Published in Journal of Applied Physics, Vol.

104, 013910, 2008

DOI: 10.1063/1.2952537

Optimization and improvement of Halbach cylinder

design
R. Bjørk, C. R. H. Bahl, A. Smith and N. Pryds

Abstract
In this paper we describe the results of a parameter survey of a 16 segmented Halbach cylinder in three
dimensions in which the parameters internal radius, rin , external radius, rex , and length, L, have been varied.
arXiv:1409.3859v1 [physics.ins-det] 8 Sep 2014

Optimal values of rex and L were found for a Halbach cylinder with the least possible volume of magnets with a
given mean flux density in the cylinder bore. The volume of the cylinder bore could also be significantly increase
by only slightly increasing the volume of the magnets, for a fixed mean flux density. Placing additional blocks
of magnets on the end faces of the Halbach cylinder also improved the mean flux density in the cylinder bore,
especially so for short Halbach cylinders with large rex . Moreover magnetic cooling as an application for Halbach
cylinders was considered. A magnetic cooling quality parameter, Λcool , was introduced and results showed that
this parameter was optimal for long Halbach cylinders with small rex . Using the previously mentioned additional
blocks of magnets can improve the parameter by as much as 15% as well as improve the homogeneity of the
field in the cylinder bore.
Department of Energy Conversion and Storage, Technical University of Denmark - DTU, Frederiksborgvej 399, DK-4000 Roskilde, Denmark
*Corresponding author: rabj@dtu.dk

External radius, rex

1. Introduction
Configurations of permanent magnets that produce a strong
Length, L
homogeneous field in a confined region of space and a very
weak field elsewhere are useful in many applications such as
particle accelerators (1), nuclear magnetic resonance (NMR)
apparatus (2) or magnetic cooling applications (3).
The design known as a Halbach cylinder is especially
good at producing this type of magnetic field. A Halbach
cylinder is a long cylinder made of a magnetic material with a
bore along the cylinder symmetry axis. The Halbach cylinder
can be characterized by three parameters; the internal and
external radii, rin and rex respectively, and the length, L. The Internal radius, rin
magnetic material around the bore is magnetized such that the Figure 1. A sketch of a Halbach cylinder showing the
direction of magnetization at any point is at an angle internal radius, rin , external radius, rex , and length, L. Also
shown as arrows is the direction of the remanent
η = 2θ (1) magnetization of the magnetic material.
from the vertical axis (4; 5). This arrangement means that a
uniform field will be created across the bore in the vertical
direction without creating, in the ideal case, any stray field effect of the finite length of the Halbach cylinder has been
outside the cylinder. Fig. 1 shows a drawing of a Halbach studied. In one example (7) the reduction in flux density due
cylinder. to a finite length Halbach cylinder was investigated, but the
It is well known that the flux density inside the bore of an field was only calculated for a single fixed length. An analyti-
infinitely long Halbach cylinder is (5) cal formula for the magnetic flux of a Halbach cylinder of any
  given length was derived, however this formula is extremely
rex
B = Br ln , (2) complicated, making it impractical for direct application. In
rin another publication (8) the effects of a finite length Halbach
where Br is the remanent flux density of the magnetic material. cylinder were also explored, as well as the effect of dividing
Halbach cylinders have previously been investigated in the Halbach cylinder into a number of segments, each with its
detail in two dimensions, but there exist only a few investi- own direction of magnetization. However, both investigations
gations of Halbach cylinders in three dimensions, where the were only performed for one specific Halbach cylinder of a
 Optimization and improvement of Halbach cylinder design — 2/10

fixed length.
100 %
In this paper the three dimensional Halbach cylinder will

Derivation from infinite segment cylinder

be investigated in greater detail, and the flux density will be 95 %
computed for a multitude of different configurations and not
90 %
only a single specific case.
First, the effect of dividing the Halbach cylinder into seg- 85 %
ments each with their own direction of magnetization will
80 %
be investigated. To measure only the effect of segmentation
the calculations are performed in two dimensions, so that any 75 %
effects from a finite length Halbach cylinder are avoided. We
70 %
then assume that the effect of segmentation in two dimensions Numerical calculation, µr=1
is similar in three dimensions. 65 % Numerical calculation , µr=1.05
Thereafter the Halbach cylinder will be investigated in sin(2π/n)/(2π/n)
three dimensions, with focus on how to build a Halbach cylin- 60 %
0 4 6 8 12 16 24 32
der with a certain mean flux density using a minimum of Number of segments
magnetic material, i.e. find the configuration of rin , rex and Figure 2. Dividing a Halbach cylinder into n segments makes
L that generates the strongest flux density for the minimum the flux density differ from that of a perfect Halbach cylinder.
amount of magnetic material. The deviation is given by Eq. 4 for ”perfect magnets”,
Finally it will be investigated if the magnetic flux density µ r = 1, while magnets with µ r = 1.05 deviate more from the
can be improved by placing additional blocks of permanent ideal Halbach cylinder.
magnets on the end faces of the Halbach cylinder.
The results of this investigation of Halbach cylinder design
are useful in many different fields, e.g. magnetic cooling (3) often made up of segments, each of which has a direction of
or tabletop NMR (9). These applications typically require a magnetization equal to the direction of magnetization of a
flux density of around 1 − 3 T, and this is also the range of continuous Halbach cylinder at the center of the segment.
flux density that we will concern ourselves with in this paper. A Halbach cylinder consisting of n such segments will
All numerical work in this paper was done using the com- have its flux density reduced to (5)
mercially available finite element multiphysics program, Com-
sin(2π/n)
sol Multiphysics(10). The Comsol Multiphysics code has B(n) = B(∞) , (4)
previously been validated through a number of NAFEMS (Na- 2π/n
tional Agency for Finite Element Methods and Standards)
where B(∞) is the flux density given by Eq. 2, i.e. with a
benchmark studies (11).
continuous magnetization.
The equation solved in the following simulations is the
We have analyzed the consequence of this segmentation
magnetic vector potential equation,
of the Halbach cylinder by computing the mean value of
∇ × (µ −1 −1 the magnetic flux density inside the Halbach cylinder bore
0 µ r (∇ × A − Br )) = 0, (3)
for a Halbach cylinder consisting of 4, 6, 8, 12, 16, 24 and
where A is the magnetic vector potential, Br is the remanent 32 segments. The calculations were performed both for a
flux density, µ 0 is the permeability of free space and µ r is the Halbach cylinder consisting of “perfect” magnets, i.e. with
relative permeability assumed to be isotropic. a relative permeability, µ r , of 1, and magnets where actual
The solver used to solve this equation on the simulation material properties were taken into account by increasing
mesh is Pardiso which is a parallel sparse direct linear solver µ r to 1.05. The magnetic material was assumed to have a
(12; 13). remanent flux density of 1.4 T, equal to standard grade N48
Boundary conditions are chosen such that the boundaries Neodymium-Iron-Boron (NdFeB) magnets (6). This value for
of the computational volume, which is many times larger than the remanent flux density will be used in the remainder of this
the Halbach cylinder, are magnetically insulating, while all paper.
other (internal) boundaries are continuous. The results of the computations together with Eq. 4 are
shown in Fig. 2. Here it is seen that Eq. 4 describes the
numerical data with µr = 1 extremely well. It is also seen that
2. Segmented Halbach cylinder
choosing a small number of segments severely limits the flux
An infinitely long Halbach cylinder is equivalent to a two density.
dimensional situation so it fulfills Eq. 2, if the direction Based on the results shown in Fig. 2 we choose, in the
of magnetization varies continuously through the magnetic calculations and computations presented in the following sec-
material as prescribed by Eq. 1. This continuous variation tions, to use a 16 segmented Halbach cylinder with µ r = 1.05.
of the direction of magnetization is often not attainable in This configuration obtains 95% of the flux density of a perfect
real-world assemblies, and therefore the Halbach cylinder is Halbach cylinder and is realizable in real-world assemblies.
 Optimization and improvement of Halbach cylinder design — 3/10

Table 1. The variation of the Halbach parameters. In total 4

Contour of <B> for r = 10 mm
in
x 10
there are 90 × 90 × 3 different configurations.

2.6
6

2.4
2

2.7

2.9
From To Stepsize
[mm] [mm] [mm]

2.8
5

2.2
Volume of bore [mm3]
L 41 130 1
rex 22 200 2 4
rin 10 30 10

2.6
2.4
2
3

2.2
Contour of <B> for rin = 20 mm
2
200
1.2

2
1.5

1.9

2
180
1.7

1
1.8
1.4

160 2 4 6 8 10
1.6
1

3 6
Volume of magnets [mm ] x 10
140 Figure 4. Contours of the mean flux density as a function of
1.
9 the volume of magnets used and the volume of the cylinder
L [mm]

120
1.2

1.8
1.5

1.

bore for rin = 10 mm. It is seen that the volume of the bore
7

100 1.
6 can be significantly increased by slightly increasing the
1.4
1

80 1.7 volume of the magnets. Note that the range is not the same on
1.5 1.6
60 1.2 1.4
the two axes. A look-up table is necessary such that each data
1.5
1.4 point (these are not shown) is uniquely tied to a specific
40 1.2
1 1.2 Halbach cylinder, i.e. a given rin , rex and L.
0.8 1 1
50 60 70 80 90 100 110 120 130
rex [mm]
5
Contour of <B> for r = 20 mm
in
Figure 3. Contours of the mean flux density for the Halbach 2.5
x 10
1
cylinders with rin = 20 mm. Each contour is labeled by its

1.8
1.2

1.7

2.05
1.6
mean flux density. As is expected the maximum flux density
1.5
1.4
is obtained by maximizing both rex and L. 2

2
Volume of bore [mm3]

1.9
Having determined the configuration to be used in the 1.5
1.8
1 1

1.7
1.2

following simulations we now proceed to investigate if there

1.6
1.5
1.4

exist optimal dimensions for a Halbach cylinder design. For

1
this three dimensional simulations must be used, in order to
study how the loss of flux density through the ends of the
cylinder bore varies with rex and L.
1

0.5 1.2

2 4 6 8 10
3. Halbach cylinder 3D study 3
Volume of magnets [mm ] 6
x 10
A parameter study of Halbach cylinder configurations have Figure 5. Contours of the mean flux density as a function of
been performed by varying the parameters L, rex and rin as the volume of magnets used and the volume of the cylinder
given in Table 1. In each of the 90 × 90 × 3 configurations bore for rin = 20 mm. The conclusion of Fig. 4 applies here
the mean flux density of the magnetic field inside the cylinder as well.
bore have been computed. The results are shown as a contour
plot of the mean flux density as a function of L and rex in Fig.
3 for rin = 20 mm. the external radius does not affect the volume of the bore.
Fig. 3 shows that the configuration producing the strongest Consequently a better way of characterizing each Halbach
mean flux density is the configuration with the largest rex and cylinder configuration is by the volume of its magnets and
L. This is in agreement with Eq. 2 and the fact that for a the volume of the bore, and then calculating contour plots
long Halbach cylinder the loss of flux through the ends of the with lines of equal mean flux density. These are shown in
cylinder will be relatively smaller than for a short cylinder. Figs. 4-6 for the three different values of rin . On Figs. 4-6
It is not sufficient to characterize a design only by the the volume of the bore scales directly with the length of the
value of the mean flux density. It should be considered that Halbach cylinder because the internal radius is kept constant
increasing the length of the Halbach cylinder increases the in each figure.
volume of the bore, thus allowing a larger sample to be placed Looking at, e.g., Fig. 5 it can be seen that for a mean flux
inside the Halbach cylinder bore. On the other hand increasing density of 1.6 T a Halbach cylinder can be constructed with
 Optimization and improvement of Halbach cylinder design — 4/10

5
Contour of <B> for r = 30 mm 6 r = 20 mm
in in
x 10 x 10
12
5.5

0.6
0.8

1.3

1.4

1.6
1.2
5

1.5
10

Volume of magnets [mm3]

4.5

1.1
Volume of bore [mm3]

4 8
3.5

1.4
1.3
1
0.6
0.8

1.2

6
3
2.5
1.1

4
2
1.5 2
1
0.6
0.8

1
0
2 4 6 8 10 0.5 1 1.5 2 2.5
3 6 <B> [T]
Volume of magnets [mm ] x 10
Figure 6. Contours of the mean flux density as a function of Figure 7. The total volume of the magnetic material as a
the volume of magnets used and the volume of the cylinder function of the mean flux density for rin = 20 mm. The data
bore for rin = 30 mm. The conclusion of Fig. 4 applies here points somewhat resemble a feather, and so this plot will be
as well. referred to as the “feather plot”. The data have been produced
in series where rex has been fixed and L has been varied. Two
of these data series have been highlighted in black and
a ∼50% increase in magnetic material but a ∼250% larger starting from the left both series can be seen to first approach
volume of the bore compared to the design using the least the rightwards edge of the “feather” and then leave it again.
amount of magnetic material. Similar plots exist for rin = 10 mm and 30 mm.
It is possible to attain this substantial increase in the vol-
ume of the bore because the latter configuration is a very long
Halbach cylinder with a small external radius, while the con- 140 140
figuration with the smallest volume of the magnets is a short
120 120
Halbach cylinder with a large external radius. In these two
configurations the shape of the bore is different, but the mean 100 100
flux density is the same. r in = 30 mm

Length [mm]
In Fig. 7 the total volume of the magnetic material is
rex [mm]

80 80
shown as a function of the mean flux density in the bore for r in = 10 mm
rin = 20 mm. In this plot there are 90 × 90 data points. Two 60 60
rin = 10 mm
data series where rex has been fixed and L has been varied are r in = 20 mm rin = 20 mm 40
40
highlighted on the plot (one could also have chosen to vary rin = 30 mm
rex and kept L fixed. The curves look the same). Here one can 20 r 20
ex
see that as L is increased the mean flux density is increased as L
well. At some point each data series becomes the rightwards 0 0
0.5 1 1.5 2 2.5 3
edge of the “feather”, and then the increase in the volume of <B> [T]
the magnets with mean flux density becomes too steep and the Figure 8. The optimal rex and L as functions of the mean flux
data points move upwards, leaving the edge of the “feather”. density in the cylinder bore for Halbach cylinders with rin =
The reason for this behavior is that the data series start 10, 20 and 30 mm. Building a Halbach cylinder with
with a short Halbach cylinder. This configuration loses a lot dimensions different from the dimensions given here means
of flux through the ends of the cylinder and so as the length that more magnetic material is used than need be, if one is
is increased the average flux density increases quite rapidly. only interested in obtaining the maximum flux density
When a certain length of the Halbach cylinder is reached there possible and does not care about the size of the cylinder bore.
is not as much to be gained by increasing the length of the
cylinder further and so the average flux density only increases
slowly as the volume of the magnets is increased. In Fig. 8 the values of rex and L are plotted as functions
As this is the case for all data series where the length of the mean flux density for the optimal points. Thus one
of the Halbach cylinder is gradually increased it is possible can directly use this figure to find the external radius and
to characterize the rightwards edge of the “feather” as the length for the Halbach cylinder with the minimum volume of
optimal configuration, i.e. the configuration with the smallest magnets at a given mean flux density. Straight lines have been
volume of the magnets at a given mean flux density. fitted to the data.
 Optimization and improvement of Halbach cylinder design — 5/10

The conclusion of this parameter investigation is twofold. Front view Side view
First, it can be concluded from Figs. 4-6 that it is possible, at a D
constant mean flux density, to increase the volume of the bore H 1 3
significantly by only increasing the volume of the magnets 1
slightly for a Halbach cylinder with a fixed rin . Secondly, the
length and external radius of the minimum magnetic material
Halbach cylinder at a given mean flux density was found and
can be read off directly from Fig. 8. Thus if one wishes 2 4
to build a Halbach cylinder with a given mean flux density 2
and the limiting factor is the price of magnetic material, one
should always choose the optimal configuration from this
figure.
Figure 9. This figure shows the height, H, angular span, φ ,
A few remarks on the precision of the simulations are in
order. With the chosen boundary conditions, i.e. a magnet- and depth, D, of the additional round blocks, colored in gray.
ically insulating computational volume, it is important that The blocks are always symmetrically placed. The black
the computational volume is large enough that the insulating arrows shows the direction of magnetization. The additional
boundaries do not effect the calculations. Also, as the solution blocks diagonally opposite each other have the same
method used is a finite element method, the mesh applied to direction of magnetization.
the geometry must be as detailed as needed for the desired
precision. The resolution of the mesh used for the simulations
direction of magnetization, at the end faces of the cylinder,
presented in this paper are chosen such that the results have a
in such a way that they do not block access to the cylinder
high degree of precision. To give an example, the mean flux
bore but still increase the flux density in the cylinder bore and
density of the cylinder bore was calculated at different preci-
ensure a low flux density outside of the Halbach cylinder.
sions for a random Halbach cylinder. The Halbach cylinder
In this section we investigate what specific design and
chosen had rin = 20 mm, rex = 102 mm and L = 70 mm, and
placement is optimal for these additional magnets. We also
a mean flux density of 1.54 T. This result, calculated using the
discuss whether it is better to use the additional magnets or
precision used throughout this paper, deviated by only 1.13%
if one might as well use the additional magnetic material for
from a simulation using 173% more mesh elements all in all,
building a larger Halbach cylinder.
and 1845% more mesh elements in the cylinder bore. The
influence of the size of the computational volume on the mean To maximize the amount of magnetic material capable of
flux density in the bore has also been tested for a number of being used in the additional blocks we use a design of the
different values of rin , rex and L and found to be less than 1%. additional blocks that follows the curvature of the cylinder
Thus we conclude that at least the relative precision of the bore, i.e. a circular design as can be seen in Fig. 9. In
numerical experiments is satisfactory. total four additional blocks are used, placed symmetrically
Although the above results are useful in choosing the around the Halbach cylinder symmetry axis. In this design an
optimal Halbach cylinder design, alternative methods for im- additional block is characterized by three parameters, namely
proving the design of a Halbach cylinder exists. The problem the angular extent of a block, φ , the block’s depth, D, and
with especially the short Halbach cylinders is that they lose a height, H. The direction of magnetization of the individual
relatively large amount of flux through the ends of the cylinder. additional block is perpendicular to the Halbach cylinder end
This is the reason that their flux density is not well described face. Furthermore the blocks diagonally opposite have the
by Eq. 2. In the next section it is investigated if it is possible same direction of magnetization.
to limit the amount of escaping flux through the ends of the The height, H, angular span, φ , and depth, D, of the
cylinder by appending blocks of permanent magnets to the additional blocks are varied to find the optimal configuration
end faces of the Halbach cylinder and thus in this way improve for several different Halbach cylinders. Calculating the flux
the design. density for each of the original 90 × 90 × 3 Halbach cylinders
with additional magnets is a too time consuming task, and
thus the calculations were only done on a few carefully chosen
4. Improving the Halbach cylinder design Halbach cylinder designs. These are given in Table 2.
The main loss of flux from the bore of the Halbach cylinder The results of the simulations are shown in Fig. 10. Here
is through the ends of the cylinder bore. It has previously the mean flux density in the bore as a function of the total
been shown (15) that by “covering” the ends of the Halbach volume of the magnetic material used in the simulated design
cylinder with magnetic blocks in the shape of an equipotential is shown. The figure shows both the Halbach cylinders with-
surface, all of the flux can be confined inside the Halbach out any additional blocks, and the simulations of the Halbach
cylinder. However this also blocks access to the cylinder bore. cylinders with additional blocks.
Instead we propose that some of the escaping flux may be One can see from the figure that placing additional blocks
confined by placing additional magnets, of a given size and on a relatively short Halbach cylinder will improve the mean
 Optimization and improvement of Halbach cylinder design — 6/10

Table 2. The additional magnets were placed on the four namely magnetic cooling. For this type of application the
different Halbach cylinders given in column 5. The Halbach cylinder must be designed such that it has a high flux
parameters of the additional blocks were varied as given in density in a large volume and with a minimum of magnetic
this table resulting in 10 × 10 × 8 different configurations of material.
the additional blocks for each Halbach cylinder. The magnetic cooling process relies on a magnetocaloric
From To Step Halbach dim- material. The temperature of such a material is increased upon
size ensions [mm] the application of a magnetic field and decreased again upon
◦
φ [ ] 10 80 10 L = 100 the removal of the magnetic field. A large number of different
H [mm] 100 30 10 rex = 100 materials have been suggested as the active component of a
D [mm] 50 5 5 rin = 20 magnetic refrigeration machine (16).
φ [◦ ] 10 80 10 L = 92 From experimental studies it is known that the adiabatic
H [mm] 130 25 15 rex = 130 temperature change of Gadolinium, the “benchmark” magne-
D [mm] 46 4.6 4.6 rin = 20 tocaloric material at room temperature, has a magnetocaloric
φ [◦ ] 10 80 10 L = 200 effect that scales with the flux density of the magnetic field
H [mm] 60 25 5 rex = 60 (18) to the power of 0.7 . This is in good accordance with the
D [mm] 100 10 10 rin = 20 power of 23 predicted by mean field theory (17).
φ [◦ ] 10 80 10 L = 50 However, it is not only the flux density inside the cylinder
H [mm] 130 25 15 rex = 130 bore that is of importance to the magnetocaloric effect. The
D [mm] 25 2.5 2.25 rin = 20 volume outside the cylinder bore where the magnetocaloric
material is placed when it is moved to the “out of field” po-
sition is also important. In order to maximise the magne-
6
x 10
r in = 20 mm tocaloric effect the flux density in this region must be as low
12 as possible. It can of course be argued that one could simply
Halbach cylinders
Blocks, L= 50 mm, rex= 130 mm move the magnetocaloric material further away than right out-
10 side the end of the cylinder bore, but this would increase the
Blocks, L= 92 mm, rex= 130 mm
Volume of magnets [mm3]

Blocks, L= 100 mm, rex= 100 mm physical size of the magnetic refrigeration machine. Finally,
8 it is important that the cylinder bore has as large a volume as
Blocks, L= 200 mm, rex= 60 mm
possible and that the volume of the magnets be as small as
6 possible.
Taking all this into account we propose to characterize a
4 configuration of magnets for use in magnetic cooling applica-
tions by the parameter
2
Vfield
Λcool ≡ (<B0.7> − <B0.7
out>) Pfield , (5)
Vmag
0
0.5 1 1.5 2 2.5
<B> [T] where Vmag is the volume of the magnets, Vfield is the volume
Figure 10. Placing additional blocks on a specific Halbach with a high flux density, i.e. the volume of the cylinder bore,
cylinder improves the mean flux density in the cylinder bore Pfield is the fraction of the total volume of the cylinder bore
but the improvement depends on L and rex of the Halbach and the volume outside the cylinder bore that is filled with
cylinder. The legend shows which Halbach cylinder the magnetocaloric material, <B0.7> is the volume average of the
additional blocks was placed on. Without the additional flux density in the high flux volume, i.e. the cylinder bore,
blocks the figure is identical to Fig. 7. to the power of 0.7 and <B0.7 out> is the volume average of the
flux density to the power of 0.7 in the region shown in Fig.
11, i.e. the volume just outside the cylinder bore where the
flux density in the cylinder bore significantly. magnetocaloric material is placed when it is moved out of the
magnetic field. It has the same size and shape as the cylinder
5. Halbach cylinders for use in magnetic bore.
The magnetic cooling parameter is shown for the Halbach
cooling
cylinders without additional blocks and with rin = 20 mm in
We have shown that using additional blocks of magnets on Fig. 12 for Pfield = 0.5 i.e. we assume that the total volume
the sides of the Halbach cylinder can increase the mean flux is only half filled with magnetocaloric material at any given
density in the cylinder bore. However in some cases the moment. Here we see that the optimal design is the Halbach
additional magnetic material might as well be used to enlarge cylinder with the largest L and smallest rex . Note that this
the Halbach cylinders external radius and in this way also design is not the overall optimal design, as it lies on the edge of
increase the flux density. We will consider this more closely in the parameter space, i.e. simulations have not been conducted
the context of one particular application for Halbach cylinders, with a larger L and smaller rex .
 Optimization and improvement of Halbach cylinder design — 7/10

Additional magnets Additional magnets Contours of Λcool for r in = 20 mm

200

0.12

0.08

0.06

0.05

0.04
180

0.03
0.1

0.02
160

140

Length [mm]
120

0.12

0.08

0.06
0.05

0.04

3
100

0.0

2
0.1

0.0
80

60

8
6

04
05
2
03

0.0
0.0

0.0
40 0.

0.
0.01

0.
40 60 80 100 120
Outside volume External radius [mm]
Figure 12. A contour plot showing the magnetic cooling
parameter, Λcool , defined in Eq. 5 for the Halbach cylinders
Cylinder bore without additional blocks and with rin = 20 mm. The
Figure 11. A side view of the Halbach cylinder with optimum design is the longest and thinnest Halbach cylinder.
additional blocks. The volume inside which <B0.7 out> is
calculated is shown. The volume is identical in shape to the
cylinder bore, also shown, and is located directly outside the
end of the cylinder bore. Also shown are the additional Halbach cylinders without blocks
Blocks on L = 50 mm, rex = 130 mm
blocks of magnets. The edges of the 16 segments that make
Blocks on L = 92 mm, rex = 130 mm
up the Halbach cylinder can also be seen on the figure.
Blocks on L = 100 mm, rex = 100 mm
1.3
Λcool / (Λcool ) org

Blocks on L = 200 mm, rex = 60 mm

1.2
There are several reasons that the long, thin Halbach cylin- 1.1
der has the largest Λcool . The primary cause is due to the fact 1
that the loss of flux through the ends of the cylinder bore is 0.9
greatly reduced in the long Halbach cylinder. Another effect 0.8
is due to its long length, the volume inside which <B0.7 out> is
0.7
calculated is also long (as previously mentioned the shape 0.6
of the cylinder bore and this volume are identical), and thus 0.5
0.4
the end furthest from the cylinder bore will only experience 50 100 150 200
a very small field, thus lowering <B0.7
out>. For practical appli-
Length of Halbach cylinder without blocks [mm]
cations one would choose to optimize Λcool under a criterium Figure 13. The relative improvement of the magnetic cooling
of a minimum flux density in the cylinder bore, i.e. find the parameter compared to the original Halbach cylinder without
Halbach cylinder with the maximum Λcool that at the same any additional blocks, for the designs listed in Table 2. The
time has a minimum flux density of e.g. 1 T in the cylinder design most improved is the short Halbach cylinder with a
bore. large external radius, i.e. L = 50 mm, rex = 130 mm. For
We are also interested in knowing what effect the addi- each Halbach cylinder the best configuration of the additional
tional blocks of magnets have on the magnetic cooling param- blocks have been marked by a star. The black crosses in the
eter. In Fig. 10 we saw that the additional blocks increase the figure are Halbach cylinder with the same rin and L as the
flux density in the cylinder bore, but this might not mean that Halbach cylinder with blocks, but with a bigger rex and no
Λcool is increased as well, as additional magnetic material is blocks. The black crosses are displaced by ±6 mm on the
also used. x-axis, to make the comparison with the Halbach cylinders
In Fig. 13 Λcool is shown for the different Halbach cylin- with additional blocks possible.
ders with additional blocks, i.e. the ones given in Table 2.
Here we can see that some configurations of the additional
blocks do increase Λcool by as much as ∼15%. Shown in the additional blocks or on enlarging the external radius of the
figure are also Halbach cylinders with no additional blocks Halbach cylinder. As one can see from the figure in three of
that have the same rin and L as the Halbach cylinders with the cases it is better to spend the additional magnetic material
additional blocks but where rex has been expanded by up to on the additional blocks.
30 mm. These are shown such that it can be estimated if it The design that is the most improved by the additional
is better to spend any additional magnetic material on the blocks is a short Halbach cylinder with a large external radius,
 Optimization and improvement of Halbach cylinder design — 8/10

Table 3. The optimal configuration of the additional blocks,

i.e. the configurations that maximize the magnetic cooling 1.5
Halbach cylinders without blocks
parameter. (Λcool )org refers to the “original” Halbach cylinder 1.4 Blocks on L = 50 mm, rex = 130 mm
without any additional blocks. Blocks on L = 92 mm, r = 130 mm
1.3 ex
Halbach φ H D Λcool Blocks on L = 100 mm, rex = 100 mm
dimensions (Λcool )org 1.2
Blocks on L = 200 mm, rex = 60 mm
[mm] [◦ ] [mm] [mm] 1.1

η/η org
rex = 130, L = 50 80 47.5 22.5 1.15
1
rex = 130, L = 92 60 61.2 27.6 1.09
rex = 100, L = 100 60 50 25 1.03 0.9

rex = 60, L = 200 60 30 10 1.00 0.8

0.7

50 100 150 200

i.e. L = 50 mm, rex = 130 mm. For the longest Halbach Length of Halbach cylinder without blocks [mm]
cylinder with the smallest external radius, L = 200 mm, rex = Figure 14. The parameter η/ηorg describing the
60 mm, the additional blocks do not improve Λcool . The homogeneity of the field in the Halbach cylinder bore. The
reason for this is that the short designs are also the ones that data point marked with a star represents the best datapoint
have the greatest loss of flux through the ends of the cylinder from Fig. 13. The black crosses are Halbach cylinders with
bore, which is exactly what the additional magnets reduce. the same rin and L as the Halbach with additional blocks but
The optimal design of the additional blocks, i.e. the an- with a larger rex and no blocks. The black crosses are
gular span, height and depth, vary between the four Halbach displaced by ±6 mm on the x-axis to make the comparison
cylinders presented in Fig. 13, and the results can be read with the Halbach cylinders with additional blocks possible.
in Table 3. Here it is seen that the optimal design of the
additional blocks seems to be an angle span of around 60◦ ,
a height in the range 45 − 60 mm and a depth in the range cylinders with additional blocks. Also shown in the figure are
22.5 − 27.5 mm, at least for the three systems used in these Halbach cylinders with the same rin and L as the Halbach with
simulations. The parameters were varied as given in Table 2. additional blocks but with a larger rex and no blocks. It can
It can thus be concluded that for a short Halbach cylin- clearly be seen that the no-block designs with larger rex have a
der with a large external radius it is possible to optimize homogeneity parameter comparable to the “original” Halbach
the magnetic cooling parameter by using additional magnets cylinder without blocks, while a number of the designs with
placed at the ends of the cylinder. However, as can be seen by additional blocks clearly improve the homogeneity of the field
comparing Fig. 12 and 13 the improvement gained by using in the cylinder bore, i.e. lower η/η org .
the additional blocks is small compared to building a long
Halbach cylinder with a small rex in the first place. For ex- 5.2 Direction of magnetization
ample the design improved the most by the additional blocks, As previously mentioned the additional blocks all have a di-
rex = 130 mm and L = 50 mm, has the magnetic cooling pa- rection of magnetization perpendicular to the end face of the
rameter improved 1.15 times. This is not as impressive when Halbach cylinder. This might not be the optimal configuration,
one considers that the magnetic cooling parameter for this so various directions of magnetization have been tested to find
Halbach cylinder has a value of 0.011 (the lower right corner the greatest enhancement of the flux density. The direction
in Fig. 12), and even multiplied by 1.15 this is still much of magnetization was given by (0, Br cos(θ ), Br sin(θ )) where
lower than the longer Halbach cylinders. However, in spe- θ was varied in steps of 1◦ . The Halbach cylinder symmetry
cific cases with a limited geometry due to the application the axis is oriented along the z-axis.
additional blocks can still be used to improve the flux density. The result indicate that the mean flux density in the bore
could only be improved by less than 1% by changing the
5.1 Homogeneity of the field direction of magnetization from the 90◦ orientation used in
In most Halbach cylinder applications it is not only the flux the preceding simulations.
density that is important but also the homogeneity of the field
in the cylinder bore.
To characterizes the homogeneity of the flux density in 6. Discussion
the cylinder bore the quantity, It is important to discuss the possible influence of the coerciv-
ity of the individual magnets in the Halbach cylinder assembly.
<B2> − <B> 2
η≡ , (6) A serious problem in this context is that the assumed linearity
<B2> of the magnets is only valid when the magnetic field is above
where the angled brackets denoting volume average, is de- the value of the intrinsic (polarization) coercivity, Hc . For
fined. In Fig. 14 this parameter is shown for the Halbach typical 1.4 T NdFeB magnets µ0 Hc is around 1.2 T at room
 Optimization and improvement of Halbach cylinder design — 9/10

µ Hc + H . Br /|Br| problem of calculating the flux density is linear in the rema-

0
>0T nence. This means that the mean flux density both inside and
0T
-0.2 T
outside the Halbach cylinder depends linearly on the rema-
-0.4 T nence of the magnets. In this paper we have used magnets
-0.6 T with a remanence of 1.4 T. If one would e.g. replace all
-0.8 T these magnets in the Halbach cylinder with magnets with a
-1 T
-1.2 T
remanence of 1.2 T the mean flux density both inside and
< -1.2 T outside the Halbach cylinder would decrease by a factor of
1.2/1.4 = 0.86. This has be verified numerically.
There are still factors that have not been taken into account.
We have for example discussed the use of additional blocks
130 mm
while taking their shape for granted. It is necessary to test
10 mm
Figure 15. A quarter of 2D Halbach cylinder with rin = 10 if the circular design used for the additional blocks is the
mm and rex = 130 mm. Shown as a color map is the proper design to use. One could just as well have used e.g. a
expression µ0 Hc + H · Br /|Br |. When this expression is square design of the additional blocks. It is also important to
negative the magnetic field intensity is stronger than the investigate the effect of the additional blocks on a much larger
intrinsic coercivity of the magnet (taken to be µ0 Hc = 1.2T) sample of Halbach designs, including designs with varying
and the magnet will be reversed. The direction of the internal radii.
magnetization of the blocks are shown as black arrows.
7. Conclusion
In this paper we found the optimal values of rex and L for
temperature. Once the reverse component of the magnetic a Halbach cylinder with a given mean flux density and rin .
field reaches this value the linearity of the magnets breaks These configurations have the smallest volume of the magnet
down and a small increase in the magnetic field will reverse possible for a given mean flux density in the cylinder bore.
the direction of magnetization of the magnet. We have not Also, we found that placing blocks of additional permanent
modeled this nonlinearity but have assumed that the linear magnets on the sides of the Halbach cylinder can improve
relation is always valid. This is of course problematic when the flux density in the cylinder bore significantly. Finally, we
the magnetic field strength is too high. The reason the non- introduced a magnetic cooling efficiency parameter, Λcool , and
linearity is not modeled is that due to hysteresis a complete showed that the additional blocks can improve this by as much
history of the magnet material would be needed, including the as 15% compared to ordinary Halbach cylinders. However
physical building of the Halbach array, and this is not possible one must always take care that the polarization coercivity, Hc ,
to model. is always higher than the flux density in the Halbach cylinder
The part of the Halbach cylinder where this is a problem is gap.
around the inner equator(14). In Fig. 15 the projection of the
magnetic field intensity along the direction of the remanent
magnetism is shown for the largest Halbach considered in
Acknowledgements
this paper. Anisotropy is typically larger than coercivity for The authors would like to acknowledge the support of the
NdFeB magnets so any reverse field component is important Programme Commission on Energy and Environment (EnMi)
(14). It can be seen that the magnetic field is strong enough to (Contract No. 2104-06-0032) which is part of the Danish
reverse the direction of magnetization of typical industry 1.4 Council for Strategic Research.
T magnets at several locations.
The problem is present for Halbach cylinder with a flux
density in the bore larger than 1.2 T. To overcome this prob-
lem one can replace the magnets in the affected volume of the
Halbach cylinder with magnets with a higher intrinsic coer-
civity. Such magnets are readily available at slightly lower
remanences, e.g. a typical industry NdFeB magnet with a
remanence of 1.2 T has µ0 Hc = 3.2 T, which is sufficiently
strong to keep the relative permeability constant and thus the
material linear. Otherwise the best solution is to remove the
magnets from this part of the Halbach cylinder and replace
them with a nonmagnetic material.
An additional remark on the conducted numerical simula-
tions is that because the Halbach cylinder consists of magnets
with a relative permeability close to one, the magnetostatic
 Optimization and improvement of Halbach cylinder design — 10/10

References
[1] M. Sullivan, G. Bowden, S. Ecklund, D. Jensen, M.
Nordby, A. Ringwall, and Z. Wolf, IEEE 3, 3330 (1998).
[2] S. Appelt, H. Kühn, F. W Häsing, and B. Blümich, Nature
Physics 2, 105 (2006).
[3] J. M. D. Coey, Journal of Magnetism and Magnetic Ma-
terials 248, 441 (2002).
[4] J. C. Mallinson, IEEE Transactions on magnetics 9 (4),
678 (1973).
[5] K. Halbach, Nuclear instruments and methods 169, 1-10
(1980).
[6] Standard Specifications for Permanent Magnet Materi-
als, Magnetic Materials Producers Association, 8 South
Michigan Avenue, Suite 1000, Chicago, Illinois 60603
[7] T. R. Ni Mhiochain, D. Weaire, S. M. McMurry, and J.
M. D. Coey, J. Appl. Phys. 86, 6412 (1999).
[8] X. N. Xu, D. W. Lu, G. Q. Yuan, Y. S. Han, and X. Jin, J.
Appl. Phys. 95 (11), 6302-6307 (2004).
[9] G. Moresi, and R. Magin, Concepts in Magnetic Reso-
nance Part B (Magnetic Resonance Engineering) 19B (1),
35 (2003).
[10] COMSOL AB, Tegnérgatan 23, SE-111 40 Stockholm,
Sweden.
[11] Comsol, Comsol Multiphysics Model Library, third ed.
COMSOL AB, Chalmers Teknikpark 412 88 G (2005).
[12] O. Schenk, K. Gärtner, W. Fichtner, and A. Stricker,
Journal of Future Generation Computers Systems 18, 69
(2001).
[13] O. Schenk, and K. Gärtner, Parallel Computing 28, 187-
197 (2002).
[14] F. Bloch, O. Cugat, G. Meunier, and J. C. Toussaint, IEEE
Transactions on magnetics 34, 5 (1998).
[15] E. Potenziani, J. P. Clarke, and H. A. Leupold, J. Appl.
Phys. 61, 3466 (1987).
[16] K. A. Gschneidner Jr, V. K. Pecharsky, and A. O. Tsokol,
Rep. Prog. Phys 68, 1479 (2005).
[17] H. Osterreicher, and F. T. Parker, J. Appl. Phys. 55, 4334
(1984).
[18] V. K. Pecharsky, and K. A. Gschneidner Jr, International
Journal of Refrigeration 29 (8), 1239 (2006).