Académique Documents
Professionnel Documents
Culture Documents
The magnetic field of a permanent magnet is calculated analytically for different geometries. The cases of a sphere, cone, cylinder, ring and
rectangular prism are studied. The calculation on the axis of symmetry is presented in every case. For magnets with cylindrical symmetry,
we propose an approach based on an expansion in Legendre polynomials to obtain the field at points off the axis. The case of a cylinder
magnet was analyzed with this method by calculating the force between two magnets of this shape. Experimental results are presented too,
showing a nice agreement with theory.
Keywords: Permanent magnet; calculation; analytical; symmetry.
∞ µ
X ¶ Evaluating in the limits and simplifying, we obtain 2/z 2 ,
` V`
Φ(r, θ) = U` r + `+1 P` (cos θ) which does not depend on r0 . Finally, performing the inte-
r
`=0 gration on this variable, we obtain
M R3 1
Φ(r, θ) = P1 (cos θ)
3 r2
M R3 cos θ 1 m cos θ
= 2
=
3 r 4π r2
Z
z−h
M xdx
Φaxis (z) =
2 [Dx2 + Ex + F ]1/2
z
where
R2 2R2 ³ z´ 2
³ z ´2
D =1+ , E = 1 − , F = R 1 − .
h2 h h h
M h2 h p i
Φaxis (z) = ±(z − h) − R 2 + z2
2(h2 + R2 )
F IGURE 2. Scheme for calculating the field produced by a conical
magnet. M hR2 (z − h)
+
2(h2 + R2 )3/2
2.2. Cone √
(h ± h2 + R2 )(z − h)
× ln √ √ (4)
According to Fig. 2, each infinitesimal dipole contributes to R2 + hz + h2 + R2 R2 + z 2
the potential on the z axis with:
1 dm cos α 1 M ρ0 dρ0 dz 0 dφ0 (z − z 0 ) Then, the induction field has the following dependence
dΦaxis (z) = =
4π |~r − r~0 |2 4π [(z − z 0 )2 + ρ02 ]3/2 on the z axis:
Therefore, the potential on the axis is given by " µ ¶
Z2π Zh µ0 M h 1 zh−R2
M B(z)= √ ∓ h
Φaxis (z) = dφ0 (z − z 0 )dz 0 2 h2 +R2 R2 +z 2
4π #
0 0 √
ρ0max 0 R2 (h ± h2 +R2 )(z−h)
Z (z ) − 2 ln √ √ (5)
ρ0 dρ0 (h +R2 )3/2 R2 +hz+ h2 +R2 R2 +z 2
×
[(z − z )2 + ρ02 ]3/2
0
0
0
Zh R(1− zh )
Z The double sign in the Eqs. (4) and (5) apply to the val-
M 0 0 ρ0 dρ0 ues of z above the apex of the magnet (z > h, upper sign)
= (z − z )dz
2 [(z − z )2 + ρ02 ]3/2
0 or below the base (z < h, lower sign). Interestingly, the
0 0
model predicts that field diverges at the apex of the magnet.
Zh
Mh M (z − z 0 )dz 0 This feature is the basis of the design of electromagnets with
= − z 0 2 1/2 truncated conical poles, which produce high magnetic fields.
2 2 [(z − z 0 )2 + R2 (1 − h) ]
0 The variation of this field along the z axis was calculated by
Kroon [11].
2.3. Cylinder
As shown in Fig. 3 (b), the potential on the z axis due to each infinitesimal dipole is given by:
1 cos θ0 1 z − z0 M z − z0
dΦaxis = dm = M dV = ρ0 dρ0 dz 0 dφ0
4π |~r − r~0 |2 4π |~r − r~0 |3 4π [(z − z 0 )2 + ρ02 ]3/2
Therefore, the total scalar potential on the axis is given by:
Z ZL ZR
M z − z0 M ρ0 dρ0
Φaxiz (z) = ρ0 dρ0 dz 0 dφ0 = (z − z 0 )dz 0
4π [(z − z 0 )2 + ρ02 ]3/2 2 [(z − z 0 )2 + ρ02 ]3/2
cylinder 0 0
F IGURE 3. (a) Schematic of a cylindrical magnet. (b) Scheme for calculating the field produced by the magnet.
2.4. Ring
#
ab
− arctan p (11)
(z + c) a2 + b2 + (z + c)2
For completeness we reproduce here the results of Yang
et al, [12] who calculated the magnetic field at any point in
space. Adapting their results to the case of a magnetization
directed along the z axis, we get:
F IGURE 5. Scheme for calculating the field of a prism-shaped mag- µ0 M F2 (−x, y, −z)F2 (x, y, z)
Bx (x, y, z) = ln
net. 4π F2 (x, y, −z)F2 (−x, y, z)
4 ab
= 0
arctan p 3. Results of magnetic fields on the z axis and
z−z (z − z ) a + b2 + (z − z 0 )2
0 2
comparison with experiments
Then, we obtain: Now we compare the theoretical calculations of magnetic
Zc fields on the symmetry axis with measurements made in the
M ab laboratory. First, from Eq. (3) we obtain the field for the
Φaxis (z) = arctan p dz 0
π (z − z 0 ) a2 + b2 + (z − z 0 )2 sphere:
−c 2µ0 M R3
B(z) = (12)
3z 3
Mathematica gives (except additive constants omitted
We measured the magnetic field using a gaussmeter with
here):
an estimated accuracy of ±5%. From a linear fitting of
" this field vs. 1/z 3 , we obtained a magnetization value
M ab µ0 M = (0.933 ± 0.044) T, which lies in the range expected
Φaxis (z) = (z + c) arctan p
π (z + c) a2 + b2 + (z + c)2 in these magnets. For example, A. Walther et. al. [13] re-
# ported a value of 1.25 T for thin films of N d2 F e14 B; in the
ab case of single crystals it has been reported [14,15] a value of
− (z − c) arctan p (10)
(z − c) a2 + b2 + (z − c)2 1.6 T at room temperature. The magnetization of the other
and F (x, y) and E(x, y) are the Incomplete Elliptical Inte- Akoun et. al [19]. Also, recently it has been possible to cal-
grals of first and second class respectively. culate the torque between these magnets [20].
To calculate the vertical component of the field, we need To realize our study, we placed the two magnets coaxially
to take as shown in Fig. 7. The vertical force was measured with a
Pasco Scientific CI-6537 sensor mounted onto a homemade
1 ∂(Aρ)
Bz (ρ, z) = universal base. The upper magnet was attached to the sen-
ρ ∂ρ sor and the lower one was attached to a mobile stage which
Unfortunately, none of the programs that we tried (Mathe- could be displaced vertically with a screw. Each 360◦ turn of
matica and Maple) were able to provide an analytical result the screw produced a displacement of 1.49 mm. This way,
in this case. As we can see, the magnetic induction field out- the distance between the magnets (a) could be varied with a
side the cylinder axis is difficult to obtain. In several previ- high precision. The sensor has an accuracy of ±0.03 N.
ous interesting works published in educational journals, the The force exerted by magnet 1 on magnet 2 is given by
Z
field of cylindrical magnets or solenoids has been calculated.
~
F = ∇(d~ ~
µ · B)
For example, Labinac et al [16] calculated the field for a thin
solenoid and a thick coil using a series expansion of Gauss magnet−2
hypergeometric functions. Derby and Olbert [17] calculated
Here, B~ is the the field produced by the lower magnet.
the field from a solenoid to study the speed of fall of perma-
nent cylindrical magnets inside a copper tube. Lerner [18] an- More specifically, the vertical force is
Z
alyzed the role of the permeable core of a finite solenoid, by ∂Bz
comparing its field with that produced by an infinite solenoid. F = M2 dV (13)
magnet−2 ∂z
In each case, the calculation involved elliptic integrals, whose
evaluation is difficult. This is where the expansion of Φaxis (z) The first step of the calculation is to expand the function
proposed in the present work will exhibit its usefulness. given by Eq. (6) in powers of z. The terms to expand can be
written in the form:
4.2. Force between two cylindrical magnets p 1 1 1 5 8
1 + x2 = 1 + x2 − x4 + x6 − x + ...
2 8 16 128
As we just showed, it is very difficult to obtain closed ex-
pressions of the field for a cylindrical magnet. To illustrate This expansion is valid for |x| < 1. For this reason, if R > L,
the method of approximation in power series, we decided three regions are defined: close (L < z < R), intermediate
to calculate the force between two cylindrical magnets with (R < z < L + R) and far (z > L + R). If L > R no close
uniform magnetization, and compare our results with experi- region exists; only the intermediate (L < z < L + R) and far
mental measurements. Before that, we want to mention that (z > L+R) regions remain. Since R1 = 1.27 cm > 0.95 cm
similar calculations have been reported previously. The force = L1 , we need to consider the three regions .
between two cubic magnets was calculated analytically by The infinite series can be truncated by taking a finite num-
ber of terms, up to a maximum power of z (`max ). The approx-
imation to the exact potential improves as more terms are in-
cluded, but the calculation of the force becomes increasingly
elaborate. The choice of `max depends on these two factors. potential and goes like 1/r, the second term corresponds to
The expansion coefficients for each zone are listed in the ap- the dipolar distribution and goes like 1/r2 , the third term is
pendix. The coefficients were taken up to the greatest value the quadrupole one and goes like 1/r3 , the octupole term
the computer could handle (`max = 10, 9 or 8). goes like 1/r4 , and so on. Therefore, it is possible to identify
Figure 8 shows the plots of the exact potential due to the Eq. (14) as the multipole expansion of the magnetic scalar
magnet 1 on its axis, and the approximation of the finite se- potential produced by the permanent magnet. As expected,
ries. As it can be seen, the approximation is quite good ex- the monople term vanishes because of the nonexistence of
cept near the borders between neighbor zones (z − L1 = 0.58 magnetic monopoles. The leading term in this equation is
and 0.93 cm), where convergence of the series is slower. As
mentioned, one uses a certain number of terms to give a good
M LR2 cos θ 1 m cos θ
balance between accuracy and computation time. In Ref. 7,
2
= ,
the author used up to 20 terms in the series he proposed, get- 4 r 4π r2
ting an excellent agreement with the exact calculation; how-
ever, he found discrepancies of the order of 100% in certain i.e., the dipole contribution. In all the geometries studied
regions. In our example, the maximum difference observed here, we expanded Φaxis (z) in a power series of 1/z around
with respect to the exact value on the axis was about 15%. zero. We found in every case that the dipole contribution
We just want to bring in the idea that in any finite series ap- given by Eq. (1) was the leading term, which makes a lot
proximation, there will be regions of very slow convergence, of sense since at large distances the magnet “looks” like a
in which there will be significant deviations from the exact point dipole. The similiarity between the electrostatic multi-
function. pole expansion and the present one is not casual: it is easy to
Now, it is interesting to take a close view at the explicit demonstrate [22] that the scalar potential due to a uniformly
expression in the far region. The series expansion is: magnetized body of arbitrary form may be written in the form
Φ = −M ~ · ∇(²0 ψ), where ψ is the electrostatic potential due
∞ ∞ µ ¶µ ¶
M X X i 2k 1/2 2k − 2 + i 1 to a uniformly charged body (with ρe = 1) of the same form
Φaxis (z) = LR 2k−1+i
2 i=1
k 2k − 2 z and dimensions. Indeed, the dipole term of the magnet ex-
k=1
pansion (14) can be obtained by applying the latest formula
Therefore, the scalar potential at any point of this region to the monople term of the expansion of Φρe (~x), with ρe = 1.
is given by: The second term of Eq. (14) can be obtained from the sec-
à ond term of the electrostatic series, an so on. Therefore, there
M LR2 cos θ L2 R2 3 cos2 θ − 1 exists a tight connection between electrostatic and magneto-
Φ(r, θ) = +
2 2 r2 4 r3 static potentials due to uniform sources, and the present cal-
! culation is a nice and educational manifestation of this char-
4L3 R2 − 3LR4 5 cos3 θ − 3 cos θ acteristic.
+ + ... (14)
16 r4
Next, we calculated the vertical component of the mag-
This expression resembles the multipole expansion of the netic induction field:
electrostatic potential Φρe (~x) outside the region containing
all source charges [9]:
1 ³ q p~ · x̂ 1 X x̂i xˆj ´
Φρe (~x) = + 2 + Qij 3 + · · ·
4π²0 r r 2 i,j r
where Z
q= ρe d3 x
is the traceless quadrupole moment tensor, ρe is the charge F IGURE 9. Force between the two cylindrical magnets as a func-
density, ²0 is the vacuum permitivity and r =| ~x |. Grif- tion of their separation distance a. We show the measured values
fiths [21] presents a nice graphic discussion of the latter ex- (°) and the calculated ones using the method described in the text
pansion. The first term is the monopole contribution to the (solid line).
` U`+
∂Φ(r, θ) 0 L 2
L
−
4
+ 16RL 6
5L8
5 − 128R7 + 256R9
7L10
Bz (r, θ) = −µ0 2R 8R3
∂z L
−R + 2RL3 3L5 5L7 35L9
1 3 − 8R5 + 16R7 − 128R9
2 4 6
315L8
and changed from spherical (r, θ) to cylindrical coordinates 2 3L
− 4R 15L 35L
3 + 16R5 − 32R7 + 256R9
(ρ, z) before performing the integral of Eq. (13) to obtain 3 L
− 4R5L3 35L5
5 + 16R7 − 32R9
105L7
2R3
the vertical force between the two magnets. The expressions 15L 2 4
735L6
4 16R5
− 175L
64R7
+ 128R 9
are too large to be reproduced here, therefore we preferred to 3L 35L3 441L5
5 − 8R5 + 16R7 − 16R9
present the results in a plot. The Mathematica file containing
35L2 735L4
the full expression of the force in the far region (the simplest 6 − 32R 7 + 128R9
3
case) can be downloaded freely ati 7 5L
16R7
− 105L
32R9
Figure 9 shows the force F as a function of a. We can 315L2
8 256R9
see that the theoretical calculation describes fairly well the 9 35L
− 128R 9
experimental behavior. Comparing Figs. 8 and 9, we see that
the largest differences between model and experiment occur
near the boundaries between zones of the potential expan- 6.2. Intermediate region
sion. Apart from these discrepancies, we can say that our
method works adequately. This methodology was used to ` U`+ V`+
2 4 6 8 10 2
calculate the force between a cylinder magnet and a super- 0 R+ L
− 8R L L 5L
3 + 16R5 − 128R7 + 256R9
7L
− R2
2R
conductor in the Meissner state [23]. The interaction between L L3 3L5 5L7 35L9
1 1 − R + 2R3 − 8R5 + 16R7 − 128R9 0
a ring magnet and any other magnet can be studied using the 1 3L2 15L4 35L6 315L8 R4
principle of superposition and a similar approach to the one 2 2R
− 4R 3 + 16R5 − 32R7 + 256R9 8
L 5L3 35L5 105L7
followed in this section. The task of calculating with this 3 2R3
− 4R 5 + 16R7 − 32R9 0
2 4
735L6 6
method the field outside the axis of symmetry for the cone 4 − 8R1 3 + 16R
15L 175L
5 − 64R7 + 128R9 − R16
magnet is left to smart challenge hunters. 3L 35L3 441L5
5 − 8R 5 + 16R7 − 64R9 0
1 35L2 735L4 5R8
6 16R5
− 32R 7 + 128R9 128
3
5L
7 16R7
− 105L
32R9
0
5. Conclusions 5 315L2 10
8 − 128R 7 + 256R9 − 7R
256
35L
An alternative method to calculate the magnetic field at any 9 − 128R 9 0
7 21R12
point in space produced by a permanent magnet with az- 10 256R9 1024
imuthal symmetry is presented. The method is based on a
power series expansion of the magnetic scalar potential value
on the axis of symmetry. The resulting series can be identified 6.3. Far region
as the multipole expansion of the potential. This approach
allowed us to obtain the force between cylindrical magnets U`+ = 0 ∀`
without using the hard-to-manage exact solutions. The mag-
` V`+
netic fields on the axis of a sphere, rectangular prism, cone,
cylinder and ring were calculated. The calculations showed 0 0
LR2
good agreement with the experiments. 1 2
L2 R2
2 2
4L3 R2 −3LR4
3 8
6. Appendix 4 2L4 R2 −3L2 R4
4
8L5 R2 −20L3 R4 +5LR6
5 16
Coefficients of the expansion of the scalar magnetic potential
8L6 R2 −30L4 R4 +15L2 R6
on the symmetry axis for a cylindrical magnet. 6 16
64L7 R2 −336L5 R4 +280L3 R6 −35LR8
7
∞ µ ¶ 128
MX + ` V`+ 8 16L8 R2 −112L6 R4 +140L4 R6 −35L2 R8
Φaxis (z) = U` z + `+1 32
2 z
`=0
i. http://www.mda.cinvestav.mx/magnetic forces/force in the far 12. Z. J. Yang, T. Johansen, H. Bratsberg, G. Helgesen and A. T.
region.nb. Skjeltorp, Supercond. Sci. Technol. 3 (1990) 591.
1. D. Castelvecchi, Sci. Amer. 306 (2012) 37-41. 13. A. Walther, C. Marcoux, B. Desloges, R. Grechishkin, D.
2. L. Holysz, A. Szczes, and E. Chibowski, J. Colloid Interface Givord, N. M. Dempsey, J. Magn. Magn. Mater. 321 (2009)
Sci. 316 (2007) 996-1002. 590-594.
3. X. F. Pang and B. Deng, Physics Mechanics Astron. 51 (2008) 14. M. Sagawa, S. Fujimura, H. Yamamoto, Y. Matsuura, and S.
1621-1632. Hirosawa, J. Appl. Phys. 57 (1985) 4094-4098.
4. R. Cai, H. Yang, J. He, and W. Zhu, J. Mol. Struct. 938 (2009)
15. J. F. Herbst, Rev. Mod. Phys. 63 (1991) 819-898.
15-19.
5. Jr. Stanley Humphries, Finite-element methods for electromag- 16. V. Labinac, N. Erceg, and D. Kotnik-Karuza, Am. J. Phys. 74
netics (2010). Electronic document available at http://www. (2006) 621-627.
fieldp.com/femethods.html 17. N. Derby and S. Olbert, Am. J. Phys. 78 (2010) 229-235.
6. P. Pao-la-or, A. Isaramongkolrak, and T. Kulworawanichpong,
Engineering Lett. 18 (2010). Online document available at 18. L. Lerner, Am. J. Phys. 79 (2011) 1030-1035.
http://www.engineeringletters.com/issues v18/issue 1/EL 18 1 19. G. Akoun and J.-P. Yonnet, IEEE Trans. Magn. 20 (1984) 1962-
01.pdf 1964.
7. R. Ravaud, G. Lemarquand, V. Lemarquand and C. Depollier,
20. H. Allag and J.-P. Yonnet, IEEE Trans. Magn. 45 (2009) 3969-
IEEE Trans. Magn. 44 (2008) 1982-1989.
3972.
8. R. Ravaud, G. Lemarquand, V. Lemarquand and C. Depollier,
PIER B 11 (2009) 281-297. 21. D. J. Griffiths, Introduction to Electrodynamics third edition,
(Prentice Hall 1999).
9. J. D. Jackson, Classical Electrodynamics, third edition, (Wiley,
1998). 22. V. V. Batygin and I. N. Topygin, Problems in Electrodynamics
10. R. H. Jackson, IEEE Trans. Electr. Dev. 46 (1999) 1050-1062. (Academic Press, London 1964), see problem 287.
11. D. J. Kroon, Electromagnets (Boston Technical Publishers Inc., 23. E. Várguez-Villanueva, V. Rodrı́guez-Zermeño, and V. Sosa,
Philips Technical Library, 1968). pp. 15-19. Rev. Mex. Fı́s. 54 (2008) 293-298.