Maxwell's Equations For Magnets: Part II: Realistic Fields

THE CERN ACCELERATOR SCHOOL
Maxwells Equations for Magnets

Part II: Realistic Fields
Andy Wolski
The Cockcroft Institute, and the University of Liverpool, UK
CAS Specialised Course on Magnets
Bruges, Belgium, June 2009
Key results from Lecture 1
In the rst lecture, we saw that multipole elds of the form:
B
y
+iB
x
= C
n
(x +iy)
n1
= C
n
r
n1
e
i(n1)
(1)
with B
z
=constant, provided valid solutions to Maxwells
equations in free space.
We also saw that such a eld could be generated by a current
owing parallel to the z-axis, on a cylinder of radius r
0
, with
distribution:
I() = I
n
cos n(
n
). (2)
In this case, the eld is given by:
B
y
+iB
x
=

0
I
n
2r
0
e
in
n
r
r
0
n1
e
i(n1)
. (3)
Maxwells Equations 1 Part 2: Realistic Fields
Key results from Lecture 1
Multipole elds can also be generated by currents owing in
wires wound around iron poles. For a pure 2n-pole eld, the
shape of the surface of the iron pole must match a surface of
constant scalar potential, , given by:
= |C
n
|
r
n
n
sinn(
n
). (4)
The eld is given by:
B = . (5)
If each pole in an ideal 2n-pole magnet is wound with N
turns of wire carrying current I, then the multipole gradient of
the eld is:
n1
B
y
x
n1
=
n!
0
NI
r
n
0
, (6)
where r
0
is the radius of the largest cylinder that can be
inscribed between the poles.
Goals for Lecture 2
In this lecture, we shall go beyond the idealised geometries and
materials that we assumed in Lecture 1, to consider more
realistic magnets. In particular, we shall:
1. deduce that the symmetry of a magnet imposes constraints
on the possible multipole eld components, even if we relax
the constraints on the material properties and other
geometrical properties;
2. consider dierent techniques for deriving the multipole eld
components from measurements of the elds within a
magnet;
3. discuss the solutions to Maxwells equations that may be
used for describing elds in three dimensions.
Allowed and forbidden harmonics
In general, from equation (1), a pure 2n-pole eld can be
written:
B
y
+iB
x
= |C
n
|e
in
n
r
n1
e
i(n1)
, (7)
where r and are the polar coordinates within the magnet, and
n
is the angle by which the magnet is rolled around the z axis.
We cannot design a realistic magnet to produce a pure 2n-pole
eld. The materials will have nite permeability and nite
dimensions, and may saturate.
More generally, a eld consists of a superposition of 2n-pole
elds:
B
y
+iB
x
=
n=1
|C
n
|e
in
n
r
n1
e
i(n1)
. (8)
However, in the design of a 2n-pole magnet, we can impose a
perfect symmetry under rotations through 2/n about the z
axis.
In fact, we see from equation (7):
B
y
+iB
x
= |C
n
|e
in
n
r
n1
e
i(n1)
, (9)
that under a rotation by /n, i.e.
n
n
+/n, the eld
changes sign:
B. (10)
Since this symmetry is imposed by the geometry of the
magnet, any multipole eld within the magnet must obey this
symmetry. This restricts the multipole components that may
be present, at least in the design.
Consider a eld given by:
B
y
+iB
x
= |C
n
|e
in
n
r
n1
e
i(n1)
+|C
m
|e
im
m
r
m1
e
i(m1)
.
(11)
If the geometry of the magnet is such that the eld simply
changes sign under a rotation through /n, then the extra
harmonic must satisfy:
e
i
m
n
= 1, (12)
therefore:
m
n
= 1, 3, 5, 7 . . . (13)
We see that only higher harmonics are allowed; and the indices
of the higher harmonics must be an odd integer multiple of the
main harmonic.
For a dipole, n = 1, so the allowed harmonics are m = 3, 5, 7. . .
For a quadrupole, n = 2, and the allowed harmonics are
m = 6, 10, 14. . .
Of course, this argument cannot tell us the strengths of the
allowed harmonics: those depend on the details of the design.
It is also a fact that a magnet, when fabricated, will never
exhibit perfectly the symmetry with which it was designed.
Therefore, a real physical magnet will generally include all
harmonics to some extent, not just the harmonics allowed by
the ideal symmetry.
However, for a carefully fabricated magnet, the harmonics
forbidden by the ideal symmetry should be small in comparison
to the allowed harmonics; and it should be possible to predict
the sizes of the allowed harmonics accurately from the design.
Measuring multipoles
Knowing the multipole components of magnets in an
accelerator is important for understanding the beam dynamics.
Construction tolerances will mean that the strengths of the
multipoles present in the magnet will dier from those in the
design.
This leads us to consider how to determine the multipole
components from measurements of the magnetic eld.
There are many possible approaches to the problem: we shall
consider two, for illustration, and only go as far as necessary to
understand some of the pros and cons.
Measuring multipoles 1: Cartesian basis
The eld may be represented as:
B
y
+iB
x
=
n
C
n
(x +iy)
n1
, (14)
where the real parts of the coecients C
n
give the normal
multipole strengths, and the imaginary parts give the skew
multipole strengths.
If we take a set of measurements of B
y
and B
x
along the x
axis, then y = 0 at each measurement point, and the normal
multipoles can be found by tting a polynomial to B
y
vs x, and
the skew multipoles can be found by tting a polynomial to B
x
vs x.
Let us see how this works with some invented data...
We construct a quadruople eld with higher harmonics 3, 4 and
5, and t a quartic to the eld data:
n actual value tted value
2 1.000 1.000
3 0.010 0.010
4 0.001 0.001
5 0.010 0.010
So far so good. But what happens if we include a higher-order
multipole that we do not t?
Now we add a higher harmonic n =6, but still t the eld data
with a quartic:
n actual value tted value
2 1.0000 0.9972
3 0.0100 0.0100
4 0.0010 0.0131
5 0.0100 0.0100
6 0.0100 -
We now have signicant errors on the n = 2 and n = 4
multipole eld strengths.
The reason for the discrepancy perhaps starts to become clear
if we plot the deviation from the nominal quadrupole eld.
With only components present up to n = 5 (left hand plot), the
t is essentially perfect. But with a higher component (n = 6,
right hand plot) that we do not t, there are signicant
residuals.
The real problem is that mathematically, the basis functions
that we use to t the data (monomials in x and, possibly, y)
are not orthogonal. This means that data constructed from
one monomial can be tted, with non-zero strength, with a
completely dierent monomial.
Although this does not invalidate the technique altogether, it
does make it a little dicult to apply accurately. Ideally, we
need to know in advance which multipole components are
present.
However, there is a more robust technique...
Measuring multipoles 2: Polar basis
Instead of expressing the eld in Cartesian coordinates, let us
write the same eld in polar coordinates:
B
y
+iB
x
=
n
C
n
r
n1
e
i(n1)
. (15)
Now suppose that we take a set of measurements of B
y
and B
x
at xed r = r
0
, but at M equally spaced steps in = 2m/M,
where m = 1. . . M.
We then notice that:
M
m=1
(B
y
+iB
x
)
m
e
2i(n
1)
m
M
=
M
m=1
n
C
n
r
n1
0
e
2i(nn
)
m
M
= MC
n
r
n
1
0
. (16)
Hence:
C
n
=
1
Mr
n1
0
M
m=1
(B
y
+iB
x
)
m
e
2i(n1)
m
M
. (17)
The advantage of using polar coordinates, over Cartesian
coordinates, is that the basis functions, e
i(n1)
, are
orthogonal. Mathematically, we have:
M
m=1
e
2i(n1)
m
M
e
2i(n
1)
m
M
=
0 if n = n
M if n = n
(18)
The orthogonality means that the value we determine for one
multipole component is completely unaected by the presence
of other multipole components.
Determining the multipole coecients C
n
amounts to carrying
out a discrete Fourier transform on the eld data measured on
a cylinder inscribed through the magnet.
A further advantage of the polar basis comes from the fact
that the radius of the cylinder on which the eld data are
collected appears as 1/r
n1
0
in the expression for the
coecients C
n
, equation (17).
Suppose that there is some error in the eld measurements.
This will lead to some error in the values of C
n
that we
determine. If we reconstruct the eld (e.g. for particle
tracking), then there will be some error in the calculated eld.
However, this error will decrease as r
n1
, as we go towards the
centre of the magnet (where the beam is).
Of course, if we try to extrapolate the eld outside the cylinder
of radius r
0
, then any errors will increase as some power of the
distance from the centre.
Advantages of mode decompositions
There are some important advantages to describing a eld in
terms of a mode decomposition, instead of a set of numerical
eld values on a grid:
A description of the eld in terms of mode coecients is
very much more compact than a description in terms of
numerical eld data.
A eld constructed from mode coecients is guaranteed to
satisfy Maxwells equations: numerical eld data are not.
Measurement noise can be smoothed by suppressing
higher-order modes.
Errors can be represented in a realistic way by introducing
higher-order modes.
A number of beam dynamics analysis tools require mode
decompositions.
Three-dimensional elds
The polar basis for tting multipole eld components
generalises nicely to three-dimensional elds. But, since we
have not so far discussed such elds at all, before showing how
the eld tting works, we need to discuss solutions to
Maxwells equations for three-dimensional magnets.
As before, the relevant equations are:

B = 0, and

B = 0. (19)
Any eld that satises these equations is a possible magnetic
eld in free space. So far, we have considered only multipole
elds, that are independent of one coordinate; but this is not
very realistic.
A eld that satises Maxwells equations (19) is given by:
B
x
= B
0
k
x
k
y
sink
x
x sinhk
y
y sink
z
z, (20)
B
y
= B
0
cos k
x
x coshk
y
y sink
z
z, (21)
B
z
= B
0
k
z
k
y
cos k
x
x sinhk
y
y cos k
z
z, (22)
where:
k
2
y
= k
2
x
+k
2
z
. (23)
There are a number of variations on this eld, for example, with
the hyperbolic function appearing in the x or z coordinates; or,
with dierent phases in x and/or z. However, the above
representation is particularly convenient for describing insertion
devices (wigglers and undulators), as we shall now discuss.
B
x
= B
0
k
x
k
y
sink
x
x sinhk
y
y sink
z
z,
B
y
= B
0
cos k
x
x coshk
y
y sink
z
z,
B
z
= B
0
k
z
k
y
cos k
x
x sinhk
y
y cos k
z
z,
Normal-conducting electromagnetic wiggler at the KEK Accelerator Test
Facility.
If we take k
x
= 0, then the eld becomes:
B
x
= 0, (24)
B
y
= B
0
coshk
z
y sink
z
z, (25)
B
z
= B
0
sinhk
z
y cos k
z
z. (26)
The above equations describe a eld that varies sinusoidally in
z, and has no horizontal (x) component at all. This is a eld
that could only occur in an insertion device with innite length,
and innite width.
A better description would take account of the fact that the
longitudinal variation of the eld will not be perfectly
sinusoidal. We can account for this by superposing elds with
dierent values of k
z
:
B
x
= 0, (27)
B
y
=

B(k
z
) coshk
z
y sink
z
z dk
z
, (28)
B
z
=

B(k
z
) sinhk
z
y cos k
z
z dk
z
. (29)
We see that if we take measurements of B
y
as a function of z
in the plane y = y
0
(for xed y
0
), then we can obtain the mode
coecients

B(k
z
) by a (discrete) Fourier transform.
This allows us to reconstruct all eld components, at all
locations within the eld.
Suppose we measure B
y
at a set of locations z = z
, where:
z
=

N
z
max
, = N, N +1, N +2, . . . , N. (30)
N is an integer, and the eld has eectively fallen to zero at
z = z
max
.
We obtain the mode coecients from:
B
n
=
1
2N cosh(nk
z
y
0
)
N
=N
B
y
(z
) sin(nk
z
z
) , (31)
where k
z
= /z
max
.
And the eld at any point can be reconstructed from:
B
y
(y, z) =
N
n=N
B
n
cosh(nk
z
y) sin(nk
z
z) , (32)
B
z
(y, z) =
N
n=N
B
n
sinh(nk
z
y) cos (nk
z
z) . (33)
Let us see what happens in a simple example again, using
invented data. The left hand plot shows the eld data
(points), and the reconstructed eld (red line). The right hand
plot shows the mode coecients, from which the
reconstruction is obtained. Measurements are made along
y = 0.25.
Mathematically, the reconstructed eld must t the data
exactly. However, the interpolation between the data points is
only reliable if the mode coecients are small for large mode
number.
If the data are too sparse, then we can determine only a small
number of modes. The previous plots were constructed from
80 measurement points. What happens if we reduce the
number of measurement points to 30?
The data are still tted exactly, but there is signicant wiggle
between the data points, where the eld is really close to zero.
Using our mode coecients, we can reconstruct the eld along
the z axis:
Note that the eld extends out further in z: the fringe eld
increases as we approach the mid-plane. Also note that the
wiggles beyond the fringe eld have been smoothed out: this
is a result of the hyperbolic dependence of the eld on y. If we
had tted the data on y = 0, the residuals would have
increased exponentially with increasing y.
If we look at the eld variation with vertical position (y), we see
that the eld variation is not sensitive to the number of modes:
Again, this is a result of the hyperbolic dependence of the eld
on the vertical coordinate.
However, if we try to extrapolate to y > y
0
, where y = y
0
is the
line of the t, then unpredictable behaviour can result...
Of course, the above example still describes a eld that is
independent of one of the coordinates. However, generalisation
to include a dependence on x is straightforward.
Generally, from (21), we can write the vertical eld component
as:
B
y
=
B
mn
cos mk
x
x coshk
y
y sinnk
z
z, (34)
where:
k
2
y
= m
2
k
2
x
+n
2
k
2
z
. (35)
The mode coecients

B
mn
(note the double index) can be
obtained from a 2D discrete Fourier transform on a plane of
xed y. As in the example, it is benecial to choose the plane
y = y
0
so that the planes y = y
0
bound the region of interest.
Outside this region, the elds will diverge.
Another issue with this choice of basis functions is the value of
the horizontal parameter k
x
= /x
max
.
Ideally, one will make measurements over a suciently wide
region that the eld drops to zero at x = x
max
.
This may be possible in principle for some simple wigglers and
undulators, but may not always be practicable, even in these
cases.
An alternative approach, and one that is preferable in many
situations, is to use a cylindrical basis for the mode
decomposition.
Cylindrical basis functions for three-dimensional elds
In cylindrical polar coordinates, a eld satisfying Maxwells
equations can be represented by:
B
dk
z
B
m
(k
z
)I
m
(k
z
) sinm cos k
z
z, (36)
B
dk
z
B
m
(k
z
)
m
k
z
I
m
(k
z
) cos m cos k
z
z, (37)
B
z
=
dk
z
B
m
(k
z
)I
m
(k
z
) sinm sink
z
z. (38)
Here, the functions I
m
(r) are modied Bessel functions: broadly
speaking, they are to regular Bessel functions as hyperbolic
trigonometric functions are to regular trigonometric functions.
The mode coecients

B
m
(k
z
) may be obtained, for example,
by a 2D discrete Fourier transform of the radial eld
component B
on the surface of a cylinder.

We can draw a direct connection between the 3D polar basis,
and the multipole decomposition in 2D. We use the fact that
the modied Bessel functions have an expansion (for small ):
I
m
() =

m
2
m
(1 +m)
+O(m+1). (39)
Therefore, if the mode coecients are given by:
B
m
(k
z
) = 2
m
(1 +m)C
m
(k
z
)
mk
m1
z
, (40)
where () is the Dirac delta function, then:
B
m
C
m
m1
sinm, (41)
B
m
C
m
m1
cos m, (42)
B
z
= 0. (43)
Then, we note that:
B
x
= B
cos B
sin, (44)
B
y
= B
sin +B
cos , (45)
to nd:
B
x
=
m
C
m
m1
sin(m1), (46)
B
y
=
m
C
m
m1
cos(m1). (47)
Hence:
B
y
+iB
x
=
m
C
m
(x +iy)
m1
. (48)
This is the familiar form for a multipole eld. Since the original
coecients

B
m
are real, it follows from (40) that the
coecients C
m
are also real, so equations (36)-(38) represent a
3D normal multipole. A skew multipole is just obtained by a
rotation through by /2.
The connection between the 3D mode decomposition in the
cylindrical basis and the usual representation of a multipole
eld gives us a nice interpretation of the mode coecients

B
m
.
The coecient

B
m
(k
z
) is essentially the longitudinal Fourier
amplitude of the 2m-pole eld in the magnet. We should
remember, of course, that multipole elds only really exist in
innitely long, uniform magnets. The representation (36)-(38),
however, can be used to describe realistic, 3D elds.
We mentioned above that the mode coecients

B
m
(k
z
) may
be obtained by a 2D discrete Fourier transform of the radial
eld component B
on the surface of a cylinder.

As usual, it is benecial to choose the radius of this cylinder to
be as large as possible: inside the cylinder, errors in the t
decrease exponentially; outside the cylinder, the errors increase
exponentially.
With good eld data, it is possible to achieve a very good t.
As an example, the following slides show ts to a model for a
permanent magnet wiggler for the TESLA damping ring. The
(nominal, on-axis) peak eld is 1.7 T, and the wiggler period is
0.4 m. The ts were obtained using 18 azimuthal and 100
longitudinal modes, on a cylinder of radius 9 mm.
Field t to TESLA damping wiggler
Field t to TESLA damping wiggler: residuals
Final remarks
Often, the apertures in accelerator magnet (particularly, in
insertion devices) are not circular: the horizontal aperture is
wider than the vertical. In such cases, a good t can be
obtained over a wider region by using an elliptical cross-section
for the cylinder, rather than a circular cross-section.
It is possible to perform an inverse Fourier transform on the
mode coecients in the longitudinal dimension z, while
retaining the mode decomposition in . Then, we obtain a
representation in which the multipole component appears as
a function of z. This approach leads to the idea of generalised
gradients.
For further information on both these topics, see the book by
Alex Dragt:
Lie methods for nonlinear dynamics with applications to
accelerator physics.
http://www.physics.umd.edu/dsat/dsatliemethods.html
Summary of Lecture 1
Maxwells equations impose strong constraints on magnetic
elds that may exist.
The linearity of Maxwells equations means that complicated
elds may be expressed as a superposition of simpler elds.
In two dimensions, it is convenient to represent elds as a
superposition of multipole elds.
Multipole elds may be generated by sinusoidal current
distributions on a cylinder bounding the region of interest.
In regions without electric currents, the magnetic eld may be
derived as the gradient of a scalar potential.
The scalar potential is constant on the surface of a material
with innite permeability. This property is useful for dening
the shapes of iron poles in multipole magnets.
Summary of Lecture 2
Symmetries in multipole magnets restrict the multipole
components that can be present in the eld.
It is useful to be able to nd the multipole components in a
given eld from numerical eld data: but this must be done
carefully, if the results are to be accurate.
Usually, it is advisable to calculate multipole components using
eld data on a surface enclosing the region of interest: any
errors or residuals will decrease exponentially within that
region, away from the boundary. Outside the boundary,
residuals will increase exponentially.
Techniques for nding multipole components in two
dimensional elds can be generalised to three dimensions,
allowing analysis of fringe elds and insertion devices.
In two or three dimensions, it is possible to use a Cartesian
basis for the eld modes; but a polar basis is sometimes more
convenient.
Appendix: The vector potential
In lecture 1, we used a scalar potential for the magnetic eld to
derive the shape for the pole face of a multipole magnet.
The scalar potential is dened such that:
B = . (49)
With this denition, the equation

B = 0 is automatically
satised. The equation

B = 0 leads to Laplaces equation
for the scalar potential:
2
= 0. (50)
However, the scalar potential is only dened in the absence of
currents. More generally, we need to use a vector potential

A.
In fact, in the most general case of time-dependent electric and
magnetic elds, we need both a scalar potential , and a vector
potential

A:
B =

A, and

E =

A
t
. (51)
Some important methods for beam dynamics analysis use the
potentials and

A, rather than the elds. It is therefore useful
to have expressions for the potentials corresponding to the
expressions for the elds we have derived in the main part of
these lectures.
For the case of interest here (a magnetostatic eld, and zero
electric eld), we can assume that is independent of position,
and

A is independent of time.
If we allow the presence of nonmagnetic materials ( =
0
)
carrying an electric current density

J, then substituting from
(51) into Maxwells equations gives:

B =

A 0, (52)
and:

B =

A (

A)
2

A =
0
J. (53)
Maxwells equation

B = 0 is automatically satised by any
vector potential

A, by virtue of a vector identity (the
divergence of the curl of any vector eld is zero).
Maxwells equation

H =

J (assuming static elds) is
satised, if the vector potential

A satises:
2

A(

A) =
0
J. (54)
Now, we observe that since

B =

A, and is
identically zero for any scalar eld , we can dene a new
potential

A
=

A+ that gives exactly the same eld as

A.
We can use this property of the elds and potentials, known as
gauge invariance, to simplify equation (54).
Suppose that we have a vector potential

A for which:

A = f, (55)
where f is some function of position. Then, if we dene:
=

A+, (56)
where satises Poissons equation:
2
= f, (57)
then

A
gives the same eld

B as

A, and:
=

A+
2
= 0. (58)
In other words, if we can solve Poissons equation (57) for ,
then we can make a gauge transformation to a vector potential
that has vanishing divergence.
Let us suppose that we nd a vector potential that has
vanishing divergence:

A = 0. (59)
Equation (59) amounts to a condition that species a
particular choice of gauge: this particular choice (i.e. with zero
divergence) is known as the Coulomb gauge. It is useful,
because equation (54) for the vector potential then takes the
simpler form:
2

A =
0
J. (60)
This is Poissons equation, which has the standard solution:
A(r) =
0
4

J(r
)
|r r
|
d
3
r
. (61)
Now, consider the potential given by:
A
x
= 0, A
y
= 0, A
z
= Re
C
n
(x +iy)
n
n
. (62)
Taking derivatives, we nd that:
A
z
x
= Re C
n
(x+iy)
n1
, and
A
z
y
= ImC
n
(x+iy)
n1
. (63)
Hence:
B =

A,
=
A
z
y
,
A
z
x
, 0
,
=
ImC
n
(x +iy)
n1
, Re C
n
(x +iy)
n1
, 0
. (64)
Therefore, we have:
B
y
+iB
x
= C
n
(x +iy)
n1
,
which is just the multipole eld (1).
We have shown that the potential:
A
x
= 0, A
y
= 0, A
z
= B
ref
n=1
(b
n
+ia
n
)
Re(x +iy)
n
nR
n1
ref
, (65)
gives the multipole eld (1):
B
y
+iB
x
= B
ref
n=1
(b
n
+ia
n
)
x +iy
R
ref
n1
.
Note also that:

A =
A
x
x
+
A
y
y
+
A
z
z
= 0, (66)
so this potential satises the Coulomb gauge condition.
Note that the longitudinal eld component derived from the
multipole potential (65) is:
B
z
=
A
y
x
A
x
y
= 0. (67)
In order to generate a solenoidal eld, with B
z
= constant = 0,
we need to introduce non-zero components in A
x
, A
y
, or both.
For example:
A
x
=
1
2
B
sol
y, A
y
=
1
2
B
sol
x. (68)
While it is convenient, for beam dynamics, to work in a gauge
with only the z component of the vector potential non-zero,
this is not possible for solenoidal elds.
Finally, we give the vector potentials corresponding to 3D
elds. In the Cartesian basis (20)-(22):
B
x
=
B(k
x
, k
z
)
k
x
k
y
sink
x
x sinhk
y
y sink
z
z,
B
y
=

B(k
x
, k
z
) cos k
x
x coshk
y
y sink
z
z,
B
z
=

B(k
x
, k
z
)
k
z
k
y
cos k
x
x sinhk
y
y cos k
z
z,
a possible vector potential is:
A
x
= 0, (69)
A
y
=

B(k
x
, k
z
)
k
z
k
x
k
y
sink
x
x sinhk
y
y cos k
z
z, (70)
A
z
=
B(k
x
, k
z
)
1
k
x
sink
x
x coshk
y
y sink
z
z. (71)
In the Polar basis (36)-(38):
B
=

B
m
(k
z
)I
m
(k
z
) sinm cos k
z
z,
B
=

B
m
(k
z
)
m
k
z
I
m
(k
z
) cos m cos k
z
z,
B
z
=
B
m
(k
z
)I
m
(k
z
) sinm sink
z
z,
a possible vector potential is:
A
m
I
m
(k
z
) cos m sink
z
z, (72)
A
= 0, (73)
A
z
=
m
I
m
(k
z
) cos m cos k
z
z. (74)

Maxwell's Equations For Magnets: Part II: Realistic Fields

Uploaded by

Copyright:

Available Formats

Maxwell's Equations For Magnets: Part II: Realistic Fields

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Maxwell's Equations For Magnets: Part II: Realistic Fields

Uploaded by

Copyright:

Available Formats

THE CERN ACCELERATOR SCHOOL

Maxwells Equations for Magnets

on the surface of a cylinder.

on the surface of a cylinder.

gives the same eld

You might also like