View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Juelich Shared Electronic Resources

Computer Physics Communications 72 (1992) 22—39 Computer Physics

North-Holland Communications

Solution of Maxwell’s equations

Michael Bartsch a, Micha Dehier a, Martin Dohlus a, Frank Ebeling a, Peter
Hahne a, Reinhard Klatt a, Frank Krawczyk a, Michaela Marx b, Zhang Mm a,
Thomas Pröpper a, Dietmar Schmitt a, Petra Schütt a, Bernhard Steffen C~
Bernhard Wagner a, Thomas Weiland a, Susan G. Wjpfb and Heike Wolter a
a Technische Hochschule Darmstadt, Fachbereich 18, Fachgebiet Theorie Elektromagnetischer Felder,
SchloJigartenstraj3e 8, W-6100 Darmstadt, Germany
b Deutsches Elektronen-Synchrotron DESY, NotkestraJ3e 85, W-2000 Hamburg 52, Germany
C Kernforschungsanlage Julich KFA, W-5170 .Jülich, Germany

A numerical approach for the solution of Maxwell’s equations is presented. Based on a finite difference Yee lattice the
method transforms each of the four Maxwell equations into an equivalent matrix expression that can be subsequently
treated by matrix mathematics and suitable numerical methods for solving matrix problems. The algorithm, although
derived from integral equations, can be consideredto be a special case of finite difference formalisms. A large variety of
two- and three-dimensional field problems can be solved by computer programs based on this approach: electrostatics
and magnetostatics, low-frequency eddy currents in solid and laminated iron cores, high-frequency modes in resonators,
waves on dielectric or metallic waveguides, transient fields of antennas and waveguide transitions, transient fields of
free-moving bunches of charged particles etc.

1. Introduction

The field ofaccelerator physics largely deals with controlled application ofelectromagnetic forces to
charged particles. These forces occur in various parts of an accelerator at differentlevels of complexity:
Magnetostatic and electrostatic fields are used to steer, focus and accelerate particle beams. RF-fields
in metallic resonators are the most common type ofaccelerating devices. CW-transmitters in the UHF
range of more than 1 MW output power are used in large numbers, pulsed tubes are being built up
to 100 MW in the S-band. Transient fields are excited by charged particles when passing accelerator
structures and cause difficult nonlinear problems (collective effects).
In the field of elementary particle physics accelerators as large as 27 kilometers in circumference are
commissioned and plans for machines beyond 180 kilometers have been presented. The cost of these
front-line accelerators is enormous. Thus it is extremely important to ensure, prior to construction,
that such a device will work as planned. In order to predict the behaviour of charged-particle beams in
accelerators, one has to solve a variety offield problems for realistic structures such as those mentioned
above. This has occasioned the developement of algorithms and program systems, which have already
proved dependable and useful in many areas of physics and engineering.

Correspondence to: T. Weiland, Technische Hochschule Darmstadt, Fachbereich 18, Fachgebiet Theorie Elektromagnetischer
Felder, Schlol3gartenstraBe 8, W-6100 Darmstadt, Germany

0110-4655/92/s 05.00 © 1992 — Elsevier Science Publishers B.V. All rights reserved
 M. Bartsch et a!. / Solution of Maxwell’s equations 23

2. The method

In order to avoid specializations of Maxwell’s equations prior to numerical solution it is advanta-

geous to solve Maxwell’s equations directly, rather than solving a partial differential equation denved
therefrom. Using SI units, E and H for the electric and magnetic field strength, D and B for the flux
densities and .1 for the current density the equations to solve read

y~E.ds=-ff~.dA~ (1)

y~H.ds=ff(~/~+J).dA~ (2)

ffB.dA=O~ (3)

jf(~+J).dA=o~ (4)

with the following relations:

D = iE, B = uH, I = ,cE + pv. (5,6,7)

A grid G is defined in the orthogonal coordinate system r = r (u, v, w) as

G={(uI,vJ,wk);ul<uI<uJ, i=2,...,I—l;v1<v3<vj,j=2,...,J--l;

Wl<Wk<WK,k=2,...,Kl}. (8)

All nodes of G are numbered linearly by

n = 1 + (i— l)M~+ (I— l)M~+ (k— l)M~, (9)

n = l,..,N = IJK. (10)

Usually one uses M~= 1, M~ = I, M,~,= IJ or any permutation of these three assignments. Figure
1 shows such a three-dimensional grid.
The functions which describe the field distribution in the volume under consideration are replaced
by discrete field values in each cell of the grid. Thus one has to allocate components to grid nodes,
lines, areas or elementary volumes. We will not choose the obvious allocation of calculating all six
field components at the grid nodes. Such an allocation would cause serious problems at surfaces of
materials where some of the field components are not continuous. Instead, the components of E are
placed at the mid-points of the edges of the grid cells and the components of B at the centre of each
face, as shown in fig. 2, extending the Yee Algorithm [1] to more general field problems [2—4].
This kind of allocation has the distinct advantage that the transition from one cell to the next
involves only continuous components, tangential E fields and perpendicular B fields (see fig. 3);
thus Maxwell’s equations are always satisfied, in this respect, even when different material fillings are
involved. We now apply a first-order integration formula in order to approximate the left hand side
24 M. Bartsch et a!. / Solution of Maxwell’s equations

UI U
2 U1 U1

/
Fig. 1. Three-dimensional grid in the orthogonal not necessarily Cartesian system (u, v, w).

E~ ,~ +

v, n +

/)J /
~bV/

Fig. 2. Allocation of unknown field components in the grid G showing the indexing.

of eq. (1) and obtain the following algebraic expression replacing the integral along the border of an
elementary cell:

E. ds = + (u~+i— ~ + (v~+i vj)Ev,n+M~ (u1+i

— — — ui)Eu,n+M~

~ (11)
 M. Bartsch et a!. / Solution of Maxwell’s equations 25

~B7~{E}

{~
Fig. 3. Two neighbouring mesh cells showing the allocation of the electric and magnetic field components in G and G.

For the right hand side of that equation, we obtain (again by the lowest-order integration formula):

ff ~.dA~W,fl(uj+l_uj)(vj+l_vj)+O([(uj+1_uj)(vj+1_vj)]2). (12)

Equating the two expressions yields a discrete replacement for the first Maxwell equation on each
surface of the grid cells and thus on every area composed of mesh cell surfaces.
In order to describe all these equations for all surfaces we introduce a basic discretization matrix
with only two bands and with elements taking only the values 0, +1 or —1:

(—1, p=m,
Pu:(Pu)m,p_~+l, p_m+M~, m,p=1,...,N. (13)
(0, otherwise,

Out of such simple matrices we combine the matrix C that replaces the contour integral operator in
eq. (1) for all mesh points:

0 —P~ P~ \
c=( Pw 0 —Pa). (14)
\—P~ Pu 0 J
All unknown components of the electric field E are put into a vector of dimension 3N = 3 (JfK):

e = ~ . .. ,Ew,N)t (15)

and we replace D by d, B by b, H by h and J by j, respectively. The topological part of the contour

integral (curl) is represented by the matrix C. The actual length of the integration path is put into a
diagonal matrix:

= Diag(Aul,...,AuN,~v1,...,~vN,i~w1,...,z~wN), (16)
26 M. Bartsch et a!. / Solution of Maxwell’s equations

ui+
1

~= {f~Oru~vJ~wk~dU
1~i<I—l,1<j<J,l<k<K (17)

Av~and z~w~
are defined accordingly. The surfaces of the grid cells compose the diagonal matrix:

DA = Diag(AU1,...AU,N,AVI,...,AVN,AWI,...,AWN), (18)
= L~v,4w~,~ = L~w,4u~,~ = ~ (19,20,21)

Finally we can give a matrix equation using all these definitions that replaces the first Maxwell
equation on all cell surfaces:

CD5e = —DAb (22)

The special allocation of field components in the grid generates a second, dual grid G in which the
magnetic flux densities are allocated as are the electric field components in the original grid G, see fig.
3.
The dual grid is defined by

G={(ü~,v~j,w~k):ül<üI<üJ,
i=2,...,I—l;~1<~j<~j,j=2,...,J—l;
tO1 <Wk <WK, k = 2,...,K— l}, (23)
üt=(u1~1+u1)/2, l~iI—l,üj=0, (24)
2, 1 <j < J 1, vY~ = 0, — (25)
= (v1÷1+ v~)/
Wk=(Wk+1+Wk)/2, l<k<K—l,ü~K=0. (26)

We can also define matrices that hold the mesh step sizes and cell areas as for the original grid G:

= Diag(i~u
1,.~ .. ~ ..,/XWN), (27)
1(~un+~un_M~)/2,2i~I—l,
= ~ iSu~/2, i = 1, (28)
(I~UflM~/2, I = I,

DA = Diag(A~,1,. .,Au,N,A~,l,...,Av,N,A~,l,.. .,Aw,N),

. (29)
A~,, = z~v~
Awn, ~ = i~w~
~ ~ = ~ ~ (30,31,32)

When solving the second Maxwell equation in a way similar to the first one, we have to face the fact
that the magnetic field H is not defined and thus has to be calculated from the flux density piecewise
along the integration path. Also, the integration over the electric flux density is no longer a simple
product of an area with a component but the sum over four parts on each of which the flux has to be
determined from the defined E and the material constants. Finally we obtain:

CD5D~’b= DA(d + j). (33)

Note that the contour integral operator on the dual grid, C, (curl) is related to C simply by
 M. Bartsch et a!. / Solution of Maxwell’s equations 27

C = Ct. (34)

The diagonal matrices which take care of the piecewise different values for the material constants are

given as
(D~)~
= 1~Un_M~+ ~~!L), (35)
2 2!Lu,n
1Uu,fl.M~
(D )2N+n = (AW,flMUEW,fl..MU + AW,n_MVEW,n_MV

+ ~ + AW,nMU_MV ~w,nM~M~ ) / (4Aw,n), (36)

(D,c)2N+n +
= (Aw,n_M~”~w,n_M~ AW,n_MVKW,n_MV

+ ~ + AW,fl_MU..MVICW,fl_MU_MV)/(4AW,fl), (37)

while ~, and ic may have different properties in the directions u, v and w. Although not originally
defined in the grid, one may relate the field strength, flux density and current density and formally
define a magnetic field and an electric flux density by

b = D,~h (B = huH), (38)

d = D~e (D = EE), (39)
j = D~e+ D,,v (I = ,cE + pv). (40)

In order to complete the transformation of Maxwell’s equations to the grid space, we approximate
the divergence equation by integrating B over surfaces of each mesh cell of G. We can then write this
equation by defining the discrete div operator on G as

S (Pu,Pv,Pw) (41)

and obtain

SDAb = 0. (42)

This equation allocates the (nonexistent) magnetic charges to nodes of G. The corresponding matrix
on the dual grid G is simply given by

t t
~—‘Pt
— “ U, tJ’ W -

So the continuity equation in the grid space reads

SDA(d+i)=O (44)

Thus electric charges are defined on nodes of the original grid G.

28 M. Bartsch et a!. / Solution of Maxwell’s equations

3. Properties of the grid equations

One of the outstanding properties of Maxwell’s grid equations is that properties of analytical solu-
tions have their exact analogues in th~grid space. The analytical identity div curl 0 reads in the grid
space as

div curl 0 ~ SC SC 0. (45)

This identity in grid space can easily be proved by showing that:

SC = (0,0,0) = (PVFW ~ ~ —P~P~). (46)

Thus the proof reduces to the equivalence of interchanged partial differentiation:

= P~P~ ~
ía a =
a a~
~—~---). (47)
This matrix identity can easily be verified. The analytical property that scalar potential fields are curl
free is also found in the grid space as

curl grad 0 ~ CtSt 0. (48)

The proof of this identity is found by simply transposing the identity (div curl 0) eq. (45). Thus
the source-freeness of the “curly” field is in that sense equivalent to the fact that the scalar gradient
field is irrotational.
These properties ofthe Maxwell grid equations not only offer a unique tool to test numerical results
for their physical- correctness but also avoid the occurrence of incorrect solutions in the calculation of
three-dimensional eigenmodes in resonators, or at least their identification [3]. This is quite important
since numerical errors in the matrix algorithms may occur and cannot generally be detected. In the
field of accelerator physics, one often investigates unknown phenomena by using such codes, so that
an incorrect solution could lead to physical misinterpretations.

4. The Maxwell grid equations

Without specifying anything about the shape of materials nor the time dependence of the fields, we
have obtained a set of matrix equations that approximate Maxwell’s equations on a double grid G, G
(R3 and ~ represent the physical spaces and IFL~. the time):

Real space EJ~3® ~ Grid space R3N ®

jE. ds = j7~.dA ~ CDe = -DAb, (49)

~H.th=ff(~+I).dA ~ CD
5h=DAd+j, (50)
 M. Bartsch et a!. / Solution ofMaxwell’s equations 29

ffB.dA=0 ~ SDAb=0, (51)

(52)

D=E ~‘ d=D~e, (53)

B=uH ~* b=D,~h, (54)
J=KE+pV ~ j=D,e+D~,v, (55)
div curl 0 ~ SC = SC = 0, (56)
curl grad 0 ~. C~S~
= CtSt = 0. (57)

5. Special cases of Maxwell equations

In order to solve these equations one has only to perform matrix manipulations and then solve the
established matrix problem numerically.

5.1. Static fields

In the caseof static fields it is in general not necessary to describe the problem by vectors. Gradients

of a scalar potential are used to derive the Poisson equation:

E = —grad~E, (58)
div(Egrad~E)= —p. (59)

Instead of numerically solving Maxwell’s equation we define the electric potentials on nodes of G as

= (~E,1,~E,2,Q
5E,3,. - . ~ (60)

and derive the corresponding equations only by matrix manipulation:

e = StD~løe, (61)
SDDADs1St~e = —q. (62)

The latterequation is ofthe order N and is the “Grid Potential Equation”. The right hand side contains
all charges on the nodes in the vector q of dimension N.
For the magnetostatic field the procedure is similar except that the “curly” part ofthe magnetic field
has to be taken out in order to allow the use of a scalar potential:

H = H~ gradq~H,
— (63)
curlH~= I, (64)
div(
1ugradq~H)= div(itH~). (65)
30 M. Bartsch et a!. / Solution of Maxwell’s equations

Here the right hand side is given by “magnetic charges” which result from the curly field but have
otherwise no physical meaning. Equation (65) is equivalent to (59) where ~u,H and div (~uH~) corre-
spond to e, E and p. One obtains the same matrix equations by allocating H components on the grid
and calculating D~similarly to DC:

h = — StDl~h (66)
CDShC = D~,j, (67)
SD,~DAD~’St~h
= SD~DAhc. (68)

This allows the use of the same iterative algorithm for solving electro- and magnetostatic problems.

5.2. Time harmonic and resonant fields

The electric fields simultaneously have to obey the following two equations, that can be directly

derived from Maxwell’s equations

curl -~- curiE = w2e E — ioncE — iwl, (69)

divE= -1-(div,cE+divj). (70)

The numerical formulation can be derived again by matrix manipulation:

(CD 2DADC )e = -iWDADPV, (71)

5D~’D~CD5+ ~WDAD,C W -

SDADCe = (SDADKe + ~IiADPv). (72)

Equation (71) represents a 3N x 3N equation system, which can usually be solved unless the stimu-
lation frequency w corresponds to a system resonance frequency. The resulting matrix, however, has
a multiply vanishing eigenvaiue for w = 0 as the existence of static solutions cannot be excluded
by curl operations only. Positive frequencies yield eigenvalues with a negative real part, which create
problems for many algorithms for the solution of equation systems. That is why the combination of
eq. (71) with derivations of eq. (72) seems to be advantageous so that a better distribution of eigen-
values can be achieved. A variety ofcombinations is possible; e.g. with any diagonal matrix D~which
vanishes for every point where eq. (72) does not, the following 3N x 3N system can be derived:

D~DVSDADCe = 0. (73)

Equation (72) vanishes for all grid-points which are not on a surface where material properties change
and for which the current density is source free. In order to convert the algorithm to a symmetric
form we must perform a diagonal transformation, which physically replaces the field strength by the
root of the local energy density (except from a scalar factor):

e’ = (DCDADS)”2e, (74)
((DCD)(DCD)t + icoD,~D~’ co2I)e~= -1WDDADPV,- (75)
 M. Bartsch et a!. / Solution of Maxwell’s equations 31

D~D~DvSD’e’= 0, (76)

with the matrices

D = (D~’D5D~’)112, (77)
D = (D;’D~D~)”2. (78)

Solving a combination ofthese linear systems for a given driving current distribution yields the field
at all locations in G. Further specialisation to e.g. low-frequency eddy currents invokes mixing ofscalar
fields and vector fields. For a loss free medium with no driving current one obtains an eigenvaiue
equation, the eigenvalues of which are the resonant frequencies squared [3]:

(DCD)(DCD)te’ = w2e’. (79)

The above mentioned multiple eigenvalue at zero can be avoided by combining eq. (79) with (76). As
the total number of eigenvaluesremains at 3N, unphysical eigenvaluesappear, which can be identified
easily by resubstituting the eigenvectors in eqs. (79) and (76).
The analytical analogue to this combination of eqs. (75) or (79) with (76) is the equation
curl curl E = grad divE V ~E and divE = 0 thus solving V 2E.
—

5.3. Fields in the time domain

For transient fields with a central time difference formula, step size öt, no currents and no losses,
one obtains the Yee-algorithm [1]. In the presence of free moving charges the algorithm has more
constraints to ensure charge conservation [5]. The upper index n denotes the time in units of ôt.
These recursive formulae allow the easy calculation of transient fields of antennas, particle beams in
accelerators [51or wave propagation problems even in the lossy case:

= b~—z~tD~’CD 5, (80)
5e”~°
= + (I — Da)D;’D~’CDsD~’b”~’(I — — Da) ~ (81)

with

Da = exp(—D~’DK1~t). (82)

The detailed setting up of the final matrix problem to be solved depends too much on the specific
problem to be explained here for all possible cases. We refer to refs. [3,4] and will demonstrate the
wide applicability of this method by means of a few examples.

6. Description of the MAFIA programs

The MAFIA programs are a set oftwo- and three-dimensional computer codes based on the Maxwell
Grid Equations. They were developed for use in the computer-aided design of particle accelerators
and are now finding wider applications in other fields such as tomography, filters, integrated circuits
and resonators. See table 1.
32 M. Bartsch et a!. / Solution of Maxwell’s equations

Table 1
Description of the MAFIA programs.

Module Description Special features Geometry

M mesh generator, translates the many predefined shapes, interactive remeshing (x,y, z), (x, y),
physical problem into mesh data (r, z)
S static solver, electro- and magne- open boundary, nonlinear magnetic material (x, y, z), (r, z)
tostatic problems properties, advanced multigrid solver for very
large meshes

R, E matrix generator, eigenvalue sol- periodic boundary conditions, resonance fre- (x,y, z), (x, y),
ver for resonator and waveguide quencies (several hundred modes), waveguide (r, z)
problems modes and propagation parameters
T2, T3 time domain solver losses, open boundary, stimulation by: initial (r, z), (x, y, z)
field, current, waveguide, incident waves; wake
potentials
TS2, TS3 particle in cell code, calculates losses, stimulation by initial field, calculation (x, y, z), (r, z)
equation ofmotion for free mov- of trajectories, velocities, charge densities, etc.
ing charges
W3 eddy current solver, frequency losses, current stimulation (x, y, z)
domain
P post processor, displays results 1, 2 and 3D graphics of scalar and vector fields, (x, y, z), (x, y),
and calculates secondary field power loss integrals, field energy, far field pat- (r, z)
quantities tern, shunt impedance

I _____ I _____ _____ _____ I

S R W3 T2 T3 TS2L TS3

Hi~ _
Fig. 4. The MAFIA system with its interrelationships.

The acronym MAFIA stands for the solution of MAxwell’s equations by the Finite Integration
Algorithm. Figure 4 shows the interconnections between the various programs ot the third release [6].
The second release ofthe MAFIA Programs, comprising M3, R3, E3 1, E32 and P3, has been distributed
to over 120 installations worldwide including most countries of Europe, USA, USSR, China, Japan,
India and Brazil. The programs have already proved their worth through comparison with theoretical
calculations and by the successful design of accelerator components. The MAFIA codes are written
in standard FORTRAN77 and currently run on IBM, CRAY, VAX, APOLLO, HP, SUN, CONVEX,
AMDAHL, FUJITSU, HITACHI and STELLAR computers, among others. The distribution center
for codes and userguides is the Technische Hochschule Darmstadt, for information contact Prof. Dr.-
Ing. Thomas Weiland.
 M. Bartsch et a!. / Solution ofMaxwell’s equations 33

Fig. 5. Netplot of the field dependant ~ in the symmetry plane of the relay.

Fig. 6. The left of this figure shows a 3D perspective of the end region of the RFQ. Only one quarter of this structure needs
to be calculated, due to existing symmetries. For this structure the power loss in a cut plane at the and of the structure is
displayed.
34 M. Bartsch et a!. / Solution of Maxwell’s equations

7. Applications

The applicability of this method and of the computer codes based on it is almost unlimited. In this
short presentation we can only give a few typical applications from the area of electrical engineering
and accelerator physics.

/T’7T~1Tj~
• I : : ,

,t
,

~1
~ ~Ittl’fti
~
~ .
- ‘
~

~4~~4~I’‘‘1
-

Fig. 7. Arrowplot of the electric field of the E

01 -mode in a circular waveguide with an asymmetrically placed dielectric rod.
(Geometry: R = 1 m, r/R = 0.5, ofI’set/R = 0.1, e11 = 1, ~r = 8, f = 119.36 MHz).

Fig. 8. Nondestructive testing of a tube by inducing an azimuthal eddy current by means of a single coil at the position z = 0.
 M. Bartsch et a!. / Solution of Maxwell’s equations 35

The first example is an application of the MAFIA magneto- and electrostatic module S [8,9]. Re-
lays are electric switching devices. The design goal for such a device might be to reduce its weight, the
needed driving current or to minimize the force at the moving armature. This nonlinear magnetostatic
problem has been calculated with our static module and also computed with the well known PROFI.
The comparison has been performed for the achievable accuracy and the computational effort. The
agreement in the results was reasonable. S has the advantage that the use of open boundary condi-
tions allows a description of this structure with less unknowns than is necessary when using standard
Dirichlet or Neumann boundary conditions. Here a possible reduction in the number of unknowns of
10% reduced the needed CPU time by about 20%. Figure 5 displays the relay and the 1u-distribution
inside a cut through the central plane ofthe relay. One can clearly see the saturation effect at the bot-
tom ofthe structure and inside the top armature, that indicate the necesity of redesigning the relay in
order to obtain reasonable behaviour.
The next example is a radiofrequency quadrupole (RFQ) that has been calculated with the MAFIA
frequency domain module R/E [10]. The device might be tuned for a certain minimum eigenmode or
in order to minimize the losses. Three-dimensional calculations are important especially for the end
region of such a RFQ. A careful design using the MAFIA mesh genarator has been carried out, using
about 75000 unknowns. The lowest eigenmode calculated, 344.78 MHz is in good agreement with
measurements and other calculations. Figure 6 displays the power loss. This loss is strongest on the

Table 2

Comparison of fi (1 /m) of ref. [7] with our method.

Mode Solution [7] fi (1/rn) Numeric solution fi (l/m)

5.649 5.649
2 5.648 5.641
3 4.204 4.19
4 3.079 3.082

i’4,1

r1r~ryv4c~

7-

Fig. 9. Same tube and same stimulation as in the above figure but the tube has a radial hole at z = 0.
36 M. Bartsch et a!. / Solution of Maxwell’s equations

..,, _____
~J.. I. ____

Fig. 10. Difference of the electric fields of the tube with and without the hole.

- ,a I 7A ~
a a a -, .. ., _ - - .

.~t ~ Lu I I ~‘—‘-—-.~L
Fig. 11. Electric field of a ~%/2-dipolewith an unsymmetric coating. Since the geometrical variation of the fields is small,
only a mesh sized ~ x 0.75)~had to be computed.
 M. Bartsch et a!. / Solution of Maxwell’s equations 37

outside wall. The power loss can be calculated by the MAFIA-postprocessor from the field solutions.
Different special approaches exist for the analysis ofthe eigenvalues and field components ofdielec-
trically loaded waveguides. It is necessary to know these values to be able to estimate the influence of
mechanical tolerances and to design filters and resonators. Table 2 compares semianalytical results for
the phase constant 48 of a calculation similar to [7] with the values achieved by our method. Figure
7 shows the electric vector components of the third mode (E01 -mode).
A typical problem for nondestructive testing (NDT) is the detection ofcracks and holes by means of
eddy currents [11]. For example a tube can be tested by moving a coil through the tube and measuring
its impedance. Material defects in the conducting tube are detected by the change in the impedance. In

Fig. 12. Far field diagram for the field distribution as shown in the last figure. The maximum of the far field amplitude is
shifted towards higher 0 values.

Fig. 13. Human test body in a three-dimensional nuclear resonance spectrometer.

38 M. Bartsch et a!. / Solution of Maxwell’s equations

the following example the azimuthally oriented coil excites an azimuthal current density field, which
is shown in fig. 8. The z-coordinate axis is coincident with the axis of the tube and the coil, the x-
coordinate axis here corresponds to the radial coordinate. In a second calculation a hole in the tube
wallwas modelled. The resulting current density distribution in the plane normal to the hole is shown
in fig. 9. The electric fields induce a voltage in the stimulation coil, the difference of the fields with
and without the hole is plotted in fig. 10.
A far field calculation example is shown for the electromagnetic field generated by a 2/2-dipole. The
left metallic stub is coated with a material with the parameters ~r = 2, Pr = 2. The dipole is excited
by a current connecting both metallic stubs. Figure 11 shows an arrowplot of the electric field. The
far field characteristic can be seen in fig. 12.

S
I ~ I? I - ‘~

Fig. 14. Arrow plot of electric field in a cross-section of the nuclear resonance spectrometer. (Due to symetry the half
structure was calculated.)

a134~4~7

Fig. 15. Absorbed power density in the same cross-section as in the above figure.
 M. Bartsch et a!. / Solution of Maxwell’s equations 39

The last application examined the influence of a human test body in a three-dimensional nuclear
resonance spectrometer, as seen in fig. 13. The ii stimulation structures are the two strip lines, mounted
on dielectric supports at each end, on either side of the test person. Figures 14 and 15 show the
calculated field (T3 [121) for a 90 MHz driving current in a cut plane through the body and the
antennas, first the electric field as an arrow plot, then the absorbed power density as a contour plot.
The field of a 2/2-antenna resonance can clearly be recognised. The woak field density within the
body is a consequence of absorption and reflection caused by the conductivity and high permeability
ofbiological tissue. Most ofthe power is absorbed at the part of the body closestto the antennas. The
area of the neck shows a local maximum, which indicates body resonance.

References

[I] K.S. Yee, Numerical solution of initial boundary value problems involving Maxwell’s equations in isotropic media, IEEE
Trans. Antennas Propag. 14 (1966) 302.
[2] T. Weiland, A discretization method for the solution of Maxwell’s equations for 6 component fields, AEU, Arch. für
Elektron. und Ubertragungstech. Electron. and Commun. 31(1977) 116.
[3] T. Weiland, On the numerical solution ofMaxwell’s equations and applications in the field ofaccelerator physics, Particle
Accel. 15 (1984) 245, and references therein.
[4] T. Weiland, On the unique solution ofMaxwellian eigenvalue problems in three dimensions, Particle Accel. 17 (1985) 227.
[5] T. Weiland, Transient electromagnetic fields excited by bunches ofcharged particles in cavities ofarbitrary shape, in: Proc.
Xl-th mt. Coni on High Energy Accelerators, Geneva, 1980 (Birkhauser Verlag, Basel, 1980) p. 570.
[6] M. Bartsch et al., MAFIA release 3.X, in: Proc. 1990 Linear Accelerator Conference, Albuquerque, LA 12004-C, Los
Alamos, p.372.
[7] K. Steinike, Calculation of the propagation characteristics of a circular waveguide excentrically loaded with a dielectric
rod, Frequenz 44 (1990) 1.
[8] F. Krawczyk und Th. Weiland, A new static solver with open boundary conditions in the 3D-CAD system MAFIA, IEEE
Trans. Magn. 22 (1988) 55.
[9] F. Krawczyk und Th. Weiland, 3D magneto- and electrostatic calculations using MAFIA-S3, in: Proc. 1st EPAC Conf.,
Rome, 1988 (World Scientific, Singapore, 1988) Vol 1, pp. 520—522.
[10] D. Schmitt and T. Weiland, IEEE Trans. Magn. 28 (1992) 1793.
[11]P. Hahne and T. Weiland, IEEE Trans. Magn. 28 (1992) 1801.
[12] M. Dehler, M. Dohius and T. Weiland, IEEE trans. Magn. 28 (1992) 1797.