Analysis of torque in nested magnetic

cylinders
Cite as: Journal of Applied Physics 86, 6412 (1999); https://doi.org/10.1063/1.371705
Submitted: 17 May 1999 . Accepted: 16 August 1999 . Published Online: 17 November 1999

T. R. Nı́ Mhı́ocháin, D. Weaire, S. M. McMurry, and J. M. D. Coey

ARTICLES YOU MAY BE INTERESTED IN

The ideal dimensions of a Halbach cylinder of finite length

Journal of Applied Physics 109, 013915 (2011); https://doi.org/10.1063/1.3525646

Optimization and improvement of Halbach cylinder design

Journal of Applied Physics 104, 013910 (2008); https://doi.org/10.1063/1.2952537

Studies of strong magnetic field produced by permanent magnet array for magnetic
refrigeration
Journal of Applied Physics 95, 6302 (2004); https://doi.org/10.1063/1.1713046

Journal of Applied Physics 86, 6412 (1999); https://doi.org/10.1063/1.371705 86, 6412

© 1999 American Institute of Physics.

JOURNAL OF APPLIED PHYSICS VOLUME 86, NUMBER 11 1 DECEMBER 1999

Analysis of torque in nested magnetic cylinders

T. R. Nı́ Mhı́ocháin,a) D. Weaire, S. M. McMurry, and J. M. D. Coey
Department of Physics, Trinity College, Dublin 2, Ireland
共Received 17 May 1999; accepted for publication 16 August 1999兲
Rotatable nested cylindrical magnets may be used to generate variable magnetic fields. A mutual
torque exists between the cylinders which is an important consideration in practical designs. We
present a thorough investigation of this torque, combining experiment, computation, and theory. In
the finite length case, in which the magnetization varies according to the prescription of Halbach, the
torque is an end effect. It is proportional to sin ␣, where ␣ is the angle defining the relative
orientation of the cylinders. Practical designs use a discrete number N of permanent magnet
segments to approximate the continuously varying magnetization of the ideal design. These
introduce higher components sin(n␣) to the angular dependence of torque, where the allowed values
of n are restricted to n⫽(kN⫾1) according to the segmented magnet torque theorem. © 1999
American Institute of Physics. 关S0021-8979共99兲06122-8兴

I. INTRODUCTION the bore, and zero field everywhere outside the cylinder. The
magnitude of the induction B in the bore2 is

冉 冊
Permanent magnet variable flux sources have some ap-
Ro
pealing advantages over conventional electromagnets. They B⫽B r ln , 共1兲
are compact and require no cooling or external power sup- Ri
ply. Long permanent magnet cylinders, known as Halbach where B r is the remanence of the material, and R o ,R i are the
cylinders or dipole rings, create a magnetic field in their outer and inner radii of the cylinder, respectively.
bore which is uniform to a very good approximation. Two Figure 1共b兲 presents a practical approximation to this
such cylinders can be coaxially nested and rotated about design, as suggested by Halbach.3 A discrete number N of
their common axis to create a variable field as described uniformly magnetized segments is used to approximate the
by Leupold and Potenziani.1 uniformly varying magnetization of the ideal design. The
Among the issues encountered in the design of such de- direction of the magnetization in any segment is constant and
vices, there is the problem of mutual torque between the two given by twice the angular position ␺ ⫽ ␥ j of the center of
cylinders. The torque is exactly zero in the idealized model the segment, where ␥ j ⫽ j2 ␲ /N for j⫽1,2...,N. We refer to
of infinitely long cylinders with continuously varying orien- this design as the ‘‘segmented cylinder.’’ Inhomogeneities in
tation of the magnetization. Thus, its origins and magnitude the field within the bore and some stray field outside the
in real systems have not been apparent. cylinder result from the segmented structure and finite length
In this article, we present an experimental, numerical, of such a magnet.
and theoretical analysis of the torque as a function of the A pair of coaxially nested cylinders, one inside the bore
relative angle between the cylinders. Principles of symmetry of the other, is shown in Fig. 1共c兲. Such a pair can generate
are applied, and a theorem is proved concerning the func- magnetic fields which are continuously variable in both mag-
tional form of the energy and torque. An analytical theory nitude and direction, as suggested in Ref. 1. The resultant
emerges which is confirmed and complemented by numerical field in the central bore is the vector sum of fields from the
simulations. There remain to be resolved some minor dis- individual cylinders. By rotating the cylinders with respect to
crepancies with experimental observations, but the present each other, the magnitude and orientation of the central field
level of understanding provides a valuable design tool. can be uniformly varied. If the dimensions of the cylinders
The design of an ideal Halbach cylinder is indicated in are chosen such that the same bore field is produced by each
cylinder alone, then the magnitude of the resultant field of
Fig. 1共a兲. The axis of the cylinder is the z axis, and the
the pair can range from zero to twice that of a single cylin-
magnetization lies in the plane perpendicular to z. The mag-
der.
nitude of the magnetization is constant, while its orientation
Commercial cylinders are constructed using rare-earth
varies continuously: at an angular position ␺ measured
magnet materials such as Nd–Fe–B. These materials have
clockwise from the y axis, the magnetization has orientation
very square hysteresis loops with a high coercivity ( ␮ 0 H c
2␺ as shown in the figure. A cylinder with continuously ⬎1.5 T) and small transverse susceptibility ( ␹⬜ ⬇0.1), so
varying magnetization direction will be termed as an ‘‘ideal that the magnetization pattern of the cylinders is essentially
Halbach cylinder.’’ An infinitely long ideal Halbach cylinder unaffected by the large fields to which it is exposed.2 Thus,
produces a uniform magnetic field in the y direction within in theoretical calculations of the field due to such a cylinder,
it is a fair approximation to take the susceptibility in any
a兲
Electronic mail: [email protected] direction to be zero.

0021-8979/99/86(11)/6412/13/$15.00 6412 © 1999 American Institute of Physics

J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6413

FIG. 2. Graphical definition of the relative angular rotation ␣ between the

FIG. 1. 共a兲 Ideal infinite length dipole ring, 共b兲 finite length, segmented cylinders when the outer cylinder is kept in a fixed position.
Halbach approximation, 共c兲 two nested cylinders; the field in the bore is the
vector sum of the fields from the individual cylinders.

4
In a typical device, two cylinders are independently ro-
⌫ i⫽ 冕再
V
共 JⴛHapp兲 i ⫹ 兺
jkl
⑀ i jk r j J l 共 ⵜ k H app
l 兲 冎 dV. 共6兲

tated by motors which drive them to the positions necessary The torque and the energy are related by
to produce a given field. A variable torque is experienced as
dE
one cylinder is rotated relative to the other. Torque places ⌫⫽⫺ , 共7兲
strain on the gears and motors turning the cylinders, and can d␣
contribute significantly to their wear and tear. Hence, it is where ␣ is the angle of rotation of the inner cylinder relative
important to have a greater understanding of its origins. to the outer, as shown in Fig. 2, and E is the energy of one
We have investigated this torque experimentally 共Sec. cylinder in the field of the other.
II兲, using computer modeling 共Sec. IV兲, and analytically
共Sec. V兲. In Sec. III, we present the background theory nec- II. EXPERIMENTAL RESULTS
essary for our analysis. We show that the torque is absent for
We measured the variation in torque as a function of ␣
infinitely long ideal cylinders, and results from field nonuni-
for a nested two-cylinder system, where ␣ is the angle
formities in nested segmented cylinders of finite length.
through which the inner cylinder is rotated from the position
The energy dE of a small element of magnetic material
where its magnetization is fully aligned with that of the
with volume dV and uniform polarization J, placed in an
outer, from Fig. 2. Both cylinders were made of segments of
external magnetic field Happ, is given by
Nd–Fe–B. The cylinder dimensions are given in Table I.
dE⫽⫺J–Happ dV, 共2兲 Each cylinder was connected, via a system of gears, to a
where J is related to magnetization M by J⫽ ␮ 0 M. For these dc motor.5 The position of each cylinder at any instant was
rigid magnets, the polarization is equal to the remanence recorded via an absolute encoder.
J⫽Br . The torque ⌫ mot generated by the dc motor is directly
The element experiences a torque d⌫m , tending to align proportional to I, the current drawn,
J with Happ, given by ⌫ mot⫽k T I, 共8兲
d⌫m ⫽JⴛH app
dV, 共3兲 where k T is the torque constant given by
and a torque d⌫ f about a fixed origin 共 V⫺IR 兲
k T⫽ . 共9兲
d⌫ f ⫽rⴛF, 共4兲 ␻m
where r gives the position of the element relative to the Here, V is the operating voltage, R is the resistance at the
origin. motor terminals, and ␻ m is the angular speed of the rotor in
共rad/s兲. This equation is obtained by equating the power in-
F⫽⫺ⵜ 共 dE 兲 ⫽ⵜ 共 J–Happ兲 dV 共5兲 put
is the force on the element due to the gradient of the applied
field, in the case where J is constant over the volume dV. TABLE I. Cylinder dimensions and properties.
In the case of concentric cylinders, and as a result of the
reciprocity theorem given in Sec. III A, the torque on the Inner cylinder Outer cylinder
system may be calculated by considering the rotation of ei- Inner radius 共mm兲 26 52.5
ther cylinder about its axis in the field due to the other cyl- Outer radius 共mm兲 47.5 110
inder which is kept stationary. The total torque ⌫ consists of Length 共mm兲 100 100
Segment number N 8 8
contributions d⌫m and d⌫ f 关Eqs. 共3兲 and 共4兲兴 integrated over
Nominal value for 1.08 1.17
the volume of the first cylinder. The i th component of the remanence 共T兲
torque is given by
6414 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

to the motor, P i ⫽VI, to the total output power P o generated

by the motor. P o is the sum of the power associated with the
torque, ⌫ mot␻ m , and the power loss, RI 2 .
If the motor is run at constant voltage, the torque gener-
ated by the motor at any instant can be calculated by moni-
toring the variations in current drawn and motor speed. This
can then be related to the torque experienced by the cylinder
via the gear ratios. The gearing system comprised a planetary
gearbox, and a worm-wheel unit.
In the measurements reported below, the outer cylinder
was held in a fixed position while the inner cylinder was
rotated. The inner motor was run at a constant voltage of 12
V. The current, inner encoder position and voltage were re-
corded at regular time intervals, and the angular speed ␻ c of
rotation of the cylinder at each of these instants was deter-
mined. The angular speed ␻ m of the motor was found from

␻ m ⫽ ␻ c r gb共 n teeth /n start兲 , 共10兲

where r gb is the gearbox ratio, n teeth is the number of teeth in

the worm wheel, and n start is the number of starting teeth in
the worm screw.
From these data, the torque ⌫ mot generated by the motor
was calculated at each instant using Eqs. 共8兲 and 共9兲. This
yielded the torque ⌫ experienced by the cylinder:

⌫⫽⌫ motr gb共 n teeth /n start兲 ␩ m ␩ ww␩ gb , 共11兲

where ␩ m , ␩ ww and ␩ gb are the efficiencies of the motor,

worm wheel, and gearbox, respectively.
The cylinder torque for a constant applied voltage of 12
V is plotted in Fig. 3共a兲 as a function of the relative orienta-
tion ␣ for two complete rotations of the inner cylinder, and
the current and angular velocity of the cylinder are plotted in
Fig. 3共b兲. In this case, the cylinder was rotating in the clock-
wise direction, so that ␣ is negative. As will be explained FIG. 3. 共a兲 Torque on the inner cylinder over two full rotations, as calcu-
more fully later, the variation in torque around one full rota- lated from Eq. 共11兲, 共b兲 current drawn by inner cylinder motor and cylinder
angular velocity for two rotations, 共c兲 current data from clockwise and an-
tion of the cylinder is essentially sinusoidal. For one half of ticlockwise rotations combined to give complete information over one com-
each cycle, the motor is working to overcome the magnetic plete rotation.
torque. In this region, an increase in load torque is reflected
by an increase in the current drawn by the motor, and a
decrease in the speed of the motor head and vice versa. For
the second half of each cycle, both the motor and the mag- at four different values of constant speed and allowing the
netic torque rotate the cylinder in the same direction, so there voltage to vary. Each of these cases yielded results consistent
is effectively zero load torque on the motor. The current with those presented above.
drawn by the motor is reduced and its speed increased. How- The variation of the current as a function of ␣ may be
ever, the maximum 共no load兲 speed, n 0 , and the minimum written as a Fourier series
共no load兲 current, I 0 , are determined by the applied voltage
and the terminal resistance of the motor, and their values are ⬁

defined for a given motor. In this case n 0 ⫽4090 rpm, and I共 ␣ 兲⫽ 兺

n⫽1
关 a n cos共 n ␣ 兲 ⫹b n sin共 n ␣ 兲兴 . 共12兲
I 0 ⫽12.3 mA. Hence, in this region of the graph, both current
and speed remain approximately constant.
Figure 3共b兲 shows variation in current I for two complete A Fourier transform of the current data is shown in Fig. 4,
clockwise rotations. Similar data were collected for rotation plotting the amplitude of the current, 冑a 2n ⫹b 2n , as a function
in the anticlockwise sense. These two sets of data were com- of the order n of the term. The dominant term is evident at
bined to indicate the structure of the torque variation over n⫽1, indicating that the variation is mainly sinusoidal.
360° cycles. The combined data for one complete rotation Higher order terms are also evident, in particular at n⫽7, 15,
are shown in Fig. 3共c兲. and 23, each with a smaller neighboring peak around n⫽9,
These measurements were repeated for constant voltages 17, and 25, respectively. That is, higher order terms are sig-
of 6, 8, and 10 V, and also while rotating the inner cylinder nificant at n⫽(kN⫾1), where N is the number of segments
J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6415

FIG. 5. Cuboid of magnetic material with polarization J may be represented

as two planes with equal an opposite surface charge density ␴.

tion for the real cylinders, that each segment has a perfectly
rigid polarization J. The magnetic induction within a block
FIG. 4. Fourier transform of data in Fig. 3共c兲, showing a dominant peak at of magnetic material is
the n⫽1 term of the Fourier series, with successive peaks at the n⫽(kN
⫾1) terms as labeled, where N⫽8 is the number of segments and k is an B⫽ ␮ 0 H⫹J. 共14兲
integer.
In a magnetostatic system,
ⵜⴛH⫽0 共15兲
on the cylinder and k is an integer. These terms in the torque so we may define a scalar potential ␸ m by
variation are modelled numerically and accounted for ana-
lytically in the following sections. H⫽⫺ⵜ ␸ m . 共16兲
Since ⵜ–B⫽0, then from Eq. 共16兲
III. THE MODEL ⫺ⵜ–J
ⵜ–H⫽⫺ⵜ 2 ␸ m ⫽ 共17兲
␮0
The numerical and analytical calculations which we
present depend on the following: and ␸ m satisfies a Poisson equation with magnetic charge
共i兲 The reciprocity theorem which allows us to find the density ␳ m ⫽⫺ⵜ–J.
torque by calculating the energy of the inner cylinder in the In the ideal case, the field is calculated from a nonzero
field of the outer; volume charge density, as in the Appendix. In the case of a
共ii兲 The rigidity of the polarization in the high rema- segmented cylinder, however, each segment has uniform and
nence, high coercivity materials used to construct the cylin- perfectly rigid J. Thus, the volume magnetic charge density
ders. This allows us to use a magnetic charge model for our ␳ m in each segment is zero. In this case, the magnet may be
field calculations, and in particular, a surface charge model represented by charged surfaces, as in the surface charge
for calculating the field due to a segmented cylinder 共Sec. model of Yonnet.7 The magnetic charge density ␴ on a sur-
IV A兲; face with unit normal n̂ is given by
共iii兲 The symmetry properties of the cylinders which lead
␴ ⫽J–n̂. 共18兲
to the ‘‘segmented magnet torque theorem’’ 共Sec. V B兲 con-
cerning the angular dependence of the torque. By way of illustration, Fig. 5 shows a cuboidal magnet rep-
resented by two charged surfaces. Analytical expressions for
A. Reciprocity
the vector field at a point, due to a charged rectangular sur-
The reciprocity theorem6 for magnetostatic energy states face have been derived by Akoun and Yonnet.8
that the energy, E, of one magnet due to the field of a second
is equal to the energy of the second due to the field of the C. Consequences of particular symmetries in Halbach
first. That is, cylinders

E⫽ 冕all space
H1 –J2 dV⫽ 冕
all space
H2 –J1 dV, 共13兲
Both the ideal and segmented designs have a bilateral
symmetry: i.e., J is symmetric about ␺ ⫽0:

where Hi is the magnetic field produced by magnet with J共 ␺ ⫽ ⑀ 兲 ⫽J共 ␺ ⫽⫺ ⑀ 兲 . 共19a兲

polarization Ji , where i⫽1,2. The inner and outer cylinders In the ideal case and the case of cylinders with an even
must experience equal and opposite torques which may be number of segments N, there is also a vertical asymmetry:
calculated from the energy using Eq. 共7兲. In our nested i.e., J is antisymmetric about ␺ ⫽ ␲ /2,
model, we choose to calculate the torque by considering the
effect on the inner cylinder of the field due to the outer one. J共 ␺ ⫽ ␲ /2⫹ ⑀ 兲 ⫽⫺J共 ␺ ⫽ ␲ /2⫺ ⑀ 兲 . 共19b兲
for any angle ⑀. Note that the angle ␺ is measured from the y
axis as in Fig. 1, and is related to the usual polar angle ␾, by
B. Magnetic charge model
␺ ⫽ ␲ /2⫺ ␾ .
As the first step in the simulation, we calculate the field Now consider the case of two infinitely long ideal cylin-
due to the outer cylinder alone 共Sec. IV A兲. The model de- ders 共with polarization direction varying continuously兲. Each
pends on the assumption, which is an adequate approxima- has zero external stray field and a perfectly uniform bore
6416 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

FIG. 6. 共a兲 Grid of points, in one plane, at which the field of the outer
cylinder is calculated. The inner cylinder is placed in the bore, and those
grid points lying within the cross section of the inner cylinder are used for
the torque calculation, 共b兲 Planes along the length of the cylinder.

field. Therefore, the field due to the inner cylinder in the

volume occupied by the outer cylinder is zero. Hence, ac-
cording to Eq. 共6兲, the outer cylinder experiences no torque.
By reciprocity, there can be no torque on the inner cylinder,
which experiences a uniform field due to the outer cylinder. FIG. 7. 共a兲 Torque of the inner cylinder as a function of angular position ␣
This implies that an ideal cylinder experiences zero torque in within the bore of the static outer cylinder, 共b兲 Energy of the inner cylinder
as a function of ␣.
a uniform field. As a consequence of the symmetries in Eqs.
共19a兲 and 共19b兲, a finite length cylinder composed of an even
number N of segments also experiences zero torque in a uni-
form field, as the torque on each segment is exactly balanced on the grid, and L is the separation between successive
by an equal and opposite torque on another segment. planes. This volume element must be sufficiently small to
The torque observed experimentally must therefore arise ensure the accuracy of the calculation.
from nonuniformities in the field, which are attributable to
the segmented structure and finite length of both inner and B. Torque calculation
outer cylinders.
The total torque on the inner cylinder is calculated by
summing over individual volume elements ⌬V, as in Eq. 共6兲.
Similarly for the total energy E:
IV. NUMERICAL SIMULATION

There are two main steps in calculating the variation of ⌫⫽ 兺i ⌫共 i 兲⫽ 兺i 共 J共 i 兲ⴛH共 i 兲⌬V⫹r共 i 兲ⴛF共 i 兲 兲 , 共20a兲
the torque as the cylinders rotate. First, the field produced in
the bore of the outer cylinder is calculated numerically, using
the surface charge model, as described in Sec. IV A. Second, E⫽ 兺i E 共 i 兲⫽⫺ 兺i J共 i 兲–H共 i 兲⌬V. 共20b兲
the torque experienced by the inner cylinder in this field is
calculated, as in Sec. IV B. In Sec. IV C, we present the Here, H(i) is the field due to the outer cylinder at the position
results of these calculations. Section IV D presents a numeri- r(i) of the i th volume element of the inner cylinder, and J(i) is
cal analysis of the nonuniformities in the profile of the field the polarization of that element. The force F(i) at each point
in the cylinder bore, and relates these to the torque. is proportional to the field gradient at that point, from
Eq. 共5兲.
A. Field calculation
For the following calculation on segmented cylindrical
C. Numerical results
magnets, each segment is treated as a combination of prisms
in order to approximate curved surfaces. The total field at In this section, the results of the calculation of ⌫ and E
any point is simply the sum of the field contributions from are presented for cylinders of the same dimensions as those
each face, calculated using the surface charge model 共Sec. used to obtain the experimental data, given in Table I. In
III B兲. On a cross section of the cylinder, we specify a grid of accordance with expectations, the only nonzero component
points, each with position vector r⬅(x,y,z) as shown in Fig. of torque is ⌫ z , about the axis of the cylinder.
6共a兲. At each of these points within the bore of the outer Figures 7共a兲 and 7共b兲 show ⌫ z and E, respectively, as
magnet, we calculate the components of the field H x , H y , functions of ␣ 共the angle through which the inner cylinder is
and H z . The field varies along the length of the cylinder as rotated from the fully aligned position, as in Fig. 2兲. The
well as across the width of the bore, and so is calculated energy reaches a minimum at the fully aligned ( ␣ ⫽0) posi-
across similar grids on a number of different, equally spaced tion. The energy data are closely fitted by a cosine function,
planes as shown in Fig. 6共b兲. Each point at which the field is E⫽⫺A cos ␣, where A is a positive constant. There are two
calculated is surrounded by an element with volume ⌬V contributions to ⌫ z in Eq. 共20a兲 关see Eqs. 共3兲 and 共4兲兴, each
⫽Ld 2 , where d is the separation between neighboring points of which is shown in Fig. 8. The ⌫ mz and ⌫ f z terms are
J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6417

tracted from the original data. In doing so, we largely re-

move the n⫽1 and n⫽7 terms. A Fourier transform of the
remaining data is shown in Fig. 9共b兲. Higher order terms can
now be seen to occur in pairs at n⫽(N⫺1),(N⫹1),(2N
⫺1),(2N⫹1),(3N⫺1),(3N⫹1), etc., where N⫽8 is the
number of segments 共i.e., peaks at n⫽7,9,15,17,23,25, etc.兲.
These terms coincide with the n⫽(kN⫾1) terms which
emerged experimentally in Fig. 4. These results will be ac-
counted for by the theory of Sec. V.
FIG. 8. Resultant torque ⌫ z is a combination of two contributions.
D. Analysis of the field profile
Numerical calculation confirms that for a segmented cyl-
found to be well approximated by A sin(␣) and ⫺2A sin(␣),
inder with even N placed in a completely uniform field, the
respectively. A good approximation to the total torque is
torque vanishes at every value of ␣. Therefore, in the case of
therefore given by ⌫ Z ⫽⫺A sin(␣).
two segmented cylinders, the existence of finite torque, and
Numerical values of the contributions to the torque from
the dependence of E and ⌫ on ␣, arise from inhomogeneities
different cross sections along the axis of the cylinder show
in the field.
that the main contributions come from the end regions. The
We have investigated these inhomogeneities by using
total value of the torque is not significantly changed by in-
the surface charge model to calculate the field in the bore of
creasing the lengths of the cylinders, while the contribution
an outer cylinder with dimensions as given in Table I. We
from the central region becomes negligible. This indicates
demonstrate separate contributions to field nonuniformities
that the torque is primarily an end effect associated with the
by varying segment number N and length L.
field inhomogeneities at the cylinder ends.
In addition to the fundamental sinusoidal variation, ⌫ z is 1. Approximation of ideal infinite length cylinder
modulated by structure associated with segmentation. In
Figure 10共a兲 shows the profile of H y in the bore midway
similar contexts, this structure is often referred to as ‘‘cog-
along the length of the cylinder (z⫽0) when the cylinder is
ging torque.’’ Figure 9共a兲 shows the amplitude of the Fourier
very long 共to approximate infinite length兲 and has a large
components of the torque data in Fig. 7共a兲. The order of the
number of segments, N⫽128 共to approximate uniformly
dominant term in the torque variation, n⫽1, corresponds to
varying magnetization兲. To the accuracy of our numerical
the leading sine term. The seventh order term is also pro-
procedure, H y remains constant at all points in this central
nounced. Higher harmonics are present, but with much
plane, as expected for an ideal cylinder of infinite length.
smaller amplitudes. To see these more clearly, we fit a curve
of the form B sin(␣)⫹C sin(7␣) to the data in Fig. 7共a兲,
where B and C are constants. The fitted curve is then sub- 2. Field variation due to finite length
Figure 10共b兲 shows the profile at z⫽0 in a cylinder with
N⫽128 to approximate uniformly varying magnetization,
but with L⫽100 mm. The magnitude of the field at ␺ ⫽0
and ␺ ⫽ ␲ is greater than that at ␺ ⫽ ␲ /2 and ␺ ⫽3 ␲ /2. As
the cylinder is made longer, the extent of this effect at the
central z⫽0 position decreases, although it persists at both
ends. This is therefore an end effect, associated with the
finite length of the cylinder.
To determine a specific form for the end effect of Fig.
10共b兲, consider the calculated magnitudes of H y and H x as a
function of angular position ␺ on a contour of fixed radius
within the bore such as that shown in Fig. 11. H y as a func-
tion of ␺ is shown in Fig. 12共a兲. This function is well fit by
H y ⫽H 0 ⫹⌬H cos共 2 ␺ 兲 . 共21a兲
Figure 12共b兲 plots H x against ␺, which is fitted by
H x ⫽⌬H sin共 2 ␺ 兲 . 共21b兲
Equations 共21a兲 and 共21b兲 are accounted for analytically in
the Appendix.
FIG. 9. 共a兲 Fourier transform of the data in Fig. 7共a兲, with dominant terms of
order n⫽1 and 7 as labeled, 共b兲 Fourier transform of the data in Fig. 7共a兲 3. Field variation due to segmentation
after a function of the form B sin(␣)⫹C sin(7␣) is fit to the data and sub-
tracted off. Peaks of order n⫽(kN⫾1) are now apparent, where N⫽8 is the Figure 10共c兲 shows the profile for an eight-segment, fi-
number of segments and k is an integer. nite length cylinder. Again the end effect is apparent, but in
6418 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

FIG. 11. Field values in Fig. 12 are taken clockwise from the starting point
around a circular contour inside the bore.

them to prove the segmented magnet torque theorem in Sec.

V B. In Sec. V C, we show analytically that the primary
variation is due to finite length, while in Sec. V D, we show
that, in the limit of very long magnets, the torque does not
increase with length. As an extension of this we identify the
way in which the torque scales with the cylinder dimensions.

A. Symmetry arguments in the ideal, finite length

case
First, we consider a pair of nested, ideal, finite length
Halbach cylinders, that is, cylinders in which the magnetic
polarization varies continuously as shown in Fig. 1共a兲. The
polarization J depends only on the angle ␺ 共measured clock-
wise from the y axis兲, while its magnitude J is independent
of position:
J共 ␺ 兲 ⫽J 关 sin 共 2 ␺ 兲 x̂⫹cos 共 2 ␺ 兲 ŷ兴 . 共22兲
If the relative orientation of the cylinders ␣ is measured from
the minimum energy configuration, where the polarizations
in the two cylinders are fully aligned 共Fig. 2兲, then we shall
show that the energy satisfies
E 共 ␣ 兲 ⫽E 共 0 兲 cos ␣ , 共23a兲
where we expect E(0)⬍0. The corresponding torque is
FIG. 10. Profiles of the y component of field at z⫽0 in the bore of 共a兲 an
dE
ideal, infinite length cylinder, 共b兲 an ideal, finite length cylinder, 共c兲 an ⌫ 共 ␣ 兲 ⫽⫺ ⫽E 共 0 兲 sin ␣ . 共23b兲
eight-segment, finite length cylinder. d␣
The proof follows:
addition, peaking in the field occurs near the boundaries be- At a point in an ideal Halbach cylinder with angular position
tween adjacent segments throughout the bore. ␺, the polarization J, Eq. 共22兲, makes angle ␺ with the radius
Figures 12共c兲 and 12共d兲 show H y and H x as functions of vector as shown in Fig. 1共a兲. So rotating the inner cylinder
␺. The peaking effects of the eight segments appear super- through an angle ␪ relative to the outer one is equivalent to
imposed on the end effect of Figs. 12共a兲 and 12共b兲. As the keeping both cylinders fixed and rotating the polarization at
number of segments is increased towards the case of an each point in the inner cylinder through angle ⫺␪, as shown
ideal, but finite length cylinder, the magnitude of the seg- in Fig. 13.
mentation peaking decreases. The energy of an element of the inner cylinder of vol-
ume dV at the point r with coordinates ( ␳ , ␺ ,z) is
V. THEORETICAL MODEL dE 共 r兲 ⫽⫺J共 r兲 –H共 r兲 dV⫽⫺JH 共 r兲 cos ␤ 共 r兲 dV, 共24兲
In this section, we demonstrate theoretically that the where J(r) is the polarization of the inner cylinder, and ␤ (r)
variation of the torque with relative orientation of the cylin- is the angle between this polarization and the field H(r) due
ders arises from field inhomogeneities due to their finite to the outer cylinder at that point. Let E( ␪ 0 ) be the result of
length and segmented structure. We apply symmetry argu- integrating Eq. 共24兲 over the volume of the inner cylinder,
ments in the case of an ideal cylinder in Sec. V A, and extend for some relative orientation of the cylinders ␣ ⫽ ␪ 0 .
J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6419

FIG. 12. 共a兲 H y around contour of 20 mm radius; 共b兲 H x around contour of 20 mm radius; 共c兲 H y around contour of 40 mm radius; 共d兲 H y around contour
of 40 mm radius; all at z⫽0 in a bore of 52.5 mm radius.

Notice that integration of Eq. 共24兲 over the angular vari- where, for convenience, we have introduced complex nota-
able ␺ gives zero if the field is entirely uniform. This follows tion. The magnitude of the polarization, and the field H and
because, in the case where H is independent of r and parallel angle ␤ at the point r are unchanged. Integrating this expres-
to the y axis, the angle ␤ (r)⫽2 ␺ ⫹ ␪ 0 . Integration over ␺ sion over the volume V of the inner cylinder will give the
then gives zero energy and zero torque, in agreement with form of the energy as
the symmetry arguments presented in Sec. III C.
E 共 ␪ ⫹ ␪ 0 兲 ⫽Re兵 f 共 ␪ 0 兲 e i␪ 其 , 共26a兲
For an arbitrary initial relative orientation ␪ 0 , the energy
after a further rotation through ␪ is found by rotating the where

冕
polarization at each point through ⫺␪. The energy of the
volume element at point r becomes f 共 ␪ 0 兲 ⫽⫺J dVH 共 r兲 exp关 i ␤ 共 r兲兴
V
dE ⬘ 共 r兲 ⫽⫺JH 共 r兲 cos关 ␤ 共 r兲 ⫹ ␪ 兴 dV
⫽ 兩 f 共 ␪ 0 兲 兩 exp关 i ␤ 1 共 ␪ 0 兲兴 . 共26b兲
⫽Re兵 ⫺JH 共 r兲 e i共 ␤ 共 r兲 ⫹ ␪ 兲 其 dV, 共25兲
Equation 共26a兲 then gives
E 共 ␪ ⫹ ␪ 0 兲 ⫽ 兩 f 共 ␪ 0 兲 兩 cos关 ␪ ⫹ ␤ 1 共 ␪ 0 兲兴 . 共26c兲
If the initial orientation is chosen to be the minimum energy
orientation, then ␪ 0 ⫽0, and the angle ␪ becomes ␣, the an-
gular displacement from the minimum energy position from
Fig. 2. The bilateral symmetry of the polarization, Eq. 共19a兲,
implies that
E 共 ␣ 兲 ⫽E 共 ⫺ ␣ 兲 共27兲
and this is only satisfied when ␤ 1 (0)⫽0, giving the results
Eqs. 共23a兲 and 共23b兲 advertised above for the energy and
torque. This result is confirmed analytically in Sec. V C.

B. The segmented magnet torque theorem

In the case of segmented cylinders the discussion above
FIG. 13. Rotating the inner, ideal Halbach cylinder by ␪ is equivalent to may be generalized to prove the following theorem:
rotating its magnetization at each point by ⫺␪. The torque between a pair of segmented cylinders of finite
6420 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

length and relative orientation ␣ is obtained by differentiat-

ing the energy with respect to ␣, where the energy may be
written
再
E 共 s 兲 共 ␣ 兲 ⫽Re c 0 exp共 i ␣ 兲 ⫹ 兺 共 c k exp关 i 共 kN⫹1 兲 ␣ 兴
k⫽1

E 共 ␣ 兲 ⫽a 0 cos ␣ ⫹ 兺 关 a k cos共 kN⫺1 兲 ␣ ⫹b k cos共 kN⫹1 兲 ␣ 兴

k⫽1
⫹c ⫺k exp关 i 共 ⫺kN⫹1 兲 ␣ 兴 兲 . 冎 共33兲

共28兲 The bilateral symmetry requirement in Eq. 共27兲 eliminates all

sine terms in Eq. 共33兲. Thus, this expression for the energy
and the summation is over all positive integers k. reduces to Eq. 共28兲, and since the orientation ␣ ⫽0 corre-
In the segmented case, the direction of the polarization sponds to a minimum energy, we should expect a 0 ⬍0. Dif-
J(s) is constant within a segment, so that rotation of the inner ferentiation of Eq. 共28兲 gives the following expression for
cylinder through angle ␪ is equivalent to rotation of J(s) the torque:
through angle ⫺␪ only when ␪ is an integer multiple of
2 ␲ /N. It is this restricted symmetry property which dictates
the form of Eq. 共28兲, as follows.
⌫ 共 ␣ 兲 ⫽a 0 sin ␣ ⫹ 兺
k⫽1
关共 kN⫺1 兲 a k sin共 kN⫺1 兲 ␣

Again, it is convenient to use complex notation and write

⫹ 共 kN⫹1 兲 b k sin共 kN⫹1 兲 ␣ 兴 . 共34兲
the energy of an element at point r, for some arbitrary ori-
entation angle ␪ 0 in the range 0 to 2 ␲ /N, as The allowed terms which emerge in this equation agree with
those seen both experimentally and numerically, shown in
dE 共 s 兲 共 r 兲 ⫽⫺J 共 s 兲 H 共 x 兲 共 r兲 cos ␤ 共 s 兲 共 r兲 dV Figs. 4 and 9, respectively.
共 s 兲 共 r兲
⫽⫺Re兵 J 共 s 兲 H 共 s 兲 共 r兲 e i ␤ 其 dV, 共29兲
C. Analytical results for the finite length effect
where, as in Eq. 共24兲, ␤ (s) (r) is the angle between J(s) and
In the previous section, we showed by symmetry argu-
H(s) at that point, and the superscript 共s兲 identifies the seg-
ments that the leading term in the torque variation is sinu-
mented case. Now after a rotation through 2 ␲ /N, the energy
soidal, as shown in Eq. 共34兲. In this section, we show ana-
of the element becomes
lytically that the leading sinusoidal term is determined by
dE 共 s 兲 共 r兲 ⫽⫺Re兵 J 共 s 兲 H 共 s 兲 共 r兲 exp关 i共 ␤ 共 s 兲 共 r兲 ⫹2 ␲ /N 兲兴 其 dV. field nonuniformities due to the finite length of the cylinders.
共30兲 Consider nested cylinders of finite length, but uniformly
varying magnetization. The outer cylinder is kept fixed and
Integration over the volume of the inner cylinder gives the the inner rotated through ␣ from the fully aligned position as
energy corresponding to the relative orientation ␣ ⫽ ␪ 0 in Fig. 2. From Eq. 共6兲, the torque d⌫ z on any volume ele-
⫹2 ␲ /N as ment of the inner cylinder, dV, centered on point r(x,y,z)
⬅r( ␳ , ␺ ,z), is given by
E 共 s 兲 共 ␣ 兲 ⫽Re兵 f 共 s 兲 共 ␪ 0 兲 exp共 i2 ␲ /N 兲 其
d⌫ z 共 ␳ , ␺ ,z 兲 ⫽ 共 J x H y ⫺J y H x 兲 dV⫹ 共 xF y ⫺yF x 兲
⫽Re兵 f 共 s 兲 共 ␪ 0 兲 exp关 i共 ␣ ⫺ ␪ 0 兲兴 其 , 共31兲

where, corresponding to Eq. 共26b兲, 冋 冉

⫽ 共 J x H y ⫺J y H x 兲 ⫹x J x
⳵Hx
⳵y
⫹J y
⳵Hy
⳵y 冊
f 共 s 兲 共 ␪ 0 兲 ⫽⫺J 共 s 兲 冕
V
dVH 共 s 兲 共 r兲 exp关 i ␤ 共 s 兲 共 r兲兴 .
冉
⫺y J x
⳵Hx
⳵x
⫹J y
⳵Hy
⳵x 冊册 dV, 共35兲

We define where the force on the element has components F i

⫽ 兺 l J l (ⵜ i H l )dV, and J lies in the xy plane.
F 共 ␣ 兲 ⫽ f 共 s 兲 共 ␪ 0 兲 exp共 ⫺i␪ 0 兲 , When the two cylinders are in the fully aligned position,
the polarization J at an angular position ␺ in each cylinder is
and Eq. 共31兲 may be written in terms of the angle ␣:
at 2 ␺ to the y axis. After rotating the inner cylinder through
E 共 s 兲 共 ␣ 兲 ⫽Re兵 F 共 ␣ 兲 exp共 i␣ 兲 其 . 共32兲 ␣, the polarization J at a point in the inner cylinder with
orientation ␺ is given by
Since F( ␣ ) depends only on ␪ 0 , we may write
J⫽J 关 sin共 2 ␺ ⫹ ␣ 兲 x̂⫹cos共 2 ␺ ⫹ ␣ 兲 ŷ兴 . 共36兲
F 共 ␣ 兲 ⫽F 共 ␪ 0 ⫹2 ␲ /N 兲 ⫽F 共 ␪ 0 兲 . In the Appendix, we have derived analytical forms of the x
and y components of the field in the bore, Eq. 共A18兲:
Thus, F( ␣ ) is periodic in 2 ␲ /N, and may be expanded in a
Fourier series: H x ⫽⌬H 共 ␳ ,z 兲 sin共 2 ␺ 兲 , 共37a兲
⫹⬁ H y ⫽H 0 共 ␳ ,z 兲 ⫹⌬H 共 ␳ ,z 兲 cos共 2 ␺ 兲 , 共37b兲
F共 ␣ 兲⫽ 兺 c k exp共 ikN ␣ 兲 .
k⫽⫺⬁ where the expressions for H 0 ( ␳ ,z) and ⌬H( ␳ ,z) are given in
Eq. 共A19兲. In the case of infinite length, ⌬H→0 and H 0
Substitution of this expansion into Eq. 共32兲 gives, for arbi- →(J/ ␮ 0 )ln(Ro /Ri) as in Eq. 共1兲. The field gradient has com-
trary relative orientation ␣, an expression of the form ponents
J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6421

⳵ H x ⳵ 共 ⌬H 兲 2⌬H D. Torque in the case of long ideal Halbach cylinders

⫽ sin ␺ sin共 2 ␺ 兲 ⫹ cos ␺ cos共 2 ␺ 兲 ,
⳵x ⳵␳ ␳ In the case of ideal Halbach cylinders of finite length,
共38a兲 the torque tends to a constant as the length is increased. It is
therefore an end effect. This is demonstrated theoretically
⳵ H x ⳵ 共 ⌬H 兲 2⌬H below.
⫽ cos ␺ sin共 2 ␺ 兲 ⫺ sin ␺ cos共 2 ␺ 兲 ,
⳵y ⳵␳ ␳ Consider first a single semi-infinite Halbach cylinder oc-
共38b兲 cupying ⫺⬁⬍z⬍0. The field at a distance z from the end of
the cylinder is given by H(z) 共suppressing the other spatial
⳵ H y ⳵ 关 H 0 ⫹⌬H cos共 2 ␺ 兲兴 variables ␳, ␺兲. For large z, this field is the sum of contribu-
⫽ sin ␺
⳵x ⳵␳ tions from distant dipoles, so it must diminish with z at least
as fast as
2⌬H

冕
⫺ cos ␺ sin共 2 ␺ 兲 , 共38c兲 ⬁
␳
共 z ⬘ 兲 ⫺3 dz ⬘ ⫽O 共 z ⫺2 兲 . 共45兲
z
⳵ H y ⳵ 关 H 0 ⫹⌬H cos共 2 ␺ 兲兴 2⌬H
⫽ cos ␺ ⫹ sin ␺ sin共 2 ␺ 兲 A Halbach cylinder of finite length L may be treated as a
⳵y ⳵␳ ␳
finite section of an infinite cylinder, which consists of the
共38d兲
finite section with a semi-infinite section added at each end.
and The field H(z) in the bore of the finite section is the differ-
ence between the uniform field H0 due to the infinite cylin-
x⫽ ␳ sin ␺ , 共39a兲 der and the contributions ⌬H1 (z) and ⌬H2 (z) from the two
semi-infinite segments:
y⫽ ␳ cos ␺ . 共39b兲
H共 z 兲 ⫽H0 ⫺⌬H1 共 z 兲 ⫺⌬H2 共 z 兲 . 共46兲
Substituting into Eq. 共35兲, and cancelling terms, we find The torque on a Halbach cylinder in a uniform field is zero
共Secs. III C and V A兲. So, in considering the torque on a
d⌫ z ⫽ 共 JH 0 sin共 2 ␺ ⫹ ␣ 兲 ⫹J⌬H sin ␣ ⫺2J⌬H sin ␣ 兲 dV.
second Halbach cylinder of length L, nested within the first
共40兲
one 共also of length L兲, there is no contribution from H0 , and
Equation 共40兲 contains two separate contributions: we can use ⫺(⌬H1 ⫹⌬H2 ) instead of H. Expressing this
torque in terms of an integral 关the element d⌫ z of Eq. 共35兲
共 J x H y ⫺J y H x 兲 ⫽ 共 JH 0 sin共 2 ␺ ⫹ ␣ 兲 ⫹J⌬H sin ␣ 兲 共41a兲 integrated over the volume of the inner cylinder兴, we see that
the contribution of each of the two terms ⌬H1 and ⌬H2 is
and independent of L as L→⬁, because of the rapid convergence
of the fields indicated by Eq. 共45兲. This equation suggests
共 xF y ⫺yF x 兲 ⫽⫺2J⌬H sin共 ␣ 兲 dV. 共41b兲
that the asymptotic form of the torque for long cylinders is of
The total torque ⌫ tot is found by integrating over the volume the form
of the inner cylinder: ⌫⬃constant⫹O 共 L ⫺1 兲 . 共47兲

⌫ tot共 ␣ 兲 ⫽ 冕冕冕 关 JH 0 sin共 2 ␺ ⫹ ␣ 兲

The order indicated is an upper bound. The symmetries con-
sidered here imply a stronger 共octopole兲 decrease of the field
than z ⫺2 , and the actual asymptotic form of ⌫ should be of
⫺J⌬H sin ␣ 兴 ␳ d ␳ d ␺ dz. 共42兲 order L ⫺3 .
After performing the angular integration, E. Scaling laws

⌫ tot⫽⫺2 ␲ J sin ␣ 冉冕 冕 冊
⌬H 共 ␳ ,z 兲 ␳ d ␳ dz , 共43兲
From the definition of energy in Eq. 共2兲, we see that as
the dimensions are scaled up, the energy must scale as
共length兲.3 We have just shown that, in the ideal case, the
i.e., torque is independent of the length of the cylinder, in the
limit of large length. The scaling law may therefore be writ-
⌫ tot⬀⫺sin ␣ . 共44兲 ten
Note that the result in Eq. 共43兲 is the combination of the two ⌫⬀R 3i 共48兲
terms in Eqs. 共41a兲 and 共41b兲, which contribute in a ratio
independent of L, for long, ideal Halbach cylinders. For seg-
⫺2:1. This is in agreement with the ratio between these con-
mented cylinders, there is a second term of order R 2i L. This
tributions found numerically in Fig. 8.
prediction is born out by numerical calculation.
This analysis demonstrates that, in the case of an ideal
cylinder of finite length, the torque varies as ⫺sin(␣) in ac-
VI. CONCLUSION
cordance with the symmetry arguments of Sec. V A, and in
agreement with experimental 共Sec. II兲 and numerical 共Sec. Halbach cylinders experience zero torque in uniform ap-
IV C兲 results. plied field. Thus, the origin of the real torque observed in
6422 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

nested systems of finite length, segmented Halbach cylinders flux from the bore, particularly at the ends where the torque
merited investigation. In Sec. V C, we saw that although is most significant. This reduction in applied field may fur-
each volume element experiences a large torque due to the ther serve to reduce the experimentally observed torque.
applied field H0 , these contributions cancel over the whole Thus the calculated value for torque acts as an upper bound
cylinder, due to symmetry. Therefore, the resultant torque is for real systems.
determined by the much smaller fluctuations in this field due
to finite length and segmentation. The total torque for cylin- ACKNOWLEDGMENTS
ders with this particular symmetry is much less than we
The authors are grateful to Magnetic Solutions Ltd. for
should expect from the magnitude of the torque on any one
providing the facilities necessary to carry out this work, with
segment.
particular thanks to Farid Bengrid and Laurent Clochard. The
The principal variation of the torque as the inner cylin-
work was partly supported by Forbairt and also by the U.S.
der is rotated through an angle ␣, while holding the outer
Navy, Contract No. ONR N00014-96-1-0767. Treasa Nı́
cylinder fixed, is sinusoidal. We have derived an analytical
Mhı́ocháin acknowledges helpful conversations with Jerome
theory of the inhomogeneities in the fields produced in the
Delamare and Orphée Cugat.
cylinder bores. This, together with symmetry arguments,
leads to the fact that the dominant sinusoidal variation in
APPENDIX: CALCULATION OF THE FIELD DUE TO A
torque is an end effect due to the finite lengths of the cylin-
FINITE LENGTH IDEAL HALBACH CYLINDER
ders.
The remaining variation in the torque is accounted for in The field due to a Halbach cylinder may be calculated
the segmented magnet torque theorem, where we use sym- from a magnetic potential ␸ m which satisfies the Poisson Eq.
metry to isolate the higher order terms, n⫽(kN⫾1), in the 共17兲. In the case of an ideal cylinder, the polarization varies
Fourier transform of the torque data 共where N is the number continuously according to Eq. 共22兲. If this equation is rewrit-
of segments in the cylinder, and k is an integer兲. This sym- ten in terms of conventional cylindrical coordinates ␳,
metry analysis says nothing about the size of the coefficients ␾ (⫽ ␲ /2⫺ ␺ ) and z, the polarization is of the form
in Eq. 共34兲. In particular, nothing is determined about spe-
cific dependencies of the coefficients on N or on the dimen- J共 ␾ 兲 ⫽J 关 sin共 2 ␾ 兲 x̂⫺cos共 2 ␾ 兲 ŷ 兴 . 共A1兲
sions of the cylinders. This may provide an interesting topic
for future study. However, the numerical calculations show
In this case, the volume magnetic charge density is
that all coefficients a k for k⭓1 are small compared with the
magnitude of a 0 , and decrease strongly with k, although this
2J sin ␾
may not necessarily be the case for alternative designs. ␳ m 共 r兲 ⫽⫺ⵜ–J⫽⫺ . 共A2兲
In the case of ideal cylinders, the torque is independent ␳
of the length L, in the limit of large L, and this leads to the
scaling law ⌫⬀R 3 , where R is a radial dimension of the This is nonzero over the volume V of the cylinder: R i ⬍ ␳
⬍R o , 0⬍ ␾ ⬍2 ␲ , and, in the case of finite length, ⫺L/2
nested cylinders. This implies that, for long ideal cylinders,
⬍z⬍L/2.
the torque is produced entirely in regions within a distance ⬃
The solution to the Poisson equation for the potential at
R from either end.
a point r is
Good agreement was found between experiment, nu-
merical calculation, and analytic theory, with respect to an-
gular variation of torque as the cylinders are rotated. The
calculated value for the magnitude of the torque was found
␸ m 共 r兲 ⫽
1
4␲␮0
冕V
dV ⬘
␳ m 共 r⬘ 兲
兩 r⫺r⬘ 兩
, 共A3兲
to be an upper bound on experimental values. Experimental
results show maximum values ranging from 6 to 13 Nm for where primed coordinates indicate source points, and the in-
different magnet assemblies of the same dimensions with an tegration is over the volume of the cylinder. The denomina-
average of approximately 8 Nm, while the numerical maxi- tor in the integrand may be expanded in terms of spherical
mum is 12.6 Nm. A number of factors may account for this harmonics:
difference, as follows.
⬁ ⫹l l
Our field calculation is based on the assumption of zero 1 1 r⬍
transverse susceptibility ␹⬜ , while in reality ␹⬜ ⬇0.1. As 兩 r⫺r⬘ 兩
⫽4 ␲
l⫽0
兺 m⫽⫺l
兺 共 2l⫹1 兲 ␮ 0 r ⬎
l⫹1
this will tend to align the magnetization direction along the
direction of the applied field, it may act to reduce the net * 共 ␪ ⬘ , ␾ ⬘ 兲 Y lm 共 ␪ , ␾ 兲 ,
⫻Y lm 共A4兲
torque experienced. Calculated fields are often as much as
10% greater than those measured in the bores of real cylin- where r ⬍ and r ⬎ are, respectively, the lesser and greater of r
ders. In part, this reflects nonrectangularity of the hysteresis and r ⬘ . This equation is expressed in spherical polar coordi-
loop. As torque is proportional to J2 , this may account for up nates, related to the cylindrical coordinates by
to 20% reduction in the torque observed compared to that
calculated. In addition, in the two cylinder system detailed in ␳ ⫽r sin ␪ , 共A5兲
Sec. II, steel ball bearings are attached to the cylinder ends to
allow them to rotate. This soft magnetic material draws some z⫽r cos ␪ ⫽ ␳ cot ␪ .
J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al. 6423

FIG. 14. Field at a point r in the cylinder bore is due to source points r⬘ in
the cylinder, some of which satisfy r⬎r⬘ , such as r⬘1 in the shaded regions,
and some of which satisfy r⬍r⬘ , such as r2⬘ in the unshaded regions.

FIG. 15. 共a兲 Field due to an ideal Halbach cylinder of finite length L may be
calculated by integrating over the volume of the cylinder; 共b兲 alternatively it
may be found by calculating, by integration, the field due to two semi-
Substitution of Eq. 共A4兲 into Eq. 共A3兲 gives the potential infinite cylinders, separated by a distance equal to the length of the finite

冉 冊
length cylinder, and subtracting this field from the field created by an infinite
⬁ ⫹l
b lm length cylinder.
␸ m 共 r兲 ⫽ 兺 兺
l⫽0 m⫽⫺l
a lm r l ⫹ Y 共 ␪,␾ 兲,
r l⫹1 lm
共A6兲

where
⬍␳⬘⬍Ro , L/2⬍z ⬘ ⬍⬁. The contribution of this shell to the
potential at a point r in the region ␳ ⬍R i , ⫺L/2⬍z⬍⫹L/2
a lm ⫽
1
共 2l⫹1 兲 ␮ 0
冕V
dV ⬘
␳ m 共 r⬘ 兲
r ⬘ l⫹1 lm
Y * 共 ␪ ⬘, ␾ ⬘ 兲, is, from Eqs. 共A2兲, 共A6兲, and 共A7兲
⬁ ⫹l
共A7兲
d ␸ m⫹ 共 r, ␪ , ␾ 兲 ⫽ 兺 兺 da lm r l Y lm 共 ␪ , ␾ 兲 , 共A9兲

冕
l⫽0 m⫽⫺l
1
b lm ⫽ dV ⬘ ␳ m 共 r⬘ 兲 r ⬘ l Y lm
* 共 ␪ ⬘, ␾ ⬘ 兲.
共 2l⫹1 兲 ␮ 0 V
where

For source points such that r ⬘ ⬎r, the first term in Eq. 共A6兲
contributes to the potential, while the second term is the
contribution from regions where r ⬘ ⬍r.
da lm ⫽
⫺2J
共 2l⫹1 兲 ␮ 0
冕 ⬘冕
L/2
⬁
dz
2␲

0
d ␾ ⬘␳ ⬘ d ␳ ⬘
sin ␾ ⬘ 1
␳ ⬘ r ⬘ l⫹1
In the case of an infinitely long ideal cylinder, the po- * 共 ␪ ⬘, ␾ ⬘ 兲.
⫻Y lm 共A10兲
tential, giving rise to the field of Eq. 共1兲, is
Using Eq. 共A5兲 to write z ⬘ and r ⬘ in terms of ␪ ⬘ and ␳ ⬘ , this

冉 冊
becomes
J Ro
␸ m⬁ 共 r, ␪ , ␾ 兲 ⫽⫺ r sin ␪ sin ␾ ln 共A8兲
冕 冕
.
␮0 Ri ⫺2J ␪1 2␲
da lm ⫽ d␪⬘ d␾⬘
共 2l⫹1 兲 ␮ 0 0 0
We wish to calculate the potential within the bore of an ideal
Halbach cylinder of finite length L. d ␳ ⬘ l⫺1
⫻sin ␾ ⬘ sin ␪ ⬘ Y lm
* 共 ␪ ⬘, ␾ ⬘ 兲, 共A11兲
At an arbitrary point r in the bore there will, in general, ␳ ⬘l
be contributions to the potential from points in the magnetic
where, from Fig. 15, the upper limit of the integration over ␪
material with r ⬘ ⬎r, such as r 2⬘ in the unshaded region of
satisfies
Fig. 14, and also from points with r ⬘ ⬍r, such as r ⬘1 in the
shaded region. To simplify the calculation and avoid the ne-
cessity of evaluating contributions from points with r ⬘ ⬍r, L
cot ␪ 1 ⫽ . 共A12兲
we proceed as follows. 2␳⬘
We calculate the potential due to two semi-infinite ideal
cylinders, one extending from z⫽L/2 to ⬁, and the other Integrating over ␾ ⬘ and substituting into Eq. 共A9兲, we obtain
from z⫽⫺L/2 to ⫺⬁, Fig. 15. From this we find the poten-
⬁
tial due to these cylinders at points in the bore between z ⫺J 共 l⫺1 兲 ! 1
⫽⫺L/2 and z⫽⫹L/2, and then subtract the result from the d ␸ m⫹ 共 r, ␪ , ␾ 兲 ⫽ 兺 P 共 cos ␪ 兲 sin ␾ r l
␮ 0 l⫽1 共 l⫹1 兲 ! l
potential for a cylinder of infinite length, Eq. 共A8兲. The field
due to a cylinder extending from z⫽⫺L/2 to z⫽L/2 at a
point within its bore is then found from the gradient of this
potential. All points within the material of the two semi-
⫻ 冉 冕 d␳⬘
␳ ⬘l 0
␪1
冊
d ␪ ⬘ sinl⫺1 ␪ ⬘ P 1l 共 cos ␪ ⬘ 兲 ,

infinite cylinders satisfy r ⬘ ⬎r, where r refers to any point 共A13兲

within the bore of the finite length cylinder, ⫺L/2⬍z where l (cos ␪⬘)
Pm is an associated Legendre function.
⬍⫹L/2, ␳ ⬍R i . Similarly, an expression is obtained for the potential
Consider a cylindrical shell of thickness d ␳ ⬘ at a fixed d ␸ m⫺ (r, ␪ , ␾ ) due to the cylindrical shell extending from z
radius ␳ ⬘ within the semi-infinite cylinder defined by R i ⫽⫺L/2 to ⫺⬁. The symmetry properties of the integrand
6424 J. Appl. Phys., Vol. 86, No. 11, 1 December 1999 Nı́ Mhı́ocháin et al.

indicate that each term of the summation in d ␸ m⫹ is equal to H x 共 r, ␪ , ␾ 兲 ⫽⌬H 共 r, ␪ 兲 sin共 2 ␾ 兲 ,

the corresponding term in d ␸ m⫺ apart from a factor
(⫺1) l⫹1 . Thus terms with even l cancel. The total potential H y 共 r, ␪ , ␾ 兲 ⫽H 0 共 r, ␪ 兲 ⫺⌬H 共 r, ␪ 兲 cos共 2 ␾ 兲 , 共A18兲
due to the two semi-infinite cylinders is obtained by integrat- H z 共 r, ␪ , ␾ 兲 ⫽⌬H z 共 r, ␪ 兲 sin ␾ ,
ing over ␳ ⬘ :
where
⬁

冉
⫺2J 共 l⫺1 兲 ! 1
兺
⬁
␸ m⫹ ⫹ ␸ m⫺ ⫽ P 共 cos ␪ 兲 sin ␾ r l ⫺J 共 l⫺1 兲 ! l⫺1
␮ 0 l odd 共 l⫹1 兲 ! l ⌬H 共 r, ␪ 兲 ⫽ 兺
␮ 0 l odd 共 l⫹1 兲 !
r I l 关共 l⫹1 兲 sin2 ␪ ⫺2 兴

⫻ 冉冕 冕 Ri
Ro d␳⬘
␳ ⬘l 0
␪1
d ␪ ⬘ sinl⫺1 ␪ ⬘ P 1l 共 cos ␪ ⬘ 兲 .冊 ⫻
d Pl
d 共 cos ␪ 兲
⫹l 共 l⫹1 兲 cos ␪ P l , 冊
冉 冊
共A14兲
J Ro
Notice that the upper limit of the integration over ␪ ⬘ is a H 0 共 r, ␪ 兲 ⫽⫺ ln ⫹⌬H 共 r, ␪ 兲
␮0 Ri
function of ␳ ⬘ , Eq. 共A12兲. The integrand of the angular in-
⬁
tegration may be rewritten using the property of Legendre 2J 共 l⫺1 兲 ! dP
functions ⫺
␮0 兺
l odd 共 l⫹1 兲 !
r l⫺1 I l
d 共 cos
l
␪兲
,

冉
2 l l! l ⬁
sinl ␪ ⬘ ⫽ P 共 cos ␪ ⬘ 兲 . 共A15兲 ⫺2J 共 l⫺1 兲 ! l⫺1
共 2l 兲 ! l ⌬H z 共 r, ␪ 兲 ⫽ 兺
␮ 0 l odd 共 l⫹1 兲 !
r I l 共 l⫹1 兲
Finally, the potential due to a cylinder of length L is
␸ mL ⫽ ␸ ⬁ ⫺ 共 ␸ m⫹ ⫹ ␸ m⫺ 兲 ⫻sin ␪ cos ␪
d Pl
d 共 cos ␪ 兲
⫺l 共 l⫹1 兲 sin ␪ P l .冊
⫽
⫺J
␮0
r sin ␪ sin ␾ ln
Ro
Ri 冉 冊 共A19兲

⬁
1
H. A. Leupold and E. Potenziani II, J. Appl. Phys. 64, 5994 共1988兲.
2J 共 l⫺1 兲 !
兺
2
H. A. Leupold, in Rare Earth Iron Permanent Magnets, edited by J. M. D.
⫹ I P 1 共 cos ␪ 兲 sin ␾ r l , 共A16兲
␮0 l odd 共 l⫹1 兲 ! l l Coey 共Oxford University Press, Oxford, 1996兲, Chap. 8.
3
K. Halbach, Nucl. Instrum. Methods 169, 1 共1980兲.
4
where Permanent magnet flux sources are available from Magnetic Solutions
Ltd., IDA Centre, Unit 13, Pearse St., Dublin 2, Ireland.

冕 d␳⬘
冕 ␪1 2 l l! P ll 共 cos ␪ ⬘ 兲 P 1l 共 cos ␪ ⬘ 兲 T. Kenjo, Electric Motors and Their Controls: An Introduction 共Oxford
5
Ro
I l⫽ d␪⬘ . University Press, Oxford 1991兲.
Ri ␳ ⬘l 0 共 2l 兲 ! sin ␪ ⬘ 6
A. Aharoni, Introduction to the Theory of Ferromagnetism 共Oxford Uni-
共A17兲 versity Press, Oxford, 1996兲.
7
J. P. Yonnet, in Rare Earth Iron Permanent Magnets, edited by J. M. D.
The magnetic field due to the finite length cylinder, H Coey 共Oxford University Press, Oxford, 1996兲, Chap. 9.
⫽⫺ⵜ ␸ mL , has components 8
G. Akoun and J. P. Yonnet, IEEE Trans. Magn. 20, 1962 共1984兲.