Closed-form expressions for the magnetic field of

permanent magnets in three dimensions

arXiv:2106.04153v3 [physics.app-ph] 1 Nov 2022

V. Ziemann, Uppsala University

November 2, 2022

Abstract
We derive a closed-form expression of the magnetic field of a finite-size
current sheet and use it to calculate the field of permanent magnets, which
are modeled through their surface current densities. We illustrate the
method by determining the multipoles and the effective length due to fringe
fields of a finite-length dipole constructed of magnetic cubes.

1 Introduction
Today permanent magnets are widely used in the construction of efficient elec-
trical motors, for example those used in battery-driven automobiles, but also for
scientific applications, such as ion traps to store and analyze rare atoms, or for
undulators to generate synchrotron radiation, and for multipole magnets to guide
and confine charged-particle beams. These magnets do not need to be powered
and are therefore very energy efficient, which makes them attractive regarding
today’s focus on sustainability. In the 1970s Klaus Halbach showed [1] how to
design multipole magnets and undulators using paper and pencil only. His ap-
proach made use of variables in the complex plane and is essentially limited to two
spatial dimensions. In order to analyze three-dimensional features we typically
use finite-element codes [2]. In this report, we offer a complementary approach
based on solving Biot-Savart’s equation for a finite-size rectangular current sheet
and modeling the magnets as an assembly of such sheets. In iron-free geometries
~
this allows us to calculate the three components of the magnetic field B(z) at any
point in space as the superposition of the contributions from the current-carrying
sheets. Since all integrations are done analytically this results in expressions that
make calculating the fields very rapid. Moreover, the field values are exact, no
approximations due to finite meshes are involved.
A key feature of permanent magnet material is that the relative magnetic
permeability µr is very close to unity, which makes them transparent to other,
externally generated fields. As a consequence, the field at a given point ~z is the

1
superposition of the fields due to all current sources. And here the second key
feature enters; the magnetic field of permanent magnets can be calculated from
the equivalent surface current density I 0 = dI/ds = Br /µ0 that is proportional
to the remanent magnetic field Br of the permanent magnet material. Here
µ0 = 4π × 10−7 A/Tm is the permeability of free space. Since Br is on the order
of one Tesla, the corresponding current densities are enormous.
These observations now guide us through the remainder of this report. We first
calculate the field of a straight current-carrying filament, followed by spreading
the filament laterally to obtain a current sheet and calculating the field it gener-
ates. In the subsequent section we briefly introduce software, based on MATLAB,
to assemble magnets that are composed of these sheets and calculate their fields.
We then use the software to illustrate the field of a single sheet, a magnet cube
assembled of four sheets, and a dipole magnet, made of four cubes, whose field
quality and fringe fields we analyze. Finally, we consider solenoid magnets which
help us to validate the calculations and the software.

2 Field of a filament
The magnetic field at point ~r2 due to a filament that carries a current I, starts
at position ~ra , and ends at ~rb is given by Biot-Savart’s law

~ µ0 I Z d~r1 × (~r2 − ~r1 ) µ0 I Z 1 [~rb − ~ra ] × [~r2 − t~rb − (1 − t)~ra ]

B(~r2 ) = = dt ,
4π |~r2 − ~r1 |3 4π 0 |~r2 − t~rb − (1 − t)~ra )|3
(1)
where we parameterize a point on the filament ~r1 by ~r1 = (1 − t)~ra + t~rb with
0 < t < 1. First we simplify the numerator in the last equality and find

(~rb − ~ra ) × (~r2 − t~rb − (1 − t)~ra ) = (~rb − ~ra ) × (~r2 − ~ra ) (2)

whereas for the denominator we obtain

|~r2 − t~rb − (1 − t)~ra )|2 = (~r2 − t~rb − (1 − t)~ra ) · (~r2 − t~rb − (1 − t)~ra )
= (~r2 − ~ra )2 − 2t(~r2 − ~ra ) · (~rb − ~ra ) + t2 (~rb − ~ra )2
= A − 2Bt + Ct2 (3)

with

A = (~r2 − ~ra )2 , B = (~r2 − ~ra ) · (~rb − ~ra ), and C = (~rb − ~ra )2 . (4)

Note that these parameters have simple geometric interpretations: A describes

the squared distance dP between the observation point P at ~r2 and the starting
point of the filament ~ra . The variable B is the product of dP , the length L of
the filament, and the cosine of the enclosed angle, whereas C is the length of the

2
 Figure 1: Current sheet.

filament squared. Inserting these expressions into the equation for the magnetic
~ we get
field B
Z 1
~ µ0 I dt
B(~r2 ) = (~rb − ~ra ) × (~r2 − ~ra ) . (5)
4π 0 (A − 2Bt + Ct2 )3/2

This integral can be found in integral tables [3] and is given by

Z
dt Ct − B
2 3/2
= √ (6)
(A − 2Bt + Ct ) (AC − B 2 ) A − 2Bt + Ct2
~ r2 ) as
such that we write B(~
" #
~ r2 ) = µ0 I (~rb − ~ra ) × (~r2 − ~ra ) √ C − B
B(~
B
+√ (7)
4π AC − B 2
A − 2B + C A
where A, B and C are given in Equation 4. Equation 7 is already quite useful
to determine the fields of systems containing a finite number of current-carrying
wires. We previously used it to design earth-field compensation coils [4].

3 Field of a finite current sheet

In order to calculate the field of a sheet, we first consider the field that is generated
by a filament that is laterally displaced by the vector

p~ = α(~rb − ~ra ) × ~n , (8)

where ~n is the vector normal √to the sheet and α is the aspect ratio of the sheet,
such that w = α|~rb − ~ra | = α C is the width of the sheet. The construction is

3
illustrated in Figure 1. Note that p~ is perpendicular to the filament; the scalar
product of ~rb −~ra and p~ is zero. This construction permits us to parameterize the
parallel-displaced filament by its end points ~ra +s~p and ~rb +s~p, where −1/2 ≤ s ≤
1/2 describes by how far the filament is displaced, as shown in Figure 1 by the thin
dotted line that is parallel to ~rb −~ra . The field from the displaced filament is then
given by replacing the parameters A, B and C by their s-dependent counterparts
given by

A(s) = (~r2 − ~ra − s~p)2

B(s) = (~r2 − ~ra − s~p) · (~rb − ~ra ) (9)
C(s) = (~rb + s~p − ~ra − s~p)2 = C .

Moreover, the cross-product before the integral becomes (~rb −~ra ) × (~r2 −~ra − √
s~p).
0
We now “spread out” the current I = I w evenly across the width w = α C
of the sheet with constant current density I 0 and sum up the contributions of
all displaced filaments, which corresponds to integrating over s. We also note
that the current density I 0 is related to the remanent magnetic field Br through
µI 0 = Br . Combining these concepts we find that the the field at point ~r2 is given
by the following integral
√ Z
~ r2 ) = B r α C 1/2 (~rb − ~ra ) × (~r2 − ~ra − s~p)
B(~ (10)
4π −1/2 A(s)C − B(s)2
 
C − B(s) B(s) 
× q +q ds
A(s) − 2B(s) + C A(s)

where A(s) and B(s) are at most quadratic functions of s. The evaluation of the
integral, which is rather lengthy and therefore deferred to Appendix A, leads to

~ r2 ) = Br αC [(C − B)J1 + BJ3 ] (~rb − ~ra ) × (~r2 − ~ra )

B(~ (11)
4π
Br αC 2
+ [(C − B)J2 + BJ4 ] ~n ,
4π
where J1 to J4 are defined in Appendix A in Equations 22 to 24. We point
out that αC is the area of the sheet, such that Br αC describes the integrated
“strength” of the sheet to excited magnetic fields.

4 Software implementation
Equation 11 describes the magnetic field vector B(~ ~ r2 ) that is caused by one sheet
that is characterized by ~ra , ~rb , ~n, α, Br in a coordinate-free way and only requires
function evaluations; no integrations are necessary which makes the calculation

4
Figure 2: Field pattern of a current sheet (left) and a four sheets assembled to
generate the field of a permanent magnet cube (right).

~ r2 ) very fast. In order to calculate the field at a given point ~r2 due to several
of B(~
sheets we only have to sum over the contributions from each of the sheets.
We therefore represent each sheet by a one-dimensional array that contains
the three spatial components of ~ra , ~rb , ~n as well as α and Br for that sheet; thus
there are eleven numbers to store for each sheet. The entire geometry is then
given as a collection of, say M , such sheets that is stored in an array of dimension
M × 11.
In order to place the sheets within the geometry, we prepare functions to
translate sheets by a vector ~a; the function simply adds ~a to ~ra and ~rb in the array
with the sheets. Likewise, three functions to rotate a sheet by an angle θ around
each of the three axes, just multiply ~ra , ~rb , and ~n in a sheet by the appropriate
rotation matrix. Using these functions, it is easy to create any geometry, very
similar to the way that three-dimensional modeling software composes complex
geometries from geometric primitives, often triangles, but also from cubes or
spheres.
Finally functions to calculate the field at a point or along a path just loop
over the sheets and add their contributions. We implemented these functions in
MATLAB, briefly explain the functions in Appendix B, and illustrate their use
with a number of simple examples in the next section.

5 Examples
As a first example we consider the field of a single sheet, which is shown on the
left-hand image in Figure 2. The sheet is shown as the blue rectangle and the
current as the green line stretching in the center of the sheet. The vertical green

5
arrow indicates the normal vector ~n of the sheet. We observe that the magnetic
field vector B~ curls around the sheet, perpendicular to the current’s direction,
pointing to the right above and to the left below the sheet. Making the sheets
very narrow causes the field to approach a circular form, just as expected from
Ampere’s law.
In the next step we assemble four quadratic current sheets to represent a
permanent magnet cube, which is shown on the right-hand image in Figure 2.
We see that the current, again shown as a green line, flows around the surface of
the cube. We use the convention that the normal vectors, again shown as green
arrows, point towards the outside of the magnet. The easy axis of the magnet is
indicated by the yellow arrow on the sheets, that is oriented perpendicular to the
current flow and points upwards. The magnetic field created by the four sheets
representing the permanent magnet cube is illustrated by the red arrows. We
observe that the field points upwards at the top of the cube, then turns over such
that it points down on the outside of the cube, then turns over again to point
upwards below the cube. We point out that the four current sheets form a square
solenoid and the field therefore corresponds to that geometry.
We reach the next level of complexity by assembling eight 5 mm-cubes with
Br = 1.47 T to form an approximation of a Halbach dipole, discussed in Ap-
pendix A.3 in [5]. The top left image in Figure 3 shows the view onto the as-
sembly from the top. The easy axis of the magnets, shown by the yellow arrows,
rotates twice when moving from one magnet to the next; in every fourth of the
eight magnets it points in the same direction. In the center of the assembly, the
magnetic field points upwards, as is expected for a dipole. The plot on the top
right shows the vertical field component By along the black dotted line shown on
the left-hand image. We observe that the field is on the order of 0.35 T and has a
small dip in the center. The image on the left-hand side in the middle row shows
a perspective view of the same geometry which makes the three-dimensional char-
acter obvious. The vertical red line in the center of the assembly indicates the
direction of the z-axis and the plot on its right shows By along this direction,
but extending over ±20 mm. We see that the peak field of about 0.35 T appears
in the center of the magnet, but the fringe field extends significantly outside the
magnet, which only has a a height of ±2.5 mm.
We determine the multipolarity of the assembly at the radius r shown by the
cyan circle on he top-left image. A tangent to the circle, illustrated by the cyan
straight line, is also shown. In order to determine the multipoles we calculate
the scalar product of the tangent vector with the magnetic field at every point
on the circle. This yields the tangential field component Bt which we display in
the top panel of the image at the bottom left plot as a function of the azimuthal
angle φ. The multipolarity, shown on the middle panel then follows from a
Fourier-transformation, as discussed in Appendix C. We find that mainly the
dipole component with multipolarity m = 1 is present and that its amplitude is
Bmax = 0.353 T. Moreover, from the real and imaginary part of the transform,

6
Figure 3: The geometry of a permanent magnet solenoid (top left) and the
vertical field in the midplane in the x-direction (top right). A 3D rendition
of the magnet (middle left) and the field along the z-axis (middle right). The
bottom left shows the field along the circle shown in cyan and the multipole
coefficients (absolute and relative) in the midplane. The bottom right shows the
corresponding integrated values.

7
Figure 4: The geometry of a permanent magnet solenoid (left) and the field on
axis (right).

we determine the roll angle of the magnet, which here turns out to be close to
zero. On the bottom panel we show the amplitudes of the higher multipoles,
normalized to Bmax . We find the decapole component (m = 5) with a relative
amplitude on the order of a percent to give the dominant contribution. Finally,
we calculate the integral of Bt along a line in the range z = ±20 mm with x and
y-positions determined by the cyan circle and display it on the upper panel in the
image on the bottom-right of Figure 3. Again, mostly the dipole component with
m = 1 is present. Moreover the field integral turns out to be 3.63 Tmm, which
allows us to determine the effective length of the magnet to be about 10.3 mm, a
value that significantly exceeds the physical length of the magnet, which is only
5 mm. Again the decapole component with a relative magnitude of about 6×10−3
gives the dominant contribution to the integrated field.
As last example we consider an axial permanent magnet solenoid with Br =
1.47 T, inner radius 3 mm, and outer radius 10 mm. It is modeled by circular
currents flowing in one direction at the inner radius and in the opposite direction
at the outer radius. The left-hand image in Figure 4 shows the assembly, here
the circular magnet is approximated by a hollow polygon with 18 sides. The
field along the central axis of two long concentric circular solenoids is known
analytically [6]. We therefore compare it to our numerical solution, based on
Equation 11, and show both on the right-hand plot in Figure 4. We observe that
the agreement is very good.

6 Conclusions
We determined an analytic expression for the magnetic field generated by a rect-
angular current sheet and used it to model permanent magnets in iron-free ge-

8
ometries. We illustrated the method by calculating the fields of simple geometries
such as a single sheet, cubes, and solenoids assembled from multiple sheets.
We expect this algorithm to be useful for rapid prototyping of three-dimensio-
nal permanent-magnet structures. It is rather fast because all integrations are
done analytically and only fields at the points of interest need to be calculated.
This is in contrast to finite-element methods. They first need to determine the
fields on a grid in the whole geometry and then interpolate the fields to the
points of interest. On the other hand is the present method limited to iron-free
geometries and to magnets that can be modeled with rectangular current sheets.
Discussions with Roger Ruber are gratefully acknowledged. In part, funding
is provided through the project Disseminating technology for cold magnets to
provide access to a wider international market that is supported by the European
Regional Development Fund (ERDF) and Region Kronoberg.

References
[1] K. Halbach, Design of permanent multipole magnets with oriented rare earth
cobald magnets, Nuclear Instruments and Methods 169 (1980) 1.

[2] Finite-element codes, for example, COMSOL https://www.comsol.com.

[3] I. Gradstein, I. Ryshik, Tables of series, products, and integrals, Verlag Harri
Deutsch, Thun, 1981.

[4] V. Ziemann, R. Wedberg, T. Peterson, A. Wiren, Retro-Fitting Earth-Field

Compensation Coils to the Vertical Cryostat GERSEMI in FREIA, Instru-
ments 2020, 4(1), 8.

[5] V. Ziemann, Hands-On Accelerator Physics Using MATLAB, CRC Press,

Boca Raton, 2019.

[6] Q. Peng, S. McMurry, J. Coey, Axial magnetic field produced by axially and
radially magnetized permanent rings, Journal of Magnetism and Magnetic
Materials 268 (2004) 165.

A The integral
In this appendix we evaluate the integral, given in Equation 10. First we consider
the cross product before the integral

(~rb − ~ra ) × (~r2 − ~ra − s~p)

= (~rb − ~ra ) × (~r2 − ~ra ) − αs(~rb − ~ra ) × [(~rb − ~ra ) × ~n]
= (~rb − ~ra ) × (~r2 − ~ra ) + sαC~n (12)

9
where we used the identity ~a × [~b × ~c] = ~b(~a · ~c) − ~c(~a · ~b) to evaluate the double
cross product in the second line. Next, we calculate
A(s) = (~r2 − ~ra )2 − 2s(~r2 − ~ra ) · p~ + s2 p~2 = A − 2Ds + s2 p~2 (13)
where we introduce D = (~r2 − ~ra ) · p~. For p~2 in the last term we find
p~2 = α2 [(~rb − ~ra ) × ~n] · [(~rb − ~ra ) × ~n]
h i
= α2 (~rb − ~ra )2~n2 − ((~rb − ~ra ) · ~n)2 = α2 C , (14)
which allows us to write
A(s) = A − 2Ds + α2 Cs2 (15)
as a second-order polynomial in the variable s. For B(s) we arrive at
B(s) = (~r2 − ~ra − s~p) · (~rb − ~ra ) (16)
= B − sα [(~rb − ~ra ) × ~n] · (~rb − ~ra ) = B ,
which is constant. Inserting these parameters into Equation 10 leads to
0
√ Z
~ r2 ) = µ 0 I α C 1/2 (~rb − ~ra ) × (~r2 − ~ra ) + αsC~n
B(~ ds (17)
4π −1/2 AC − B 2 + C 2 − 2sCD + α2 C 2 s
" √ √ #
C(C − B) C(B)
√ +√ .
AC − 2BC + C 2 − 2sCD + α2 C 2 s2 AC − 2sCD + α2 C 2 s2
Close inspection of the integrand shows that the second-order polynomial in the
denominator of the first line only differs by a constant term from the polynomials
under the roots in the second line. Introducing the abbreviations
R1 = a1 +b1 s+c1 s2 with a1 = AC −2BC +C 2 , b1 = −2CD, c1 = α2 C 2 (18)
and
R2 = a2 + b2 s + c2 s2 with a2 = AC, b2 = −2CD = b1 , c2 = α2 C 2 = c1 (19)
we can write
0
"
~ r2 ) = µ 0 I αC Z 1/2
(~rb − ~ra ) × (~r2 − ~ra ) + αsC~n C − B
B(~ ds √
4π −1/2 p1 + R1 R1
#
(~rb − ~ra ) × (~r2 − ~ra ) + αsC~n B
+ √
p2 + R2 R2
0
" Z 1/2
µ0 I αC ds
= (~rb − ~ra ) × (~r2 − ~ra )(C − B) √
4π −1/2 (p1 + R1 ) R1
Z 1/2
sds
+~nαC(C − B) √
−1/2 (p1 + R1 ) R1
Z 1/2
ds
+(~rb − ~ra ) × (~r2 − ~ra )B √
−1/2 (p2 + R2 ) R2
Z 1/2 #
sds
+~nαCB √ (20)
−1/2 (p2 + R2 ) R2

10
with p1 = −(B − C)2 and p2 = −B 2 . We denote the integrals by
Z 1/2 Z 1/2
ds sds
J1 = √ , J2 = √ ,
−1/2 (p1 + R1 ) R1 −1/2 (p1 + R1 ) R1
Z 1/2 Z 1/2
ds sds
J3 = √ , J4 = √ . (21)
−1/2 (p2 + R2 ) R2 −1/2 (p2 + R2 ) R2
We note that J3 and J4 resemble J1 and J2 , only the constant parameters ai , bi , ci ,
and pi for i = 1, 2 differ. We therefore only consider J1 and J2 in the following.
As a matter of fact, closed expressions for these integrals are given as number
2.284 in [3]. In particular J1 and J2 can be determined from
Z
Us + V U 2V c1 − U b1
U J2 +V J1 = √ ds = I1 (s)− q I2 (s) (22)
(p1 + R1 ) R1 c1 c21 p1 [b21 − 4(a1 + p1 )c1 ]

with I1 (s) given by [3]

 q 
 √1 arctan R1
for p1 > 0
p1 p1√ 
I1 (s) =  √
−p − (23)
√1
2 −p1
ln √−p1 +√R
1 1
R1
for p1 < 0 .

With d1 = b21 − 4(a1 + p1 )c1 the second contribution I2 (s) is given by [3]
 q 
p1 b1√+2c1 s

 − arctan d1 R1
for p1 d1 > 0


 √ √ √ 
1 −d1 R1 + p1 (b1 +2c1 s)
I2 (s) = 2i
ln √−d1 √R1 −√p1 (b1 +2c1 s) for p1 d1 < 0 and p1 > 0 (24)

 √ √ √ 
1 + −p1 (b1 +2c1 s)
 1 ln √dd11 √R

√ for p1 d1 < 0 and p1 < 0

2i R1 − −p1 (b1 +2c1 s)

where R1 is a function of s, given in Equation 18. We point out that the sign
of I2 (s) in Equation 22 is reversed with respect to [3] to make the numerically
evaluated integral consistent with the value given in Equation 24. Inserting J1
through J4 in Equation 20 and reordering terms then leads to the result stated
in Equation 11 in the main body of this report.

B MATLAB functions
The following MATLAB functions implement the functionality described in Sec-
tion 4 and are used to generate the plots shown in the main body of this report.
Note that sheets is a M × 11 array that describes M sheets.
• B=Bsheets(sheets,r2): returns B(~
~ r2 ) for the sheets defined by the input
argument.
• sheets=make cubez(a,alpha,Br): returns the four sheets for a square
block with width a and height αa with the easy axis pointing in the z-
direction.

11
 • sheets=make brick(a,b,c,Br): returns the four sheets for a rectangular
block with sizes a, b, and c in the x, y, and z direction, respectively. The
easy axis points in the z-direction.

• sheets=make polygon(n,r,h,Br): returns the n sheets for a polygon with

n sides, where r specifies the distance from the center to the corners and h
is the height in the z direction. The easy axis also points in the z-direction.

• sheets=make polygon hollow(n,ri,ro,h,Br): returns the 2n sheets for a

hollow polygon with n sides, where ri specifies the distance from the center
to the corners of the inner polygon and ro to the outer polygon; h is the
height in the z direction. The easy axis also points in the z-direction. This
function creates “solenoidal” permanent magnets with single call.

• sheets=sheets translate(sheets,a): returns the sheets provided as in-

put after adding a vector ~a to ~ra and ~rb of the sheets.

• sheets=sheets rotate x(sheets,theta): returns the sheets provided as

input after the corresponding vectors ~ra , ~rb , and ~n of the sheets are rotated
by the angle θ. There are corresponding functions for rotations around the
y and z-axis.

• draw sheet(sheet): draws one sheet as a blue surface and adds a green
line to indicate the direction of the current flow. A yellow arrow indicates
the easy axis.

• B=field along line(sheets,line): returns an array of B(~

~ r2 ) at the se-
quence of points ~r2 provided in the array line.

The use of these functions is best illustrated in the code to prepare a Halbach
multipole. The code below first prepares a 5 mm cube and the rotates it such
that the easy axis points in the vertical direction. After defining the number
of cubes and the multipolarity we loop over the cubes. Inside the loop we first
rotate the cube as needed to generate the desired multipolarity m, then translate
it radially along the x axis by 8 mm before rotating into place a second time.
Last, we add the sheets of the currently handled cube to the sheets which, after
the loop completes, contains all the sheets that define the multipole.

cube=make_cubez(5,1,Br);
cubey=sheets_rotate_x(cube,-90);
MM=8; % number of cubes
m=1; % multipolarity (m=1: dipole, 2: quadrupole)
for k=0:MM-1
tmp=sheets_rotate_z(cubey,-k*m*360/MM);
tmp=sheets_translate(tmp,[8;0;0]);

12
 tmp=sheets_rotate_z(tmp,-k*360/MM);
if k==0
sheets=tmp;
else
sheets=[sheets;tmp];
end
end

C Determining the multipolarity

On magnet test benches the multipolarity of magnets is often determined from
the voltage that is induced in a coil that rotates inside the magnet. In a dipole the
polarity changes once per revolution and the voltage shows a spectral component
at the revolution frequency. In a quadrupole, the polarity changes twice per
revolution and the spectrum shows a peak at the second revolution harmonic,
whose amplitude is proportional to the quadrupole gradient.
Here, we use a simplified version of this procedure and calculate the projection
of the magnetic field onto a (normalized) vector that is tangent to a circle with
radius r inside the magnet. This is illustrated by the cyan circle and tangent
vector visible on the top-left image in Figure 3. In much the same way as on a
measurement bench, the number of polarity changes when traveling around the
circle is related to the multipolarity m of the magnet. The amplitude B̂ of the
Fourier harmonic directly gives the magnetic field harmonic on the circle. For
example, in a quadrupole, B̂ is related to the gradient g by g = B̂/r.
Moreover, if we denote the real and imaginary part of the largest Fourier
harmonic by am + ibm , the roll angle φ̂ of the magnet around the z-axis (pointing
upwards on the top-left image in Figure 3) is given by φ̂ = (1/m) arctan(bm /am ).
In the legend of the images in the bottom row this value is also reported. It is,
however, very small, because the magnet was not rotated.
If we use the field on the cyan circle, we measure the harmonic at a given
z-position, similar to what a short rotating coil would do. If, on the other hand,
we integrate the field along a line extending in the z-direction and perpendicular
to positions on the circle, we determine the harmonics of the integrated field,
similar to what a long coil would do. This information is reported on the bottom
left image in Figure 3.

13