ðx cÞ2 þ y2 ¼ a2
with the straight line. We find that the expression for x2 is the same as the one for x1,
except that the first term in the numerator is +c instead of –c. Thus we have
2c
Δx ¼ x2 x1 ¼
1 þ tan2 ϕ
Δy ¼ y2 y1 ¼ Δx tan ϕ:
The resulting thickness of the conductor is

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tðϕÞ ¼ ðΔxÞ2 þ ðΔyÞ2
¼ 2c cos ϕ:
Thus the overlapping circular conductors represent another form of a cosine current
distribution. Quadrupole fields can be designed in the same manner using over-
lapping elliptical conductors.[8]
Figure 4.11 Annular current block.
Example 4.9: on-axis field of an annular sector with constant current density
Consider the annular sector conductor shown in Figure 4.11. For a field point on the
axis we have
!
ρ ¼ a cos ϕ ^x a sin ϕ ^y :

ð a 2 ð ϕ2
! ð0; 0Þ ¼ μ0 σ ^z ða cos ϕ ^x þ a sin ϕ ^y Þ
B a dϕ da
2π a1 ϕ1 a2
ð a 2 ð ϕ2
μ0 σ
¼ ðcos ϕ ^y sin ϕ ^x Þ dϕ da:
2π a1 ϕ1
Thus the on-axis field of the annular sector is
! μσ
B ð0; 0Þ ¼ 0 ða2 a1 Þ½ðsin ϕ2 sin ϕ1 Þ^y þ ðcos ϕ2 cos ϕ1 Þ^x :
2π
(4.50)
We see that the field for the case of constant current density is directly proportional to
the radial thickness of the conductor.
Example 4.10: field due to a rectangular conductor

Consider a rectangular conductor with constant current density σ in the z direction, as
shown in Figure 4.12. We substitute
!
ρ ¼ ðxo xÞ ^x þ ðyo yÞ ^y
Figure 4.12 Rectangular conductor.
into Equation 4.48 and find that
! μ0 σ ½ðxo xÞ ^y ðyo yÞ ^x
dB ¼ dS: (4.51)
2π ðxo xÞ2 þ ðyo yÞ2
Looking at By , we have
ðb ðd
μ0 σ ðxo xÞ
By ¼ dx dy:
2π a c ðxo xÞ2 þ ðyo yÞ2
Integrating first over x, we evaluate7

ð x2
ðxo xÞ
I1 ðx1 ; x2 Þ ¼ dx
x1 ðxo xÞ2 þ ðyo yÞ2
1
¼ ln½ðxo xÞ2 þ ðyo yÞ2 xx21 :
2
Then integrating over y, we find8
ð y2
I2 ðy1 ; y2 ; αÞ ¼ ln½ðxo αÞ2 þ ðyo yÞ2 dy
y1
y2
2 2 1 yo y
¼ ðyo yÞln½ðxo αÞ þðyo yÞ þ 2ðyo yÞ2sðxo αÞ tan
sðxo αÞ y1 ;
where s = ±1. In order to get a physical solution, we need to choose s ¼ þ1 when

xo > α and s ¼ 1 when xo < α. This is equivalent to taking the absolute value of
ðxo αÞ. Thus we can write By as
7 8
GR 2.175.1. GR 2.733.1.
μ0 σ
By ¼ ½I2 ðc; d; bÞ I2 ðc; d; aÞ
4π

μσ yo c
¼ 0 ðyo cÞ ln½ðxo bÞ2 þ ðyo cÞ2 þ 2jxo bj tan1
4π jxo bj

2 2 1 yo d
ðyo dÞ ln½ðxo bÞ þ ðyo dÞ 2jxo bj tan
jxo bj

2 2 1 yo c
ðyo cÞ ln½ðxo aÞ þ ðyo cÞ 2jxo aj tan
jxo aj

2 2 1 yo d
þ ðyo dÞ ln½ðxo aÞ þ ðyo dÞ þ 2jxo aj tan :
jxo aj
From the symmetry of Equation 4.51, we see that the result for Bx is the negative of
the result for By with the substitutions
x↔y
xo ↔yo
ða; bÞ↔ðc; dÞ:
Thus Bx is given by

μ0 σ 2 2 1 xo a
Bx ¼ ðxo aÞ ln½ðxo aÞ þ ðyo dÞ þ 2jyo dj tan
4π jyo dj

xo b
ðxo bÞ ln½ðxo bÞ2 þ ðyo dÞ2 2jyo dj tan1
jyo dj

2 2 1 xo a
ðxo aÞ ln½ðxo aÞ þ ðyo cÞ 2jyo cj tan
jyo cj

2 2 1 xo b
þ ðxo bÞ ln½ðxo bÞ þ ðyo cÞ þ 2jyo cj tan :
jyo cj
Using these expressions for the field, we show in Figure 4.13 a scan of the vertical
field component along the x axis for a square conductor centered at the origin.
The calculation using these equations fails when the observation point is located at
one of the four corners of the rectangle.
We can find the multipoles produced by an annular sector conductor block

analogously to the procedure used for current sheets in Section 4.5. The normal
multipoles are given by
ðð
μ cos nϕ
Bn ¼ Jða; ϕÞ n1 da dϕ: (4.52)
2π a
Figure 4.13 Vertical field from a square conductor with sides of length 2 centered
at the origin.
The skew multipoles can be found in a similar manner and are given by
ðð
μ sin nϕ
An ¼ Jða; ϕÞ n1 da dϕ: (4.53)
2π a
4.8 Ideal multipole current block

Consider an annular current block with a pure multipole current density J.
We assume that the block is composed of a radial distribution of ideal multipole
current sheets extending from a1 to a2 . We find the vector potential for the block by
integrating the potential for a multipole sheet. The current density in the block is
assumed to be proportional to cos mϕ.
There are three cases, depending on the radius of the field observation point r.
Case 1: r < a1
The vector potential, Equation 4.35, for the ideal multipole sheet is
μI0 r m
Azsh ðr; θÞ ¼ cos mθ:
2m a
Then the vector potential for the current block is

ð a2 m
μJ0 r
Az ðr; θÞ ¼ cos mθ a da:
2m a1 a
4.8 Ideal multipole current block 101
For m ≠ 2, this can be written as

mþ2
μJ0 m a2 amþ2
Az ðr; θÞ ¼ r cos mθ 1
: (4.54)
2m m þ 2
When m = 2, we have

μJ0 r2 a2
Az ¼ cos 2θ ln : (4.55)
4 a1
Case 2: r > a2
When the observation point is always larger than the radius of any of the multipole
sheets, we use Equation 4.37 for the vector potential of the ideal multipole sheets.
μI0 am
Azsh ðr; θÞ ¼ cos mθ:
2m r
The vector potential of the current block is

ð a2
μJ0
Az ðr; θÞ ¼ cos mθ am a da:
2m rm a1
Evaluating the integrals, we have

mþ2
μJ0 m a2 amþ2
Az ðr; θÞ ¼ r cos mθ 1
: (4.56)
2m mþ2
Case 3: a1 < r < a2

In the third case, where the observation point can be inside the current block, we
must break the radial integration into two parts, depending on the relative positions
of r and the multipole sheet.
ðr ð a2
μJ0
Az ðr; θÞ ¼ cos mθ rm amþ1 da þ rm amþ1 da :
2m a1 r
Evaluating the integrals, we have for m ≠ 2,

mþ2 mþ2
μJ0 m r amþ2 m a2 rmþ2
Az ¼ cos mθ r 1
þr : (4.57)
2m mþ2 m þ 2
For the case m = 2, we have instead

4 a
μJ0 2 r a1
4
2
Az ¼ cos 2θ r þ r ln
2
: (4.58)
4 4 r
Figure 4.14 Radial scan of the vertical component of the magnetic field of an ideal
dipole (m =1, a1 = 0.2 m, a2 = 0.3 m, J0 = 100 A/mm2).
The components of the magnetic field in each of these regions is easily com-
puted from these vector potentials. Figure 4.14 shows a radial scan of the
vertical component of the magnetic field along the midplane for an ideal dipole.
The current block extends from 0.2 m to 0.3 m in this example. The field is
constant and purely vertical inside the aperture, as expected. The field changes
direction inside the current block. The strength of the field slowly falls off
outside the block.
4.9 Field from a magnetized body

Let us consider the field intensity outside a magnetized body.9 We saw from
Equation 3.32 that the scalar potential for a magnetic body is
ð !0 !
1 M ·R
Vm ¼ dV 0 :
4π R3
In the two-dimensional case, let us assume M only has x and y components.

The scalar potential is
ð ð∞ ð∞
1 0 Rx 0 0 Ry 0
Vm ¼ Mx 3
dz þ My 3
dz dS0 ;
4π ∞ R ∞ R
9
We will consider H inside a magnetized body in Chapter 9.
4.9 Field from a magnetized body 103
where Rx ¼ x x0 , etc. Define the transverse distance in the plane with z = 0 as

!
ρ ¼ Rx ^x þ Ry ^y :
The integrals over z0 are10
ð∞ ð∞
Rx 0 dz0 2Rx
dz ¼ R x ¼ 2 :
∞ R
3
∞ fρ2 þ z0 2 g
3=2 ρ
Then the two-dimensional potential is given by

ð !0 !
1 M ·ρ
Vm ¼ dS0 : (4.59)
2π ρ2
The two-dimensional field is

ð !0 !
! 1 M ·ρ
Hm ¼ r dS0 :
2π ρ2
Using the vector relation B.2,
!
!0 !
ρ !0 !
ρ ρ ! ! !
ρ
r M · 2 ¼ M r 2 þ 2 r M0 þ M0 · r 2
ρ ρ ρ ρ
!
ρ !
þ 2 · r M0 : (4.60)
ρ
The gradient operator only acts on unprimed coordinates, so the second and fourth
terms on the right-hand side vanish.
In the first term, we have using Equation B.6
!
ρ 1 1
!
r 2 ¼ 2 r ρ þr 2 ! ρ: (4.61)
ρ ρ ρ
The first term here vanishes because ρ is radial and so its curl vanishes. In the
second term

1 1 1
r 2 ¼ ^x ∂x 2 þ ^y ∂y 2
ρ ρ ρ
2 !
¼ 4 ρ:
ρ
Thus the cross-product of the last two factors in Equation 4.61 is 0, and the second
term also vanishes.
10
GR 2.271.5.
Thus only the third term in Equation 4.60 survives and we have
ð !
! 1 !0 ρ
Hm ¼ ð M ·rÞ 2 dS0 : (4.62)
2π ρ
Expanding the dot product,

ð ! !
! 1 0 ρ 0 ρ
Hm ¼ Mx ∂x 2 þ My ∂y 2 dS0 :
2π ρ ρ
Writing out the vector ρ in terms of its x and y components, taking the derivatives,
combining terms, and dropping the primes, we find that the two-dimensional field
from an iron element is given by
ð "! ! #
! 1 M 2ðM · ! ρÞ !
Hm ¼ ρ dS: (4.63)
2π ρ2 ρ4
4.10 Superconductors
High-field magnets are usually energized using superconducting cables. In the
superconducting state, the resistivity vanishes, so a large current can flow through
the magnet coils without losing power due to Joule heating. In a superconducting
material, attractive forces between pairs of electrons are transmitted through
vibrations in the material lattice.[11] The operating conditions for superconductiv-
ity lie below the surface of a three-dimensional space of temperature, magnetic
field, and current density. The limiting values on each of the three axes are called
the critical values. When the superconductor is not in the superconducting state, it
is said to be in the normal state. There are several classes of superconducting
materials. Type I materials exhibit the Meissner effect, where any external magnetic
field is excluded from the interior of the superconductor. A magnetization is
generated in the superconductor that just cancels the external field. This remains
true as the external field is increased until it reaches the critical field Hc. Type I
superconductors are perfect diamagnetic materials.
Some alloys of intermetallic compounds form what are called type II super-
conductors. Two important examples are NbTi and Nb3 Sn. The critical current
density for these materials at 4 K is shown as a function of the magnetic flux density
in Figure 4.15.[12] NbTi is a useful material for fields up to ~9 T at 4 K, whereas
Nb3 Sn can be used up to ~22 T. These materials have two critical magnetic fields,
Hc1 and Hc2 , which can be much larger than Hc1 . For H < Hc1, the material
4.11 End fields 105
10 000
JE [A / mm2]
1000
Nb3Sn
NbTi
100
5 10 15 20 25
B [T]
Figure 4.15 Engineering current density at 4 K as a function of the magnetic field.
completely excludes the external flux and it behaves like a type I superconductor.
For Hc1 < H < Hc2, flux begins to penetrate into the material and the magnitude of
the magnetization begins to fall, but electrically it still has zero resistivity.
The magnetic flux enters the superconductor in the form of discrete, quantized
flux lines known as fluxoids. The field still vanishes in the material surrounding the
fluxoids. The fluxoids can move because of Lorentz forces, creating heat. To stop
the fluxoids from moving, inhomogeneities known as pinning centers must be
introduced into the lattice. Finally when H > Hc2 , the material returns to the
normal state.
Magnets with superconducting cables must deal with the problem of persistent
currents.[13] As the current in the magnet is ramped up or down, eddy currents11
are induced in the superconductor. These induced shielding currents produce a field
that opposes the change in the field caused by the magnet’s power supply. Because
of the lack of resistance in a superconductor, the decay times for the induced
currents are very long. The persistent currents produce undesired sextupole and
higher multipoles that can be particularly significant at low field values.
4.11 End fields

Although this chapter has mainly been concerned with transverse fields that are
uniform along the z direction, real currents exist as closed loops, so we must
comment on what happens at the end of this type of magnet. At the ends of
a dipole, the conductor on one side of a pole must bend in such a way that it returns
11
Eddy currents will be discussed in more detail in Section 10.4.
in the opposite direction at the symmetrical location on the other side of the pole.
There are several standard end configurations. If the aperture in the end regions
does not need to be open, the simplest configuration is the racetrack coil.[10]
The coil bends around an arc, maintaining the same vertical position as the coils in
the straight part of the magnet. The bedstead end bends the conductors by 90° as
quickly as possible and then crosses over the pole at a fixed z location. This
configuration does keep a clear aperture in the end regions, but the sharp bend in
the conductor may not be acceptable for mechanical reasons. Another type of end
that maintains a clear aperture is the saddle end.[10] In this case, the conductor
turns cross over the pole following an arc that is spread out over z. The magnetic
design of real coil ends is usually done numerically using the Biot-Savart equation.
The end turns introduce additional multipole contributions to the field of
a magnet. However, it is possible to define a new set of multipole coefficients
defined in terms of the field components integrated along the axis of the magnet.
Flux theorems have been developed that can relate these integrated multipoles to
the geometry of the end turns.[14] The z locations where the cross-over for various
conductors begin can be adjusted using spacers to help balance the integrated
multipoles in the magnet.[15]
References
[1] W. Smythe, Static and Dynamic Electricity, 2nd ed., McGraw-Hill, 1950, p. 65.
[2] R. Gupta, Field Calculations and Computations, School at Centre for Advanced
Technology, Indore, India, 1998, p. 32.
[3] N. Schwerg & C. Vollinger, Analytic models for the calculation of the iron yoke
contribution in superconducting accelerator magnets, CERN Accelerator Technology
Department report CERN/AT 2007-33, 2007.
[4] P. Schmuser, Superconducting magnets for particle accelerators, in M. Month &
M. Dienes (eds.), The Physics of Particle Accelerators, AIP Conf. Proc. 249, vol. 2,
1992, p. 1106.
[5] G. Morgan, Shaping of magnetic fields in beam transport magnets, in M. Month &
M. Dienes (eds.), The Physics of Particle Accelerators, AIP Conf. Proc. 249, vol. 2,
1992, p. 1256–1259.
[6] M. Wilson, Superconducting Magnets, Oxford University Press, 1983, p. 32.
[7] J. Coupland, Dipole, quadrupole and higher order fields from simple coils, Nuc. Instr.
Meth. 78:181, 1970.
[8] A. Jain, Basic theory of magnets, in S. Turner (ed.), CERN Accelerator School on
Measurement and Alignment of Accelerator and Detector Magnets, CERN 98–05,
1998, p. 25.
[9] G. Parzen, Random errors in the magnetic field of superconducting dipoles and
quadrupoles, Part. Acc. 6:237, 1975.
[10] M. Wilson, op. cit., p. 27–28.
[11] Ibid., p. 279.
[12] https://nationalmaglab.org/magnet-development/applied-superconductivity-center.
References 107
[13] P. Schmuser, op. cit., p. 1123–1141.

[14] F. Mills & G. Morgan, A flux theorem for the design of magnet coil ends, Part. Acc.
5:227, 1973.
[15] R. Palmer & A. Tollestrup, Superconducting magnet technology for accelerators, in
J. Jackson, H. Gove & R. Schwitters (eds.), Annual Review of Nuclear and Particle
Physics, vol. 34, Annual Reviews Inc., 1984, p. 269–270.
5
Complex analysis of transverse fields
In this chapter, we continue the discussion of transverse fields that are determined
by the location of the conductors. In the “central” region, far from the magnet ends,
the powerful methods of complex analysis1 can be applied to the calculation of
potentials, magnetic fields, multipoles and forces. Many of the topics in this chapter
are based on a series of important papers by Richard Beth and by Klaus Halbach.
Beginning with the field from a line current, we consider methods for calculating
the fields from current sheets. Then we use the complex form of Green’s theorem to
express the fields of block conductors in terms of contour integrals.
5.1 Complex representation of potentials and fields

We define the complex potential function as
WðzÞ ¼ uðx; yÞ þ ivðx; yÞ:
The real and imaginary parts of W must satisfy the Cauchy-Riemann equations,
which are expressed in Cartesian coordinates as
∂u ∂v
¼
∂x ∂y
(5.1)
∂u ∂v
¼ :
∂y ∂x
From Equations 3.2 and 3.26 in free space in two dimensions, we have
∂Az ∂Vm
Bx ¼ ¼ μ0
∂y ∂x
(5.2)
∂Az ∂Vm
By ¼ ¼ μ0 :
∂x ∂y
1
A brief summary of some important results from the theory of complex variables is given in Appendix E.
108
5.1 Complex representation of potentials and fields 109
These equations can be put into the form of the Cauchy-Riemann equations by
associating
u ¼ Az
v ¼ μ 0 Vm :
Thus in two dimensions, the vector and scalar potentials are related to each other as
the real and imaginary parts of the complex potential function
WðzÞ ¼ Az þ iμ0 Vm : (5.3)
Now consider the derivative of the complex potential. A complex function with
a continuous derivative is known as an analytic function. A complex derivative
must give the same result independent of the manner that Δz approaches 0. In the
case when Δz ¼ Δx, we have
dW ∂W ∂Az ∂Vm
¼ ¼ þ iμ0 ¼ By iBx :
dz ∂x ∂x ∂x
If we had chosen Δz ¼ iΔy instead, we would obtain the same expression for B.
So in either case we find
dW
i ¼ Bx iBy :
dz
Defining the complex magnetic field as2
BðzÞ ¼ Bx þ iBy ; (5.4)
we find the relation between the magnetic field and the potential is
dW
B ðzÞ ¼ i ; (5.5)
dz
where B* is the complex conjugate of B.[1]

We can transform the magnetic field between Cartesian and polar coordinates by
using the complex rotation variable. Let
Bc ¼ Bx þ iBy
Bp ¼ Br þ iBθ
2
Unfortunately, Beth and Halbach use different definitions for the complex magnetic field H and use different
systems of units, so some care must be exercised in comparing their results. We follow Halbach’s conventions
here in defining the components of H the same way as normal complex variables and using the SI system of
units.
110 Complex analysis of transverse fields
be the Cartesian and polar representations of a complex variable. Defining

R ¼ eiθ ¼ cos θ þ i sin θ;
we can transform between the two representations using

B p ¼ R Bc
(5.6)
Bc ¼ RBp :
Consider the analytic function

f ðzÞ ¼ uðx; yÞ þ ivðx; yÞ: (5.7)
Differentiating the first Cauchy-Riemann Equation 5.1 with respect to x, we have
∂2 u ∂2 v
¼ :
∂x 2 ∂x∂y
The fact that the second partial derivative has to exist follows from the analytic
nature of f ðzÞ.[2] Differentiating the second Cauchy-Riemann equation with
respect to y gives
∂2 u ∂2 v
¼ :
∂y2 ∂x∂y
Combining these equations, we find that

∂2 u ∂2 u
þ ¼ 0:
∂x2 ∂y2
Thus uðx; yÞ satisfies the Laplace equation. Similarly, we can differentiate the first
Cauchy-Riemann equation with y and the second with x to show that vðx; yÞ also
satisfies the Laplace equation. It follows that the real and imaginary parts of any
analytic function satisfy the Laplace equation.
Returning to Equation 5.7, consider the two curves
uðx; yÞ ¼ α1
vðx; yÞ ¼ β1 ;
where α1 and β1 are fixed values. Differentiating u with respect to x, we find
∂u ∂u dy
þ ¼ 0:
∂x ∂y dx
The slope of the curve is

dy ∂x u
mα ¼ ¼ :
dx ∂y u
5.2 Maxwell’s equations in complex conjugate coordinates 111
Differentiating v with respect to x, we find

∂v ∂v dy
þ ¼0
∂x ∂y dx
and the slope of this curve is

∂x v
mβ ¼ :
∂y v
The product of the slopes is

∂x u ∂x v
mα mβ ¼ :
∂y u ∂y v
Rewriting the numerator using the Cauchy-Riemann equations, we find

ð∂y vÞð∂y uÞ
mα mβ ¼ ¼ 1:
∂y u∂y v
From analytic geometry, this is the condition that indicates that two lines are
perpendicular. Thus the real and imaginary parts of an analytic function describe
orthogonal curves. This indicates in particular that the equipotential lines for Az and
Vm cross at right angles.
5.2 Maxwell’s equations in complex conjugate coordinates

Instead of defining complex variables as functions of x and y, it is sometimes more
convenient to use z and z* as the independent variables. These are known as
complex conjugate coordinates.[3] We can write the partial derivatives with respect
to x and y in terms of these variables as
∂ ∂ ∂
¼ þ
∂x ∂z ∂z
(5.8)
∂ ∂ ∂
¼i :
∂y ∂z ∂z
The corresponding derivatives with respect to z and z* are
∂ ∂ ∂
2 ¼ i
∂z ∂x ∂y
(5.9)
∂ ∂ ∂
2 ¼ þi :
∂z ∂x ∂y
We can use Equation 5.8 to write the magnetostatic Maxwell equations in complex
coordinates. The divergence equation
∂x Hx þ ∂y Hy ¼ 0
becomes [1]
∂H ∂H
þ ¼ 0; (5.10)
∂z ∂z
where H ¼ Hx þ iHy is the complex magnetic field intensity. The curl equation
∂x Hy ∂y Hx ¼ Jz ≡ σ
can be transformed using Equation 5.8 into the form

i∂z H þ i∂z H ¼ σ:
This can be further simplified using Equation 5.10, resulting in two forms for the
curl equation.[1]
∂H
2i ¼σ
∂z
(5.11)
∂H
2i ¼ σ:
∂z
Operating on the complex potential in Equation 5.3, we find
∂W
2 ¼ ð∂x Az ∂y μ0 Vm Þ þ ið∂x μ0 Vm þ ∂y Az Þ:
∂z
The expressions in parentheses are the Cauchy-Riemann equations, which are thus
compactly incorporated into the expression [4]
∂W
¼ 0: (5.12)
∂z
The r operator can be written as
∂
r¼2
∂z
(5.13)
∂
r ¼ 2
∂z
and the Laplacian is
∂2
r2 ¼ 4 : (5.14)
∂z∂z
5.3 Field from a line current 113
Figure 5.1 Geometry of a line current.
5.3 Field from a line current

Consider a single filament of current that crosses the x-y plane at the location z and
an observation point zo , as shown in Figure 5.1. The displacement between these
two points is
zo z ¼ Reiα : (5.15)
We know from Equation 1.26 that the field of the line current is
! μ0 I
B ¼ ð^x sin α þ ^y cos αÞ: (5.16)
2πR
Every filament in a magnet must have a return filament of the opposite polarity
somewhere. It is convenient to assume that all filaments have their currents return
through a filament at the coordinate origin. Considering Equation 3.9 for the vector
potential of a line current, we assume the complex potential for a filament and its
return is given by
μ0 I
W¼ ½lnðzo zÞ lnðzo 0Þ:
2π
The second term in this equation is constant for a given field point. In the cross-
section of any real magnet, there are equal numbers of filaments with positive and
negative currents. Thus summed over all the filaments in a magnet, the second
terms cancel out. The first terms do not cancel out in general because the filaments
have different positions z. Thus the potential for the line current is
μ0 I
Wðzo Þ ¼ lnðzo zÞ: (5.17)
2π
Using Equations 5.5 and 5.15 with the derivative operating on the coordinates of
the field point zo , we get the magnetic field
μ0 I 1
Bx i By ¼ i
2π zo z
μI
¼ i 0 eiα (5.18)
2πR
μI
¼ i 0 ðcos α i sin αÞ;
2πR
which agrees with the result in Equation 5.16.
From Equation 5.15, the complex logarithm is
lnðzo zÞ ¼ lnR þ iα:
Using Equations 5.3 and 5.17 we can confirm that the vector potential for a line
current is
μ0 I
Az ¼ lnR (5.19)
2π
and find that the scalar potential is

I
Vm ¼ α
2π
(5.20)
I 1 yo y
¼ tan :
2π xo x
Although Equations 5.19 and 5.20 appear very different, they both lead to the fields
in Equation 5.18.
Assume that a line current is located at position z. Then, according to
Equation 5.18, the field intensity at the observation point zo is
I 1
H ðzo Þ ¼ i : (5.21)
2π zo z
Integrate H* over observation points around any closed contour that encloses the
point z.
þ þ
I 1
H dzo ¼ i dzo :
2π zo z
According to Cauchy’s integral formula,3
3
See Appendix E.
þ
1
dzo ¼ 2πi:
zo z
Thus we find that the complex form of the Ampère law is

þ
H dzo ¼ I; (5.22)
where I is the total current enclosed by the contour.

Now consider a line current in the vicinity of a plane surface of infinite perme-
ability iron. As we saw in Chapter 2, the effect of the iron on the field of a conductor
filament is equivalent to the presence of an image filament on the other side of the
iron surface. The direction of the image current is the same as the conductor
current. In the case of a circular boundary of radius R, the positions of the conductor
and image filaments are
z ¼ ρeiϕ
R2 iϕ R2 (5.23)
zI ¼ e ¼ :
ρ z
5.4 Field from a current sheet

We can consider a current sheet as a collection of parallel line currents. The sheet is
assumed to have a finite width, but to have infinitesimal thickness. Then using
Equation 5.17, the potential for the current sheet is
ð
μ0
Wðzo Þ ¼ KðsÞln½zo zðsÞds; (5.24)
2π
where s is the arc length along the sheet and K ¼ dI=ds is the sheet current density.
Example 5.1: potential for a straight sheet with constant K

Consider the straight sheet with width b shown in Figure 5.2. The current density is
K ¼ bI and the filaments making up the sheet are located at
zðsÞ ¼ z1 þ seiθ :
It follows that dz ¼ eiθ ds and
z2 ¼ z1 þ beiθ : (5.25)
From Equation 5.24, the potential is

ð
μ0 I
Wðzo Þ ¼ ln½zo zeiθ dz:
2πb
Figure 5.2 A straight current sheet.
We can write
lnðzo zÞ ¼ lnðz zo Þ þ lnð1Þ

¼ lnðz zo Þ þ iπ:
The second term is independent of zo and gets absorbed into the constant term for the
potential. Defining u ¼ z zo , we get
ð u2
μI
Wðzo Þ ¼ 0 eiθ lnu du:
2πb u1
Substituting for eiθ from Equation 5.25 and evaluating the integral4 gives
μ0 I
Wðzo Þ ¼ ½u2 lnu2 u1 lnu1 u2 þ u1 :
2πðz2 z1 Þ
The last two terms in the square bracket give
u2 þ u1 ¼ ðz2 zo Þ þ ðz1 zo Þ

¼ z1 z2 :
This term is also independent of zo and gets absorbed into the constant term for the
potential. Thus the potential for the straight sheet is given by [5]
μ0 I
Wðzo Þ ¼ ½u2 ln u2 u1 ln u1 : (5.26)
2πðz2 z1 Þ
Summing up the contributions to the magnetic field from the field of individual
line currents given in Equation 5.21, we find the field of a current sheet is given by
4
CRC 377.
ð
i KðsÞ
H ðzo Þ ¼ ds; (5.27)
2π zðsÞ zo
where s is the distance along the sheet. If the position along the sheet is specified by
the polar angle ϕ, this can be written as
ð
i dI=dϕ
H ðzo Þ ¼ dϕ: (5.28)
2π zðϕÞ zo
It is possible to determine a unique current distribution dI=dϕ for circular or elliptic
current sheets that can produce any desired two-dimensional field compatible with
Maxwell’s equations in the magnet aperture.[6]
Example 5.2: field due to a circular arc sheet with constant current density
Let us consider a current sheet in the form of a circular arc, as shown in Figure 5.3.
The magnetic field from the sheet is given by Equation 5.28.
ð ϕ2
i dI dϕ
H ðzo Þ ¼ :
2π dϕ ϕ1 zo zðϕÞ
Since z ¼ aeiϕ , we can write this as

1 dI
H ðzo Þ ¼ I;
2π dϕ
where5
ð z2
dz
I¼
z1 z ðzo zÞ
(5.29)
2i 1 zo 2z
¼ tan i :
zo zo
Figure 5.3 Circular arc current sheet.
5
GR 2.172.
Therefore the field is

z2
i dI 1 zo 2z
H ðzo Þ ¼ tan i :
πzo dϕ zo z1
Using the relation6

1 1 þ iz
tan1 z ¼ ln ;
2i 1 iz
we can write the field of the circular arc as

i dI z2 z1
H ðzo Þ ¼ ln ln : (5.30)
2πzo dϕ zo z2 zo z1
From Equation 5.28, the field of the arc conductor at the origin is
ð ϕ2
i dI dϕ
H ð0Þ ¼
2π dϕ ϕ1 a eiϕ
1 dI iϕ2
¼ ½e eiϕ1 :
2πa dϕ
If the angular arc completes a full circle, we have a current shell and the integral in
Equation 5.29 becomes
þ
dz
I¼ :
zðz zo Þ
A point where the denominator of the integrand becomes zero is called a pole. If zo is
inside the circle, then the contour integral has simple poles at z = 0 and z ¼ zo .
The residue7 for the pole at z ¼ 0 is
1 1
lim z ¼
z→0 zðz zo Þ zo
and the residue for the pole at z ¼ zo is
1 1
lim ðz zo Þ ¼ :
z→zo zðz zo Þ zo
Therefore, by the residue theorem, the value of the integral is zero and the field inside
the shell vanishes. When zo is outside the shell, the integral only has the pole at z = 0
and the residue theorem gives
6 7
GR 1.622.3. See Appendix E.

1 2πi
I ¼ 2πi ¼ :
zo zo
Therefore, the field outside the shell is
i dI
H ðzo Þ ¼ :
zo dϕ
Since the total current is
dI
I ¼ 2π ;
dϕ
the field can be written as
I
H ðzo Þ ¼ i :
2πzo
This is the same as Equation 5.21 for the field of a line current located at the center of
the shell.
Let us apply the Ampère law, Equation 5.22, for an infinitesimal rectangular
contour across a current sheet, as shown in Figure 5.4. Then we have,
H1 ðzo Þdz H2 ðzo Þdz ¼ dI;
where dI is the current enclosed in the contour. In the limit where the distance
perpendicular to the sheet approaches 0, the path of the observation
points approach the path along the sheet and this results in the “current sheet
theorem.”[7]
dI
H1 ðzÞ H2 ðzÞ ¼ : (5.31)
dz
In addition to determining fields, this result has been used for calculations of
magnetic stored energy and Lorentz forces.[8]
Figure 5.4 The sheet theorem.

5.5 cos ϕ current sheets

We consider here two examples of using complex methods to study the properties
of current sheets that have a cos ϕ azimuthal current distribution.
Example 5.3: field from cos ϕ current distribution using contour integration
Consider a closed circular sheet of radius a. We have
dI
¼ I0 cos ϕ
dϕ
z ¼ aeiϕ
eiϕ þ eiϕ z þ z
cos ϕ ¼ ¼ :
2 2a
þ
i I0 z þ z dz
H ðzo Þ ¼
4πa z zo i z
þ
I0 1 z
¼ þ dz (5.32)
4πa z zo zðz zo Þ
I0
¼ ½I1 þ I2 :
4πa
It follows from Cauchy’s theorem that the first integral

þ
dz 2πi if zo < a
I1 ¼ ¼ :
z zo 0 if zo > a
Using the method of partial fractions,[9] the denominator of the second integral can
be written
1 A B
¼ þ
zðz zo Þ z z zo
1 ¼ ðz zo ÞA þ zB:
Equating powers of z, we find that
1
A¼
zo
1
B¼ :
zo
Then we can write
I2 ¼ I3 þ I4 ;
5.5 cos ϕ current sheets 121
where
þ
1 z
I3 ¼ dz
zo z
ð
ia 2π iϕ
¼ e dϕ ¼ 0
zo 0
and
þ
1 z
I4 ¼ dz: (5.33)
zo z zo
Using z ¼ a2 =z ; we can write this as
þ
a2 dz
I4 ¼
zo zðz zo Þ
þ þ
a2 1 dz 1 dz
¼ þ :
zo zo z zo z zo
For zo inside the contour, the factor in square brackets vanishes because of the
residue theorem and I4 ¼ 0: Then from Equation 5.32,
H ðzo Þ ¼ Hx iHy
I0 I0
¼ ½2πi þ 0 ¼ i :
4πa 2a
From this, we see that the field inside the current sheet is
Hx ¼ 0
I0 (5.34)
Hy ¼ :
2a
The field is only in the vertical direction and has constant magnitude everywhere
inside the sheet in agreement with Equation 4.36.
For zo outside the contour, we have8
þ
z
I4 ¼ dz
zo ðz zo Þ
ð
ia2 2π 1
¼ dϕ
zo 0 a e zo
iϕ
a2
¼ 2πi:
z2o
8
GR 2.313.1.
Substituting these results back into Equation 5.32, the field outside the contour is

I0 a2 I0 a
H ðzo Þ ¼ 0 2 2πi ¼ i 2 :
4πa zo 2zo
If we write zo ¼ x þ iy and multiply the numerator and denominator by ðzo Þ2, we find
that
I0 a I0 a
H ðzo Þ ¼ 4
xy i 4 ðx2 y2 Þ:
r 2r
Equating real and imaginary parts, we find the Cartesian field components outside the
sheet are
I0 a I0 a
Hx ¼ 4
xy ¼ 2 sin 2ϕ
r 2r (5.35)
I0 a 2 I0 a
Hy ¼ 4 ðx y Þ ¼ 2 cos 2ϕ:
2
2r 2r
On the midplane (y = 0), H is positive, along the y direction, and falls off with distance
like 1=x2 .
Example 5.4: field from cos ϕ current distribution using the sheet theorem
Assume again that we have a circular sheet with radius a. The current elements are
located at
z ¼ aeiϕ
dz ¼ izdϕ;
so we have
dI dI dϕ I0
¼ ¼ i cos ϕ
dz dϕ dz z
iϕ
I0 e þ eiϕ
¼ i
z 2
I0 z a

¼ i þ :
z a z
Using the current sheet theorem, Equation 5.31,

I0 I0 a
H1 ðzÞ H2 ðzÞ ¼ i þ :
2a 2z2
The field inside the sheet Hin must be finite at z ¼ 0 and for current in the positive
z direction in the first quadrant of the circle, the field must go in the negative
y direction. Therefore we identify H2 with Hin and get
5.6 Green’s theorems in the complex plane 123
I0
Hin ðzÞ ¼ Hin;x þ iHin;y ¼ i :
2a
Equating real and imaginary parts, we find that
Hin;x ¼ 0
I0 (5.36)
Hin;y ¼
2a
in agreement with Equation 5.34. We identify the field exterior to the current sheet
Hext with H1 in the sheet theorem.
I0 a
Hext;x i Hext;y ¼ i
2z2
5.6 Green’s theorems in the complex plane

So far we have examined the fields due to current filaments and current sheets.
We next want to proceed to the case of conductors with finite cross-sectional areas.
However, before doing that, we need to review some important theorems that
allows us to replace two-dimensional integrations over the conductor surface
with contour integrals around the boundary of the surface. Besides the practical
importance of reducing computation times in numerical calculations, this allows us
to make use of some powerful results from the theory of complex contour
integration.
Recall from Equation 3.79 that Green’s theorem in the plane is
ðð þ
∂Q ∂P
dx dy ¼ ðP dx þ Q dyÞ;
∂x ∂y
where Pðx; yÞ and Qðx; yÞ are continuous functions with continuous partial
derivatives in a region R that is bounded by a curve C. Define the complex
function
Fðz; z Þ ¼ Pðx; yÞ þ iQðx; yÞ:
Using Equation 5.9, the derivative of F can be written as

∂F ∂P ∂Q ∂P ∂Q
2 ¼ þi þ : (5.37)
∂z ∂x ∂y ∂y ∂x
The closed integral of F around C is
þ þ þ
F dz ¼ ðP dx Q dyÞ þ i ðQ dx þ P dyÞ:
Applying Green’s theorem in the plane, we have

þ ðð
∂P ∂Q ∂P ∂Q
F dz ¼ i þi þ dx dy:
∂x ∂y ∂y ∂x
Replacing the integrand on the right-hand side using Equation 5.37, we find the first
complex Green’s theorem.[1]
ð þ
∂F 1
dS ¼ F dz (5.38)
∂z 2i
Following similar arguments, we have

∂F ∂P ∂Q ∂P ∂Q
2 ¼ þ þi þ
∂z ∂x ∂y ∂y ∂x
and
þ ðð
∂P ∂Q ∂P ∂Q
F dz ¼ i þ þi þ dx dy:
∂x ∂y ∂y ∂x
After substitution, we obtain the second complex Green’s theorem.[1]
ð þ
∂F 1
dS ¼ F dz : (5.39)
∂z 2i
5.7 Field from a block conductor

We next want to consider the case of a block conductor, which we define as one
with finite cross-sectional area. If we consider the conductor block as made up from
an array of current filaments, we can use Equation 5.21 and express the field as
ð
i σ
H ¼ dS; (5.40)
2π zo z
where σ is the current density in the block. If we assume the current density is
constant, we can rewrite this as
ð
iσ dS
H ¼ : (5.41)
2π z zo
Powerful methods have been developed that allow the fields from block conductors
to be evaluated using contour integration.[1] Consider the Green’s theorem,
Equation 5.38. For our application, the integrand of the surface integral is asso-
ciated with the expression for the magnetic field in Equation 5.41. The integral has
5.7 Field from a block conductor 125
Figure 5.5 Contour for the Green’s theorem calculation.
a singularity for the case when zo is inside C, as shown in Figure 5.5. We can isolate
the singularity by constructing a small circular contour C1 around it. Then we can
use Green’s theorem in the region between the two contours to transform the
surface integral into a contour integration. However, this requires evaluation on
both contours C and C1 . Halbach proposed adding a constant term to the function
F in Green’s theorem that is (a) analytic in R and (b) makes the contour integration
around C1 vanish.[1] He assumed that F could be written as the product of two
functions F1 and F2 with the property
∂F ∂F2 ðz Þ
¼ F 1 ðzÞ ; (5.42)
∂z ∂z
where F1 contains the singularity. Then F must have the form
Fðz; z Þ ¼ F1 ðzÞ½F2 ðz Þ F2 ðz o Þ: (5.43)
Following this procedure, we define

iσ 1
F1 ðzÞ ¼
2π z zo
F2 ðz Þ ¼ z :
Then for use in Green’s theorem, we have

iσ 1
F¼ ðz z o Þ
2π z zo
and
∂F iσ 1

¼ ;
∂z 2π z zo
which is the integrand from Equation 5.41. Applying Green’s theorem, we find that
þ
1 iσ 1
H ¼ ðz z o Þdz;
2i 2π z zo
which simplifies to [1, 10]
þ
σ z z o
H ¼ dz: (5.44)
4π z zo
Let us confirm that Equation 5.44 does indeed vanish for the circular contour C1.
Let
z zo ¼ reiθ :
Then for the contour C1,
þ ð 2π
z z o
dz ¼ ir eiθ dθ ¼ 0:
z zo 0
Thus we can ignore the contours around isolated singularities inside the conductor
region and only evaluate Equation 5.44 on the outer boundary of the conductor.
Other quantities of interest can also be conveniently expressed in terms of
contour integrals. For example, the area A of a current block is given by [10, 11]
þ
1
A¼ z dz: (5.45)
2i
Expressions have also been derived for the stored energy.[1, 12]
5.8 Block conductor examples

We consider three examples of using Equation 5.44 to find the field of a block
conductor. The first example, the cylindrical conductor, was treated already in
Chapter 1 using the Ampère law. Even though the calculation presented here is
considerably more complicated, we carry it out to demonstrate some of the techni-
ques involved and to compare with a result where we know the answer. The other two
examples cannot be computed straightforwardly using the Ampère law.
Example 5.5: field of a solid cylindrical conductor

Assume we have a solid cylindrical conductor with radius a, as shown in Figure 5.6.
Let
z ¼ aeiϕ
zo ¼ reiθ :
Case 1: zo inside the conductor 127
Figure 5.6 Solid cylindrical conductor.

ð
σ 2π aeiϕ reiθ
H ¼ i aeiϕ dϕ
4π 0 aeiϕ reiθ (5.46)
σa
¼i ½a I1 reiθ I2 ;
4π
where
2ðπ
dϕ
I1 ¼
aeiϕ reiθ
0
(5.47)
ð
2π
eiϕ
I2 ¼ dϕ:
aeiϕ reiθ
0
Case 1: zo inside the conductor

When zo is inside the conductor, we can use theorems from complex analysis
to evaluate the integrals. Define u ¼ eiϕ and β ¼ zo =a. Then
þ
1 du
I1 ¼ : (5.48)
ia ðu βÞu
We would like to convert the denominator into a simple pole so that we can use
the residue theorem to evaluate the integral. To do this, expand the denominator
using the method of partial fractions. Then we can write Equation 5.48 as
þ þ
1 1 du 1 du
I1 ¼ þ :
ia β u0 β uβ
Applying the residue theorem, we obtain

1 1 1
I1 ¼ 2πi þ ¼ 0:
ia β β
Turning next to I2 ,
þ
1 du
I2 ¼
ia ðu βÞ
we apply the residue theorem and find

1 2π
I2 ¼ 2πi ð1Þ ¼ :
ia a
Returning now to Equation 5.46,
σa 2π
H ¼ i ðr eiθ Þ
4π a
σr
¼ i ½cos θ i sin θ:
2
Thus the field inside the conductor is

σr
Hx ¼ sin θ
2
σr (5.49)
Hy ¼ cos θ:
2
Case 2: zo outside the conductor

In this case, there are no singularities inside the contour, so we can treat I1 and
I2 as ordinary integrals. Performing the first integration gives9
1 2π
I1 ¼ iϕ lnðreiθ þ aeiϕ Þ 0 :
ire iθ
The logarithm term cancels because it has the same value at 0 and 2π. Thus we
find that
2π
I1 ¼ :
reiθ
9
GR 2.313.1.
Case 2: zo outside the conductor 129
For the integral I2 , let β ¼ r=a and u ¼ eiϕ . Then

þ
1 du
I2 ¼
ia u βeiθ
1 2π
¼ ln eiϕ βeiθ 0 :
ia
The second term in the logarithm is constant and the first term has the same
value at the two limits. Therefore, I2 ¼ 0, and Equation 5.46 gives
σa2
H ¼ i
2reiθ
σa2
¼ i ½cos θ isin θ:
2r
Equating real and imaginary parts, we find the field outside the conductor is
σa2
Hx ¼ sin θ
2r
(5.50)
σa2
Hy ¼ cos θ:
2r
The field of an elliptical block conductor has also been found using similar
methods.[12, 13]
Example 5.6: field outside a rectangular conductor

Assume we have a rectangular conductor oriented at an angle θ with respect to the
x axis, as shown in Figure 5.7. We look for the field at the observation point zo.
In terms of the variables (z, z*), a straight line from the vertex n to vertex n þ 1 has the
equation [1, 10]
z z
z ¼ z n þ Δz
n
; (5.51)
Δz
where
Δz ¼ znþ1 zn : (5.52)
Since the rectangle has four sides and has to close, we identify z5 ¼ z1. For the side
beginning with vertex 1, we define
Δz aeiθ
β1 ¼ ¼ iθ ¼ e2iθ ;
Δz ae
Figure 5.7 Rectangular conductor.
where a is the length of side 1. The second equation comes from considering z1 as the
origin of a line to z2 in polar coordinates. Note that β is a constant because it is defined
in terms of fixed z locations. In a rectangle, the angles of the other sides with respect
to the x axis increase by 90° at each of the vertices. Thus for the side from vertex 2 to
vertex 3,
β2 ¼ e2iðθþπ=2Þ ¼ eiπ e2iθ ¼ β1 :
Similarly we find,
β3 ¼ β1
β4 ¼ β1 :
ð z2
σ ½z 1 þ β1 ðz z1 Þ z o
H ¼ dz þ
4π z1 z zo
with similar expressions for the remaining three sides. Defining the constant
αn ¼ z n βn zn z o ;
we can write
σ
H ¼ ½α1 I1 þ β1 I2 þ : (5.53)
4π
For points zo outside the contour,
ð z2
dz
I1 ¼
z1 z zo
¼ ½lnðz zo Þzz21
Case 2: zo outside the conductor 131
and10
ð z2
z
I2 ¼ dz
z1 z zo
¼ ½z þ zo lnðz zo Þzz21 :
Substituting into Equation 5.53,

σ z2 zo
H ¼ α1 ln þ β1 ½z2 þ zo lnðz2 zo Þ z1 zo lnðz1 zo Þ þ
4π z1 zo

σ z2 zo z2 zo
¼ α1 ln þ β1 zo ln þ β1 ðz2 z1 Þ þ
4π z1 zo z1 zo

σ z2 zo
¼ ½α1 þ β1 zo ln þ β1 ðz2 z1 Þ þ :
4π z1 zo
Writing out the third term for all four sides gives
β1 ½ðz2 z1 Þ ðz3 z2 Þ þ ðz4 z3 Þ ðz1 z4 Þ ¼ 2β1 ½z1 þ z2 z3 þ z4 :
For a rectangle, the directed line segments
z4 z3 ¼ ðz2 z1 Þ;
so this term cancels. Thus we find the field at zo due to the rectangular conductor
block is [5, 10]
σ X4
H ¼ hn
4π n¼1
(5.54)
znþ1 zo
hn ¼ ½ðzn zo Þ βn ðzn zo Þ ln :
zn zo
Example 5.7: on-axis field for annular sector conductor

For our last example, consider a conductor with the shape of an annular sector, as
shown in Figure 4.11. We look for the field at the center of the circular arcs. Thus we
have
z ¼ reiϕ
zo ¼ 0
and the contour in Equation 5.44 can be broken into the four parts.
10
GR 2.112.1.
(ð ð ϕ2
σ r2
reiϕ1 iϕ1 r2 eiϕ
H ð0Þ ¼ e dr þ r2 i eiϕ dϕ
4π r1 reiϕ1 ϕ1 r2 eiϕ
ð r1 ð ϕ1 )
reiϕ2 iϕ2 r1 eiϕ
þ e dr þ r1 i eiϕ dϕ :
r2 reiϕ2 ϕ2 r1 eiϕ
Simplifying and performing the integrals, we get

(ð ð ϕ2 ð r1 ð ϕ1 )
σ r2
H ð0Þ ¼ eiϕ1 dr þ i r2 eiϕ dϕ þ eiϕ2 dr þ i r1 eiϕ dϕ
4π r1 ϕ1 r2 ϕ2
σ
¼ ðr2 r1 Þðeiϕ2 eiϕ1 Þ:
2π
Expanding the exponentials, we find that the field of the annular sector conductor is
σ
H ð0Þ ¼ ðr2 r1 Þ½ðcos ϕ2 cos ϕ1 Þ iðsin ϕ2 sin ϕ1 Þ; (5.55)
2π
which agrees with Equation 4.50. Note that the field strength is proportional to the
radial thickness.
5.9 Field from image currents

We now consider the magnetic field produced by a current distribution in the
presence of infinite permeability iron. The case of a filament near a planar iron
surface is shown in Figure 5.8. The current in the filament induces image currents
on the surface of the iron, which can be represented by an equivalent image
filament inside the iron. We have seen in Chapter 2 that the image current is in the
same direction as the conductor filament and is located the same distance from
the iron surface as the conductor filament. The field of the image filament is
given by
Figure 5.8 Line current near an iron slab.

5.9 Field from image currents 133
I 1
HI ðzo Þ ¼ i : (5.56)
2π zI zo
The location of the image filament is
zI ¼ d þ ðd xÞ þ i y
¼ 2d z :
For a uniform distribution of current, we have
ð
σ dS
HI ðzo Þ ¼ i :
2π 2d z zo
To convert this surface integral to a contour integral, we use the complex Green’s
theorem, Equation 5.39. Choosing
iσ z
F¼ ;
2π 2d z zo
we find that the contribution to the field from the image current in a planar iron
surface is
þ
σ z
HI ðzo Þ ¼ dz : (5.57)
4π 2d z zo
We are also interested in the image currents near a circular iron surface at radius
R, as shown in Figure 5.9. From Chapter 2, we know the image current is in the
same direction as the conductor current. If ρ is the distance of the conductor
filament from the center of the circle, then the image filament is a distance R2 =ρ
from the center. Thus we have
z ¼ ρeiϕ
R2 iϕ R2
zI ¼ e ¼ :
ρ z
Figure 5.9 Line current near a circular iron cavity.

For a sheet conductor, the contribution of the images in a circular iron cavity to the
field is a sum over the corresponding image currents. Using Equation 5.27, we
obtain
ð
i KðzÞ
HI ðzo Þ ¼ dz
2π zI zo
ð (5.58)
i Kz
¼ dz;
2π R2 zo z
where KðzÞ ¼ dI=dz.
Example 5.8: image field for cos ϕ current sheet in an iron cavity
Let us examine the image field at the origin for a closed circular sheet with radius
a and a cos ϕ angular current distribution. When zo = 0, the integrand does not have
a singularity. Applying Equation 5.58,
ð 2π
i I0 cos ϕ aeiϕ
HI ð0Þ ¼ izdϕ
2π 0 iz R2
ð
iI0 a 2π cos ϕ
¼ dϕ
2πR2 0 eiϕ
ð
iI0 a 2π eiϕ þ eiϕ
¼ dϕ:
4πR2 0 eiϕ
After the integration, we find the contribution of the image field at the origin is
I0 a
HI ð0Þ ¼ i ¼ HIx iHIy ;
2R2
which gives the field components
HIx ¼ 0
I0 a (5.59)
HIy ¼ :
2R2
The contribution of the image field is in the same direction that we saw in
Equation 5.34 for the field of the conductor. The enhancement of the field due to
the presence of the iron is
H ð0Þ þ HI ð0Þ
Eð0Þ ¼
H ð0Þ
(5.60)
a2
¼ 1þ 2:
R
This shows that the iron cavity can contribute up to a factor of 2 to the field at the
origin.
For a block conductor with constant current density, the image current in circular
iron is
ð
iσ z
HI ðzo Þ ¼ dS:
2π R2 zo z
Using Equation 5.39 for Green’s theorem and choosing
iσ zz
F¼ ;
2π R2 zo z
we find that the field due to the image current in the circular iron is [1]
þ
σ zz
HI ðzo Þ ¼ dz : (5.61)
4π R2 zo z
Example 5.9: on-axis image field in circular iron for annular sector conductor
Consider an annular sector conductor extending from radius r1 to r2 inside an iron
cavity of radius R. The image field at the origin is given by Equation 5.61 with zo ¼ 0.
þ
σ
HI ð0Þ ¼ zz dz
4πR2
( ð r2 ð ϕ2 ð r1 ð ϕ1 )
σ
¼ e iϕ1 iϕ
r dr ir2 e dϕ þ e
2 3 iϕ2
r dr ir1 e dϕ :
2 3 iϕ
4πR2 r1 ϕ1 r2 ϕ2
Evaluating the integrals and simplifying gives the field contribution due to the iron.
σ
HI ð0Þ ¼ ðr3 r31 Þ½ðcos ϕ2 cos ϕ1 Þ iðsin ϕ2 sin ϕ1 Þ: (5.62)
6πR2 2
Note that this expression has the same sign and angular dependence as the field from
the conductor given in Equation 5.55. The presence of the iron gives the enhancement
factor at the origin [14]
r22 þ r1 r2 þ r21
Eð0Þ ¼ 1 þ : (5.63)
3R2
5.10 Multipole expansion

Since the magnetic potential, Equation 5.3, is an analytic function, it can be
expanded in a power series
X
∞
Wðzo Þ ¼ wn zno :
n¼0
The magnetic field can then be expressed as

i dW
H ðzo Þ ¼
μ0 dzo
X ∞
in
¼ o :
wn zn1
n¼1
μ0
Redefining the coefficients, we write the field as the power series

X
∞
H ðzo Þ ¼ o :
cn zn1 (5.64)
n¼1
The field can also be expressed in terms of the integral in Equation 5.41.
ð
i σ
H ðzo Þ ¼ dS
2π z zo
ð
i σ
¼ dS:
2π z 1 zo
z
Expand the factor in the denominator in a geometric series.
ð
i σ zo zo 2
H ðzo Þ ¼ 1þ þ þ dS
2π z z z
This series converges for observation points inside the magnet aperture up to the
closest conductor. Equating this expression with Equation 5.64 gives
X ð ∞ n1
∞
i σX zo
cn zn1
o ¼ dS
n¼1
2π z n¼1
z
∞ ð n1
i X σ zo
¼ dS:
2π n¼1 z z
The zo factor cancels from both sides of the equation. Then matching term by term,
we find
ð
iσ n
cn ¼ z dS: (5.65)
2π
We can convert this surface integral into a contour integral by using the Green’s
theorem, Equation 5.39, with
iσ z1n
F¼ :
2π 1 n
Thus Equation 5.65 becomes [1]

þ
1 iσ z1n
cn ¼ dz
2i 2π 1 n
þ (5.66)
σ
¼ z1n dz
4πðn 1Þ
for n > 1. For the case n = 1, we return to Equation 5.65 and find
ð
iσ 1
c1 ¼ dS:
2π z
This time we use the Green’s theorem Equation 5.38 with
iσ
F¼ z
2πz
to find that [1]
þ
σ z
c1 ¼ dz: (5.67)
4π z
Example 5.10: multipoles for an annular sector conductor

We consider an annular sector conductor with radius between r1 and r2 that has
constant current density σ. Let z ¼ reiϕ : For multipoles with n > 1, we have using
Equation 5.66
(ð ð ϕ2
σ r2
iϕ1 1n iϕ1
cn ¼ ðre Þ e dr i ðr2 eiϕ Þ1n r2 eiϕ dϕ
4πðn 1Þ r1 ϕ1
ð r1 ð ϕ1 )
þ ðreiϕ2 Þ1n eiϕ2 dr i ðr1 eiϕ Þ1n r1 eiϕ dϕ :
r2 ϕ2
After performing the integrations and simplifying the algebraic results, we find
that
σ
cn ¼ ðr2n r12n Þðeinϕ2 einϕ1 Þ: (5.68)
2πnð2 nÞ 2
Because of the factor in the denominator, this relation cannot be used when n = 2. For
that case, we return to Equation 5.66 and find
þ
σ
c2 ¼ z1 dz :
4π
For the annular sector, performing the integrals and summing terms, we find the
quadrupole multipole is

σ r2
c2 ¼ ln ðe2iϕ2 e2iϕ1 Þ: (5.69)
4π r1
We can get the n = 1 term from Equation 5.67. The dipole multipole is
σ
c1 ¼ ðr2 r1 Þðeiϕ2 eiϕ1 Þ: (5.70)
2π
Errors in the construction of magnet coils can lead to the introduction of

additional unwanted multipole contributions to the field.[1, 15] These errors can
include left-right and up-down asymmetries in the shape of the coils, displace-
ments, rotations, and errors in the excitation currents.
5.11 Field due to a magnetized body

We next look at the magnetic field produced by a magnetized body. This is the case,
for example, for a permanent magnet with net magnetization in the x-y plane.
Consider a pair of parallel filaments with currents flowing in opposite directions
located a distance d apart, as shown in Figure 5.10. There is a net field component
in the x-y plane, oriented perpendicular to the axis connecting the two filaments.
Such an arrangement is known as a current doublet.[16, 17] The field for the two
filaments is

iI 1 1
H ðzo Þ ¼ :
2π z2 zo z1 zo
Let
d ¼ z2 z1 ¼ jdjeiα
zd ¼ ½ðz1 þ z2 Þ;
where α is the angle between d and the x axis. Substituting, we find
2 3
iI 6
6 d 7
7:
H ðzo Þ ¼
2π 4 25
d
ðzd zo Þ2
4
Recall that the magnetic dipole moment is
m ¼ IA ¼ I l d;
5.11 Field due to a magnetized body 139
Figure 5.10 Model for a magnetized body.
where l is a unit distance along the z direction. Define m0 as the magnetic moment
per unit length. Then in the limit as d → 0,
I d→m0
d2
→0
4
and the field at zo due to the doublet at z is
i m0 eiα
H ðzo Þ ¼ :
2π ðz zo Þ2
We now want to express the field in terms of the magnetization M. The direction
of m0 is rotated by π/2 with respect to the direction of d. Let us define β to be the
direction of M with respect to the x axis.
π
β¼α
2
eiα ¼ eiβ eiπ=2 ¼ ieiβ :
Then summing up all the magnetic moments in the magnetized body, we have [18]
ð
1 M
H ðzo Þ ¼ dS: (5.71)
2π ðz zo Þ2
If the magnetization is constant in the body, we can convert this to a contour

integral by using the Green’s theorem, Equation 5.39. Defining
M 1
F¼ ;
2π z zo
we find that the field due to the magnetized body is [18]

þ
M dz
H ðzo Þ ¼ : (5.72)
4πi z zo
Example 5.11: triangular block with constant magnetization

Consider a triangular block of magnetic material with vertices fz1 ; z2 ; z3 g. Assume
the magnetization has the constant value M. Define
Δzn ¼ znþ1 zn :
Since the triangle is closed, z4 ¼ z1 . The slopes of the sides are
Δz n
βn ¼ ;
Δzn
so we can change the integration variable in Equation 5.72 from z* to z for side
n through the relation
dz ¼ βn dz:
The field produced by the block is

ð z2
M β1
H ðzo Þ ¼ dz þ
4π i z1 z zo
M
¼ ½β lnðz2 zo Þ β1 lnðz1 zo Þ þ :
4π i 1
Collecting terms, the field of the triangular magnetized block is
M
H ðzo Þ ¼ fðβ3 β1 Þlnðz1 zo Þ þ ðβ1 β2 Þlnðz2 zo Þ
4πi
þðβ2 β3 Þlnðz3 zo Þg: (5.73)
5.12 Force
The vector force dF on a current filament is
! ! !
dF ¼ I dl B :
If the filament is directed along the z direction, B is in the x-y plane, and so is F.
The force can then be written as the complex variable F ¼ Fx þ iFy , where
dFx ¼ μ0 IHy dz
dFy ¼ μ0 IHx dz:
Define f to be the force per unit length in the z direction.

dF
f ¼ ¼ iμ0 I H
dz
For a distributed current distribution, we can generalize this as
ð
f ¼ iμ0 σ HdS: (5.74)
Using Equation 5.11 for σ, we have

ð
∂H
f ¼ 2μ0 H dS:
∂z
To express this as a contour integral, use the complex Green’s theorem,
Equation 5.39, with
F ¼ μ0 H 2 ;
which gives [1]
þ
iμ0
f ¼ H 2 dz : (5.75)
2
This shows that the transverse force per unit length is proportional to the square of
the magnetic field intensity. Examples of complex force calculations can be found
in references.[1, 19]
5.13 Conformal mapping

Operating on a complex variable z with some function f
w ¼ f ðzÞ
produces another complex variable w. This can be interpreted as a mapping from
the z plane onto another w plane. Suppose that two curves in the z plane intersect at
a point with the angle θ between them. A mapping is called conformal if the two
corresponding curves in the w plane also intersect with the same angle θ between
them. If f ðzÞ is an analytic function with df =dz≠0 inside a region R, then the
mapping is conformal. Conformal mappings have the property that the function in
the w plane is also analytic, so the real and imaginary parts of the mapped function
are solutions of the Laplace equation.
Conformal mapping can frequently be used to transform a problem with com-
plicated boundaries in the z plane, for example, into a simpler problem in the upper
half-plane or the interior of the unit circle in the w plane. Once the solution is found
for the problem in the w plane, an inverse mapping z ¼ gðwÞ can be used to obtain
the solution to the original problem. The theory of conformal mapping is a major
subject in its own right. We only have space here to briefly introduce the subject
and present a few examples. Fortunately, approximately half the book by Binns and
Lawrenson is devoted to using conformal mapping in the solution of electric and
magnetic field problems.[20] The interested reader can find many useful examples
there.
The bilinear transformation combines the operations of translation, rotation,
stretching, and inversion.[21]
αz þ β
w¼ ;
γz þ δ
where α, β, γ, and δ are complex numbers with the property that

αδ βγ ≠ 0:
This transformation can map circles and lines in the z plane into circles and lines in
the w plane. It can be used, for example, to map a pair of separated circles to
concentric circles. The bilinear transformation has the property that a quantity
known as the cross-ratio is conserved.
ðw w1 Þðw2 w3 Þ ðz z1 Þðz2 z3 Þ
¼ (5.76)
ðw w3 Þðw2 w1 Þ ðz z3 Þðz2 z1 Þ
This expression can be used to create a transformation that maps three given points
in the z plane to three corresponding points in the w plane. An important bilinear
transformation that maps any point zo in the upper half of the z plane into the
interior of the unit circle in the w plane is given by [22]

iθ0 z zo
w¼e : (5.77)
z z o
The points on the x axis are mapped to the boundary of the circle.
Example 5.12: line current in an iron cavity

Suppose we have a line current at the point w1 inside a circular cavity with unit radius
that is made from infinitely permeable iron, as shown in Figure 5.11. We use
Equation 5.77 to map between the physical situation in the w plane and the upper
half of the z plane. To determine the two unknown constants θ0 and zo , we associate
the points
Figure 5.11 Line current in an iron cavity.
z ¼ i↔w ¼ 0
z ¼ ∞↔w ¼ 1;
which requires that

zo ¼ i
eiθ0 ¼ 1:
This gives the specific mapping function between the planes
iz
w¼ :
iþz
The known line current at w1 maps to a line current at z1 in the z plane and the circular
iron boundary maps to an iron plane along the real axis in the z plane. The mapping
between the two line currents is
i x1 iy1
w1 ¼ u1 þ iv1 ¼ :
i þ x1 þ iy1
Normalizing the denominator, we find
½1 x21 y21 þ i ½2x1

u1 þ iv1 ¼ :
x21 þ ð1 þ y1 Þ2
The real and imaginary parts of this equation can be solved for x1 and y1 as
½2v1 þ i ½1 u21 v21

x1 þ iy1 ¼ :
u21 þ v21 þ 2u1 þ 1
In the z plane, we know there is an image current below the iron plane at the location
z2 ¼ z 1 . We can then use the mapping function to find the location w2 of the image
current in the w plane. After some algebraic simplifications, we find that
Figure 5.12 Schwarz-Christoffel transformation.
v1 þ iðu1 þ 1Þ
w2 ¼ u2 þ iv2 ¼
v1 þ iðu21 þ v21 þ u1 Þ
u1 þ iv1
¼ :
u21 þ v21
Representing w1 and w2 in polar coordinates, we find that
1
r2 ¼
r1
θ2 ¼ θ1 ;
which agrees with the result from the method of images.
Suppose that the boundary of some region in the z plane is made up of a series of
straight line segments, as shown in Figure 5.12. The line segments meet at the
vertices z1 ; z2 , . . . It is possible to map this boundary to the real axis in the w plane
by using the Schwarz-Christoffel transformation,[23] which takes the form of the
differential equation
dz
¼ Gðw u1 Þα1 =π1 ðw u2 Þα2 =π1 ðw un Þαn =π1 ; (5.78)
dw
where G is a complex constant and the αi are the interior angles. The points
u1 ; u2 , . . . on the real axis of the w plane correspond to the vertices in the
z plane. The interior of the figure in the z plane maps to the upper half of the
w plane.
Example 5.13: potential of a line current near the corner of two perpendicular
planes
Consider a line current near the perpendicular intersection of two infinitely permeable
plane surfaces, as shown in Figure 5.13. We solve the problem by using the Schwarz-
Christoffel transformation, which in this case takes the form
Figure 5.13 Line current near a corner. ABC and abc lie on infinitely permeable
boundary surfaces. The line current is at z1 .
dz
¼ Gðw u1 Þ1=2 ;
dw
since the vertex angle α1 ¼ π=2. Integrating this equation, we find
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z ¼ 2G w u1 þ H;
where H is another complex constant. Solving for w, we get
z2 2H z þ H 2
w u1 ¼ :
4G2
Breaking this equation into real and imaginary parts, leads to
x2 y2 2H x þ H 2 þ 2iðx y H yÞ
u u1 þ iv ¼ : (5.79)
4G2
We choose three points A, B, C on the boundary in the z plane and demand that they
correspond to three points a, b, c along the real axis in the w plane according to the
following prescription:
A : x ¼ 0; y ¼ 1↔a : u ¼ 1; v ¼ 0
B : x ¼ 0; y ¼ 0↔b : u ¼ 0; v ¼ 0
C : x ¼ 1; y ¼ 0↔c : u ¼ 1; v ¼ 0
Applying these constraints to Equation 5.79, we find that
u1 ¼ 0
H¼0
4G2 ¼ 1
and the resulting transformation equation is
w ¼ z2 :
Figure 5.14 Image current in the w plane.
Figure 5.15 Line current and three images in the z plane.
In the w plane, the potential is due to the line current w1 and the image current due to
the plane boundary of the infinitely permeable material, as shown in Figure 5.14.
The potential is given by
μ0 I
Wðwo Þ ¼ ½lnðwo w1 Þ þ lnðwo w 1 Þ:
2π
Transforming back to the z plane, we have
μ0 I
Wðzo Þ ¼ ½lnðz2o z21 Þ þ lnðz2o z 2
1 Þ
2π
μ I
¼ 0 ln½ðzo z1 Þðzo þ z1 Þ þ ln½ðzo z 1 Þðzo þ z 1 Þ
2π
μ I
¼ 0 lnðzo z1 Þ þ lnðzo þ z1 Þ þ lnðzo z 1 Þ þ lnðzo þ z 1 Þ :
2π
This shows that the potential in the z plane is due to the physical line current at z1
together with three image currents,[24] as shown in Figure 5.15. The four currents lie
on a circle centered at the corner of the iron surfaces.
5.14 Integrated potentials 147
5.14 Integrated potentials

Suppose that Φ(r, θ, s) is a scalar potential describing some three-dimensional
magnetic configuration. Assume the conductors have a finite extent in the
s direction, so that Φ vanishes as s → ±∞. Define
ð∞
Eðr; θÞ ¼ Φðr; θ; sÞ ds:
∞
Taking the derivative with respect to r gives

ð∞ ð∞
∂E ∂Φ
¼ ds ¼ Br ds; (5.80)
∂r ∞ ∂r ∞
while the derivative with respect to θ yields

ð∞ ð∞
1 ∂E 1 ∂Φ
¼ ds ¼ Bθ ds: (5.81)
r ∂θ ∞ r ∂θ ∞
We can likewise define Aðr; θ; sÞ as the vector potential describing the same three-
dimensional magnetic configuration. Since the conductors have a finite extent in
the s direction, As also vanishes as s→ ±∞. Define
ð∞
Fðr; θÞ ¼ As ðr; θ; sÞ ds:
∞
Considering the integral of Br , we find that

ð∞ ð
!
Br ds ¼ ðr A Þr ds
∞
ð
1 ∂As ∂Aθ
¼ ds
r ∂θ ∂s
ð ∞
1 ∂
¼ As ds Aθ :
r ∂θ ∞
Assuming that Aθ has the same value at ±∞, we find that

ð∞
1 ∂F
Br ds ¼ : (5.82)
∞ r ∂θ
Similarly, the integral of Bθ
ð∞ ð
∂Ar ∂As
Bθ ds ¼ ds
∞ ∂s ∂r
∞ ð
∂
¼ Ar As ds;
∞ ∂r
so that
ð∞
∂F
Bθ ds ¼ : (5.83)
∞ ∂r
Equating the expressions for the integrated values of Br and Bθ , we find that
1 ∂F ∂E
¼
r ∂θ ∂r
∂F 1 ∂E
¼ :
∂r r ∂θ
These two equations have the same form as the Cauchy-Riemann equations in polar
coordinates. Thus F and E represent the real and imaginary parts of the analytic
potential function
WðzÞ ¼ Fðr; θÞ þ iEðr; θÞ:
This potential can be used to describe the influence of a magnet end on the field
quality of a long magnet.[25]
References
[1] K. Halbach, Fields and first order perturbation effects in two dimensional conductor
dominated magnets, Nuc. Instr. Meth. 78:185, 1970.
[2] K. Miller, Introduction to Advanced Complex Calculus, Dover, 1970, p. 66.
[3] M. Spiegel, Complex Variables, Schaum’s Outline Series, McGraw-Hill, 1964, p. 7,
69–70, 83.
[4] R. Beth, Complex methods for three-dimensional magnetic fields, IEEE Trans. Nuc.
Sci. 18:901, 1971.
[5] R. Beth, Complex representation and computation of two-dimensional magnetic
fields, J. Appl. Phys. 37:2568, 1966.
[6] R. Beth, Elliptical and circular current sheets to produce a prescribed internal field,
IEEE Trans. Nuc. Sci. 14:386, 1967.
[7] R. Beth, Some extensions of complex methods for two-dimensional fields, Proc. 6th
Int. Conf. on High Energy Accelerators, Cambridge Electron Accelerator Lab,
Cambridge, MA, 1967, p. 387.
[8] F. Toral, et al., Further developments on Beth’s current sheet theorem: computation of
magnetic field, energy and mechanical stresses in the cross section of particle accel-
erator magnets, IEEE Trans. Appl. Superconductivity 14:1886, 2004.
[9] M. Protter & C. Morrey, College Calculus with Analytic Geometry, Addison-Wesley,
1964, p. 465–468.
[10] R. Beth, Evaluation of current produced two-dimensional magnetic fields, J. Appl.
Phys. 40:4782, 1969.
[11] M. Spiegel, op. cit., p. 114.
[12] R. Beth, Analytic design of superconducting multipolar magnets, Proc. 1968 Summer
Study on Superconducting Devices and Accelerators, Brookhaven National
Laboratory, p. 843–859.
References 149
[13] R. Beth, An integral formula for two-dimensional fields, J. Appl. Phys. 38:4689, 1967.
[14] J. Blewett, Iron shielding for air core magnets, Proc. 1968 Summer Study on
Superconducting Devices and Accelerators, Brookhaven National Laboratory,
p. 1042–1051.
[15] A. Jain, Basic theory of magnets, in S. Turner (ed.), CERN Accelerator School on
Measurement and Alignment of Accelerator and Detector Magnets, CERN 98–05,
1998, p. 1.
Problems, 2nd ed., Pergamon Press, 1973, p. 48–49.
[17] M. Green, Modeling the behavior of oriented permanent magnet material using
current doublet theory, IEEE Trans. Mag. 24:1528, 1988.
[18] K. Halbach, Design of permanent multipole magnets with oriented rare earth cobalt
material, Nuc. Instr. Meth. 169:1, 1980.
[19] R. Beth, Currents and coil forces as contour integrals in two dimensional magnetic
fields, J. Appl. Phys. 40:2445, 1969.
[20] K. Binns & P. Lawrenson, op. cit., chapters 6–10 and appendix III.
[21] K. Miller, op. cit., p. 202–204.
[22] M. Spiegel, op. cit., p. 203–204, 216.
[23] J. Dettman, Applied Complex Variables, Dover, 1984, p. 260–265.
[24] K. Binns & P. Lawrenson, op. cit., p. 43.
[25] F. Mills & G. Morgan, A flux theorem for the design of magnet coil ends, Part. Acc.
5:227, 1973.
6
Iron-dominant transverse fields
In the previous two chapters, we have been discussing transverse field magnets
where the field shape is controlled by the distribution of the conductors. When
iron was present, it served mainly to enhance the strength of the field produced
by the conductors. In this chapter, we examine transverse field magnets where
the primary roles of the conductor and the iron are reversed. Here the shape of
the field is determined by the shape of the iron surface and the conductors are
used to excite the field in the iron.[1, 2] In addition, the iron reduces the
reluctance in the magnetic circuit, allowing a larger useful field for a given
number of amp-turns from the conductor. These types of magnets typically
have a maximum field less than 2 T, so that iron saturation effects do not
destroy the field quality. We will mainly be concerned with the calculation of
the magnetic fields and do not consider the many engineering considerations
necessary to actually build magnets of this type.
6.1 Ideal multipole magnets

If the permeability of the iron is very large (μr ~ 1000), it is a useful approximation
to assume that μr is infinite. In that case, the magnetic flux density B must
be perpendicular to the iron surface. The shape of the iron surface in the transverse
plane coincides with an equipotential line for the scalar potential. Then, since the
equipotential lines of the real and imaginary parts of the complex potential W
are orthogonal, the magnetic field follows from the equipotential lines for the
vector potential. Each positive pole of the magnet acts like a source of magnetic
field, while the negative poles act like a sink where the magnetic field returns back
into the iron.
The shape of the iron pole piece for an ideal 2n-multipole magnet is determined
by the complex potential for the multipole, which can be found from conformal
mapping to have the form
150
6.1 Ideal multipole magnets 151
Figure 6.1 Iron surfaces and field lines in a dipole magnet.
WðzÞ ¼ cn zn
¼ cn rn einθ ;
where the constant cn gives the strength of the potential. The magnetic field for the
ideal multipole is given by
dW
B ¼ i
dz
¼ i n cn zn1 :
The simplest example of an ideal multipole is the dipole (n = 1), which has the
complex potential
W ¼ c1 z
¼ c1 ðx þ i yÞ:
Figure 6.1 illustrates the iron surface and the lines of magnetic field for a dipole
magnet. The dipole has two poles with opposite polarity. Taking the real and
imaginary parts of W, the vector and scalar potentials are
A z ¼ c1 x
μ0 Vm ¼ c1 y:
We see that the iron surface is given by the equipotential

c1 y ¼ h;
where the constant h identifies a particular surface. The vector potential is given by
the equipotential
c1 x ¼ k;
152 Iron-dominant transverse fields
Figure 6.2 Iron surface of an ideal quadrupole, ignoring the asymptotic tails.
where the constant k identifies a particular equipotential line. The magnetic field for
the dipole is
∂Az
By ¼ ¼ c1 ;
∂x
which is a constant.
The iron surfaces for higher order ideal multipoles can be found in a similar
manner. The ideal quadrupole (n = 2) has
W ¼ c2 z2 ¼ c2 ðx2 y2 þ 2 i x yÞ
Az ¼ c2 ðx2 y2 Þ
μ0 Vm ¼ 2c2 x y:
The iron surface for the ideal quadrupole is the hyperbola

h
xy¼ ;
2c2
as shown in Figure 6.2. In polar coordinates, we can write the equation of the
hyperbolic surface as
r2 sin 2θ ¼ a2 ;
where a is the radius to the center of a pole and θ is measured from the positive
x axis. There are four poles around the perimeter of the magnet that alternate in
polarity. The surface hyperbola for a normal quadrupole has asymptotes along the
x and y axes. The field components inside the aperture are
Bx ¼ 2c2 y
By ¼ 2c2 x:
The vertical field on the midplane varies linearly across the aperture and is an
example of a gradient field.
The symmetry properties of a magnet are determined by the symmetry of the
poles. In an ideal normal 2N-multipole, the poles are located at the azimuthal
angles
π
ϕk ¼ ð2k 1Þ ; k ¼ 1; 2; . . . ; 2N (6.1)
2N
The polarities of the poles alternate in direction. The spacing between the poles
is π/N.
6.2 Approximate multipole configurations

It is not possible to build an ideal multipole magnet because the equipoten-
tial surfaces extend to infinity. Thus one is faced with approximating the
ideal surface as well as possible to meet the field quality requirements for
the magnet. Any approximation leads to the presence of additional allowed
multipoles. The strength of the normal multipoles are proportional to cos ϕk .
The conductor must be wound around the poles in such a way that the polarities of
adjacent poles are in opposite directions. In order to get the poles to alternate in
sign, we need
π
cos ϕk þ ¼ cos ϕk :
N
This is the same requirement that we saw in Section 4.6 for current distributions, so
the allowed multipole components m are again given by
m ¼ N ð2n þ 1Þ; n ¼ 0; 1; 2; . . .
Halbach has described methods for determining the effects on the multipole
coefficients of iron saturation and perturbations in the fabrication or construction of
iron-dominated magnets.[3, 4] These methods involve determining the effect of the
perturbation on the scalar potential associated with the pole surface. Among the
effects he considers are azimuthal and radial displacements of the poles and
modifications in the shape of the pole surface. For example, the addition of an
iron shim with thickness profile hðϕÞ modifies the unperturbed scalar potential
approximately by
δVm ≈ hðϕÞ Hr ðϕÞ;
where Hr ðϕÞ is the field on the unperturbed surface.

Figure 6.3 Window frame dipole.
6.3 Dipole configurations

Dipole magnets are commonly used to bend charged particle beams and for
experiments requiring a uniform field.[5] The window frame dipole, shown in
Figure 6.3, is a common configuration.[2, 6, 7] The coils approximate two
parallel infinite current sheets, which we saw in Equation 4.33 produces
a uniform vertical field. In the window-frame approximation, the field is very
uniform across the aperture up to the vicinity of the coils. The field inside the
coils falls off approximately linearly, reaching zero at the outer edge of the coils.
If we look at the Ampère law around the dotted path indicated in Figure 6.3, we
find that
ð
! !
NI ¼ H ·dl
B0 h B0 Liron
¼ þ ; (6.2)
μ0 μ
where NI is the number of amp-turns in the coil, B0 is the field on the midplane at
the center of the aperture, h is the gap between the iron boundaries, and Liron is the
path length in the iron. Since μ μ0 and the typical path length in the iron is at
most a few times greater than the gap, we have
h Liron
:
μ0 μ
The field produced by a window frame dipole is then

μ0 NI
B0 ≃ : (6.3)
h
Note that the field strength is inversely proportional to the size of the gap.
Figure 6.4 Cross-section through a dipole coil. The aperture of the magnet is
located at negative x in this figure.
For finite permeability, the iron will undergo saturation at high field strengths,
the second term in Equation 6.2 may no longer be negligible, and the field in the gap
will be smaller than indicated by Equation 6.3. The field is lower at the center of the
magnet compared to the field at the edges. This creates a small positive sextupole
component in the field. Increasing the width of the pole beyond the useful aperture
can improve the field quality.
Figure 6.4 shows a cross-section through one of the coils. The force on the
conductor is
ð
! ! !
F ¼ J B dV:
The current density can be written as

NI B0
J¼ ¼ :
wh μ0 w
Assuming the field falls off linearly across the coil

x
BðxÞ ¼ B0
w
and using
dV ¼ hL dx;
we can write [1]

ðw
B20 hL
F¼ x dx
μ0 w 2 0
B20 hL
¼ :
2 μ0
50 50
40 40
30 30
20 20
10 10
0 0
0 10 20 30 40 50 60 70 80 90 100
Figure 6.5 POISSON model of a window frame dipole. The dimensions are in
centimeters.
Thus the force per unit length acting on the coil is
F B20 h
¼
L 2μ0
and the transverse pressure is
B20
P¼ :
2μ0
Field calculations involving finite permeability iron have to be done using

computer programs. Figure 6.5 shows a model1 of a window frame dipole made
with the program POISSON.2 This figure shows one quarter of the cross-section of
the magnet. The box in the lower left corner is an air region, which is the useful
aperture in the magnet. The box to the right of the aperture is the conductor region.
The remaining region is assumed here to be made of 1010 alloy steel. The contour
lines show the direction of the magnetic field, which are vertical and fairly uniform
in the aperture. POISSON breaks the iron into a large grid of points where the
vector potential is computed. The relative permeability at each grid point is
determined from a B-H curve. The one used here has a maximum μr value of
2,755. The program produces a self-consistent solution of Maxwell’s equations.
Table 6.1 summarizes the results for three values of the field B0 at the center of the
aperture. The second column shows the minimum value of the relative permeability
1
This is a model of the 18D72 bending magnet that was built at Brookhaven National Laboratory.
2
We will discuss the POISSON program in more detail in Chapter 11.
Table 6.1 Summary of POISSON calculations for the win-

dow frame dipole
B0 [T] Minimum μr x0.001 [cm] F3

1.56 65 21.2 3.5 10−5
2.07 13 9.3 4.0 10−4
2.59 4 3.0 3.1 10−3
2.8
2.7
2.6 B0 = 2.588 T
2.5
BY [ T ]
2.4
2.3
2.2
2.1 B0 = 2.065 T
0 5 10 15 20
x [cm]
Figure 6.6 Magnetic field along the midplane aperture for the window frame
dipole.
at any of the iron grid points. This quantity depends on the field strength in the iron
and becomes smaller as the iron saturates at higher fields.
The third column shows the half-width of the good field region, defined here as
the distance at which the field exceeds B0 by more than 103 T. The last column
shows the fractional contribution of the sextupole compared to the dipole contribu-
tion to the field. The strength of the field across the half-aperture is shown in
Figure 6.6 for the two higher values of B0 . The field is smallest at the center of the
aperture and grows as it approaches the conductor.
Another common dipole configuration is the H-dipole,[2, 6, 7] shown in
Figure 6.7. The coils are recessed and hidden from direct view of the useful part
of the magnet aperture. This makes the field less sensitive to errors in the coil
location. The field is not as uniform as that in the window frame configuration.
The error multipoles in a fixed, useful aperture decrease exponentially with
increasing pole width. The iron near the edge of the pole is the first area that
exhibits saturation. The good field region can also be extended by adding or
subtracting material at the outer edges of the pole, rounding the corners, or by
Figure 6.7 Cross-section of an H-dipole.
tapering the side edge of the iron. Leakage flux, which circulates around the
conductors, can also cause saturation in the iron near the coils, so the magnetic
field in the pole piece is maximum at the center of the pole. This creates a negative
sextupole component in the field in the aperture.
The effective width of the field is ≈w þ g. This causes the flux from the midplane
to get squeezed going into the poles,
wBpole ≈ ðw þ gÞB0 ;
where B0 is the field on the midplane at the center of the aperture. The field on the
pole is then
g
Bpole ≈ 1 þ B0 :
w
The H-dipole has good field quality and mechanical stability. Simple racetrack-
shaped coils can be used to excite the field in the iron.
The C-dipole, shown in Figure 6.8, is a configuration that allows good
access to the magnet aperture from the side.[2, 6] The field in the iron can
be excited with simple racetrack coils. However, the requirement for accessi-
bility leads to a number of disadvantages. The necessary volume of the iron
yoke is larger than for an H-dipole. There may be considerable leakage flux
surrounding the conductors. The asymmetry in the yoke makes the mechanical
stability worse. At high field levels, the attractive force between the poles can
be quite large. Shims may be required at the edges of the poles to get
acceptable field quality. There is a nonuniform magnetic field across the
aperture, although this may be a desirable feature for applications that require
a gradient field component. The field is smaller on the outside side of the gap
than on the inside. The lack of left-right symmetry allows even harmonics to
also be present in the field between the pole pieces. The fringe fields between
the pole pieces extend outward by about a gap length on both sides of the pole
pieces.
Figure 6.8 Cross-section of a C-dipole.
The effective length of a dipole, taking into account its end windings, is
ð
1 ∞
Leff ¼ BðzÞ dz
B0 ∞ (6.4)
≃ Liron þ h;
where B0 is the dipole strength in the center of the magnet. The quantity h is
a length proportional to the aperture of the dipole, which takes into account the
fringe field extending beyond the iron.
Conformal mapping techniques have been used to improve the modelling of the
fringe field from dipole magnets.[8] Maps were used to transform the field from
single and double-sided pole pieces with a uniform gap to the upper half of the
complex plane.
6.4 Quadrupole configurations

Quadrupole magnets are frequently used for focusing charged particle beams.[5]
We saw in Section 6.1 that an ideal quadrupole requires an infinitely long hyper-
bolic iron surface. A common method for terminating the iron boundary is to use
symmetric cutoff angles θ1 , as shown for half of a symmetric pole in Figure 6.9.[6]
The surface at the cutoff angles proceeds outwards along a radius. The equation of
the hyperbolic surface relative to the centerline of the pole is
r2 cos 2θ ¼ a2 ;
Figure 6.9 Cutoff angle θ1 for the iron surface in a quadrupole magnet.
Figure 6.10 Loop through the quadrupole.
where a is the radius to the center of the pole. The cutoff angle θ1 can be selected to
be ~27° in order to eliminate the first allowed error multipole B6. In this case, the
radius to the cutoff point is given by
r1
≃ 1:12:
a
The choice of the angle θ1 also determines the amount of space available for the
conductor. Excitation of the iron poles by the conductor can be determined by using
the Ampère law around the path shown in Figure 6.10.
ða
BðrÞ B0 Liron
NI ≃ dr þ :
0 μ0 μ
The contribution from the path in the iron may be neglected since μr 1.
The contribution for the path along the x axis vanishes because the field is
perpendicular to the path. Thus we have,
ða
gr
NI ≃ dr
0 μ0
ga2
¼ ;
2μ0
where NI is the amp-turns around a pole and g is the quadrupole field gradient. Thus
the gradient is given by
2μ0 NI
g¼ (6.5)
a2
and the pole tip field is
Bpole ¼ ga
2μ NI (6.6)
¼ 0 :
a
Saturation in the iron affects the area near the conductor first.
Figure 6.11 shows a POISSON model3 of a quadrupole with hyperbolic pole
pieces. The figure shows 1/8 of the symmetric cross-section. The 45° boundary
splits one of the four poles in half. The region in the vicinity of the origin is the open
aperture. The rectangular box on the side of the pole for x ~14 to 26 cm is the
conductor, which wraps around the pole and returns on the opposite side of the
symmetric half pole piece. The pole piece is part of the iron yoke that provides
a flux path to the symmetric adjacent pole.
High-field dipoles and quadrupoles require pole piece materials with a large
value for the saturation magnetic flux density. A number of soft magnetic materials
with large Bsat are listed in Table 6.2. Also listed are the initial and peak values for
the permeability and the coercivity. The resistivity of the material is important for
considerations of eddy current losses in time-varying operations.
Quadrupoles have also been constructed by approximating the hyperbolic sur-
face with a circular cylinder.[6] Consider a circle tangent to the hyperbola at the
center of the pole, as shown in Figure 6.12. The circular surface is continued out to
a cut-off angle θ1 with respect to the center of the pole and then extends outward
along a radius. Let a be the shortest distance from the center of the magnet to the
pole and R be the radius of curvature of the circle, which is centered at C. Then
R ¼ R sin θ1 þ a sin θ1
a sin θ1
¼ :
1 sin θ1
The radius of the cutoff point is

r1 ¼ ðR þ aÞcos θ1 :
3
This model is an example file that is part of the POISSON code distribution.
Table 6.2 Magnetic alloys with large Bsat [9]
Alloy Composition1 Bsat [T] Initial μr Max μr Hc [Oe]2 ρe [μΩ-cm]

35Co,1Cr 2.42 650 10,000 0.63 20
Supermendur 49Co,2 V 2.40 800 70,000 0.23 40
Vanadium 49Co,2 V 2.35 800 6,000 2.20 40
permendur
Iron 2.14 150 5,000 1.00 10
Silicon steel 0.5Si 2.05 280 3,000 0.90 28
silicon steel 3Si 2.01 290 8,000 0.70 47
grain-oriented 3Si 2.01 1,400 50,000 0.09 50
Si steel
1
In percent, balance is Fe; 2 1 Oe = 1 10−4 T / μ0 :
0 5 10 15 20 25 30 35
30
25
20
15
10
0
0 5 10 15 20 25 30 35
Figure 6.11 POISSON model of a quadrupole with hyperbolic pole pieces.

Dimensions are in centimeters.
References 163
Figure 6.12 Quadrupole magnet with circular iron surface.
One solution to these equations, which makes the first allowed multipole harmonic
B6 ¼ 0; is
θ1 ¼ 31:5
r1
¼ 1:785
a
R
¼ 1:094:
a
Conformal mapping techniques have been used to simplify the design of
quadrupoles and higher order multipole magnets.[10] The desired higher-order
multipole is mapped to a dipole geometry, where it is easier to understand what
effects proposed modifications make to the field in the useful aperture. It is also
possible to numerically determine the field quality in the higher multipole
aperture more accurately by computing the multipole coefficients in the trans-
formed geometry.
References
[1] J. Tanabe, Iron-Dominated Electromagnets, World Scientific, 2005.
[2] T. Zickler, Basic design and engineering of normal-conducting, iron-dominated
electromagnets, in Proc. CERN Accelerator School on Magnets, Bruges, Belgium,
CERN-2010-004, June 2009, p. 65.
[3] K. Halbach, First order perturbation effects in iron dominated two dimensional
symmetrical magnets, Nuc. Instr. Meth. 74:147, 1969.
[4] K. Halbach, Fields and first order perturbation effects in two dimensional conductor
dominated magnets, Nuc. Instr. Meth. 78:185, 1970.
[5] A. Banford, The Transport of Charged Particle Beams, Spon Ltd., 1966, p. 81–84,
100–103.
[6] G. Parzen, Magnetic Fields for Transporting Charged Beams, Brookhaven National
Laboratory Report 50536, 1976.
[7] G. Fischer, Iron-dominated magnets, AIP Conf. Proc. 153:1147–1151, 1987.

[8] P. Walstrom, Dipole magnet field models based on a conformal map, Phys. Rev.-Spec.
Top. Accel. Beams 15:102401, 2012.
[9] Magnetically soft materials, American Society for Metals (ASM) Handbook, vol. 2,
1990, p. 761–781.
[10] K. Halbach, Application of conformal mapping to evaluation and design of mag-
nets containing iron with nonlinear B(H) characteristics, Nuc. Instr. Meth. 64:278,
1968.
7
Axial field configurations
In this chapter, we consider field configurations that have an axial field component.
In straight channels, these fields are azimuthally symmetric around the system axis
and only have axial and radial components. The basic example of this type of
configuration is the closed circular current loop. Combinations of current loops can
be used to produce desired axial field profiles. The current loop can also be
extended axially to generate an ideal sheet solenoid. We conclude the chapter
with a discussion of bent solenoids. When the bent channel forms a closed ring,
we obtain the toroid configuration.
7.1 Circular current loop

We recall from Equation 1.18 that the on-axis field of a circular current loop with
radius a is
μ0 Ia2
Bz ¼ ; (7.1)
2fa2 þ z2o g3=2
where I is the current in the loop and zo is the distance of the observation point along
the z axis from the plane of the loop. We now consider the determination of the
vector potential in the case when the observation point P is not restricted to lie
along the z axis, as shown in Figure 7.1. We define a coordinate system where the
x axis lies directly below the observation point P. By symmetry, the vector potential
only has a ϕ component and cannot depend on the azimuthal coordinate ϕ.
The vector potential is given by
þ
μ0 I ds
Aϕ ðρ; zÞ ¼ :
4π R
165
166 Axial field configurations
Figure 7.1 Current loop geometry.
An arbitrary element of current has the Cartesian coordinates (a cos ϕ; a sin ϕ; 0), so
n o1=2
R ¼ ðρ a cos ϕÞ2 þ ða sin ϕÞ2 þ z2
1=2
¼ ρ2 þ a2 2aρ cos ϕ þ z2 :
The contribution from a current element at ϕ makes the same contribution to the
vector potential as the element at –ϕ. In addition, the contribution of each of these
elements to the vector potential at P is proportional to cos ϕ. Therefore we can write
ð
μ0 I π cos ϕ
Aϕ ðρ; zÞ ¼ a dϕ:
2π 0 fρ þ a 2aρ cos ϕ þ z2 g1=2
2 2
Making the substitutions

ϕ ¼ π þ 2θ
cos ϕ ¼ 1 þ 2sin 2 θ;
we can write the vector potential as

ð
μ Ia 0 2sin 2 θ 1
Aϕ ðρ; zÞ ¼ 0 2 dθ:
2π π=2 fρ2 þ a2 þ z2 2aρð2sin 2 θ 1Þg1=2
The integral is symmetric in θ, so we can translate the limits of integration. After

rearranging the terms in the denominator, we get
ð π=2
μ Ia 2sin 2 θ 1
Aϕ ðρ; zÞ ¼ 0 dθ:
π 0
1=2
fða þ ρÞ2 þ z2 4aρ sin 2 θg
Define
4aρ
k2 ¼ : (7.2)
ða þ ρÞ2 þ z2
Then we have
ð π=2
μ Ia k 2sin 2 θ 1
Aϕ ðρ; zÞ ¼ 0 pffiffiffiffiffiffiffiffi dθ
π 4aρ 0 f1 k2 sin 2 θg
1=2
(7.3)
μ Ia k
¼ 0 pffiffiffiffiffiffiffiffi ½2 I1 I2 ;
π 4aρ
where1
ð π=2
sin 2 θ
I1 ¼ 1=2
dθ
0 f1 k2 sin 2 θg
KðkÞ EðkÞ
¼
k2
and2
ð π=2
1
I2 ¼ 1=2
dθ
0 f1 k2 sin 2 θg
¼ KðkÞ:
The function KðkÞ is the complete elliptic integral of the first kind and EðkÞ is the
complete elliptic integral of the second kind.3 Substituting back into Equation 7.3,
we find

μ0 Ia k 2 2
Aϕ ðρ; zÞ ¼ pffiffiffiffiffiffiffiffi 1 KðkÞ 2 EðkÞ ;
π 4aρ k2 k
which can be written in the form [1, 2]

rffiffiffi
μI a k2
Aϕ ðρ; zÞ ¼ 0 1 KðkÞ EðkÞ : (7.4)
πk ρ 2
1 2 3
GR 8.112.5. GR 8.112.1. Properties of complete elliptic integrals are discussed in Appendix F.
The components of the magnetic field in cylindrical coordinates are

∂Aϕ
Bρ ¼
∂z
Bϕ ¼ 0
1 ∂
Bz ¼ ðρAϕ Þ:
ρ ∂ρ
In order to evaluate these field components, we need the derivatives of the para-
meter k defined in Equation 7.2. We have [1]
∂k zk3
¼
∂z 4aρ
(7.5)
∂k k k3 k3
¼ :
∂ρ 2ρ 4ρ 4a
We also need the derivatives4 of the complete elliptic integrals KðkÞ and EðkÞ with
respect to their parameter k.
∂K E K
¼
∂k k ð1 k2 Þ k
(7.6)
∂E E K
¼ :
∂k k k
Evaluating the derivatives together with a lot of algebra,5 we find that [1, 2]
" #
μ0 I z a2 þ ρ2 þ z2
Bρ ¼ KðkÞ þ EðkÞ (7.7)
2π ρfða þ ρÞ2 þ z2 g1=2 ða ρÞ2 þ z2
and
" #
μ0 I 1 a2 ρ2 z2
Bz ¼ KðkÞ þ EðkÞ : (7.8)
2π fða þ ρÞ2 þ z2 g1=2 ða ρÞ2 þ z2
In the limit ρ → 0, k2 ¼ 0 and the elliptic integrals in Equation 7.8 equal π/2. Then
it is straightforward to show that Bz approaches Equation 7.1 for the axial field on
the axis. Using l’Hopital’s rule and the series expansions
4 5
GR 8.123.2,4. A computer algebra program is really useful here!
π π 2
EðkÞ ≃ k þ
2 8
π π
KðkÞ ≃ þ k2 þ ;
2 8
it is also possible to show that Equation 7.7 for Bρ approaches 0 on the axis, as it
should.
In the preceding derivation, we have gone through a standard approach of
calculating the vector potential and taking its derivatives to find the field compo-
nents. We have done this to illustrate several useful mathematical properties
involving the use of elliptic integrals. We should note, however, that it is possible
in this case to solve the Biot-Savart equation for the fields directly since the
required integrals are known.[3]
Besides the solution given here in terms of KðkÞ and EðkÞ and cylindrical
coordinates, the problem of the circular current loop has been solved using
a number of alternative methods. The vector potential for the circular loop can be
written in terms of Bessel functions as [4]
ð∞
μ Ia
Aϕ ðρ; zÞ ¼ 0 J1 ðkaÞ J1 ðkρÞ ekj z j dk: (7.9)
2 0
For some applications, it is more convenient to solve for the vector potential of the
current loop in spherical coordinates. Spherical solutions for the vector potential
and field have also been given in terms of elliptic integrals.[5, 6] However in
spherical coordinates, it is sometimes more natural to expand the solutions in
Legendre functions. The vector potential for the current loop for r < a is given in
this case as [1]
μ0 I X
∞
sin α r n 1
Aϕ ðr; θÞ ¼ Pn ðcos αÞ P1n ðcos θÞ; (7.10)
2 n¼1 nðn þ 1Þ a
where P1n is an associated Legendre function and α is the polar angle of the loop. For
r > a, the radial factor in this equation must be replaced with
anþ1
:
r
This type of expansion makes it easier to show that the field of the current
loop approaches that for a magnetic dipole in the limit when r >> a. It is
also possible to solve for the field components directly from Maxwell’s
equations.[7]
Figure 7.2 Mutual inductance of two current loops.
Example 7.1: mutual inductance of two coaxial current loops

Consider two coaxial current loops separated by a distance d, as shown in Figure 7.2.
The mutual inductance between the two loops is the flux intercepted by loop 2 for
a given current in loop 1. Thus we have
ð
Φ2 1
M¼ ¼ Aϕ1 ðb; dÞ ds2 :
I1 I1
The vector potential for loop 1 at points along loop 2 can be found using Equation 7.4
rffiffiffi
μI a k2
Aϕ ðb; dÞ ¼ 0 1 KðkÞ EðkÞ ;
πk b 2
where k 2 follows from Equation 7.2.
4ab
k2 ¼ :
ða þ bÞ2 þ d2
Since the value of Aϕ is constant for all the points on loop 2, the mutual inductance
is [8]

pffiffiffiffiffi 2 2
M ¼ μ0 ab k KðkÞ EðkÞ : (7.11)
k k
7.2 Radial expansion of the on-axis magnetic field

Consider a longitudinal distribution of azimuthally symmetric current sources. It is
useful in some cases to express the off-axis values of the magnetic field as
a function of the magnetic field along the system axis of the current distribution.
This could be used, for example, to synthesize some desirable field distribution or
7.2 Radial expansion of the on-axis magnetic field 171
for rapid optimization of current source parameters. We begin by assuming we have

a scalar potential
Ω ¼ μ 0 Vm
that does not depend on the coordinate ϕ and can be written as the power series
X
∞
Ωðρ; zÞ ¼ cn ðzÞ ρn :
n¼0
We demand that Ω satisfy Laplace’s equation in the region from the axis up to the
location of the closest coil
r2 Ω ¼ 0;
where the Laplacian operator is given in cylindrical coordinates by

1
r2 ¼ ∂ρ ðρ ∂ρ Þ þ ∂2z :
ρ
Substituting the series for Ω into Laplace’s equation and bringing the operator
inside the summation sign, we get
X
n ∂ cn
2
c nρ
n n
2 n2
þρ ¼ 0:
∂z2
In order to satisfy this relation, we need to get cancellations between the cn terms of
order n and second derivative terms two orders higher than n. Therefore let us
demand that
∂ 2 cn
cnþ2 ðn þ 2Þ2 ρn ¼ ρn :
∂z2
Making the substitution n→n 2, we can write the coefficient in terms of the
recursion relation
1 ∂2 cn2
cn ¼ ; n ≥ 2: (7.12)
n2 ∂z2
We know that the radial component of the magnetic field has to vanish on the axis
of the system. Since
∂Ω
Bρ ¼ ;
∂ρ
we find that
Bρ ¼ c1 þ 2c2 ρ þ 3c3 ρ2 þ
Therefore, we must have c1 ¼ 0, and since Equation 7.12 relates c1 to all the
higher odd terms, the series expansion for Ω can only contain even terms. Thus Ω
has the form
X
∞
Ωðρ; zÞ ¼ c2n ðzÞ ρ2n ;
n¼0
where
∂2n c0
ð1Þn
c2n ðzÞ ¼ ∂z2n : (7.13)
2 ðn!Þ2
2n
The numerical factors in the coefficient can be checked by comparing the values
from Equation 7.13 with the values from recursively using Equation 7.12. Define
the magnetic field on the system axis as

∂Ω ∂c0
B0 ðzÞ ¼ Bz ð0; zÞ ¼ ¼ :
∂z ρ¼0 ∂z
Then the off-axis axial field component is [9]
X∞
ð1Þn ∂2n B0 2n
Bz ðρ; zÞ ¼ 2 ∂z2n
ρ (7.14)
n¼0 2 ðn!Þ
2n
and the off-axis radial field component is

X
∞
ð1Þnþ1 ∂2nþ1 B0 2nþ1
Bρ ðρ; zÞ ¼ ρ : (7.15)
n¼0 2
2nþ1
n!ðn þ 1Þ! ∂z2nþ1
In cases involving loops and solenoids, where the on-axis fields are known
analytically, it is possible using this method to achieve high accuracy in computing
the field out to radial distances ~70% of the coil radius.[9]
7.3 Zonal harmonic expansions

The solution of Laplace’s equation in spherical coordinates for azimuthally sym-
metric current distributions can be expressed as a series of zonal harmonic
Figure 7.3 Geometry for zonal harmonic calculations.
functions.[10, 11] Computations of magnetic fields using these expansions can be

faster than calculations using elliptic integrals. In addition, the harmonic expansion
allows easier optimization, where, for example, we can get better field quality by
eliminating leading error terms in the series.
Consider the spherical coordinate system shown in Figure 7.3. The z axis is the
polar axis of symmetry. We choose a source point z0 along the z axis as the origin of
the coordinate system. We define a central region extending from the origin to
a radius rc that is the shortest distance to the edge of any current element. We define
the remote region to extend from the radius rr , which is the longest distance from
the origin to any part of a current element, to infinity. The zonal harmonic expan-
sion can be written as convergent series for r < rc and for r > rr. The expansion for
a given source point is divergent in the region rc < r < rr . However, it is possible
to extend the region of validity by moving the source point. The conductors are
azimuthally symmetric around the z axis. The field point F is defined to have the
spherical coordinates r and θ.
The magnetic scalar potential V ¼ Vm is a solution of Laplace’s equation. Let us
define u ¼ cos θ. The zonal harmonic solution in the central region is
X
∞
V¼ cn rn Pn ðuÞ;
n¼0
where Pn ðuÞ is the Legendre polynomial6 of order n. We choose to write the

coefficients cn in the following manner in order to simplify the expressions for
the magnetic field.
c0 ¼ Vðz0 Þ
1
cn ¼ Bc ; n > 0:
μ0 n rcn1 n1
6
Some important properties of Legendre functions are discussed in Appendix D.
The unknown quantities are now contained in the coefficients Bcn , which are called the
source terms for the central region. These quantities will be related later to the fields
produced by various current elements, such as circular loops. An important feature of
the zonal harmonic method is that the source terms for a given zo only depend on the
coordinates of the current source. Thus once the source terms have been calculated,
they can be used repeatedly in the series expansions for different field points.
With these definitions, V in the central region is written as [10]

rc X∞
Bcn1 r n
V ¼ Vðzo Þ Pn ðuÞ:
μ0 n¼1 n rc
In order to compute the cylindrical components of the magnetic field, we first make
use of the following relations from Figure 7.3.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ ρ2 þ ðz zo Þ2
∂z r ¼ u
∂ρ r ¼ sin θ
and
z zo
u¼
r
1 u2
∂z u ¼
r
u
∂ρ u ¼ sin θ:
r
The axial component of the field is

X∞
Bcn1 n 0 1 u2
Bz ¼ rc r P n ðuÞ þ P n ðuÞn rn1
u ;
n¼1
n rnc r
where P0 n is the derivative of Pn with respect to its argument u. The quantity in

square brackets is
½ ¼ rn1 fP0n ðuÞ ð1 u2 Þ þ nuPn ðuÞg:
We can use the recursion relation for Legendre polynomials7
ð1 u2 Þ P0n ¼ nPn1 unPn (7.16)
7
GR 8.914.2.
to write
½ ¼ rn1 nPn1 ðuÞ:
Substituting back into the equation for Bz, we find

X ∞ n1
r
Bz ¼ c
Bn1 Pn1 ðuÞ:
n¼1
r c
Changing the index to m ¼ n 1, the axial field in the central region can be written as
X∞ m
r
Bz ¼ Bmc
Pm ðuÞ: (7.17)
m¼0
rc
Following a similar procedure, the transverse field component in the central region is
X
∞
Bc
Bρ ¼ rc n1
½rn1 sin θ ðuP0n þ nPn Þ:
n¼1
n rnc
Using the recursion relation [12]

nPn ¼ uP0n P0n1
and shifting the series index again, we find the transverse field in the central region is
X ∞ m
Bcm r
Bρ ¼ sin θ P0m ðuÞ: (7.18)
m¼0
m þ 1 rc
In the remote region, we write the scalar potential in the form
rr X∞
Brnþ1 rr nþ1
V ¼ V0 ð∞Þ þ Pn ðuÞ:
μ0 n¼1 n þ 1 r
The axial field is

X∞
Brnþ1 rr nþ2 0
Bz ¼ Pn ð1 u2 Þ Pn u ðn þ 1Þ :
n¼1
nþ1 r
Using the recursion relation Equation 7.16 and shifting the series index, we find the
axial field component in the remote region is
X
∞ r mþ1
r
Bz ¼ Brm Pm ðuÞ: (7.19)
m¼2
r
The transverse field component in the remote region is

X∞
Brnþ1 rr nþ2
Bρ ¼ sin θ uP0n ðn þ 1ÞPn :
n¼1
nþ1 r
Using the recursion relation [13]

ðn þ 1ÞPn ¼ P0nþ1 uP0n
and shifting the series index, we find that the transverse field component in the
remote region is
X∞
Brm rr mþ1 0
Bρ ¼ sin θ Pm ðuÞ: (7.20)
m¼2
m r
Now that we have determined the series representations of the field components
in the central and remote regions, we turn to the calculation of the source terms.
Consider a field point on the z axis in the central region with z > zo . In this case,
θ = 0, ρ = 0, u = 1, and r ¼ z zo . From Equation 7.17, we have
X
∞
Bc
B0 ðzÞ ¼ Bz ð0; zÞ ¼ n
ðz zo Þn :
n¼0
rnc
When n = 0 and z ¼ zo , we see that the coefficient
Bc0 ¼ Bz ð0; zo Þ
is the axial magnetic field at the source point. The Taylor series around the point
z0 is
X∞
1 dn B0 n
B0 ðzÞ ¼ n z
ðz zo Þ : (7.21)
n¼0
n! dz o
Comparing the two series for B0, we find that [11]
rnc dn B0
Bcn ¼ ðzo Þ: (7.22)
n! dzn
Suppose that the current source is a circular loop, as shown in Figure 7.4. For
a current loop, we have rc ¼ rr ¼ rL . From Equation 7.1, the axial field at the field
point F is
μ0 Iρ2L
B0 ðzÞ ¼ : (7.23)
2d3
Figure 7.4 Source terms for a circular current loop.
Define ζ ¼ z zo . From the law of cosines, we have

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d ¼ r2L þ ζ 2 2 rL ζ uL :
We make use of the following series expansion for Legendre polynomials8
1 X
∞
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ hn Pn ðuÞ;
1 þ h 2hu n¼0
2
where h < 1. Differentiating both sides of this equation with respect to u, we find
1 X
∞
3=2
¼ hn P0nþ1 ðuÞ: (7.24)
f1 þ h2 2hug n¼0
In the central region, let

ζ
h¼ < 1;
rL
so that
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d ¼ rL 1 þ h2 2 huL :
Using Equation 7.24, we find that
1 1X ∞
¼ hn P0nþ1 ðuL Þ:
d3 r3L n¼0
8
GR 8.921.
Substituting this into Equation 7.23 and comparing with the general series expan-
sion Equation 7.17 for the case of a field point on the axis, we can conclude that the
source term for the circular loop in the central region is [11]
μ0 Iρ2L 0
Bcn ¼ P ðuL Þ: (7.25)
2r3L nþ1
Note that this expression does not depend on any parameters of a field point.
We follow an analogous procedure in the remote region to find that
μ0 Iρ2L 0
Brn ¼ P ðuL Þ: (7.26)
2r3L n1
7.4 Multiple coil configurations

Combinations of current loops are often used to create regions of space with some
desired magnetic properties. We can generalize the current source as a “small” coil
with N turns, provided the size of the coil is small compared with the separation
between the coils. In this case, we replace the loop current I in the field equations with
the product NI. The classic example of a multiple coil configuration is the Helmholtz
pair, where two coils are used to create a region of approximately uniform field.
Consider the arrangement of two coaxial current loops shown in Figure 7.5.
The two loops are perpendicular to the z axis and have the same radius a and the
same current I. The axial field of each loop is given by Equation 7.1. In a coordinate
system with the origin midway between the coils, the axial field of the two loops
can be written
μ0 I a 2
Bz ð0; zÞ ¼ FðzÞ; (7.27)
2
where
1 1
FðzÞ ¼ : (7.28)
2 3=2 3=2
fa2 þ ðb zÞ g fa2 þ ðb þ zÞ2 g
In the Helmholtz configuration, the spacing 2b between the coils equals the radius
a of the coils. The field of a Helmholtz pair at the origin is
μ0 I 8
Bz ð0; 0Þ ¼ pffiffiffi
a 5 5
(7.29)
μI
≃ 0:7155 0 :
a
Figure 7.5 A two-coil configuration.
The axial field between the coils falls off slowly with z. At the center of the current
loops the field is
a μ I 1 1

Bz 0; ¼ 0
1 þ pffiffiffi
2 a 2 8
μ0 I
≃ 0:6768 :
a
In the vicinity of the origin, the axial field can be expanded in the Taylor series,
Equation 7.21. The field uniformity is determined by the leading-order terms in this
expansion. Because of the symmetry of the coil arrangement, all the odd power
terms in the series have to vanish. The virtue of the Helmholtz configuration is that
the second derivative term in the expansion also vanishes. Thus the leading
correction in the Taylor series is the fourth order term, which is proportional to
∂ 4 Bz μI
≃ 19:8 05 :
∂z 4 a
Thus in the vicinity of the origin, the axial field is [14]
z 4
μ0 I
Bð0; zÞ ≃ 0:7155 0:825 þ : (7.30)
a a
The field at any point off the axis can be found by using the elliptic integral
solutions for the current loop given in Equations 7.7 and 7.8.[15] In the plane
midway between the coils, the field only has an axial component because of
symmetry. Defining the scaled radius u = ρ/a, the elliptic integral parameter is
16u
k2 ¼
4u2 þ 8u þ 5
Figure 7.6 Helmholtz coil configuration.
and the axial field is

2μ0 I 1 3 4u2
Bz ðu; 0Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi KðkÞ þ 2 EðkÞ :
πa 4u2 þ 8u þ 5 4u 8u þ 5
By numerically evaluating this expression, the field is found to be quite uniform in the
vicinity of the axis.[16] It falls off to 99.93% of the on-axis value at a radius of 0.2a.
The Helmholtz pair arrangement has the geometric property that the two coils lie
on the surface of a sphere, as illustrated in Figure 7.6. For this configuration, we have
tan θ ¼ 2
a
sin θ ¼ ;
r
so the radius of the sphere is
a
r¼ ≃ 1:118 a:
sinðtan1 2Þ
The Helmholtz pair also has interesting asymptotic behavior.[17] Expanding the
on-axis field in powers of a/z, the field at large distance is given by
μ0 I a2 3μ0 I a2
Bz ð0; zÞ ≃ þ ð4b2 a2 Þ þ
z3 2z5
The leading term is the magnetic dipole term. However, the next term in the series
vanishes under the Helmholtz condition a = 2b.
An inverse Helmholtz pair has the currents in the two coils flowing in opposite
directions. In this case, the field at the origin vanishes, and the leading multipole
term is the field gradient. The optimum gradient for fixed radius a is [18]
Figure 7.7 Maxwell tricoil configuration.
dBz 48 μ I
¼ pffiffiffi 02
dz 25 5 a
μI
≃ 0:8587 02 :
a
If practical constraints demand it, other gradient optimizations are possible for
fixed spacing b or for fixed radius r2 ¼ a2 þ b2 .[19]
The classic design using three coils is the Maxwell tricoil, shown in Figure 7.7.
The tricoil has a pair of identical coils and a third coil with larger radius in the
symmetry plane between the coil pair. The three coils all lie on the surface of
a sphere. This design improves on the field quality from the Helmholtz pair by also
making the fourth-order term in the Taylor series vanish. Thus the first correction
term is sixth-order. Maxwell’s solution is
rffiffiffi
4
a¼ R
7
rffiffiffi
3
b¼ R
7
49
I¼ I0 :
64
The magnetic field at the origin is
μ0 I
Bz ð0; 0Þ ¼ 60 :
R
An improved three coil design with three circular coils of the same radius has
a larger uniform field region than Maxwell’s design.[15] Another sixth-order
Figure 7.8 Sheet model of a solenoid.
design used three square coils.[20] An exhaustive study of multi-coil systems using
the methods of zonal harmonics has identified many uniform and gradient field
configurations with higher-order error corrections.[21]
7.5 Sheet model for the solenoid

We consider here the field from a solenoid in the approximation that the radial
thickness of the solenoid coil can be neglected. We assume that the sheet conductor
is composed of circular current loops that are extended in the axial direction for the
length L of the solenoid, as shown in Figure 7.8. We compute the on-axis field at the
location z of a finite length solenoid by integrating the field of a current loop
ð L=2
μ0 Ia2
Bz ð0; zÞ ¼ n o3=2 n dt ;
L=2 2
2 a þ ðz tÞ
2
where n is the number of turns per unit length. Performing the integration9 gives
" #
μ0 nI z þ L=2 z L=2
Bz ð0; zÞ ¼ : (7.31)
2 fa2 þ ðz þ L=2Þ2 g1=2 fa2 þ ðz L=2Þ2 g1=2
This can be written in the form

μ0 nI
Bz ð0; zÞ ¼ ðcos β2 cos β1 Þ; (7.32)
2
where βi are the angles subtended at location z on the axis of the solenoid to the outer
edges of the two ends of the current sheet. The field in the center of the solenoid is
9
GR 2.264.5.
L
Bz ð0; 0Þ ¼ μ0 n I pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : (7.33)
L2 þ 4a2
In the limit of an infinitely long current sheet, this expression reduces to the field of
an ideal solenoid, Equation 1.28.
Bz ¼ μ0 n I:
There is a close connection between the derivatives of the on-axis solenoid field
and the on-axis fields of the current loops at the ends of the solenoid.[9, 10] In the
coordinate system in Figure 7.8,

dBSolenoid ð0; zÞ L L
z
¼ n Bz
Loop
0; z þ Bz
Loop
0; z (7.34)
dz 2 2
The off-axis expansion method discussed in Section 7.2 can be used in conjunction
with Equation 7.34 to find the field of a sheet solenoid.[9]
We turn next to calculating the field of a solenoid at any point, including points
off the symmetry axis. We will perform a direct calculation of the field using the
Biot-Savart equation
! !
! μ0 I dl R
dB ¼ :
4π R3
Consider the solenoid geometry shown in Figure 7.9. Because of the azimuthal
symmetry of the current, the field is also azimuthally symmetric. Thus for
Figure 7.9 Geometry of the sheet solenoid.

mathematical simplicity, we are free to choose the field point F to lie directly above
the x axis. The distance from the source current element to the field point is
!
R ¼ ðρ a cos ϕ0 Þ ^x a sin ϕ0 ^y þ ðz z0 Þ ^z ; (7.35)
while the current element is

!
dl ¼ a sin ϕ0 dϕ0 ^x þ a cos ϕ0 dϕ0 ^y : (7.36)
We first compute the axial component of the field. Taking the z component of the
cross-product in the Biot-Savart equation, we find that the ϕ0 dependence only
involves terms in cos ϕ0 . Thus symmetric current elements with respect to the x axis
make identical contributions to the integral. We have
ð ð
μ0 I 0 a L=2 π a ρ cos ϕ0
Bz ¼ dz0 dϕ0
2π L=2 0 fa2 þ ρ2 2aρ cos ϕ0 þ ðz z0 Þ2 g3=2
ð
μ0 I 0 a π
¼ ða ρ cos ϕ0 Þ I1 dϕ0 ;
2π 0
where I 0 is the sheet current density,

ð L=2
dz0
I1 ¼ 3=2
L=2 fe 2 z z0 þ z0 2 g
and we define
e ¼ a2 þ ρ2 2aρ cos ϕ0 þ z2 : (7.37)
Define the distances from the observation point to the two ends of the solenoids in
terms of the new variables
L
z1 ¼ z
2
L
z2 ¼ z:
2
After doing the integration, we get10
z2
I1 ¼ Ωðz1 Þ;
ðe z2 Þfa2 þ ρ2 2aρ cos ϕ0 þ z22 g
1=2
10
GR 2.264.5.
where we use the symbol Ω here as a shorthand notation that means a second term
similar to the first, but with L replaced by –L, i.e., the other end of the solenoid.
Then we have
ðπ
μ I 0a a ρ cos ϕ0 z2 0
Bz ¼ 0 0 1=2 dϕ Ωðz1 Þ:
2π 0 ða2 þ ρ 2 2aρ cos ϕ Þ a þ ρ þ z2 2aρ cos ϕ
2 2 2 0
Change the integration variable using
cos ϕ0 ¼ 1 þ 2 x2
2 (7.38)
dϕ0 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi dx:
1 x2
This gives
ð1
μ0 I 0 a a þ ρ 2ρx2 z2
Bz ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi dx Ωðz1 Þ:
π 0
2
½ða þ ρÞ 4aρx2 fða þ ρÞ2 þ z2 4aρx2 g1=2 1 x2
2
We can put the integral into a standard form by defining

4 aρ
k2 ¼ (7.39)
ða þ ρÞ2 þ z22
and
4 aρ
n¼ : (7.40)
ða þ ρÞ2
We find that
μ0 I 0 a z2
Bz ¼ 2
½ða þ ρÞ I2 2ρ I3 Ωðz1 Þ;
πða þ ρÞ fða þ ρÞ2 þ z2 g1=2
2
where11
ð1
dx
I2 ¼
0 ð1 n x2 Þfð1 k2 x2 Þ ð1 x2 Þg1=2
¼ Πðk; nÞ:
The function Πðk; nÞ is the complete elliptic integral of the third kind. The other
integral is
11
GR 8.111.4.
ð1
x2
I3 ¼ dx:
0 ð1 n x2 Þfð1 k2 x2 Þ ð1 x2 Þg1=2
This can be evaluated by writing it in the form

ð
1 1 1 nx2 1
I3 ¼ dx
n 0 ð1 n x2 Þfð1 k2 x2 Þ ð1 x2 Þg1=2
1 1
¼ KðkÞ Πðk; nÞ;
n n
where12
ð1
dx
KðkÞ ¼ : (7.41)
0 fð1 k 2 x2 Þ ð1 x2 Þg1=2
Substituting, we find that the axial field of the solenoid is [22]
μ0 I 0 a z2 h aρ i
Bz ¼ KðkÞ þ ðΠðk; nÞ KðkÞÞ Ωðz1 Þ:
π ða þ ρÞfða þ ρÞ2 þ z2 g1=2 2a
2
(7.42)
Selecting instead the x component of the cross-product in the Biot-Savart

equation, the transverse component of the solenoid field is
ð ð
μ0 I 0 a L=2 π ðz z0 Þ cos ϕ0
Bρ ¼ dz0 dϕ0
2π L=2 0 fa2 þ ρ2 2aρ cos ϕ0 þ ðz z0 Þ2 g3=2
ð
μ I0a π
¼ 0 cos ϕ0 ½z I1 I4 dϕ0 :
2π 0
The integral over z0 involves the integral I1 that we have already considered and the
integral13
ð L=2
z0
I4 ¼ 3=2
dz0
L=2 fe 2 z z0 þ z0 2 g
zL
e
¼ 2
( 2 )1=2 ΩðLÞ;
L
ðe z2 Þ a2 þ ρ2 2 aρcos ϕ0 þ z
2
12 13
GR 8.111.2 and 8.112.1. GR 2.264.6.
where e was defined in Equation 7.37. Substituting these back into the equation for
Bρ and simplifying, we get
ðπ
μ I 0a cos ϕ0
Bρ ¼ 0 dϕ0 Ωðz1 Þ:
fa2 þ ρ2 þ z22 2aρ cos ϕ0 g
2π 1=2
0
Making the change of variable in Equation 7.38, we find

ð1
μ0 I 0 a 1 2x2 1
Bρ ¼ dx Ωðz1 Þ
π fða þ ρÞ2 þ z2 g1=2 0 fð1 k2 x2 Þ ð1 x2 Þg1=2
2
μ0 I 0 a 1
¼ ½2 I5 KðkÞ Ωðz1 Þ;
π fða þ ρÞ2 þ z2 g1=2
2
where k2 was defined in Equation 7.39. The remaining integral is14

ð1
x2
I5 ¼ 1=2
dx
0 fð1 k 2 x2 Þ ð1 x2 Þg
KðkÞ EðkÞ
¼ :
k2
Substituting and simplifying, we find the transverse component of the solenoid
field is [22]
1=2
μ I 0 fða þ ρÞ2 þ z22 g
Bρ ¼ 0 ½2ðKðkÞ EðkÞÞ k2 KðkÞ Ωðz1 Þ: (7.43)
4π ρ
Equations 7.42 and 7.43 are exact solutions for the sheet solenoid that are valid
for all points in space, except for observation points at the same radius as the
current sheet. For an arbitrary field point with the cylindrical coordinates (ρ, ϕ, z),
we can write the Cartesian values of the transverse field as
x
Bx ¼ Bρ cos ϕ ¼ Bρ
ρ
(7.44)
y
By ¼ Bρ sin ϕ ¼ Bρ :
ρ
An alternate solution for the field components has also been given in terms of
related functions known as generalized complete elliptic integrals.[23]
The vector potential for the sheet solenoid may also be expressed exactly in
terms of elliptic integrals [22, 24] as
14
GR 3.153.5.
Figure 7.10 The dependence of Bz on ρ at the center of a solenoid with L = 20 cm,

a = 10 cm, I 0 ¼ 105 A/m.
"
μ I 0 z2 1=2
Aϕ ¼ 0 fða þ ρÞ2 þ z22 g ðKðkÞ EðkÞÞ
4π ρ
# (7.45)
ða ρÞ2
1=2
ðΠðk; nÞ KðkÞÞ Ωðz1 Þ;
fða þ ρÞ2 þ z22 g
where k and n are given by Equations 7.39 and 7.40, respectively. The vector
potential and field for the sheet solenoid may also be expressed as sums of Bessel-
Laplace integrals [25, 26] or in terms of modified Bessel functions.[27]
Example 7.2: radial dependence of the axial solenoid field

Let us examine the dependence of Bz on ρ at the center of a sample solenoid.
The results from using Equation 7.42 are shown in Figure 7.10. Note the reversal
of the field direction at the radius of the sheet.
Example 7.3: mutual inductance and axial force between a solenoid and a loop
Once the vector potential and the field components for the circular current loop and
the sheet solenoid are known, the mutual inductance between a solenoid and a current
loop can be computed as
ð
ΦL 1
MðS; LÞ ¼ ¼ Aϕ ðSÞ dl
IS IS
2πρL
¼ Aϕ ðSÞ;
IS
where the symbols S and L refer to the solenoid and the loop.
7.6 Block model for the solenoid 189
Assume the current is flowing in the same direction in the loop and in the solenoid.
The axial force acting on a current loop due to the magnetic field from the solenoid is
ð ð 2π
! !
Fz ðS; LÞ ¼ IL dlL B ðSÞ ¼ IL ρL Bρ ðSÞ dϕ
0
¼ 2πIL ρL Bρ ðSÞ;
where the minus sign indicates that the force tries to pull the loop and the solenoid
together.
7.6 Block model for the solenoid

In cases where the accuracy of the field calculated from the sheet model for the
solenoid is inadequate, it may be necessary to take into account the radial thickness
of the coils. Consider the cross-section of a block solenoid shown in Figure 7.11,
where the coil extends from an inner radius a to an outer radius b. We can find the
on-axis axial field by integrating Equation 7.31 for the field due to a current sheet
ð
μ0 J b z þ L=2
Bz ð0; zÞ ¼ dr ΩðLÞ;
2 a fr2 þ ðz þ L=2Þ2 g1=2
where again Ω is used as a shorthand for the expression in the first term with
L replaced by –L. Performing the integral,15 we find for points along the symmetry
axis
( " 1=2 #)
μ0 J b þ fb2 þ ðz þ L=2Þ2 g
Bz ð0; zÞ ¼ ðz þ L=2Þ ln 1=2
ΩðLÞ: (7.46)
2 a þ fa2 þ ðz þ L=2Þ2 g
The axial field at the center of the solenoid is

" 1=2
#
μ0 J L b þ fb2 þ L2 =4g
Bz ð0; 0Þ ¼ ln : (7.47)
2 a þ fa2 þ L2 =4g1=2
The off-axis field from the block solenoid is usually treated by summing over the
fields from a set of current sheets, using one of the methods we have previously
discussed. For example, the block conductor may be simulated using a radial
distribution of current sheets expressed in elliptic integrals.[22] It is also possible
to express the thick solenoid field in terms of a radial expansion of the on-axis field,
[9] as a series of zonal harmonics,[10, 11, 21] or in terms of Bessel-Laplace integrals.
15
GR 2.271.5.
Figure 7.11 Block model of a solenoid.
[25, 26, 28] The good field region calculated for a solenoid from a properly designed
block conductor is frequently larger than that for a sheet solenoid.[21]
The flux leaving a solenoid travels outside and returns through the opposite end.
As a result, the fringe field on the outside of the solenoid can be quite significant.
If the fringe field is unacceptable, it can be reduced by adding supplemental bucking
coils or by using iron shielding. Figure 7.12 shows a POISSON model for the
magnetic field in a typical solenoid. The figure shows 1/4 of a plane projection
through the solenoid. The vertical axis is the centerline of the solenoid. The field is
symmetric on both sides of the vertical axis and both sides of the horizontal axis.
The program used Dirichlet boundary conditions on the left, right, and top borders,
and Neumann boundary conditions on the bottom border. The figure on the left
shows the field from just the coil, while the figure on the right illustrates the reduction
in the exterior field from adding a cylindrical iron return yoke. The current in the coil
was the same for both figures.
7.7 Bent solenoid

So far we have been considering configurations where current loops or solenoids
have been symmetrically configured along a straight axis. In the cylindrical coordi-
nate system we have been using, the current has been azimuthally symmetric along ϕ,
the system axis has been along z, and the magnetic field only has components along ρ
and z. We now generalize this to consider configurations where a solenoid, for
example, is bent to follow a circular axis. The magnetic field of a bent solenoid
channel is conveniently defined in terms of a rotating coordinate system that follows
some reference curve, as shown in Figure 7.13. In the curvilinear description of
orthogonal coordinate systems,[29] changes in the values of the coordinates
(u1 ; u2 ; u3 ) are related to the distance element by
7.7 Bent solenoid 191
(a) (b)
180 180
160 160
140 140
120 120
100 100
80 80
60 60
40 40
20 20
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Figure 7.12 Magnetic field of a solenoid coil (left); field for a coil surrounded by
an iron return path (right).
Figure 7.13 Frenet-Serret coordinate system.
ds2 ¼ h21 u21 þ h22 u22 þ h23 u23 ;
where (h1 ; h2 ; h3 ) are a set of scale factors. In the Frenet-Serret coordinate system
considered here,[30, 31] the reference curve is a circle and the origin of the unit
vectors moves along the circle. The unit vector s is in the bending plane and tangent
to the circle. The unit vector r is in the bending plane and perpendicular to s.
The unit vector v is perpendicular to the bending plane. The curvilinear scale
factors are
hr ¼ hv ¼ 1
r
hs ¼ 1 þ ;
ρ
where ρ is the radius of curvature.

We can approximate the scalar potential in the magnet aperture in terms of
a power series. To compute the first-order fields, we must include second-order
terms in the potential,
V ¼ μ0 Vm ≃ V00 þ V10 r þ V01 v þ V20 r2 þ V11 rv þ V02 v2 :
We also include terms in the potential that allow for the possibility of superimposed
transverse fields. The gradient of V is defined as
1
rV ¼ ∂r V ^r þ ∂s V ^s þ ∂v V ^v :
hs
Thus the magnetic field components are
Br ≃ V10 þ 2V20 r þ V11 v

Bv ≃ V01 þ V11 r þ 2V02 v
1 0 0 0 0 2 0 0 2
Bs ≃ ðV00 þ V10 r þ V01 v þ V20 r þ V11 rv þ V02 v Þ;
hs
where primes indicate derivatives with respect to s. Recalling the midplane expan-
sions of the transverse field components given in Equation 4.9, we can associate the
potential terms with the multipole field coefficients16
V01 ¼ B1
V10 ¼ A1
V11 ¼ B2
2V20 ¼ A2 :
Thus to the first-order, the field components are
Br ≃ A1 A2 r þ B 2 v
Bv ≃ B1 þ B2 r 2 V02 v
(7.48)
1
Bs ≃ ðbs A0 1 r þ B0 1 vÞ;
hs
16
In the case where a charged particle has to follow the reference path in the horizontal plane, we must have the
horizontal dipole field A1(s) = 0.
7.8 Toroid 193
where we identify the on-axis component of the axial field as

0
bs ¼ V00
and assume the transverse dipole fields can vary with s.

At this point, we still have one unidentified potential term V02 in Equation 7.48,
so we demand that the field components also satisfy the divergence relation
!
r· B ¼ 0. In the coordinates discussed here, this can be written as
1 1
∂r ðhs Br Þ þ ∂s Bs þ ∂v Bv ¼ 0:
hs hs
Inserting the field components from Equation 7.48 and using

1 r
≃1 ;
hs ρ
we find the constraint

1
A2 A1 2 V02 þ b0s ¼ 0:
ρ
Thus the first-order vertical field component is

0 A1
Bv ≃ B1 þ B2 r bs A2 v: (7.49)
ρ
If no superimposed transverse fields are present, the first-order axial field in the
bent channel is
r
Bs ≃ bs bs : (7.50)
ρ
It is also possible to define an expansion for the field that makes use of “curved
multipoles” that directly correspond to the solution of Laplace’s equation in the
curved coordinate system.[32]
7.8 Toroid
When the bent solenoid channel is extended to form a closed circular ring, we have
a toroid, as shown in Figure 7.14. The current loops from the solenoid are centered
on the circular system axis and the plane of the loops lie in the ρ-z plane.
^ depend on the azimuthal location around
The direction of the unit vectors ^ρ and ϕ
the toroid. Since the coils are closer together on the side nearer to the center of
Figure 7.14 Geometry of the toroid from above.
curvature, we expect that the field inside the bent solenoid will have a gradient with
respect to the system axis, in agreement with Equation 7.50.
Because of the symmetry of the configuration, all of the field components must
be independent of the azimuthal angle ϕ. Let the mean radius of the toroid equal
b and the radius of the current loops equal a. Then a simple application of
the Ampère law on the midplane (z = 0) shows that Bϕ ¼ 0 for ρ < b a and for
ρ > b þ a since no net current is enclosed in a circular path in those regions.
However, applying the Ampère law on a circular path on the midplane, we find
the field inside the toroid is
μ0 N I
Bϕ ¼ ; (7.51)
2πρ
where N is the number of conductor turns around the circumference and ρ is the
radius of the path. This shows that the field varies like 1=ρ inside the toroid.
A cross-section of the toroid at some azimuthal angle ϕ is shown in Figure 7.15a.
The angle α gives the location of an element of the current loop. The current
element has the Cartesian coordinates
xl ¼ ðb þ a cos αÞ cos ϕ
yl ¼ ðb þ a cos αÞ sin ϕ
zl ¼ a sin α
and the directions

dlx ¼ a sin α cos ϕ dα
dly ¼ a sin α sin ϕ dα
dlz ¼ a cos α dα:
References 195
(a) (b)
Figure 7.15 (a) Cross-section of the toroid at the azimuthal location ϕ; (b) location
of the field observation point F.
The location of the field observation point (ρ, 0, z) is shown in Figure 7.15b. The
resulting distance vector is
!
R ¼ ½ρ ðb þ a cos αÞcos ϕ ^x ðb þ a cos αÞsin ϕ ^y þ ðz a sin αÞ ^z :
Applying the Biot-Savart law to any point inside the toroid shows that Bz and Bρ
vanish. It follows that the field inside the toroid has to have the form
B ¼ Bϕ ðρ; zÞ:
The analytic results for Bϕ are complicated [33] expressions defined in terms of
integrals of elliptic integrals. Alternatively, one could examine the field inside the
toroid by evaluating one of the integrals in terms of complete elliptic integrals and
performing the other integral numerically.
References
[2] J. Stratton, Electromagnetic Theory, McGraw-Hill, 1941, p. 263.
[3] R. Good, Elliptic integrals, the forgotten functions, Eur. J. Phys. 22:119, 2001.
Wesley, 1962, p. 156.
[5] R. Schill, General relation for the vector magnetic field of a circular current loop:
a closer look, IEEE Trans. Magnetics 39:961, 2003.
[6] J. D. Jackson, Classical Electrodynamics, Wiley, 1962, p. 142.
[7] D. Redzic, The magnetic field of a static current loop: a new derivation, Eur. J. Phys.
27:N9, 2006.
p. 329.
[9] R. Jackson, Off-axis expansion solution of Laplace’s equation: application to accurate

and rapid calculation of coil magnetic fields, IEEE Trans. Electron Devices 46:1050,
1999.
[10] M. Garrett, Axially symmetric systems for generating and measuring magnetic fields,
J. Appl. Phys. 22:1091, 1951.
[11] F. Gluck, Axisymmetric magnetic field calculation with zonal harmonic expansion,
Progress in Electromagnetics Research B 32:351, 2011.
equation 12.25.
[13] Ibid., equation 12.24.
[14] L. Eyges, The Classical Electromagnetic Field, Dover, 1980, p. 140.
[15] J. Wang, S. She & S. Zhang, An improved Helmholtz coil and analysis of its magnetic
field homogeneity, Rev. Sci. Inst. 73:2175, 2002.
[16] J. Higbie, Off-axis Helmholtz field, Am. J. Phys. 46:1075, 1978.
[17] E. Purcell, Helmholtz coils revisited, Am. J. Phys. 57:18, 1989.
[18] P. Murgatroyd, Optimum designs of circular coil systems for generating non-uniform
axial magnetic fields, Rev. Sci. Inst. 50:668, 1979.
[19] P. Murgatroyd & B. Bernard, Inverse Helmholtz pairs, Rev. Sci. Inst. 54:1736, 1983.
[20] R. Merritt, C. Purcell & G. Stroink, Uniform magnetic field produced by three, four
and five square coils, Rev. Sci. Inst. 54:879, 1983.
[21] M. Garrett, Thick cylindrical coil systems for strong magnetic fields with field or
gradient homogeneities of the 6th to 20th order, J. Appl. Phys. 38:2563, 1967.
[22] M. Garrett, Calculation of fields, forces and mutual inductances of current systems by
elliptic integrals, J. Appl. Phys. 34:2567, 1963.
[23] N. Derby & S. Olbert, Cylindrical magnets and ideal solenoids, Am. J. Phys. 78:229,
2010.
[24] L. Cohen, An exact formula for the mutual inductance of coaxial solenoids, Bulletin
Bureau Standards 3:295, 1907.
[25] J. Conway, Exact solutions for the magnetic fields of axisymmetric solenoids and
current distributions, IEEE Trans. Mag. 37:2977, 2001.
[26] J. Conway, Trigonometric integrals for the magnetic field of a coil of rectangular cross
section, IEEE Trans. Mag. 42:1538, 2006.
[27] T. Tominaka, Magnetic field calculation of an infinitely long solenoid, Eur. J. Phys.
27:1399, 2006.
[28] V. Labinac, N. Erceg & D. Kotnik-Karuza, Magnetic field of a cylindrical coil, Am.
J. Phys. 74:621, 2006.
[30] S. Y. Lee, Accelerator Physics, World Scientific, 1999, p. 33–34.
[31] C. Wang & L. Teng, Magnetic field expansion for particle tracking in a bent solenoid
channel, Proc. 2001 Particle Accelerator Conference, p. 456. Available from www
.jacow.org.
[32] S. Mane, Solutions of Laplace’s equation in two dimensions with a curved long-
itudinal axis, Nuc. Instr. Meth. A 321:365, 1992.
[33] N. Carron, On the fields of a torus and the role of the vector potential, Am. J. Phys.
63:717, 1995.
8
Periodic magnetic channels
In the previous chapters, we considered magnets that for the most part had
continuous transverse or longitudinal fields along some system axis. In this chapter,
we look instead at magnetic channels where the on-axis field is periodic. Periodic
field configurations are used for focusing charged particle beams and for produc-
tion of radiation at light sources. We begin by considering the field produced by
helical conductor windings. Then we examine several examples where we demand
some desired field configuration along the axis and then find off-axis field compo-
nents that satisfy Maxwell’s equations.
8.1 Field from a helical conductor

A helically wound conductor can produce a periodic field. The parametric equa-
tions of a helix are
x ¼ a cos ϕ
y ¼ a sin ϕ
λ
z¼ ϕ;
2π
where a is the radius and λ is the period of the helix. We define the axial
wavenumber as k ¼ 2π=λ: We parameterize the nature of the helix by the angle α
in Figure 8.1.[1] We have
Δz λ 1
tan α ¼ ¼ ¼
a Δϕ 2πa ka
since z progresses by a distance λ as the azimuthal angle goes once around the
circumference. With this definition, α = 0 corresponds to the limiting case when the
helix reduces to a circular loop. It is convenient to write λ as a function of a.
197
198 Periodic magnetic channels
Figure 8.1 Definition of the helical angle α.
Figure 8.2 Helical conductor geometry at a fixed value of z.
λ ¼ 2πa tan α
The azimuthal angle and axial distance are connected through the helix constraint
z ¼ a ϕ tan α: (8.1)
Consider the cross-section through the helix shown in Figure 8.2. A single
conductor follows a helical path in a current sheet of radius a. Let the observation
point F have cylindrical coordinates (ρ, ϕ, z) and the current element at the
location S on the conductor have coordinates (a; ϕ0 ; z0 ). The current element is
given by
!
dl ¼ a sin ϕ0 dϕ0 ^x þ a cos ϕ0 dϕ0 ^y þ a tan ϕ0 dϕ0 ^z
and the distance vector is

!
R ¼ ðρ cos ϕ a cos ϕ0 Þ ^x þ ðρ sin ϕ a sin ϕ0 Þ ^y þ ðz a ϕ0 tan αÞ ^z :
Taking into account the constraint between z0 and ϕ0 , the direct evaluation of
B using the Biot-Savart equation only requires an integration over ϕ0 : The
integration limits for a winding of finite length can be found using Equation 8.1.
Although the resulting integral for the general case is complicated, the solution for
observation points along the axis of the helix is fairly straightforward.[1]
The vector potential for an infinitely long helix is given by
ð !
! μ K ða; ϕ0 ; z0 Þ 0
A ðρ; ϕ; zÞ ¼ 0 dS : (8.2)
4π R
The sheet current density only has components in the ϕ0 and z0 directions. The pitch
angle α for the helical winding can be written as
Kz 0
tan α ¼ ;
K ϕ0
so the components of the current density are related by
Kϕ0 ¼ ka Kz0 :
Thus the current density is [2]
! I
K ða; ϕ0 ; z0 Þ ¼ ð^z 0 þ kaϕ^0 Þ δðϕ0 kz0 εÞ; (8.3)
a
where ε is the azimuthal angle of the winding at z0 ¼ 0. The Dirac delta function
enforces the constraint between changes in z0 and changes in ϕ0 .
We can write the periodic delta function in Equation 8.3 as a Fourier series. Let
τ ¼ ε þ kz0 . Then
X
∞
f ðϕ0 Þ ¼ δðϕ0 τÞ ¼ a0 þ ½an cos nϕ0 þ bn sin nϕ0 :
n¼1
The coefficients are

ð
1 π 1
a0 ¼ δðϕ0 τÞ dϕ0 ¼
2π π 2π
ðπ
1 1
an ¼ δðϕ0 τÞcos nϕ0 dϕ0 ¼ cos nτ; n>0
π π π
ðπ
1 1
bn ¼ δðϕ0 τÞ sin nϕ0 dϕ0 ¼ sin nτ:
π π π
Thus the delta function can be expressed as
1 1X ∞
δðϕ0 τÞ ¼ þ ½cos nτ cos nϕ0 þ sin nτ sin nϕ0
2π π n¼1
" # (8.4)
1 X∞
0
¼ 1 þ 2 cos nðϕ τÞ :
2π n¼1
The distance from the current element to the observation point can be written as
1=2
R ¼ fa2 þ ρ2 2 aρ cosðϕ ϕ0 Þ þ ðz z0 Þ2 g : (8.5)
We need to express the unit vectors in Equation 8.3 in terms of unit vectors in the
coordinate system for the observation point. The axial unit vectors are identical,
z^0 ¼ ^z . The azimuthal unit vector is given by
ϕ^0 ¼ ^ρ sin ðϕ ϕ0 Þ þ ϕ

^ cos ðϕ ϕ0 Þ: (8.6)
Thus there are in general nonvanishing components of the vector potential in all
three dimensions.
Substituting Equations 8.3–8.6 into Equation 8.2, the axial component of the
vector potential is given by
" #
X
ðð 1þ2
∞
cos nðϕ0 kz0 εÞ
μ0 I
a dϕ0 dz0 :
n¼1
Az ðρ; ϕ; zÞ ¼ 1=2
4π 2πa 0 2
fa2 þ ρ2 2 aρ cos ðϕ ϕ Þ þ ðz z0 Þ g
(8.7)
For ρ < a, evaluation of the integrals give [2]
μ0 I μ IX∞
Az ðρ; ϕ; zÞ ¼ ln a þ 0 Kn ðnkaÞ In ðnk ρÞ cos nðϕ kz εÞ (8.8)
2π π n¼1
and for ρ > a
μ0 I μ IX∞
Az ðρ; ϕ; zÞ ¼ ln ρ þ 0 Kn ðnk ρÞ In ðnkaÞ cos nðϕ kz εÞ : (8.9)
2π π n¼1
The functions Kn and In are modified Bessel functions of order n.

The radial component of the vector potential is given by

" #
X
ðð sin ðϕ ϕ0 Þ 1 þ 2
∞
cos nðϕ0 kz0 εÞ
μ0 I k
a dϕ0 dz0 :
n¼1
Aρ ðρ; ϕ; zÞ ¼ 1=2
4π 2π fa2 þ ρ2 0
2 aρ cos ðϕ ϕ Þ þ ðz z0 Þ2 g
(8.10)
For ρ < a, evaluation of the integrals give [2]
μ0 Ika X
∞
Aρ ðρ; ϕ; zÞ ¼ ½Knþ1 ðnkaÞ Inþ1 ðnkρÞ Kn1 ðnkaÞ In1 ðnkρÞ
2π n¼1 (8.11)

sin nðϕ kz εÞ
and for ρ > a
μ0 Ika X
∞
Aρ ðρ; ϕ; zÞ ¼ ½Knþ1 ðnkρÞ Inþ1 ðnkaÞ Kn1 ðnkρÞ In1 ðnkaÞ
2π n¼1
(8.12)
sin nðϕ kz εÞ :
The azimuthal component of the vector potential is given by

" #

ðð cosðϕ ϕ0 Þ 1 þ 2 X∞
0
cos nðϕ kz0 εÞ
μ0 I k
a dϕ0 dz0 :
n¼1
Aϕ ðρ; ϕ; zÞ ¼
4π 2π 0 0 2 1=2
fa þ ρ 2 aρ cosðϕ ϕ Þ þ ðz z Þ g
2 2
(8.13)
For ρ < a this has the solution [2]
μ0 I kρ
4π
μ0 I ka X
∞
þ ½Knþ1 ðnkaÞ Inþ1 ðnkρÞ þ Kn1 ðnkaÞ In1 ðnkρÞcos nðϕ kz εÞ
2π n¼1
(8.14)
and for ρ > a
μ0 I ka2
4πρ
μ0 I ka X
∞
þ ½Knþ1 ðnkρÞ Inþ1 ðnkaÞ þ Kn1 ðnkρÞ In1 ðnkaÞcos nðϕ kz εÞ :
2π n¼1
(8.15)
The solution for the magnetic field components can be found by taking the curl
of A. For the case ρ < a the field components are [2, 3, 4]
μ0 Ik2 a X
∞
Bρ ðρ; ϕ; zÞ ¼ nK0 ðnkaÞ In0 ðnkρÞ sinðnðϕ kz εÞÞ
π n¼1 n
μ0 Ika X
∞
In ðnkρÞ
Bϕ ðρ; ϕ; zÞ ¼ nKn0 ðnkaÞ cosðnðϕ kz εÞÞ (8.16)
π n¼1 ρ
μ0 Ik μ0 Ik2 a X
∞
Bz ðρ; ϕ; zÞ ¼ nK0 ðnkaÞ In ðnkρÞ cosðnðϕ kz εÞÞ;
2π π n¼1 n
while the solution for the case ρ > a is
μ0 Ik2 a X
∞
Bρ ðρ; ϕ; zÞ ¼ nK0 ðnkρÞ In0 ðnkaÞ sinðnðϕ kz εÞÞ
π n¼1 n
μ0 I μ0 Ika X
∞
Kn ðnkρÞ
Bϕ ðρ; ϕ; zÞ ¼ þ nI 0 ðnkaÞ cosðnðϕ kz εÞÞ (8.17)
2πρ π n¼1 n ρ
μ0 Ik2 a X
∞
Bz ðρ; ϕ; zÞ ¼ nI 0 ðnkaÞ Kn ðnkρÞ cosðnðϕ kz εÞÞ:
π n¼1 n
Primes on the Bessel functions indicate derivatives with respect to the

arguments.
An example of the variation of the field components for a helical conductor is
shown for one period in Figure 8.3. The calculations were done using
Equations 8.16. In this case, the magnitude of the transverse components are
small compared to the axial component. At radii large compared to a, the azimuthal
field component becomes dominant and the field approaches that of a straight wire.
The on-axis field of the helical conductor can be found by evaluating
Equation 8.16 at ρ = 0. Using the relations
(a)
(b)
Figure 8.3 (a) The dependence of Bz along one period of the helix; (b) the depen-
dence of Bρ (solid) and Bϕ (dashed) versus z. The parameters used here were
λ = 10 cm, a = 10 cm, ρ = 5 cm, ϕ = 0, ε = 0, I ¼ 105 A, and N = 40 terms in the
sums.
In ð0Þ ¼ 0 for n > 0

In0 ð0Þ ¼ 0 for n > 1
I10 ð0Þ ¼ ½;
the on-axis field is [1, 3, 5]
μ0 Ik2 a 0
Bρ ð0; 0; zÞ ¼ K1 ðkaÞ
2π
Bϕ ð0; 0; zÞ ¼ 0
μIk
Bz ð0; 0; zÞ ¼ 0 :
2π
There is a close relationship between these results and our previous results for
the field of a solenoid. In practice, the conductor in a solenoid is wound in many
helical layers over a cylindrical form. The helical pitch length λ for a solenoid is
very small. In the previous chapter, the field for a solenoid was derived by assuming
that the field came from a longitudinal distribution of parallel infinitesimal current
loops. In the limit that k → ∞, it can be shown by taking the asymptotic limits for
the Bessel functions that Equation 8.16 approaches the on-axis field of an infinitely
long solenoid
Bz ð0; 0; zÞ ¼ μ0 nI;
where n = 1/λ is the number of turns per unit length.[6] In the opposite limit where
k → 0, the helical fields approach that of a straight conductor.
8.2 Planar transverse field

A planar wiggler has an on-axis transverse field component that oscillates in
a fixed plane. It is called a wiggler because a charged particle beam moves
back and forth in this type of field and can be used, for example, to produce
electromagnetic radiation for light sources. Assume we want an on-axis field
given by
By0 ¼ B0 cos ðγz φÞ

(8.18)
Bx0 ¼ Bz0 ¼ 0;
where z is the direction of the system axis and φ is an initial phase.

The coefficient γ is related to the wavelength of the field oscillation λ by γ ¼
2π=λ: We saw in Chapter 3 that solutions of Laplace’s equation in rectangular
coordinates (1) can be written as products of trigonometric and hyperbolic sines
and cosines and (2) that these solutions must have at least one trigonometric and
one hyperbolic factor. Since By is assumed to be non-zero on the axis, we must
choose the cosine and hyperbolic cosine functions for the general solution. Once
we have specified that the z dependence is a cosine function, there are three possible
combinations for the product of the x and y dependences. Let us consider the
solution where
By ¼ B0 cosðαxÞ coshðβyÞ cosðγz φÞ:
In free space, the div B = 0 equation gives
∂x Bx þ B0 β cosðαxÞ sinhðβyÞ cosðγz φÞ þ ∂z Bz ¼ 0:

8.2 Planar transverse field 205
In order to satisfy this equation for all x, y, and z, Bx and Bz must have the form
Bx ¼ f sinðαxÞ sinhðβyÞ cosðγz φÞ

Bz ¼ g cosðαxÞ sinhðβyÞ sinðγz φÞ;
where f and g are unknown factors. Substituting these field expressions back into
the divergence equation gives the constraint
f α þ βB0 þ gγ ¼ 0: (8.19)
The x component of the curl B = 0 equation gives the relation
γB0
g¼ ;
β
while either the y or z component of the curl equation gives
αB0
f ¼ :
β
Substituting these values for f and g into Equation 8.19, we find the wave number
constraint
γ2 ¼ α2 þ β2 : (8.20)
This can be written in terms of the period of the field variation as
4π2
λ2 ¼ :
β2 α2
For a periodic solution, we have the additional constraint that β > α. The solution
for the planar transverse field is then [7]
αB0
Bx ¼ sinðαxÞ sinhðβyÞ cosðγz φÞ
β
By ¼ B0 cosðαxÞ coshðβyÞ cosðγz φÞ (8.21)
γB0
Bz ¼ cosðαxÞ sinhðβyÞ sinðγz φÞ:
β
The other two solutions consistent with Equation 8.18 have the transverse
dependences
coshðαxÞ cosðβyÞ
and
coshðαxÞ coshðβyÞ:
These solutions can be derived following the same procedure used above. Each
solution has a unique relation among the wavenumbers.[7]
It’s important to keep in mind that this type of derivation only represents part of
the problem. What we have shown is that our desired on-axis field profile is a valid
solution of Maxwell’s equations. However, what we have not considered is
a configuration of conductors that actually produces that desired field.
The obvious trial solution here would be a series of transverse permanent magnet
or electrically excited magnetic poles that alternate in direction along the system
axis. Oftentimes the field from a realistic coil distribution can only approximate the
desired field. We define the problem of finding a current distribution that produces
a specified magnetic field as an inverse problem to distinguish it from the situation
encountered using the Biot-Savart formula, where we find the magnetic field
produced by a given current distribution. Finding a suitable current distribution
frequently involves using numerical optimization methods.
8.3 Helical transverse field

Consider an on-axis transverse field that rotates around the system axis, analogously
to the magnetic field vector in circularly polarized light. It is convenient to look for
a solution in a cylindrical coordinate system that rotates around the system axis, as
shown in Figure 8.4. In this system, z and ϕ are coupled and the on-axis field is
Bρ0 ¼ B0 cosðγz ϕ εÞ
(8.22)
Bϕ0 ¼ Bz0 ¼ 0;
where ε is an initial phase shift. For points off the axis, we look for a solution for Bρ
with the form
Figure 8.4 Rotating cylindrical coordinate system.

8.3 Helical transverse field 207
Bρ ¼ C FðρÞ cosðγz ϕ εÞ; (8.23)
where C Fð0Þ ¼ B0 and F is an undetermined function of ρ. Since the ρ

and ϕ coordinates are separated by 90°, we expect the solution for Bϕ to have
the form
Bϕ ¼ C GðρÞ sinðγz ϕ εÞ; (8.24)
where G is another undetermined function. We know from Chapter 3 that the

solution of Laplace’s equation in cylindrical coordinates must involve Bessel
functions. Calculating the ρ component of the curl B = 0 equation allows us to
obtain an expression for Bz.
Bz ¼ C γ ρ GðρÞ sinðγz ϕ εÞ: (8.25)
Calculating the ϕ component of the curl equation lets us determine a relation

between the unknown functions F and G.
FðρÞ ¼ ∂ρ ½ρ GðρÞ: (8.26)
Using Equations 8.23–8.26, we can write the div B = 0 equation directly in terms
of G.
1 1
½ρ ∂2ρ ðρGÞ þ ∂ρ ðρGÞ G γ2 ρ G ¼ 0: (8.27)
ρ ρ
Rearranging terms, this can be written as
∂2 ðγ ρ GÞ ∂ðγ ρ GÞ
γ2 ρ2 þγρ ½1 þ ðγ ρÞ2 ðγ ρ GÞ ¼ 0: (8.28)
∂ðγ ρÞ 2 ∂ðγ ρÞ
This is the differential equation for the modified Bessel1 function I1 , where the
unknown variable is γ ρG and the argument of the Bessel function is γ ρ. Thus we
have
γ ρ GðρÞ ¼ I1 ðγ ρÞ
and the unknown function G is
I1 ðγ ρÞ
GðρÞ ¼ : (8.29)
γρ
1
Some properties of the modified Bessel functions are described in Appendix C.
We can now find F from Equation 8.26.

∂I1 ðγ ρÞ
FðρÞ ¼
∂ðγ ρÞ
Using a recursion relation,2 we can write this as

1
FðρÞ ¼ I0 ðγ ρÞ I1 ðγ ρÞ; (8.30)
γρ
where I0 is the modified Bessel function of order 0. Substituting this back into
Equation 8.23, we find

1
Bρ ¼ C I0 ðγ ρÞ I1 ðγ ρÞ cosðγz ϕ εÞ:
γρ
Near the axis, I0 and I1 have the series expansions3
1 2
I0 ðuÞ ≃ 1 þ u þ (8.31)
4
and4
1 1 3
I1 ðuÞ ≃ uþ u þ (8.32)
2 16
Thus near the axis, we take the leading terms for I0 and I1 and find that
C
Bρ ≃ cosðγz ϕ εÞ:
2
Comparing this with Equation 8.23 gives C ¼ 2B0 . The solution for the helical
transverse field is then [8]

1
Bρ ¼ 2B0 I0 ðγ ρÞ I1 ðγ ρÞ cosðγz ϕ εÞ
γρ
I1 ðγ ρÞ (8.33)
Bϕ ¼ 2B0 sinðγz ϕ εÞ
γρ
Bz ¼ 2B0 I1 ð γ ρÞ sinðγz ϕ εÞ:
Note that the argument for the Bessel functions involves the longitudinal wave-
number γ.
2 3 4
AS 9.6.26. AS 9.6.12. AS 9.6.10.
8.4 Axial fields 209
It was suggested that this solution could be produced by winding conductors in

a helical manner over a cylindrical bore tube.[8] Between adjacent helical turns,
a second helix could be laid down with the current running in the opposite
direction. However, exact calculations showed that Equation 8.33 is only
a reasonable approximation for the field of this interleaved helical winding con-
figuration when a/λ is larger than ~0.3, where a is the winding radius.[3, 5]
8.4 Axial fields

As a final example of constructing a desired on-axis field, we consider the periodic
axial field
Bz0 ¼ B0 cosðγz εÞ
Bϕ0 ¼ Bρ0 ¼ 0
in a cylindrical coordinate system. By symmetry, Bϕ vanishes everywhere. Because

of the periodic behavior in z, we suspect that the off-axis solution must contain
terms proportional to the modified Bessel functions. Therefore we look for the
simplest possible solution
Bz ¼ B0 I0 ðαρÞ cosðγz εÞ:
Since the div B = 0 equation involves ∂z Bz , we know that Bρ must be proportional to

sin ðγz εÞ. Thus we have
1
∂ρ ðρBρ Þ ¼ B0 γI0 ðα ρÞ sinðγz εÞ:
ρ
Multiplying by ρ and integrating both sides, we get

ðρ
ρBρ ¼ B0 γ sinðγz εÞ ρ I0 ðαρÞ dρ þ C;
0
where C is an integration constant. The remaining integral can be evaluated as5

ðρ
ρ
ρ I0 ðαρÞ dρ ¼ I1 ðαρÞ:
0 α
Thus
γ C
Bρ ¼ B0 I1 ðαρÞ sinðγz εÞ þ :
α ρ
5
AS 11.3.25.
Since Bρ must vanish at ρ = 0, we must have C = 0. To find the relationship between

the wavenumbers α and γ, we look at the ϕ component of the curl B = 0 equation,
which gives
γ2 ∂I0 ðαρÞ
I1 ðαρÞ ¼ 0:
α ∂ρ
Using6
∂I0 ðαρÞ
¼ α I1 ðαρÞ;
∂ρ
we find that α = γ. Thus the solution for the periodic axial field is
Bρ ¼ B0 I1 ðγρÞ sinðγz εÞ
Bϕ ¼ 0 (8.34)
Bz ¼ B0 I0 ðγρÞ cosðγz εÞ:
A likely conductor configuration for producing this field is a series of solenoids

along the axis that alternate in direction. In fact, it is possible to design block
solenoids that give an excellent approximation to a sinusoidal axial field along the
axis.[9]
References
[2] T. Tominaka, Vector potential for a single helical current conductor, Nuc. Instr. Meth. A.
523:1, 2004.
[3] T. Tominaka, M. Okamura & T. Katayama, Analytic field calculation of helical coils,
Nuc. Instr. Meth. A. 459:398, 2001.
[4] R. Hagel, L. Gong & R. Unbehauen, On the magnetic field of an infinitely long helical
line current, IEEE Trans. Mag. 30:80, 1994.
[5] S. Park, J. Baird, R. Smith & J. Hirshfield, Exact magnetic field of a helical wiggler,
J. Appl. Phys. 53:1320, 1982.
[6] T. Tominaka, Magnetic field calculation of an infinitely long solenoid, Eur. J. Phys.
27:1399, 2006.
[7] D. Sagan, J. Crittenden, D. Rubin & E. Forest, A magnetic field model for wigglers and
undulators, Proc. 2003 Part. Accel. Conf., Portland, OR, p. 1023.
[8] J. Blewett & R. Chasman, Orbits and fields in the helical wiggler, J. Appl. Phys.
48:2692, 1977.
[9] R. Fernow & R. Palmer, Solenoidal ionization cooling lattices, Phys. Rev. Special
Topics – Accel. and Beams 10:064001, 2007.
6
AS 9.6.27.
9
Permanent magnets
Permanent magnets are ferromagnetic materials that retain significant magnetiza-

tion after the magnetizing current is removed.[1] In applications where space is
constrained, permanent magnets can provide stronger fields than electromagnets,
require no power source, and do not need cooling. After discussing the properties
of bar magnets and magnetic circuits, we consider some models of the properties of
magnets made from rare earth compounds. We conclude with a discussion of
assemblies of permanent magnets, which can be used to produce desired multipole
fields.
9.1 Bar magnets

Let us consider a cylindrical sample of permanent magnet material with uniform
magnetization M along the axis of the cylinder, as shown in Figure 9.1. Recall from
Equation 3.23 that the vector potential can be written as
ð ! ð !
! μ0 M n^ μ rM
A ¼ dS þ 0 dV:
4π R 4π R
Since M is uniform here, the second term vanishes. In addition, the first term
vanishes on the flat end faces. Thus A only depends on the surface contributions
around the sides of the cylinder. We assume the sources for the potential are
azimuthally directed Amperian currents with current density
! ¼! M ^n
Ka
^
¼ M ϕ: (9.1)
Thus the field in a bar magnet is analogous to the field in a solenoid, where the
Amperian currents here take the place of the conduction currents in a solenoid.
The magnetic flux density in the axial direction is then given by Equation 7.32
211
212 Permanent magnets
Figure 9.1 Bar magnet.
μ0 nI
Bz ð0; zÞ ¼ ðcos β2 cos β1 Þ;
2
where βi are the angles from the observation point along z to the two outer edges at
the ends of the cylinder. Making the substitution Ka ¼ n I; we get
μ0 K a
B¼ ðcos β2 cos β1 Þ
2
μM
¼ 0 ðcos β2 cos β1 Þ (9.2)
2
BR
¼ ðcos β2 cos β1 Þ;
2
where BR is the remanent field for the permanent magnet. For a point P inside the
magnet, cos β1 < 0 and cos β2 > 0. Thus B points along the positive z direction.
From Gauss’s law for a pillbox on one of the end faces, B must also be directed
along +z outside the magnet. We can find the magnetic intensity H from
μ0 H ¼ B μ0 M:
Thus [2]
Ka
HP ¼ ðcos β2 cos β1 Þ Ka
2
(9.3)
Ka
¼ ðcos β2 cos β1 2Þ:
2
Since the two cosine terms are both smaller than one, this expression is negative.
Thus H inside the magnet points in the opposite direction from M and B.
Now let us consider the situation at the point Q on the end face of the magnet.
In this case cos β2 ¼ 0 and
9.1 Bar magnets 213
μ0 K a
BQ ¼ cos β1 : (9.4)
2
On the inside surface of the end face, we find from Equation 9.3 that
Ka
HQin ¼ ðcos β1 þ 2Þ;
2
which points along −z. On the outer surface of the end face, M = 0 and we find from
Equation 9.4 that
Ka
HQout ¼ cos β1 ;
2
which points along +z. The behavior for H is similar to the case of the electric field
from a surface distribution of charge. It is sometimes convenient when modeling
bar magnets to assume that the end faces contain a distribution of fictitious
magnetic charges or “poles.” In terms of magnetic charges, we can express the
surface and volume charge densities as [3]
!
σm ¼ M ·^n
!
ρm ¼ r·M (9.5)
and the scalar potential as
ð ð
1 σm 1 ρm
Vm ¼ dS þ dV: (9.6)
4π R 4π R
Thus in a bar magnet, we can assume that B comes from the Amperian currents
flowing azimuthally around the cylinder and that H comes from magnetic charges
on the flat end faces. In a real bar magnet, M is typically weaker near the end faces
than it is in the central region. In this case, there will also be contributions to the
field from the volume integrals above.
Consider a bar magnet with radius a that is much smaller than its length L. This is
sometimes referred to as a “magnetic needle.”[4] We look at the field on the axis
outside the magnet at location z. We have
( 2 )1=2
a
cos β2 ¼ 1 þ :
z L=2
For a L,
2
1 a
cos β2 ≃ 1 þ
2 z L=2
and we have a similar expression for cos β1 with L → –L. Then from Equation 9.2,
μ0 M a 2 μ0 M a2
BðzÞ ¼ 2
:
4ðz L=2Þ 4ðz þ L=2Þ2
If we define the strength of the magnetic charge as

qm ¼ πa2 M; (9.7)
we can write the magnetic intensity as

qm qm
HðzÞ ¼ 2
:
4πðz L=2Þ 4πðz þ L=2Þ2
We interpret this as the field due to a positive magnetic charge at face 2 and a
negative charge at face 1. The field strength falls off like the inverse square distance
from the charge.
9.2 Magnetic circuit energized by a permanent magnet

Consider the arrangement shown in Figure 9.2, where a pair of permanent magnets
surround an open gap on one side and are connected by an iron path on the other.[5]
We know from the Ampère law that
þ
! !
H · dl ¼ 0
around the circuit since there are no conduction currents. Assuming the gap is small
and that the leakage flux is negligible, this gives
Figure 9.2 Magnetic circuit containing permanent magnets.

9.2 Magnetic circuit energized by a permanent magnet 215
Hg Lg Hm Lm þ Hi Li ¼ 0;
where the subscripts g, m, and i refer to gap, magnet, and iron. The quantity Hm Lm
takes the place of NI in magnetic circuits powered by conductors. Neglecting
leakage, the flux is constant around the loop, so we have
ΦB ¼ Bi Ai ¼ Bg Ag ¼ Bm Am ;
where A is the cross-sectional area. Combining these equations, we have

Hg Lg Hi Li
Hm Lm ¼ Bg Ag þ
Bg Ag B g Ag

Lg Li (9.8)
¼ Bm Am þ
μ0 Ag μAi
¼ ΦB R;
where the series reluctance is

Lg Li
R¼ þ :
μ0 Ag μAi
Equation 9.8 is analogous to Ohm’s law with Hm Lm corresponding to the applied

voltage.
The magnetic energy stored in the gap is
Wg ¼ ½ μ0 B2g Ag Lg
¼ ½ H g L g Bg Ag :
Since μ μ0 , we can neglect the reluctance through the iron. Then
Hm Lm ≃ Hg Lg
and we can write

Wg ≃ ½ Bm Hm Am Lm : (9.9)
Thus the stored energy in the gap is proportional to the BH product and the volume
of the permanent magnet.
Since B and H point in opposite directions, permanent magnets operate in
the second quadrant of the hysteresis curve. To maximize the energy stored in the
gap, it is desirable to operate at a point where the BH product is maximum.
Rewriting Equation 9.8 as
3.5
2.5
1.5
0.5
0
0 1 2 3 4 5
Figure 9.3 POISSON model of a magnetic circuit energized by a pair of perma-

nent magnets.
Bm Lm
¼ ;
H m Am R
the quantities on the right-hand side can be adjusted to get the B-H “load line” for
the magnetic circuit to pass through the point where the BH product is maximum.
Figure 9.3 shows a simple POISSON model1 of a magnetic circuit energized by
a pair of permanent magnets. The figure shows one quarter of a symmetric circuit.
The x and y axes are symmetry planes in this figure. The use of a 2D program such
as POISSON assumes that the configuration is uniform over a large distance in the
third dimension (into the figure).
The permanent magnet material is located in the rectangle near the origin.
The magnetization was oriented in the vertical direction. Considerable field fring-
ing is evident in the air gap for this simple geometry. The amount of fringing
depends strongly on the type of permanent magnet material that is used.[1]
9.3 Material properties

Characteristic properties of some permanent magnet materials are given in
Table 9.1. Alnico is an alloy of iron with aluminum, nickel, and cobalt. It has
1
The calculation was actually made with the PANDIRA program in the POISSON distribution.
9.4 Model for rare earth materials 217
Table 9.1 Properties of permanent magnet materials [6, 7]
BR [T] HC [kOe]1 HCi [kOe] BHmax [MG Oe]

Alnico 0.83–1.25 0.6–6.4 0.6 – 1.9 1.4–5.5
Ceramic ferrite 0.22–0.39 1.9–3.2 2.5–3.3 1–3.6
SmCo 0.81–1.15 7.2–10.6 9–18 16–32
NdFeB 0.98–1.35 7.5–12.8 8–26 24–45
1
1 Oe = 1 10–4 T/ μ0 :
Figure 9.4 Second quadrant of the hysteresis curve.
a high remanent field but is easily demagnetized and has a large leakage flux.[1]
Ferrite contains iron oxides (Fe2 O3 ). It has a low BR, but it is a cheap material.
Ceramic ferrites are compounds of barium or strontium ferrite. Samarium and
neodymium are rare earth elements.[8] They have large values of ðBHÞmax , can
produce a large field from a compact magnet, are very resistant to demagnification,
and have small leakage flux.
9.4 Model for rare earth materials

Usually there is some direction in a crystalline material along which the moments in
the material tend to align. This is referred to as the “easy” magnetization direction.
A permanent magnet can be produced so that the maximum magnetization is along
some desired direction. A useful model has been developed for describing the mag-
netic properties of rare earth materials.[1, 9] Figure 9.4 shows the second quadrant of
a typical hysteresis curve for the easy direction. It is a good approximation to assume
the relative permeability in this region is equal to 1, so we have the relation
BR ≃ μ 0 H C :
In this case, the fields from assemblies of blocks can be linearly superimposed.
If the magnetization is anisotropic, the relation between B and H can be written as
Bk ¼ μk Hk þ BR
(9.10)
B? ¼ μ ? H ? ;
where k refers to the easy direction in the material and ? refers to the direction
perpendicular to it. For convenience, we define the reluctivity γ = 1/μ. Then from
BR
Hk ¼ γk Bk
μk (9.11)
≃ γk Bk HC :
These equations can be combined into the vector relations

! ! ! !
B ¼ μk H k þ μ? H ? þ B R (9.12)
and
! ! ! !
H ¼ γk B k þ γ? B ? H C : (9.13)
The vectors BR and HC are directed in opposite directions along the easy axis.
Taking the divergence of Equation 9.12, we find that
! ! !
r·ðμk H k þ μ? H ? Þ ¼ r· B R ≡ ρm ; (9.14)
which relates H to the density of magnetic charges. Taking the curl of Equation 9.13
in a region with no conduction currents, we get
! ! ! !
r ðγk B k þ γ? B ? Þ ¼ r H C ≡ Jm ; (9.15)
which relates B to the Amperian currents. Thus the material can be treated
!
magnetically as vacuum together with either a charge density r · B R or
!
a current density r H C : For homogeneous materials, these charges or currents
vanish everywhere except on the surface.
The scalar potential for the permanent magnet material can be written as
ð
1 ρm
μ 0 Vm ¼ dV
4π R
ð !
1 ðr· B R Þ
¼ dV; (9.16)
4π R
9.5 Rare earth model in two dimensions 219
where R is the distance between the source point and the field point. If the material
is inhomogeneous, we can define G = 1/R and make use of Equation B.3
! ! !
r·ðG B R Þ ¼ G r· B R þ B R ·rG
to write Equation 9.16 in the form

ð
1 h ! ! i
μ 0 Vm ¼ r·ðG B R Þ B R ·rG dV:
4π
Using the divergence theorem on the first part of the integrand,

ð ð
! !
r·ðG B R Þ dV ¼ G B R ·^n dS ¼ 0;
which vanishes because we can choose the surface to lie outside the material where
BR ¼ 0. We also have
!
R
rG ¼ 3 :
R
Thus the scalar potential for inhomogeneous material is
ð! !
1 B R· R
μ 0 Vm ¼ dV: (9.17)
4π R3
For homogeneous material, we can use the divergence theorem in Equation 9.16
and find that
ð!
1 B R ·^
n
μ 0 Vm ¼ dS
4π R
! ð
B R n^
¼ · dS; (9.18)
4π R
where the final integral only involves geometric quantities.
9.5 Rare earth model in two dimensions

If the permanent magnet pieces are uniform over a long distance in the z direction
and the magnetization does not have a component along z, then it is possible to
make a two-dimensional approximation of the fields.[9] From Equation 5.40, the
field at the observation point zo is
ðð
μ σ
B ðzo Þ ¼ 0 dx dy: (9.19)
2πi zo z
We assume the remanent field can have both x and y components
BR ¼ BRx þ i BRy :
From the curl B = μ0 J equation, we have
∂x BRy ∂y BRx ¼ μ0 σ:
Substituting back into Equation 9.19, we get
1
B ðzo Þ ¼ ½I1 I2 ; (9.20)
2πi
where
ðð
∂x BRy
I1 ¼ dx dy
zo x i y
ðð
∂y BRx
I2 ¼ dx dy:
zo x i y
Considering the first integral, we integrate over x by parts with

1
u¼
zo x i y
dv ¼ ∂x BRy dx;
which gives
ð" ð #
BR y BR y
I1 ¼ dx dy
zo x i y ðzo x i yÞ2
þ ðð
BR y BR y
¼ dy dx dy:
zo x i y ðzo x i yÞ2
The first term vanishes for a line integral evaluated outside the material where
BR ¼ 0. Thus we have
ðð
BR y
I1 ¼ dx dy:
ðzo x i yÞ2
9.6 Multipole expansion for continuously distributed material 221
For the integral I2 , we integrate over y by parts to find

ðð
BR x
I2 ¼ i dx dy:
ðzo x i yÞ2
Substituting back into Equation 9.20, we find that [9]

ðð
1 BR
B ðzo Þ ¼ dx dy: (9.21)
2π ðzo zÞ2
If the material is homogeneous,

ðð
BR 1
B ðzo Þ ¼ dx dy
2π ðzo zÞ2
we can apply the complex Green’s function, Equation 5.39, with
BR 1
F¼ :
2π zo z
The field for a homogeneous block can then be written in terms of the contour
integral
þ
BR dz
B ðzo Þ ¼ : (9.22)
4π i zo z
It is also possible by direct integration to write this in the alternate forms [9]
þ
BR dx
B ðzo Þ ¼
2π i zo z
þ
BR dy
¼ : (9.23)
2π zo z
9.6 Multipole expansion for continuously distributed material

We next consider the problem of assembling permanent magnet material to
make a 2D multipole magnet.[9, 10] In a multipole magnet, we try to make some
multipole order N as large as possible, while at the same time making all the other
orders small. We will mainly be concerned here with the magnetic field inside the
aperture of the magnet. The field in general is given by Equation 9.21. In order to
study the multipole structure of the field, it is convenient to first expand the
denominator in a power series. We start with
1 1 X∞
zm
¼
¼ o
:
zo z z 1 zzo m¼0
zmþ1
If we let n = m + 1, then
1 X∞ n1
zo
¼ : (9.24)
zo z n¼1
zn
Taking the derivative with respect to z of both sides gives
1 X
∞
n zn1
¼ o
: (9.25)
ðzo zÞ2 n¼1
znþ1
Substituting back into Equation 9.21, we have

ð
1 X∞
n1 BR
B ðzo Þ ¼ nz dS:
2π n¼1 o znþ1
We identify the nth multipole field contribution as

ð
n BR
Bn ¼ dS; (9.26)
2π znþ1
so the field in the aperture has the multipole expansion

X
∞

B ðzo Þ ¼ o :
Bn zn1 (9.27)
n¼1
It is possible to express Bn as a contour integral by defining

BR n
F¼ z
2π
in the Green’s theorem in Equation 5.39, which gives
þ
BR dz
Bn ¼ : (9.28)
4π i zn
Now let us consider the design of an ideal multipole magnet. We use a polar
coordinate system with z ¼ r eiϕ . We assume the material is located in an
annular region between the radii r1 and r2 . Assume that we have the freedom
to specify the direction of the easy axis everywhere in the material. Let the easy
axis in the material located at angle ϕ be rotated through an angle βðϕÞ with
9.6 Multipole expansion for continuously distributed material 223
respect to its direction at ϕ = 0. The contribution to the field from the material at the
angle ϕ is
Br ðϕÞ ¼ eiβðϕÞ BR ð0Þ:
From Equation 9.26, the multipole field contribution is

ðð
n BR expfi½βðϕÞ ðn þ 1Þϕg
Bn ¼ r dϕ dr: (9.29)
2π rnþ1
The first term in the square brackets comes from the easy axis distribution
and the second term comes from the azimuthal dependence in zn+1 in
Equation 9.25. We can make the multipole as large as possible by making
the quantity in square brackets equal to 0. This determines the constraint on
the easy axis angles
βðϕÞ ¼ ðn þ 1Þ ϕ: (9.30)
Multipole fields in the aperture can be produced by choosing n to be a positive

integer N. For example, N = 1 produces a uniform dipole field in the aperture with
no field outside the permanent magnet ring. Choosing n to be a negative integer on
the other hand, produces a field outside the ring and no field in the aperture.[11] For
example, n = –1 produces a dipole field outside the ring. Further control over the
field can be obtained by surrounding the permanent magnet ring with an additional
soft-iron ring.[11]
Confining our attention to 2N-multipole fields inside the aperture, setting n = N in
Equation 9.30 and substituting into Equation 9.29 we find
ð
n BR r2 1
Bn ¼ I dr;
2π r1 rn
where
ð 2π
I¼ ei ðNnÞϕ dϕ
0

2π for n ¼ N
¼
0 for n ≠ N:
For a dipole assembly, N = 1, and we have

r2
B1 ¼ BR ln ; (9.31)
r1
while for higher order multipoles, N ≥ 2, we get

ð r2
dr
BN ¼ N BR N
r1 r
" N1 # (9.32)
N BR 1 r1
¼ 1 :
N 1 r1 N1 r2
The 2N-multipole field at the point zo is given by
B ðzo Þ ¼ BN zN1
o :
9.7 Segmented multipole magnet assemblies

In practical terms, the continuous distribution of easy axis directions discussed in
the previous section can only be approximated by using a finite number of
magnetized blocks. As a result, the strength of the desired multipole is reduced
and unwanted multipole orders are introduced.
Assume that the multipole magnet is constructed from M geometrically identical
rare earth trapezoidal blocks.[9] Each block is separated from its nearest neighbor
by the angle 2π/M around the z axis. In each block, the easy axis direction is chosen
to approximate the ideal angle given in Equation 9.30. The direction of the easy
axis inside a block is made to rotate by the angle ðN þ 1Þ 2π=M from a given block
to its following neighbor. For example, Figure 9.5 shows a dipole magnet (N = 1)
made up of M = 8 rare earth blocks. In this assembly, neighboring blocks are
Figure 9.5 Dipole magnet made up from an assembly of trapezoidal-shaped

permanent magnets.
9.7 Segmented multipole magnet assemblies 225
separated geometrically by the angle π/4 and the easy axis rotates by π/2 from block
to block.
Let Cn be the contribution to the multipole Bn for some reference block. Then the
contribution of block m to Bn is

2π 2π
Cn exp i m ðN þ 1Þ exp i m ðn þ 1Þ ;
M M
where the first exponential comes from the rotation of the easy axis and the second
exponential comes from the definition of Bn . The sum of all M blocks gives
X
M1
Nn
Bn ¼ C n exp i 2πm ;
m¼0
M
where m = 0 refers to the reference block. If the quantity ðN nÞ=M is a positive or

negative integer, then by writing the exponential in terms of cosines and sines, we
see that the value of the sum is M. If ðN nÞ=M is not an integer, then the value of
the sum is zero.[9] Thus we have from Equation 9.27
X
∞
B ðzo Þ ¼ M Cn zN1
o ; (9.33)
ν¼0
where ν is an index to the set of allowed multipoles. The multipole order that
corresponds to a given ν is given by [9]
n ¼ N þ νM: (9.34)
Next we turn to finding the reference multipole Cn . Substituting the series

expansion Equation 9.24 into the contour integral 9.23, we find we can express
Cn as
þ
BR dx
Cn ¼ :
2π i zn
Figure 9.6 shows the geometry of a trapezoidal block. Only the two segments
marked 1 and 2 contribute to the contour integral. On path 1 we have
y ¼ x tan α;
where α = π/M. On path 2 we have
y ¼ x tan α:
Figure 9.6 Trapezoidal block.
Thus the coefficient can be written as

ð r2 ð r1
BR dx dx
Cn ¼ n þ n
2π i r1 ðx i x tan αÞ r2 ðx þ i x tan αÞ
ð r2 ð r1
BR 1 dx 1 dx
¼ n þ n :
2π i ð1 i tan αÞ r1 x n ð1 þ i tan αÞ r2 xn
We can express
e inα
ð1 i tan αÞn ¼ :
cosn α
Performing the integrals and rearranging terms, we get

" n1 #
BR cosn α r1
Cn ¼ 1 ðeinα einα Þ:
2π i ðn 1Þrn1
1 r 2
Writing the exponential term as a sine function, we find that Cn for the trapezoidal
block is
" n1 #
BR r1
Cn ¼ 1 cosn α sin nα: (9.35)
π ðn 1Þr1n1 r2
Substituting this back into Equation 9.33, we find that the field inside the aperture is
given by
" n1 #
X ∞ n1
zo n r1
B ðzo Þ ¼ BR 1 Kn ; (9.36)
ν¼0
r 1 n 1 r2
References 227
Table 9.2 Properties of a quadrupole magnet assembly
M K2 B2 ðr1 Þ [T] 1st allowed harmonic [T]

4 0.32 0.454 B6 ðr1 Þ ¼ 0:030
8 0.77 1.095 B10 ðr1 Þ ¼ 0:086
12 0.89 1.270 B14 ðr1 Þ ¼ 0:086
16 0.94 1.336 B18 ðr1 Þ ¼ 0:077
where n is related to ν by Equation 9.34 and we define the segmentation efficiency

factor as [9]

nπ
π sin M
Kn ¼ cosn : (9.37)
M nπ
M
The factor Kn measures how well a segmented magnet with M blocks approximates
the idealized magnet with continuous variation of the easy axis that we described in
the previous section.
Example 9.1: quadrupole magnet assembly

As an example, let us consider a quadrupole (N = 2) assembly made up of trapezoidal
blocks. We assume BR ¼ 0:95 T and that r2 =r1 ¼ 4. We choose the observation point
on the inner edge of the permanent magnet material, i.e., zo ¼ r1 . Table 9.2 shows
some properties of the assembly as a function of the number of blocks M used around
the circumference. The second column gives the segmentation efficiency factor for
the quadrupole moment. We see that the efficiency increases quickly with the number
of blocks used in the assembly. The third column gives the fundamental multipole
contribution to the field. Already with 12 blocks, this contribution is almost 90% of
the ideal unsegmented value. The last column shows the contribution to the field from
the first allowed harmonic term. The strength of this term is ~7% of that for the
fundamental.
Assemblies of permanent magnet blocks can be used to make many magnetic
configurations, including dipoles, quadrupoles, sextupoles, solenoids, and periodic
transverse fields.[1, 10]
References
[1] J. Bahrdt, Permanent Magnets Including Undulators and Wigglers, Proc. CERN
Accelerator School on Magnets, Bruges, Belgium, June 2009, CERN-2010-004,
p. 185.
[2] D. Tomboulian, Electric and Magnetic Fields, Harcourt, Brace & World, 1965, p. 240.
[3] L. Eyges, The Classical Electromagnetic Field, Dover, 1980, p. 139.
[4] W. Smythe, Static and Dynamic Electricity, 2nd ed., McGraw-Hill, 1950, p. 434.
p. 409.
[6] High Performance Permanent Magnets, Magnet Sales and Manufacturing, Inc.,
Culver City, CA, 1993.
[7] A. Chao & M. Tigner, Handbook of Accelerator Physics and Engineering, World
Scientific, 1999, p. 366.
[8] J. Becker, Permanent magnets, Sci. Am. 223:92, 1970.
[9] K. Halbach, Design of permanent multipole magnets with oriented rare earth cobalt
material, Nuc. Instr. Meth. 169:1, 1980.
[10] A. Chao & M. Tigner, op. cit., p. 468.
[11] Q. Peng, S. McMurry & J. Coey, Cylindrical permanent magnet structures using
images in an iron shield, IEEE Trans. Mag. 39:1983, 2003.
10
Time-varying fields
Until this point, we have only examined magnetic effects due to steady currents or
magnetic materials in stationary configurations. In this chapter, we will partially
relax this constraint by considering phenomena where there are slow variations in
current or magnetic flux. By slow, we mean slow enough that we can ignore all
effects of electromagnetic radiation. We begin with a discussion of Faraday’s law,
which presents another connection between electric and magnetic phenomena.
This is followed by a more detailed discussion of the energy associated with
a magnetic field, including the energy loss from the hysteresis cycle in ferromag-
netic materials. We find that Faraday’s law leads to the production of eddy currents
in some materials, while the skin effect can restrict currents to a layer near the
surface. We introduce the displacement current, which finally allows us to give
a complete set of Maxwell’s equations for stationary media. We conclude the
chapter with a brief discussion of magnetic measurements.
10.1 Faraday’s law

Michael Faraday discovered that a changing magnetic flux through a wire circuit
C induced a voltage in the wire.
dΦB
V∝ (10.1)
dt
The changing flux could be due to changing the current in C itself, changing the current
in a second, nearby circuit, moving a second circuit or permanent magnet with respect
to C, or changing the shape of C. Here we will mostly consider effects due to explicit
changes in the current, in which case we can replace the total time derivative in
Equation 10.1 with a partial derivative. According to Lenz’s law, the voltage induced
by the changing flux is such as to induce a current that gives rise to an additional flux
that opposes the original change in flux. Thus Faraday’s law can be written as
229
230 Time-varying fields
Figure 10.1 Electromotance induced in a rectangular loop.
þ ð
!! ∂ !
ℰ ¼ E ·dl ¼ B ·^n dS; (10.2)
∂t
where E is the electric field intensity and the surface S is bounded by the closed
circuit. The field E acts on a distance element dl in its rest frame. Because of the
tangential boundary condition on E, it follows that C can refer to any closed loop in
space, not just a physical circuit.[1] The line integral1 on the left side of
Equation 10.2 is called the electromotance ε. If the flux links a coil with N turns,
the electromotance must be multiplied by N. Since the contour integral is non-zero,
the induced electric field in this case is nonconservative, i.e., work is done on
a charge going around the contour.
Example 10.1: ε induced in a current loop

Consider a rectangular loop near a wire with increasing current I, shown in
Figure 10.1. The time-dependent field from the wire is
! μ ^
B ðtÞ ¼ 0 IðtÞ ϕ:
2πρ
The flux through the square loop is

ð Rþw
μ0 IðtÞ dρ
ΦB ¼ h
2π R ρ

μ0 IðtÞ Rþw
¼ h ln
2π R
1
Historically, ε has also been referred to as an emf or electromotive force.
10.1 Faraday’s law 231
and the electromotance is

μ0 dI Rþw
ε¼ h ln :
2π dt R
Using Stokes’s theorem in Equation 10.2, we find

ð ð
! ! ∂ ! !
ðr E Þ· dS ¼ B · dS
∂t
ð !
∂ B !
¼ · dS :
∂t
Then since the surface S is arbitrary, we find the differential form of Faraday’s law is
!
! ∂B
r E ¼ : (10.3)
∂t
This equation is valid at any point in space. We can relate this to the vector potential by
! ∂ !
r E ¼ ðr A Þ
∂t
!
∂A
¼ r ;
∂t
so that
!!
! ∂A
r E þ ¼ 0:
∂t
Since its curl vanishes, the quantity in parentheses must be the gradient of a scalar
function, which we denote Ve .
!
! ∂A
rVe ¼ E þ :
∂t
Thus the electric field

!
! ∂A
E ¼ rVe : (10.4)
∂t
can arise from static charge distributions or from time-varying magnetic fields.
From Equation 1.32, the inductance is related to the flux by
L I ¼ N ΦB :
Taking the time derivative of both sides, we find that an alternative definition
of L is
ε
L¼ : (10.5)
dI=dt
10.2 Energy in the magnetic field

We return now to the subject of the energy associated with a magnetic field.
Consider a current element in an isolated loop together with an associated power
source. Suppose that we want to increase the current in the loop from 0 up to some
value I. For each step in the process of raising the current, the source must produce
a voltage change
!
dVe ¼ rVe ·dl
across the current element and the source must supply the power
dP ¼ I dVe
!
¼ J dA ðrVe ·dl Þ;
where A is the cross-sectional area of the conductor. Since J and dl are parallel, we
can use Equation 10.4 to write
!!
! ∂A !
dP ¼ E þ · J dτ;
∂t
where, to minimize confusion, we use dτ in this section to represent the volume

element. The total power provided by the source for the full loop is then [2]
ð ! !
! ! ∂A !
P¼ E·J þ · J dτ:
∂t
The first term on the right side is the power used to compensate for energy losses
from heating in the conductor. The second term is the power used to set up the
magnetic field associated with the increasing current, which is the subject of
interest here. If we let W represent the energy stored in the magnetic field, then
10.2 Energy in the magnetic field 233
ð !
dW ∂A !
¼ · J dτ: (10.6)
dt ∂t
Consider a small volume element of the conductor where J can be considered

constant.2 Then we can write A as the product of J with a factor that only depends
! !
on the geometry. Thus we can assume that A ¼ α J , where α is a constant. Then
substituting
!
∂ !! ! ∂A
ð A · J Þ ¼ 2J ·
∂t ∂t
into Equation 10.6, we find the energy stored in the magnetic field is
ð
!!
W ¼ ½ J · A dτ: (10.7)
! !
If the current distribution is a current loop, we let J dτ→I dl and Equation 10.7
becomes
ð
!!
W ¼ ½ I A ·dl :
This can be expressed in terms of the magnetic flux by
W ¼ ½ I ΦB : (10.8)
Returning again to Equation 10.7, we can use the curl H = J equation to write the
energy as
ð
! !
W ¼ ½ ðr H Þ· A dτ:
Using the vector identity B.4, we find

ðh
! ! ! !i
W¼½ H ·ðr A Þ r·ð A H Þ dτ:
Rewriting the first term in terms of B and using the divergence theorem in
the second, we get
2
J.D. Jackson, Classical Electrodynamics, Wiley, 1962, p. 176.
ð ð
!! ! !
W ¼ ½ H · B dτ þ ½ H A ·^n dS:
Looking at the surface integral, we know that the field from a conductor element falls
off like 1=R2 and the vector potential falls off like 1=R, while the surface area only
grows like R2 . By evaluating at a sufficiently large distance, the second integral
vanishes. Thus the energy stored in the magnetic field from conduction currents is
ð
!!
W ¼ ½ B ·H dτ (10.9)
and the magnetic energy density in the field is

!!
wB ¼ ½ B ·H : (10.10)
The energy of a permeable body with magnetization M in an applied magnetic

field Ba can be expressed as [3]
ð
!!
W ¼ ½ M · B dτ: (10.11)
10.3 Energy loss in hysteresis cycles

Consider a Rowland ring containing a ferromagnetic sample, as discussed in
Section 2.5. If we increase the current in a conductor wound around the sample,
we get an induced electromotance that opposes the change in current. The extra
power expended by the source is
dW dΦB
¼ NI
dt dt
dB
¼ NIA
dt
NI dB
¼ Al ;
l dt
where N is the number of conductor turns, A is the cross-sectional area of the
sample, l is the mean circumference of the ring, and B is the average flux density
inside the sample. Using the Ampère law, this can be written
dW dB
¼ HV ;
dt dt
where V is the volume of the sample.
Now consider the hysteresis loop shown in Figure 10.2. The energy supplied by
the source in moving from point a to point b along the loop is
Figure 10.2 Energy loss in a hysteresis loop.
ðb
Wab ¼ V H dB:
a
Since dB is the independent variable, the value of this integral is the area projected
on the B (vertical) axis in the figure. Going from point b to point c along the loop,
I is in the same direction, but is decreasing. Thus the electromotance changes sign
and some energy is returned to the source.
ðc
Wbc ¼ V H dB:
b
The sum of these two integrals is the area inside the hysteresis loop in the first
quadrant. If we continue this analysis for a complete cycle, we find that the net
energy lost in the ferromagnetic material per cycle is [4]
þ
W ¼ V H dB: (10.12)
This energy loss can be minimized by choosing ferromagnetic materials with

a narrow hysteresis loop.
10.4 Eddy currents

Faraday’s law shows that time-varying magnetic fields produce a voltage in
materials such as conductors, iron, or mechanical supports. If a closed path
exists inside the material, this voltage can drive currents, known as eddy
currents.[5] The eddy currents can in turn create new magnetic fields that are
superimposed over the original field. Eddy currents can be used for a number of
desirable purposes, including displacement and position measurements, induc-
tion heating, magnetic shielding, levitation, and braking. On the other hand,
undesirable effects from eddy currents include resistive power losses, Lorentz
forces, multipole errors in a desired field, and a time lag in reaching an

equilibrium field value.
Starting from Faraday’s law
!
! ∂B
r E ¼ ;
∂t
multiplying both sides by the electrical conductivity σ and taking the curl, we find
! ∂ !
r ðr σ E Þ ¼ σ ðr B Þ:
∂t
We can write Ohm’s law in the form
! !
J ¼ σE: (10.13)
The range of current densities over which this linear relation holds depends on the
material. Thus we have
! ∂ !
r ðr J e Þ ¼ σ μ ðr H Þ;
∂t
where Je is the eddy current density. We can use the vector identity B.7 on the left
side of this equation and the curl H = J equation on the right side to get
!
! ! ∂J e
rðr· J e Þ r2 J e ¼ σ μ :
∂t
Since the divergence term on the left side vanishes, we find that [6]
!
2! ∂J e
r J e ¼ σμ : (10.14)
∂t
This is a form of the diffusion equation. The rate of build-up of the eddy currents is
controlled by the factor σμ.
If instead, we begin by taking the curl of the Ampère law, we find
! !
r ðr H Þ ¼ r J e
!
¼σr E:
Applying Equation B.7 on the left-hand side and Faraday’s law on the right, we
obtain the equation
!
2! ∂H
r H ¼ σμ : (10.15)
∂t
Figure 10.3 One-quarter of a symmetric H-dipole.
Thus the magnetic field associated with the eddy currents also satisfies a diffusion
equation with the same characteristic constant. If one specifies the time depen-
dence for H and the geometry of the configuration, the diffusion equation can be
solved for the spatial and time dependence of the magnetic field due to eddy
currents.[5] This may lead to a series of terms, each with its own characteristic
time dependences.
Example 10.2: time constant for eddy currents in a solid iron core
Consider a long H-dipole with a solid iron yoke, as shown in Figure 10.3. For slow
time changes, eddy currents can flow throughout the volume of the iron yoke
surrounding the coil.[7] The magnetic flux from the eddy currents is not symmetric
with the flux from the coils, which causes the iron saturation to vary with transverse
position.
Consider a path through the iron yoke at the midplane in the region 0 ≤ x ≤ d.
Assuming there is no leakage flux, all of the return flux from the conductor has to
pass across this path. Assume the current in the conductor is changing with time.
Then the magnetic field in the vicinity of the path is in the y direction, the induced
electric field is in the z direction, and on the midplane both fields are only functions
of x and t.
! ¼ B ðx; tÞ ^y
B y
!
E ¼ Ez ðx; tÞ ^z :
The eddy currents flow parallel to the z axis until they reach the magnet end,
where they reverse direction and flow back at the symmetric (x, y) position on
the other side of the magnet. From the Ampère and Ohm’s laws, we have
∂By ðx; tÞ
¼ σ μ Ez ðx; tÞ;
∂x
while from Faraday’s law
∂By ðx; tÞ ∂Ez ðx; tÞ

¼ :
∂t ∂x
Applying the Laplace transform to the time variable for these two equations,[8]
we get
∂x By ðx; pÞ ¼ σ μ εz ðx; pÞ (10.16)
and
p By ðx; pÞ ¼ ∂x εz ðx; pÞ; (10.17)
where p is the variable conjugate to t in the Laplace transform. Taking the derivative
of Equation 10.16 with respect to x and substituting Equation 10.17, we get
∂2x By ðx; pÞ ¼ σ μ p By ðx; pÞ:
Defining k 2 ¼ σμp, the solution for the magnetic field consistent with the boundary
conditions is [7]
By ðx; pÞ ¼ B0 cosh kx:
The field across the return yoke is asymmetric and is larger on the side nearer the coil.
At the edge of the path closest to the conductor, we have
pffiffiffiffiffiffiffiffiffiffi
kd ¼ d σμp
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ σ μ p d2
pffiffiffiffiffiffi
¼ p τ;
where [7]
τ ¼ σμd2 : (10.18)
The variable τ has the dimensions of time. It gives a characteristic time for eddy
current effects in this configuration. Note that it depends quadratically on the width
d of the return yoke.
Eddy currents can be suppressed by restricting the rate of change of the desired
field or by constructing the magnet in such a way that potential eddy current loops
are minimized. Magnet yokes are frequently constructed by assembling thin iron
laminations for this reason.
10.5 Skin effect 239
10.5 Skin effect

Consider a current density with the periodic time variation
! ! iωt
J ¼ J0e ;
where ω is the angular frequency and J0 only depends on the spatial dimensions.
The diffusion equation for the current density, analogous to Equation 10.14, is
!
2!
∂J
r J ¼ σμ
∂t
!
¼ i ωσ μ J :
Defining ζ 2 ¼ iωσμ, we obtain
! !
r2 J ζ 2 J ¼ 0: (10.19)
Now assume that the current is flowing along a conducting slab that occupies the
space y ≤ 0. Then the component of J flowing in the z direction, for example, is
Jz ¼ Jz0 eiωt :
Applying Equation 10.19 to this, we find
d2 Jz0
¼ ζ 2 Jz0 :
dy2
This differential equation has the solution
Jz0 ¼ JS eζ j y j ;
where JS is the spatial dependence of the current density on the surface of the slab.
Using the relation
i ¼ ½ ð1 þ 2 i 1Þ ¼ ½ ð1 þ iÞ2 ;
we can write
1 pffiffiffiffiffiffiffiffiffi
ζ ¼ pffiffiffi ð1 þ iÞ ωσμ:
2
The boundary condition for large | y | eliminates the negative solution for ζ. Thus
rffiffiffiffiffiffiffiffiffi
ωσμ
ζ ¼ ð1 þ iÞ :
2
Defining the skin depth as

sffiffiffiffiffiffiffiffiffi
2
δ¼ ; (10.20)
ωσμ
the solution for the current density is [9]

jyj jyj
Jz ¼ JS exp exp i ωt : (10.21)
δ δ
We see that the current density decreases exponentially with distance into the
surface. In addition, there is a phase shift of the current flowing inside the material
with respect to the current flowing on the surface. These effects scale with the skin
depth parameter δ. For a copper conductor with current varying at 1 kHz, the skin
depth is ~2.1 mm.
10.6 Displacement current

!
We have seen in Chapter 1 that r· J ¼ 0 in magnetostatic problems. However,
once we allow for time variations, the conservation of charge requires
! ∂ρ
r· J þ ¼ 0; (10.22)
∂t
where ρ is the electric charge density of free (i.e., unbound) charges. Therefore,
when time variation is allowed, the divergence of the conduction current density no
longer needs to vanish. From electrostatics, we know that [10]
!
r· D ¼ ρ; (10.23)
The vector D is called the electric flux density3 and is related to the electric field
intensity by
! !
D ¼ε E (10.24)
for linear materials, where ε is the permittivity. Taking the time derivative of
Equation 10.23, we get
!
∂ρ ∂D
¼ r· : (10.25)
∂t ∂t
3
Historically, the vector D is also known as the electric displacement.
Table 10.1 Maxwell’s equations
Differential form Integral form

! ð ð
r· D ¼ ρ ! !
D · dS ¼ ρ dV
! þ ð !
! ∂B !! ∂ B !
r E ¼ E ·dl ¼ · dS
∂t ∂t
ð
! ! !
r· B ¼ 0 B · dS ¼ 0
! þ ð !!
! ! ∂D !! ! ∂D !
r H ¼ J þ H ·dl ¼ J þ · dS
∂t ∂t
!
Comparing Equations 10.22 and 10.25, we see that the quantity ∂ D =∂t acts
like an additional kind of current. Thus we define the displacement current
density as
!
! ∂D
Jd¼ : (10.26)
∂t
Taking this into account, the Ampère law must then be modified as [11]
!
! ! ∂D
r H ¼ J þ : (10.27)
∂t
This shows that a magnetic field can also be produced by a time-varying electric
field.
At this point, we can summarize the complete set of Maxwell’s equations for
stationary media in Table 10.1. It is important to keep in mind that writing the
equations in this form assumes the validity of the constitutive relations
! ¼μ!H
B
! !
J ¼σ E
! !
D ¼ε E:
10.7 Rotating coil measurements

Magnetic fields can be measured using a number of techniques. Nuclear magnetic
resonance (NMR) probes are used for high-precision measurements.[12] Hall
effect probes are simple to use and are commercially available.[13] Other common
methods of measuring the magnetic field are based on electromagnetic induction.
Figure 10.4 Field measurement with a radial coil.
One technique, which relates directly with our previous discussions of the multi-
pole content of fields, involves measurements in long magnets with large aperture
using a rotating coil.[14] The azimuthal component of the field can be measured
using a radial coil, the principle of which is shown in Figure 10.4. The flux through
the wire loop with N turns is
ð r2
ΦB ðθÞ ¼ N L Bθ ðr; θÞ dr
r1
X
∞ ð r2
¼ N L ðAn sin nθ þ Bn cos nθÞ rn1 dr
n¼1 r1
X∞ n
r2 rn1
¼ N L ðAn sin nθ þ Bn cos nθÞ ;
n¼1
n
where we have used Equation 4.8 to express the azimuthal field in terms
of multipole field components. If the coil rotates at a constant rate, we have θ ¼
ω t and
dΦB dΦB dθ dΦB

¼ ¼ω :
dt dθ dt dθ
The induced voltage in the coil from Faraday’s law is then

X
∞
VðθÞ ¼ ω N L ðAn cos nθ Bn sin nθÞ ðrn2 rn1 Þ: (10.28)
n¼1
Performing a Fourier analysis on the voltage signal allows the multipole coeffi-
cients to be determined.
Figure 10.5 Field measurement with a tangential coil.
ð 2π X
∞ ð 2π
VðθÞ sin mθ dθ ¼ ωNL ðr2 r1 Þ An cos nθ sin mθ dθ
n n
0 n¼1 0
ð 2π
Bn sin nθ sin mθ dθ ¼ ωNLðrn2 rn1 Þ Bn π δmn :
0
Thus
ð 2π
1
Bm ¼ VðθÞ sin mθ dθ: (10.29)
π ωN L ðrm
2 r1 Þ
m
0
Similarly we find that

ð 2π
1
Am ¼ VðθÞ cos mθ dθ: (10.30)
π ωN L ðrm
2 r1 Þ
m
0
It is possible to do a similar analysis on the radial component of the field Br using

the rotating tangential coil illustrated in Figure 10.5. Using Equation 4.7, we have
ð θþδ
ΦB ðθÞ ¼ NL Br ðR; θÞ R dθ
θδ
X∞
Rn
¼ 2NL sin nδ ðAn cos nθ þ Bn sin nθÞ:
n¼1
n
The induced voltage in this case is

X∞
VðθÞ ¼ 2ωN L Rn sin nδ ðAn sin nθ þ Bn cos nθÞ: (10.31)
n¼1
Performing a Fourier analysis, we find that

ð 2π
1
Bm ¼ VðθÞ cos mθ dθ (10.32)
2πωNLRm sin mδ 0
and
ð 2π
1
Am ¼ VðθÞ sin mθ dθ: (10.33)
2πωNLRm sin mδ 0
References
Wesley, 1962, p. 159.
p. 351–356.
[3] J. Stratton, Electromagnetic Theory, McGraw-Hill, 1941, p. 126–128.
[5] G. Moritz, Eddy currents in accelerator magnets, in D. Brandt (ed.), Proc. CERN
Accelerator School on Magnets, Bruges, Belgium, June 2009, CERN-2010-004,
p. 103.
p. 340–341.
[7] K. Halbach, Some eddy current effects in solid core magnets, Nuc Instr. Meth.
107:529, 1973.
[8] E. Kreyszig, Advanced Engineering Mathematics, Wiley, 1962, chapter 4.
[9] J. Marion, Classical Electromagnetic Radiation, Academic Press, 1965, p. 149.
[11] J. D. Jackson, Classical Electrodynamics, Wiley, 1962, p. 177–178.
[12] C. Reymond, Magnetic resonance techniques, in S. Turner (ed.), CERN Accelerator
School on Measurement and Alignment of Accelerator and Detector Magnets, CERN
98-05, 1998, p. 219.
[13] J. Kvitkovic, Hall generators, in S. Turner (ed.), ibid., p. 233.
[14] A. Jain, Harmonic coils, in S. Turner (ed.), ibid., p. 175.
11
Numerical methods
Until now, we have mostly considered magnetostatic problems that have analytic
solutions. In practice, this has usually restricted our choice of problems to situa-
tions where one of the problem boundaries coincides with a coordinate system axis
and to solutions that can be written as products of functions of a single coordinate.
Frequently, more complicated problems can only be solved using numerical meth-
ods.[1] A number of commercial and freeware programs are available for solving
magnetic problems. A lot of effort has been devoted to making many of these
programs accurate, efficient, and user-friendly. If such a program is available and
can address the problem under consideration, it is often the best choice to use it.
However, there are occasions when new code must be written to solve a problem.
It is also important to have some basic understanding about the methods involved in
obtaining these solutions. In this chapter, we will examine three numerical methods
that have been used for solving boundary value problems involving the Poisson
equation: finite differences, finite elements, and integral equations. In each case,
the analytical equation or its solution is approximated in some way that leads to
a matrix equation for the unknown potential or field. We conclude the chapter with
a discussion of inverse problems and optimization techniques.
11.1 Finite difference method

In the finite difference method, the continuous space of the problem domain is
replaced with a grid of discrete points called nodes.[1] The grid, which is com-
monly rectangular or polar, must extend over the whole space of the problem. This
usually requires grid points outside all conductors and iron, out to a point where the
potential has some assumed value, typically zero. Symmetries in the configuration
may be used to reduce the required grid size. For some problems, it may be
necessary to use reduced grid spacing in regions where the field gradient is large
or where high accuracy is required.
245
246 Numerical methods
Unknown quantities are calculated at the nodes. Derivatives defined on the

continuous domain of the physical problem are replaced with difference approx-
imations defined in terms of the values of the unknown function at the nodes. For
example, assume that u is an unknown function and that a one-dimensional
problem space has been discretized with the grid spacing h ¼ Δx. The Taylor series
for the node at location x þ h is
∂u h2 ∂2 u
uðx þ hÞ ¼ uðxÞ þ h þ þ Oðh3 Þ: (11.1)
∂x x 2! ∂x2 x
Ignoring second- and higher-order terms, the first derivative can be approximated
in terms of the node values by
∂u uðx þ hÞ uðxÞ
≃ : (11.2)
∂x h
This is called the forward difference because it involves the next higher node than
the node at x. Similarly taking h→ h, we find the backward difference is
∂u uðxÞ uðx hÞ
≃ : (11.3)
∂x h
We can obtain an approximation for the derivative that is accurate through the h2
term in the Taylor series by calculating
h2 00 h2
uðx þ hÞ uðx hÞ ≃ u þ hu0 þ u u þ hu0 u00 :
2 2
The central difference approximation for the first derivative is then
∂u uðx þ hÞ uðx hÞ
≃ : (11.4)
∂x 2h
We can approximate the second derivative from its definition as

∂2 u 1 uðx þ hÞ uðxÞ uðxÞ uðx hÞ
≃
∂x2 h h h
(11.5)
1
¼ 2 ½uðx þ hÞ 2uðxÞ þ uðx hÞ:
h
Example 11.1: one-dimensional Poisson equation

To illustrate the basic concepts of the finite difference method, let us consider the
solution of the one-dimensional Poisson’s equation
∂2 u
¼f
∂x2
Figure 11.1 Nodes on a line.
in the greatly simplified situation shown in Figure 11.1. The line is discretized into 7
nodes. Let u refer to the unknown quantity, which we assume satisfies the Dirichlet
boundary conditions u0 ¼ 0 and u6 ¼ 0. Assume that the source function f has the
value f3 ¼ v at the center of the line and is 0 otherwise. The values of u at the five
interior nodes are the unknown quantities. Using Equation 11.5, each interior node
satisfies the equation
uðx þ hÞ 2uðxÞ þ uðx hÞ ¼ h2 f :
We can write the equations for the five unknowns in the form of a matrix equation
Cu ¼ g: (11.6)
For the case here, we have

2 32 3 2 3
2 1 0 0 0 u1 0
6 1 2 1 0 0 7 6 u2 7 6 0 7
6 76 7 6 7
6 0 1 2 1 0 7 6 u3 7 ¼ 6 v 7;
6 76 7 6 7
4 0 0 1 2 1 5 4 u4 5 4 0 5
0 0 0 12 u5 0
which has the solution

2 3
½ h2 v
6 h2 v 7
6 7
u ¼ 6 3=2 h v 7
6 2
7:
4 h2 v 5
½ h v2
For a square grid in two dimensions, let us designate the node under consideration
as node 0 and its four nearest neighbors as nodes 1–4, as shown in Figure 11.2.
We can write the Laplacian operator in terms of the values at the five nodes as [2]
1
r2 u ≃ ½uðx h; yÞ þ uðx þ h; yÞ þ uðx; y hÞ þ uðx; y þ hÞ 4uðx; yÞ:
h2
(11.7)
It is also possible to write a generalized version for the two-dimensional Laplacian

where the distances from a given node to each of its nearest neighbors can be
different. If h is the characteristic grid spacing, then [3]
Figure 11.2 Node structure in two dimensions.
Figure 11.3 Node near a curved boundary.

2 u1 u2 u3 u4 1 1
r2 u ≃ þ þ þ þ u0 ; (11.8)
h2 pð p þ rÞ qðq þ sÞ rð p þ rÞ sðq þ sÞ pr qs
where p, q, r, s are dimensionless scaling factors for the spacings from node 0 to its
nearest neighbors. Using this expression for problems where the physical boundaries
of conductors and iron are parallel to the x and y axes, it is possible to set up the
equations for the interior nodes together with nodes coinciding with the boundaries.
Higher-order difference equations for the Laplacian are also possible.[4]
A complication arises in setting up the node equations for nodes adjacent to
boundaries that do not align exactly with the grid spacing, for example nodes next
to circular boundaries in a rectangular grid. For Dirichlet boundary conditions, we
can make use of the fact that we know the value of uðx; yÞ on the boundary.
Consider a node 0 adjacent to the boundary shown in Figure 11.3. The two-
dimensional Laplacian operator acting at u0 can be approximated as [5]

2 uA uB u3 u4 1 1
r u≃ 2
2
þ þ þ þ u0 ; (11.9)
h sð1 þ sÞ tð1 þ tÞ 1 þ s 1 þ t s t
where s and t are dimensionless scale factors.

There are also complications in setting up the difference equations when the
problem requires Neumann boundary conditions.[6] Here we will only consider the
situation shown in Figure 11.4, which is a planar boundary in a square grid between
Figure 11.4 Boundary between regions with different permeabilities.
two regions (a) and (b) with different permeabilities. Let us consider an arbitrary
point 0 along the boundary. Since point 0 is part of region a, the Laplace equation is
Aa1 þ A2 þ Aa3 þ A4 4A0 ¼ 0; (11.10)
where we use A for the unknown function here. For region a, node 1 is fictitious and
must be eliminated from the final difference equation. Node 0 is also a part of
region b, so we have
Ab1 þ A2 þ Ab3 þ A4 4A0 ¼ 0; (11.11)
where node 3 is fictitious in region b. The Neumann boundary condition at node 0 is

1 Aa1 Aa3 1 Ab1 Ab3
¼ :
μa 2h μb 2h
Substituting for Aa1 from Equation 11.10 and Ab3 from Equation 11.11, we find the
difference equation at boundary point 0 is
2μa 2μb
4A0 Ab1 A2 Aa3 A4 ¼ 0: (11.12)
μa þ μb μa þ μb
If region b has infinite permeability, then the difference equation simplifies to
4A0 A2 2Aa3 A4 ¼ 0: (11.13)
Interpolation must be used when a value of some quantity u is required at

a location away from one of the nodes. Suppose we want the value of the function
uðx; yÞ, as shown for a rectangular grid in Figure 11.5. The simplest scheme for
estimating the value of u is bilinear interpolation. We first determine which rectangle
in the grid that the desired point is located in. Then defining the variables
Figure 11.5 Bilinear interpolation.
x x1
s¼
x2 x1
y y1
t¼ ;
y2 y1
we can approximate the value of uðx; yÞ as [7]
uðs; tÞ ≃ ð1 sÞð1 tÞu1 þ sð1 tÞu2 þ stu3 þ ð1 sÞtu4 : (11.14)
This expression varies continuously in x and y and reduces correctly to the node
values at the corners of the rectangle.
The discrepancy between the result from using the difference approximation and
the exact result from solving the differential equation is known as the truncation
error. The error can be estimated by examining the first term in the Taylor series that
was neglected in deriving the difference formula under consideration. For a square
mesh, the error on the second derivative goes like
2h2 ∂4 u
∼ :
4! ∂x4 0
The error is proportional to h2, so one method of improving the accuracy in a finite
difference calculation is to reduce the mesh spacing. We can monitor the improve-
ment in accuracy by finding the maximum absolute value for the difference
eij ¼ uhij2 uhij1 ;
where the superscript refers to the mesh spacing used for the solution and ði; jÞ
refers to nodes common to both mesh spacings. This approach is ultimately limited
by the growth in the size of the coefficient matrix and by rounding errors in the
numerical calculations. An alternative approach for increasing the accuracy of the
calculation is to use higher-order difference equations.
The quality of a solution can be monitored by calculating the residual for each of
the interior nodes. For a general node for the Poisson equation, the residual is
defined as
Ri; j ¼ 4 ui; j ui; jþ1 ui; j1 uiþ1; j ui1; j h2 μi; j Ji; j : (11.15)
If the difference equation exactly satisfies Poisson’s equation, the residual should
be 0.
For problems using iteration algorithms, we can compute
eij ¼ unij un1

ij
for the unknown function at the node ði; jÞ, where the superscript refers to the
iteration number. For these methods, one can estimate the quality of the solution by
calculating the difference at all the nodes. Let M refer to the absolute value of the
largest difference in the mesh.
M ¼ max jeij j
For the 5-point Laplacian operator in Equation 11.7, the error ε between the exact
solution of the difference equation and the approximate solution after n iterations is
bounded by [8]
Mρ2
ε ≤ ; (11.16)
4h2
where ρ is the radius of the smallest circle that encompasses the entire field region.
In problems where iron saturation is a consideration, the permeability of the iron
can be made a variable at each of the nodes in the iron regions.[9] The permeabil-
ities are stored on a separate mesh. After each iteration of the potential, the field in
the iron region is updated. A table of B-μ values can be used to relate the perme-
ability to the field at the node. The mesh of permeability values is then updated
using, for example, an under-relaxation algorithm.
11.2 Example solution using finite differences

As a simple example, let us consider a rectangular conductor with constant
current density J close to a sheet of iron with permeability μr ¼ 100, as shown
in Figure 11.6.
Assume that the conductor and the sheet are uniform in the z direction, so that
a two-dimensional analysis is justified. Assume that the figure is up-down sym-
metric, so that the x axis lies in a symmetry plane. We solve the problem using finite
differences on a square 200 × 200 mesh. For simplicity, we have chosen the
boundaries of the conductor and the iron sheet to line up with node locations.
The problem requires Dirichlet boundary conditions on the left, right, and top outer
borders, where we set Az ¼ 0. Because of the up-down symmetry, the bottom
border requires a Neumann boundary condition.
Figure 11.6 Conductor close to an iron sheet.
Figure 11.7 Finite difference node patterns.
In applying the Poisson equation here, four types of node patterns are required,
as shown in Figure 11.7. In each case, node 0 refers to the node we are currently
evaluating. For a general interior node where all the neighbor nodes are in the same
region, we apply the pattern (a), which results in the relation
4A0 ¼ A1 þ A2 þ A3 þ A4 þ f ; (11.17)
where f ¼ h2 μ J for nodes inside the conductor and 0 otherwise. For nodes on the
symmetry plane, pattern (b) gives
4A0 ¼ A1 þ 2A2 þ A3 þ f :
For the left side of the iron sheet, we can use Equation 11.12 for pattern (c) with
μa ¼ 1 and μb ¼ 100. For the right side of the sheet, we use Equation 11.12 for
pattern (d) with μa ¼ 100 and μb ¼ 1.
For problems with a very large number of unknown nodes, it is not practical to
solve the matrix equation using direct methods. Instead iterative methods must be
used. A common method is to use the Successive Overrelaxation (SOR) algorithm.
[10, 11] Let us define Anj;k to be the value of the potential at the interior node at
location ðj; kÞ after n iterations. On the next iteration, we update the value of the
potential according to the prescription

j;k ¼ ð1 αÞ Aj;k þ α Aj;k ;
Anþ1 n
(11.18)
where α is called the overrelaxation parameter. For efficient convergence, we need

1 < α < 2. The optimal value for α is problem dependent, but the value α ~ 1.7,
which we use here, is typical. The quantity A j;k is the solution for Ajk from the
appropriate nodal solution of Poisson’s equation. For example, using
Equation 11.17 for a general node, the SOR relation is
α h nþ1 i
Aj;k ¼ ð1 αÞAj;k þ
nþ1 n
A þ Ajþ1;k þ Aj;k1 þ Aj;kþ1 þ fj;k :
n nþ1 n
4 j1;k
Thus the updated value of the potential has two contributions. The first term is an
adjustable fraction of its value on the previous iteration. The second term is
a fraction of the Poisson equation residual at the node, calculated from the values
of the potential at the neighbor nodes. Note that the calculation of the residual uses
values for two nodes that have already been updated for a given iteration and values
for two nodes from the previous iteration. The iterations continue until

Anþ1 An
j;k j;k
max ≤τ
Anj;k
over all the interior nodes.1 The tolerance τ ¼ 105 was used in this example. This
criterion was satisfied after 7,511 iterations.
The magnetic field was calculated at the center of every square formed by four
neighbor nodes, as shown in Figure 11.8.
1
Bx ¼ ∂ y A z ¼ ðA1 þ A2 þ A3 A4 Þ
2h
1
By ¼ ∂x Az ¼ ðA1 þ A2 A3 A4 Þ:
2h
The results of the calculations for the magnetic field are shown in Figure 11.9.
1
Or just the difference in values if the potential is 0.
Figure 11.8 Magnetic field calculation.
Figure 11.9 Magnetic field pattern for the example finite difference problem.
11.3 Finite element method

In the finite element method, the problem space is completely subdivided into a set of
subregions called finite elements.[1, 12] The potential in each element is represented
by an interpolation function that is defined in terms of the potential values at the
nodes of the element. The Poisson equation and its boundary conditions can be
formulated in terms of energy functionals. The minimization of this functional
generates a set of algebraic equations that can be solved directly or through iterative
techniques. The method is quite flexible since there is considerable freedom in
choosing element shapes to match boundary and interface geometries.
For simplicity, we restrict our discussion here to two-dimensional problems.

The nonlinear Poisson equation can be written in the form

∂ ∂A ∂ ∂A
γ þ γ ¼ J;
∂x ∂x ∂y ∂y
where γ is the reluctivity and A and J only have nonvanishing components in the
z direction. This differential equation can be expressed in terms of the energy func-
tional [13]
ðð ð B þ
∂A
F¼ γb db JA dx dy A dl; (11.19)
0 ∂n
where b is the magnitude of the magnetic field.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2ffi
∂A ∂A
b¼ þ : (11.20)
∂x ∂y
The line integral in Equation 11.19 vanishes since we require that the potential
satisfy either Dirichlet or Neumann boundary conditions everywhere on the
boundary. If the reluctivity is constant over an element, we can perform the
integration over b to get the simplified energy functional
ðð ( " 2 2 # )
γ ∂A ∂A
F¼ þ J A dx dy: (11.21)
2 ∂x ∂y
The simplest two-dimensional finite element is a triangle, as shown in

Figure 11.10. We assume the potential varies linearly inside the element.
A ¼ c1 þ c2 x þ c3 y: (11.22)
Figure 11.10 Triangular finite element.

If we write this expression for each of the three nodes, we have three equations that
can be solved for the three unknown coefficients ci in terms of the potentials and
coordinates at the nodes. Substituting the result back into Equation 11.22, we find

ðx2 y3 x3 y2 Þ þ ðy2 y3 Þx þ ðx3 x2 Þy
A ¼ A1
2S

þA2 (11.23)
2S

þ A3 ;
2S
where S is the area of the triangle.
S ¼ ½½ðx2 y3 x3 y2 Þ þ ðy2 y3 Þx1 þ ðx3 x2 Þy1 (11.24)
The coefficients of the node potentials in this equation are known as shape functions,
ζ.[14] Thus we can also write the interpolation function for the potential as
A ¼ ζ 1 A1 þ ζ 2 A2 þ ζ 3 A3 : (11.25)
The shape function ζ 1 has the properties that
ζ 1 ðx1 ; y1 Þ ¼ 1
ζ 1 ðx2 ; y2 Þ ¼ 0
ζ 1 ðx3 ; y3 Þ ¼ 0
and similarly for ζ2 and ζ3.

In order to evaluate the simplified energy functional in Equation 11.21, we need
the derivatives of A from Equation 11.23.
∂A ðy2 y3 ÞA1 þ ðy3 y1 ÞA2 þ ðy1 y2 ÞA3
¼
∂x 2S
(11.26)
∂A ðx3 x2 ÞA1 þ ðx1 x3 ÞA2 þ ðx2 x1 ÞA3
¼
∂y 2S
ðð (
γ ½ðy2 y3 ÞA1 þ ðy3 y1 ÞA2 þ ðy1 y2 ÞA3 2
F¼
2 4S2
)
½ðx3 x2 ÞA1 þ ðx1 x3 ÞA2 þ ðx2 x1 ÞA3 2
þ dx dy
4S2
ðð
JA dx dy:
The solution of the field equations is equivalent to finding a function A that

minimizes this energy functional.[15] The potentials at the three nodes may be
considered to be the parameters of the functional F for a given element. Thus we
require that
ðð
∂F γ 2½ðy2 y3 ÞA1 þ ðy3 y1 ÞA2 þ ðy1 y2 ÞA3 ðy2 y3 Þ
¼
∂A1 2 4S2

2½ðx3 x2 ÞA1 þ ðx1 x3 ÞA2 þ ðx2 x1 ÞA3 ðx3 x2 Þ
þ dx dy
4S2
ðð
∂A
J dx dy ¼ 0
∂A1
with analogous expressions for the derivatives with respect to A2 and A3 .

The integrand for the first integral is independent of x and y and the integrand for
the second integral may be evaluated using Equation 11.23. Thus we have
∂F γ h i
¼ ðy2 y3 Þ2 A1 þ ðy3 y1 Þðy2 y3 ÞA2 þ ðy1 y2 Þðy2 y3 ÞA3
∂A1 4S
γ h 2
i
þ ðx3 x2 Þ A1 þ ðx1 x3 Þðx3 x2 ÞA2 þ ðx2 x1 Þðx3 x2 ÞA3
4S
ðð
½ðx2 y3 x3 y2 Þ þ ðy2 y3 Þx þ ðx3 x2 Þy
J dx dy ¼ 0
2S
with analogous expressions for the derivatives with respect to A2 and A3. For
elements containing current, the second integral can be evaluated by assuming
that J is constant and that x and y are evaluated at the centroid of the triangle.
x1 þ x2 þ x3
xc ¼
3
y1 þ y2 þ y3
yc ¼ :
3
In this case, the numerator in the last term is 2S/3, so the integral has the value JS/3.
Thus minimization of the functional over the triangular element leads to the matrix
equation
2 32 3 2 3
C C12 C13 A1 1
γ 4 11 J S
C21 C22 C23 54 A2 5 ¼ 4 1 5: (11.27)
4S 3
C31 C32 C33 A3 1
The coefficient matrix C is symmetric with six unique elements.

Figure 11.11 Two neighboring 2D elements.
C11 ¼ ðy2 y3 Þ2 þ ðx3 x2 Þ2

C12 ¼ ðy3 y1 Þðy2 y3 Þ þ ðx1 x3 Þðx3 x2 Þ
(11.28)
C22 ¼ ðy3 y1 Þ2 þ ðx1 x3 Þ2
C33 ¼ ðy1 y2 Þ2 þ ðx2 x1 Þ2 :
Each triangular element introduces an analogous set of equations. However, if

N is the total number of elements, the number of unknown potentials is less than
3N because all of the elements share boundaries with neighbor triangles. For
example, if we consider the two elements shown in Figure 11.11, the first element
introduces three unknown potentials while the second element only adds one more.
The resulting set of equations can be solved for the potentials using direct or
iterative methods.
Setting up a realistic finite element problem involves a great deal of careful
bookkeeping and computations.[13, 15] The problem space must be completely
covered by the set of finite elements. The elements and nodes must be indexed and
the association of each element with its corresponding nodes, current, and perme-
ability must be clearly established. The boundary conditions must be imposed on
the appropriate subset of the nodes. The coefficients for Equation 11.27 must be
determined and an appropriate method used for solving the resulting system of
equations. Additional iterative techniques must be applied if the problem contains
saturable iron.
11.4 Integral equation method

Thus far we have discussed numerical methods for solving the Poisson differential
equation directly and for solving the potentials by minimizing the energy func-
tional for the magnetostatic field. Here we examine a third approach where the
unknown potentials or sources of the field are expressed in terms of an integral
equation. A major advantage of the integral equation method over methods based
on the solution of a differential equation is that the mesh only needs to encompass
the iron region (and possibly the conductor region if the current density is not
uniform).[16, 17] The boundary condition at infinity follows naturally and does not
have to be imposed at the edge of a mesh. An important disadvantage is that the
resulting matrix equation is dense, so the solution time grows rapidly as the number
of elements is increased. Also the flux density computed near the iron elements can
be strongly affected by the discretization. There are many ways to formulate
a solution of Poisson’s equation using integral equations.[17, 18, 19] In addition,
it is also possible to formulate procedures which combine differential and integral
equation techniques.[20]
We describe here an integral equation approach that uses the magnetization of
iron elements as the unknown function.2 Recall that the magnetization is related to
the magnetic field intensity by
!
! B ! !
M ¼ H ¼ χðHÞ H ; (11.29)
μ0
where χ is the susceptibility. The field intensity has contributions from both
conductor currents and from the magnetization in the iron. The total field inten-
sity is
! ! !
H ¼ H c þ H m;
so the magnetization is given by

! ! !
M ¼ χ ðH c þ H m Þ:
The contribution Hc can be calculated for simple conductor configurations using

the complex variable techniques given in Chapter 5 or directly from the Biot-
Savart law
ð ! !0 !
! ! 1 J ðr Þ R
H cð r Þ ¼ dV 0 ; (11.30)
4π R3
! !
where R ¼ !
r r0 : Using Equation 3.32, the field due to the magnetization is
ð !
! ! 1 ! !0 R
H mð r Þ ¼ r M ð r Þ· 3 dV 0 : (11.31)
4π R
2
This procedure was adopted by a group at the Rutherford High Energy Laboratory in the development of the
GFUN program.
Break the iron region into N elements and assume the magnetization is constant
over the area of each element. The magnetization at some element i will depend on
the field at i due to the conductors and on the field at i due to the magnetization from
all the elements. Thus we have
" ðX ! #
N
! ! 1 ! R ij
M i ¼ χi H ci ri M j · 3 dVj : (11.32)
4π j¼1
Rij
We define the contribution to the field at element i due to the magnetization at

element j in terms of the coupling constants
ð !
1 ^ R ij
Gij ¼ ri M j · 3 dVj : (11.33)
4π Rij
It is important to note that the components of G depends on the directions of the unit
vectors used to define M, but do not depend on the magnitude of M. We can then
rewrite Equation 11.32 as
" #
! ! X
N
!
M i ¼ χi H ci þ Gij M j :
j¼1
Rearranging this equation, we have

! X N
Mi ! !
Gij M j ¼ H ci ;
χi j¼1
which can be written in the standard matrix equation form [17]

!
XN
δi j ! !
Gij M j ¼ H ci : (11.34)
j ¼ 1
χj
This is a set of algebraic equations for the N unknown magnetization elements.

The number of unknowns is 2N for two-dimensional problems and 3N for three
dimensions. The field components due to the conductors at each iron element can
be calculated directly, so the right-hand side of Equation 11.34 is known. However,
the problem is generally nonlinear because χ depends on the field due to the
unknown magnetizations.
Returning to the definition of the coupling constant in Equation 11.33, we
know from the vector identity B.2 that the gradient of the scalar product can be
expanded as
!! " ! !# ! !!
^ j· R ij ^ j ri R ij R ij ^ j ·ri Þ R ij
^ j Þ þ ðM
ri M ¼M þ 3 ðri M
R3ij R3ij Rij R3ij
"! #
R ij ^ j:
þ 3 ·ri M
Rij
Two of the terms vanish because the derivative in field coordinates acts on the unit
vector along the magnetization in source coordinates. Another term vanishes because
of the curl operator acting on the linear vector R. Thus only the third term on the
right-hand side remains. Writing this out in terms of components, we have
! !
ðM^ j ·ri Þ R i j
R3i j
2 3
ðx x Þ^
x þ ðy y Þ^
y þ ðz z Þ^
z
¼ M ^ jx ∂ix þ M ^ jz ∂iz 4
^ jy ∂iy þ M i j i j i j 5
2 2 2 3=2
fðxi xj Þ þ ðyi yj Þ þ ðzi zj Þ g
^ jx h
M !i M^ jy h !i M ^ jz h !i
¼ R 2
i j ^
x 3ðx i x j Þ R þ R 2
i j ^
y 3ðy i y j Þ R þ R 2
i j^z 3ðzi z j Þ R :
R5i j R5i j R5i j
After inserting this expression into Equation 11.33, we can identify the three-
dimensional coupling constants
ð 2 2
1 Rij 3ðxi xj Þ
Gix; jx ¼ dVj
4π R5ij
ð
1 3ðxi xj Þðyi yj Þ
Gix; jy ¼ dVj
4π R5ij
and similarly for the other components.[21, 22] The G coupling matrix is sym-
metric. There are constraints on the sum of the diagonal elements.[21]

0 if i ≠ j
Gix; jx þ Giy; jy þ Giz; jz ¼
1 if i ¼ j
In two dimensions, Mj is uniform along z, the field observation point has zi ¼ 0,

and ri and Mj only have x and y components. We find the two-dimensional
coupling constants by integrating the three-dimensional couplings over z
ðð !!
1 ^ j ·ri Þ R ij
Gij ¼ ðM dzj dSj ;
4π R3ij
where dSj is the two-dimensional area element. This can be written in the form
ð
1 ^ ^ jy I2 dSj ;
Gij ¼ ½M jx I1 þ M (11.35)
4π
where
ð∞
R2i j ^x 3ðxi xj Þ½ðxi xj Þ ^x þ ðyi yj Þ ^y zj ^z
I1 ¼ dzj
∞ R5i j
and
ð∞
R2i j ^y 3ðyi yj Þ½ðxi xj Þ^x þ ðyi yj Þ ^y zj ^z
I2 ¼ dzj :
∞ R5i j
Let rij be the distance between the observation point and the centroid of the iron
element in the x-y plane. Then the integral I1 can be broken into the three simpler
integrals3
ð∞
dzj 2
3=2
¼ 2
∞ frij þ zj g
2 2 rij
ð∞
dzj 2
5=2
¼ 4
∞ fr2
ij þ zj g
2 3rij
ð∞
zj
dzj ¼ 0
∞ fr2 þ z 2 g5=2
ij j
with the result that

2 4ðxi xj Þ !
I1 ¼ 2
^x r ij
rij r4ij
2 4ðyi yj Þ !
I2 ¼ 2
^y r ij :
rij r4ij
Inserting these results into Equation 11.35, we find the two-dimensional coupling
constants are the dimensionless, geometric factors
3
GR 2.271.5, 2.263.3, 2.271.7.
ð
1 ðxi xj Þ2 ðyi yj Þ2
Gix; jx ¼ dSj
2π r4ij
ð
1 2ðxi xj Þðyi yj Þ
Gix; jy ¼ Giy; jx ¼ dSj (11.36)
2π r4ij
ð
1 ðyi yj Þ2 ðxi xj Þ2
Giy; jy ¼ dSj :
2π r4ij
There are also constraints on the sum of the two-dimensional diagonal

elements.[21]

0 if i ≠ j
Gix; jx þ Giy; jy ¼
1 if i ¼ j
Once we know the coupling constants G, we can solve Equation 11.34 to find the
magnetization in each of the iron elements. Then the field at any position can be
found from the sum of the fields due to all the current elements together with the
sum of all the fields due to the iron magnetizations. In applications where saturation
in the iron is important, the permeability of all the iron elements must be recom-
puted using the magnetizations and a μ–H table for the iron material. The process is
then iterated until the maximum change in permeability in any element is less than
some tolerance value.
Example 11.2: setting up the integral equations for a dipole configuration

We will illustrate the two-dimensional integral equation algorithm by considering
a simple example of currents and iron blocks arranged in a dipole configuration. Once
the current and iron magnetization has been determined in the first quadrant, the
dipole symmetry constrains the geometry and polarity of the currents and magnetiza-
tions in the other quadrants, as illustrated in Figure 11.12. The currents have polarities
ð1Þ ð1Þ
{I, –I, -–I, I} in the four quadrants. If we let ðMx ; My Þ refer to the magnetization of
an iron element in the first quadrant, then the dipole symmetry requires that
ð2Þ ð3Þ ð4Þ ð1Þ
My ¼ My ¼ My ¼ My
ð2Þ ð4Þ ð1Þ
Mx ¼ Mx ¼ Mx
ð3Þ ð1Þ
Mx ¼ Mx ;
where the numeral superscripts refer to the quadrants. Making use of the dipole
symmetry allows us to treat only the magnetization components in the first quadrant
as unknowns. This can be important in problems with large numbers of iron elements
since it reduces the size of the matrix equation by a factor of four.
Figure 11.12 Dipole configuration of a current and iron magnetization.
Assume, for example, that there are two iron elements in the first quadrant. We can
find the right-hand side of Equation 11.34 using the techniques in Chapters 4 and 5.
We evaluate the field at the centroids of each of the iron elements. The magnetization
has a constant magnitude and direction in each element. The coupling constants for
the field due to the magnetizations can be found from Equation 5.72.
þ
M dz
H ðzo Þ ¼ :
4π i z zo
We can determine the coupling constants Gxx and Gyx by evaluating H* with M = 1
and the coupling constants Gxy and Gyy by evaluating H* with M ¼ i.
The field components due to the magnetization in each iron element can be found
from

Hmx ¼ Gxx Mx þ Gxy My

(11.37)
Hmy ¼ Gyx Mx þ Gyy My :
This gives the field anywhere outside the iron block. However, when the observation
point is inside the block, the numerical procedure must ensure that H and M point in
opposite directions, as they must inside a magnetic material. For the dipole config-
uration, we define the matrix coefficients C as sums over the coupling constants in the
four quadrants. For example,
Ci x j x ¼ Gð1Þ ð2Þ ð3Þ ð4Þ

xx Gxx þ Gxx Gxx ;
where the minus signs take into account the reversal in the sign of Mx in the second
and fourth quadrants. The other three coefficients are similarly defined. The matrix
Equation 11.34 can be written for the case of two iron elements as
11.5 The POISSON code 265
ðs1 C1x1x ÞM1x C1x1y M1y C1x2x M2x C1x2y M2y ¼ Hc1x
C1y1x M1x þ ðs1 C1y1y ÞM1y C1y2x M2x C1y2y M2y ¼ Hc1y
C2x1x M1x C2x1y M1y þ ðs2 C2x2x ÞM2x C2x2y M2y ¼ Hc2x
C2y1x M1x C2y1y M1y C2y2x M2x þ ðs2 C2y2y ÞM2y ¼ Hc2y ;
where si ¼ 1=ðμri 1Þ and the numeral subscripts refer to the two iron elements in
the first quadrant.
After solving the matrix equation, M is known for all the iron blocks. The
contribution of the iron to the field at any location can be found using
Equation 11.37, where G is now evaluated for the desired field point.
11.5 The POISSON code

We have shown results from the POISSON code4 a number of times previously
in this book. POISSON is one of the earliest examples of a finite element
program. We give a brief description here of the method used in the code for
solving the two-dimensional Poisson equation.[23, 24] The user defines the
boundaries and properties of the physical regions in the problem, together with
the boundary conditions at the borders of the problem space. The program then
automatically sets up an irregular triangular mesh where every interior node is
surrounded by six triangles. All boundaries and interfaces between regions lie
on mesh lines. The current density is assumed to be constant in each triangle in
a conductor region and the permeability is assumed to be constant in each
triangle in an iron region.
The code does not solve the Poisson equation directly. Instead, the solution
algorithm makes use of the Ampère law
þ ð
!! !
H ·dl ¼ J ·^n dS:
Allowing for saturation in the iron, this can be written as

þ ð
!! !
γðBÞ B ·dl ¼ μ0 J ·^n dS;
where γ is the reluctivity. The vector potential is assumed to only have a z

!
component and to satisfy r· A ¼ 0: Expressing B in terms of the vector potential,
we get
4
http://laacg.lanl.gov/laacg/services
Figure 11.13 Relation among nodes in the POISSON algorithm.
þ ð
∂A ∂A ! !
γðBÞ ^x ^y ·dl ¼ μ0 J ·^n dS:
∂y ∂x
The contour around each mesh point follows a twelve-sided path through the
interior of the six surrounding triangles. After a lengthy calculation,[24]
a difference equation for the potential at node 0 can be derived in terms of the
potentials at the six neighbor nodes as
X6
μ X6
Ai wi þ 0 Ji Si
i¼1
3 i¼1
A0 ¼ :
X 6
wi
i¼1
In this equation, Si is the area of the triangle. The wi are coupling coefficients that
involve the parameters of the triangles on adjacent sides of the line connecting A0 to
neighboring node i. Looking at the diagram in Figure 11.13,
w1 ¼ ½ðγ1 cot θ1 þ γ2 cot θ4 Þ
with similar expressions for the other five couplings.
The vector potential varies linearly inside any triangle. As a result, the magnetic
field is constant over the area of the triangle. Values for the potential are updated
using the successive over-relaxation algorithm. The new values of the field are then
used to estimate new values for γ and for the couplings w. An under-relaxation
algorithm is used to update the final values of the couplings for each iteration
winþ1 ¼ ð1 αÞwin þ αwinew ;
where the relaxation parameter satisfies 0 < α < 1.

As an illustration of using POISSON, we return to the simple problem discussed
in Section 11.2. The input commands to define the problem are shown in
Table 11.1 POISSON input commands for the example problem
Example: rectangular conductor near iron sheet

&reg kprob=0, ! Poisson or Pandira problem
icylin=0, ! rectangular coordinates
mode=-1, ! iron has fixed finite permeability
fixgam=0.01, ! reluctivity
dx=0.3,dy=0.3, ! mesh size intervals
nbslo=1, ! Neumann boundary condition on lower edge
nbsup=0, ! Dirichlet boundary condition on upper edge
nbslf=0, ! Dirichlet boundary condition on left edge
nbsrt=0 & ! Dirichlet boundary condition on right edge
&po x=0.0,y=0.0 & ! problem domain
&po x=100.,y=0.0 &
&po x=100.0,y=100.0 &
&po x=0.0,y=100.0 &
&po x=0.0,y=0.0 &
&reg mat=1,cur=19500. & ! conductor
&po x=30.0,y=0.0 &
&po x=50.0,y=0. &
&po x=50.0,y=30. &
&po x=30.0,y=30.0 &
&po x=30.0,y=0.0 &
&reg mat=2 & ! iron sheet
&po x=70.0,y=0.0 &
&po x=72.0,y=0.0 &
&po x=72.,y=70.0 &
&po x=70.,y=70.0 &
&po x=70.0,y=0.0 &
Table 11.1. The first REG command defines the problem domain and specifies the
mesh size and the boundary conditions. The PO commands define points around
the boundary of regions. The second region defines the conductor and specifies the
current. The third region defines the iron sheet. POISSON sets up a triangular mesh
using this information and then solves the Poisson equation using the SOR algo-
rithm. For this example, the program used 112,896 mesh points, converged in 1,160
iterations, and had an average residual of 5 107 . The resulting field distribution,
shown in Figure 11.14, agrees qualitatively well with the finite difference result in
Figure 11.9. The shielding effects of the iron sheet are clearly apparent in the figure.
11.6 Inverse problems and optimization

We have previously defined the inverse problem as finding a current distribution
that generates a specified magnetic field configuration. We discussed several
problems of this type in Chapter 8. The solution of inverse problems is simplified
0 20 40 60 80 100
100
90
80
70
60
50
40
30
20
10
0
0 20 40 60 80 100
Figure 11.14 Field distribution for the example problem.
by constraining the coil geometry. For example, by specifying that the unknown
currents lie on a cylindrical current sheet, it is possible to use Fourier-Bessel
transforms to find the azimuthal and longitudinal current components that produce
a specified target field inside a magnet aperture.[25] Another interesting approach
used a numerical variational process to modify the contours of a uniform current
density block conductor.[26] The target field was expressed in terms of a multipole
expansion of the transverse field in the aperture. Higher-order multipoles were
minimized by varying the geometry of the outer boundary of initial circular or
elliptical current blocks.
A powerful technique for solving inverse problems is to make use of numerical
optimization methods. Let us consider in more detail the numerical solution for two
interesting inverse problems. As the first example, assume we have a solenoid
channel with a constant axial field B1 and that we need to design an interface region
to a second solenoid channel with constant axial field B2 . Assume the interface has
length L measured from the center of the last magnet in the first channel to the
center of the first magnet of the second channel. Assume in addition that the
transition region and second channel have to accept the full magnetic flux present in
the first channel. Then the desired field profile in the interface region must satisfy
the four constraints
Bz ð0Þ ¼ B1
Bz ðLÞ ¼ B2
dBz
ð0Þ ¼ 0
dz
dBz
ðLÞ ¼ 0:
dz
A model field profile that satisfies these constraints is
B1
Bz ðzÞ ¼ ; (11.38)
1 þ c z2 þ d z3
where
3 ðB1 B2 Þ
c¼
B2 L2
2 ðB1 B2 Þ
d¼ :
B2 L3
If r1 is the inner radius of the coils in the first channel, then the requirement for
constant flux puts an additional constraint on the allowed inner radius of the
downstream coils.
sffiffiffiffiffiffiffiffiffiffiffi
B1
rðzÞ ≥ r1 :
Bz ðzÞ
For example, let the coil C1 be the last solenoid in a 10 T channel with a fixed
inner radius of 10 cm and coil C14 be the first solenoid in a 2 T channel. Assume the
transition region is 7 m long and contains 12 solenoids that are 45 cm long,
separated by 5 cm, and have adjustable inner radius, radial thickness, and current
density. The axial field for each solenoid uses Equation 7.46. The merit function
f for the minimizer compares the desired value of the field at N locations zi from
Equation 11.38 with the calculated sum of the fields from all the coils, each with
a set of parameters aj .
" ! #2
X
N X
14
f ¼ Bz ðzi ; aj Þ Bz ðzi Þ :
i¼1 j¼1
Figure 11.15 The optimized axial field (dots) and the desired field profile (line) in
the transition region between two solenoid channels.
The calculation shown here used N = 36. Minimization of this function used
methods that do not require calculation of the derivatives.[27] The initial mini-
mization was done using a simplex algorithm. The most useful parameters to adjust
were the current densities and the inner radii of the coils. The axial parameters are
severely constrained here by the chosen geometry for the transition region. After
a preliminary solution had been found, the Powell direction-set method was used
for the final minimization. The optimized axial field is compared with the desired
field profile in Figure 11.15.
As a second example of optimization, let us consider the design of the central
section of a long dipole magnet with a circular cross-section. Assume that field
quality in the dipole aperture is the matter of concern and that we want to minimize
the strength of the first four allowed harmonics of the dipole field. We saw in
Chapters 4 and 5 that the multipole coefficients depend on the limiting angles of
annular conductor blocks. In order to eliminate four multipoles we will need to use
at least three blocks. We choose here a conductor design with two radial layers,
each of which has two conductor blocks, as shown in Figure 11.16. The contribu-
tion to the multipoles from an annular conductor block with constant current
density was given in Equations 5.68–5.70. We again use a minimization algorithm,
where the merit function is now given by
" ! #2
X
4 X
16
f ¼ wi bn ðni ; aj Þ e
b n ðni Þ :
i¼1 j¼1
Figure 11.16 Conductor blocks in the first quadrant of a dipole magnet.
The index i sums over desired multipole orders, the index j sums over coils, and the
normalized multipole ratio is defined as
Bn ½T=mn1 rn1
ref ½m
n1

bn ¼ :
B1 ½T
The eb n factors are the desired values of the multipole ratios, which we take here as
0. The aj are the set of parameters that describe conductor block j. The parameters
of the coils in quadrants 2–4 are related to the parameters of the coils in quadrant 1
by the dipole symmetry. For this calculation the adjustable parameters are the end
angles of the blocks nearest the midplane and the start and end angles of the second
block in each layer. The start angle of the two blocks nearest the midplane are made
as close to 0 as possible to maximize the dipole field. The wi are weights that
determine the importance of satisfying the constraint on multipole ni . The reference
radius used for the multipole calculations was 2/3 of the magnet aperture. After
minimization, the allowed multipole ratios b3 ; b5 ; b7 , and b9 have strengths ~104 .
In the design of actual magnets,[28] a need for high precision field quality
may require that allowed multipoles higher than b9 are also minimized.
In addition, the conductor may have to be described in terms of individual
turns of the cable separated by the appropriate insulation thickness, instead of
the continuous conductor blocks used here. This introduces the additional
constraint that there must be an integral number of turns in a conductor block.
In addition, if the coils are surrounded by an iron shell, saturation effects, which
cause the multipole strength to vary with the excitation current, may have to be
taken into account.
References
[1] A. Frisiani, G. Molinari & A. Viviani, Introduction, in M. Chari & P. Silvester
(eds.), Finite Elements in Electrical and Magnetic Field Problems, Wiley, 1980,
p. 1–10.
[2] N. Gershenfeld, The Nature of Mathematical Modeling, Cambridge University Press,
1999, p. 86–91.
Problems, Pergamon Press, 2nd ed., 1973, p. 246.
[4] D. Jones, Methods in Electromagnetic Wave Propagation, vol. 1, Oxford University
Press, 1987, p. 113.
[5] G. Smith, Numerical Solution of Partial Differential Equations: Finite Difference
Methods, 2nd ed., Oxford University Press, 1978, p. 216.
[6] K. Binns & P. Lawrenson, op. cit., p. 268.
[7] W. Press, S. Teukolsky, W. Vetterling & B. Flannery, Numerical Recipes in Fortran,
2nd ed., Cambridge University Press, 1992, p. 117.
[8] W. Milne, Numerical Solution of Differential Equations, Dover, 1970, p. 217.
[9] G. Parzen & K. Jellett, Computation of high field magnets, Part. Acc. 2:169, 1971.
[10] W. Press, et al., op. cit., p. 857–860.
[11] K. Binns & P. Lawrenson, op. cit., p. 260–265.
[12] N. Gershenfeld, op. cit., p. 93–101.
[13] M. Chari, Finite element solution of magnetic and electric field problems in electrical
machines and devices, in M. Chari & P. Silvester (eds.), Finite Elements in Electrical
and Magnetic Field Problems, Wiley, 1980, p. 87–107.
[14] R. Gallagher, Shape functions, in M. Chari & P. Silvester (eds.), ibid., p. 49–67.
[15] P. Silvester & M. Chari, Finite element solution of saturable magnetic field problems,
IEEE Trans. on Power Apparatus and Systems 89:1642, 1970.
[16] J. Simkin & C. Trowbridge, Three dimensional nonlinear electromagnetic field
computations using scalar potentials, IEE Proc. 127:368, 1980.
[17] J. Simkin & C. Trowbridge, Magnetostatic fields computed using an integral equation
derived from Green’s theorems, Proc. Compumag, Rutherford Appleton Laboratory,
Oxford, 1976, p. 5.
[18] C. Trowbridge, Applications of integral equation methods to the numerical solution of
magnetostatic and eddy current problems, in M. Chari & P. Silvester (eds.), op. cit.,
p. 191–213.
[19] C. Trowbridge, Progress in magnet design by computer, Proc. 4th Int. Conf. Magnet
Technology, Brookhaven National Laboratory, 1972, p. 555.
[20] B. McDonald & A. Wexler, Mutually constrained partial differential and integral
equation field formulations, in M. Chari & P. Silvester (eds.), op. cit., p. 161–190.
[21] M. Newman, C. Trowbridge & L. Turner, GFUN: an interactive program as an aid to
magnet design, Proc. 4th Int. Conf. Mag. Tech., Brookhaven National Laboratory,
1972, p. 617.
[22] F. Mingwu, S. Hanguang & W. Jingguo, Some experiences of using integral equation
method to calculate magnetostatic fields, IEEE Trans. Mag. 21:2185, 1985.
[23] A. Winslow, Numerical solution of the quasilinear Poisson equation in a nonuniform
triangle mesh, J.Comp. Phys. 1:149, 1966.
[24] J. Billen & L. Young, Poisson-Superfish, Los Alamos National Laboratory report LA-
UR-96–1834, 2006, p. 577–596.
[25] R. Turner, A target field approach to optimal coil design, J. Phys. D: Appl. Phys. 19:
L147, 1986.
References 273
[26] G. Morgan, Two dimensional, uniform current density, air core coil configurations for
the production of specified magnetic fields, Proc. 1969 Part. Accel. Conf., Washington
DC, p. 768.
[27] W. Press, et al., op. cit., p. 402–413.
[28] E. Bleser, et al., Superconducting magnets for the CBA Project, Nuc. Instr. Meth.
Phys. Res. A 235:435, 1985.
Appendices
A
Symbols and SI units [1, 2]
Symbol Quantity Unit Dimension

I current A Q T−1
K sheet current density A/m Q T−1 L−1
J volume current density A/m2 Q T−1 L−2
B magnetic flux density T M T−1 Q−1
ΦB magnetic flux Wb = T m2 M L2 T−1 Q−1
μ0 permeability of free space =4π 107 T m/A M L Q−2
M magnetization A/m Q T−1 L−1
H magnetic intensity A/m Q T−1 L−1
A vector potential Wb/m = T m M L T−1 Q−1
Vm scalar potential A Q T−1
L, M self, mutual inductance H = Wb/A M L2 Q−2
ρ electric charge density C/m3 Q L−3
V potential difference V M L2 T−2 Q−1
E electric field intensity V/m M L T−2 Q−1
σ conductivity (Ω m)−1 T Q2 M−1 L−3
ε permittivity farad/m T2 Q2 M−1 L−3
D electric flux density coulomb m2 Q M−2
F force N = J/m M L T−2
W stored energy J=Nm M L2 T−2
P power W = J/s M L2 T−3
References
[1] E. R. Cohen (ed.), The Physics Quick Reference Guide, American Institute of Physics,
1996, p. 37–47.
[2] D. Halliday & R. Resnick, Physics for Students of Science and Engineering, Wiley,
1962, appendix G.
274
B
Vector analysis
Vector analysis plays an essential role in describing the theory of magnetic phe-
nomena.[1, 2] A vector V is a quantity that has both a magnitude and a direction.
A scalar S is a quantity that only has an associated magnitude. Vector fields are
functions that describe a physical quantity at every point in space.
The vector differential operator (del) is
∂ ∂ ∂
r¼ ^x þ ^y þ ^z :
∂x ∂y ∂z
When r is applied to a scalar function, it results in a vector known as the gradient.
∂S ∂S ∂S
rS ¼ ^x þ ^y þ ^z :
∂x ∂y ∂z
The gradient gives a measure of the rate of change of a vector. The dot product of r
with a vector forms a scalar known as the divergence.
! ∂Vx ∂Vy ∂Vz

r· V ¼ þ þ :
∂x ∂y ∂z
Roughly speaking, the divergence gives a measure for the spreading out of
a function away from a localized source. The Laplacian is an important operator
that describes the second derivative of a scalar function and is given by
∂2 S ∂2 S ∂2 S
r2 S ¼ r · rS ¼ þ þ :
∂x2 ∂y2 ∂z2
It is also useful to define the Laplacian of a vector function, which is given in

Cartesian coordinates as
!
r2 V ¼ r2 Vx ^x þ r2 Vy ^y þ r2 Vz ^z :
275
276 Vector analysis
The cross product of r with a vector forms another vector known as the curl.

^x ^y ^z
!
r V ¼ ∂x ∂y ∂z :

Vx Vy Vz
The curl gives a measure of the tendency of the vector to circulate around some
source. According to Helmholtz’s theorem,[3] a vector function that is bounded at
infinity can be uniquely defined by specifying its divergence and its curl.
If we consider a volume of space V enclosed by a surface S, then we find that any
changes in a vector W inside the volume must match the flux of W through the
boundary surface. This is the basis for an important result known as Gauss’s diver-
gence theorem.[4]
ð ð
! !
r·W dV ¼ W ·^n dS;
where n is the normal vector to the surface. If, on the other hand, we break up the
surface S into a number of smaller areas and look at the net result of the circulation
in all the subareas, we find that the circulations cancel in the interior of the region
and only give a net result around the perimeter of S. The result is known as Stokes’s
theorem.[5]
ð þ
! !!
ðr W Þ·^n dS ¼ W ·dl
Some other integral relations involving the gradient, divergence, and curl are less
common, but still useful.[6]
ð ð
rP dV ¼ P ^n dS
ð þ
!
^
n rP dS ¼ P dl
ð ð
! !
r W dV ¼ W ^n dS;
where P is a scalar function and S is the surface that bounds the volume V.
The differential vector operators for cylindrical and spherical coordinate systems
are given in Table B1.
Some important vector identities are
! ! ! ! !! ! !!
A ðB C Þ ¼ B ðA ·C Þ C ðA ·B Þ (B.1)
Vector analysis 277
Table B1 Vector operators in cylindrical and spherical coordinates [6]
Cylindrical Spherical
1 ^ þ ∂z S ^z 1 1
rS ∂ρ S ^ρ þ ∂ϕ S ϕ ∂r S ^r þ ∂θ S ^θ þ ^
∂ϕ S ϕ
ρ r r sin θ
! 1 1 1
r· V ∂ρ ðρ Vρ Þ þ ∂ϕ Vϕ þ ∂z Vz ∂r ðr2 Vr Þ
ρ ρ r2
1
þ ½∂θ ðVθ sin θÞ þ ∂ϕ Vϕ
rsin θ
!
r V 1 ^ 1
∂ϕ Vz ∂z Vϕ ^ρ þ ð∂z Vρ ∂ρ Vz Þ ϕ ½∂θ ðVϕ sin θÞ ∂ϕ Vθ ^r
ρ rsin θ
1 1
þ ½∂ρ ðρ Vϕ Þ ∂ϕ Vρ ^z þ ½∂ϕ Vr sin θ∂r ðr Vϕ Þ^θ
ρ rsin θ
1 ^
þ ½∂r ðr Vθ Þ ∂θ Vr ϕ
r
1 1 1
r2 S ∂ρ ðρ∂ρ SÞ þ 2 ∂2ϕ S þ ∂2z S ∂r ðr2 ∂r SÞ
ρ ρ r2
1
þ 2 ∂θ ðsin θ ∂θ SÞ
r sin θ
1
þ ∂2ϕ S
r2 sin 2 θ
!! ! ! ! ! ! ! ! !
rð A · B Þ ¼ A ðr B Þ þ B ðr A Þ þ ð A ·rÞ B þ ð B ·rÞ A (B.2)
! ! !
r·ðS V Þ ¼ V ·rS þ S r· V (B.3)
! ! ! ! ! !
r·ð A B Þ ¼ B ·ðr A Þ A ·ðr B Þ (B.4)
!
r·ðr V Þ ¼ 0 (B.5)
! ! !
r ðS V Þ ¼ rS V þ S r V (B.6)
! ! !
r r V ¼ rðr· V Þ r2 V (B.7)
r rS ¼ 0 (B.8)
! ! ! ! ! ! ! ! ! !
r ð A B Þ ¼ A ðr· B Þ B ðr· A Þ þ ð B ·rÞ A ð A ·rÞ B (B.9)
278 Vector analysis
References
p. 636–649.
[2] J. Marion, Classical Electromagnetic Radiation, Academic Press, 1965, p. 447–456.
[3] L. Eyges, The Classical Electromagetic Field, Dover, 1980, p. 387.
p. 13–16.
[5] Ibid., p. 21–22.
[6] E. R. Cohen (ed.), The Physics Quick Reference Guide, American Institute of Physics,
1996, p. 162.
C
Bessel functions
Using the method of separation of variables for the Laplace equation in cylindrical
coordinates gives rise to Bessel’s equation.[1, 2]

d dR
ρ ρ þ ðk2 ρ2 n2 Þ R ¼ 0:
dρ dρ
In this equation, R ¼ RðρÞ and k and n are separation constants. The parameter
n must be an integer to keep the azimuthal dependence of the solution single-
valued, i.e., we must have
ΦðϕÞ ¼ Φðϕ þ 2πnÞ:
Bessel’s equation is a second order differential equation that has two indepen-
dent classes of solution. One class involves Bessel functions of the first kind,[3]
RðρÞ ¼ Jn ðkρÞ: The behavior of the first three Bessel functions Jn are shown as
a function of kρ in Figure C1. All functions of this type are well-behaved at ρ = 0.
They are oscillatory with a decreasing amplitude that approaches zero as kρ → ∞.
The first root of the function J0 ðxÞ occurs at x = 2.405, where x = kρ. The first root of
J1 ðxÞ occurs at x = 3.832. The series expansion is
X
∞
ð1Þk xnþ2k
Jn ðxÞ ¼ :
k ¼ 0
k! ðn þ kÞ! 2
The Bessel functions satisfy the recurrence relation

2n
Jn ðxÞ ¼ Jn1 ðxÞ þ Jnþ1 ðxÞ;
x
while the derivatives satisfy the relation
2 J 0 n ðxÞ ¼ Jn1 ðxÞ Jnþ1 ðxÞ:
279
280 Bessel functions
Figure C1 Bessel functions of the first kind for n = 0, 1, 2.
Figure C2 Bessel functions of the second kind for n = 0, 1, 2.
The derivative of J0 is given by
dJ0 ðxÞ
¼ J1 ðxÞ:
dx
The other class of solutions to Bessel’s equation are the Bessel functions of
the second kind,[4] RðρÞ ¼ Nn ðkρÞ: The behavior of the first three Bessel functions
Nn are shown as a function of kρ in Figure C2. These solutions are also oscillatory
with decreasing amplitude that approach zero as kρ → ∞. However, they diverge at
ρ = 0, so they cannot be used in magnetostatics for any region that contains the
Bessel functions 281
origin. The Nn ðxÞ functions satisfy the same recurrence relations as Jn ðxÞ. The
derivative of N0 is given by
N 0 0 ðxÞ ¼ N1 ðxÞ:

If in applying the method of separation of variables for the Laplace equation in
cylindrical coordinates, we require that the solution along z is oscillatory, then the
separation parameter for the axial and radial terms must have the opposite sign
from that used in deriving the Bessel differential equation. This leads to the radial
equation

d dR
ρ ρ ðk2 ρ2 þ n2 Þ R ¼ 0:
dρ dρ
The solutions of this equation are known as modified Bessel functions. This same
equation can be produced by replacing k with i k in the ordinary Bessel equation.
One class of radial solutions involves the modified Bessel function In ðkρÞ.[5]
The behavior of the first three modified Bessel functions In are shown in Figure C3.
All functions of this type are well behaved at ρ = 0. They are related to the ordinary
Bessel functions by
Iν ðxÞ ¼ iν Jν ði xÞ:
The series expansion is

X
∞
1 xnþ2k
In ðxÞ ¼
k¼0
k! ðn þ kÞ! 2
Figure C3 Modified Bessel functions In ðkρÞ for n = 0, 1, 2.

282 Bessel functions
Figure C4 Modified Bessel functions Kn ðkρÞ for n = 0, 1, 2.
and it satisfies the recursion relations1

2n
In ðxÞ ¼ In1 ðxÞ Inþ1 ðxÞ
x
2I 0 n ðxÞ ¼ In1 ðxÞ þ Inþ1 ðxÞ:
The other class of solutions for the modified Bessel’s equation are the functions
Kn ðkρÞ: The behavior of the first three modified Bessel functions Kn are shown in
Figure C4. These solutions diverge at ρ = 0, so they cannot be used in any region
that contains the origin. The functions Kn satisfy the recursion relations2
2n
Kn ðxÞ ¼ Kn1 ðxÞ Knþ1 ðxÞ
x
2K 0 n ðxÞ ¼ Kn1 ðxÞ þ Knþ1 ðxÞ:
References
[1] F. Bowman, Introduction to Bessel Functions, Dover Publications, 1958.
[2] M. Abramowitz & I. Stegun (eds.), Handbook of Mathematical Functions, Dover
Publications, 1972, chapter 9.
p. 573–584.
[4] Ibid., p. 596–601.
[5] Ibid., p. 610–619.
1 2
GR 8.486.1,8.486.2. GR 8.486.10,8.486.11.
D
Legendre functions
Separation of variables for the Laplace equation in spherical coordinates gives the
partial differential equation
1 1
∂θ ðsin θ ∂θ YÞ þ ∂2ϕ Y þ lðl þ 1Þ Y ¼ 0
sin θ sin θ
2
for the angular dependence. The solution of this equation is given in terms of the
spherical harmonic functions Ylm [1, 2]
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð2l þ 1Þ ðl mÞ! m
Yl m ðθ; ϕÞ ¼ P ðcos θÞ eimϕ ;
4π ðl þ mÞ! l
where l and m are integers and Pm l is an associated Legendre function. Allowed

values of m are all integers in the range –l ≤ m ≤ l. Values of the spherical harmonics
for negative m are given by

Yl;m ¼ ð1Þm Yl;m ;
where the asterisk denotes complex conjugation. The spherical harmonics for l ≤ 2
are given in Table D1.
The polar angle part ΘðθÞ of the solution to the Laplace equation has to satisfy
the second order, ordinary differential equation

d 2 dΘ m2
ð1 x Þ þ lðl þ 1Þ Θ¼0
dx dx 1 x2
where x = cos θ. The solutions of this equation are called associated Legendre
functions of the first and second kind,
fPm
l ðxÞ; Ql ðxÞg:
m
283
284 Legendre functions
Table D1 Spherical harmonics
l m Ylm
0 0 1
pffiffiffiffiffi
4π
rffiffiffiffiffi
3
1 0 cos θ
4π
rffiffiffiffiffi
3
1 1 sin θ ei ϕ
8π
rffiffiffiffiffiffiffiffi
5
2 0 ð3 cos2 θ 1Þ
16π
rffiffiffiffiffi
15
2 1 sin θ cos θ ei ϕ
8π
rffiffiffiffiffiffiffiffi
15
2 2 sin2 θ e2i ϕ
32π
Only the functions of the first kind have convergent power series over the complete
range 0 ≤ x ≤ 1, so we choose
ΘðθÞ ¼ Pm
l ðcos θÞ:
The associated Legendre functions can be calculated from
ð1Þm dlþm
l ðxÞ
Pm ¼ l ð1 x2 Þm=2 lþm ðx2 1Þl :
2 l! dx
Associated Legendre functions with negative m are related to functions with

positive m by
ðl mÞ! m
Pm
l ðxÞ ¼ ð1Þ
m
P ðxÞ :
ðl þ mÞ! l
The associated Legendre functions for l ≤ 3 and m > 0 are given in Table D2.
In problems with azimuthal symmetry, we have m = 0. In this case, the associated
Legendre functions reduce to the ordinary Legendre polynomials.
P0l ðcos θÞ ¼ Pl ðcos θÞ

Legendre functions 285
Table D2 Associated Legendre functions
l m Plm
1 1 sin θ
2 1 3 cos θ sin θ
2 2 3 sin2 θ
3
3 1 ð5 cos2 θ 1Þ sin θ
2
3 2 15 cos θ sin2 θ
3 3 15 sin3 θ
Figure D1 Legendre polynomials for l ≤ 4.
The Legendre polynomials form a complete set of orthogonal functions over the
interval 1 ≤ cos θ ≤ 1. The behavior of the Legendre polynomials for l ≤ 4 are
shown in Figure D1 as a function of x. Legendre polynomials satisfy the recurrence
relation
ðl þ 1Þ Plþ1 ðxÞ ¼ ð2l þ 1Þ x Pl ðxÞ l Pl1 ðxÞ
and their derivatives satisfy the recurrence relation
ðx2 1Þ P0l ðxÞ ¼ l x Pl ðxÞ l Pl1 ðxÞ:

286 Legendre functions
References
chapter 12.
[2] M. Abramowitz & I. Stegun (eds.), Handbook of Mathematical Functions, Dover
Publications, 1972, chapter 8.
E
Complex variable analysis
We present here a brief summary without proofs of some of the important

results from complex analysis that are relevant to the material covered in this
book.[1, 2]
Complex variables
In a Cartesian coordinate system, the complex variable z is given by
z ¼ x þ i y;
pffiffiffiffiffiffiffi
where x is called the real part of z, y is called the imaginary part of z, and i ¼ 1.
In polar coordinates, z can be written in the form
z ¼ r eiθ
¼ r ðcos θ þ i sin θÞ;
where r is called the modulus of z and θ is the argument of z. The De Moivre
formula is useful for evaluating powers of z.
ðcos θ þ i sin θÞn ¼ cos nθ þ i sin nθ
The complex conjugate of a complex variable z is
z ¼ x i y:
The real and imaginary parts of a complex number can be written as

z þ z
ℝeðzÞ ¼
2
z z
ImðzÞ ¼ :
2i
287
288 Complex variable analysis
Care is required in working with the complex counterparts of some real func-
tions. An important function in magnetostatics is the complex logarithm function.
This is defined as
w ¼ ln z
¼ lnðr eiθ Þ
¼ ln r þ i ðθ þ 2πnÞ;
where n ¼ 0; 1; 2; …. This function has multiple branches of angular
width 2π, depending on the value of n.[3] We customarily compute this function
using the principal branch where n = 0 and where θ is in the range –π < θ ≤ π. In this
case, the function changes discontinuously when crossing the negative x axis,
which is called a branch cut.
Complex differentiation
The derivative of the complex function F is defined as [4]
Fðz þ ΔzÞ FðzÞ

F 0 ðzÞ ¼ lim ;
Δz→0 Δz
provided that the limit exists and is independent of the manner in which Δz
approaches 0. If the derivative of F exists at all points throughout some planar
region R, we say that the function is analytic in the region.[5] Examples of analytic
functions include polynomials, exponentials, trigonometric, and hyperbolic func-
tions. The real and imaginary parts of an analytic function are harmonic, i.e., they
satisfy the Laplace equation.
Points where a function F(z) is not analytic are called singularities. A singularity
in F(z) at a point z0 is called a pole of order n if [6]
lim ðz z0 Þn FðzÞ
z→z0
exists and is not 0.

An important property of analytic functions is that constraints exist between
their real and imaginary parts.
Theorem E.1 (Cauchy-Riemann) [7] (Necessity) If a function f ðzÞ ¼ uðx; yÞþ

ivðx; yÞ is analytic in some domain D, then u and v have continuous first partial
derivatives in D and satisfy the Cauchy-Riemann equations
∂u ∂v
¼
∂x ∂y
∂u ∂v
¼ :
∂y ∂x
Complex variable analysis 289
(Sufficiency) If a function f ðzÞ ¼ uðx; yÞ þ ivðx; yÞ is defined in D, if u and

v have continuous first partial derivatives in D and if the Cauchy-Riemann
equations hold in D, then f(z) is analytic in D.
For a region not including the origin, the Cauchy-Riemann equations can be written
in polar coordinates as
∂u 1 ∂v
¼
∂r r ∂θ
1 ∂u ∂v
¼ :
r ∂θ ∂r
Series
Theorem E.2 (power series) [8] Let f(z) be analytic on a domain G and let zo
be an arbitrary point of G. Let d ¼ dðz0 Þ be the distance between zo and the
boundary of G. Then there exists a power series
X
∞
f ðzÞ ¼ cn ðz zo Þn
n¼0
that converges to f(z) on the disk |z – zo| < d.
A power series can be differentiated or integrated term-by-term within its radius of

convergence.
Theorem E.3 (Taylor series) [9] Let f(z) be analytic and single-valued in an
open region G. Let a be any point in G and let C be a circle with center at a, which
together with its interior lies entirely in G. Then at every point z in C, the series
f 00 ðaÞ f ðnÞ ðaÞ

f ðaÞ þ f 0 ðaÞðz aÞ þ ðz aÞ2 þ þ ðz aÞn þ
2! n!
converges to f(z).
In other words, f(z) can be written as a Taylor series that converges in the region
jz aj < R; where R is the radius of convergence.
Theorem E.4 (Laurent series) [10] Let f(z) be analytic for the annular region
G: R1 < jz zo j < R2
and let C be any simple closed contour lying inside G and having zo in its interior.
Then for points z in G, the function f(z) may be expanded in the series
X
∞
f ðzÞ ¼ ck ðz zo Þk ;
k¼∞
where
þ
1 f ðzÞ
ck ¼ dz
2πi ðz z0 Þkþ1
and the integration is along the contour C.
The Laurent series is valid in a region surrounding, but not including, a singularity.
Note that this series includes negative values of k. The coefficient c1 has special
significance and is known as the residue.
Complex integration
Theorem E.5 [11] If f ðzÞ ¼ uðx; yÞ þ ivðx; yÞ is continuous on a simple smooth

arc from points a to b, then the integral exists and is given by
ð ðb
f ðzÞ dz ¼ ðu þ ivÞ ðdx þ i dyÞ:
a
Theorem E.6 (Cauchy integral theorem) [12] If f(z) is analytic in a simply

connected domain D, then
þ
f ðzÞ dz ¼ 0
on every simple closed path in D.
If instead of f(z), we consider the contour integral of f(z) / (z – zo), then we have the
following theorem.
Theorem E.7 (Cauchy’s Integral Formula) [13] Let f(z) be analytic within and
on a simple closed contour C. Then, if zo is a point inside C,
þ
1 f ðzÞ
f ðzo Þ ¼ dz:
2πi ðz zo Þ
This gives the value of f ðzo Þ at the singularity zo inside a region in terms of the
contour integral around the boundary C.
Complex variable analysis 291
Theorem E.8 [14] Let f(z) be analytic within and on a simple closed contour
C. Then all derivatives of f ðzo Þ exist at a point zo inside C and are given by
þ
ðnÞ n! f ðzÞ
f ðzo Þ ¼ dz:
2πi ðz zo Þnþ1
Theorem E.9 (Residue theorem) [15] Let f(z) be analytic within and on a simple
closed contour C, except for a finite number of isolated singularities inside C. Let σ be
the sum of the residues at the singular points of f(z) that lie inside C. Then
þ
1
f ðzÞ dz ¼ σ:
2πi
In other words, the value of the contour integral is 2πi times the sum of the
residues for the enclosed singularities. For a pole of order n, the residue can be
found as [16]
1 dn1
a1 ¼ lim n1 ½ðz aÞn f ðzÞ:
ðn 1Þ! z→a dz
In the case of a simple pole (n = 1), the residue is given by
a1 ¼ lim ðz aÞ f ðzÞ:

z→a
Conformal mapping
We can define a function F that maps a complex variable z into a variable w in
another two-dimensional space.
w ¼ FðzÞ:
Assume that two curves that cross at a point zo in the z space are separated by an
angle θ. A mapping w ¼ FðzÞ is conformal, or angle preserving, if the mapped
curves in the w space cross at the point wo ¼ Fðzo Þ with the same angle θ.
Theorem E.10 [17] A mapping defined by an analytic function F(z) is conformal,

except at points where the derivative F 0 ðzÞ is zero.
Theorem E.11 (Riemann mapping theorem) [18] Let D be a simply connected

domain with at least two boundary points. Then there exists a simple function
w ¼ FðzÞ which maps D onto the unit disk |w| < 1. If we specify that a given point
zo in D maps into the origin and a given direction at zo is mapped into a given
direction at the origin, then the mapping is unique.
Theorem E.12 (Schwarz-Christoffel transformation) [19] Let R be a polygon

in the w plane having vertices at w1 ; w2 ; . . . ; wn with corresponding interior angles
α1 ; α2 ; . . . ; αn respectively. Let the points w1 ; w2 ; . . . ; wn map into the points
x1 ; x2 ; . . . ; xn on the real axis of the z plane. Then the transformation
dw
¼ Aðz x1 Þα1 =π1 ðz x2 Þα2 =π1 ðz xn Þαn =π1 ;
dz
where A is a complex constant, maps the interior of the polygon in the w plane onto
the upper half of the z plane and maps the boundary of the polygon onto the real axis
of the z plane.
References
Wesley, 1962, p. 61–72.
[3] J. Dettman, Applied Complex Variables, Dover, 1984, p. 58–60.
[4] K. Miller, Introduction to Advanced Complex Calculus, Dover, 1970, p. 48–52.
[5] M. Spiegel, Complex Variables, Schaum’s Outline Series, McGraw-Hill, 1964, p. 63.
[6] Ibid., p. 67.
[7] W. Kaplan, Advanced Calculus, Addison-Wesley, 1952, p. 510–512.
[8] R. Silverman, Introductory Complex Analysis, Dover, 1972, p. 204–205.
[9] K. Miller, op. cit., p. 114–115.
[10] J. Dettman, op. cit., p. 163–164.
[11] Ibid., p. 81–82.
[12] W. Kaplan, op. cit., p. 516.
[13] J. Dettman, op. cit., p. 113.
[14] Ibid., p. 114.
[15] K. Miller, op. cit., p. 132–133.
[16] M. Spiegel, op. cit., p. 172–173.
[17] E. Kreyszig, Advanced Engineering Mathematics, Wiley, 1962, p. 719.
[18] J. Dettman, op. cit., p. 256–259.
[19] M. Spiegel, op. cit., p. 204, 218–219.
F
Complete elliptic integrals
An elliptic integral is an integral that can be written in the form [1]

ð pffiffiffiffiffiffiffiffiffi
Rðx; f ðxÞÞdx;
where R is a rational function and f is a third- or fourth-order polynomial in x. All

integrals of this type can be written in terms of the three standard forms.
ðθ 0
0 dθ
Fðk; θ Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 1 k2 sin2 θ
ð θ 0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0
Eðk; θ Þ ¼ 1 k2 sin2 θ dθ
0
ðθ 0
dθ
∏ðk; n; θ 0 Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi :
0 ð1 þ n sin θÞ 1 k2 sin2 θ
2
Each of these integrals depends on a parameter k called the modulus that satisfies
k2 ≤ 1. The third type of integral also depends on a second parameter n called the
characteristic.[2] When the upper limit of integration is
π
θ0 ¼ ;
2
these functions define the complete elliptic integrals of the first, second, and third1
kinds.
1
One should be aware that the complete elliptic integral of the third kind is sometimes defined with a negative
sign before the factor n in the denominator.
293
294 Complete elliptic integrals
ð π=2
dθ
KðkÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 1 k2 sin2 θ
ð π=2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
EðkÞ ¼ 1 k2 sin2 θ dθ
0
ð π=2
dθ
∏ðk; nÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 ð1 þ n sin θÞ 1 k2 sin2 θ
2
Moreover, these functions can alternatively be defined in polynomial form as

ð1
dx
KðkÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 ð1 x2 Þð1 k2 x2 Þ
ð 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
1 k 2 x2
EðkÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi dx
0 ð1 x2 Þ
ð1
dx
∏ðk; nÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi :
0 ð1 þ nx Þ ð1 x Þð1 k x Þ
2 2 2 2
It is important to emphasize that, despite the awkward nomenclature, the complete

elliptic integrals are functions of k, and in the case of the third kind, also a function of n.
The complete elliptic integrals K and E can be expressed in terms of the infinite
series
2 2
π 1 13
KðkÞ ¼ 1 þ k þ
2
k þ
4
2 2 24
2 2 2
π 1 k 1 3 k4
EðkÞ ¼ 1 þ ;
2 2 1 24 3
where k2 < 1.[2] Efficient numerical algorithms have been developed to calculate
the complete elliptic integrals.[3]
The dependences of the complete elliptic integrals of the first and second kinds are
shown as a function of k in Figure F1. Both functions have the value π/2 for k = 0.
The function EðkÞ has the value 1 for k = 1, while KðkÞ approaches ∞ as k → 1.
The behavior of the complete elliptic integral of the third kind for several values of n is
shown as a function of k in Figure F2. The function ∏ðk; nÞ increases as k increases
for all values of n. For a given value of k, the function increases as n becomes more
negative.
If the vector potential is defined in terms of complete elliptic integrals, we
need to take derivatives to find the magnetic field. In this case, we need to know
Complete elliptic integrals 295
Figure F1 Dependence of the functions KðkÞ and EðkÞ on the modulus k.
5.0 n = –0.7
4.5
4.0
n = –0.3
3.5
Π (k,n)
3.0 n=0
2.5
n = 0.3
2.0
1.5
0.0 0.2 0.4 0.6 0.8 1.0

k
Figure F2 Behavior of the complete elliptic integral of the third kind.
the derivatives of the complete elliptic integrals with respect to their

arguments.2
∂KðkÞ EðkÞ KðkÞ

¼
∂k kð1 k Þ
2 k
∂EðkÞ EðkÞ KðkÞ
¼
∂k k k
2
GR 8.123.2, GR 8.123.4.
296 Complete elliptic integrals
For the complete elliptic integral of the third kind, the derivatives are given by
[4, 5]
∂∏ðk; nÞ k
¼ ½EðkÞ ð1 k2 Þ∏ðk; nÞ
∂k ð1 k Þðk2 nÞ
2

∂∏ðk; nÞ 1 k2 n k 2 n2
¼ EðkÞ þ KðkÞ ∏ðk; nÞ :
∂n 2ðn 1Þðk2 nÞ n n
References
[1] CRC Standard Mathematical Tables, 14th ed., Chemical Rubber Co., 1965, p. 282.
[2] M. Abramowitz & I. Stegun (eds.), Handbook of Mathematical Functions, Dover,
1965, p. 590–591.
[3] W. Press, S. Teukolsky, W. Vetterling & B. Flannery, Numerical Recipes in Fortran,
2nd ed., Cambridge University Press, 1992, p. 254–261.
[4] National Institute of Standards and Technology, Digital Library of Mathematical
Functions, http://dlmf.nist.gov, equation 19.4.4.
[5] http://functions.wolfram.com/EllipticIntegrals/EllipticPi/introductions/CompleteElliptic
Integrals/05
Index
Allowed harmonics 92 Complex field examples

Ampere law annular block (on-axis) 131, 135
complex 115 block conductor 124
for B 14, 26 circular arc 117
for H 26 cos(ϕ) sheet 120
Amperian currents 24 current sheet 116
Analytic function 109, 288 cylindrical block 126
Area of current block 126 doublet 139
Associated Legendre functions line current 114
58, 283 magnetized body 139
multipole 151
B H curve 28 rectangular block 129
B H product 215, 217 triangular block 140
Bar magnet 211 Complex force 140
Bedstead coil 106 Complex multipoles 135
Bent solenoid 190 Complex potential 108
Bessel functions 56, 279 Complex potential examples
Bilinear interpolation 249 current sheet 115
Bilinear transformation 142 dipole 151
Biot-Savart law 6, 25 line current 113
Boundary conditions line current near iron corner 146
current sheet 17 multipole 150
regions with different μ 33 quadrupole 152
scalar potential 51 straight sheet 115
vector potential 44, 75 Complex rotation 109
Boundary value problems 59 Conformal mapping 141, 159, 163, 291
Boundary value problem examples line current in iron cavity 142
conductor near iron sheet 251, 266 line current near iron corner 144
cos(mθ) sheet in iron cavity (A) 85 Constitutive relations 241
cos(mθ) sheet in iron cavity (Vm) 87 Continuity equation 2
line current in iron cavity 77 Coulomb gauge 41
permeable sphere in field 63 Coupling constants 260
rectangular conductor in slot 59 Curie law 28
Curl operator 276
Cauchy integral formula 290 Current 2
Cauchy-Riemann equations 288 Current sheet 3
Circular current loop 10, 165 Curvilinear coordinates 190
Coercivity 30, 31, 162, 217
Complex conjugate coordinates 111, 287 De Moivre formula 287
Complex field 109 Derivatives (numerical) 246
image 132 Diamagnetism 27
297
298 Index
Diffusion equation 236 energy loss 235
Dirac delta function 13, 199
Dipole magnets Images method
C-dipole 158 circular boundary 37, 79, 115, 144
excitation 154 plane boundary 34, 133
H-dipole 157, 237 Inductance
symmetries 90, 263 coaxial solenoids 20
window frame 154 ideal solenoid 19
Dirichlet problem 59, 248 mutual 19, 170, 188
Displacement current 241 self 19, 232
Divergence operator 275 Integral equation method 68, 258
Divergence theorem 276 Integrated potentials 147
Domains 28 Intrinsic coercivity 30, 217
Doublet 138 Inverse Helmholtz coils 180
Inverse problem 206, 267
Easy axis 217 Iron enhancement 36, 79, 86, 134
Eddy currents 235
Effective length 159 Langevin equation 28
Electric field intensity (E) 230 Laplace equation 41, 53
Electric flux density (D) 240 cylindrical coordinates 55
Electric scalar potential 231 polar coordinates 71
Electromotance 230 rectangular coordinates 54
Elliptic integrals 293 spherical coordinates 57
End fields 105 Laplacian operator 247, 275
Energy stored in magnetic field 20, 215, 232 Laurent series 289
Legendre functions 283
Faraday’s law 229 Legendre polynomials 177, 284
Ferromagnetism 28 Lenz’s law 229
Finite difference method 245 Linear materials 27
Finite element method 254 Lines of induction 3, 63, 66
Finite element (triangular) 255
Flux linkage 19 Magnetic charges 53, 213
Force Magnetic circuit 31
between current loops 4 C-shaped electromagnet 32
between two parallel wires 8 with permanent magnets 214
between solenoid and loop 189 Magnetic field
complex 140 block 93, 124
in terms of inductance 20 circulation 13, 26
magnetized body 50 complex 109
on charged particle 5 current sheet 80, 116
on conductor 155 divergence 11
on localized current 49 flux density (B) 3
Frenet-Serret coordinates 191 harmonic expansion 72
intensity (H) 26
Gauss’ law for magnetism 11 radial expansion 170
Gradient field 161 Magnetic field from conductors
Gradient operator 275 annular sector 97, 131
Green’s function 68 circular loop 10, 168
Green’s theorem 67 circular arc sheet 118
complex 124 cos(ϕ) sheet 121
cylindrical block 14, 126
Harmonic functions 41 dipole 33, 154, 158
Helical wiggler fields 208 finite plane sheet 115
Helix 197 helix 202
relation to solenoid 204 Helmholtz coils 178
Helmholtz coils 178 ideal multipole sheet 84
Helmholtz theorem 276 infinite plane sheet 8
Hysteresis 30 line current 7, 114
Index 299
Maxwell tricoil 181 Paramagnetism 28
overlapping circular 95 Periodic axial fields 209
pair of loops 178 Permanent magnet
parallel current sheets 82 assemblies 224
quadrupole 90, 161 circuit excitation 214
rectangular block 97, 129 materials 216
solenoid 15, 182, 189 rare earth model 217
toroid 31, 194 Permeability 4, 27, 29, 31, 162
Magnetic field from image currents Permittivity 240
annular sector in iron cavity 135 Persistent currents 105
block in circular iron 135 Perturbations 153
cos(ϕ) sheet in iron cavity 134 Planar wiggler fields 204
line current 132 Poisson equation 41, 255
line current in plane boundary 35, 133 energy functional 255
sheet in circular boundary 134 POISSON program
Magnetic field from magnetized bodies algorithm 265
bar magnet 212 conductor near iron shield 266
dipole moment 47 permanent magnet 216
doublet 139 quadrupole 161
magnetized block 25, 104, 139, 264 sample input file 267
permeable sphere 66 solenoid 190
rare earth materials 218, 221, 224 window frame dipole 156
triangular block 140 Poles (complex) 288, 291
Magnetic flux 3, 40 Potential
Magnetic measurements 241 complex 108
Magnetic moment 6, 47 scalar 51
Magnetic needle 213 vector 39
Magnetic units 274
Magnetization 23 Quadrupole magnets
Maxwell’s equations 241 circular 161
complex 112 excitation 160
Meissner effect 104 hyperbolic 152
Modified Bessel functions 202, multipole 152
208, 281 permanent magnet assembly 227
Multipole field components
annular sector 99, 137 Racetrack coil 106
approximations 90, 153 Radial axial field expansion 170
complex 137 Refraction at current sheet 18
dependence on current distribution 88 Reluctance 31, 215
dimensionless coefficients 75, 271 Reluctivity 218
errors 91, 93, 106, 138 Remanence 30, 217
inside magnet aperture 74 Residual 250
normal 89 Residue 290
rare earth materials 221 Residue theorem 291
rotating coil 243 Resistivity 30, 161
skew 89 Rotating coil measurements 242
symmetries 89, 93 Rowland ring 32
trapezoidal block 226
Saddle coil 106
Neumann problem 59, 248 Saturation
Non-allowed harmonics 93 cause 29
effects 31, 86, 155, 157, 162, 237
Ohm’s law 236 modelling 251, 258, 263
Optimization Scalar potential 51
matching solenoid channels 268 Scalar potential examples
minimizing higher multipoles 270 cos(mθ) sheet in circular cavity 87
Orthogonal equipotentials 111 dipole 151
300 Index
Scalar potential examples (cont.) Toroid 193
inside magnet aperture 73 Torque 5
line current 114
magnetized body 52, 102, 213 Vector identities 276
permeable sphere 66 Vector potential
quadrupole 152 block conductors 94
rare earth magnets 219 circulation 39
Schwarz-Christoffel transformation current sheet 81
292 definition 39
line current near perpendicular divergence 41
planes 144 Vector potential examples
Segmentation efficiency 227 bifilar conductors 43
Segmented magnet assemblies circular loop 167, 169
224 cos(mθ) sheet in circular cavity 85
Separation of variables 53 dipole 151
Shape functions 256 helical conductor 200
Sheet theorem 119 ideal multipole sheet 84
Skin depth 240 ideal multipole block 100
Skin effect 239 inside magnet aperture 72
Soft magnetic alloys 30 line current 42, 77
Solenoid magnet line current in circular cavity 77
block model 189 line current in slot 59
derivative of field 183 localized current 47
sheet model 182 magnetized body 49
Spherical harmonics 283 quadrupole 152
Successive overrelaxation 253 sheet solenoid 187
Successive underrelaxation 266
Superconductors 104 Wiggler 204, 206
Susceptibility 27
Stokes Theorem 276 Zonal harmonic expansion 172

Fernow RC Principles of Magnetostatics

Uploaded by

Copyright:

Available Formats

Fernow RC Principles of Magnetostatics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fernow RC Principles of Magnetostatics

Uploaded by

Copyright:

Available Formats

PRINCIPLES OF MAGNETOSTATICS

The subject of magnetostatics—the mathematical theory that describes the forces

richard c. fernow received his PhD at Syracuse University for work on

Cambridge University Press is part of Cambridge University Press & Assessment,

3.7 Scalar potential for a magnetic body 52

7 Axial ﬁeld conﬁgurations 165

My career at Brookhaven National Laboratory began in 1978, when I was hired to

AS M. Abramowitz & I. Stegun (eds.), Handbook of Mathematical

I would like to thank several of my former colleagues, especially Gerry Morgan,

1.2 Magnetic forces

where the constant μ0 ¼ 4π 107 is known as the permeability of free space,4 dl is a

Note that the force only acts on moving charges.

Figure 1.1 Rectangular current loop in an external ﬁeld.

For the example here,

1.3 The Biot-Savart law

where K is the surface current density. Likewise, for a volume distribution of

Figure 1.2 Current in a long straight wire.

Example 1.1: ﬁeld from an inﬁnitely long straight wire

the ﬁeld at point P due to the current in the wire is

Thus the magnetic ﬁeld due to the current in the wire is

Figure 1.3 Force between two parallel wires.

Example 1.2: force between two parallel wires

Example 1.3: ﬁeld above an inﬁnite current sheet

Figure 1.4 Field above an inﬁnite current sheet.

The ﬁeld is given by

Figure 1.5 Field along the axis of a current loop.

Example 1.4: on-axis ﬁeld due to a circular current loop

1.4 Divergence of the magnetic ﬁeld

and taking the divergence of both sides, we have

Thus we ﬁnd that the divergence of the magnetic ﬁeld vanishes.

1.5 Circulation of the magnetic ﬁeld

Taking the curl of both sides of this equation, we ﬁnd

Then since r does not operate on J 0 , we have

Consider for the moment the relation

In the second integral in Equation 1.22 when R ≠ 0, we have

in spherical coordinates. Then since R ¼ j! r r0 j, we ﬁnd that r2 R1 ¼ 0:

Although r2 R1 is undetermined when R = 0, the integral of this expression is

1.6 The Ampère law

Example 1.5: a long cylindrical conductor

Thus the ﬁeld inside the conductor is

Example 1.6: ideal solenoid

where n is the number of conductor turns per unit length, or that

Similarly, on the path cdgh we ﬁnd

Bz ðrÞ ¼ μ0 nI; (1.28)

Figure 1.8 Gaussian pillbox across a current sheet.

1.7 Boundary conditions at a current sheet

Thus the normal component of B must be continuous across a current sheet.

Thus the change in the ﬁeld across the sheet is

B2t B1t ¼ μ0 K (1.30)

Figure 1.9 Fields near a current sheet.

Figure 1.10 Refraction of the magnetic ﬁeld crossing a current sheet.

The tangential component of B changes by an amount proportional to the current

The corresponding angle with the normal in region (1) is given by

Example 1.7: self-inductance of an ideal solenoid

The ﬂux in the solenoid is

Example 1.8: mutual inductance of two coaxial solenoids

and the mutual inductance is

1.9 Energy stored in the magnetic ﬁeld

Figure 1.11 LR circuit.