INTRODUCTION TO RF LINEAR ACCELERATORS (LINACS)1)

P.M. Lapostolle

Abstract
The history of linear accelerators is now more than 50 years old and
after a difficult start presented many remarkable successes. Theory and
technology encounter difficult problems briefly presented here with the
solutions so far developed on orbit stability, focusing, RF structure and
beam dynamics computations. Questions still remain however about
high intensity limits which, when answered, might still open a more
brilliant future for these types of machine.

1. BRIEF HISTORY OF LINACS

The first accelerators built for nuclear physics were of electrostatic type; such machines
were efficient but limited in voltage due to electrical breakdown.

In 1924 a proposal was made by Ising to add several accelerations without having
anywhere the total voltage.
The method was based on the use of drift tubes and time varying fields, as sketched in Fig. 1.

Fig. 1 Drift tube accelerator of Ising

Along the axis of metallic tubes, charged particles can drift without being subject to any
electric field except at their ends, in the gaps between two consecutive drift tubes, according to
their respective voltage V. If an accelerating voltage is applied initially to all the tubes and is
switched off on each of them between the time of entrance and exit of a charged particle, the
particle will receive in each gap a succession of accelerations. In practice, voltage pulses are
more appropriate, as used in present day induction linacs; needless to say, however, that at the
time of the proposal shape and timing of high voltage pulses were not good enough to produce
a useful operation.

In 1928, Widere proposed to replace voltage pulses by an RF voltage (with a constant

frequency, the drift tube lengths have to increase with the of the particles). The method was
tested on a single drift tube (input + output) and with 25 kV RF peak voltage, it was possible to
observe an acceleration of 50 keV for singly-charged Na and K ions.

In 1931 Sloan and Lawrence built a real linac of 30 drift tubes giving to Hg ions an
energy of 1.25 MeV; lengthening it later in 1934 to 36 drift tubes and increasing their voltage
they reached 2.8 MeV. the intensity was of course very low and the beam quality not specified:
R.F. voltage differed from Ising pulses, having no real flat top, phase stability (see Section 2)
was not yet discovered and focusing was not ensured, except maybe by ES lens effect (see
Section 5).
No further development occurred until the war, due to the lack of proper high power RF
technology (limited then to 10 MHz) and to the discovery of the cyclotron. The length of the
Sloan and Lawrence machine approached 2 m with a max of only a few thousandths; with the

1) More details on the subject can be found in a Los Alamos report LA 11601 MS, Proton Linear Accelerators,
or in a CERN Yellow Report (in French) CERN 87-09
same RF wavelength of 30 m acceleration of protons would have led to a prohibitive length. In
cyclotrons, on the contrary, the spiralling of the trajectories together with some focusing effect
allowed the succession of many accelerations over a limited extent, leading to the possibility of
producing 10 MeV protons.

The development of radars offered, after the war, pulsed high voltage equipment in the
metric and centimetric wave ranges; science and technology of electromagnetism and beam
dynamics of already high level were then available.

The parallel development of circular accelerators (synchrocyclotrons and synchrotrons)

was also helpful with the discovery of phase stability.

Ginzton, Hansen, Chodorow, Slater, Walkinshaw used 3 GHz to accelerate electrons.

Alvarez and Panofsky used 200 MHz for protons.

2. LONGITUDINAL MOTION. PHASE STABILITY. ACCELERATION BY

A TRAVELLING WAVE

For a given geometry of the drift tubes which corresponds to a certain rate of acceleration,
particles must receive at each gap an exact energy gain and the voltage must have an exact value.
The RF voltage V applied is larger and there are two phases per RF period for which the
voltage has the right value Vs (see Fig. 2). When the field is rising the phase is stable since a
particle arriving too early will be less accelerated and slip slightly in phase until the next gap;
vice versa for a late particle. The other phase is unstable. The stable phase is called the
synchronous phase s and one has
V s = Vcos s

with = 0 corresponding to the crest (proton case).

Fig. 2 Phase instability

In a first approach one can replace the discrete gap configuration by an equivalent
continuous interaction (in Section 5 will appear a justification). One may say that the field
distribution along the axis in the successive gaps presents an analogy with a standing wave
pattern, sum of a forward and a backward wave. One would then consider the interaction of the
forward synchronous wave with the particles.
 Introducing relativity symbols and letting s , s , s refer to the synchronous particle,
one has
W
=1+
m0 c 2

and putting
= s

one can write, following the path of a particle, along the axis

m0 c 2 d ( ) qET
w ds
=
w
(cos cos s )
d 1 1
= =
ds c s c s3 s3

whence
d s3 s3 d
m0 c3 = qET (cos cos s )
ds ds

Forgetting about the change in s one then obtains

2qET s3 s3
2 +
m0 c
(sin cos s ) = Ct

where ET is the amplitude of the synchronous wave.

Such a motion derives from the Hamiltonian

2
(sin cos s )
qET
H= 3 3
c 2 s s m0 c 2

Neglecting the change in s is only valid for heavy particles (protons and ions) for which
it is slow enough. It corresponds to an acceleration fighting against a constant breaking force
(or to a forced pendulum model as often presented for circular machines).

One gets from it the usual stability bucket of Figs. 3 and 4; Fig. 5 shows the phase space
plot relative to an operation with s = 0 (fixed point as used by Sloan and Lawrence).
Taking into account the change in s opens up the bucket and gives the so-called golf
club, see Fig. 6 (notice that the coordinates used are not conjugate).
Fig. 3 Classical stability bucket

Fig. 4 Non-accelerating bucket

Fig. 5 Fixed-point operation

 Fig. 6 Golf club

For electrons increases very rapidly and becomes close to 1 over a short distance.
Most of the accelerating structure (apart from a buncher) is a constant velocity = 1 structure.

The previous derivation must be modified; keeping instead of , it leads to:

1
(cos cos 0 )
eET
= 2 1 =
1+ m0 c

where refers now to a 0 of the field expressed as ET sin (electron linac convention).
Figure 7 shows a corresponding phase space plot. One may notice that the 0 of a curve
corresponds to its position for .

Introducing, as is the custom

eET eET
= 2 = 2
m0 c m0 c

one can see that for 2 the curve corresponding to 0 = /2 which provides the maximum
acceleration can start from W = 0. For < , however, there can be low energy electrons just
slipping in phase and not accelerated (Fig. 7).

Fig. 7 Electron acceleration phase plot for <

3. TRANSVERSE MOTION. DEFOCUSING ACTION OF THE
ACCELERATING FIELD

The fields at the entrance and exit of a gap produce focusing and defocusing actions (see
Fig. 8). If the field is rising the overall effect is defocusing.

Fig. 8 Focusing and defocusing actions in a gap

Calling on the travelling wave accelerator concept one may transform the EM field of the
synchronous wave in a frame moving at the same velocity (supposed constant). A TM
accelerating field becomes electrostatic. Then

Ez E Ey E
= x = 2 r
z x y r

If there is a longitudinal stability in z, there is instability in the transverse direction.

This has been a strong handicap for ion linacs until AG focusing was invented and
applied (thin foils were destroyed by breakdown and thick wire grids, if transparent enough,
introduced bad aberrations). New methods exist now, in particular the RFQ principle for low
.

For electrons, which become very quickly relativistic, the transverse effects are very
weak.

4. LINAC ACCELERATING STRUCTURES

(Only a general review is given here; details can be found in more specialized lectures, in
particular on ion linacs.)

In a circular waveguide, the phase velocity of the waves ph is always larger than 1. It is
then necessary to slow them down. Periodic loading is used and/or field concentration or
reorientation via drift tubes.

The structure must also provide space to install quadrupole focusing (at least at low for
ions and protons).

Stability of the field distribution is also a concern.

<< 1 (ions)
 Sloan and Lawrence structure (present design, see Fig. 9a); the large drift tubes can
house quadrupoles (frequency from 10 to 100 MHz).

H-type structure (see Fig. 9b); the field of the transverse electric mode is made
accelerating through the drift tube configuration; since there is not much space for focusing, the
fixed point operation (see Fig. 5) is used, with focusing and rebunching from place to place
(frequency up to 100 MHz).

Fig. 9a Fig. 9b

0.02 < < 0.1

Quasi Alvarez structure (see Fig. 10); the large drift tubes (almost 2 long) can house
quadrupoles; the rate of acceleration is lower than with the H-type structure (frequency up to
200 MHz); high intensities can be accelerated.

Fig. 10

0.03 < < 0.4

Alvarez structure (most common for protons, see Fig. 11); a quadrupole is put in each
drift tube. Such a structure allows very large intensities (frequency between 100 and
400 MHz). For approaching 0.5 the RF losses increase, the drift tubes becoming resonant
like /2 antenna.

Fig. 11
> 0.4 (protons)
 Side-coupled cavities (Fig. 12). At this the focusing is installed in between sections of
cavities (frequency from 600 to 1200 MHz).

Fig. 12
= 1 (electrons)

Iris-loaded cavity (Fig. 13) with three or four irises per wavelength (frequency usually
3 GHz).

Fig. 13

Other types of structures

Superconducting (or normal temperature) independent cavities for heavy-ion boosters: a

few gaps or even a helix. Very flexible for various particles and different : for each velocity
and particle (charge and mass) the phase of the independent cavities is correspondingly
adjusted.

RFQ (low- protons and ions): cavity excited on a quadrupolar EM mode with profiled
vanes or rods to provide acceleration.
5 . DETAILED PARTICLE DYNAMICS COMPUTATION
 Acceleration through a gap
Transit time factor
Simulation codes

The initial approach (Panofsky equations) was developed from the experience gained on
electron beam tubes with grids (see Fig. 14). In the case of grids the field in a gap can be
uniform of value E0. With open holes on the contrary it penetrates inside and can have, on the
axis, a distribution as shown on the figure.

Fig. 14 Accelerating gap

With the initial approach (with grids) if one assumes a particle crosses the gap with a
constant velocity v with the phase in the middle, the energy gain is:

+g/ 2
z
W = q E0 cos + dz = qVT cos
v
g/ 2

with
V = E0 g

voltage across the gap, and

sin / 2
T=
/2
where

= g / v

is the transit time (in phase) through the gap and T is called the transit time factor (always
< 1): it is the reduction in acceleration with respect to what the voltage V would give.

In the real case (without grids) it is possible to express the amplitude of the field Ez (z) on
the axis (as obtained from measurements or rather from a computer) with the help of a Fourier
integral:
+

T (kz ) cos kz z dz
V
Ez ( z,r = 0) =
2
with the inverse relation
 +
VT ( kz ) = Ez ( z,0) cos kz z dz

from which

dT ( kz ) +
V = zEz ( z,0) sin kz z dz
dkz

In such an expression the field is represented as the sum of an infinite set of travelling
waves of amplitude proportional to T(kz) (in a standing wave configuration only two waves are
present). Each wave T(kz) is an EM wave the complete distribution of which can be known
from Maxwell's equation (assuming circular symmetry) in such a way that the complete field
distributions can be derived throughout the gap inside of a cylinder of radius a, the hole radius:
+
T (kz )I0 (kr r ) cos kz z cos (t + )dkz
V
Ez ( z,r,t ) =
2
+
T ( kz ) z I1 ( kr r ) sin kz z cos ( t + )dkz
V k
Er ( z,r,t ) =
2 kr
+
T ( kz ) I1 ( kr r ) cos kz z sin ( t + )dkz
V w
cB ( z,r,t ) =
2 ckr
with

kr2 = kz2 2 / c 2

With such expressions, making use of inverse Fourier relations, one can easily compute
the energy gain. Along a parallel to the axis, at the distance r, one has:

W = qVT ( kz ) I0 ( kr r ) cos

where kz = /v and T(kz) is the amplitude factor relative to the synchronous wave; the fact
that only the synchronous wave interacts justifies the approach used in Section 2. It also
explains a correction often introduced in the expression of the transit time, as follows:

sin / 2 I0 ( kr r )
T=
/ 2 I0 ( kr a)

with, for , a slight correction to take into account the effect of the chamfer of the hole.

With the field equations above one can compute the phase change across the gap,
accounting for the change in velocity, and the transverse motion. It is also possible to consider
a trajectory with a slope, such that
r = r0 + r'0 z

Such integrals lead to expressions involving T(kz) and its first and second derivatives with
respect to kz, according to the relation given above (see the report LA-11601-MS, page 80).
 These expressions are used in common codes like PARMILA and MAPRO, valid
however only for protons and ions.

A more recent code, DYNAC, valid for all particles (including electrons), with a better
accuracy, introduces instead of the radius r the reduced radius

dR
R = r z with R' =
dz
leading, in the paraxial approximation, to:
2
q 1 Ez q 2 +2 2
R"R + R Ez = 0
2m0 c3 3 3 t 2m0 c 2 4 4
such that R and R' are conjugate (which was not the case for r and r').

This equation shows, in addition to the phase dependent term usually defocusing (see
Section 3), a focusing term of the electron lens type, mainly important for electrons but also for
very low velocity particles (explaining probably the weak focusing observed by Sloan and
Lawrence, see Section 1).

Such a term for a uniform field just leads by integration to the classical solution and the
relation between r and R; it can however in the case of localized fields (gaps) exhibit an
appreciably enhanced effect, not at all in contradiction with the considerations developed in a
moving frame at constant velocity.

The derivation of the equation in R, not given in the report LA-16601-MS, is developed
in the Appendix.

6. SPACE CHARGE EFFECTS. INTENSITY LIMITS

Particles of the same charge repel each other. Space charge field tends to increase the
beam size and hence entails a risk of loss.

According to Poisson's law a linear space charge field requires = Cte.

The only solution for a uniform density, only possible for a continuous beam is the so-
called Kajchinsky-Vladimirsky (K.V.) distribution, which is a surface distribution in the 4D
space. Such a distribution is, of course, not very physical but fortunately it happens, from
energy consideration, that the K.V. beam equations are satisfied approximately (over short
enough distances) by r.m.s. dimensions (size and emittances) for all distributions (for a K.V.
distribution r.m.s. dimensions are half the real ones and r.m.s. emittances one quarter of the
real ones). Such a property is extremely useful, in simulation studies for instance, for finding
good matching conditions.

When the intensity increases, various phenomena occur, due to the non-linear character of
space charge forces.
Beam emittance can transfer from one coordinate to another and total emittance tends to
increase.

Around the beam core a "halo" develops; when trying to scrape it, it reappears. Even if
such a halo contains only one or a few per cent of the particles or even less this can be a very
serious problem for large intensity machines.

For small average intensity on the other hand, the emittance (beam quality) is the main
concern. A practical recipe which leads to good operation is as follows:
 AG focusing is usually specified by the phase advance per focusing period of the
incoherent oscillations; calling 0 the value for zero intensity and for full beam
one must have 0 < 90 and usually 0 in the range of 60 or 70 with /0 > 0.4.

In addition focusing and longitudinal stability should be adjusted, as far as possible,

in such a way that transverse kinetic energy and longitudinal oscillation energy
remain as close as possible to avoid emittance transfer.

Several simulation codes are available to study numerically the space charge effects. The
beam is represented by a few thousand of macroparticles and several methods are used to
compute their space charge field:

A fast Fourier transform (FFT) routine to solve Poisson's law when one can
assume a circular symmetry for the bunch (PIC, particle in cell code)

A numerical integration using analytical expressions valid for an ellipsoidal bunch

shape (Ellipsoidal profile code)

A point-to-point computation which does not make any assumption on the geometry
but requires some care to avoid "collisional effects" (Particle-to-particle interaction
code); this last code is much slower when increasing the number of macroparticles.

Present theoretical approaches to study the details and analyze the phenomena strongly
related to highly non-linear space charge fields are making use of the modern theories of
stochasticity with resonance overlap effects and Arnold's diffusion. They are still under
development.
BIBLIOGRAPHY

G. Ising, Ark. Mat. Astron. Fyz. 18 (1924) 1.

R. Widere, Archiv fr Electrotechnik 21 (1928) 387.

D.H. Sloan and E.O. Lawrence, Phys. Rev. 38 (1931) 2021.

D.H. Sloan and W.M. Coates, Phys. Rev. 46 (1934) 539.

L. Alvarez, The design of a proton linear accelerator, Phys. Rev. 70 (1946) 799.

J.P. Blewett, Phys. Rev. 88 (1952) 1197.

L. Alvarez et al., Berkelez proton linear accelerator, Rev. Sci. Instrum. 26 (1955) 111 and 210.

V.I. Veksler, A new method of acceleration, J. Phys. USSR 9 (1945) 153.

I. Kapchinskij and V. Vladimirskij, Proc. Int. Conf. on High Energy Accelerators, Geneva,
1959 (CERN, Geneva, 1959), p. 274.

E.M. MacMillan, The synchrotron, Phys. Rev. 68 (1945) 143.

L. Smith, Linear accelerators, Handbuch der Physik, Band XLIV (Springer, Berlin, 1959), p.
341-389.

H. Bruck, Acclrateurs circulaires de particules (Presses Univ.de France, Paris, 1966).

[English translation by Los Alamos: Circular particle accelerators, LASL.LA TR 72-10 Rev.
(1972).]

A. Lichtenberg, Phase space dynamics of particles (J. Wiely and Sons, New York, 1969).

E. McMillan, The relation between phase stability and first order focusing, Phys. Rev. 80
(1950) 493.

L. Teng, AG focusing for linacs, Rev. Sci. Instrum. 25 (1954) 264.

L. Smith et al., Focusing in linacs, Rev. Sci. Instrum. 26 (1955) 220.

E. Nolte et al., The Munich heavy ion post accelerator, IEEE Trans. Nucl. Sci. NS-26 (1979)
3724.

T. Fukushima et al., Measurement of interdigital H type linac, Proc. Linear Accelerator Conf.,
Santa Fe, 1981 (LA 9234C, Los Alamos, 1982), p. 296.

E. Knapp et al., Standing-wave high-energy linacs, Rev. Sci. Instrum. 39 (1968) 979-991.

D. Bhne, The UNILAC, Proc. Linear Accelerator Conf., Chalk River, 1976 (AECL 5677,
Chalk River, 1976), p. 2.

L. Smith, Beam dynamics in induction linacs, Proc. Linear Accelerator Conf., Santa Fe, 1981
(LA 9234C, Los Alamos, 1982), p. 111.
T. Fessenden, Induction linacs for HIF, Proc. Linear Accelerator Conf., Seeheim, 1984 (GSI
84-11, Darmstadt, 1984), p. 485.

A. Schempp et al., A heavy ion post accelerator, Proc. Linear Accelerator Conf., Montauk,
1979 (BNL 51134, Brookhaven, 1980), p. 159.
L. Bollinger et al., Concept of a superconducting linac, Proc. Linear Accelerator Conf.,
Seeheim, 1984 (GSI 84-11, Darmstadt, 1984), p. 217.

I. Kapchinskij et al., The linac with space uniform quadrupole focusing, IEEE Trans. Nucl.
Sci. NS-26 (1979) 3462.

J. Potter et al., RFQ research at Los Alamos, IEEE Trans. Nucl. Sci. NS-26 (1979) 3745.

R. Stokes et al., RFQ dynamics, IEEE Trans. Nucl. Sci. NS-26 (1979) 3469.

K. Halbach et al., Properties of the code SUPERFISH, Proc. Linear Accelerator Conf., Chalk
River, 1976 (AECL 5677, Chalk River, 1976), p. 122, and SUPERFISH, Part. Accel. 7,
(1976) 213.

M. Bell et al., Numerical computation of field distribution, Proc. Linear Accelerator Conf.,
Batavia, 1970 (NAL, Batavia, 1971), p. 329.

W. Panofsky et al., Berkeley proton linac, Rev. Sci. Instrum. 26 (1955) 119.

P. Lapostolle, Equations de la dynamique des particules, rapport CERN AR/Int. SG5/11

(1965).

A. Carne et al., Design equations in a linac, Proc. Linear Accelerator Conf., Los Alamos, 1966
(LA 3609, Los Alamos, 1966), p. 201.

M. Prom and M. Martini, Computer studies of beam dynamics, Part. Accel. 2 (1971)
289-299.

I. Hofmann, Generalized three-dimensional equations for the emittance and field energy of
high-current beams in periodic focusing structures, Part. Accel. 21 (1987) 69.

P. Lapostolle, Relations nergtiques dans les faisceaux continus, rapport CERN ISR DI/71-6
(1971). [English translation by Los Alamos: LA TR 80-8 (1980.]

F. Sacherer, r.m.s. envelope equations with space charge, IEEE Trans. Nucl. Sci. NS-18
(1971) 1105.

T. Wangler, Relation between field energy and r.m.s. emittance, IEEE Trans. Nucl. Sci. NS-
32 (1985) 2196.

D. Warner, Accelerating structure of CERN Linac, Proc. Linear Accelerator Conf., Chalk
River, 1976 (AECL 5677, Chalk River, 1976), p. 49.

M. Weiss, The new CERN Linac, Proc. Linear Accelerator Conf., Montauk, 1979 (BNL
51134, Brookhaven, 1980), p. 67.

R. Chasman, Numerical calculation of transverse emittance growth, IEEE Trans. Nucl. Sci.
NS-16 (1969) 202.
R. Jameson, Equipartitioning in linacs, Proc. Linear Accelerator Conf., Santa Fe, 1981 (LA
9234C, Los Alamos, 1982), p. 125.

P. Lapostolle, Etude numrique de charge d'espace, rapport CERN ISR/78-13 (1978).

APPENDIX
EQUATION OF THE TRANSVERSE MOTION WITH THE REDUCED
RADIUS IN THE PARAXIAL APPROXIMATION

The equation

d ( mvr )
= q( Er vz B ) (A.1)
dt
can be written

d (r' ) =
q
m0 c
( Er cB )dt = k (r, z,t )dt (A.2)

Putting

( )
1/ 4
R = r = r 2 1 and R' = dR / dz (A.3)

one has

5/ 4 d
( ) ( )
1/ 4 R
r' = R' 2 1 2 1 (A.4)
2 dz
and

d
(
(r' ) = R" 2 1 )
1/ 4

dz
7/ 4 d 2
( ) ( )
3/ 4
+ 2 1
R 3R 2 2
+ 1 (A.5)
2 4 dz
3/ 4 d 2

R
2
(
2 1
dz 2
)
Now, from Eq. (A.2), in the paraxial approximation

q Er B
k (r, z,t ) = k ( z,t )r = c r (A.6)
m0 c r r

with, according to Maxwell's equations

Er 1 Ez B 1 Ez
= and = 2 (A.7)
r 2 z r 2c t
But one has from Eq. (A.2)

(r' ) = ( )
d k r, z,t
(A.8)
dr c
and, for the longitudinal motion
 d
m0 c 2 = qEz (A.9)
dz
so that one also has

d 2 dEz E q Ez
m0 c 2 2 =q =q z + (A.10)
dz dz z c t

Eventually one obtains

2
d2R q 1 Ez q 2 +2 2
R + R Ez = 0
dz 2 2m0 c3 3 3 t 2m0 c 2 4 4
FUNDAMENTALS OF ION LINACS
M. Weiss
CERN, Geneva, Switzerland

Abstract
The operating principles of ion RF linear accelerators are described,
based on the concept of 'slow waves' generated in 'loaded cavities'.
Fundamental cavity parameters are presented, and equations
describing the action of electromagnetic fields on the particles are
given. Space charge influences are touched upon only briefly. Various
types of linear accelerators are introduced according to their use at
different particle energies and frequencies.

1. INTRODUCTION
Linear accelerators (linacs) accelerate particles on a linear path. Usually linacs operate
with sinusoidally varying electromagnetic fields and are called RF linear accelerators. RF
fields are created in a bounded volume, known as cavity, and the cavity operates either as a
waveguide (the electromagnetic energy enters the cavity at one end, is dissipated partly along
the cavity walls which are of finite conductivity, and exits at the other end to be dissipated in
a matched load), or as a resonator (the electromagnetic energy is reflected back and forth on
the end walls). Accelerators working on these principles are called travelling- and standing-
wave linacs, respectively. RF linacs operate in the frequency range of a few MHz up to
several GHz. They are suitable for the acceleration of light particles, like electrons (at
10 MeV electrons already have the relativistic velocity = 0.999), as well as of heavier
particles, ions (protons at 10 MeV have = 0.145). Ions are slow particles and ion linacs
usually operate in the MHz region whilst electron linacs operate in the GHz region. The
structure of the cavity differs when used for slow or rapid particles. However, the basic
principles of operation are always the same. This paper deals with ion linacs which have a
vast field of applications, such as physics research, nuclear waste transmutation, breeding of
nuclear fuel, materials study (neutron spallation sources) and medical applications (cancer
treatment).
Sections 2 and 3 are essential for understanding the principles of operation of linear
accelerators with RF electromagnetic fields. The accelerating structures have been treated as
lossless for better clarity; lossy structures would be treated in a similar way. Section 4
outlines the basic cavity parameters, whilst Sections 5 and 6 deal with some more common
accelerator types. Section 7 shows how the action of electromagnetic fields on particles is
computed, and Section 8 touches the problem of space charge influences and the beam
loading of cavities. Section 9 treats a particular accelerator called the radio-frequency
quadrupole (RFQ), which is in wide use all over the world. Section 10 gives some concluding
remarks.

2. EMPTY CAVITIES: WAVE TYPES AND MODES

In free space, the electromagnetic wave has electric and magnetic field vectors
perpendicular to the direction of propagation. Such a wave type is called the transverse
electromagnetic wave (TEM). In bounded media, where the boundary is a perfect conductor,
such a wave type is not possible because the boundary conditions cannot be satisfied. On the
boundary, the tangential component of the electric field E t , as well as the normal component
of the magnetic field Bn , have to be zero. These conditions are satisfied if the wave has one of
the field components in the direction of propagation (the longitudinal or z-axis of the
coordinate system). If it is the electric field component Ez , the wave type is called transverse
magnetic (TM); if it is the magnetic field component B z the wave type is called transverse
electric (TE), [1].
Wave propagation in cavities can be explained in terms of reflections from wall to wall;
in this way one has a field component in the direction of propagation. Not all the angles of
wave incidence and reflection are allowed; only a discrete number of them satisfy the
conditions. For each wave type, one can have certain wave modes, which are indicated by
subscripts. For example, in a rectangular cavity one can have cavity modes
TEmn or TM mn ,
m and n indicating the number of half waves in the x and y direction, respectively (see Fig. 1).
If there is a third subscript (in resonators), it indicates the number of half waves in the z
direction.

Fig. 1 TE01 wave in a rectangular cavity

In linacs one distinguishes between two wave velocities: the phase velocity ph and the
group velocity g . The former is the velocity of the wave phenomenon, like that of rapid wave
propagation along the shore, resulting from an obliquely incident sea wave. Electromagnetic
waves, moving with the velocity c and falling obliquely on the side walls of a cavity, would
result in a longitudinal (along the cavity) wave phenomenon moving with a phase velocity
bigger than the velocity of light c. Of course, the wave energy does not move in the
longitudinal direction with the phase velocity, but with the much slower group velocity.
However, for the acceleration of particles in accelerators it is the phase velocity which counts.
To see this in more detail, we shall write the wave equation in cylindrical coordinates
(accelerators usually have a circular cross section) for a rotationally symmetric
electromagnetic field like TM 01, where the subscripts indicate the number of field variations
azimuthally, and the number of zeros of the axial field radially (0 < r R max), respectively:

2 Ez 1 Ez 1 2 Ez
+ r =0.
z2 r r r c2 t 2

The solution is usually given in the form of a product of functions of one variable:
Ez (z,r,t) = Z(z) R(r) T(t) ,
where the functions Z(z) and T(t) are of the form:

Z(z) exp (jkz) ; T(t) exp (jt) .

The factor k in the exponent represents the phase advance of the wave per unit length. The
factor represents the phase change per unit time or the angular frequency. The exponent
j( t kz) is typical for travelling waves. Counting z from the position of the crest of the wave
at time t = 0, one stays on the crest if during motion the condition

t kz = 0 ,
remains satisfied. One has
z
= = ph > c,
t k
as we already know. Inserting the assumed solution for E z into the wave equation, one
obtains an equation for R(r) :

d 2 R 1 dR 2
2 + + 2 k2 R = 0 .
dr r dr c
14243
K r2

This is the Bessel equation of zero order (due to the assumed field symmetry), and the
solution is given by the Bessel function of first kind and zero order:
R(r) = AJ0 (Kr r) .
At the boundary (cylinder of radius Rmax = a), Ez (z,a,t) must be zero, hence:

J0 (Kr a) = 0 Kr a = 2.405 (first zero of the Bessel function).

To satisfy the boundary conditions, the value of Kr is fixed. From its definition:

= K2 + k2 .
2

c r

This is the very important dispersion relation for empty cavities and for a given wave type
and mode. Plotting = f (k), we obtain the dispersion or Brillouin diagram presented in
Fig. 2.

ph =
k

d
g =
dk

Fig. 2 Dispersion (Brillouin) diagram for empty cavities

The plotted curve is a hyperbola and the slope of the radius vector from the origin to a
point on the curve gives the phase velocity ph = /k ; all the points above the asymptote, for
which ph = c, have ph > c. The slope of a point on the hyperbola gives the group velocity
[2]
d
g = .
dk
The lowest frequency for which the boundary conditions are satisfied is c. At this frequency
the phase advance k, as well as the group velocity, are zero, whilst the phase velocity is
infinite. To accelerate particles we need a longitudinal electric field but as the electromagnetic
wave is a time varying field we must maintain a synchronism between the wave and the
particle, which means that the phase velocity ph of the wave and the velocity of the particle
p must be equal. One sees immediately that in order to use electromagnetic waves for
acceleration one must find methods to slow the waves down.
3. LOADED CAVITIES: SPACE HARMONICS, TRAVELLING AND STANDING
WAVE STRUCTURES
One method to slow down the waves is to 'load' the cavity by introducing some periodic
obstacles into it [3], as shown in Fig. 3. The study of the solution of the wave equation in
such a case is based on two essential points:

Fig. 3 Disc-loaded cavity (schematic)

i) Floquet's theorem for periodic structures, which states that in a given mode of oscillation
and at a given frequency, the wave function is multiplied by a constant such as e jkL , when
moving from one period to the next. A simple function that satisfies Floquet's theorem is
ejkz.

ii) The complicated boundary conditions cannot be satisfied by a single mode, as was the
case with an empty cavity, but by a whole spectrum of space harmonics, which is in fact a
Fourier series applied to a periodic case. We now have:

Ez (r, z,t) = F(r, z)e j( t kz)

F (r, z + L) = F(r, z)
n
F(r, z) = an (r)e j(2 n/ L)z .

Introducing these expressions in the wave equation for rotationally symmetric waves we
get:
n d 2 an (r) 1 dan (r)
e jt e j(k +2 n/ L)z 2 + + Krn
2
an(r) = 0 ,
dr r dr
with

2 2 n 2
2
Krn = k+ ,
c L
and

ph = = .
k + 2 n / L kn
For each n we have a travelling wave with its own velocity ph, which is slowed down
compared to the case of an empty cavity. In principle we can find an n such that ph = p ,
which is what we wanted. For ph < c, one has Krn
2
< 0 and so Krn = jkr . The solution of the
Bessel equation is now expressed by the Bessel functions of imaginary argument, which are
called modified Bessel functions:
 an (r) = An I 0 (kr r)
where A n is a constant.
It is not easy to find the functional relationship between and kn for the dispersion
diagram of a periodic structure. However, this diagram must reflect the periodicity of the
structure and it will be drastically different from that of an empty cavity, even in the region of
n = 0. Figure 4 shows the dispersion diagram of loaded cavities.

Fig. 4 Dispersion diagram of a periodic structure (loaded cavity); unloaded cavity

case shown for comparison

Interesting features can be observed:

i) for a given wave type and mode, there is a limited passband of possible frequencies from
c to ; at both ends of the passband, the group velocity g is zero;
ii) for a given frequency, one has an infinite series of space harmonics, from n = to n =
+. All space harmonics have the same group velocity, but different phase velocities;

iii) when the electromagnetic energy propagates only in one direction (full curves on
diagram), we have a travelling wave accelerator (TW); when the energy is reflected back
and forth at both ends of the cavity (full and dotted curves), we have a standing wave
accelerator (SW).

vi) the group and phase velocities do not have to be in the same direction; if they are, one has
a forward wave and if they are in opposite directions, one has a backward wave.

4. FUNDAMENTAL CAVITY PARAMETERS

In the design of linear accelerators one is concerned with a certain number of
parameters, of which the fundamental ones are the following [4,5]:

i) Transit-time factor T
L j( z / p )
0 Ez (z,0)e dz
T= L .
0 Ez (z,0) dz
The transit-time factor T indicates how well a field Ez accelerates particles. The integral in the
denominator is equal to E0 L, where E0 is the average field in a cell, and L the cell length. The
significance of the transit-time factor will became clearer in Section 7.
ii) Effective shunt impedance per unit length Zseff

Zseff =
( E0T )2 [M m 1 ] .
dP / dz
The effective shunt impedance indicates how efficient the acceleration for a given dissipated
RF power dP/dz is, where P is the power flowing in the structure and E 0 T is the amplitude of
the relevant space harmonic.

iii) Quality factor Q

w
Q= ,
dP / dz
where w is the stored energy per unit length.
The ratio Zseff/Q = (E 0 T)2 / w is also an important parameter as it depends only on the
geometry of the structure and not on the quality (conductivity) of the walls; it indicates how
much acceleration one has for a given stored energy.

iv) Group velocity g

P
g = .
w
P is the power flowing in the structure; as w E2 , a small g is usually preferred for better
acceleration efficiency.

v) Frequency f
The choice of the frequency of an accelerator depends on many factors. In general, with
higher frequencies one has a better Zseff (Zseff 1/2 ), but the aperture for the beam gets
smaller.

vi) Filling time t

L dz
tF = ,
0 g (z)

where here L is the length of the cavity.

The factor tF indicates the time needed for the electromagnetic energy to fill the cavity; the
above formula is valid for TW accelerators. For SW accelerators, where the electromagnetic
energy builds up progressively in time one has
Q
tF .

vii) Duty factor
The duty factor indicates the fraction of time during which the accelerator is in active mode.
Linear accelerators usually operate in a pulsed regime where the pulse is repeated a certain
number of times per second. The duty factor is given by the product of the pulse length and
repetition rate.

5. SOME TYPICAL ION LINACS

Before describing some ion linear accelerators, it seems appropriate to analyse briefly
the TW and SW modes of operation. The TW mode is usually used for particles which
approach the velocity of light; the SW mode is used for slower particles. Both modes can be
understood by observing the dispersion diagram of Fig. 5:
 Fig. 5 Dispersion diagram of periodic structures
i) the working point of the TW mode is usually around point A (phase advance per period of
the structure /2; the phase velocity c, the group velocity corresponds to the tangent to
the curve at point A);
ii) the working points of the SW mode must have phase advances of 0 or , modulo 2 (as for
example points B and C); it is only at these points that the direct (full line) and reflected
wave (dotted line) have the same phase velocity, so they can both be used to accelerate
particles. A disadvantage is that at these points the group velocity is equal to zero; we shall
come to this problem later.
Figure 6 shows a disc-loaded ion linac which operates in the SW mode with a phase
advance of . The electric field Ez has opposite directions in successive cells. The 'noses'
around the beam aperture are present in order to increase the transit-time factor T (or the
amplitude of the space harmonic which is relevant for the acceleration). Figure 7 shows the
Alvarez or drift-tube linac (DTL), where the field E z is in the same direction in all cells.
Hence the DTL operates in the 0 mode (in fact, for the beam, the mode is 2 ). As the
azimuthal magnetic field is also in the same direction in all cells, the walls separating the cells
could have been left out. The DTL therefore has a higher shunt impedance and is better than
the disc-loaded structure at relativistic velocities of 0.04 < < 0.4. Another advantage is that
the drift tubes can house focusing elements (like magnetic quadrupoles), very important at
low particle velocities. Figure 8 shows a photo of drift tubes taken out of CERN Linac 2 (for
protons, output energy 50 MeV, frequency 200 MHz).

Fig. 6 Disc-loaded linac (schematic)

 Fig. 7 DTL linac (schematic)

Fig. 8 Linac 2: girders with drift tubes

At lower velocities, 0.01, a very convenient accelerator is the radio-frequency
quadrupole, RFQ. This widespread accelerator will be treated in some detail in Section 9. For
higher ion velocities, > 0.4, a convenient structure is the side-coupled linac, SCL [6],
shown schematically in Fig. 9. Its particularity is that in addition to accelerating cells it also
contains coupling cells, placed sidewise. The field vector Ez is in phase opposition in adjacent
accelerating cells. The coupling cells form another chain of oscillators and the SCL is
therefore a 'biperiodic structure'. Biperiodic structures have interesting properties: if the
accelerating and coupling cells are properly tuned, their passbands in the dispersion diagram
are brought to a confluence and form a curve as shown in Fig. 10. The original operating
point of accelerating cells, with the group velocity zero (A), becomes a point with a final
group velocity (B). Hence, energy can flow and the structure is said to be compensated.
(Energy in non-compensated SW structures must also flow to cover RF losses in the structure
walls; to render this possible, the electromagnetic field pattern gets slightly distorted.
Compensated structures avoid this field pattern distorsion.)
 Fig. 9 Side-coupled linac, SCL (schematic)

Fig. 10 Confluence of two passbands

The DTL can also be compensated, and a second chain of oscillators can be brought
into the cavity in the form of post couplers (see Fig. 11). The inductance of the post coupler
and its capacity towards the drift tube are tuned so as to bring the two passbands to
confluence. CERN Linac 2 is compensated in this way.

Fig. 11 DTL cavity compensated with post couplers (schematic)

6. STRUCTURE COMPUTATIONS: DRIFT TUBE LINAC
Linear accelerators are periodic structures with complicated boundary conditions, and to
design them (geometry corresponding to a given frequency and field pattern) one uses
computer programs. To illustrate this procedure we take as an example the computation of the
DTL, which operates with a TM 010 field type and mode. The DTL is composed of one or
several tanks (cavities), each tank containing a certain number of cells, coupled together, and
all resonating at the same frequency. Each cell is composed of a gap and two half drift tubes,
as seen in Fig. 7. The length of the cells increases according to the velocity of the particles,
which have to cross a cell in one RF period. One computes only a certain number of cells, for
various velocities, and then interpolates the results. Owing to symmetry it suffices to solve
the wave equation, with proper boundary conditions, in only a quarter of a cell, as shown in
Fig. 12. (In fact a cell is not symmetric because of the stem which holds the drift tube.
However, one first neglects the stem and introduces it only later as a perturbation.) To
increase the precision of calculations, the mesh size can be diminished in regions where the
field vectors have to be determined more accurately (see Fig. 13). In addition to fields, 2-D
programs like SUPERFISH [7] or URMEL [8] compute also other cavity parameters, such as
Zseff, Q, power loss in the walls PL, average accelerating field E 0 , stored energy w, transit-
time factor T, etc. Some of these parameters are used in beam dynamics calculations.

Fig. 12 Computation of DTL cells by SUPERFISH

Fig. 13 SUPERFISH calculations with a variable mesh size

7. ACTION OF ELECTROMAGNETIC FIELDS ON PARTICLES
In Section 6 we discussed how one computes an accelerator cell from the
electrodynamic point of view. Now we shall analyse how the electromagnetic field acts on
particles. As an example we consider the situation in a gap of a DTL accelerator (see Fig. 14).
The field pattern is TM 010 and thus we have the field components Ez , Er , and B [9].

Fig. 14 Cell of a DTL accelerator (schematic)

In Section 3 we expressed the Ez field by a Fourier series. In the same way, but to be
more general, we now express this field by a Fourier integral

Ez (r, z) = Ak I0 (kr r) cos kz dk ,

where cos kz has replaced the exponential function because Ez is an even function of z and
1/ 2
2 2
kr = k
c
.

By inversion of the Fourier integral one gets
1
Ak I0 (kr r) = Ez (r, z) cos kz dz .
2

To determine the constant Ak , we shall assume that at the drift tube bore radius a, the Ez
field has a constant value in the gap, and zero within the drift tube:
g g
Ez (a, z) = const = E , z .
2 2
Solving the integral we get

Eg sin ( kg / 2) 1
Ak = ,
2 kg / 2 I0 (kr a)
and for the field

Eg sin ( kg / 2) I0 (kr r)
Ez (r, z) = cos kz dk .
2 kg / 2 I0 (kr a)
The gain in energy of a particle q, when passing through the i-th cell of the DTL, is computed
as follows:
 Wi Wi1 = q LLi / 2/ 2 Ez (r, z,t) dz = q LLi / 2/ 2 Ez (r, z) cos ( t + ) dz ,
i i

where L i is the cell length and is the RF phase (counted from the crest of the wave) when
the particle crosses the mid gap (t = 0). At a given time t, the particle will be at the position z
given by

z = p t ,
where p is the velocity of the particle, considered here as constant. Substituting t = z/p in
the integral one gets

z z
Wi Wi1 = q LLi / 2/ 2 Ez (r, z) cos cos sin sin dz .
i
p p
We recognize that the first term under the integral gives the inverse Fourier integral
multiplied by cos , and the second term is zero, as the even function Ez is multiplied by the
odd function sin z/p . As the inverse Fourier integral has already been computed (note that k
= / p ), we have

sin ( kg / 2) I0 (kr r)
Wi Wi1 = q Eg cos .
kg / 2 I0 (kr a)

The result is interesting: q Eg cos would be the energy gain due to an alternating
voltage Eg traversed at a phase from the peak; the factor

sin ( kg / 2)
,
kg / 2
indicates the reduction in the energy gain due to the finite time the particle travels across the
gap; the factor
1
,
I0 (kr a)
is the additional reduction due to the fact that the field Ez on the axis penetrates into the drift
tube aperture and is weaker than E, the field at the drift tube bore radius. Of course

L / 2 Ez (0, z) dz = Eg .
L/2

The field off axis is always stronger than the field on axis [factor I 0 (kr r)].
Substituting Eg with E0 L (E 0 is the average longitudinal field on axis in a cell), we
finally get

Wi W i1 = q E0 L T(k,r) cos ,
with

sin ( kg / 2) I0 (kr r)
T(k,r) = = T(k) I0 (kr r) .
kg / 2 I0 (kr a)
T(k,r) is the transit-time factor off axis (on axis it is T (k)). The transit-time factor makes it
possible to compute the energy gain in the whole gap, without the necessity of integrating the
motion of particles point by point.
 Knowing E z , one can compute the other electromagnetic field components: E r with the
divergence theorem
r 1 E
div E = (rEr ) + z ,
r r z
giving
1 r Ez
Er (r ) = r dr ,
r 0 z
and B via
v
v 1 E
rot B = 2 ,
c t
giving
B 1 Er
= 2 .
z c t
One also has to make use of a property of Bessel functions:
x I 0 (x)dx = xI 1 (x) .

The formulae for all the TM010 field components are written below:
E0 L
Ez (r, z,t) = T(k)I0 (kr r) cos kz cos( t + ) dk
2
E L k
Er (r, z,t) = 0 T(k) I1 (kr r) sin kz cos( t + ) dk
2 kr
E0 L 1
B (r, z,t) = T(k) I1 (kr r) cos kz sin ( t + ) dk .
2 kr

The components Er and B produce a radial force on the moving particle. The change
of the radial momentum pr of the particle in a cell is

z z dz
pr,i pr,i1 = q LLi / 2/ 2 Er (r, z) cos + p B (r, z) sin + .
i
p p p
The computation proceeds in an analogous way as with Wi W i1.
8. BEAM LOADING OF CAVITIES: SPACE CHARGE
In Section 7 we have seen how the electromagnetic field in the cavity acts on particles.
Particles also act on the fields, and this interaction between beam and cavity is called beam
loading of the cavity [10]. For reasons of phase stability [11,12], the beam bunch (particles in
the linac are grouped into bunches) and the RF field are not in phase: the bunch crosses the
gap before the crest of the electric RF field. Therefore beam loading has a resistive as well as
a capacitive component. The RF power supplied by the RF generator must, in addition to
structure losses, cover also the 'beam loading losses'. During the passage of the beam in the
cavity, an RF amplitude and RF phase control is in action, and additional RF power is
correctly supplied to keep the original pattern of the electromagnetic field in the cavity. A
feedback system acts on the tuners to keep the operating frequency stable.
 The space charge of the beam which loads the cavity also produces a repulsive action
between particles. As the charge distribution in the beam is not uniform, the space charge
forces are non-linear. The question arises: how can one match a beam with non-uniform
forces to a structure where the forces (as for example of focusing quadrupoles) are
predominantly linear? The answer is that it can be done provided one considers the root-
mean-square (r.m.s) values of the particle distribution in phase space. Without going into
much detail here, we just mention that Sacherer [13] has demonstrated that the evolution of
r.m.s values depends only on the linearized part of self forces, and that all beams having the
same intensity and the same r.m.s values behave in the same way. Therefore, once we know
the r.m.s values of a real beam it can be replaced by an equivalent one having the same r.m.s
values, but a uniform density distribution. Such a beam can be matched.
The r.m.s beam sizes are computed via the second momenta of a normalized
distribution function f(x,x,y,y,z,z)

( )
1/ 2
x = x 2 = ... (x x)2 f (x, x , y, y , z, z ) dx dy dz dx dy dz ,

where x is the first moment (average value), usually zero. The same is valid for other phase
space coordinates. The r.m.s beam emittance in the phase plane (x,x) is defined as

[ ]
1/ 2
rms = x 2 x 2 (xx )2 .

9. THE RFQ LINEAR ACCELERATOR

The RFQ accelerator is a very particular ion linac because it accepts an unbunched
(continuous) beam, bunches it, focuses it and accelerates it [14,15]. All these actions are
performed with RF fields: no external focusing is needed and the bunching efficiency
approaches 100%. The RFQ is a Russian invention (Kapchinskij and Teplyakov), but it
became popular worldwide only after the Los Alamos National Laboratory, LANL,
established a design method and built a prototype in 1980 [16]. This ion accelerator is the best
one in the low energy range, from a few keV to a few MeV per atomic mass unit.
The operation of the RFQ can best be understood by considering a long electric
quadrupole with an alternating voltage on it, as shown in Figs. 15 and 16. Particles moving
along the z-axis and staying inside the RFQ for several periods of the alternating voltage,
would be exposed to an alternating gradient focusing (Fig. 15). If the tips of the quadrupole
electrodes are not flat but 'modulated' in an appropriate way, a part of the electric field is
'deviated' into the longitudinal direction and this field can be used to bunch and accelerate
particles (Fig. 16).

Fig. 15 Long electric quadrupole with alternating voltage

 Fig. 16 Modulation of RFQ electrodes to create longitudinal fields
It is very instructive to analyse the RFQ from the beam dynamics and electrodynamics
point of view.

2.1 Beam dynamics

The particles move along the z-axis of the RFQ, and thus the beam dynamics considers
only the small central region around this axis. The electrodes form a well defined boundary
and, due to symmetry, the magnetic field on the axis is zero and nearly zero in the close
neighbourhood. Therefore, in this region one can replace the wave equation by the much
simpler Laplace equation for the electric field potential U(r, ,z):

1 U 1 2U 2U
r + 2 + =0 .
r r r r 2 z2
The general solution of this equation has the form:
V n n l
U(r, , z) = [ Aon r 2n cos2 n + Aln I2n (lkr) cos2 n cos lkz] ,
2
where
l + n = 2p + 1, p = 0,1,2,...
V/2 is the electrode potential with respect to the axis
I2n (x) is the modified Bessel function of order 2n
k = 2/ is the relativistic factor and the wavelength
The general solution contains all the harmonics in an infinite series, but an RFQ is usually
well described with only a few harmonics. The lowest-order solution has only two terms, one
of each infinite series:

U(r, , z) =
V
2
[ ]
A01 r 2 cos2 + A10 I0 (kr) cos kz .

The first term is the potential of an electric quadrupole (focusing term); the second,
containing cos kz, is linked with the acceleration. Constants A01 and A10 are determined by
imposing that the potential U along the electrode stays constant and equal to V/2:
 m2 1
A10 = ,
m 2 I0 (ka) + I0 (mka)
1
A01 = 2 [1 A10 I0 (ka)] = 2 .
a a

The constants A 10 and A 01 are expressed by geometric parameters, a, the minimum distance of
the electrode from the axis, and m , the modulation factor. Increasing m one gets more
acceleration; decreasing a one gets more focusing. Multiplying the equation for A 01 by V one
gets:
V + A10 I0 (ka)V = V .
This equation tells us that part of the electrode voltage V is required for focusing ( V), and
another part for acceleration [A10 I 0 (ka)V]. As the electrode modulation can be produced in a
very precise manner (e.g. by a numerically-controlled milling machine), we can obtain fields
which act on the beam as we want. The field components, derived from the two-term
potential function, are:
U
= [2 A01 r cos2 + kA10 I1 (kr) cos kz ]
V
Er =
r 2
1 U
E = = V A01r sin 2
r
U V
Ez = = kA10 I0 (kr) sin kz .
z 2
In addition, all these fields have to be multiplied by a time factor, such as sin ( t + ), to take
care of their a.c. character. The computation of the energy gain per cell is done as with the
DTL linac. The RFQ operates with a phase advance per cell of , hence the cell length is
/2, i. e. the particles cross it in half of the RF period.
9.2 Electrodynamics
For the beam dynamics it was sufficient to consider a small region around the axis and
deal with the Laplace equation. For the electrodynamics one has to consider the whole cavity
and solve the wave equation. The RFQ operates with the TE210 field configuration, which is
schematically shown in Fig. 17, for an empty cavity as well as for a loaded cavity (electrodes
of such a shape are called vanes). The modulation of the vanes is neglected as it has little
effect on the design of the cavity. The electric field is transverse, and the magnetic field is
longitudinal and of opposite sign in adjacent quadrants. An electric equivalent scheme for the
four-vane RFQ is shown in Fig. 18. To compute the cavity with a 2-D program like
SUPERFISH, it suffices to consider, due to symmetry, only one half of a quadrant (see
Fig. 19). In the region where the fields have to be known with a greater precision, the mesh is
finer.

Fig. 17 TE 210 field pattern in empty and loaded cavities

 Fig. 18 Equivalent scheme for a four-vane RFQ

Fig. 19 SUPERFISH calculations of a half quadrant of a four-vane RFQ

So far we have left out one question: what happens to the magnetic field at both ends of
the RFQ? It cannot be perpendicular to the end covers of the cavity (perfect conductors), it
must be parallel to them. In fact the magnetic field of a quadrant satisfies this condition by
turning around the vane as shown in Fig. 20. These end regions of the RFQ are computed
with 3-D programs like MAFIA [17].
Apart from the four-vane RFQ, one very often also uses a four-rod RFQ [18],
developed by the University of Frankfurt and shown in Fig. 21. This RFQ resembles a
transmission line and, indeed, the cavity which encloses it affects very little the resonant
frequency. This can be an advantage when operating at lower frequencies ( 100 MHz),
where a four-vane RFQ would require too large a cavity. Figure 22 shows a design by
MAFIA of a four-rod type RFQ with symmetric supports, built by the Laboratori Nazionali di
Legnaro, LNL, Italy, for the CERN lead-ion project [19].
 Fig. 20 End region of a four-vane RFQ and its equivalent circuit

Fig. 21 A four-rod type RFQ developed by the University of Frankfurt

Fig. 22 The lead-ion RFQ designed with MAFIA

10. CONCLUSION
The basic features of RF ion linear accelerators have been described in this paper. In
order to accelerate particles, a synchronism between the electromagnetic wave and the
velocity of particles was required. The concept of the slowing down of waves linked with
loaded cavities was essential for understanding how linacs function. The presentation of
space harmonics in the dispersion diagram for periodic structures made clear the choice of
the working point for either the TW or the SW structure. The fact that SW structures operate
only with phase advances per period of 0 or became clear when considering that both the
direct and the reflected wave have to contribute to the acceleration of particles.
A few examples of ion linear accelerators operating in the SW mode have been
presented. All of them, with the exception of the SCL, usually operate in the frequency range
of one hundred to a few hundreds of MHz. Structure computations, as well as the
computation of the action of the electromagnetic field on particles, have been presented with
the aim to show the methods applied. The notion of space charge and beam loading complete
this introduction to RF ion linacs.
REFERENCES

[1] C. Ramo, J.R. Whinnery and T. van Duzer, in Fields and waves in communication
electronics (John Wiley, New York, 1965) p. 374.

[2] J.C. Slater, in Microwave electronics (D. van Nostrand, New York, 1951) p. 395.

[3] Ibid., pp. 169177.

[4] G.A. Loew and R.B. Neal, Accelerating structures, in Linear Accelerators, eds.
P.M. Lapostolle and A.L. Septier (North Holland, Amsterdam, 1970), pp. 4752.

[5] G. Dme, Review and survey of accelerating structures, in Linear Accelerators (see
Ref. [4]).

[6] E.A. Knapp, High energy structures, in Linear Accelerators (see Ref. [4]), pp. 601616.

[7] K. Halbach and R.F. Holsinger, Superfish A computer program, Part. Accel. 7 (4)
(1976) pp. 213222.

[8] T. Weiland, On the computation of resonant modes in cylindrically symmetric cavities,

Nucl. Instrum. Methods 216 (1983) pp. 229248.

[9] P. Lapostolle, Introduction la thorie des acclrateurs linaires, CERN 8709 (1987).

[10] D. Boussard, Beam loading, Proc. CAS, Queens College, Oxford, 1985; CERN 8703
(1987).

[11] P.M. Lapostolle, Introduction to RF linear accelerators (Linacs), these proceedings.

[12] M. Weiss, Introduction to RF linear accelerators, Proc. CAS, University of Jyvskyl,

Finland, 1992; CERN 9401 (1994).

[13] F. Sacherer, RMS envelope equations with space charge, CERN/SI/Int. DL/7012.

[14] I.M. Kapchinskiy, Strahldynamik in den Sektionen mit rumlich uniformer

Fokussierung, GSItr14/76 (Translation from Russian).

[15] M. Weiss, Radio frequency quadrupole, Proc. CAS, Aarhus, Denmark,1986; CERN 87
10 (1987).

[16] R.H. Stokes et al., The radio frequency quadrupole, Proc. 11th Int. Conf. on High
Energy Accelerators, Geneva, 1980 (Birkhauser, Basle, 1980).

[17] R. Klatt et al., Mafia A three dimensional electromagnetic CAD system, Proc. Linear
Accelerator Conf., 1986, SLAC report 303, pp. 276278.

[18] A. Schempp et al., Zero-mode RFQ development in Frankfurt, Proc. Linear Accelerator
Conf., Seeheim, Germany, 1984, ed. N. Angert (ISSN: 01714546, GSI8411, 1984)
p. 338.

[19] A. Lombardi et al., Comparison study of RFQ structures for the lead ion Linac at
CERN, Proc. EPAC, Berlin, 1992, eds. H. Henke, H. Honeyer and C. Petit-Jean-Genaz
(Editions Frontires, Gif-sur-Yvette, 1992) pp. 557559.
RF SYSTEMS FOR LINACS
W. Pirkl
CERN, Geneva, Switzerland

Abstract
An overview of RF systems for linacs is given, problems associated
with beam loading are discussed and examples of RF system
installations presented.

1. RF POWER REQUIREMENTS

In linear accelerators the beam has to gain its energy during a single pass through the
structure. This leads to high requirements on accelerating fields and on overall RF power. A
typical linac has therefore installed RF power in the range of tens to hundreds of Megawatts. It
is then justified to consider linacs as RF-oriented machines, in contrast to circular accelerators
which are rather magnet-oriented.

In the present context the RF system proper starts at the RF connection to the structure.
The latter can be of the standing- or travelling-wave type, but whatever the mechanisms of
power build-up and energy transfer to the beam may be there are two distinct requirements to be
covered: delivering power to create and maintain a well defined field pattern in the structure
(wall power Pwall), and replacing the power transmitted to the particle beam (beam power
Pbeam), i.e.

PRF = Pwall + Pbeam

All wall losses can be thought of as concentrated in a single resistor, the "shunt impedance" r
defined as
r = Uge
2
/ (2 * Pwall )

where U ge is the effective structure voltage seen by a beam of given velocity, i.e. taking into
account the transit time factor T that expresses the voltage reduction due to the sinusoidal field
variation during beam-passage time. This is the "equivalent circuit" definition of the shunt
impedance hence the factor of 2 in the denominator which stems from the fact that the power
generated by a sinusoidal RF voltage is a factor of two smaller than the power generated by
DC.
On the other hand, the beam power is given by

Pbeam = Uacc * I1 / 2
Uacc = Uge * cos
where Uacc is the average energy gain of the beam expressed in eV, the stable phase angle
and I1 the spectral component of the beam current at the RF frequency. For short bunches this
component is close to twice the DC current I0 so that the beam power is approximately Uacc*I0.

Note that there is no third component representing additional losses caused by

transporting the beam power through the structure. This is due to the fact that it rides on top of
the very large reactive energy flows related with the fields so that the additional transport losses
can be neglected for all practical purposes.

2. BEAM LOADING

To study the effects of beam-structure interaction from the external RF system point of
view the equivalent circuit of Fig. 1 below is often used, idealizing the structure as a resonator
with a single accelerating gap across which the full effective gap voltage U ge is applied. This
resonator is driven by two sources, namely the RF system and the beam, which are supposed
to act all directly on the gap plane. While this model is a fairly accurate representation of a
simple buncher or even a complete standing wave tank it is not directly applicable to standing-
wave structures.

Fig. 1 Equivalent circuit of the cavity (single-gap resonator)

The corresponding phasor diagram is given in Fig. 2 for a data set corresponding
approximately to the conditions of Tank 1 at the CERN linac 2. As phase reference the
effective gap voltage has been chosen and positioned at 0 in the first quadrant. For phase
stability reasons the center of the bunch crosses the midplane of the accelerating gap(s) not at
the instant of maximum voltage Uge but at some time before. This time corresponds to the
"stable-phase angle" of the RF, whose value according to the "linac" definition which counts
this phase from the top amounts to -30 in the example. (The beam current is situated in the
third quadrant due to the sign convention in the diagram defining it as a generator rather than a
load.)

According to Kirchhoff's first law, the current to be fed by the RF amplifier is

Iamp = Icav Ibeam = Iloss + Ireact Ibeam

where I loss stands for the loss current (in phase with the voltage causing driving wall losses)
and Ireact the reactive current of the cavity (orthogonal to voltage causing lossless reactive
power exchange). This is represented by the phasor triangle in the fourth quadrant. Since the
amplifier requires a resistive load for optimal working conditions, the cavity is intentionally
detuned to provide a reactive component that just cancels the reactive component of the beam
current. This sets the necessary angle of detuning (47.3 in the example, i.e. slightly higher
than one half-bandwidth).

Now that the complex impedance of the detuned resonator is determined one can also, as
an alternative, analyze the circuit by superposition of the voltages. Responses of the resonator
to the amplifier current alone (Uamp) and to the beam current alone (Ubeam) are combined to
give again the effective gap voltage U ge as shown by the phasors in the first quadrant. (It is
assumed here for simplicity that the generator is of infinite output impedance, i.e. that it acts as
an ideal current source.)

It can be seen that the amplifier with unchanged drive would induce, in the absence of the
beam, a resonator voltage Uamp that is grossly out of amplitude and out of phase. After the
beam has arrived the resonator voltage would only gradually approach the wanted voltage U ge
in a (spiral) trajectory that takes several cavity time constants to complete. During this transient
the beam would not receive the proper energy gain and be useless.
 Fig. 2 Phasor diagram for the cavity

What is needed is some circuitry that would:

change the amplifier drive in amplitude and in phase such that it results in Icav before the
beam arrives and in Iamp thereafter,

set a cavity tuning angle that compensates the reactive part of the beam current.

These tasks are normally covered by three different regulation systems or servos, namely for
amplitude, phase and tuning. Combining amplitude and phase servos by a single overall RF
feedback is in principle possible (and applied in circular accelerators) but has not yet found
wide application in linacs.

3. ANATOMY OF A FULLY REGULATED RF CHAIN

Figure 3 shows a block diagram of the Tank 1 RF chain in CERN linac 2 as an

illustration of how the three servos mentioned above can be integrated into the system.
 Fig. 3 Block diagram of a fully regulated RF chain

3 . 1 Forward RF path (bold lines in Fig. 3)

A low-level RF signal of about 2 W arrives at the input plane with a well-defined phase
from the phase reference line. The signal then travels through the detector box, 6-bit phase
shifter, phase and amplitude modulator, is gradually amplified in the transistor amplifier, power
amplifier, drive amplifier and final amplifier before arriving at the feeder loop of the resonator at
sufficiently high power to cover tank requirements. Somewhere in the forward RF path there
have to be means to adjust output phase and amplitude at high speed in response to the electrical
correcting signals out of the servos. This is straightforward for the phase which is in general
well transmitted through almost any amplifier chain so that a low-level phase shifter can be used
at the input. However, the amplitude modulation is quite critical since most high power
amplifiers exhibit some kind of saturation and over saturation, where the RF output stalls or
even decreases for increased drive. In the present case these adverse effects are avoided by a
special current stabilisation in the final amplifiers so that the amplitude modulator can be
implemented as a low-level device at the input of the chain. In other machines the amplitude
linearity problem is solved by controlling the RF output amplitude via the plate supply voltage
using a high-voltage floating modulator with one or more series tubes.

3 . 2 Amplitude and phase servo

A sample of the tank RF field is taken from the tank via a coupling loop, cleaned in a
bandpass filter and routed to the detector box ( lines in Fig. 3). It carries information on
tank amplitude as well as tank phase. Both are demodulated in the detector box and appear at
the output in the form of video signals (A (amplitude) and (phase with respect to the input
plane)). In the feedback control module they are compared to the command value ATANK for
tank amplitude generated in the control system, and to 0 as implied command value for the tank
phase. The resulting error signals are amplified and fed to the amplitude and phase modulator,
thus closing the loop.
For constant beam current the amplitudes of the error signals are predictable so that
computer-generated "feedforward" signals can be injected for open-loop correction. This can
also reduce the transient at the start of the beam. Inputs for such feed-forward signals are
provided at the feedback control, but not used in normal operation. The absolute phase of the
tank with respect to the input plane is stabilized to some value that depends on the phasing
between cavity and detector box. Phase adjustment could be implemented by an adjustable
phase shifter in this line, but the unavoidable mismatches would also produce unwanted
amplitude variation at the same time. It is therefore preferable to set the operating phase by #1
in the reference signal phase feeding the chain. A motor-driven line stretcher "trombone" with a
resolution of 0.25 degrees is foreseen for this purpose.

The 6-bit digital phase shifter #2 is provided to center the initial phase of the main signal
path within the limited operating range of the fast phase modulator.

3 . 3 Tuning servo

To sense the tuning state of the resonator the phase of the voltage across its gap(s) has to
be compared with the phase of the forward wave in the feeder line. This information is
extracted by a directional coupler and fed via a low-pass filter and 6-bit phase shifter #2 to the
detector box where information on resonator phase is already available. An error signal is
generated and fed to the tuning control, which in turn actuates the tuner motors if the tuning
state is outside an adjustable error band. Phase shifter #3 allows setting of the tuning angle in
5.6 steps to optimize the power transfer between amplifier and cavity; it has virtually NO
influence on the phase seen by the beam due to the action of the phase servo.

4. OVERALL STRUCTURE OF THE RF SYSTEM

A linac RF system consists in general of many independent RF chains for a number of

reasons: accelerating tanks cannot exceed a certain length to avoid beam-breakup; power
amplifiers cannot exceed practical power limits; different functions like bunching, debunching
and diagnostics require independent settings; different harmonic numbers in the structure
require amplifiers of different frequency, etc.

The individual RF chains have to be linked together with constant phase-relationship that
should be unaffected by manipulations on any of its members. This is achieved by a phase
reference line whose outputs are strongly decoupled from each other, preferably through
directional devices to keep the power loss in reasonable limits. Figure 4 shows the overall
structure of CERN linac 2. Each chain receives its input signal via a directional coupler of
20 dB coupling and 40 dB directivity, such that the effective separation is as high as 60 dB.
The phase reference line is running in along the tanks, with couplers mounted in close
proximity to the outputs of the return loops. The two cables carrying respectively the reference
and the return signals are of the same type, tied and laid together so that a common variation in
length due to temperature differences will not influence the phasing of the resonator.

The power stages for Tanks 2 and 3 deserve special mention since there are two identical
amplifiers with separate feeder lines and coupling loops to work into a single tank. This
configuration has been chosen in order to provide two feed points into the relatively long tanks
and to improve this way the inter-tank transient conditions under heavy beam loading. From a
purely RF-plumbing point of view the two amplifiers could also have been connected to the
tank by a dedicated combiner, as explained in section 5.4 below.

Figure 5 gives an impression of the physical layout of such a chain:

 Fig. 4 RF system of CERN linac 2

Fig. 5 Physical layout of the RF chain for tank 2 in CERN linac 2

 The four racks at the left contain respectively: the computer interface and a HV supply for
the pre-predriver; low-level circuitry, servos and the pre-prediver amplifier itself (5 kW);
the HV power supply for the prediver; the prediver (50 kW).

the three cylinders to the right are the covers of identical triode power amplifiers used as
driver and composite final stage for tank 2. The height of these amplifiers is over three
meters, their lower part extends into a service gallery that also contains the infrastructure
(water supplies, blowers etc.) as well as the RF feeder lines and high-power phase
shifters (trombones).

the enclosure against the right wall contains the HV modulator for the final stages which
is implemented as a pulse-forming network discharged by an ignition.

The layouts for tanks 1 and 3 are similar. Buncher and debuncher chains consist only of
the four standard racks since the power delivered by the "predriver" is sufficient to power the
corresponding resonators.

5. SPECIAL TOPICS

Now that a general picture has been obtained some special topics can be discussed that
influence the performance of the RF system.

5 . 1 Limits of the regulation systems

5.1.1 Loop delay

We consider as approximation a servo system with a single time constant, together with
an overall time delay delay (caused by amplifiers, RF feeder lines, structure, return cable etc.).
The only limit on servo performance is then the overall loop delay delay. It is well known from
the theory of servomechanisms that a certain "phase margin" is needed at the unity-gain
frequency to assure satisfactory transient behaviour and to avoid instabilities. Assuming a
permissible phase margin of 45, the total phase at unity gain frequency must not exceed
(180 - 45) = 135 from which 90 are already reserved for the 6db/octave "rolloff" from the
leading time constant (the cavity in most cases). In other words the unity-gain frequency fug is
limited to a value where the delay contributes no more than about 45, i.e. since

f ug * delay * 360 = 45

( )
f ug < 1 / 8 * delay .
Knowing fug one can derive servo, the servo time constant after which a step perturbation is
reduced by a factor of e = 2.718. Theory of the simple model then predicts
( )
servo = 1 / 2 * * f ug = 8 * delay / (2 * ) 1.3 * delay .
Hence the servo time constant is always somewhat longer than the overall delay in the loop, a
result that is also intuitively evident.

5.1.2 Cross coupling of servos

Phase and amplitude servos are not independent as can be seen in the phasor diagram
Fig. 6a (left). The two components of the gap voltage, induced by amplifier and beam, are
presented in the same form as earlier in Fig. 2. Suppose that a test signal is injected in the
amplitude loop so that the voltage Uamp generated by the amplifier is (slowly) varied with
 Fig. 6 Cross-coupling of servos

constant phase. The resulting gap voltage U ge will however not only have a component in
amplitude A' but also in phase '; an unwanted return signal will therefore be seen in the
phase servo.

Similarly, a pure phase variation of the RF drive as in Fig. 5b (right) will lead to wanted
phase component " together with unwanted cross-coupling amplitude component A".

Another source of cross coupling of amplitude and phase is inherent in most high power
amplifiers due to non-linearities.

Since both servos have to react equally quickly they need similar cut-off frequencies.
Cross-coupling introduces additional phase shifts at these critical spots with the result that the
achievable frequency limit for both servos combined is much lower than for one servo alone.
The tuning servo, as the third element in the regulation system can fortunately be designed for
significantly slower response since it has only to cope with temperature drifts or air pressure
variations. Its crossover frequency is therefore much lower and will not interfere with the fast
systems.

5.1.3 Non-linearities in the RF chain

All power amplifiers have a tendency to saturate at the upper end of their operating range
i.e. their differential amplification is reduced or may even change sign. The loop amplification
of the amplitude servo varies accordingly. Despite linearization circuits in the feedback
controller there remains a residual effect that sets an upper limit to the applicable loop gain.

5.1.4 Overall performance of the servomechanisms

Due to the effects described above, the practical performance of the servomechanisms is
an order of magnitude lower than the theoretical limit set by the loop delay alone. Typical loop
gains are in the range of 20 40 dB, together with time constants of the order of microseconds.
This leads to operating precision of about 1% in amplitude and 1 in phase even under heavy
beam loading.

5 . 2 Choice of RF power source

It has been seen that the RF system of a linac is comprised of a number of individual RF
chains of very different output power requirements. It is then useful to keep the same basic
structure (like interlocks, computer controls, servomechanisms, etc.) for all these individual
chains and provide a set of modular amplifier blocks that are combined to provide the necessary
output power. A short overview of industrial power amplifier elements is presented in Fig. 7.
Legal power limits for different TV and broadcast services are also indicated since industry
 Fig. 7 Approximate RF average power limits of different active devices (source: VALVO
catalogue)

designs the main line of readily available off-the-shelf components within these limits. Letter
codes for main radar bands have been added since related equipment is and was the
crystallisation point for many linac RF systems. It may be worth remembering that the very
widespread operating frequency of 202.56 MHz had initially been chosen because of the
availability of surplus World War II radar components.

Some comments on the different amplifier categories:

5.2.1 Semiconductors

The "Transistor amplifier limit" is a quickly moving target since semiconductor amplifiers
with bipolar or FET-devices are in constant progress. They have replaced most of the tube
circuitry in all preamplifier stages and are gaining acceptance as final stages for certain
applications such as bunchers and debunchers. From the point of view of circuit design they
are low-voltage, low-impedance, low-Q devices.

Semiconductor components are inherently unforgiving concerning overvoltages; since

they are also quite sensitive to load conditions, they need adequate protection elements like
circulators which, however, make the amplifier an ideally-matched source.

Advantages: compact and reliable packages with virtually no maintenance; very wide
bandwidths possible.
Disadvantages: Tend to have high electrical delay times; are radiation sensitive (less important
in linacs than in other accelerators).

5.2.2 Tubes (Triodes and Tetrodes)

 Tubes use a control grid to modulate the density of an internal electron beam which, after
acceleration, transmits its RF energy to the external load circuit. Transit-time effects reduce the
effectiveness of grid control at high frequencies and mandate small system dimensions of
limited power carrying capability. Tubes are the workhorse for linacs up to about 300 MHz.
Gains are moderate, about 10 15 dB per stage for triodes and up to 20 dB for tetrodes. They
are generally run in "grounded-grid" configuration which reduces the gain but is the only
practical possibility to assure stability above a few MHz for triodes and above about 100 MHz
for tetrodes. From the point of view of circuit design, tubes are high voltage, high impedance,
high-Q devices.

A typical tube circuit provides coaxial input and output resonators to tune out the tube
capacitances over a relatively small bandwidth. The two resonators are often implemented in
"reentrant" configuration, i.e. they share one cylindrical metallic wall with inner/outer faces
assigned to inner/outer resonator (Fig. 8). It is of course the skin effect that separates the two
circuits at RF frequencies although they are at DC, galvanically connected by the conducting
wall.

Advantage: Relatively robust devices.

Disadvantage: Limited lifetime of the active elements, need regular maintenance.

Fig. 8 Structure of a triode amplifier

5.2.3 Klystrons

Klystrons are the active elements of choice for the higher linac frequency bands. They
overcome the limits of gridded tubes by the use of velocity modulation (rather than density
modulation). The internal electron beam is initially directed through a pillbox-type cavity where
its velocity is modulated with the frequency of the input signal. After a certain drift space the
faster electrons overtake the slower ones and create regions of higher density. Passive RF
cavities provided at these regions further enhance the beam structure by the self-induced electric
field. The RF beam power is finally transferred to an internal RF cavity from where it is
extracted to the external load.
Klystrons are very high voltage, high-Q devices. The long internal drift space of the
beam needs transverse focusing by solenoids or permanent magnets. As the length of this drift
space increases with the RF wavelength there is a practical lower limit for the operating
frequency. Figure 9 shows the "Klystron Skyline", the pictorial representation of the units
from one large European manufacturer. Also shown is the largest commercially available
klystron, almost 6 m in length for operation at 200 MHz. The large dimension has so far
hindered its use in low-frequency linacs.

Advantages of klystrons: High gain of 40 50 dB (about 10 dB per internal cavity).

High lifetime.

Disadvantages: Supply voltages of 100 200 kV necessitate special technology; X-ray hazard
exists if not properly shielded; pulsed high-power klystrons tend to be noisy electrically and
acoustically. In general only moderate DC-to-RF conversion efficiency ~ 50%.

5.2.4 Other RF power sources.

The upper power limit in klystrons is ultimately determined by the output resonator. Its
dimensions scale directly with the wavelength for a given cavity mode; at higher frequencies
some critical part (usually the coupling slot) becomes breakdown-limited due to the small
dimensions involved. Gyroklystrons, gyratrons and similar construction principles push this
limit upwards by the use of transverse rather than longitudinal extraction mechanisms which
allow larger cavity dimensions; their development is in full swing.

Two-beam accelerators first generate an auxiliary beam which runs in parallel with the
physics beam and serves as a distributed power source. A prominent example for extremely
high operating frequency is the CERN CLIC proposal which is presented in section 6.3.

5 . 3 Amplifier matching and resonator transients

According to a well known theorem in circuit theory, maximum power transfer between
source and load occurs if their impedances are complex conjugate to each other. This is
perfectly valid for low-level, linear circuitry but additional constraints of dynamic range have to
be considered at high power. Take for example an amplifier for P = 100 kW whose tube as
active element presents 15 k internal (source) impedance and supports a maximum RF
voltage U = 10 kV. 15 k is the optimum load at low level, but only up to the output voltage
limit of 10 kV corresponding to 5 kW. Beyond this point the load impedance has to be reduced
to avoid saturation and sparking; full power can only be achieved at a load impedance of
U2/(2*P)= 500 with drastically reduced amplification.
In virtually all active elements the optimal load resistance for maximum output power
differs by one or two orders of magnitude from the internal source resistance. Looking back
from the feeder line into the amplifier, this internal source impedance appears at a voltage-
standing-wave ratio (VSWR) of 10 to 100. It is assumed that the tube-to-feeder coupling is
arranged such that the optimal load for maximum output power corresponds to the characteristic
impedance of the line.

The load impedance of cavities and tanks is not constant but exhibits a sharp resonance in
the frequency domain. It is customary to refer to cavity impedances at the "Detuned Short"
(DS) plane of the feeder line. This has been implicitly been done in Fig. 1. The input
impedance begins and ends at 0 for the fully detuned cavity, and describes the well known
Fig. 9 The 'klystron skyline' of one European manufacturer
circle going through the center of the Smith-diagram for perfect match at the resonance
frequency (Fig. 10).

Fig. 10 Influence of the line length between amplifier and cavity: maxi Q/mini Q

Fourier transform allows one to derive the behaviour in the time domain for transient
conditions, i.e. for abrupt switching-on of the cavity by a gated sinusoid at the resonance
frequency. The apparent instantaneous cavity impedance is zero at the beginning, follows the
real axis, and finally stops at the steady-state resistance after several cavity time constants.

If the transformed source impedance is situated on this low-impedance part of the real
axis (Fig. 10 middle), then there is perfect match at least at one moment of the transient. The
overall average power transfer during the filling phase is optimal and the fastest possible
risetime results. In the frequency domain, the bandwidth is widest due to the strongest
damping of the resonator by the amplifier. The term "Mini-Q" has been coined for this regime
 Inversely, if the transformed source impedance is on the high-impedance part of the real
axis, there is never a perfect match; the average power transfer is worst; the risetime is slowest;
the bandwidth smallest. This is designated as "Maxi-Q". These two extreme cases represent
the only two points where the overall system is at "resonance", i.e. where the amplitude
response in the frequency domain peaks at the operating frequency and where there is no
beating in time domain. These conditions should be set for best overall performance. The
"Mini-Q" point is the preferred mode of operation, but is there is a tendency for amplifier
sparking in case of cavity breakdowns. Therefore the more forgiving "Maxi-Q" mode has
sometimes to be set during conditioning.

It should be pointed out that the above qualitative description is strictly valid only for low-
level conditions. Saturation effects modify more or less the behaviour at nominal operating
levels but the basic effects remain valid. As a practical consequence, a high-power phase
shifter (trombone) is in general necessary between amplifier and cavity for the optimisation of
transient behaviour (and not to adjust the overall phase at an ill-conditioned spot as is
sometimes suspected).

Some generators, especially semiconductors and klystrons, have stringent requirements

for the load impedance that preclude the direct connection of a cavity in pulsed regime. Non-
reciprocal ferrite devices such as circulators or insulators have to be provided in these cases to
absorb all load reflections. The transient behaviour becomes then independent of feeder line
length.

5 . 4 RF plumbing

The RF feeder line between the power amplifier and the structure transports in general
considerable RF power and has to be designed with some precaution. Except for very high
duty cycle it is the peak-pulse power rather than the average power that sets the limits.

In coaxial systems the support of the inner conductor is often the weakest spot.
Accumulation of dust and metallic particles, e.g. stemming from contact wear on phase shifters
(trombones), may substantially reduce the flashover rating in particular on horizontally oriented
disks. The use of metallic /4 stubs eliminates this problem altogether since there are no
insulating surfaces to be contaminated. It is of course prudent to choose the largest line
diameter possible. This is more an economic than a technical problem, since coaxial line
diameters of up to 300 mm having peak-power capabilities of tens of Megawatts can be used
around 200 MHz without problems from higher-order modes. Contact erosion from RF
currents has been a problem at some machines, this is again alleviated by large diameters
resulting in lower current densities. Pressurisation by air or SF6 is sometimes used to increase
the peak power capability, at least permanent venting with dry air is advisable as a preventive
measure. This complicates the mechanical design since continuity of the air channel has to be
assured and excessive leaks avoided.

In waveguide systems the dimension of the feeder is in general predetermined by the

operating frequency. Pressurisation or evacuation are the means to achieve high peak-power
capability; sometimes both possibilities coexist. The number of vacuum/gas windows, their
position with respect to voltage/current maxima under transient conditions, as well as particular
requirements of the elements to be interconnected have to be considered in the system layout.
Waveguide flanges have to be carefully assembled to avoid RF leakage.

5 . 5 Power combining and splitting

5.5.1 Direct combining

It may occur that a single generator cannot provide the required output power; power
combining using several sources is then a solution. This method is the rule in semiconductor
amplifiers where a great number of individual, identical output stages are working together in
one common output. Another example is a 200 MHz RF installation at CERN, delivering
1 MW continuous power combined from 32 commercially available TV transmitters.
Conversely, one large klystron may be able to feed several accelerating sections; power
splitting is then the answer. This is the case in most electron machines.

The preferred basic building block for both power combining and splitting is some
directional device in the form of a "bridge transformer", "directional coupler", "magic tee",
"decoupled junction" etc. which exists in many coaxial or waveguide variants. If correctly
imbedded in the circuit they can either provide virtual independence between the individual
sources or loads, or can implement matched source or load conditions even if assembled from
fully reflecting devices. One or several absorbers (also called terminations or loads) are
necessary to annihilate the unwanted reflections. Detailed descriptions can be found in
microwave textbooks; practical examples are given below in the paragraph describing other
machines.

5.5.2 Pulse compression

The majority of Linacs are used at the relatively low duty cycle of a few per mil or per
cent, where the generators are operated up to the permissible peak power but may still have
considerable reserves in terms of average power (i.e. pulse length). One approach is then to
generate a longer pulse than ultimately needed and compress it in time, thus trading excess
pulse length for additional peak power. An original scheme to achieve pulse compression in
accelerator environment was devised by Z.D. Farkas at SLAC under the name of SLED, a
scheme now in widespread use in many Linacs under different names.

Its function is best explained by a DC analogy (see Fig. 11): Imagine a DC generator
charging a capacitor. Reversing the polarity between generator and capacitor, in other words
changing the generator phase by 180, provokes a transient whose peak amplitude can attain
many times the current available from the generator alone.

Fig. 11 RF pulse compression by the SLED principle

The conversion of this principle to RF uses high-Q resonators instead of a capacitor as
energy storage elements. To avoid the main problem of reflections back into the generator a
balanced scheme with two resonators is used. A 3-dB directional coupler serves as connecting
element such that all reflections emanating symmetrically from the two resonators are added at
the output to the accelerating section but cancelled at the generator port. Both resonators are
first filled by the generator whose phase is then switched by 180 to initiate the transient of
multiple amplitude; this transient appears again only at the output port. The amount of energy
gain depends mainly on the ratio between cavity and structure time constants and is typically
between a factor of two and four.

6. FURTHER EXAMPLES OF LINAC RF SYSTEMS

6 . 1 LIL

LIL is the acronym for the LEP Injector Linacs providing 500 MeV electrons and
positrons for LEP, the world's largest accelerator (27.2 km ring circumference). The RF
system is typical for electron machines, with high RF frequency (2998.55 MHz), short pulse
length (4.5 microseconds) and relatively high pulse rate (100 Hz).

Figure 12 shows the overall structure: A phase reference line working at high power
feeds six active klystron lines through directional couplers, via amplitude and phase control
circuitry. Due to the high gain of the klystrons no intermediate amplifier stages are necessary;
on the other hand, their strongly non-linear characteristics together with the short RF pulse
length and the delay time in the travelling-wave structure precludes fast closed-loop control.

Fig. 12 Overall structure and RF networks of LIL

The first RF line (MDK 03) feeds standing-wave prebuncher and buncher cavities
exhibiting a sizeable transient; the klystron is therefore protected from load reflections by a
circulator. The following lines (MDK 13, 25, 27, 31, 35) are in principle designed for the
pulse compression scheme, even though lines MDK25 and 35 are only prepared for, but not yet
equipped with, the device. SLED has by the way been rebaptised LIPS (LIL Power Saver) in
the CERN environment. Since the accelerating structure is of the travelling-wave type with
relatively large bandwidth, the transient is sufficiently mild that these klystrons can be
connected directly without circulator. A single klystron feeds two or four segments of the
accelerating structure; the RF power is distributed in "canonical" way, i.e. by sets of 3-dB
couplers arranged in N layers with 2N output ports of equal power (here N = 1 or 2, for two or
four outputs). Figure 13 shows the klystron and Table 1 its characteristics.
 Fig. 13 Valvo 3-GHz for LIL. Upper part: RF output above drift tube surrounded by
focusing solenoid (X-ray shield removed). Lower part: klystron cathode with filament
transformer (normally immersed in tank with insulating oil).

Table 1
Main electrical characteristics of Thomson TH 2094 or Valvo YK 1600 klystrons

Peak output power 35 MW

Duration RF output pulse (max.) 4.5 s
Repetition frequency 100 Hz
Peak beam voltage 285 kV
RF drive power 250 W
Efficiency 45 %

6 . 2 TESLA

The object of the TESLA study is the proposal for a superconducting linear collider with
the highest accelerating gradient that can be obtained with present-day technology. Figure 14
shows a schematic diagram for the proposed test facility (TTF) consisting of two cryo units.
Each cryo unit carries eight cavities which in turn comprise nine RF cells fed by a common
power coupler. A single klystron is foreseen as RF source, its power has to be equally
distributed to the 16 cavities that form the load. Due to their very high Q a circulator is
mandatory to assure acceptable loading conditions to the klystron. The machine will work in L-
band at 1300 MHz; this frequency is unusual for accelerators but heavily used for long-range
radar so that high-power components are available. The klystron is provided by Thomson, a
major European manufacturer. Its data are given in Table 2.
 Fig. 14 Layout of the RF distribution in the Tesla Test Facility (TTF). One 5-MW klystron
feeds two cryo units. (Source: A. Gamp, report TESLA 93-43).

Table 2
Measured parameters of the TH 2104C klystron

Frequency 1300 MHz

Pulse length 547 s
Repetition rate 40 Hz
Beam voltage 130 kV
Beam current 89 A
Perveance 1.94 10-6
Drive power 45 W peak
Output power 5.1 MW peak 102 kW average
Body dissipation 3.2 kW max.
Efficiency 44.7 %
Gain 50.5 dB
Bandwidth 7 MHz

The number of load points, 16, is the 4th power of two and would therefore permit a
four-layer "canonical" power distribution with 3-dB couplers as in the LIL machine. Another
method is proposed instead that allows a much cleaner physical layout: after one initial split
into two branches by a "magic tee" the power is further distributed by in-line directional
couplers that run along the accelerating structure similar to a phase reference line. A minor
disadvantage of this approach is the necessity of couplers with unequal coupling factors, which
can however be easily implemented considering the very small bandwidth required.

6 . 3 CLIC

A radically different way to create the required RF power is used in CLIC (CERN Linear
Collider), a CERN proposal for the next generation of linear colliders. There are strong
arguments in favour of increasing the operating frequency for such machines: firstly, the
permissible gradient increases with the frequency according to Kilpatrick's sparking criterion;
secondly, for fixed total voltage and length, the average RF power consumption for copper
structures at room temperature scales as 1/f2. Arguments against high frequency include
undesirable wake-field effects and alignment difficulties due to the small dimensions involved.
The frequency of 30 GHz is proposed as a compromise on the high side of the spectrum. The
required RF power is as high as 40 MW per accelerating segment.

Standard generator tubes are ruled out as energy sources at this frequency, but the two-
beam accelerator approach is a valid alternative (Fig. 15). A high-intensity, low-energy drive
beam of very short bunches is generated in a dedicated drive linac, transported in parallel to the
main structure and used as distributed energy source. Transfer units extract 30 GHz energy for
the accelerating structure proper at regular intervals. In some project versions for a long
collider, superconducting reacceleration units are foreseen to restore the average energy lost
after several hundred segments. This assembly is equivalent to a conventional RF system in its
function as power source.

Fig. 15 Principle of two-beam accelerator (CLIC).

The low-intensity, high-energy physics beam is accelerated in a conventional structure

consisting of segments of 84 iris-loaded accelerating cells. Waveguides connect the two parts
as shown in Fig. 16. A prototype of the accelerating structure installed at the CERN test facility
CLIC has already achieved the nominal accelerating gradient (73 MV/m average, 103 MV/m
peak).

CONCLUSIONS

RF systems for linacs come in a large variety of structures. Their output power is
generally pushed to the maximum limit feasible by state-of-the-art RF technology, with a
general trend towards higher frequencies. Extensive closed-loop control is applied wherever
component characteristics and signal time delays permit.
 Fig. 16 Proposed layout of CLIC. One transfer structure of the drive beam (rear) feeds two
accelerating structures for the main beam (front).

* * *

BIBLIOGRAPHY

S. Turner [ed.], RF Engineering for Particle Accelerators, Proc. CAS CERN Accelerator
School, Oxford, 1991, CERN 92-03.

C. Zettler, RF systems for accelerators, CERN 87-10 (1987), pp. 231-245.

Meinke, Gundlach, Taschenbuch der Hochfrequenztechnik, 4 Auflage Springer 1986, ISBN 3-

540-15393-4.

Proceedings of linear accelerator conferences, organized bi-annually by different accelerator

centers in turn. Latest edition: 1992 Linear Accelerator Conference Proceedings, Accelerator
Physics Branch, Chalk River Laboratories, AECL-10728 (2 Volumes).
INDUCTION LINACS

J. de Mascureau
Commissariat a l'Energie Atomique, CESTA, France

Abstract
Induction linacs represent a somewhat unusual way of accelerating to
medium energies particle beams of extremely high intensities. They
generally operate in long-pulse mode where applications require high
power beams. In this chapter we shall present the principle and main
features of induction linacs and give information on the related problems
of beam transport and the means to diagnose them. Some specific
applications will be briefly reviewed.

1. INTRODUCTION

Development of induction linacs to accelerate very high current particle beams (multi kA)
to energies in the tens of MeV range is a relatively recent outcome of a principle which was
expressed before the last world war. In fact the idea was discussed by Bouwers in 1939 [1] but
the first realization was achieved by N. Christofilos in 1964 at LLNL [2]. This Astron
accelerator provides a 5 MeV, several hundred amperes electron beam to confine plasmas for
the magnetic fusion program. During the early 1970's at the Lawrence Berkeley Laboratory
(LBL) was built the Electron Ring Accelerator (ERA) for the goal of proton acceleration.
Meantime significant technological development has also occured in the Soviet Union in the
induction accelerator field [3]. Other machines were built during the seventies and the eighties at
LLNL and among them the Experimental Test Accelerator (ETA) and the Advanced Test
Accelerator (ATA). The last one, designed for beam propagation studies, was a 50 MeV,
10 kA, 50 nsec, device with burst-mode capability up to 1 kHz.

It must be noted that the LLNL has been during nearly twenty years a leading laboratory
for linear induction accelerator development. This was closely related to the progress of the Free
Electron Laser (FEL) program in this laboratory [4]. The last device built at LLNL is the
ETA II/ETA III accelerator operated in an FEL experiment aimed towards high power
microwave production for the MTX Tokamak magnetic fusion facility [5]. ETA III is a 6 MeV,
2 kA accelerator capable of multiburst operation (up to 50 pulses at 2 kHz) with improved
characteristics (beam quality and reliability). During the same period other laboratories were
involved in the development of induction linac technology and namely the Naval Research
Laboratory (NRL) and the National Bureau of Standards (NBS) [6].

More recently, during the last decade, new developments have appeared in China and in
Japan at ILE, KEK and JAERI[7,8,9]. In France, since 1988, the Commissariat a l'Energie
Atomique (CEA) has undertaken an induction linac development program at the CESTA
laboratoy, first for FEL application with the LELIA facility (2.5 MeV, 1 kA, 50 ns) [10,11] and
then for Flash - Radiography with the AIRIX project now under construction (16 MeV,
3.5 kA, 60 ns) [12]. Many papers have been published on the subject of designing induction
linacs in conference proceedings or elsewhere. Some of them are listed in treferences [1316].

2. PRINCIPLE OF OPERATION

In principle the accelerating unit consists of an induction cell that provides the particle
beam passing through with an eV energy increase and an electric generator supplying the cell
with a High Voltage (HV) pulse of V amplitude. By simply stacking n induction cells (and n
HV pulse generators) the incoming beam with an eVo energy exits with an e(Vo+nV) energy
according to the schematic drawing (Fig. 1). Induction linacs allow beam acceleration up to
several tens of MeV by using HV pulse generators operating at a few hundred kV voltage. As
we shall discuss further, the main advantage of induction linacs is their ability to accelerate
long-pulse (tens of ns to s) high-intensity (multi-kA) beams. Another specific feature is the
total electrical insulation of the apparatus, the high voltage appearing only inside the induction
cells.

Fig. 1 Schematic of an induction linac

An induction cell looks like a cylindrical metallic box with a hollow center where the beam
is transported and accelerated under vacuum. As shown in Fig. 2 the HV from the electrical
generator is applied to a circuit containing magnetic cores. The electrical insulation is obtained
via a concentric gap area and a ring ensuring both electrical and vacuum insulation of the beam
pipe, which is under vacuum, from the oil-cooled magnetic cores. Moreover a guiding magnet
is enclosed in the box to allow beam focusing while two trim coils permit small beam position
corrections.

Fig. 2 Induction cell

2 . 1 The transformer analogy

In order to understand the induction cell principle, the analogy of the 1:1 transformer is
usually presented as a first approach. In this comparison the primary winding is the metallic
structure of the cell, the beam itself playing the part of the secondary winding. When a square
pulse of V o voltage and t duration is applied to the cell a magnetic field is created into the
cores that induces on the beam an accelerating voltage of the same amplitude and duration.
When n cells fed by identical electric pulses are coupled, the beam is given an nVo accelerating
voltage as in a transformer with n primary windings and a single secondary. With this analogy,
represented in Fig. 3, we can understand how, in an induction linac, a high accelerating voltage
can be reached by simply stacking along the beam axis many cells fed by a pulse of moderate
voltage.

However only the gross electric balance is concerned by this approach. To go further in
the design of the cell and the magnetic cores we have to study the current and voltage
distribution in the accelerator.

Fig. 3 The transformer analogy

2 . 2 Current and voltage distribution

When the high voltage pulse (Vo, t) from the coaxial transmission line enters the cell, it
meets on the one hand a capacitive element constituted by the accelerating gap, and on the other
hand an inductive element constituted by the magnetic material. The I supply current is then
divided into an Ib charging current flowing into the gap and the inner wall of the cell, and an Im
magnetization current flowing around the magnetic cores (see Fig. 4). If these cores have been
properly magnetized before the pulse, they present a large inductance Zm limiting the evolution
of the Im current and preventing the line from being short-circuited on the electric ground.

The Ib charging current equals the beam current and therefore is limited by its impedance
which depends on the upstream diode characteristics. It has to be noticed that this current loops
in the HV generators by the means of coaxial cables. It never surrounds the magnetic cores
unlike the transformer analogy has assumed.

Let us now consider the voltage distribution at different places of the apparatus and to
begin with, at the cable meeting point with the cell, assuming the HV generator and the cell are
 Fig. 4 Current distribution

impedance matched. The potential difference between the power supply point (A) and the
ground (B) then equals the Vo voltage delivered by the HV generator.

V A-VB = 0
r
The same potential difference is obtained by computing the line integral of the E electric
field along the loop along which the Im current flows.
r r
Vo = E.dl

According to the Faraday law r

r r B r
Vo = E.dl = .dS (1)
s t

r
where S is the sectional area of the magnetic cores and B the magnetic induction produced by
the Im current. This last equation shows that the Vo cell voltage is linked with the magnetic and
r
geometric characteristics of the cores. If we neglect the radial variations of B the integration of
the Eq. (1) between to and to +t leads to :

Vo t S[ B(to + t) B(to )] (2)

We know that the magnetic induction inside the cores varies according to a hysteresis
cycle as shown in Fig. 5. Taking care to preset the cores at the Br remanence field before each
pulse, it is possible to get a maximum variation
Bs = Br + Bs
where Brs is the value of B at saturation.

Fig. 5 Hysteresis cycle Fig. 6 Characteristic dimensions

of a ferrite core of an induction cell

The equation (2) can be written as the following inequality :

Vo t S. Bs (3)

This can be used for the dimensioning of the magnetic cores according to the high voltage pulse
characteristics.

2 . 3 Considerations for the design of magnetic materials

Ferrites are often used as magnetic material for their mechanical and magnetic
characteristics. Other materials have been used and among them Fe-Ni alloys and Metglass
(from Allied Chemical). We currently have used PE 11B Ni-Zn ferrites from TDK. The
maximum B s for this material is 6 kG. New PE 16 ferrite allows larger B s of the order of
7 kG. For Metglass 2605 SC the Bs value is much larger (~ 25 kG) but this material has to
be used in ribbons of small thickness (~ 20 m) with Mylar insulation between consecutive
layers.

The role of the magnetic material is essential in the induction cell operation. Let us try to
see how it is possible to dimension these cores. If we consider the area they fill in the cell, it
could be regarded as a coaxial transmission line of length l ended by a short circuit Fig. 6. The
internal conductor of the line is formed by the cylinder of radius ri, the external conductor by the
cylinder of radius re and the short circuit by the SC plate. In this line, at point A, enters the
voltage wave (Vo,) associated with current Im such that :

Vo
Im = (4)
Zm

Zm forming the characteristic impedance of the line.

As in a classical line, as the wave advances in the magnetic material, the V o supply
voltage is propagated lengthwise along the internal cylinder whereas the external cylinder and
the SC plate are kept at the ground potential. When the wave reaches the plate, the whole
cylinder is virtually held at the Vo voltage. At this time the wave is reflected by the short circuit
with a polarity change for the voltage. As it comes back, the internal cylinder voltage goes
progressively to zero up to the A feeding point. Accordingly the V o voltage can be held at the
line terminals no longer than the time of a journey there and back for the wave along the cell at
the velocity 1 / .

In these conditions a pulse of duration will be used to best advantage by imposing a

magnetic core length such as

t 2l (5)

where and are respectively the dielectric constant and permeability of the magnetic material.

On the other hand the pulse duration must not be shorter than a single transit time of the
wave, otherwise some magnetic cores would not see the pulse and would be useless. This
condition can be expressed by
t l (6)

It must be pointed out that Eqs. (5) and (6) hold only when the Im magnetization current
remains smaller thant the Is saturation current. If not, a saturation wave is created with a
velocity larger than the electromagnetic wave velocity, and the magnetic cores get saturated in a
time shorter than the pulse duration.

This condition Im Is can as well be written as

H m = Hs (7)

Im Log(re / ri )
Hm = (8)
2 re ri

If the magnetic cores can be considered as an ideal inductance L

l
L= Log(re / ri ) (9)
2
then

dIm
Vo = L (10)
dt
and Eq. (10) leads to

Vo t L. Is

Vo . t Hs l (re ri )

Considering the hysteresis cycle of the magnetic core

B + Bs
r
Hs
we obtain again Eq. (3)

Vo t S. Bs
 The meaning of Eq. (3) is now clearer: when cells are dimensioned in such a way that this
condition is fulfilled, then saturation of the magnetic cores is prevented during the high-voltage
pulse by keeping the magnetization current lower than the saturation current.

To sum up, the transmission line analogy allows a better understanding of the phenomena
occuring in the magnetic cores which play a double part:

- holding the gap at the charging potential by delaying the cell short circuit,
- keeping the cell external structure at the ground potential during the whole high voltage
pulse.

The previous considerations are helpful in dimensioning the magnetic ( Bs ) and

geometrical (l,re ri ) characteristics of the cores. We shall give in another way information
about and the internal radius ri.

2 . 4 Cell coupling efficiency

During the flat top of the voltage pulse, the current from the HV pulse generator is divided
into two lines in the cell. The efficiency can be expressed as :

Ib
=
Ib + I m

Assuming the cores to be properly simulated by an inductance Zm and the beam by a

resistance R, then we have
Zm
= (11)
Zm + R

This efficiency is good provided that Zm is large compared to R. According to Eq. (9)
one condition is to get as large as possible. Another condition is to improve the re / ri ratio,
but this ratio also depends on other requirements. However the true effciency is lower due to
another resistance which is added in parallel with the cell in order to protect it against over
voltages. Indeed if operational problems occur, for example if the beam gets out of step with the
high voltage pulse, this additional resistance (we call it the "Ballast" resistor) prevents the gap
from seeing an over voltage which could result in damaging breakdown. The actual efficiency
can drop off to values under fifty percent.

2 . 5 High voltage generator

This aims at providing the induction cell with HV pulses having specific characteristics:

- high intensity, since I Ib + Im

- t and Vo fitted to the magnetic core characteristics. For electron induction linacs values in the rang
experimentally used.
- Flat-top pulse is an usual requirement in order to have a monokinetic beam accelerated.
This can have important repercussions on the beam transport in the accelerator.
- High repetition rate is needed in some cases, especially when high mean power
application is required.

If high repetition rate operation is not required, classical technology with charged
Blumlein and spark-gap is available. As an example Fig. 7 shows the result obtained with the
HV voltage pulse generator developed in our laboratory for the AIRIX program and designed
for a 250 kV standard operation with a 1% flatness over 70 ns.

Fig. 7 Experimental results of the AIRIX high-voltage pulse generator

If high repetition rate operation is needed, then the use of a generator with magnetic
switches instead of spark-gaps appears unavoidable. These magnetic switches are made with
magnetic cores which operate as saturable materials. They again use ferrites or Metglass and
present the advantage of a low jitter operation up to a few kHz. Moreover by placing several
magnetic switches in series a pulse compression can be realized.

3. BEAM TRANSVERSE DYNAMICS IN INDUCTION LINACS

The problems arising from beam transport along the induction linac are mainly due to its
specific characteristics :

- High intensity beams up to multi-kiloampere values are currently accelerated and

transported. This is due to the fact that the major part of the current does not flow
through the magnetic cores. The beam current which is given by the diode
characteristics of the injector has, in fact, no other limitation than the power of the
pulse generator.
- Acceleration of long pulses is achievable as long as the cell (magnetic cores) is fitted to
the pulse generator according to Eq (3). As an example the induction cells for our
AIRIX facility are designed to provide a 250 kV acceleration to an electron beam of
3.5 kA and 60 ns duration.
The phenomena which appear during the propagation can lead to a degradation of the
beam quality from:

- an emittance growth due to space charge effects,

- an effective emittance growth due to transverse motions originating in chromatic
effects and Beam Break-Up (BBU) instabilities.

We briefly give some results concerning these phenomena and what can be done to
minimize their effects.

3 . 1 Space charge effects

 These find their origin in the non-uniform radial distribution of the charges in the beam.
This results in a transformation of the potential energy into transverse kinetic energy. The effect
is a redistribution of the electric charges in the beam which causes an emittance growth as
shown in the next equation:
I 1
n2 R2 [U(z) U(o)] (12)
Io
where n is the normalized emittance of the beam
R is the beam radius (mean quadratic radius)
I is the beam current and Io is the limit current
and are the relativistic parameters.

The quantity in square brackets represents the variation of the electrostatic energy. As the
beam travels through the accelerator the beam profile has the tendency to get uniform which
implies that U(z) goes to zero.

In order to minimize these effects which appear mainly in the case of high currents at low
energy the obvious solutions consists in:

- injecting into the linac particles with as high an energy as possible (large )
- transporting a small radius beam.

But in the present case it is not possible to minimize I since it is a major feature of the induction
linac. Another way is to consider the injection of an electron beam with a uniform radial
distribution, but this is a characteristic of the injector. In practice this effect is generally
important only at the entrance of the linac and compared to the following effects becomes
negligible as the energy increases.

3 . 2 Chromatic effects

These are often referred to as "corkscrew" motions especially in the North-American

litterature. Their origin is double :

- The particle energy has a E variation during the pulse.

- And the focusing system is not perfect (mechanical alignment, design of the focusing
magnets,...).

The result is a motion of the beam centrod. Due to the length of the particle pulse in an
induction linac, it induces a global emittance growth when this effect is integrated over the
whole pulse duration.
In order to minimize these effects care has to be taken of the beam quality with regard to
its energy spread over the pulse duration. This is why we specify a E 1% for the electron
beam injected in to our AIRIX induction linac.

Another consequence is the need for an excellent quality focusing system along the
accelerator. As was indicated in Fig. 2 each induction cell has to be provided with a guiding
magnet. An example is shown in Fig. 8 representing the focusing magnet design of the AIRIX
induction cells. As can be seen in this figure soft-iron homogenizer rings mechanically
concentric to the beam pipe have been introduced. The effect is to reduce the transverse
component of the magnetic field inside the solenod. Measurements have shown an
improvement in the ratio of the transverse-to-axial components which reduce to values under
10-4 all over the solenod axis.
 Fig. 8 Focusing magnet design of the AIRIX induction cell

In the same way, the mechanical design of the cells is an important concern in that the
mechanical axis has to be aligned with the magnetic axis since only the latter has an influence on
the beam. Moreover all the cells must be aligned together in order to minimize the dispersion of
their axes not only for off-axis displacements but also for tilt effects. These considerations of
cell alignment are also concerned with beam break-up instabilities.

3 . 3 Beam Break-Up (BBU) Instability

This comes from the interaction of the beam with transverse electromagnetic fields in the
gap cavities. The BBU mechanism can be briefly summarized as follows:

- an off-axis beam propagation in the induction cell induces the excitation of transverse
electromagnetic fields in the gap cavity.
- These electromagnectic fields then react on the beam to induce transverse motions at the
frequencies of the electromagnetic modes. Consequently the next cavities get excited
from step to step.

BBU instability is an exponentially growing phenomenon inducing transverse displacements

which can be approximated by the following equation :
I Z ( o )
x xo exp moc 2 (13)
Io k E
where Z is the transverse impedance of the cell cavity and depends on:
- the gap geometry
- the materials used in the cavity
is the pulsation of the electromagnetic mode
E is the cell accelerating gradient
k is the wave number of the guiding magnet.

In order to prevent the growth of BBU instabilites or to minimize their effects the
following recommendations must be taken into account:

- on-axis propagation of the beam in the accelerator,

- accurate cell alignment (see 3.2),
 - reduce the transverse impedance of the cell, especially for the dominating mode.
It depends on the gap geometry according to
a
Z (14)
D2
where a is the width of the gap and D is the beam pipe diameter.

This impedance depends on the dielectric constant of the insulator. A smaller leads to
lower values of Z . For this purpose a plastic insulator, like Rexolite ( 2.5) must be
preferred to an alumina insulator ( 9).

- increase the cell gradient ,

- increase the k wave number, that is to say use a large B guiding field, but an
optimization has to be found with other effects which can lead to adjusting the B values
along the accelerator according to, for example :
B (15)

The use of an electromagnetic code is necessary to calculate the transverse impedances. In

Fig. 9 are shown the impedances corresponding to different designs of the gap cavity for the
AIRIX program with the PALAS code. The "VACUUM" case concerns a cell in which the
ferrites are under vacuum and there is no insulator, "REXOLITE" the technology designed for
the use of a specific plastic insulator, and "ALUMINA" the technology when a pure alumina
insulator is brazed on the body of the cell ensuring a perfect insulation between the ferrite area
in oil and the beam area in vacuum. In this example we should notice that the technological
characteristics (mechanical, electrical,...) have to be compared with the corresponding cell
behaviours concerning the BBU instability.

4. DIAGNOSTICS

This is not the place to deal with the accelerator diagnostics in general. But, as we have
seen previously, induction linacs are characterized by very high intensity beams and long
pulses. These specific features induce some problems of beam transport along the apparatus. In
order to minimize the effects according to the recommendations given in the last section the
accelerator designers are required to accurately measure some important parameters of the beam
and namely the beam emittance, the electron energy and energy spread, the beam position and
profile, and the beam current. As we have seen they also have to take care about the alignment
of each component (guiding magnet and cell) and of the whole apparatus.

To give an example, the 60 cells of the AIRIX induction linac have been designed and
will be built and aligned in order to allow the 60 corresponding magnetic axes to be held in a
100 m radius cylinder and a 500 rad half-angle cone. These specifications involve
sophisticated means of machining, assembly and alignment.
 Fig 9 Transverse impedances of AIRIX cells calculated with the PALAS code

An original method for checking the possible discrepancy between the mechanical and
magnetic axes of a cell is the so-called stretched-wire technique. It uses a conductive thin wire
perfectly aligned on the mechanical axis of the cell. A short electric pulse is injected at the input
of the wire. When this pulse goes across the magnetic field area inside the cell it encounters a
magnetic force which induces in the wire a small oscillation. The observation of this small
displacement with an adequate motion detector at the other end of the wire gives a signature
closely related to the kind of misalignment between the magnetic field lines of the solenod and
the mechanical axis of the cell. This technique is schematized in Fig. 10.

As induction linacs are often single-pulse machines, the usual scanning-wire techniques
cannot be employed. A good time resolution is required to analyse the evolution of the
phenomena during the pulse since high frequency modes (BBU) are likely to develop.

In fact most of the beam diagnostics consist of the optical observation of the interaction of
the (electron) beam with a specific target. The definition of the interaction and hence of the
target is related to the desired efficiency, directivity, spatial resolution, and temporal
resolution of the optical phenomena which often depend on the energy and intensity of particles
in the beam. Commonly used interactions are scintillation and Cerenkov radiation, but for high
energy beams, that is to say E 10 MeV, Optical Transition Radiation (OTR) presents
advantages as far as time resolution and directivity are concerned. Moreover scintillation and
Cerenkov radiation involve volume interaction and therefore can use beam disturbing converters
while OTR is a surface phenomenon, induced by an index step, and is likely to use thin
windows transparent to high-enough energy particles. But further studies have to be carried
through before the use of OTR can be generalized.

At the CESTA laboratory we developed a magnetic spectrometer in order to measure

accurately the time evolution of the electron beam energy for the AIRIX induction linac. In the
image plane of the electromagnet the electrons undergo a Cerenkov interaction into an optical
fiber sheet which leads the information to a streak camera. A first optical-fiber sheet allows the
observation of a broad spectrum versus time, while another sheet of closer fibers can give a
 Fig. 10 The stretched-wire technique for cell alignment

high resolution ( E ~ 4 keV) in a chosen domain of the spectrum. This zoom effect appears
very usefull for adjusting the flat-top part of the electron pulse injected into the AIRIX linac.

The other basic diagnostic tool for induction linac users is the emittance meter. We use the
well-known pepper-pot technique with a thin (0.5 mm) Bicron scintillator as analyser and a
gated camera to record the optical information as a two-dimensional image with a 5-ns
resolution. At LANL a streak camera is used giving the evolution of one spatial dimension
versus time. The same technique of recording can lead to the measurement of the beam position
and dimension with time resolution using Cerenkov or scintillation converters.

The beam intensity, and the position of the beam centrod, can be reached from B-loop
measurements. This device uses a small loop placed on the beam pipe, perpendicular to the
field of the electron beam. By integrating the loop signals to obtain the sum or the difference of
diametrically opposite loops it is possible to measure the total beam current or the beam centrod
displacements (charge density center) versus time. These so-called Beam Position Monitors
(BPM), can, if carefully designed and built and accurately calibrated, lead to high-resolution
time measurements of the beam-centrod motions closely related with the chromatic effects or
even with the high frequency modes of the BBU instabilities.

5. APPLICATIONS

The specific applications of induction linacs result form their main characteristics:

- High intensity (multi-kA) and high brightness (small emittance)

 2I
B= (16)
( n )2
- Long pluse ( 50 ns)
- Small energy variations during the pulse
- Repetition rate capability
- High peak power and high average power.

Since the first machine, built in 1964, many applications of the induction linacs have been
developed. As we have seen in the first section they mainly concern the USA and the USSR but
more recently important programs began in eastern Asia (China, Japan, Corea) and in Europe
(France). We will not review here all the facilites and all the subjects of interest but only give an
overview of the main applications and focus attention on two particular subjects (see 5.1 and
5.2).

At LLNL first studies were carried out for the goal of beam propagation in the atmosphere
(ATA) but programs rapidly turned to FEL applications.

Research on magnetic confinement fusion was concerned at the beginning with the
ASTRON program and recently with the MTX experiment. At LBL an ion induction linac is
being developed for inertial confinement fusion. At LLNL the FXR facility was built for high
power flash-radiography application [17] and at Los Alamos the DARHT induction accelerator
is designed to improve significantly the performances of FXR [18]. At CESTA, the CEA is
now designing and building the new AIRIX induction linac, similar to DARHT, for flash
radiography.

But potential applications of the induction linac are not so limited: New studies are going
on for the injector of the future CLIC (CERN Linear Collider) where an induction linac could be
used as a drive beam in a Two-Beam Accelerator (TBA) concept [19]. At the other extremity
industrial applications can be developed where high brightness beams are required which cannot
be reached with classical diode-type generators. They include all kinds of irradiation of
materials and particularly sterilization [20].

5 . 1 Free Electron Laser (FEL) application

The induction linac is a good driver for FEL experiments where high power output
radiation is required. Since the FEL wavelength depends on the electron energy and the wiggler
parameters ( w , w ) according to the equation
1 eB
2
= w2 1 + w w (17)
2 2 2 mc

in most cases the resulting wavelength is in the microwave or Infrared (IR) domains. In these
conditions induction linac characteristics (small emittance and energy spread) can lead to a good
efficiency in the FEL interaction. Moreover the high beam intensity and brightness with high
repetition rate capability can lead to high peak power and high average power radiation
emission.

An example of an induction linac FEL application was the MTX Tokamak experiment at
LLNL. The microwave radiation at 140 GHz from the ETA II induction linac driven FEL was
used to heat the plasma of the MTX Tokamak. In this experiment the ETA II accelerator has
been improved (ETA III) in order to generate up to 50 pulses at a 2-kHz repetition rate. The
FEL used the IMP wiggler in the amplification mode of a gyrotron source at 140 GHz. The
accelerator operation was stable enough to allow 25 microwave pulses to be injected in the
Tokamak with a 1-GW power.
5 . 2 Flash radiography with induction linacs

Electron induction linacs can generate a powerful flash of hard X-Rays in order to make
high quality radiographic images of dense, rapidly-moving objects. Here the goal is to produce
a large amount (dose) of X-Rays by focusing on a small spot ( ~ 2 mm) into a conversion
target a high intensity electron beam of convenient duration ( t ~ 100 ns).
The quality of the radiographic source is given by a figure of merit (FM).

D
FM (18)
2

where D is the dose of X-Rays and the diameter of the X-Ray source.

As the dose is proportional to the beam intensity, multi-kiloampere induction linacs can be
used where other accelerators cannot produce a good quality beam with a sharp focusing ability.
For Bremsstrahlung radiation from a target of Z atomic number the dose is approximated by

D ZIV 2.8 t (19)

where V is the electron energy on the target and which has to be high enough to produce the
desired dose of X-Rays on the object.

Small emittance is needed to achieve a good quality beam. We have seen that, in order to
limit the emittance growth along the accelerator care has to be taken to reduce the chromatic
effects and the BBU instabilities. The pulse duration is a compromise between dose and
blurring due to the object velocity. Two new accelerator projects for this purpose are now in
progress: DARHT at Los Alamos USA, and AIRIX at CESTA near Bordeaux, France.
REFERENCES

[1] A. Bouwers, Electrische Hochstspannungen, Verlag Julius Springer, Berlin (1939).

[2] N.C. Christofilos et al, Rev. Sci. Instr. 35 (1964) 886.

[3] A.I. Anatskii et al., Atomn Energ. 21 (1966) 439.

[4] R.J. Briggs, LLNL, UCID 21 639 (1989).

[5] C.J. Lasnier et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) vol II, p. 1554.

[6] J.E. Leiss et al, Particle Accelerator 10 (1980) 223.

[7] T. Akiba et al, Proc. Int. Conf. on High Power Particle Beams, Kobe (Japan) 1986.

[8] J. Kishiro et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) Vol I, p 673.

[9] S. Kawasaki et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) Vol I, p. 676.

[10] J. Bardy et al., Proc. of the 12th Int. FEL Conf., Paris 1990 (North Holland, 1991)
p 311.

[11] Ph. Eyharts et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) Vol I, p 670.

[12] J. de Mascureau et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) Vol I, p 697.

[13] S. Humphries, Principles of Charged Particle Acceleration, Wiley Interscience

Publication, New York 1986 p 283.

[14] D. Keefe, Induction Linacs, LBL 21 528 (1986).

[15] C.A. Kapetanakos and P. Sprangle, Physics Today, February 1985, p 58.

[16] D. Keefe, Particle Accelerator 11 (1981) 187.

[17] B. Kulke et al., LLNL, Proc. 15th IEEE Power Modulator Symposium,
Baltimore 1982.

[18] M. Burns et al., Proc. Int. Conf. on High Power Particle Beams, Washington 1992
p 283.

[19] R. Corsini, J. Gardelle, J. Grenier, Proc. of the Particle Accelerator Conf.,

Washington 1993 (IEEE 1993) Vol I, p 626.

[20] Y.S. Cho et al., Proc. of the Particle Accelerator Conf., Washington 1993
(IEEE 1993) Vol I, p 679.
CYCLOTRON MAGNET CALCULATIONS
Y. Jongen and S. Zaremba
Ion Beam Applications s.a.
Avenue Jean Lenoir 6
1348 Louvain-la-Neuve, Belgium
Abstract
Dierent aspects of a cyclotron magnet design, and not only calcula-
tions, are reviewed. The design is an iterative process starting from
a simple model which requires a vision of the complete cyclotron
and an integration of all subsystems. Finally detailed cyclotron
magnet calculations are described.

1 INTRODUCTION
Many accelerator physicists tend to merely reduce the cyclotron design to cyclotron
magnet calculations. This quite common error should be avoided. Therefore major issues
of cyclotron magnet design, not only calculations, will be discussed in this paper. The
type of cyclotrons discussed here are xed eld, xed frequency machines with azimuthally
varying magnetic eld (hills and valleys) provided by sectors (poles) and such as described
in textbooks, for example, in [1].
Cyclotron magnet design requires a vision of the complete cyclotron. Dierent sub-
systems of cyclotron (RF (radiofrequency) system, vacuum, injection, extraction) interact
with each other and always interact with the magnet. Constraints imposed by a previ-
ously completed design of the cyclotron magnet can create enormous diculties later
when designing other elements.
Like any other design it is natural to start with a crude and simple model which
is iteratively rened as the design process converges. Finally very detailed and time-
consuming calculations and an optimization of the cyclotron magnet are performed.

2 CYCLOTRON MAGNET QUALITIES AND FUNCTIONS

It seems that the time is over for cyclotrons to be designed and constructed mainly
for nuclear physics experiments. The last few years conrm that the majority of currently
constructed cyclotrons and probably all future cyclotrons will be designed for specic
applications, and generally, for a single, unique task.
The most important consequences of this trend are that:
 versatility of cyclotrons is no longer an important issue;
 cost, simplicity, reliability and low radiation doses are very important for the cy-
clotron users and for the cyclotron designers;
 weight, occupied space and consumed power are often less important to cyclotron
users than presumed by designers who see the minimization of these parameters
as an interesting challenge.
 Radioisotope production is the most frequent application of recently constructed
cyclotrons. Low kinetic energy protons or deuterons ( 40 MeV) are usually required
by radioisotope producers. On the other hand the therapeutic applications require high
kinetic energies of accelerated particles (up to hundreds of MeV).
A well designed cyclotron magnet should ensure:
 isochronous conditions during acceleration;
 axial (vertical) and radial focusing of the beam ;
 an operation point away from dangerous resonances or a fast passage of the beam
through the resonance(s) zone(s).

2.1 Isochronism and magnetic eld shape

The maximum kinetic energy of accelerated ions determines the magnetic rigidity
of particles. Magnetic rigidity indicates values of a bending radius and the magnetic eld
at this radius. A reasonable choice for the magnetic eld and the bending radius allows
a rst estimation of a cyclotron magnet radius. The magnetic rigidity as a function of the
kinetic energy is expressed by the equation:

B
 = 3001
Z
T 2 + 2
T
E0
h i
1=2
(1)
where:
B - magnetic eld (Tesla);
 - bending radius (m);
Z - charge state of particles, with respect to electron charge;
T , E0 - kinetic energy and the rest mass (MeV).
Ions should stay synchronized with a given phase of the cyclotron RF system during
acceleration. Therefore the magnetic eld should ensure a constant rotation period of ions
in the cyclotron. The magnetic eld with this feature is called isochronous.
The isochronous average magnetic eld increases with radius r.
v
u
u
B (r) = B0
(r) = B0
t1 +  2(r) (2)
1  2(r)
where:
 (r) = v(r)=c; c - light velocity and (r) = 1 + T (r)=E0;
B0 - the magnetic eld in the center of the cyclotron.
The requested shape of the magnetic eld can be obtained by dierent methods:
 using steel;
The increase of a ratio hill/valley with radius r (increase of the azimuthal length
of sectors) is the easiest way to obtain the isochronous magnetic eld. Another
possibility is to decrease the gap between poles with radius r (azimuthal shims) but
it is not recommended when a high beam intensity is required.
 OPERA-3d
Pre-processor 2.3

Figure 1: The removable pole edge attached to the magnet sector of the IBA C30 cy-
clotron. The mesh grid size is smaller in regions where high magnetic eld gradients are
present.

Magnet pole edges that can be dismounted, mechanically corrected and once again
assembled, provide a very eective way to produce the isochronous magnetic eld
shape, as in gure 1. This method almost always guarantees the convergence to the
isochronous magnetic eld when applied as an iterative process consisting of three
steps: (1) magnetic eld measurements, (2) correction calculations, (3) pole edge
mechanical correction. The correction of pole edges at the given radius r is based
on the relation between the relative frequency error of the particles and the relative
tolerance of the magnetic eld

B (r)=B (r) = 2(r)

fp(r)=fp (r) (3)

where:
B (r) - the magnetic eld error, B (r) - the magnetic eld value,
fp(r) -
frequency error and fp(r) - the rotation frequency of particles in the cyclotron.
Unfortunately these methods can only be applied to cyclotrons accelerating a single
type of particles to a xed nal kinetic energy. Dierent kinetic energies of ions
are provided by the beam extraction at dierent radii. In this case the extraction
of particles by stripping method is more suitable than by electrostatic de ection
method.
 using movable iron shims;
Moveable iron shims can be used when two dierent xed magnetic elds for two
dierent types of ions are required. The nal shape of iron shims is obtained using
the same procedure as described above. IBA Cyclone 18/9 and Cyclone 30/15
cyclotrons operate according to this principle, see gure 2.
 OPERA-3d
Pre-processor 2.3

OPERA-3d
Pre-processor 2.3

Figure 2: Moveable radial shims are placed in the middle of two opposite valleys, parallel
to the cyclotron median plane. Shims close to or far from the cyclotron median plane
produce the magnetic eld for H or D ions respectively.
  using trim-coils;
The exibility of the magnetic eld modied by trim-coils permits the acceleration
of dierent types of ions to dierent nal kinetic energies with the beam extraction
at the same nal radius. The extraction of particles by the electrostatic de ection
is often preferred in this case.
Trim-coils require careful tuning using an adequate number of power supplies. The
cyclotron is more complicated so its reliability diminishes. It is also more expensive
and usually consumes more power.

2.2 Tolerances and imperfections

Any discrepancy of a real cyclotron magnetic eld and the isochronous magnetic
eld results in a shift of accelerated particles with respect to the chosen phase of the
cyclotron RF system. An acceptable phase shift of the particles determines the tolerance
of the magnetic eld isochronism. The equation below gives a relation between the phase
shift and the relative magnetic eld imperfection:

d(sin ) = 2

h
BdB
n (4)

iso
where:
 - phase shift of particles;
h - harmonic mode of the cyclotron operation;
n - number of turns;
dB - amplitude of the magnetic eld imperfection;
Biso - nominal value of the isochronous magnetic eld
The phase shift is given by the integral of equation 4. Therefore small, often op-
posite signs and local imperfections of the magnetic eld can be accepted. They can
not be accepted when they create local, positive gradients of the magnetic eld resulting
in imaginary values of the axial betatron frequency z and consequently in axial blow-up
of the beam, see section 2.3.
The acceptable maximum value of the phase shift depends on the type of particle
extraction, the geometry of the accelerating electrodes (azimuthal length of the dees) and
the harmonic mode of operation.
Multiturn extraction allows larger phase shift. Values of the maximum accepted
phase shift are between 20 45 in currently constructed cyclotrons, assuming that par-
ticles are still suciently accelerated in all accelerating gaps. The tolerance on the mag-
netic eld
B=B  10 4 is frequently obtained by substitution of the accepted phase
shift to equation 4. Such requirements are possible to fulll for the cyclotron magnet and
its power supply.
For a single turn extraction without at-topping, the accepted values of the phase
shift are much smaller, in the range of few degrees. The substitution of such phase shift
gives the tolerance of the magnetic eld
B=B  10 7 . Clearly it will be dicult to create
the magnetic eld with such stability and to nd the magnet power supply providing these
characteristics.
The cyclotron magnetic eld B changes due to changes of the excitation current I
of the main coil(s). Generally in measurements
B=B are smaller than
I=I due to the
saturation of the magnet steel. A value of
B=B greater than
I=I more than likely
indicates problems with the cyclotron magnet and/or the measurements system.

2.3 Axial (vertical) and radial focusing of the beam avoiding dangerous res-
onances
Axial (vertical) focusing of the beam can be provided by:
 negative eld index i.e., the magnetic eld decreasing with the radius;
 azimuthally varying magnetic eld i.e., magnet sectors creating hills and valleys
of the magnetic eld;
 spiralization of magnet sectors.
The betatron frequencies can be approximated by the following equations [2] :

z2 = k + N N
2  
2 1

F
1 + 2
tan2  (5)

r2 = 1 + k + (N 2 31)

N(N 2 4)
F
1 + tan2 

2  
(6)
where:
z - axial (vertical) betatron frequency;
r - radial betatron frequency;
k = r=hB i
dhB i=dr - radial eld index;
N - number of symmetry periods of the cyclotron;
 - sector spiral angle;
F - magnetic eld utter at the radius r which can be dened using the calculated
averages of the magnetic eld hB i and of the magnetic eld square hB 2i at the radius r.
The cyclotron magnetic eld at the given radius r, having the N -fold rotational symmetry,
can be presented as Fourier series. The utter of the magnetic eld can also be dened
using coecients of Fourier series Ai; Bi; i = 1; 2; ::::

F = hB ihB i2hB i = 12
(A2i + Bi2)

2 2 1
X
(7)
i=1

The presented analytical formulas 5{7 are approximative. Correct values of the
betatron frequencies have to be calculated by programs tracking particle trajectories.

2.4 Resonances
Resonance is encountered when
K
r + L
z = P (8)
where K , L and P are integers. P is the symmetry of the driving term and the sum
jK j + jLj is the order of the resonance. Resonances of order 1, 2 or 3 are driven by a dipo-
lar, quadrupolar or sextupolar component of the guiding magnetic eld respectively.
 The amplitude of betatron oscillations grows at a resonance. Imperfection resonance
(K
r = P , L
z = P ) eects are dependent on the amplitude and the radial extension
of imperfection in the cyclotron. Coupling resonances (K; L 6= 0) are important when
large radial oscillations are generated. The axial acceptance of the cyclotron is generally
much smaller than the radial and the beam is easily lost. The third order coupling
Walkinshaw resonance r 2z = 0 frequently encountered in cyclotrons is considered
as one of the most destructive.
The operation point (r ; z ) moves as the particle kinetic energy increases and can
cross the resonance line(s). Resonances are dangerous when the crossing is slow (low
kinetic energy gain per turn) and/or when the driving term is large but the fast passage
through the resonance zone does not greatly deteriorate a beam quality.
It is possible to show (e.g. [1]) that the eld index is k = 2 1, when the magnetic
eld is isochronous. Therefore the radial betatron frequency r (equation 6) can be
expressed as r  +


, where dots replace terms dependent on N , F and .
Fundamental resonances given by the equation r = N=2 determine the minimum
number of symmetry periods (sectors) of the cyclotron. The frequency of radial betatron
oscillation starts from the value of 1 ( = 1) at the cyclotron center. Cyclotrons with
N = 2 sectors will be unstable from the center and therefore they do not exist.
Due to the same fundamental resonance the theoretical maximum kinetic energy
of protons is limited to 469 ( = 3=2) and 938 MeV ( = 4=2) for cyclotrons with
3 or 4 symmetry periods (sectors) respectively. Practically the maximum kinetic energy
of particles will be smaller because terms dependent on N , F and  increase the radial
betatron frequency r .
Very high kinetic energies require more symmetry periods in the cyclotron. To avoid
the fundamental resonance r = 3=2, it was necessary to mill a deep groove in each of three
sectors of the AGOR cyclotron [3]. This ingenious solution multiplied by two the number
of sectors, without a signicant decrease of the average magnetic eld in the cyclotron.

3 MAGNET DESIGN
3.1 Decision strategies
Decisions need to be taken concerning the choice of:
 gap between magnet poles (hills of the magnetic eld);
 magnetic eld level between magnet poles (hills of the magnetic eld);
 magnet pole radius;
 gap in the valleys;
 number of symmetry periods (sectors).

3.2 Gap between magnet poles

Very often the choice of the gap between magnet poles is a compromise between the
two situations presented below.
 A small gap between magnet poles reduces the required number of ampere-turns
in the cyclotron main coil and allows the last orbit of accelerated particles to be brought
closer to the pole edge. This reduces the radius of the pole and consequently diminishes
the overall weight of the magnet which is approximatively proportional to rpole 3
. A small
axial gap limits the radial extension of the magnetic eld hole in the cyclotron center
created for the central region elements or the axial injection system. The defocusing
radial gradient of the magnetic eld in this region can deteriorate the beam by axial
blow-up.
A large gap between magnet poles increases the available space for the beam os-
cillating in axial (vertical) plane. Also such a gap facilitates the installation of dierent
elements of the cyclotron:
 diagnostic probes
 stripper(s)
 central region components: ion source(s) or in ector, dee bridge
 electrostatic de ector(s) in superconducting cyclotrons
It is never possible to keep the cyclotron eld isochronous up to the limit of the
magnet pole. The magnetic ux beyond the last orbit is useless but has to be guided
through the return yoke. An elliptic gap between the poles (closed beyond the last orbit),
allows the last orbit to be brought within a few millimeters of the pole edge according
to the magnetostatic theorem that the ellipsoid void in an innite bloc of uniformly
magnetized steel will exhibit a constant magnetic eld, dependent on the shape of the
ellipsoid.
In this case the magnetic eld is isochronous to the magnetic shunt closing the gap.
Unfortunately, a small tunnel has to be made for the extracted beam. The magnetic eld
in the tunnel is dicult to calculate and even more dicult to measure. Calculations
show that gradients of this magnetic eld are strong and nonlinear.

3.3 Number of sectors

The choice of the number of sectors is quite controversial. At rst sight three
sectors (symmetry periods) seem to be the best choice for low kinetic energy cyclotrons.
Such structure was often used in old cyclotrons where the large gap between magnet
poles provided many possibilities for the installation of the necessary subsystems. Also
it was used in recently constructed super-conducting cyclotrons where the gap was small.
Super-conducting magnet valleys were lled with the RF system cavities and the small
axially extraction system was installed between magnet poles. A disadvantage of this
solution is that three coupled resonators usually oscillate in phase so only harmonic modes
H = 3; 6, ... are possible.
Our personal opinion is that four sector (four-fold rotational symmetry) geometry
of the cyclotron seems to be more practical. Two opposite valleys can be used by RF
cavities operated in phase in harmonic mode H = 2; 4; 6, ... or at 180 in H = 1; 3; 5,...
Two other valleys can be used for other devices: ion source, electrostatic de ector, internal
target or movable iron shims.
3.4 Initial magnet calculations
Initial calculations start from the estimation of the cyclotron magnet pole dimensions
using equation 1. The chosen average magnetic eld value hB i at the extraction radius
is the basis of the rst estimation of the magnetic eld in the hill Bhill , in the valley Bvalley
and the fraction of the hill  on one symmetry period according to the equation:
hB i = Bhill + (1 )Bvalley (9)
After adding the chosen number of sectors N one can verify the focusing properties
of the cyclotron. The utter F and slightly optimistic the betatron frequencies r ; z
can be determined from the analytical equations 5{7. The reasonable minimum value
of the axial betatron frequency z is about 0:10 0:15 when one wants to keep a small
gap between magnet poles. It is necessary to change one or more parameters or introduce
a spiralization of sectors when the value of z is too small. Then new iteration of analytical
calculations can follow.
Schematic drawings can facilitate the next step of magnet calculations. First it is
necessary to estimate a total magnetic ux in the cyclotron using the average magnetic
eld hB i and the pole radius. The pole dimensions are extended beyond the last orbit.
The part of magnetic ux of this zone can be redirected by adding a chamfer on the outer
pole edge. It should be remembered that in the median plane the eective eld boundary
is about 0:6 gap width outside the pole edge. Then, the value of the magnetic eld in the
ux return (return yoke) has to be determined taking into consideration the magnetic
properties of the used steel. The ux return shape can be designed when cross-sections
of the ux return and the coil are known.
The coil cross-section is calculated from the number of ampere-turns in the cyclotron.
The simple equation n
I  8
105
B
l where n
I is the number of ampere-turns (A), B is
the magnetic eld (T) and l is the gap width (m) which determines the number of ampere-
turns in the cyclotron gap. Then a certain number of ampere-turns for the steel should
be added. Added ampere-turns can be calculated assuming the constant value of the
magnetic permeability in the magnet steel, or just estimated. Our experience conrms
that it is dicult at this moment to nd a correct value of the magnetic permeability
due to strong non-linearities of the magnetic eld in magnet steel. An estimation can be
made by adding a generous percentage of ampere-turns to avoid later problems with the
coil power supply (e.g. 30% 50% of ampere-turns). The total number of ampere-turns
gives the choice of coil cross-section and the current density. From the current density
one can estimate the coil cooling requirements. Coil shape design is possible with the last
information.
After nishing this step of magnet calculations, it is necessary to develop a complete
preliminary shape of the cyclotron with all subsystems included. Holes and ports required
by other cyclotron elements have to be taken into account to evaluate changes in the
cyclotron magnet structure.
The calculations described above are simple, analytical and allow an initial layout
of the cyclotron magnet. More sophisticated tools are required for further optimization
of the cyclotron magnet.
 40.0 UNITS
Length : MM
Z [MM] Flux density : TESL
Field strength : AM
35.0 Potential : WBM
Conductivity : SMM
Source density : ACM2
Power : WATT
30.0 Force : NEWT
Energy : JOUL
Mass : KG

25.0
PROBLEM DATA
C235PROTONTHERAPY
Linear elements
20.0 Axi-symmetry
Modified R*vec pot.
Magnetic fields
5796 elements
15.0 3018 nodes
30 regions

10.0

5.0

0.0
1065.0 1075.0 1085.0 1095.0 1105.0 1115.0
R [MM]

OPERA-2d
Pre and Post-Processor 1.1

Figure 3: The detailed discretization of the elliptic gap between cyclotron magnet poles
in the 2-D model. The cyclotron median plane determines a lower boundary line of the
model.
3.5 Detailed magnet calculations
Detailed magnet calculations, using a nite elements method | or a similar method
with the same kind of accuracy | must rst conrm the validity of the initial design.
They are then mainly used to uncover weak points of the design and help correcting them.
Finally, they can serve to optimize the design taking into account all important and re-
quested particular characteristics of the magnet. Detailed magnet design and calculations
require the use of two-dimensional (2-D) and three-dimensional (3-D) computer codes.
The list of available programs is presented in [4],[5].
The maximum number of elements in two- and three-dimensional calculations is
of the same order. The two-dimensional model is only the plane and the three-dimensional
model should be a volume comprising the symmetry period of the cyclotron (a part
of it, when mirror symmetries are possible). Hence two-dimensional calculations oer
the possibility of creating elements having much smaller grid size than three-dimensional
ones and ensure a better accuracy of cyclotron magnet details. Even with smaller grid
size of the 2-D model fewer nodes will be used than in the 3-D model.
Figure 3 presents the part of the two-dimensional model mesh in the IBA C235
protontherapy cyclotron with the closed elliptic gap. The same geometry model was
developed and calculated in three dimensions. The analysis of the same plane in the 3-D
model shows that one element only represented the part of the cyclotron magnet pole
between radii 1100 and 1105 mm and up to 6 mm from the cyclotron median plane.
The execution time of programs 2-D and 3-D is proportional to n3 where n is the
number of created node points. Only application of all possible symmetries of the cy-
clotron magnet signicantly contributes to a decrease of the total number of nodes in the
3-D model. Sometimes the construction of 3-D models which do not respect strictly
model symmetries can decrease the number of created nodes and the computing time
but the solution should be treated with limited condence. Often excessively long com-
puting times (tens of hours) and the impossibility to create more accurate model due to
a lack of computer memory, encourage designers to develop more exactly two-dimensional
models.
Conductors created in two-dimensional calculations are very simple and their mag-
netic eld is calculated very quickly. Programs 3-D oer, except for a library of the
standard shape conductors, practically unlimited possibilities for creating shapes of con-
ductors as a sequence of bricks with variable cross-section. The magnetic eld of coils
having very exotic shapes (e.g. particular trim-coils) can be calculated this way but take
a hopelessly long time (tens of hours).
2-D codes essentially solve x-y problems in which the third dimension is innite or at
least long enough to be considered as innite. Programs 2-D serve also to solve axisym-
metric (cylindrical) three-dimensional problems. Cyclotrons with an azimuthally varying
eld are hardly axisymmetric, but the method of stacking factors can be used in regions
where the magnet structure diers from axisymmetric (sectors and valleys, important
holes in the yoke). Figure 4 shows the application of this method. The stacking factor
SF is dened as the fraction of the circle occupied by the real ferromagnetic material.
Properties of the pseudo-material lling regions with the stacking factor are given by the
B-H curve equation [6]:
Bpseudo = 0
H + (B 0
H )  SF (10)
where B and H are related by the curve of B-H for the real ferromagnetic material.
Programs 2-D are useful to study details or local modications which produce local
eects. At rst the large model solution is calculated. Then a zoom of the area of interest
from the large model is made (as e.g. in gure 3). Vector potential values corresponding to
the large model are assigned to boundary points of the zoomed model. The analysis of the
eects of perturbations has to take into account the conservation of the total magnetic
ux which results in a balance of a magnetic eld increase at some places and a magnetic
eld decrease somewhere else in the model.
Three-dimensional calculations are more time and memory-consuming than two-
dimensional. Three-dimensional models are the closest to reality. Therefore results of mea-
surements are expected to be close to results of calculations. It was observed in many
cases that absolute results of calculations and measurements dier by 2 3%. This dier-
ence can be explained by the absolute numerical precision of the software, by dierences
of the B-H curves and still some geometrical dierences between the model and the real
cyclotron magnet. Grid size dimensions are smaller in regions where high magnetic eld
gradients are expected (see e.g. gures 1, 2). Despite grid size variations small disconti-
nuities of the magnetic elds are sometimes observed due to the limited number of nodes
in the model mesh.
 800.0 UNITS
Length : MM
Z [MM] Flux density : TESL
Field strength : AM
700.0 Potential : WBM
Conductivity : SMM
Source density : ACM2
Power : WATT
600.0 Force : NEWT
Energy : JOUL
Mass : KG

500.0
PROBLEM DATA
C235PROTONTHERAPY
Linear elements
400.0 Axi-symmetry
Modified R*vec pot.
Magnetic fields
5796 elements
300.0 3018 nodes
30 regions

200.0

100.0

0.0
0.0 100.0 300.0 500.0 700.0 900.0 1100.0
R [MM]

OPERA-2d
Pre and Post-Processor 1.1

Figure 4: Dierent stacking factors in distinct regions simulate the variation of the
hill/valley ratio of the IBA 235 MeV protontherapy cyclotron featuring an elliptical gap
between its poles.

The comparison of dierences between calculations of two similar models and two
corresponding measurements (e.g. dierent coil current, shift of movable shims) very often
shows a perfect agreement [7],[8].
Results of 2-D and 3-D magnetic eld calculations are used to create maps of the
magnetic eld in the cyclotron median plane or in three-dimensions around this plane.
Then the map of the magnetic eld becomes a part of input data to programs calculating
trajectories of particles and properties of the magnetic eld [5], [9]. The most important
part of the output of these programs present values of the frequency error, the integrated
phase shift of particles and betatron frequencies. The magnet model has to be changed
when one or more parameters are not acceptable. Our experience shows that during the
creation of the 3-D model one should foresee possibilities of unexpected changes of the
model geometry without the necessity to construct the model from scratch once again.
Fully acceptable values of the frequency error and the integrated phase shift of par-
ticles ensures the isochronism of the magnetic eld. Correct values of betatron frequencies
conrm the focusing properties of the magnetic eld and the absence of dangerous reso-
nances during acceleration.
Magnetic calculations are followed by the analysis of mechanical deformations due
to the weight of magnet elements, magnetic forces between them and forces of the air
pressure when the vacuum is produced in a cyclotron vacuum chamber. The request
for minimal deformation and practical aspects of handling determine the partition of the
magnet.
References
[1] J. J. Livingood - Principles of Cyclic Particle Accelerators; D. Van Nostrand Comp.,
1961.
[2] H. L. Hagedoorn and N. F. Verster - Orbits in an AVF Cyclotron; Nuclear Instru-
ments and Methods 18,19 (1962) 201{228; North-Holland Publishing Co.
[3] A. Laisne - The realisation of the magnetic circuit of the AGOR cyclotron; Pro-
ceedings of the 2nd European Particle Accelerator Conference, EPAC 90, Nice, June
12{16, 1990, p. 1172{1174.
[4] G. G. Molinari - Existing Commercial Tools and Criteria of Selection; in Short Course
Computer Aided Electromagnetic Analysis and Design: a Technical Introduction,
Munich, November 1990.
[5] Computer Codes for Particle Accelerator Design and Analysis: A Compendium; LA-
UR-90-1766, Los Alamos Accelerator Code Group, Second Edition, May 1990.
[6] W. Beeckman - PE2D Modelling of the Cyclone230 Cyclotron; part of Cyclotron De-
sign with PE2D and TOSCA softwares at IBA; Proceedings of VECTOR FIELDS Eu-
ropean User Meeting, Ion Beam Applications, Louvain-la-Neuve, Belgium, September
1992.
[7] S. Zaremba - TOSCA Modelling of the Cyclone30 Cyclotron; part of Cyclotron Design
with PE2D and TOSCA softwares at IBA; Proceedings of VECTOR FIELDS Euro-
pean User Meeting, Ion Beam Applications, Louvain-la-Neuve, Belgium, September
1992.
[8] W. Beeckman and D. Vandeplassche - private communication.
[9] M. M. Gordon - Computation of Closed Orbits and Basic Focusing Properties for
Sector-Focused Cyclotrons and the Design of \CYCLOPS"; Particle Accelerators,
vol. 16, p. 39{62, 1984.
INJECTION INTO CYCLOTRONS
P. Mandrillon
Laboratoire du cyclotron, Centre Antoine Lacassagne, Nice, France

Abstract
A great variety of injection processes exist depending on the type of
cyclotron considered. Designing the central region where the injection
is achieved requires careful studies which are briefly presented.
Among the possibilities for external injection in compact machines, the
axial method which is frequently used is discussed in more detail.

1. INTRODUCTION

Most of the early cyclotrons and synchrocyclotrons were equipped with internal ion
sources of the open type, as shown on Fig. 1, located below the median plane of acceleration of
the particles. This early "axial injection" mode gave a very poor beam quality.
Therefore the early selected points of
attack for a great improvement of the internal
beam quality of cyclotrons was the central
region including the internal source itself and its
immediate surroundings. It was rapidly
understood that a spatially-defined, hooded-arc
source (Fig. 2a and 2b) permits the dee and
dummy dee electrodes to be brought very close
to the source slit, leaving a sufficiently narrow
acceleration gap to allow full initial energy pick-
Fig. 1 Old open source of the CERN Synchrocyup. This improvement programme was carried
out at the CERN 600-MeV synchrocyclotron and brought outstanding performances: the
internal beam intensity was increased by a factor 10 (the repetition rate of the frequency
modulation programme was also increased), the radial oscillation amplitude was reduced by
another factor 10. The resulting extraction efficiency was therefore raised from 3% up to 70%.

The acceleration of heavy ions or polarised particles which require ion sources which
could not be confined in the median plane of the cyclotron led to external injection by: either
entering the cyclotron perpendicular to the median plane, the so-called axial injection, or
reaching the central region of the cyclotron in the median plane, the so-called horizontal
injection.

Nowadays many types of cyclotrons are running, dedicated to specific aims. Even
cyclotrons could be used as injectors, bringing the beam in on the inner radius of separated-
sector booster cyclotrons. The PSI four-separated-sector-injector cyclotron [1] which produces
a 72-MeV, 1.5-mA proton beam to feed the 590-MeV, 8-sector ring cyclotron is certainly the
best example of what could be achieved as a sophisticated high energy injection process.

2. THE CENTRAL REGION

The injection process should bring the beam into the central region of the cyclotron where
the transfer of the injected beam into the cyclotron stable-acceleration region should be
Fig. 2a Vertical cross section of the central Fig. 2b Horizontal motion of the reference
region of the improved CERN SC particle in the equipotential
distribution of the confined central
region of the CERN SC

achieved. It is the region where the beam quality is defined. The design of this region requires
careful studies which are briefly presented here.

This region extends to a radius where the beam dynamics are no more governed by the
complex 3-dimensional electric field produced by the accelerating electrodes but mainly by the
properties of the magnetic field which are not phase-dependent. Therefore the quality of the
beam produced by a cyclotron is the result of the careful design of this most critical part of this
type of accelerator. Of course the design and the studies to be carried out are strongly
dependent on the injection energy. Hence two kinds of central region should be considered
depending on the type of cyclotrons: Compact machines where the injection process is at low
energy and separated-sector machines where it occurs at high energy.

2 . 1 Compact cyclotron

Several steps are required to design the geometry of the central region.

2.1.1 First phase: Hard-edge gap approximation

The initial arrangement of the electrodes (the so-called dees and anti-dees) is defined,
assuming hard-edge accelerating gaps with a uniform electric field and a uniform central
magnetic field.

The general equation of motion of an ion (q, m) accelerated by an electric field

E = E0 sin(h + 0 )ei is:

d 2 x qE0 qB dy
2 = sin cos +
dt m m dt
(1)
d 2 y qE0 qB dx
= sin sin
dt 2 m m dt
 RF
where h = harmonic number =

= angular phase = t
= electric phase = h + 0
= argument of the electric field

= +
2 2
= angular aperture of the dee

Fig. 3 Hard-edge gap

The solution of (1) is easily obtained using the complex Z = x + iy

dZ
+ iZ = Kei cos(h + 0 ), K = cste (2)
d
The motion of the centre of curvature C could be easily interpreted:

dZ dZ
= i(Zc Z) iZc = + iZ (3)
d d
Hence:

Zc ( f ) Zc (i) = iKei if [cos(h + 0 )] (4)

r r
Equation (4) shows that the displacement of C, Ci C f , during the gap traversal is perpendicular
r
to E . Therefore, the centre-of-curvature motion could be constructed very easily using this
property by a simple algorithm.

Figure 4 presents the results of these simple hard-edge geometrical constructions in the
median plane of a three-dees geometry. The injection point I (Fig. 4, top) could be easily
determined by minimizing the distance of the barycentre of the centres of curvature motion
(Fig. 4, bottom) to the machine centre.

2.1.2 Second phase: Orbit calculations in the real gap

With these initial gap positions a geometry is constructed and the distribution of the
equipotential is calculated using a 3-dimensional Laplace equation solver. A convenient method
[2, 3] to calculate the orbit in the complex electric field distribution is to make use of the
equipotential-lines distribution instead of calculating the components of the electric field. The
acceleration gap is decomposed into micro-gaps defined by two adjacent equipotential lines in
the median plane, the direction (but not the module) of the electric field is calculated and the
motion of the orbit centre within the micro-gap is deduced by a simple geometrical construction
as shown in Fig. 5.
 Fig. 4 Orbit and centre-of-curvature motion in the hard-edge gap
approximation of a three-dees geometry

Figure 6 presents a typical reference trajectory in the median plane of a superconducting

cyclotron using three accelerating electrodes, three deltas, with the equipotential distribution in
the six gaps. The advantage of this method of discretization compared to the usual integration
of the equations of motion using the whole information of the electric field is its improved
speed and its easy physical interpretation. This method gives a direct insight into the motion of
the orbit centre during the gap traversal, making modifications easier to improve the centreing
of the beam. In this second phase several readjustements of the geometry are needed to
optimise the shape of the electrodes.
 i = transit Pi Pi+1
Egi = energy gain in the micro-gap

= Vii+1 cos (h i + i )
2 2Egi E0
i+1 = i + B2 c 2
2

0
r
C C perpendicular Er P
i i+1 ( i+1 )
i+1 = i + h i
Fig. 5 Micro-gap traversal E0 = rest mass of the particle
c = velocity of light

Fig. 6 Horizontal motion of three particles in the real gaps of a three-dees geometry (the
reference central trajectory and the 50 mrad orbits)
2.1.3 Last phase: Vertical motion
 The electric axial focusing or defocusing is strong in the injection region where the
velocity of the particle is small. These effects are strongly dependent on the phase of the ions
with respect to the RF voltage. They have been analysed in many basic papers using the thin-
lens approximation in a 2D geometry.

The two types of effects which should be considered for the vertical motion during a dee
traversal are presented in Fig. 7. The spatial effects due to the typical equipotential distribution
are shown in the vertical cross section of the dee-dummy dee system (top). When the particle
enters the gap (point E) the field is focusing. The time variation effect of the electric field
during the gap traversal is shown on the bottom figure. For a given dee aperture and an
acceleration on the harmonic mode h, the superposition of these two effects could be interpreted
as a succession of alternatively focusing and defocusing thick lenses.

Fig. 7 Vertical motion during the dee traversal

The precise analysis of these effects requires a detailed knowledge of the 3-D potential
distribution. The discretisation method for the horizontal motion (micro-gaps) has been
extended by J.P. Schapira to 3-dimensions in the framework of the AGOR superconducting
cyclotron central region studies.

The complete analysis of the vertical motion requires detailed knowledge of the magnetic
field map. Local modifications of this field might be necessary to increase the magnetic
focusing in the first turns of the injection region (the so-called field bump, centring coils etc.).
Therefore, in the final phase of the central-region calculations, the adjustment of the magnetic
field will imply slight modifications of the gaps.
2 . 2 Separated-sector cyclotron (SSC)
 The isochronous cyclotrons considered here consist of magnet sectors separated by empty
valleys. The radiofrequency accelerating structures, which could be either delta (providing two
gaps) or single-gap cavities, are situated in the valleys, between the magnets. Therefore the
magnet gaps could be quite small. This fact improves the vertical focusing properties. The
SSC require an injected beam at low energy and are proposed when a multi-stage acceleration is
necessary. A great variety of injectors are still existing for separated sector cyclotrons: external
sources at high voltage (e.g. the Cockroft-Walton at PSI), tandem Van de Graff (e.g. the
injector of the HMI SSC in Berlin), together with cyclotrons (the injector of the GANIL SSC in
France or the injector of the 200-MeV SSC in South Africa), etc.
The injection in these machines should bring the beam through a magnetic field-free
central region and then guide it into the first centred orbit. The injection process is often quite
difficult because the magnetic field should bend the injected beam more than the internal beam.
The injection layout is made of different kinds of channels: either electrostatic channel or
magnetic channel to bend the beam on the first trajectory. In this type of cyclotron, the large
space available between the sectors provide the necessary space to house powerful accelerating
structures and the radial separation between the injected orbit and the inner turn in the cyclotron
is sufficiently large to install injection septa.
Therefore to optimise the layout of the injection systems it is necessary to study the
following aspects.
beam optics and matching between the injector and the accelerator: the matching of the
injected beam properties to the so-called eigen-ellipses of the betatron transverse phase
space of the cyclotron orbit, is an essential condition for getting a good beam quality.
Figure 8 presents the typical eigen ellipses for a four-sector magnet cyclotron at the centre
of the hill and at the centre at the valley.

Fig. 8 Eigen ellipses for a four-sector magnet cyclotron

shaping the injection channel which is usually made of different bending and inflection
elements. In these calculations the stray field of the magnet sector should be taken into
account.
 mechanical implementation of the elements: the required space to house the bending
elements and the necessary diagnostic stations to measure the beam quality makes the
mechanical design of this region difficult and the central region of separated sector
cyclotrons is often crowded.

Figure 9 shows the typical example of the central region layout of the four separated-
sector 200-MeV cyclotron in Faure [4] (NAC, South Africa). The beam is pre-accelerated by
an 8-MeV compact cyclotron and is injected radially along a valley into the central region. The
injection system consists of three injection elements: two bending magnets (BM1 and BM2)
and a magnetic inflection channel (MIC) which is mounted in the pole-gap of a sector magnet.

Fig. 9 Injection layout of the four-sector NAC-SSC

Since the eigen-ellipse transformation and the position of the injected equilibrium orbit are
defined by the magnetic properties of the sector magnet, it is important to inject the beam with a
phase ellipse (for the two transversal phase spaces) corresponding to an eigen-ellipse, exactly
onto this orbit to obtain the best quality. This ensures periodic beam envelopes inside the
cyclotron with a minimum of amplitude modulation and a good matching of the injected beam
[5] to the acceptance of the cyclotron.

Numerical orbit calculations take into account that fringe fields and trajectories could be
traced backwards from the centred orbits in the cyclotron, e.g. at the first valley centre line.
For the geometry shown in Fig. 9 the phase ellipses at three angles ( = -45, 0 and 45) are
presented in Fig. 10.

3. INJECTION FROM EXTERNAL SOURCES

The need to inject beams of heavy ions, negative ions and also polarised ions has led to
many different injection concepts, either in the median plane of the cyclotron or axially. High-
energy injection, heavy ions can be injected in the horizontal plane to reach the required radius
and then be stripped and placed onto the stable orbit from which they can be accelerated. This
method, proposed by C. Tobias [6] was developed at Orsay [7]. It is still successfully used
[8, 9] for injecting heavy ions from a tandem into a superconducting cyclotron.

In this section the axial method which is widely used will be discussed in more detail.
Horizontal injection processes are also presented.
 Fig. 10 Phase ellipses along the injection trajectory of the NAC SSC

3 . 1 Axial injection

The first successful axial injection was achieved by Powell [10] at the Birmingham
isochronous cyclotron. It is now widely used [11] for various reasons:

the emergence of new sources: ECR source which produces heavy ion beams with
higher charge states than the traditional internal PIG sources, polarised ions source and
multicusp for the production of negatively charged ions which are cumbersome and
therefore which could not be installed in the central region of the cyclotron.

obtaining a high quality of beams: getting a high-brightness beam accelerated by the

cyclotron requires a 6D matching, i.e. in particular using a buncher operating at the
cyclotron frequency.

increasing the reliability of the cyclotron: better vacuum for H- life-time of the ion
source.

An axial injection is made of three parts:

the external beam transport from the ion source up to the top of the magnet yoke, i.e. a
region where the stray field of the cyclotron is negligible.

the transport inside the yoke where the solenoidal field of the cyclotron should be
considered.
the inflector and the central region of the cyclotron where the beam is bent into the median
plane and further accelerated. These two regions should be considered together.
3.1.1 The external transport line
 Optical systems used for beam transport and matching are standard. Nevertheless
provision should be made not only for achromatism but also for beam emittance matching to the
eigen-ellipse at the entrance of the magnet yoke.

3.1.2 Transport inside the yoke

This transport line is made of a bunching section and of focusing elements. The axial
magnetic field should be taken into account. In the design phase this field could be obtained
with a fairly good accuracy by using the POISSON programme for solving this 2-D cylindrical
magnetostatic problem.
Several problems occur during this transit through the yoke, mainly a degradation of the
beam emittance due to the coupling of the two transverse motions and some debunching effects
because of the different length of orbit between the buncher and the median plane due to the
helical motion of particles in the axial field. For a high quality injected beam, short bunching
distances and low divergence are recommended. Figure 11 shows the axial injection layout of
the MEDICYC cyclotron in Nice.
3.1.3 The inflectors
Three types of electrostatic inflector could
be considered:
the electrostatic mirror which is made of
two plates inclined at about 45. An
opening must be provided in one of the
electrodes for the entrance and the exit of
the beam. In this type of inflector the
particle does not travel on an equipotential
plane and therefore the velocity changes
during the transit.
the spiral inflector developed in Grenoble
by Pabot-Belmont [12] is in fact a rotating
cylindrical deflector to take into account
the magnetic field of the cyclotron. The
velocity remains constant as the particle lie
on an equipotential surface, the electric
field being always perpendicular to the
velocity.
the hyperboloid inflector proposed by
Muller [13] which is the simplest of a
family of second-order surfaces. Here
also the velocity is constant.
For each type of inflector the central
trajectory of the particle is obtained by the
integration of the general equation:

F = q E+ q v x B (5)
the electric field E required to bend an ion
(q, m) is given by :
Fig. 11 The MEDICYC axial injection line qB2 F( ) / m

whereB is the central magnetic field,

the magnetic radius,
 the transit angle in the inflector, characteristic of the inflector type given in the
following table.

Inflector type Transit angle

Mirror 1/ <
Spiral /2 no limitation
Hyperboloid /2 6
fixed, =
2
3.1.3.1 The mirror

By analytical integration of Eq. (5) with B, E y and Ez constant we get the parametric
equations of the central trajectory.

x = Ey B ( t sin t)

y = Ey B (1 cos t) (7)

z = ( / 2B)Ez t 2 v0t + z0

with = qB m

The projection on the horizontal plane x0y is a part of a cycloid generated by the circle of radius
Ey B. For a +90 bend where z = z' = 0 at the exit of the mirror, the geometry is fixed by
the two parameters (angle respect to 0z) and the height h = z0 of the entrance point.

Ey sin 2
tg = =
Ez 2 (8)


h = z0 =
2

The beam entering the inflector is characterised by trajectories dispersed in position, angle,
energy and time with respect to the central ray given by (7). Therefore the beam is specified by
six parameters (two transversal and the longitudinal planes). The transfer matrix is given by
[14]:
 S 0 c 2 s 0 0
c s
0 c st 0 t


2 0 0 0 0
T= s
(9)
2t 1
2s 2st
0 0 0


c 2s s 0 1 0
2s
0 0 0 0 0 1
Where s = sin 2, c = cos 2, t = s / c..

The electrostatic mirror is attractive for its simplicity (easiness of construction) and for the small
space required to house it in the central region of the cyclotron. This last property could make it
a good choice for a superconducting cyclotron where the available space is small.

3.1.3.2. The spiral inflector

This is characterised by the fact that in the vertical plane, the electric field is radial and
perpendicular to the velocity. Under the influence of the vertical magnetic field this plane
rotates. Integration of (5) gives:

a 2 cos(K + 1) cos(K 1)
x= 2 +

2 1 K K +1 K 1

a sin(K 1) sin(K + 1) (10)
y=
2 K 1 K + 1
z = a(1 sin )


Where a = radius of curvature of the electric field:

v0t
= vertical angle
a
a
K=

Therefore this deflector is also defined

by two parameters.

The first-order optical properties of the spiral inflector could also be derived by using a
transfer matrix [15] which is more complicated to obtain than in the case of the mirror. The
equations cannot be integrated analytically and are calculated by numerical methods. The
s s 1
transfer matrix F for a small step becomes 1 + F.s where C = cos , S = sin , k =
a a

0 1 Ck 0 0 0
S 2 k 2 0 Sk a 0 0 Sk

Ck 0 0 1 0 0
F=
0 2 a
(11)
Sk a 0 0 0
Sk 0 1 a 0 0 1

Ck a 0 0 1 a 0 0

3.1.3.3 The hyperboloid inflector

A hyperbolic potential given by the following expression

1 1
V= KZ 2 Kr 2 (12)
2 4

constant is the simplest potential which satisfies V = 0 and which possesses a radial
symmetry. The parametric equations of the central trajectory which is on an equipotential
surface are

{ }
1/ 2
r = r0 1 + 1 / 2sin 2 ( kt )

( )
= 6 / 2 kt tan {(
1
) }
6 / 2 tan( kt ) (13)

z = (1 / 2)r0 sin( kt ), (0 kt / 2)

where t is the independent time variable, and v0 and k are constants which are determined by
the constraints
r0 = 2 6 ( mv0 ) / qB = 2 6Rm
(14)
k = qB / m 6 = / 6

where m is the ion mass, q is the ion charge, and v0 is the injection velocity of the ion.

The main advantage of this inflector is its simplicity and good optical properties. It has a
fixed transit time and then only one adjustable parameter. Therefore it is often used with an off-
axis injection. Its dimensions are much larger than the dimensions of the two other inflectors
and require more space in the central region. Usually it cannot be used in superconducting
compact cyclotrons.
3 . 2 Horizontal injection
 Among the possibilities for injection in the median plane of compact cyclotrons are high-
energy injection of heavy ions by change of charge in the central region, cyclodal motion along
a sector edge, while another simple possibility for injecting protons or deuterons is neutral
beam injection.
3.2.1 Injection of heavy ions by stripping
The principle of this method is quite simple
and is illustrated in Fig. 12. The parameters
of this injection method could be very easily
calculated using a hard-edge approximation
which assumes a uniform field B in the disk
of radius R e corresponding to the extraction
radius. A beam of ions (mass A, charge Zi)
is injected with an incident angle . If we
assume a non-relativistic beam, we obtain
the following equations:
Ti
Stripping radius rs = Re
Tf

Zs R2 + rs2 (2Zs / Zi 1)
Fig. 12 Injection of heavy ions by stripping i = r s ; sin = e
Zi 2Re rs Zs / Zi

with Zs the electric charge after stripping and Zi the electric charge before stripping.

3.2.2 Injection by precession along a sector edge

A horizontal injection scheme without channels has been proposed and successfully
realised for the Delft four-sector cyclotron (this method was used for the first time at the
Lebedev Institute [16] in Moscow). A 200-keV proton beam drifts along a sector edge, as
shown in Fig. 13, towards the cyclotron central region where the electric field of the dee starts
to increase the orbit radius.

Fig. 13 Precession injection in the Delft isochronous cyclotron

Beams of A have been accelerated up to the extraction radius with injected beam
intensities between 1 and 2 mA. A strong first-harmonic magnetic bump, produced by
extension of sectors 1 and 4 is used to centre the beam. However, this injection method is no
longer used.
 3.2.3 The neutral-beam injector

In order to reach the extraction energy in a low number of turns in a superconducting

cyclotron [17], high accelerating peak voltages should be used. This requires sufficiently large
gaps in the central region so it could be better to eliminate any internal ion source or inflector for
an axially-injected beam. Moreover the available space is small due to the high magnetic field.

These difficulties associated with injection may be circumvented by injecting a beam of

neutral hydrogen atoms of appropriate kinetic energy, (typically 100 keV) along the median
plane of the cyclotron in a direction perpendicular to the magnetic field as shown in Fig. 14.
Near the centre of the magnet the beam passes through a charge converter comprising of a very
thin foil of carbon or other solid material, which has the effect of stripping off the electrons
from the hydrogen atoms, giving a beam of protons of suitable energy at the correct position,
ready to be accelerated by the cyclotron. By adjusting the position of the stripping foil and the
direction of arrival of the neutral beam, the first orbits can be easily centred in the cyclotron.

The neutral beam to be injected into the

cyclotron is prepared by accelerating a beam
of hydrogen ions and passing it through a
charge converter consisting of a cell
containing gas at low pressure. It is an
advantage to start with a pre-accelerated beam
of negative hydrogen ions rather than
protons, because the cross-section for
producing neutral atoms is 10 times larger in
this case. For example for kinetic energy 100
keV the cross-section for neutralising
negative hydrogen ions in hydrogen gas is
2 10 16 cm 2 , so for a cell of length 50 cm at
normal temperature the required pressure is
1.7 mT. Tests show that about 50% of the
H beam can be converted to neutral
hydrogen. The scattering in the gas cell will
be about 0.6 mrad, corresponding to 1.5 mm
at the stripper foil (1 standard deviation). The
Fig. 14 Central region of the superconducting other important advantage of this neutral
cyclotron with the neutral injected beam beam injection system is the possibility of
bunching the injected beam before this gas cell, and adding a fast beam switch-off by means of
electrostatic deflecting plates.
The most critical component is the stripper foil near the centre of the cyclotron which
converts the neutral beam to free protons. It must be thin enough to limit the loss of energy.
On the other hand it must be strong enough and resistant to thermal and radiation damage.
Carbon foils, 25 nm thick, give satisfactory performance when measured on a test facility.
Over a dozen 25 nm carbon stripper foils have been tested in the Cyclotron Laboratory in Nice,
and have typically given over 20 A x hours of useful proton beam before breaking. The
cyclotron could be equipped with a rapid foil changing mechanism.
The multiple scattering angle in the 25 nm carbon foil will be 24 mrad (1 standard
deviation). So with 1.5 mm beam width at the stripper the emittance of the injected proton
beam will be 36 mm. mrad (1 standard deviation).

ACKNOWLEDGEMENTS
The author is grateful to N. Fietier for reading the manuscript and is very much indebted
to C. Armando, P. Montalant and C. Bajard for their help in preparing the manuscript.
REFERENCES
[1] S. Adam et al., The commissioning of PSI Injector 2 for high intensity, high quality
beams, Proceedings of the 13th Int. Conf. on Cyclotrons, Vancouver, July 1992, p. 36
[2] P. Mandrillon , Proceedings of the 9th Int. Conf. on cyclotrons, p. 307 and internal
report IPNO-GTA/85-01, IPN Orsay, France.
[3] P. Mandrillon and J.P. Schapira, 10th Int. Conf. on Cyclotrons and their applications,
East-Lansing, April 1984, p. 332.
[4] National Accelerator Centre, South Africa, Annual Report, NAC/AR/80-01.
[5] A. Chabert et al., Chromatic correlations at injection and related ejection problems in
separated sector cyclotrons, IEEE Vol NS-26, N3, June 1979.
[6] C. Tobias, Phys. Rev. 85, 764 (1952).
[7] A. Cabrespine, Eindhoven Colloquium on AVF Cyclotrons, April 1965.
[8] F. Alessandria et al., Status report on the heavy ion facility at LNS, Proceedings of the
13th Int. Conf. on Cyclotrons, Vancouver July 1992, p. 90.
[9] C.B. Bigham et al., First operation of the Chalk River superconducting cyclotron NIM,
A254, p. 237 (1987).
[10] A. J. Cox et al., Nucl. Instr. and Methods, 18, 19 (1992).
[11] G. Ryckewaert, Axial injection systems for cyclotrons, 9th Int. Conf. on cyclotrons and
their applications, Caen, Sept. 1981, p. 241.
[12] J.L. Belmont, Axial injection and central region of the AVF cyclotron, Summer school on
accelerator technology, 1986 RCNP-Osaka.
[13] R.W. Mller, Novel inflectors for cyclic accelerators, NIM 54, 29 (1967).
[14] J.P. Schapira, Proprietes optiques du miroir electrostatique, Rapport interne Orsay,
IPNO-GEPL/83-03.
[15] R. Baartman, Matching of ions sources to cyclotron inflectors, EPAC, Rome, June 6-10,
1988, p. 947-948.
[16] V.A. Gladyshev et al., Sov. Atom. Ener. (trans.) 18, 3, 268 (1965).
[17] P. Mandrillon et al., A compact facility for high energy proton therapy based on a
superconducting cyclotron, EPAC 94, London, 27 June to 1 July 1994.
EXTRACTION FROM CYCLOTRONS

J.I.M. Botman and H.L. Hagedoorn

Eindhoven University of Technology, Cyclotron Laboratory, Eindhoven, The Netherlands

Abstract
Several methods of extraction of ions in cyclotrons, such as stripping
extraction and resonance extraction, are considered. An extraction
efficiency of nearly 100% can be achieved by providing a very high
energy gain per turn or alternatively, for H- cyclotrons, by stripping.
Only simple models are used to explain certain extraction features.
Orbit centre coordinates are introduced. Examples of extraction devices
for several cyclotrons are given.

1. INTRODUCTION

There are various types of cyclotrons for different applications. Extraction mechanisms
and extraction devices are naturally adapted to the specific use of these cyclotrons. One may
differentiate between stripping extraction for H- cyclotrons and the extraction of positive ions
employing an electric septum and a magnetic channel. In general, positive-ion cyclotrons for
analytical purpose such as element analysis or material characterization, or for nuclear physics,
do not require a large beam current but, instead, need a good beam quality and energy
resolution. Here mostly resonance extraction methods are employed.

For the production of short-lived radionuclides high beam currents are neccessary, not
only because of favourable cost per unit of product, but also because of the short lifetime.
Compact cyclotrons that accelerate positive ions for isotope production generally have a limit in
extracted beam current of about 100 A. (Separated-sector cyclotrons can generate a much
higher external current.) The internal beam of such a cyclotron yields a spot of a few mm2 on a
target. This means that with an energy of 30 MeV a beam power of 3 kW may hit the
electrodes. Given the geometry of the extraction electrodes this approaches the maximum
allowable output, certainly if one wants to have a high reliability of machine operation. The
availability of negative ion sources with high output has given the opportunity to accelerate
beams with high intensities e.g. 500 A, and extract them via stripping. Here, in general,
requirements on beam quality are less important than for cyclotrons for analytical applications.

We consider several methods of extraction, using only simple theory, and we give
examples of extraction devices. The general discussion is restricted to isochronous cyclotrons,
unless synchrocyclotrons are mentioned explicitely. We will start, however, with an older idea
of extraction for early synchrocyclotrons.

It may be noted that because of the expanding orbits, extraction from cyclotrons in general
has somewhat different aspects than for example fast and slow extraction techniques for
synchrotrons, although resonances are often exploited in both cases. Slow extraction for
synchrocyclotrons does have similar features to that for synchrotrons, however.

General theory and information on extraction from cyclotrons has been given by
Heikkinen, Joho, Hagedoorn and others [13].

2. ON DEFLECTION AT n = 1 IN THE SYNCHROCYCLOTRON

In this section we briefly consider a paper from 1951 by Hamilton and Lipkin [4], with
the title given above, in which a method was proposed for extraction from early
synchrocyclotrons.
 The synchrocyclotrons used until then, with a purely rotationally symmetric magnetic field
falling off slowly radially, had a low dee voltage, with a subsequent small turn separation. By
providing a coherent precessional motion of the orbits (i.e. the orbit centres move slowly about
the magnet axis) via a field error or via mispositioning of the ion source, a larger orbit
separation near the deflector was obtained and an extraction efficiency (i.e. the ratio of extracted
to internal current) of typically 10% was achieved. The precessional motion leads to a large
energy spread of the extracted beam, via high frequency phase mixing (see section 4.4).

For a cylindrically symmetric field, radial and axial particle motion are governed by the
well known relations of Kerst and Serber, yielding stable motion for 0 < n < 1, with n the field
index
n= 0
r dB
B(r0 ) dr r0

where B(ro) is the axial field at radius ro [5]. Hamilton and Lipkin made a plot of B(r)r versus r
for the synchrocyclotron. The top of the curve represents the highest energy attainable, and also
gives the radius for which n = 1. They proposed to accelerate beyond the point n = 1, to utilize
the inherent radial instability beyond this, so that orbits are automatically radially separated.

Now consider radial oscillations. It is easy to show that for radial oscillation of particles
about an equilibrium orbit at radius ro and with maximum and minimum radius rb and ra
rb

( Br )dr = B(r0 )r0

1
rb r a
ra

i.e. the particles have equal areas at either side of ro in the Br versus r plot (Fig. 1a). Particles
for which equal areas do not exist, such as in Fig. 1b, are extracted at escape point re. This can
also be seen in Fig. 2, in which the equivalent potential well for radial oscillations is drawn.
The 'energy levels' correspond to amplitudes of radial oscillations.

Fig. 1 (Br) versus r for the synchrocyclotron, with ranges of radial oscillations (see text)
Hamilton and Lipkin argued that the best situation is the one in which no coherent radial
oscillations exist throughout acceleration from source to final radius. This would lead to equal
escape points for all particles, and equal final energy. As an example, for their cyclotron the
value of Br at the n = 1 point is 8.5% above the value at n = 0.2 for which r = 2 z , with
r and z the radial and axial oscillation frequencies. (This coupling resonance could not be
passed using the precessional extraction method because of energy exchange from radial to
vertical motion and consequent beam loss.) Accordingly their final energy was higher. In
present day thinking, for extraction one would provide a coherent oscillation amplitude near the
n = 0.2 point, lifting the oscillation energy level from 1a to 1b in Fig. 2.

Fig. 2 Equivalent potential well for radial motion, and energy levels corresponding to situation
in Figs. 1a and 1b.

3. STRIPPING EXTRACTION

During the last years the vast majority of commercially available cyclotrons serve to
deliver proton beams in the energy range of 3 to 30 MeV [6]. They are mostly used for the
production of short-lived radio-isotopes for medical diagnosis and for positron emission
tomography. New areas of applications, such as neutron radiography, are being explored.
Cyclotrons for pure nuclear physics are hardly built anymore, and if so their design is guided by
the receiving laboratory. For these commercially available cyclotrons, it is advantageous to
accelerate H- beams, where convenient use can be made of stripping extraction, and where
intensity problems, typical for the cyclotrons for analytical purposes or for nuclear physics,
have been resolved. H- cyclotrons are capable of accelerating particles with beam currents in
excess of several 100 A. Since 1975 the TRIUMF cyclotron, 520 MeV, 200 A, is the well
known big machine accelerating H - beams, and employing extraction by stripping [7]. It is a
major example in demonstrating the feasibility of H- acceleration for isotope production
cyclotrons.

In the extraction process the negative hydrogen ion beam simply passes a thin carbon foil
(e.g. pyrolytic graphite, typically 50 to 200 g/cm2), which strips off the electrons. Figure 3
shows extraction orbits for the IBA Cyclone 30 cyclotron [8].
 Fig, 3 Extraction orbits in the IBA Cyclone 30

3 . 1 Some important features of stripping extraction are:

a wide energy range may be covered by changing the foil position in the magnetic
field (the precise foil position is defined by demanding a common cross over point
outside the cyclotron magnet for extracted beams of different energies);
at the cross over point, a combination magnet guides the beam into a single
external beam transport line;
a large internal beam currents can be handled, nowadays typically 300 to 500 A,
with 100% extraction efficiency; activation of cyclotron components is minimal,
allowing machine access shortly after machine operation;
the foil lifetime typically exceeds 2.104 Ah;
the cyclotron fringe field is traversed with reversed rotation direction and with a large
angle, hence there is no radial defocusing (which is a problem for extraction from
positive-ion cyclotrons);
simultaneous extraction of two beams to two external channels is possible with two
foils corresponding to different energies and with different beam currents, e.g. for
the production of two different isotopes at the same time.

3 . 2 Energy spread

The energy spread of the extracted beam of H- cyclotrons may be estimated as follows.
Consider a radial phase space area at a radius r in the cyclotron, belonging to a kinetic energy T.
This area just touches the stripper foil, see Fig. 4. We assume a good centreed beam in an ideal
isochronous field. Particles lying in the part of the phase space area that overlaps with the foil
are stripped and extracted from the cyclotron. Other particles have to make more revolutions to
reach the stripper, and gain extra energy. As the energy is proportional to the radius squared
[5], the energy spread of stripped particles is given by T / T = 2r / r with r the radial width
of the phase space area. As an example, for r = 5 mm at a radius of 500 mm the relative
energy spread is 2%. Hence for achieving a small energy spread of the external beam it is of
importance to have a good internal beam quality.
 Fig. 4 Radial phase space area reaching the stripper foil

4. BEAM ORBITS

4 . 1 Orbit separation

Now we consider the extraction from cyclotrons with the help of extraction devices such
as an electrostatic deflector, to direct the accelerated particles away from the cyclotron magnetic
field. A large separation between successive turns is desired for reaching the highest extraction
efficiency. In this section we give a simplified picture of orbit separation in a cyclotron, and we
consider some methods to increase the separation [3,9].

Taking cylindrical coordinates, the radial position of a particle at an azimuth in the

cyclotron, is given by:
r ( ) = r0 ( ) + x ( ) sin(vr + 0 ) , (1)

where ro( ) is the radial position of the equilibrium orbit at that azimuth, x( ) is the radial
oscillation amplitude, and 0 is an arbitrary phase angle. For incoherent oscillations the
amplitude x( ) is given by
r0 ( ) x ,

with r0 ( ) the radial beta function for radius r0 and at azimuth and x the radial emittance.
However, we now look at the radial position of the centre of the beam for a coherent oscillation,
which has been induced by creating a field perturbation. Moreover we consider this at a fixed
azimuth i , i.e. the azimuth of the extractor entrance position. We restrict ourselves to the
case of vr near 1; as vr for isochronous cyclotrons, with the relativistic factor, this is the
case for most cyclotrons. (The analysis can be adapted easily for the case of higher values of vr
e.g. near 1.5 or 2.)

Rewriting Eq. (1), the radial position at i = 2n as a function of turnnumber n, is

given by:
r ( i ) = r0 ( i )_ x sin(2 n(vr 1) + 0 ) , (2)
where for convenience vr-1 has been taken since vr is close to 1. The separation between two
successive turns is:

r ( i ) = r0 ( i ) + x sin(2 n(vr 1) + 0 ),
(3)
+2 (vr 1) x cos(2 n(vr 1) + 0 ) .

In this equation the three terms on the right-hand side represent the orbit separation due to
different effects. We discuss them briefly, below.

The first term in Eq. (3) gives the orbit separation due to acceleration. The kinetic energy
T of accelerated particles in the cyclotron is given by

T r 2B2 ,

with r and B the average radius and field. Hence, neglecting the increase in the magnetic
field, the turn separation due to acceleration is
r 1 T
,
r 2 T
where T is the energy gain per turn. As an example for a final energy of T = 30 MeV,
T = 100 keV, and an extraction radius of 0.5 m, we find r = 0.83 mm. This is a rather
small number, e.g. when compared with a radial beam width of for instance 4 mm.

Especially in separated sector cyclotrons with a high energy gain per turn, the turn
separation near extraction due to acceleration may be sufficient. The PSI Injector II cyclotron
[10] is a good example of this. At the extraction energy of 72 MeV the orbit separation is
1.9 cm, with an energy gain per turn of 1 MeV. In fact the large turn separation is also crucial
as the machine is designed to handle currents in excess of 1 mA (the use of a flat-top cavity is
also important here).

Apart from the turn separation by acceleration which is automatically present, there are
two other mechanisms of forced separation which Eq. (3) reveals. The second term on the
right-hand side in this equation gives the orbit separation by an increase in the oscillation
amplitude. This can be accomplished by creating first or second harmonic magnetic field
perturbations in a region where vr is close to 1:

Bp ( x, ) = C1 (r ) cos( 1 (r )) + C2 (r ) cos(2 2 (r ))

where e.g. C1 and 11 are the amplitude and phase of the first harmonic. This method can be
regarded as providing dipole errors with closed-orbit distortions growing for tune values near
integers, or providing gradient errors being most sensitive and leading to an increase in the beta
function for half integer tune values [11]. This last method is the regenerative extraction method
[12-14], and has been proposed and applied for synchrocyclotrons. Magnetic components
called peeler and regenerator with positive and negative field gradients are introduced on certain
azimuths in the cyclotron, and the tune enters the vr = 1 stopband. Moreover, in
synchrocyclotrons one can program the RF such that the beam is stretched and that slow
extraction can be applied [15]. So-called second harmonic extraction has been applied in a
14.5 MeV compact isochronous cyclotron [16], where the number of turns in the fringe field,
where vr < 1, was too small to allow making efficient use of the forced orbit separation
enhancement mechanism discussed below.
 4.4.1. Precessional extraction [3,17,18]

The coherent oscillation as given by Eq. (2) describes a precessional orbit motion with
oscillation amplitude x. The turn separation due to precession is given by the third term on the
right-hand side of Eq. (3); the maximum turn separation is given by 2(vr-1)x. The effect is
fully exploited by accelerating the beam far into the fringe field of the cyclotron, where (vr-1)
has a substantial value. The side effect is that a higher final energy is reached in a given
cyclotron field. As an example: when a coherent oscillation amplitude x of 3 mm has been built
up (e.g. by a first harmonic perturbation near vr = 1) and acceleration takes place until vr = 0.8,
the maximum turn separation due to precession is 3.8 mm. This can be added to the value of
0.83 mm of the example of separation by energy gain.

Accelerating the beam far into the fringe field often means passing the vr = 2 v z coupling
resonance. Energy can be exchanged from the radial to the vertical motion, blowing up the
beam vertically and leading to beam loss. If the radial oscillation amplitude is not too large, and
if the resonance is passed in only a few revolutions, vertical amplitude increase is avoided [3].
In practice, a coherent radial oscillation amplitude of the same size as the incoherent amplitude,
is a good criterion for efficient extraction. Another reason for requiring not too large radial
oscillation is avoiding strong non linear effects.

4 . 2 The orbit centre

Radial oscillations of particles w.r.t. the equilibrium orbit can conveniently be described
by the movement of their orbit centres, see Fig. 5. (In a Hamiltonian formulation of betatron
oscillations in cyclotrons the orbit centre coordinates are the canonical coordinate and
momentum.)

Denoting the angle deviation by p x = px / p0 , where px is the transverse and p0 is the

overall momentum, the phase space area is equivalent to the area of orbit centres (x,y). The
radial motion of a particle is represented by a slowly moving orbit centre motion. If vr = 1 the
orbit centre remains fixed in place; if vr - 1 > 0 the orbit centre moves with frequency -(vr - 1),
i.e. rotates opposite to the particle motion. The actual position of the particle is found by adding
the radius of curvature of the equilibrium orbit at its azimuth to the orbit centre position.

Fig. 5 Relation between orbit centre coordinates and position and radial momentum deviation of
the particle from the central orbit

4 . 3 The effect of a 1st harmonic

We give an expression for the radial oscillation amplitude induced by a first harmonic
field perturbation. Let a positive magnetic field bump B extend over an azimuthal interval
along the negative x-axis, let the particle motion be anticlockwise, and let vr > 1 (see Fig 6).
 Fig. 6. Orbit centre shift y due to a magnetic field bump along the negative x-axis

The 1st harmonic is

1
A1 =

Bd =

B .
0

The field bump gives a change in radial momentum:

B
p x =
B
and an orbit centre shift:
B
y1 = R .
B
In one turn a hypothetical particle with oscillation amplitude A, see Fig. 7, has its orbit
centre displaced by:

y2 = 2 (vr 1) A .

When y1 = y2 one finds a new equilibrium orbit, because the orbit is closed. The orbit
centre of this equilibrium orbit (middle of the closed curve) is given by x = A. The beam
starts oscillating around the new orbit centre. One finds:
1 B
x = R
2 (vr 1) B
,
R 1
= 1 (vr 1)
2
in which 1 is the relative first harmonic 1 = A1/B.

Often
1 R
x =
vr2 1
is given. Since ( r - 1) is small, this is about the same.
As an example: taking 1 = 10-4, R = 1 m and r - 1 = 0.01, one finds an orbit centre
shift, i.e. a radial oscillation amplitude, of x = 5 mm.

Fig. 7 Shift of the equilibrium orbit due to a magnetic field bump

4 . 4 HF phase mixing

When a coherent oscillation amplitude is given to the beam in the centre of the cyclotron,
e.g. by an ion source position error, orbit centres at the final radius for particles with different
HF phases (HF) lie on a circular band. After n turns the orbit centre of a particle shifts by an
angle:

= 2 (vr 1) dn .
0

Particles with different HF phases have to make a different number of turns to reach the
final radius. Hence the corresponding orbit centres have a different azimuth. This is the
phenomenon of HF phase mixing. It implies a deteriorated beam quality and an increased
energy spread of the extracted beam (see section 3).

In most cases in the precessional extraction process, where a coherent oscillation is

induced on the passing of the vr = 1 resonance, the spreading of orbit centres for different HF
phases due to HF mixing, is small for an originally well centreed beam, as in general the
number of turns from the vr = 1 resonance till extraction is not so large. So in general the
coherence of the radial oscillations persists for particles with a different HF phase. This is one
of the favourable features of the precessional extraction method. In contrast to this, an example
showing the orbit centre off-set for different HF phases has been given by Laxdall [19],
(Fig. 8) in a study of H- extraction from the TRIUMF cyclotron (with many turns from the
vr = 3/2 resonance till extraction).

Fig. 8 Radial scans for the TRIUMF cyclotron using a differential probe showing the beam
density pattern for an optimized phase band and for a phase band off-set by 24. Phase
dependent amplitude growth at the r =3/2 resonance and a cos dependence in the rate of
advance of the precession centre account for the difference.

5. PRECESSIONAL EXTRACTION, NUMERICAL SIMULATION

In this section a numerical example regarding the precessional extraction method is

presented, and some remarks about it are given [9]. The calculations refer to the AVF cyclotron
of the Eindhoven University [20].

The equations of motion are integrated up to the extraction radius, starting before the
vr = 1 radius. The measured magnetic field data are used, and a first harmonic field
perturbation
B(r, ) = C1 (r ) cos( 1 (r ))

is included to provide a coherent oscillation at vr = 1. The tune diagram for the machine is given
in Fig. 9, in which the average radius is indicated. The beam is extracted at and the calculations
start at r = 48 cm.
 Fig.9 Tune diagram for the Eindhoven cyclotron, with the average radius indicated
It is seen that several resonances are passed during the extraction process, of which
vr = 1 and vr = 2vz are the most important ones. Figure 10 gives the motion of a grid of
particles in radial phase space.

Fig. 10 Precessional extraction. Evolution of particles with HF = 30 in radial phase space for
the Eindhoven cyclotron. The extractor aperture in phase space has been calculated (solid
aperture); an optimum location is indicated (aperture in broken lines).

The following remarks and observations can be made:

The calculation starts about 10 turns before vr = 1, 'turn 0'.

The entrance of the extractor (as indicated by a box) lies near a hill-valley boundary.
Therefore the orbits are directed inwards by about 50 mrad. This explains the negative
vertical scale.
The turn separation is about 0.5 mm near vr = 1. It increases rapidly.
A cycloidal motion can be seen due to both acceleration and the presence of a radial
oscillation.
A coherent oscillation has been created with an amplitude of ~ 3 mm.
Between n = 26 and n = 32 there is one radial oscillation. Hence vr 1 1 / 6 or vr
0.83.
There is a large beam separation between turn 34 and 35 of about 6.5 mm.
The septum width (0.4 mm) and the deflector aperture are indicated. The turn
separation is sufficient compared to the septum width.
There is a non-linear distortion of the phase plane area.
The energy spread of the extracted beam is about 2 eVdee, which is the energy
difference between turn 34 and 35, with Vdee the dee voltage (the cyclotron has a
single-dee RF system).
There is an energy position correlation along the cycloid, as this curve describes an
oscillation about equilibrium orbits with increasing energy (or radius).
Particles with another HF-phase HF lie on the same cycloid, with more or less turns,
to reach the extractor, as the HF-phase mixing is not effective, see section 4.4. With
the remark above this implies the quoted energy spread of 2 eVdee for the entire beam.

Figure 11 shows a differential probe measurement for this cyclotron in the extraction
region. The separation between the probe fingers is 2 mm. The figure reveals the radial
oscillation near extraction.

Fig. 11 Differential probe scan near extraction for the Eindhoven cyclotron

6. MULTI- AND SINGLE-TURN EXTRACTION

In general, particles with a different value of the HF-phase complete a different number of
revolutions before reaching the extraction septum and having the final energy. In a purely
isochronous machine the HF phase of a particle remains constant, which explains this. This
leads to multi-turn extraction. For the Eindhoven cyclotron, as an example, the number of turns
may vary by about 30, with a total number of about 300. Single-turn extraction on the other
hand can be obtained when the HF phase width is restricted, e.g. with the help of slits in the
centre of the cyclotron. This leads to a reduced energy spread of the extracted beam. Multi-
and single-turn extraction has been described by several authors [21, 22].

The kinetic energy T and the radius r of particles on equilibrium orbits after N turns is
given by:
T = T0 (1 1 / 2 2 ) ,
r = r0 (1 1 / 4 2 ) ,

where T 0 and r0 belong to a central HF phase 0, and where is the phase difference with
respect to 0. A plot of the energy for orbits near the extractor as a function of HF phase is
given in Fig. 12. This figure also shows the energy levels belonging to the extractor opening,
i.e. of the septum and the electrode. The top of the paraboles is shifted due to the effect of the
fringe field. It is seen that several turns get extracted, with an energy pattern for the external
beam as given in Fig. 13. Single-turn extraction is obtained for a phase width restricted to =
2 - 1 (see Fig. 13), or for || < arccos(N/N + 1).

For example with N = 180, || < 6, single-turn extraction for a wider phase range and
with a small resulting energy spread can be obtained by applying a higher harmonic cavity in
addition to the fundamental RF frequency. This is the flat-topping principle [5]. Important
examples of this are found in the PSI injector II and ring cyclotrons [10,23]. The additional
advantage is of course the possibility to extract the beam with 100% efficiency. In Eindhoven a
special minicyclotron (ILEC) has been constructed with an RF phase acceptance five times
larger due to the flat-topping technique, and an energy spread less than 0.1% [24]. This
machine serves especially for PIXE and PIXE microbeam applications.

Fig. 12 Kinetic energy versus HF phase near extraction

Fig. 13 Energy profile for the extracted beam

For the main Eindhoven cyclotron, which is a multiturn machine, single-turn experiments
have been made by restricting the RF phase width with radially selecting slits in the cyclotron
centre [25]. The accelerating voltage was set to have about 180 turns, hence the phase width
had to be within 12. Single-turn extraction was found by observing 100% variation in the
external beam current as a function of the accelerating voltage, see Fig. 14. This implies that the

Fig. 14 Single-turn extraction demonstrated by 100% external beam variations as a function of
accelerating voltage
number of turns in the cyclotron can be selected. In this situation the FWHM relative energy
spread of the external beam is 0.85 10-3, compared to 0.310-2 for multiturn extraction. The
single-turn effect is extremely sensitive to variations in the magnetic field, e.g. a variation of
B/B = 2.10-4 of the main magnetic field destroys the single-turn extraction. However, in this
situation the effect of a first-harmonic field perturbation in the centre of the cyclotron created by
the inner harmonic coils, could be compensated rather completely by a first harmonic of the
outer harmonic coils. Increasing the beam current through the slits destroys the single-turn
extraction: space charge effects at low energy increase the RF phase width.

7. DEFLECTION DEVICES

The extraction hardware for a cyclotron usually consists of an electrostatic deflector

followed by a magnetic channel. The curvature of the electrodes of the deflector must
correspond to the shape of the orbits of extracted ions. An angle kick of typically 50 to 100
mrad is provided. The inner electrode, septum, is at earth potential. At the entrance it has a
thickness of a few tenths of a mm, increasing to several mm at the exit. A V-slit at the entrance
is often used to distribute the deposited power over a larger area. The septum is water cooled.

Smith and Grunder [26] have given a criterion for the product of electric field E and
potential V for a cyclotron deflector in order to avoid electric discharges: VE < 1.5 104
(kV)2/cm. This is shown in Fig. 15 together with deflector voltage values for several
cyclotrons. The maximum sustainable voltage in a magnetic field is 20-30% lower than without
magnetic field [27].

Fig. 15 Maximum electric field and voltage as a function of the deflector gap for cyclotron
deflectors. The operating deflector voltage for several cyclotrons has been indicated.

In order to reduce the cyclotron magnetic field along the extraction paths and to provide a
horizontally focusing action for compensating the defocusing action of the cyclotron fringe
field, passive [28] or active [29] magnetic channels are employed. In the design of a magnetic
channel it is important that it's magnetic field is strongly decreased in the area of the last
accelerated orbits.

Figure 16 shows a passive magnetic channel which is housed within the accelerating
electrode of the ILEC cyclotron [24]. Figure 17 shows the extraction elements in the AGOR
cyclotron [30], consisting of an electrostatic deflector ESD, an active normal conducting
channel EMC1, and a superconducting channel EMC2. Moreover two superconducting
focusing and steering elements are located on the passage of the magnet yoke. Figure 18 gives
the conductor arrangement for the EMC1.
Fig. 16 Calculated magnetic field and its gradient as produced by a passive magnetic focussing
channel in the ILEC cyclotron. The figure also shows a vertical cross-section through the
channel.

Fig. 17 Median plane view of the AGOR cyclotron with its extraction elements

Fig. 18 Cross-section of the conductors of EMCl. The cyclotron centre is at the left, the circle
marks the axis of the extracted beam. The distance between conductors 4 is about 1.5 cm
4: deflecting coil B 0.2 T; 2: gradient coil dB/dx 13 T/m; 3: compensation of stray field
at long range; 1: compensation of stray field at short range
8. CONCLUSION

Several methods of extraction from (isochronous) cyclotrons have been discussed. Most
important are stripping extraction, resonance free and precessional extraction. Some hardware
aspects have been given.

ACKNOWLEDGEMENTS

We thank W. Kleeven (IBA), M. Kruip (Oxford Instruments), H. Schreuder (KVI) and

Bob Laxdall (TRIUMF) for useful information and comments.

REFERENCES

[ 1] P. Heikkingen, Injection and extraction for cyclotrons, CAS, CERN 94-01,Vol. II,
(1994), 819.

[ 2] W. Joho, Extraction from medium and high energy cyclotrons, Proc. 5 Int. Conf. on
Cycl., Butterworths, London, (1971), 159.

[ 3] H.L. Hagedoorn and P. Kramer, Extraction studies in an AVF cyclotron, IEEE Trans.
Nucl. Sci. NS-13, No. 4, (1966), 64.

[ 4] D.R. Hamilton and H.J. Lipkin, On deflection at n = 1 in the synchrocyclotron, Rev.

Sci. Instr., Vol. 22, No. 10 (1951), 783.

[ 5] T. Stammbach, Introduction to cyclotrons, these proceedings.

[ 6] V. Bechtold, Commercially available compact cyclotrons for isotope production, Proc. 13

Int. Conf. on Cycl., Vancouver 1992, World Scientific, (1993), 110.

[ 7] G. Dutto, The TRIUMF 520 MeV cyclotron, Ibid, 138.

[ 8] Y. Jongen et. al., Construction of the Louvain La Neuve 30 MeV 500 A H-cyclotron,
Proc. 11 Int. Conf. on Cycl., Ionics, Tokyo, (1987), 275.

[ 9] F. Schutte et. al., Numerical calculations in the extraction process of the Eindhoven AVF
cyclotron, Nucl. Instr. Meth. 127, (1975), 317.

[10] U. Schryber et. al., Status report on the new injector at SIN, 9 Int. Conf. on Cycl., Les
Edition de Physique (France), (1981), 43.

[11] E. Wilson, Transverse beam dynamics, CAS, CERN 94-01, Vol. I., (1994), 131.

[12] J.L. Tuck and L.C. Teng, Regenerative deflector for synchrocyclotrons, Phys. Rev. 81,
(1951), 305.

[13] K.J. Le Couteur, The regenerative deflector for synchrocyclotrons, Proc. Phys. Soc.
London, B64, (1981), 1073.

[14] N.F. Verster, Regenerative extraction from the 150 MeV synchrocyclotron at the
Laboratoire Curie, IICC, (1959), 224.

[15] D.W. Storm, A survey of synchrocyclotron projects, IEEE Trans. Nucl. Sci., Vol. NS-
26, No. 2, (1979), 1970.
[16] J.M. van Nieuwland, Extraction of particles from a compact isochronous cyclotron,
Thesis TU Eindhoven, (1972).

[17] H.L. Hagedoorn and N.F. Verster, The extraction of the beam of the Philips AVF
cyclotron, CERN 63-19, (1963), 228.

[18] M.M. Gordon and H.H. Blosser, Orbit calculations on the extraction system for the
MSU cyclotron, Ibid, 236.

[19] R.E. Laxdall et.al., Beam test results using the auxilary accelerating cavity, Proc. 13 Int.
Conf. on Cycl., World Scientific, (1992), 442.

[20] N.F. Verster et. al., The Philips prototype AVF cyclotron, CERN 63-19, (1963), 43.

[21] M.M. Gordon, Single turn extraction, IEEE Trans. Nucl. Sci. NS-13, No. 4, (1966),
48.

[22] H.L. Hagedoorn et. al., Some factors determining the beam quality of AVF cyclotrons,
Proc. 5 Int. Conf. on Cycl., Butterworths, London, (1971), 274.

[23] Th. Stammbach, Experience with the high current operation of the PSI cyclotron facility,
Proc. 13 Int. Conf. on Cycl., World Scientific, (1992), 28.

[24] W.J.G.M. Kleeven, Theory of accelerated orbits and space charge effects in an AVF
cyclotron, Thesis Eindhoven University, (1988).

[25] J.I.M. Botman, Central region study for a moderate energy cyclotron, Thesis Eindhoven
`University, (1981).

[26] B.H. Smith and H.A. Grunder, Electrical design of electrostatic deflectors for sector-
focused cyclotrons, CERN 63-19, (1963), 304.

[27] F. Broggi et. al., The electrostatic deflectors for the Milan superconducting cyclotron, 12
Int. Conf. on Cycl., Word Scientific, (1991), 350.

[28] G. Bellomo, Design of passive magnetic channels, 13 Int. Conf. on Cycl., World
Scientific, (1992), 592.

[29] K. Pieterman et. al., The AGOR superconducting extraction channel EMC2, ibid., 585.

[30] H.W. Schreuder, The AGOR cyclotron past the half-way mark, 13 Int. Conf. on Cycl.,
World Scientific, (1992), 21.
MEDICAL APPLICATIONS OF ELECTRON LINEAR
ACCELERATORS
A.Wambersie* and R.A. Gahbauer**

* Service de Radiothrapie - Oncologie, Cliniques Universitaires St-Luc, Universit

Catholique de Louvain, Brussels, Belgium
** Division of Radiation Oncology, Ohio State University Hospitals, Columbus, Ohio, USA

Abstract
The role of radiation therapy in cancer management is first defined:
today among the cancer patients who are cured ( 45 %) about one half
are cured by radiation therapy applied alone or in combination with
surgery or chemotherapy. The experience accumulated in decades
shows that, to be efficient, the radiation treatment must be delivered
with a high physical selectivity. At present, electron linear accelerators
are the primary equipment of a modern radiotherapy department, and a
large proportion of the patients are treated with a linear accelerator for at
least part of the treatment. Photon beams of about 6-20 MV have, in
general, a sufficient penetration in the tissues to treat most of the
tumours with an adequate physical selectivity. A single beam is never
used alone, but a combination of several beams adequately oriented
allow the radiation-oncologist to deliver the prescribed dose to the target
volume without exceeding the tolerance of the surrounding normal
tissues. Modern linear accelerators also allow us to apply electron
beam therapy. Electron beam therapy is suitable for the treatment of
superficial lesions (e.g., skin tumours) and is also the best available
irradiation treatment for Intra-Operative-Radiation-Therapy (IORT).
Finally, conformation therapy, which is developed in many well-
equipped and well-staffed centres, will probably further improve, in the
near future, the efficiency of linear accelerators at least for some tumour
locations.

1. INTRODUCTION ROLE OF RADIATION THERAPY IN CANCER

MANAGEMENT

Electron linear accelerators today constitute the core of the equipment of a modern
radiation therapy department. Nowadays, the majority of the patients referred to a radiation
therapy department are treated with a linear accelerator for at least part of their treatment. It is
likely that this will remain true for the foreseeable future. Linear accelerators thus play, and will
keep playing, a significant role in cancer management in general and are responsible for the
therapeutic success obtained in many tumour treatment. To illustrate the situation, Table 1 gives
the number of linear accelerators at present used in France for radiotherapy applications (their
type and energy). The numbers of the other radiotherapy units are also given for comparison
[1].

Cancer has an increasing impact on the death rates in the Western world, as well as in the
developing countries. For example, in 1980, 730 000 deaths were attributed to cancer in the
countries of the European Union and there were 1 186 000 new cancer cases diagnosed in that
year alone [2, 3]. These estimates exclude non-melanoma skin cancer, which, although a rare
cause of death, nevertheless demands medical care. Even with the current prevention programs,
the numbers will further increase within the next two decades. Similar figures have been
published for North America.
Table 1
 Number of radiotherapy units in France, on 1 January 1994 1)

A. LINEAR ACCELERATORS
GECGR Mev Other companies Total
4-6 MV NEPTUNE 6 18 Philips (SL 75/5) 8
ORION 5 Clinac 600 C 4 31
10 MV NEPTUNE 10 27
SATURNE 1M 3 30
15 MV SATURNE 15 5
SATURNE 1 9 Philips (SL 18) 4 34
SATURNE 41 16
20 MV SATURNE 20 9 Philips (SL 75/20) 3
SATURNE 11 9 Siemens Mevatron 19
SATURNE 42 4 Clinac 2100 C 4 48
25-40 MV SAGITTAIRE 32 MV 10
SAGITTAIRE 40 MV 2 17
SATURNE 25 MV 5
20-25 MV SATURNE 111 4 4
25 MV SATURNE 43 35 Philips (SL 75/25) 9 44
Total Linear Accelerators 157 51 208
(+ 19)
B. OTHER TYPES OF RADIATION THERAPY UNITS
RX 50-100 kV 116 (-6)
RX 150-305 kV 34 (-8)
Cobal Units (including 1 gamma knife) 144 (-19)
Betatrons 1 (-2)
Cyclotron neutron therapy 1
Cyclotron neutron + protontherapy 1
Synchrotron protontherapy 1
1) The differences since 01.01.1993 are given (+) or (-)

Today, at the first consultation, approximately 65% of the patients have apparently
localized tumours. About 2/3 of these are cured either by surgery, radiotherapy, or a
combination of both treatments. In this group of patients with probable but unproved subclinical
metastatic disease, chemotherapy used as an adjutant treatment may prolong life and maybe cure
some additional patients.

Among the other 35% of patients arriving at the first consultation with already inoperable
or metastatic disease, only about 5% will be cured by combined treatment including
chemotherapy and immunotherapy as well as radiotherapy and/or surgery (Table 2). Although
there is promising progress in the field of medical oncology, this cure rate is largely limited to
solid paediatric tumours, leukaemias, lymphomas, and testicular tumours. These tumours
represent only about 5% of all cancers seen in a general population [4-6]. Although these
percentages [5] are useful as an indication of the contribution of the different techniques to
cancer cure, they will become progressively less relevant to the extent that combination of these
different techniques is more and more successfully applied.

It is axiomatic that one must control the local disease if one aims ultimately at curing any
cancer patient. In fact, it has been shown that 1/3 of the patients who die of cancer had
uncontrolled local disease. If local failure could be reduced by 50%, one could expect a
10-15% improvement in cure rate [7].

Surgical techniques have already reached a very high level. Further improvement will be
seen in a reduction of mutilating procedures (limb-sparing operation, breast-conserving therapy,
reduction of colostomies, and urinary diversions). On the other hand, wider excisions are still
foreseeable as a result of safer anaesthesiology, intensive care support, and improvement in
reconstructive surgery.

Table 2
Summary of the present situation concerning cancer cure rate

Patients Cure rate %

With localised tumour: 65
cured by surgery 22
cured by radiotherapy 12
cured by combination of surgery and radiotherapy 6
With inoperable or metastatic disease: 35
cured by combined treatment including chemo- and immunotherapy 5
Total 100 45

Furthermore, the combination, on a wider scale, of surgery with irradiation will help to
increase the local control rate. In that respect, Fletcher (M.D. Anderson Hospital, Houston) has
shown that, after radical surgery, doses of 50 Gy given over 5 weeks are able to eradicate
expected occult infestation ("subclinical disease") in the lymphatic nodes for cancer of the breast,
upper respiratory and digestive track and some pelvic tumours, in more than 90% of the cases.
Doses of 50 Gy do not exceed the tolerance limit of many normal tissues.

Today, about 70% of the cancer patients are referred to a radiation therapy department,
either for a radical treatment (aiming at a local control) or after surgery (also aiming at a local
control) or for a palliative treatment often in combination with chemotherapy (e.g., a painful
bone metastasis) [8].

2. RATIONALE FOR RADIATION THERAPY

Soon after the discovery of x-rays by Rntgen in 1895 and of radium by Marie Curie in
1898, it became evident that ionizing radiations could sterilise malignant tumours and thus cure
cancer patients. It also became rapidly evident that, above a certain dose level, x-rays induce
damage to the normal tissues they traverse.

The effects produced by radiation are not instantaneous. In particular, tumour shrinkage
after irradiation is a process extending over weeks or even months. This is due to the fact that
cells, doomed to die, do not disappear immediately: they remain present for hours or days;
sometimes they are able to undergo a few divisions before disintegration and elimination. This
progressive tumour shrinkage reduces the risk of acute complications, such as haemorrhage after
electro-coagulation. It is a major advantage of irradiation for cancer management [9].

Since local control probability increases with dose, radiation must reach the whole
volume(s) of tissues invaded by malignant cells at a sufficiently high dose level to be effective.
This goal must be reached without inducing severe and irreversible sequelae in the surrounding
normal tissues. A first approach to reach this goal is the improvement of the physical selectivity.
2 . 1 Physical selectivity of a therapeutic irradiation

The physical selectivity of irradiation is defined as the ratio of the dose to the "tumour"
relative to the dose to the surrounding "normal tissues". It can be improved by varying the
nature and energy of the radiation or the beam arrangement (number and orientation of the
beams, and their size and shape as in conformation therapy).
 The best historical illustration of the importance of the physical selectivity in radiation
therapy is the impressive improvement in the clinical results achieved in the 1950's60's when
200 kV x-rays (orthovoltage) were replaced by cobalt-60 gamma rays and high-energy x-rays,
as shown in Table 3. The high-energy x-rays mentioned in this table are x-rays ( 23 MV)
produced by Van de Graaff machines, betatrons or electron linear accelerators. Linear
accelerators have played an important role in the improvement of the physical selectivity in
radiation therapy in the last decades, as will be discussed in detail in the following sections.

Table 3
Improved survival of patients with several types of cancer treated with megavoltage radiotherapy

Survival (%) after

Type of cancer 200 kV x-rays High-energy x-rays
Hodgkin's disease 3035 7075
Cancer of the cervix 3545 5565
Cancer of the ovary 1520 5060

Cancer of the bladder 05 2535

Cancer of the prostate 515 5560
Seminoma of the testis 6570 9095
Embryonal cancer of the testis 2025 5570

Cancer of the nasopharynx 2025 4550

Cancer of the tonsil 2530 4050
Retinoblastoma 3040 8085
From the Conquest of Cancer, Report of the National Panel of Consultants of the Committee on
Labor and Public Welfare United States Senate, November 1970, p. 51 [9].

Today, with modern techniques, most critical normal tissues such as brain, eyes, spinal
cord, kidneys, liver, etc. may be completely avoided or at least irradiated at levels well below
tolerance (except, of course, in special cases depending on the tumour location: e.g. the normal
brain is the normal tissue at risk when a brain tumour is treated, similarly the spinal cord is the
normal tissue at risk when some cervical or mediastinal tumours are treated, etc.).

Physical selectivity is a second specific advantage of radiation therapy compared to

chemotherapy in which all the tissues in the body are exposed to the toxic drug.

The real situation is however more complex. Outside the limits of the detectable tumour
(i.e. the "gross tumour volume" as defined below), there is in general some "subclinical"
involvement. A larger tissue volume must then be irradiated: not only the gross tumour volume
itself but also a surrounding safety margin and, in some cases, the regional lymph node areas.
Several types of volumes thus have to be defined and identified.
2 . 2 Concepts and definitions of volumes used in radiation therapy

Gross Tumour Volume (GTV)

Radiation oncologists first identify the gross tumour volume (GTV), which is the gross
palpable or visible extent and location of the cancer disease (ICRU Report 50, [10]). The shape,
size and location of the GTV is determined by means of different diagnostic methods such as
clinical examination (e.g., inspection, palpation, endoscopy) and various imaging techniques
(e.g., x-ray, CT, ultrasound, MRI, etc.). The gross tumour volume consists mainly of
malignant cells, connective tissue and blood vessels, and necrotic areas.

Clinical Target Volume (CTV)

 Clinical experience has shown that around the GTV there is in general "subclinical
involvement", i.e., individual malignant cells, small cell clusters, or microextensions which
cannot be detected by the clinical or imaging procedures. Thus a safety margin surrounding the
GTV must be irradiated to ensure local control of the disease. The size of the safety margin
depends on the type of tumour, its location, and its tendency to invade the surrounding normal
tissues (histology, grading, etc.). The GTV together with this safety margin consisting of
tissues with presumed or proven subclinical involvement is defined as a Clinical Target Volume
(CTV).

In addition to the safety margin around the tumour, other volumes with presumed or
proven subclinical spread (such as regional lymph nodes) may also require irradiation. The
Clinical Target Volume is thus the tissue volume that contains a demonstrable GTV and/or
presumed/proven subclinical microscopic malignant disease. This volume must be treated at an
adequate dose level (and time-dose pattern) to achieve the aim of therapy, cure, or palliation.

Planning Target Volume (PTV)

To ensure that all tissues included in the CTV receive the prescribed dose, one has, in
principle, to plan to irradiate a volume geometrically larger than the CTV. It is the Planning
Target Volume or PTV.

The additional safety margin, included in the PTV, results from a number of factors:

- movements of the tissues which contain the CTV (e.g., with respiration), as well as
movements of the patient.

- variations in size and shape of the tissues that contain the CTV (e.g., different
fillings of the bladder, rectum, stomach).

- all variations and uncertainties in beam geometry and patient-beam geometry. There
are some uncertainties in the beam sizes, shapes and directions, as well as in the
relative position of the beam with respect to the patient, the CTV and the normal
tissues.

- all uncertainties in dose distribution, especially in or close to the penumbra region,

or where inhomogeneities have to be taken into account (e.g., beam penetration for
electron beams).

Delineation of the PTV is a matter of compromise and depends upon the judgement of the
radiation-oncologist, who ultimately bears the responsibility for this decision. In particular, too-
large safety margins will result in unnecessary side effects and complications.

2 . 3 IMPROVEMENT OF THE DIFFERENTIAL EFFECT. A PROBLEM FOR

THE RADIOBIOLOGIST

As defined above, the safety margins contained in the CTV and in the PTV consist
primarily of normal tissues and, only for a small proportion, of invading cancer cells. Because
of the presence of these cells, the safety margin should be irradiated in principle to the highest
possible dose to prevent a local recurrence, but the tolerance of the normal tissues included in the
safety margin limits the dose which can be prescribed.

The difficulty cannot be solved by improving the physical selectivity of the irradiation; it
requires an improvement of the differential effect, and this brings us from the field of physics to
the field of radiobiology. Improvement of the differential effect implies, for a given (physical)
dose, increasing the effect of the cancer cell population and/or reducing the effects on the normal
tissues [11, 12].

The possibility of improving the differential effect by adequately selecting the irradiation
parameters is the third advantage of radiation therapy for cancer management.

Historically the first, and presently the most efficient, method for improving the
differential effect was the fractionation of the dose, as initiated in 1919 by Regaud, Ferroux and
Coutard at the Fondation Curie in Paris. Today, five fractions of 2 Gy each per week are used
as the conventional fractionation for most treatments.

As a further refinement, the fractionation could be adapted to the growth characteristics of

the individual tumour (such as, the tumour doubling time, or better the potential doubling time,
TPOT). In that respect, an EORTC study has shown that a shorter overall time brings a benefit
for some head and neck tumours with short TPOT. This implies the administration of several
small ( 1.5 Gy) fractions per day, since large fraction sizes are known to be harmful.

Combination of radiation with radiosensitizers is another method to improve the

differential effect. Drugs such as misonidazole are used to sensitize selectively hypoxic cancer
cells. Other drugs, such as actinomycin D, bleomycin, adriamycin, 5-fluorouracil, and, more
recently, cis-platinum have been used as radiosensitizers in the treatment of oesophagus, lung,
or intestinal tumours. Besides any synergistic effects, radiation and drugs have several
complementary actions which further justify their therapeutic association.

A third possibility for improving the differential effect is to replace x-rays (and other low-
LET radiations) by high-LET radiations (ICRU Report 45 [10]). High-LET radiations bring a
benefit for well-differentiated, slowly growing tumours and/or for tumours containing a large
proportion of hypoxic cancer cells [13-15].

2 . 4 ACCURACY REQUIRED IN RADIATION THERAPY

Whatever the improvement in the radiobiological differential effect, the doses which are
needed to control a malignant tumour are often of the same order of magnitude as the tolerance
doses for the normal tissues. In addition, radiobiological and clinical evidence indicates that the
dose-effect relations for tumour control are steep (Table 4) [1618]. For some tumours, a dose
variation of a few percent can modify significantly the observed local control rate. The dose-
effect relations are even steeper for normal tissue complications (Table 5) [17]. For these two
reasons, accurate dosimetry is needed, and, in 1976, the ICRU made the following
recommendations: "the available evidence for certain types of tumours points to the need for an
accuracy of 5% in the delivery of an absorbed dose to a target volume if the eradication of the
primary tumour is sought" (ICRU Report 24 [10]). More recently (1987), Mijnheer et al. [17]
recommended an accuracy in absorbed dose delivery of 3.5% (one standard deviation, for the
dose at the specification point, for radical treatment).

Over the years, the ICRU has played an important role in the world-wide improvement of
dosimetry in radiation therapy. An important step in that direction was the definition of
quantities and of a system of units (ICRU Report 33 [10]). Besides quantities and units, the
ICRU has focused a great part of its efforts on the selection of procedures suitable for the
measurement and application of these quantities. This implies a critical review of the existing
procedures, value judgement, and finally the recommendation of selected procedures. These
should be well described and codified, and widely accepted in the different countries.

Several ICRU Reports have dealt with dosimetry protocols for photon therapy: Reports
14, 17, 23, 24 and 42 [10]. For electron beam therapy, Report 35 remains a world-wide
reference. Other ICRU Reports have contributed to a better accuracy and uniformity of
dosimetry procedures in radiation therapy: Reports 44, 46 and 48.
 Table 4
Relative steepness of the dose-effect curve for local tumour control. The steepness is expressed
as the relative increase in absorbed dose (in %) producing a change in tumour control probability
from 50 to 75%.

Site of tumour Steepness (%)

Supraglottic larynx T2 and T3 (Shukovsky) 5
Larynx T3 (Stewart and Jackson) 6
Supraglottic larynx all stages (Hjelm-Hansen et al.) 11

Larynx all stages (Hjelm-Hansen et al.) 12

Bladder T4B (Battermann et al.) 13
Epidermoid carcinoma head and neck (Cohen) 13

Supraglottic larynx T1 and T2 (Ghossein et al.) 13

Skin and lip (Strandqvist) 17
Supraglottic larynx T2 and T 3, revised analysis of the Shukovsky date
(Thames et al.) 17

Nasopharynx T1 and T2 (Tokars and Griem) 18

Nasopharynx (Moench and Philips) 19
Lymphoma (Fuks and Kaplan) 21

Retromolar trigone/anterior faucial pillar T1 and T2 (Thames et al.) 21

Bladder all stage (Morisson) 26
Base of tongue T1 and T2 (Thames et al.) 31

Tonsillar fossa T3 and T4 (Thames et al.) 32

Hodgkin (Kaplan) 46
Base of tongue T3 and T4 (Thames et al.) 50

Table 5
Relative steepness of the dose-effect curve for normal tissue reaction The steepness is expressed
as the relative increase in absorbed dose (in %) from 25% to 50%

Normal tissue reaction Steepness (%)

Major chronic complications of the larynx (Harwood and Tierie) 2
Peripheral neuropathy (Stoll and Andrew) 3
Late skin damage (Battermann et al.) 4

Late intenstinal damage (Battermann et al.) 4

Brachial plexus (Svensson et al.) 5
Radiation pneumonitis (van Dijk et al.) 6

Skin reaction (Turesson and Notter) 7

Major complications of the intestine and bladder (Morrison) 9
Skin and lip (Strandqvist) 10

Myelitis (Phillips and Buschke) 15

Major and non major complications of the larynx (Ghossein et al.) 17

3. SHORT DESCRIPTION OF A MEDICAL LINEAR ACCELERATOR

In medical linear accelerators, the charged particle is an electron and the RF accelerating
electric field oscillates at about 3000 MHz. For comparison, radio waves in standard amplitude
modulation broadcast channels oscillate at about 1 MHz [19].
 The electrons are thermo-ionically emitted from a concave metal cathode at 1000 C and
accelerated in the gun to about 1/4 the velocity of light by a pulsed dc electric field. They are
then formed (coalesced) into a pencil beam by a convergent electric field between the gun
electrodes. The RF electric field in the accelerating tube then forms the electron stream into
bunches and accelerates them to more than 99% of the velocity of light, increasing their mass
(e.g., by a factor of 13 at 6 MeV). The electrons then bombard the target (e.g., tungsten).
When an electron enters a tungsten target, the positively-charged nuclei of the tungsten atoms
pull on the negatively-charged electron, shaking it violently as it passes by one such tungsten
atom after another. These high-energy electrons give up a portion of the energy at each
successive atom that they penetrate and emit hard, penetrating x-rays in the forward direction.
Two sets of tungsten blocks allow the sizes of the radiation beam (square or rectangular
sections) to be adjusted.

All modern medical electron linacs employ an isocentric gantry. The accelerator wave
guide is mounted in the gantry, either (relatively) parallel to the gantry axis if a beam bending
magnet is employed, or perpendicular to the gantry axis if a beam bending magnet is not
required (low energy, e.g. 4 MV x-ray beam). A simplified block diagram of a medical linear
accelerator is presented in Fig. 1.

More and more modern linear accelerators are equipped with a multileaf collimator which
typically consists of two sets of 20 leaves which can be moved independently allowing the
application of irregularly-shaped beams better adapted to the complex shape of the target volume
(Fig. 2) [20].

Fig. 1 Simplified block diagram of a medical linear accelerator

4 THERAPY WITH HIGH-ENERGY PHOTON BEAMS

The most important characteristics of a therapeutic photon beam are the penetration in the
irradiated tissues (depth/dose curve), the skin sparing effect and the penumbra. For clinical
applications, a single beam is seldom applied; and, for each patient, the optimal combination
involving several (or many) photon beams has to be selected.
Fig. 2 Principle of a multileaf collimator. It consists of two sets of 20 leaves which can be
moved and adjusted independently. As a result, beams of any shape can be applied to closely
match the shape of the target volume. Target volumes with complex shapes can be treated with
minimum irradiation of the surrounding normal tissues [20].

4 . 1 Beam penetration depth/dose curves

The penetration of the beam increases with energy (Fig. 3). The depth/dose curves
describe the variation of absorbed dose (relative to the maximum dose) as a function of depth, in
the irradiated medium, on the beam axis.

Fig. 3 Depth/dose curves for therapeutic photon beams of different energy. Low-energy x-ray
beams (100 kV) are still in use and are suitable for treating skin lesions. Orthovoltage x-rays
(200-500 kV) are totally inadequate to treat deep seated tumours; they are nearly completely
abandoned. Cobalt-60 is satisfactory for many clinical situations but x-ray energies between 8
and 24 MV are needed to obtain adequate depth dose/curves. In addition to an improvement of
the beam penetration with energy, there is a sparing effect which also increases with energy.
Epidermis only is spared with cobalt-60 but the subcutaneous tissues are also protected with 8-
24 MV photons.

X-rays of about 100 kV ("low-energy" x-rays) are still in use, but only for the treatment of
superficial (skin) lesions. Conventional x-rays ("orthovoltage", 200-250 kV) were the only
treatment available up to 19551960, before cobalt-60, betatrons and linear accelerators were
available.
 Cobalt-60 provides irradiation conditions which are satisfactory for a large proportion of
clinical situations. Cobalt-60 emits gamma rays at 1.17 and 1.33 MeV respectively. Due to the
simple construction of the machine, cobalt units are the ideal equipment for developing
countries.

Linear accelerators of about 8 MV and above allow most of the tumours to be treated with
good physical selectivity. There is no clear advantage in increasing the x-ray beam energy
beyond 20-24 MV.

The importance of beam energy in irradiating deep seated tumours is better illustrated on
Fig. 4, where the doses are normalized at the depth of the centre of the target volume.

Fig. 4 The depth/dose curves for different photon beam energies are normalized at 12 cm in
depth (e.g. difficult situation for an obese patient). This presentation better illustrates the
excessive dose delivered to the normal tissues located in front of the PTV when 200 kV x-rays
or (to a lesser extent) cobalt-60 are used.

4 . 2 Skin sparing

When only 200 kV x-rays were available, the acute skin reactions often prevented the
radiation-oncologist from delivering the appropriate dose to the deep seated tumours.

For photon energies higher than 23 MV, the maximum absorbed dose is not observed at
the surface of the irradiated medium, but a few millimetres below the surface. This skin sparing
effect is due to a lack of "electronic equilibrium" (ICRU 24 [10]).

For cobalt-60, the acute skin reactions are dramatically reduced. Furthermore, for photon
energies of about 8 MV and above, there is a significant reduction of the absorbed doses at the
level of the subcutaneous tissues which reduces the risk of late fibrosis. A reduction of the
absorbed dose at the level of the subcutaneous tissues also facilitates healing of the surgical
wound if surgery is needed after irradiation.
4 . 3 Penumbra

Ideally, the absorbed dose within the geometrical limits of the photon beam should be
homogeneous and the absorbed dose outside the beam as low as possible. Unfortunately, at the
border of the beam, there is a narrow region with a steep dose gradient: it is the "penumbra
region".

The penumbra width is defined, at a given depth, as the distance between two selected
isodose levels, i.e., between the 80% and 20% isodoses or between the 90% and 10% isodoses
(100% being the maximum absorbed dose on the beam axis at that depth). In all cases, the two
isodose levels used for defining the penumbra should be specified (Fig. 5).

Fig. 5 Dose distribution obtained with a single 8 MV photon beam (beam size 10 cm x 10 cm).
Transverse section through the patient body; the planning target volume (PTV) is indicated by
the hatched area. The dose is specified at the centre of the PTV (100%), it varies within the
PTV from 114% to 86%. From ICRU 50 [10].

4 . 4 Beam combination

An acceptable dose distribution cannot be achieved with only a single photon beam. The
normal tissues in front of the target volume receive too high a dose, and the normal tissues
behind the target volume also receive a significant dose (see again Figs 3 and 4).

In the "cross-fire" technique, several beams that intersect at the level of the target volume
are used; as a result, the surrounding normal tissues at risk are irradiated only by one beam.
The number and the orientation of the beams are selected by the radiation-oncologist; it is a
matter of judgement and experience to select the optimal beam arrangement for each particular
patient. Some examples of treatment schemes are presented in Figs 69.

5. ELECTRON-BEAM THERAPY

Besides photon-beam therapy, modern linear accelerators allow electron-beam therapy to

be applied. For that purpose, in the accelerator head, the target is removed and the beam of
accelerated electrons is directly oriented towards the patient and the PTV.
Fig. 6. Transverse section through the Fig. 7. Transverse section through the
pelvis, with the display of the dose thorax, with the display of the dose
distribution resulting from four beams: two distribution resulting from a moving beam
sets of two parallel opposed beams treatment with an 8-MV photon beam (large
orthogonal to each other ("box technique"). arc 300). The PTV is indicated by the
The PTV is indicated by the hatched area. hatched area. The dose is specified at the
The dose is specified (100 %) at the centre centre of the target volume (100 %) and on the
of the PTV and on the beam axes. The dose central axis of the moving beam. The dose
within the PTV ranges from 103 % to 95 %. variation within the PTV ranges from 102 %
This beam arrangement can be used for the to 80 %. This irradiation technique can be
treatment of a locally-extended cervix used e.g. for the treatment of a mediastinal
carcinoma. From ICRU 50 [10]. tumour (oesophagus as presented in the
figure) or a prostatic tumour. From ICRU 50
[10].

Fig. 8 Three equally weighted 8-MV photon Fig. 9 Irradiation of a lateral lesion (e.g.
beams are converging towards the centre of floor of the mouth). Two orthogonal 8-MV
the PTV. Wedge filters (45 ) are used for beams with 45 wedge filters are used.
beams # 2 and 3 to achieve a homogeneous From ICRU 50 [10].
dose distribution throughout the PTV. From
ICRU 50 [10].

5 . 1 Electron beams
 The penetration of the electron beams in the tissues is much shallower than that of the x-
ray beams and, in addition, can be adjusted by varying the energy of the incident electrons
(Fig. 10). Therefore, electron-beam therapy is used to treat superficial or semi-deep-seated
tumours extending (close) to the skin surface. Beyond the depth of the maximum, the dose falls
off rapidly (but to a lesser extent when the initial energy increases). Figure 11 presents, as an
example, the dose distribution for a 20 MeV electron beam. Treatment energies range from
about 4 to 2025 MeV, but some accelerators reach higher energies (e.g., 35 MeV for a
Sagittaire and 50 MeV for the race-track microtron).

Fig. 10 Central-axis depth/dose curves for Fig. 11 Isodose curves for a 20 MeV electron
electron beams of different energy, from 6 beam in a water phantom (10 x 10 cm;
to 22 MeV (field size: 10 cm x 10 cm; source/skin distance: 100 cm). Scattering of
source/skin distance: 100 cm). The steep the electrons in the medium is responsible for
fall-off of the dose, especially at the low the widening of the isodose curves with
energies, protects the normal tissues located increasing depth.
beyond the PTV.

Electron beams are used for 1025 % of the patients referred to the radiation therapy department,
this proportion varying from country to country and from centre to centre depending on the local
treatment policy. Electrons are often used in combination with photon beams (e.g., as a boost
against the residual tumour). Two specific applications of electron beams deserve to be
mentioned.

5.2.1 Total skin electron therapy

A skin cancer, mycosis fungoides, is most often treated in this way. The aim of the basic
treatment is to irradiate the total skin envelope as homogeneously as possible. The depth of the
lesions suitable for this type of treatment varies with the stage and type of disease and/or the
body surface. This may lead to the use of different beam penetrations. When tumourous
lesions are present, there may be a need for a special boost and/or shielding.

The maximum depth of the PTV varies from approximately 5 to 15 mm in most of the
lesions. For the most frequent indications, with localised and even generalized plaques, the
PTV is located within the first 5 mm. Infiltrated plaques, ulcerations, and tumourous lesions
justify an individual estimate of the thickness of the lesions whenever possible.

Different technical approaches have been developed using multiple fields, arc therapy,
patient rotation, patient translation, according to the available local equipment and facility
(Figs 12, 13) [21]. In the past, low-energy x-rays were the only possibility, but they tend to be
less frequently used since the implementation of low-energy electron beam techniques using
linear accelerators. As a general rule, the integral dose should be kept as low as possible.
Usually, the incident electron beam energy at the level of the patient skin ranges from 3 to 7
MeV with nominal energy ranging from about 4 to 10 MeV (i.e., accelerator energy at the exit
window). Such energy losses are due to the material in the treatment head, the air gap between
the exit window and the patient, and the use of scattering screens. The use of scatter energy
degrader (typically a 1 g. cm-2 lucite panel) close to the patient improves dose uniformity,
particularly on oblique body surfaces, and results in a higher surface dose and shallower depth
dose due to the increase of the angular spread of the original electron beam.

Fig. 12 Total skin electron therapy six field Fig. 13 Total skin electron therapy. Patient
arc therapy. Geometry of the treatment set-up stances for the anterio, posterior, and two
of electron arc therapy, showing the arc of the angled dual-field exposures.
direction. Distribution of the six fields for full
circumferential coverage is also indicated on a
typical transverse section.

5.2.1 Intra-operative radiation therapy (IORT)

With this technique, electron beams are used to deliver a large single-dose fraction during
surgical procedure to a well defined PTV. The dimensions of the CTV are defined accurately by
the surgeon and the radiation oncologist together. During the procedure, mobile sensitive
tissues are displaced out of the beam using localizers, in order to decrease normal tissue toxicity.

The purpose of the procedure is usually to treat presumed subclinical disease after
macroscopically radical resection. The exact shape, size and location of the CTV can thus only
be defined during surgery.
 Different types of localizers have been described, according to the local equipment and
facilities. The dose distribution depends strongly on the design of such localizers, applicators,
and other accessories, as well as on patient/beam positioning accuracy. Most irradiation
techniques involve lucite applicators linked physically to the head of the treatment machine
(docking system), but systems using laser beams and an optical set up (non-docking system) are
also in use (Figs 14, 15).

In general, the nominal incident electron beam energy ranges from 6 to 20 MeV (beam
energy at the exit window), giving energies between 3 and 15 MeV at the patient surface.
Flattening filters are used to provide a homogeneity of 10 % within the geometrical beam; they
also act as scattering foils leading to additional decrease in the beam energy at the patient level.
Nominal energy thus becomes irrelevant, and beam quality can only be described using
depth/dose distribution on the geometrical axis of the beam (assuming an homogeneous phantom
parallel to the applicator end). In some cases, combination of adjacent beams has to be used to
treat a PTV larger than the available localizers.

Fig. 15 Electron beams for IORT. The

dose distribution varies with the localizer, in
particular with the angle between the
localizer (or beam) axis and the distal plane
of the localizer (perpendicular or oblique,
left and right figure respectively). With an
oblique beam, a higher electron energy is
Fig. 14 Shematic drawing of the Lyon IORT needed to treat a given tissue thickness.
system

6. THE FUTURE: NEW DEVELOPMENTS AND TRENDS IN RADIATION

THERAPY

As mentioned above, the efficiency of radiation therapy may be improved either by

increasing the differential effect, the physical selectivity, or both.

Several attempts have been made to improve the "radiobiological differential effect":

optimizing the fractionation scheme as a function of the characteristics of the

individual tumours;

combining radiation with new radiosensitizing drugs, but also selecting their optimal
combination (e.g., time schedule for administration of both radiation and drugs).
 However, it is probably through the development of new irradiation techniques that the
most significant progress may be expected. An increase in total dose to the target volume,
without exceeding the tolerance dose of the surrounding normal tissues, has always been the key
to a better clinical result [7, 16, 18].
A promising approach is the replacement of conventional x-rays by new types of radiation,
at least for some tumour sites. Among them, fast neutrons and protons are the most commonly
used, and, to a lesser extent, heavy ions.

Fast neutrons, which are high-LET radiation, aim at improving the differential effect
(ICRU Report 46 [10]). Radiobiology has shown that they are specifically efficient against
hypoxic (cancer) cells or against cells which accumulate in the resistant phases of the mitotic
cycle.
Recently, clinical observation has shown that neutrons bring a benefit for the treatment of
some well-selected types of tumours [13, 14, 22]. The most convincing results were obtained
for locally-extended inoperable salivary-gland tumours for which the conclusions of a NCI
study are summarized as follows: "fast neutron radiation therapy alone is the treatment of choice
for salivary-gland tumours that cannot be completely resected and/or where the surgical risk of
facial nerve damage is high" [15, 23]. Also, for locally extended prostatic adenocarcinoma, two
randomized trials, initiated by the RTOG in the US, have shown the superiority of fast neutrons
(used alone or as "mixed schedule") compared to conventional x-rays [23, 24].

Proton-beam therapy improves the physical selectivity of the irradiation, but no benefit is
to be expected from an improved differential effect [7]. However, the impressive clinical results
obtained for selected tumour sites (in particular, uveal melanoma) by the pioneer centres have
encouraged several other centres to initiate proton therapy programs. The number of proton-
therapy facilities is increasing: 4 in 1980, 6 in 1985, 10 in 1990, and 16 in 1994. Other
facilities are planned.

Technological development has allowed the engineers to offer cyclotrons that are compact,
reliable, and affordable. These compact cyclotrons, superconducting or based on conventional
technology, can be installed within the hospital. They will soon certainly facilitate the
development of neutron- and proton-beam therapy.

Heavy ions combine the advantages of neutrons and protons: they enable the radiation
oncologist to deliver a high-LET therapeutic irradiation with an excellent physical selectivity
comparable to that of proton beams. Heavy ions were applied at the Lawrence Berkeley
Laboratory, where 299 patients were treated with neon ions. Unfortunately, the facility was
closed recently. However, an ambitious heavy-ion therapy program was started at the NIRS in
China, Japan, in 1994. A third program will start at the GSI in Darmstadt, Germany, in 1996.

Unfortunately, for the present and the near future, the number of patients who could
benefit from these new types of beams will be limited. Therefore, it is important to improve and
to optimize the available techniques. In that respect, stereotactic radiosurgery, conformation
radiotherapy, intra-operative radiotherapy, total skin irradiation, and modern developments in
brachytherapy have already brought encouraging results for some tumour types. These
techniques are being applied in an increasing number of large, well-equipped and well-staffed
radiotherapy departments. Their safe application requires accurate dosimetry and careful quality
control procedures. Some of these approaches (e.g., stereotactic radiosurgery and conformation
radiotherapy) involve optimizing the use of linear accelerators and improving their technical
characteristics.

Stereotactic radiosurgery is used mainly for the treatment of lesions limited in size and
located in the head (e.g., tumours of the base of the skull). By using a large number of narrow
beams intersecting at the level of the lesion or by using arc therapy, high doses can de delivered
to the lesion while keeping the dose to the surrounding normal structures relatively low. This
procedure requires accurate localization of the lesion, accurate dose computation, and accurate
beam/patient positioning (Fig. 16) [25].
 Conformation therapy can be defined as an effort to optimize the treatment by adequately
selecting the number and orientation of the beams, their size and shape. In other words, the
volume of tissues actually irradiated should match as closely as possible the PTV, even if it is
complex in shape.

Since the early days of radiation therapy, some kind of conformation therapy was an
objective. Impressive examples were given already in the forties, for example by Baclesse and
his colleagues at the Fondation Curie in Paris, especially for some tumours of the head and neck
area. However, the poorly penetrating beam available at that time prevented the full success of
these efforts.

a) b)
Fig. 16 Stereotactic radiosurgery. a) Overview of patient treatment. A special collimator and an
accurate head support system is added to the linear accelerator. Treatment uses both gantry and
turntable rotation. b) Head schematic showing the three coordinate axes and the four arcs of a
typical cranial treatment [25]

Today, four sets of new means are available to the radiation-oncologist:

a) complete patient data: typically in the form of a series of CT slices with detailed
anatomical information (MRI or other imaging techniques). The location, extent and
position of the target volume relative to the organs at risk or anatomical reference points
can be known accurately.

b) with high-energy, penetrating x-ray beams, it is possible to reach any point in the body,
even in depth. Furthermore, the beams are well collimated; their size and shape can be
adapted to the target volume. With complex beam arrangements, different beam shapes
can be applied.

c) with the modern treatment planning systems (3-D), it is possible, even for a complex
treatment, to visualise rapidly the resulting dose distribution and thus to optimize the beam
arrangement based on reliable information.
d) lastly, with some sophisticated equipment, the treatment machine can be driven
automatically by the computer of the treatment planning system.
 An example of the possible development in the field of conformation therapy is given in
Fig. 17 where the target volume consists of the iliac and para-aortic lymph node areas. In
different transverse planes, the shape of the target volume and its position relative to the patient
contour and reference anatomical points vary significantly. This implies that the optimum beam
arrangement (number and orientation of the beams) may be different within the different
transverse planes. Furthermore, in a given plane, since the target volume is not spherical in
shape, the size and shape of the beams should be different according to their orientation [26].
Other approaches are possible, combining for example beams which are no longer coplanar.

Fig. 17 Example of conformation therapy. Dynamic treatment is used to irradiate a PTV

consisting of the retroperitoneal para-aortic lymph nodes as well of the iliaque lymph nodes. In
each transverse section, the number, size, shape and orientation of the beams is optimized.
Irregular-shaped beams can be used. The movements of the gantry, treatment head, and
collimator are computer controlled. There is, in addition, a slow (computer-controlled) patient
translation [26].

In conclusion, an optimal use of the available modern linear accelerators, in particular

conformation therapy, will improve the outcome of the radiotherapy treatments in well-selected
patient groups. The side effects and complications will certainly be reduced. Optimization of
the treatments with linear accelerators will be facilitated by further improvement in their technical
performance and reliability. In addition, several companies have designed specific types of
linear accelerators for specific purposes (e.g. stereotactic radiosurgery) [25, 27].

Finally, the techniques of stereotactic radiosurgery and conformation therapy described

above can be combined with the introduction of new beam qualities (e.g. protons, neutrons,
heavy ions). The benefit of both approaches could then be added. Application of neutron,
proton and heavy-ion beams using conformation therapy techniques has actually been achieved
in some centres for certain well-selected tumour types [24, 27].

* * *

REFERENCES

[1] A. Laugier, ACRIM 1994, Annuaire de la Cancrologie/Radiothrapie et des Imageries

Mdicales en France, Hpital Tenon, Paris 20, 1994.
[2] C.S. Muir and P. Boyle, The Burden of Cancer in Europe, Eur. J. Cancer, Vol. 26, no
11/12, 1111-1113, 1990.

[3] L. Davis, D. Hoel, J. Fox and A. Lopez, International trends in cancer mortality in
France, West Germany, Italy, Japan, England and Wales, and the USA, The Lancet, 336,
474-481, August 25, 1990.

[4] V.T. Devita, Progress in cancer management, Cancer, 51, 2401-2409, 1983.

[5] V.T. Devita and D. Korn, Progress against cancer, New Engl. J. Med., 315, 964, 1986.

[6] R. Doll, Are we winning the fight against cancer? An epidemiological assessment, EACR
- Mhlboch Memorial Lecture, Eur. J. Cancer, Vol. 26, No. 4, 500-508, 1990.

[7] H.D. Suit, Potential for improving survival rates for the cancer patient by increasing the
efficacy of treatment of the primary lesion, Cancer, 50, 1227-1234, 1982.

[8] Commission of the European Communities, Concerted action: cancer treatment with light
ions in Europe - EULIMA, Final Report - Part 1: General Feasibility Study; A.
Wambersie, P. Chauvel, G. Gademann, J.P. Gerard, R. Sealy, Socio-economic study,
pp. 2-39, CEC, rue de la Loi 200, 1049, Brussels, 1992.

[9] A. Wambersie and G.F. Whitmore, Radiation therapy, ICRU News, International
Commission on Radiation Units and Measurements (ICRU), Inc., 1, 15-20, 1995.

[10] INternational Commission On Radiation Units And Measurements (ICRU):

- Report 14: Radiation dosimetry: x-rays and gamma rays with maximum photon energies
between 0.6 and 50 MeV (1969).
- Report 17: Radiation dosimetry: x-rays generated at potentials of 5 to 150 kV (1970).
- Report 23: Measurement of absorbed dose in a phantom irradiated by a single beam of x
or gamma rays (1973).
- Report 24: Determination of absorbed dose in a patient irradiated by beams of x or gamma
rays in radiotherapy procedures (1976).
- Report 33: Radiation quantities and units (1980).
- Report 35: Radiation dosimetry: electron beams with energies between 1 and 50 MeV
(1984).
- Report 42: Use of computers in external beam radiotherapy procedures with high-energy
photons and electrons (1987).
- Report 44: Tissue substitutes in radiation dosimetry and measurement (1989).
- Report 45: Clinical neutron dosimetry - Part 1, Determination of absorbed dose in a
patient treated by external beams of fast neutrons (1989).
- Report 46: Photon, electron, proton, and neutron interaction data for body tissues (1992).
- Report 48: Phantoms and computational models in therapy, diagnosis and protection
(1993).
- Report 50: Prescribing, recording and reporting photon beam therapy (1993).
ICRU, Inc., 7910 Woodmont Avenue, Suite 800, Bethesda, Maryland 20814-3095,
USA.

[11] M. Tubiana, J. Dutreix and A. Wambersie, Radiobiologie, Hermann Ed., Paris, 1986.

[12] M. Tubiana, J. Dutreix and A. Wambersie, Introduction to radiobiology, Taylor &

Francis, London, 1990.

[13] A. Wambersie, Neutron therapy: from radiobiological expectation to clinical reality,

Radiation Protection Dosimetry, 44, 379-395, 1992.
[14] A. Wambersie, F. Richard and N. Breteau, Development of fast neutron therapy
worldwide, Acta Oncologica, 33, 261-274, 1994.

[15] G.E. Laramore, J.M. Krall, T.W. Griffin, W. Duncan, M.P. Richter, K.R. Saroja, M.H.
Maor and L.W. Davis, Neutron versus photon irradiation for unresectable salivary, gland
tumors: final report of an RTOG-MRC randomized clinical trial, Int. J. Radiation
Oncology Biol. Phys., 27, 235-240, 1993.

[16] R.C. Ultrasun, Does improved depth dose characteristics and treatment planning correlate
with a gain in therapeutic results? Evidence from past clinical experience using
conventional radiation sources, Int. J. Radiation Oncology Biol.Phys., 22, 235-239,
1992.

[17] B.J. Mijnheer, J.J. Battermann and A. Wambersie, What degree of accuracy is required
and can be achieved in photon and neutron therapy? Radiotherapy and Oncology, 8, 237-
253, 1987.

[18] H.D. Thames, T.E. Schultheiss, J.H. Henry, S.L. Tucker, B.M. Dubray and
W.A. Brock, Can modest escalations of dose be detected as increased tumor control? Int.
J. Radiation Oncology Biol.Phys., 22, 241-246, 1992.

[19] C.J. Karzmark, S. Nunan and E. Tanabe, Medical Electron Accelerators, McGraw-Hill,
Inc., Health Professions Division, New York, 1993.

[20] A. Brahme: Optimal setting of multileaf collimators in stationery beam radiation therapy.
Strahlentherapie und Oncologie, 164: 343-350, 1988.

[21] AAPM, American Association of Physicists in Medicine, C.J. Karzmark (ed): Total skin
electron therapy; technique and dosimetry, Report # 23, American Institute of Physics,
New York, 1987.

[22] W.J. Koh, J.M. Krall, L.J. Peters, M.H. Maor, G.E. Laramore, C.M. Burnison,
L.W. Davis, S. Zink and T.W. Griffin, Neutron vs. photon radiation therapy in
inoperable regional, non-small cell lung cancer: results of a multicenter randomized trial,
Int. J. Radiation Oncology Biol. Phys., 27, 499-505, 1993.

[23] National Cancer Institute (NCI), Fast neutron radiation therapy in the United States: a
twenty-year NCI sponsored research program (NTCWG annual report, 1991), National
Cancer Institute, Bethesda, MD 20892, USA, 1991.

[24] M. Austin-Sempour, R. Caplan, K. Russell, G. Laramore, J. Jacky, P. Wootton,

S. Hummel, K. Lindsley and T. Griffin, Impact of a multileaf collimator on treatment
morbidity in localized carcinoma of the prostate, Int. J. Radiation Oncology, Biol.Phys.,
30, 1065-1071, 1994.

[25] W. Lutz, K.R. Winston and N. Maleki: A system for stereotactic radiosurgery with a
linear accelerator Int. J. Radiation Oncology Biol.Phys., 14, 373-381, 1988.

[26] L.M. Chin, P.K. Kijewski, G.K Svensson, J.T. Chaffey, M.B. Levene and
B. Bjarngard: A computer-controlled radiation therapy machine for pelvic and paraaortic
nodal areas. Int. J. Radiation Oncology Biol.Phys., 7, 61-70, 1981.

[27] H.D. Kogelnik (Ed), Progress in Radio-Oncology V, International Proceedings Division,

Monduzzi Editore, 40128 Bologna, Italy, 1995.
APPLICATION OF HIGH INTENSITY LINACS
TO SPALLATION SOURCES
H. Lengeler
CERN, Geneva, and European Spallation Source (ESS), KFA, Jlich

Abstract
High intensity linear H- accelerators are at present studied for
applications in pulsed neutron spallation sources. The time structure
imposed on the linac beam offers the possibility of time-of-flight
measurements in neutron scattering and allows large peak intensities.
For the production of extremely short neutron pulses (sec) a
combination of a linac with a high intensity accumulator or accelerator
ring must be used. Low loss injection in the ring asks for acceleration
of H- ions instead of protons. The planned European Spallation Source
(ESS) is taken as an example for discussing the specific problems and
the design principles of a pulsed high intensity H- linac.

INTRODUCTION

High intensity proton and deuteron linacs have found during the last years a renewed
interest for pulsed neutron spallation sources [1], fusion material studies and nuclear waste
transmutation [2] or energy production. Peak intensities reached already 20 years ago in the
most powerful low energy proton linacs (~ 100 mA) would be almost adequate for these
applications but average intensities are low by a large factor. The most powerful high energy
proton linac in operation is LAMPF at Los Alamos with a peak intensity of 17 mA and an
average intensity of about 1 mA at 800 MeV. A considerable increase of peak and average
intensity will be necessary for the new applications considered.

1. NEUTRON SCATTERING AND SYNCHROTRON RADIATION

APPLICATIONS

Neutron scattering and synchrotron radiation have led in the past to major breakthroughs
in the study of the structure and dynamics of condensed matter. In Table 1 a few characteristic
properties of both types of radiation are given and their complementary for the study of
condensed matter appears clearly. Neutron scattering has found applications in a wide range of
condensed matter studies such as [3]

- Crystal structures with large periodic cells

(proteins, enzymes,...)
- Lattice dynamics
- Phase transitions
- Diffusion in solids
- Metal-H2 systems
- Interfaces
- High-Tc materials
- Magnetic materials
- Polymers
- Defect structures (stress and fatigue)
- Fullerines
- Liquids and quantum liquids
- Soft matter

Table 1
 Comparison of neutron scattering and synchrotron radiation

n-scattering Synchrotron radiation

Interaction principally with nuclei Interaction mainly with electrons

Energy and momentum of neutrons can be Same wavelengths like neutrons

matched simultaneously to length scales but much higher energies less suited
appropriate for atomic structures (1 to 10 ) for study of atomic excitations.
and to energies for excitations in condensed E ~ keV: study of outer electron shells,
matter (neV to eV) (static and dynamic behaviour) chemical states (electron spectroscopy)
Large momentum transfer (and scattering angles) E = 10 - 100 keV: atomic arrangements
are possible and allow the exploration of large (structure analysis e.g. in biological materials)
periodic cells with high resolution

Weak interaction, large penetration depth Small penetration depth, interaction mainly
in matter, bulk properties can be studied; concentrated near surfaces
complex sample environment is possible,
small radiation damage in biological substances

Large response differences to different Isotopes hardly distinguished

isotopes (in particular H and D) Very weak response of light elements,
Good response of light atoms in particular H

Magnetic moment of neutrons couples Polarisation possible

to magnetic moment of atoms;
short wavelength variations of magnetic
materials can be studied

Intensities and source brightness much Large spectrum of photons

smaller than for synchrotron radiation Extremely high source fluxes and brightness;
Because of weak interaction, small emittances allow very
n-scattering has always been an high lateral resolution
intensity limited field Ultrashort pulses (psec) with high repetition
rates (MHz)

2. NEUTRON SOURCES

For many years nuclear reactors have been the main source of neutrons, producing a
continuous flow of particles. This line of sources has culminated in the construction of
dedicated high flux reactors. As an example one may mention the ILL reactor at Grenoble with
a thermal power of 58 MW and a useful neutron flux of 1015n/cm2xs. However, as fission is a
highly exothermic effect releasing about 200 MeV of energy for every useful neutron (i. e. not
needed to sustain the chain reaction) very high power densities are produced in the reactor core
and limit ultimately the obtainable neutron flux. High flux reactors are near the end of their
development potential and one does not expect neutron fluxes an order of magnitude above the
ones reached at the ILL-reactor.
During the last years another type of neutron source has found growing interest: the so
called spallation source. Neutrons are produced by high energy protons (~ 1 GeV) hitting a
heavy metal target and exciting nuclei to energies where neutrons are "evaporated" (spallation).
Typical production rates for 1 GeV protons are of the order of 25 neutrons per incident proton
(in Pb) and correspond to an energy consumption of about 40 MeV per neutron.

Neutrons are produced in a large range of energies reaching up to the energy of the
incident protons and have to be slowed down in a hydrogen rich moderator to energies
adequate for the study of condensed matter (Table 2). They emerge from the moderator with a
"white" Maxwellian velocity distribution and a wide range of wavelengths.

Table 2
Typical parameters of neutrons for condensed matter studies

Cold Thermal Epithermal

Energy (meV) 1 25 1000
Temperature (K) 12 290 12000
Wavelength () 9 1,8 0,29
Velocity (m/sec) 440 2200 14000

3. PULSED SPALLATION SOURCES

One of the most important advantages of spallation sources is that one can impose on the
proton beam, and thereby on the neutron flux, almost any time structure.

Compressing the neutron pulse in a short time interval allows at given average power
an increased peak power. This is a considerable advantage for most types of n-scattering
instruments where performances have been often intensity limited in the past. If the proton
pulse is made sufficiently short, the neutron pulse can be used as a primary resolution element:
by using time-of-flight methods each neutron arrives at the detector with a time tag related to its
energy. In this way a band of neutron energy can be used simultaneously. It avoids the
neutron monochromatisation applied with CW sources which decreases considerably neutron
intensities at the detector and it allows a more efficient use of the primary neutron pulse. This is
illustrated conveniently in a time of flight diagram for the neutrons shown in Fig. 1. One
concludes from this figure that:

a small pulse length should be used for increasing the time resolution of measurements.
Arguments based on the slowing down and lifetime of neutrons in the moderator show that
it is desirable to keep pulse durations below a few sec;

small repetition rates of pulses (typically 50 Hz) should be used in order to avoid the
overlap of slow neutrons from one pulse with fast neutrons from the next pulse (frame-
overlap). This is particularly important for cold neutrons.

Detectors can work with a fixed geometry because angular scans are replaced by energy
(time) scans. Gating out the highly penetrating background of fast neutrons time-correlated to
the short proton pulse improves the signal to background ratio of detectors. It is today accepted
that most of neutron scattering instruments can profit from the time structure of neutron pulses
and that short pulses will trigger new developments and improvements of the existing
instruments.

Over the past ten years [1] a number of pulsed and CW spallation sources have become
operational (Table 3). We mention in particular the most powerful pulsed spallation source
ISIS at the Rutherford Laboratory (UK). It combines a 70 MeV linac with a 800 MeV rapid
cycling synchrotron and has reached a beam power of 160 kW with 450 nsec pulses. A similar
source (AUSTRON) is at present under planning in Austria [4].

The SINQ spallation source at the Paul Scherrer Institute (CH) is a CW source operated
with a 590 MeV isochronous cyclotron and is laid out for a maximum beam power at the target
of ~ 900 kW [5]. Powerful pulsed spallation sources are also under study at ANL (Argonne)
and at LAMPF (Los Alamos) where a beam power of about 1 MW is considered. Recently a
Fig. 1 Time-of-flight diagram of neutrons produced by a short proton pulse and their velocity
(time) distribution at the detector

two year research study / supported by the European Community / has been started for an even
more ambitious project: A European Spallation Source (ESS) with the following design
parameters [6]:
5 MW average beam power at the target(s), corresponding to the average flux of the high
flux reactor at ILL, Grenoble
~1 sec pulse length
10 and 50 Hz repetition rate at two target stations, respectively.
Table 3
A few existing and planned pulsed and CW spallation sources

Average Repetition Beam power Energy of one Number of

Facility Accelerators current rate at target pulse at target protons per
at target (Hz) (kW) (kJ) pulse at
(A) target

IPNS, Argonne 20 30 10 0.33

50 MeV Linac 4x1012

500 MeV Synchr.

KENS, KEK 10 20 5 0.25
40 MeV Linac 3.1x1012

500 MeV Synchr.

60 12 80 7
LANCSE, 800 MeV Linac 5x1013
Los Alamos
+ storage ring

ISIS, 200 50 160 3.2

70 MeV Linac 2.5x1013
Rutherford Lab.
800 MeV Synchr.

Austron II (planned) 125 25 200 8.2

130 MeV Linac 3.2x1013
[4]
1.6 GeV Synchr.
European Spallation
1.334 GeV Linac 5.2x1014
Source ESS 3.750 50 5.000 100 (more than
1.334 GeV one ring)
(planned)
Accumulator
SINQ 590 MeV 1.500 CW 900 - -
Cyclotron

Presently this goal cannot be realised by a linear proton accelerator alone. One therefore
considers the use of a pulsed linac combined with a accumulator or an accelerator ring filled by
multiturn injection and emptied by fast one-turn ejection for reaching the desired peak power
and pulse lengths (Fig. 2). The beam power of 5 MW at the target can be produced by an
adequate combination of proton energy and average current; an energy range of 0,8 to 3 GeV is
at present under discussion.

4. THE PROPOSAL FOR A EUROPEAN SPALLATION SOURCE (ESS)

Recently a very first parameter set for ESS has been proposed; it combines a H-linac of
1.33 GeV with two accumulator rings in parallel. A few parameters are given in Table 4 and a
schematic view of the linac is given in Fig. 3.
The parameter range is amongst others restricted by the following constraints:

A high average beam power and a low duty cycle for the neutron scattering experiments
Beam dynamics especially at low energy where the beam is heavily space charge dominated
The use of high intensity accumulator rings for compressing pulses to ~ 1 sec
 Fig. 2 Schematic view of a pulsed neutron source

a) 1) H--ion source, 2) RFQ, 3) Drift tube-linac, 4) High energy linac, 5) Accumulator ring
with H--charge exchange injection, 6) Beam transport to target, 7) Target, 8) Moderators,
9) Neutron channels
b) Time structure of beams at linear accelerator and at target

Table 4
Tentative parameters for ESS

H- energy (GeV) 1.33

Linac duty cycle (%) 6 to 10

Linac repetition rate (Hz) 50

Linac pulse length (ms) 1.23

Linac peak current (mA) 107

(with 60 % chopping efficiency)

Linac average current (mA) 3.8

Linac rms emittance ( rad.m) 2 to 4

Linac p/p after debunching 0.7 . 10-3

Accumulator ring circumference (m) 163

Number of circulating protons in one ring 2.6 x 1014

Energy of proton pulse at target ~ 100
from two rings (kJ)

Length of proton pulse (sec) ~1

Fig. 3 Schematic layout: a) LAMPF linac (1972), b) Proposal for ESS-linac (1994), c) time
structure of chopped linac beam and of beam ejected from an accumulator ring at ESS

Present experience with high intensity proton and deuteron linacs makes it advisable not
to exceed a peak current per rf pulse of about 100 mA. This value is also limited by the
performance of present H- sources and requires already the use of two H- sources in parallel.
Similarly the low energy section up to about 7 MeV is doubled, necessitating the use of a beam
funnel.
The use of high intensity accumulator rings asks for very low injection losses down to a
level of kW only. This is the reason for using charge-exchange (H- --> H+) injection by a thin
stripping foil; it implies of course acceleration of H- in the linac [7]. The application of a low
loss H--injection limits the number of injection turns per pulse to about 1000. This keeps the
average number of passages of a given particle through the stripping foil sufficiently low (about
1.5 to 2!). Another limit is given by fast growing instabilities in the high intensity rings with
2.6 x 1014 circulating protons (exceeding by a factor 10 the corresponding figure at ISIS). As
a consequence the linac pulses should not exceed a length of about 1 to 2 msec. The
combination of this pulse length with a repetition rate of 50 Hz leads to a duty cycle around 5 to
10 %. A direct use (without rings) of linac pulses at a third target station is considered. For
increased flexibility it is planned to design the linac for a somewhat larger duty cycle.

One turn ejection by a fast kicker needs a time gap in the circulating proton current of
about 200 nsec and requires therefore chopping of the linac H- beam. The chosen time
structure is 400 nsec/270 nsec with a chopping efficiency of 60 %.

In order to keep the ring aperture (and ring costs) as small as possible, emittance increase
along the linac should be kept as small as possible.
The high average beam power of 5 MW asks for a careful choice of particle energy. The
chosen proton energy arises from a compromise between many conflicting requirements
amongst others average beam current, power density and radiation damage at target
components, target (neutron) efficiency and overall cost. A first choice is 1.33 GeV but it may
still be changed by the requirements of target performances and component life time. Whatever
the choice of energy is, the large beam power of 5 MW implies two more important design
principles
A high acceleration efficiency, especially for the high energy part of the linac
Very small relative beam losses for keeping activation of linac components below a level
where "hands on" maintenance and repair will still be possible without intolerable time
delays

5. THE ESS LINAC LAYOUT (Fig. 3)

As already pointed out it is considered for the ESS linac to double the low energy section
for decreasing space charge effects and in particular blow-up of longitudinal and transverse
emittances.

Behind the H- ion sources with a maximum current output of 60 to 70 mA one or two
RFQ's may be used for an adequate acceleration, focusing and bunching of the beam and its
shaping for the following DTL. Because of low energies, beam losses of the order of 10 % can
be tolerated in this section. Two more beam elements a chopper and a funnel have to be
integrated in the low energy lines.
The beam chopper will be a pulsed deflecting device, whereas the beam funnel will be a
deflecting cavity operating at the same frequency as the RFQ's and at half the frequency of the
following DTL. RFQ's, chopper and funnel have to be integrated in the beam lines with
focusing elements. As can be seen from Fig. 4, longitudinal and lateral space will be at a
premium in such a layout, especially for longitudinal focusing. More studies and tests will be
needed to arrive at a satisfying layout for the low energy sections. A classical drift tube linac
(DTL) operated at 350 MHz will be used for acceleration between 5 and 100 to 130 MeV.
Because of the large duty cycle of 5 to 10 % a careful design of drift tube and quadrupole
cooling will be necessary.
For the acceleration above 130 MeV up to 1.33 GeV ( = 0.5 to 0.9) two possibilities are
considered:
A normal conducting version with side coupled cavities of the type developed for LAMPF
and operated extensively since 1972. A frequency of 700 or possibly 1050 MHz is under
discussion. The possible use of disk-and-washer cavities or of annular coupled
structures is considered
A superconducting version with iris loaded cavities. This cavity type is already used
extensively in large e storage rings but has not yet been developed and used for the
velocity range = 0.5 to 0.9. Sc cavities could present the following advantages: If a
frequency of 350 MHz is used a frequency jump always a source of increased beam
losses can be avoided between DTL and the high energy section. A much larger iris
opening could be used (e.g. 12 cm as compared to few cm for a 700 Hz side-coupled
cavity). Although we cannot yet predict quantitatively the decrease of activation by beam
losses at larger iris openings one may hope that large openings combined with strategically
placed scraper families may offer more flexibility for reducing beam losses inside sc
cavities and other beam components. Sc cavities will also allow to increase the acceleration
efficiency [9] because of their low losses. For typical nc and sc cavity parameters and CW
currents of the order of 100 mA the gain may be as high as a factor 2. The picture does
however look less promising if only short pulses (msec) can be used.
Fig. 4 Possible layout of a 2.5 MeV transport line between an RFQ and a DTL with periodic
focusing and bunching. The position of quadrupoles (Q), quadrupoles triplets (T),
bunches (B), a chopper (CH) and its beam collector (C) is indicated. Above, the
transverse and longitudinal beam envelopes. Note that the chopper and collector are
placed at a triple waist of the beam envelope [8].
The filling times of sc cavities under ESS conditions are of the order of msec and rf
pulses have to be made longer for reaching adequate phase and amplitude stability during
the beam pulses.

6. BEAM LOSSES

One of the most important design aspects of high intensity proton accelerators is to
produce small beam losses; component activation along the linac should be kept sufficiently low
to permit maintenance and repair with short notice access.
At energies above 120 MeV activation is mainly produced by spallation neutrons and is
proportional to E0.9. At energies between ~10 and 120 MeV activation is lower but depends in
a complex way on the energy and nuclear cross sections of the material. The range of protons
in Cu or Nb is approximately:
at 10 MeV 0.3-0.4 mm
at 100 MeV 11 - 13 cm
at 1 GeV 60 cm

There exist reliable data on beam losses and component activation at LAMPF [10]. For
the high-energy part of the accelerator beam losses are kept below 1 nA/m, corresponding at
800 MeV and after 1 day of deactivation to a dose rate near the cavities of 20 mrem/h. This
dose rate will already limit exposure times of persons to a few hours. In copper cavities the
main activity is due to Cu64 with a half-life of 13 h. Activation in Nb components decays with
a half-life of about 1 month.
Beam losses in high intensity linear accelerators are today under intensive study. One
uses multiparticle tracking codes and tries to describe losses and in particular halo formation by
additional analytical modelling. Codes taking into account space charge forces in a (simplified)
transverse and longitudinal focusing channel can predict today relative beam losses down to
about 10-3. A much better understanding of beam behaviour, in particular halo formation will
be needed to cope with relative beam losses in the range of 10-4 to 10-5 needed for the ESS
linac. Avoiding beam losses at this level will be a major challenge for the realisation of high
intensity accelerators.
Mismatches and misalignments are major sources for emittance growth and particle losses
(Figs. 5, 6). Changes in rf frequency and cavity type are particularly prone to mismatches.
Non-linear forces always produce filamentation of beams and the concomitant increase of
effective emittance is particularly large if mismatches are involved. Compared to the LAMPF
accelerator a much more careful matching at all stages of acceleration is foreseen for keeping
beam losses at ESS small.

Fig. 5 Beam loss along the LAMPF linac as a function of proton energy. Maxima of beam
losses occur at transition DTL-SCS (201 MHz --> 805 MHz, 100 MeV) and at change
of transverse focusing (212 MeV) [10].

Fig. 6 Activation versus distance from the start of the LAMPF SCS. As expected the loss is
greatest at the quadrupole doublets where the beam diameter are largest [10].
7. A SPECIAL PROBLEM IN A H- BEAM: Lorentz-stripping

In a spallation source the injection from a linac into the ring accelerator is done by
stripping of H- ions because this process permits small injection losses. Consequently one has
to produce and accelerate H- ions in the linac. As the extra electron in the H- ion is very weakly
bound it can be dissociated by the electric field created in its rest-system by an external magnetic
field and in particular by bending magnets and by the quadrupole fields needed for focusing
along the linac. The probability of neutralisation [11] depends on the H - kinetic energy and on
the strength of the quadrupole field which increases linearly with the distance from the axis.
For a typical focusing quadrupole in the high region one needs a gradient of about 20 T/m and
a quadrupole length of 30 cm (independently of the use of nc or sc cavities). In Fig. 7 the loss
rate per quadrupole doublet as a function of H- kinetic energy and of the distance from the axis
is shown [12]. This loss rate should not exceed a value of about 1 nA/m at each doublet. With
an average current of 3.8 mA at 1.33 GeV this corresponds to a loss rate per doublet of 0.25 x
10-6 and forces one to keep the beam radius at the doublets well below 2 cm. With focusing
quadrupoles between all cryostats the large iris openings of sc cavities cannot be exploited for
larger beam dimensions because of Lorentz-stripping. To alleviate the problem the use of
doublet focusing between cavities has been proposed.

Fig. 7 Neutralisation rate of H- in a typical quadrupole doublet for an ESS-linac as a function

of H- ion kinetic energy. Parameter: distance of particles from axis inside
quadrupoles in cm [12].

8. CONCLUSIONS

Pulsed accelerator driven spallation sources for neutron scattering are a promising
alternative to CW reactor sources, because they allow high peak neutron intensities to be
reached and a more efficient use of neutrons by time of flight methods. This is an important
advantage for a field which has been traditionally intensity limited. There exist already a
number of pulsed spallation sources and plans for even higher intensity sources with beam
powers of the order of MW, the most ambitious being the European Spallation Source (ESS)
with a beam power of 5 MW, a pulse length of ~ 1 sec and a repetition rate of 50 Hz. With
present day technologies this can only be obtained by a combination of an RF-linac and an
accumulator or accelerator ring. For the linac peak and average currents of 100 mA and a few
mA resp. at 1 GeV will be needed, well exceeding performances of existing proton linacs.

Because of space charge limits the low energy section of the linac will consist of two
beam lines combined by funnelling. The high energy end may use either normal conducting or
superconducting accelerating cavities. Special problems arise because H- ions instead of
protons have to be produced and accelerated so that a low-loss charge-exchange injection into a
ring can be applied. Fast one turn ejection from the ring also necessitates a chopped beam in
the linac. Besides avoiding emittance increase by space charges, low beam losses and a high
acceleration efficiency will be dominating design criteria for the linac and the ring.

The high intensity linac can be considered for many aspects as a prototype for linacs to be
used in nuclear waste transmutation and energy production. Although there much higher
average beam currents are considered (100 mA) problems would be somewhat alleviated
because CW operation with protons will be possible.

ACKNOWLEDGEMENTS

I would like to thank my colleagues from the ESS group, in particular K. Bongardt and
M. Pabst, for many fruitful discussions.

* * *
REFERENCES

[1] I.S.K. Gardner, EPAC, Rome 1988, p. 65

[2] J. Schapira, this Accelerator School

[3] J. Finney in Advances of Accelerator Physics and Technologies, Ed. H. Schopper,

World Scientific 1993, p. 624

[4] H. Aiginger et al, EPAC Berlin (1992) and M. Regler, private communication

[5] G. Bauer, Proc. ICANS XI, Tsukuba, 1990, 41

[6] H. Lengeler, Proc. ICANS XII, Abingdon, May 1993, Rutherford Appleton Laboratory,
Report 94-025, volume II, p-1

[7] G.H. Rees, CAS, 5th General Acc. Phys. Course, Jyvskyl, Finland, 1992, CERN
94-01 p. 731

[8] K. Bongardt, KFA Jlich, private communication

[9] H. Lengeler in [3], p. 321

[10] Proceedings: Technical Review for Next Generation Spallation Neutron Source, Los
Alamos 1993, p. 68

[11] A.J. Jason, D.W. Hudgings and O.B. van Dyck, IEEE NS -28 (1981) 2704

[12| M. Pabst, KFA Jlich, private communication

CYCLOTRONS IN RADIOTHERAPY
P. Mandrillon
Laboratoire du cyclotron, Centre Antoine Lacassagne, Nice, France

Abstract
Radiation therapy is one of the most frequently used methods for treating
cancer patients. Therefore improving the quality of the treatment by
using new irradiation technologies is one of the radiotherapist's constant
concerns. The use of accelerators for delivering protons, heavier
charged particles and secondary beams such as neutrons brings further
improvement in dose distribution and in interesting biological properties
for radiations producing a high Linear Energy Transfer (LET).
Cyclotrons have been present in this field since the early history of
radiotherapy and they could play a major role for future dedicated
hospital-based facilities which include not only an accelerator but also
specific beam-transport and beam-delivery systems. Modern cyclotron
technology exists today to meet all of the requirements within a
reasonable budget. Specific performances and several cyclotron designs
are presented.

1. INTRODUCTION

Heavy particles (protons, light ions, neutrons) offer a unique opportunity for improving
the quality of radiation therapy. Their use has been pioneered over four decades at Berkeley
where there was a very active collaboration between nuclear physicists, cyclotron designers,
physicians and biologists and this fact is well illustrated by the early history of the cyclotron and
its application in radiotherapy.

Ernest Lawrence, inventor of the cyclotron, started in the late thirties a collaborative work
with his brother John [1], a young physician, Paul Aebersold and Richard Stone [2] on the use
of neutrons in the treatment of cancers. But there was at this time no basic biological
information available to provide a rationale for the use of neutrons in preference to X-rays.
Therefore an unexpectedly high incidence of late morbidity appeared. A renewed interest
occurred in the early sixties following a great deal of radiological work carried out by
M. Catterall [3] and D. Bewley in the Hammersmith Hospital in London. This work explained
the radioresistance of tumours and the role of the LET. Therefore the specifications required for
neutron therapy are now well known and modern cyclotrons are the standard tool for this
application.

In 1946, Robert Wilson [4] mentioned that the properties of beams of mono-energetic
charged particles such as protons and ions (deposition of a large fraction of their kinetic energy
in a small volume at the end of their range (Bragg peak and distal dose fall off), small lateral
scattering) could lead to a new radiotherapeutic tool. Again the high energy cyclotrons in
Berkeley [5-6] were used for the first treatment with charged particles.

Table 1 presents a short review of cyclotron-based facilities running a medical programme

(most of them part time). The figures on patient statistics are from the Newsletter "Particles"
issued in January 1994, which is sponsored by the proton therapy co-operative group (PTCOG)
and edited by Janet Sisterson from the Harvard Cyclotron Laboratory. The type and the
characteristics of the cyclotrons have been added in the two last columns.
Table 1
Operating cyclotron facilities
 Institution, Place First Treated patients Accelerator type Maximum
treatment (date of total) energy
(MeV)
Harvard, Mass. (USA) 1961 6010 Dec. 1993 Synchrocyclotron 160
LBL 184 inch (USA) (1) 1954 2200 1991 Synchrocyclotron 932-a
Uppsala (Sweden) (2) 1957 107 June 1993 SF-synchrocyclotron 200
Dubna (Russia) 1964 108 Aug. 1992 Synchrocyclotron 680
St Petersbourg (Russia) 1975 719 June 1991 Synchrocyclotron 1000
PSI (Switzerland) 1984 1363 May 1993 Isochronous cyclotron 72
Clatterbridge (UK) (3) 1989 463 Jan. 1994 Isochronous cyclotron 62
Louvain-la-Neuve (B) (3) 1991 21 Nov. 1993 Isochronous cyclotron 85
Nice (France) (3) 1991 370 May 1994 Isochronous cyclotron 65-H-
Orsay (France) (4) 1991 235 May 1993 Synchrocyclotron 200

(1) From 1975 to 1987 the old pioneer 184-inch synchrocyclotron was also used for AVM treatment and
earlier for pituitary irradiations.
(2) The old 185 MeV synchrocyclotron has been converted into a 200 MeV sector-focused synchrocyclotron
(3) These modern lower energy isochronous cyclotrons are also used for neutron therapy.
(4) Old synchrocyclotron improved in 1973. The maximum energy ( 200 MeV) will be used in the future for
large-field treatment. It is presently used at lower energies (73 MeV) for eye treatment.

Table 2 presents planned facilities and new designs based on cyclotrons for high energy
proton therapy. The PSI (Switzerland) facility will be operational in 1994. Proton therapy is
also scheduled or proposed in several nuclear physics research centres using sophisticated
variable-energy cyclotrons. The four last lines present new designs for dedicated hospital-based
cyclotrons which are proposed by research centres and European industrial companies (IBA-
Belgium and SIEMENS-Germany). The characteristics of these designs are presented in the last
section.
Table 2
Planned facilities and new designs for proton therapy based on cyclotrons

Institutions Cyclotron type E max Remarks

(MeV)
PSI, Villigen (CH) Degraded beam from the 590-MeV 250 Dynamic beam spreading
ring cyclotron New gantry
Ready in 1994
KVI/AGOR (NL) Superconducting compact 200 Variable energy for
nuclear physics
Uppsala (S) SF synchrocyclotron 200 Gantry project
HMI, Berlin (D) Separated-sector cyclotron ~ 75 Eye treatment
TRITRON, Munich First fully superconduting cyclotron 64 Eye treatment
(D) (RF + Magnet)
MSU (USA) Superconducting compact 250 Dedicated
NAC (South-Africa) Superconducting Ring Cyclotron 200 Dedicated
IBA (B) Room temperature cyclotron 235 New industrial designs for
SIEMENS/CAL Superconducting compact 238 dedicated hospital-based
facilities

To date the cyclotrons which were used for medical purposes have been machines originally
built for nuclear physics research. It is time now for accelerators dedicated to medical use, with
sufficient energy to reach deep-seated tumours. Modern cyclotrons associated with specific
beam delivery systems could play a key role for the development of hospital-based facilities
aiming at treating a large number of patients.

2. PHYSICAL AND BIOLOGICAL RATIONALE

Therapy with heavy particles (protons, light ions, neutrons) is based on two factors:
- improved physical selectivity for charged particles, which means the delivery of a
homogeneous dose to the tumour volume while minimizing the dose to the surrounding
healthy tissues.
- improved biological effectiveness for light ions and neutrons due to the dense ionizing
tracks produced by these particles.

2 . 1 Physical selectivity

Figure 1 presents a comparison (from Dr. H. Blattmann, PSI) of depth-dose distribution

for the various types of radiation used in radiotherapy.
The advantages of protons are based on two
characteristics:
- The dose delivered by a proton rises as the
proton slows down, reaching about 4 times the
entrance dose near the stopping point, the '
Bragg peak', and is zero beyond that point.
Protons in a mono-energetic beam have very
nearly the same range (the straggling is weak)
and then deliver the maximum dose at the same
depth which is determined by their kinetic
energy.
- Protons do not deviate very much from a
straight line so that very sharp collimation is
possible (the multiple scattering is also weak).
A range modulation is necessary to deliver a
uniform dose to tumours whose depth exceeds the
width of the Bragg peak but even a range modulated
proton beam has a very favorable dose profile to treat
Fig. 1 Depth-dose distribution for deep seated lesions.
various types of radiation

2 . 2 Biological effectiveness

Densely ionizing tracks produced by light ions or by neutrons, when they hit a cell,
produce more concentrated local damage and have a more immediately lethal effect than the
scattered minor lesions inflicted by weakly ionizing particles. As a result, these high-LET
radiations exhibit differences in their biological properties when compared to low-LET radiation
(photons and protons) which may be summarized as follows:
a) Reduction in the differences in radiosensitivity for different types of cell.
b) Reduction in the differences in radiosensitivity related to the phase of the cell in the mitotic
cycle.
c) Less repair phenomena and, as a consequence, less difference between the responses of
the cell populations to fractionated irradiation.
Therefore all cell populations, in all conditions, tend to respond in a similar way when exposed
to high-LET radiation. This has important consequences for hypoxic cells, less sensitive to
radiation, and often present at the centre of large, poorly-vasculated tumours. (Radiochemical
reactions in water are in many cases mediated by free-radical oxygen atoms, so it is not
surprising that the absence of molecular oxygen reduces the sensitivity.) The effect of oxygen is
less important for high-LET particles.
 It follows that high-LET (light ions and neutrons) treatment is particulary useful for
radioresistant tumours, insensitive to normal X-rays. These will often be large,
deoxygenated tumours with slowly growing cells.

Figure 2 shows the two quantities

which express these biological effects
relatively to photons: the Oxygen
Enhancement Ratio (OER) and the
Relative Biological Efficiency (RBE)
versus the LET. For light ions the
increased ionization density leads not
only to a dose rising with depth, but also
to a change in the radiation quality.
Radiation at higher LET has an increased
RBE up to about 200 keV/mm, beyond
which the RBE falls. Consequently the
biologically effective dose at a depth,
relative to the entrance region, increases
even more than the physical dose.

Fig. 2 Biological effects relative to different photons

3. MEDICAL REQUIREMENTS FOR A FACILITY

The main requirements could be summarised as follows:

Accelerator is highly reliable, easy to operate and repair time is low.
Possibility to use different directions of the beam: a variable incidence beam (the so-called
rotating gantry) is suitable.

3 . 1 Requirements for heavy charged particles

- Range in tissues: minimum 3 cm for eye treatment, typically 20 cm, maximum 25 cm

(even more if passive beam spreading devices are used!). This specification fixes the
energy range.
- Maximum dose rate at the tumour: 5 Gray/minute in a 2 litre volume. This fixes the
maximum intensity (cf. Table 3)

- Maximum irradiated field: 30 x 30 cm2.

Energies, magnetic rigidity for 20 cm range and intensities (pps) required to deliver a dose
of 5 Gray over a volume of two litres in one minute are listed in Table 3. From the energy
required it is clear that the proton machines and light-ion machines will be quite different in size
and cost: the cyclotron is a good competitor with linacs and small synchrotrons for proton
therapy, it is close to the limit for carbon and the synchrotron is the best candidate for neon.

Table 3
Energy, magnetic rigidity for 20 cm range and intensities required to deliver
a dose of 5 Gray over a volume of 2L in one minute for different particles

Particle Energy Magnetic rigidity Intensity required

(Tm) (pps)
(MeV/n)
 p 175 2.00 2.10 10
C 330 5.67 10 9
Ne 470 6.98 5.10 8

3 . 2 Requirements for neutron therapy

Many aspects need to be considered in order to specify the characteristics of the accelerator
and the most relevant are summarised below:

Skin protection: this fixes the position of the build-up >2 cm

Depth dose distribution: the 50% isodose should be, at least, at 15 cm depth (see Fig. 3)

treatment time not exceeding 4 minutes to obtain treatment conditions similar to photons.
Taking into account the geometry of the collimator which defines the distance between the
target and the patient, this condition fixes the intensity of the beam delivered on the target.

Fig. 3 Depth-dose distribution of the MEDICYC neutron beam

A fixed-frequency isochronous cyclotron, giving a fixed energy beam, is able to produce

large intensity. This quality has made the medium-energy cyclotron the work horse of neutron
therapy programmes. A 60-MeV, 30-A proton beam is considered nowadays to be the
standard beam to meet these requirements (deuterons might also be used, e.g. the
superconducting cyclotron installed in the Harper Hospital in Detroit).
4 . CHARACTERISTICS OF A CYCLOTRON FOR PROTON THERAPY

The development of radiotherapy with proton beams in the 200-250 MeV range is a
challenge to the accelerator community. Designing a cost effective accelerator, easily installed in
a hospital and simple to operate is an overriding concern. A small footprint of the accelerator
and its associated switchyard and gantries is an important factor to reduce the total cost of the
facility. Therefore cyclotrons are good competitors with synchrotrons and linear accelerators for
reasons of energy, magnets, intensities and operation, and these aspects will be described in the
following sections.

4 . 1 Energy
 The choice of a fixed energy makes the design simple. The magnet can then be optimised
and trimming coils are not necessary. The accelerator is at a fixed radio frequency and all the
settings of the beam lines are fixed.

4 . 2 Magnet

In order to reduce the dimensions of the cyclotron a high magnetic field should be chosen.
This choice has two important consequences :

It is impossible to accelerate negatively-charged ions due to electromagnetic stripping.

(The MEDICYC high field compact cyclotron in Nice accelerates H - ions up to 65 MeV,
very close to the limit of stripping losses by the magnetic field.)

Superconducting magnets are attractive because they are about half the weight of room
temperature magnets.

4 . 3 Intensity

An isochronous cyclotron provides a continuous beam with ample intensity. This is a key
factor for reliability. Also the beam intensity is easily controlled. The beam can be very stable
and this is an important advantage for a dynamic beam-spreading system. To fully benefit from
these advantages an external injection system makes the cyclotron very flexible. On the
MEDICYC cyclotron which uses an axial injection at 33 kV for the H- beam, a feedback control
of the extracted beam intensity is installed using a few volts on the bunching cavity. A beam
stability better than 1% could be easily achieved (cf. Fig. 4).

4 . 4 Operation

No sophisticated controls are needed. A simple programmable Logic Controller is

sufficient. The operation requires less complexity and this reduces the manpower costs.

5. PATIENT TREATMENT USING A CYCLOTRON

The main characteristics of a cyclotron beam are its continuous time structure (apart, of
course, from the RF structure at several tens of MHz) and the good extracted beam emittances
and energy dispersion. These properties have important consequences for the beam delivery
system and for the equipment, the so-called rotating gantry, which provides entry of the beam
into the patient from any arbitrary direction specified by therapy planning.
 Fig. 4 MEDICYC intensity fine adjustment ( ) and control (V)

5 . 1 The beam delivery system

The acceleration of particles takes only 10 to 20 microseconds. So the beam can be turned
on and off rapidly by acting on the injection process (the device needed can be extremely
simple if one uses an external injection line). Furthermore it is possible to set up a dynamic
control of the beam intensity with reference to a fast ionisation chamber in the extracted beam
[7], with the effect that the beam can be stabilised at any required level with a response time of
less than 100 microseconds. Therefore the beam intensity could be strictly constant to better
than 2% over short and long periods, and it can be set to any desired level from a few percent to
the nominal level. This process can be fully automatic, a separate safety circuit ensuring that no
misoperation of this servomechanism can lead to an overdose for the patient. If a failure is
detected the beam can be turned off in less than a few tens of microseconds.

These characteristics have important consequences for the beam delivery systems
associated with a cyclotron. Any-beam delivery system must accomplish a three-dimensional
scanning of the tumour which requires:

lateral deflection of the beam,

variable range,
variable exposure time to achieve a uniform dose,
position-sensitive monitors,
rapid switch-off in case of malfunction.

The lateral deflection of the beam for irradiating large fields can be obtained either by
passive scattering using a contoured scatterer (cf. Fig. 5b) following the methods and results of
B. Gottschalk [8] or by active methods using a magnetic field to deflect the beam. Figure 5a
shows the price to pay for that method which leads to increasing the cyclotron energy to
compensate for the range losses in the scatterer.

F. Farley [9] has proposed a very simple method which is easy to implement with a
cyclotron and is shown in Fig. 6. It uses a dynamic beam spreading mode by a deflecting
magnet and requires:
`
Fig. 5a Performance of beam-spreading methods for illuminating a given field

Fig. 5b Beam spreading with a contoured scatterer

 An upstream modulator for range scanning at 50 Hz and which will spread the beam to a
Gaussian spot (fixed or variable, it does not matter!).

A magnet which scans this spot linearly along a line perpendicular to the axis at a low
frequency (1 Hz typical, so only a low power is needed in the magnet which has a small
gap because the deflection is in one plane only).

The patient is displaced steadily at 5 mm/sec. The combination of these two effects creates a
raster scan with uniform intensity. Of course the usual patient collimator and bolus move with
the patient.

Fig. 6 Simple dynamic beam spreading by moving the patient

This simple method is not difficult to implement and could be easily extended to spot
scanning and conformal therapy, due to the simplicity of intensity control of the cyclotron which
has the advantage of using a rapid feedback for eliminating fluctuations and for varying the
intensity under computer control of the intensity on the injection line. At PSI [10] a discrete spot
scanning technique has been developed. For each position the dose is applied by switching on
and off the beam with a kicker magnet which is located within the gantry.

5 . 2 Rotating gantries

In order to reduce the dose to the surrounding healthy tissue a standard technique
conventionally used in radiotherapy is to treat the tumour several times with beams coming from
different directions. This device is called a rotating gantry: the beam comes in along the axis, is
guided away from the centre and bent back towards the axis to cross it orthogonally. If the beam
line ends at the axis (i.e. the patient is located on the axis) the gantry is said to be isocentric.
This component represents a significant fraction of the total cost of the facility. The energy of
the beam extracted from the cyclotron is degraded at the exit of the gantry, so the settings of all
the magnets are fixed. In contrast with a synchrotron where the energy is changed at the
accelerator this whole beam line needs to be reset exactly. This factor could reduce the costs of
this equipment.

Several types of gantry have been studied in the past years, Fig. 7 shows schematically
their main properties and typical dimensions.
 Fig. 7 Various gantry designs

6. DEDICATED CYCLOTRON DESIGNS

6 . 1 Cyclotrons for neutron therapy

The high intensity required for this radiotherapy is easily achievable from isochronous
cyclotrons. Figure 8 shows the median-plane view of the 65 MeV room temperature cyclotron
designed and installed in Nice. Use has been made of H- acceleration to facilitate the extraction
system [11]. A 62-MeV commercial cyclotron designed by Scanditronix is installed in the
Clatterbridge Douglas Centre. A compact superconducting cyclotron rotating around the patient
has been constructed by H.G. Blosser [12] and is running in the Harper Hospital in Detroit
(USA). The 40 MeV deuteron maximum energy orbit has a radius of only 30 cm (cf. Fig. 9).
The structure of this cyclotron weighs less than 25 tons and this allows the entire cyclotron to be
mounted directly on a rotating gantry so that neutrons produced in an internal target can be
directed at a supine patient from any direction.

6 . 2 Cyclotrons for proton therapy

For proton therapy at 6090 MeV, useful for treatments such as the ocular melanoma,
several medium-energy cyclotrons are operating. This energy domain is close to that of neutron
therapy and therefore several cyclotrons are running both programmes; Clatterbridge (UK) and
Nice (F) which are fully dedicated machines, Louvain-la-Neuve (B) which is running part time
for medicine, are typical examples.
There is now an increasing number of new projects aimed at treating many tumour types at
any depth and therefore cyclotrons in the 200 MeV range are good candidates. These machines
could use either a room-temperature magnet or a superconducting magnet. The characteristics of
four designs are presented below.

6.2.1 A room-temperature design

A commercial design [13] by the Ion Beam Applications company in Belgium has chosen a
high field (2.15 Tesla on the extraction radius) produced by conventional coils (cf. Fig. 10 and
Table 4). This cyclotron and the associated gantry will be installed in the coming years in the
new proton therapy facility to be built at Massachusetts General Hospital in Boston (USA).
Fig. 8 Median-plane view of the Fig. 9 Median-plane view of the
MEDICYC 65-MeV cyclotron rotating cyclotron in Detroit

Characteristics of the IBA 235-MeV cyclotron

Magnet
Number of sectors 4
Yoke external diameter 4.30 m
Current density in the coil 155 A/cm2
Total weight 220 tons
Power 200 kW

RF
Two dees in opposite valleys
Frequency 107.17 MHz
Harmonic mode 4
Dee voltage injection 60 kV
Dee voltage extraction 130 kV
RF power 85 kW

Miscellaneous
Fig. 10 IBA 235-MeV, room-temperature Injection: Internal
cyclotron. source

Table 4
6.2.2 Compact superconducting cyclotron designs

Figure 11 shows a compact superconducting sector-focused cyclotron capable of

producing a continuous beam of 238 MeV protons. It has three spiral sectors with a
supplementary groove along the centre of each sector, making it more like a six-sector machine
and thus avoiding the well known r = 3/2 resonance (this technique has been well established at
the AGOR facility in Holland). Three RF cavities, galvanically connected in the central region,
operate on the third harmonic of the particle frequency. They provide a large energy gain per
turn in order to complete the acceleration in about 400 turns. Table 5 gives the main
characteristics of this design which was made by SIEMENS and The Cyclotron Laboratory of
CAL in Nice [14].
 Fig. 11 The CAL/SIEMENS 238-MeV, superconducting cyclotron.

Table 5
Main characteristics of the CAL/SIEMENS 238-MeV superconducting cyclotrons

Magnet
Number of sectors 3 i.e. "pseudo 6 sectors"
Yoke external diameter 3.10 m
Current density in superconducting coil 6800 A/cm2
Total weight 80 tons
Power including refrigerator 35 kW

RF
3 dees in galvanically connected
Frequency 110MHz
Harmonic mode 3
Dee voltage injection 80 kV
Dee voltage extraction 130 kV
RF power 120 kW

Miscellaneous
Injection: Two options axial or radial
(neutral beam + central stripper)

H.G. Blosser from Michigan State University (USA) has designed a 250-MeV four-
sector, compact, superconducting cyclotron for proton therapy [15]. Figure 12 shows a
horizontal section of this machine where magnet hills and dees repeat with 90 spacing. Protons
are produced by an internal ion source.
 Fig. 12 The MSU 250-MeV superconducting cyclotron

6.2.3 A superconducting-ring cyclotron design

Figure 13a shows the layout of a 200-MeV superconducting-ring cyclotron proposed by

the National Accelerator Centre in South Africa [16]. The magnet is made of four 15-ton sectors
with superconducting S-coils located in separate cryostats at the radial boundary of each sector.
A small internal ion source is located in the central region. The design should, in principle,
allow extraction at different energies by moving radially the extraction elements located in the
low-field valleys. Figure 13b presents the median-plane distribution of the magnetic field with
isofield contours at 0.3 T intervals from 0.6 to 4.2 T within a radius of 1.2 m.

Fig. 13 The NAC 200-MeV cyclotron. a) Layout b) median-plane magnetic field.

6 . 3 Booster cyclotron
 Several neutron therapy centres presently using low-energy proton therapy are examing the
potential for boosting the beams for high-energy proton therapy in the 200-MeV range. With
minor modifications any cyclotron can accelerate negatively-charged ions. It is easy to extract a
neutral beam using a thin stripper. This beam will have good optical properties, in particular a
small divergence, allowing it to be transported a long way and traverse the magnetic field of a
compact superconducting accelerator. This neutral beam is injected by a second stripping foil at
the appropriate radius. Figure 14 presents a schematic layout of such a booster [17]. An
interesting feature is that all commercial cyclotrons dedicated to the production of radio-isotopes
are accelerating negatively-charged ions and could be used as injectors.

Fig. 14 Boosting an existing cyclotron

7. CONCLUSIONS

Cyclotrons are good candidates for radiotherapy with heavy particles.

7 . 1 Intensity control: flexible, fast and safe

As is well known a cyclotron is capable of accelerating large intensities. Therefore

isochronous cyclotrons are well-suited accelerators for neutron therapy. Moreover, for proton
therapy the acceleration takes only 10 to 20 microseconds (typical value), so the beam can be
turned on and off rapidly by acting on the low energy end of the machine (external injection line
if any or ion source). Furthermore a dynamic control of the intensity can be easily realised.

The system is safe because in the case of a failure or danger condition the beam can be
turned off in less than 20 microseconds by cutting the injected beam. During this delay the
patient would receive at most one millionth of the standard dose.

7 . 2 Costs saving

A standard proton therapy facility based on a cyclotron (small footprint of the equipment)
reduces the cost of the building. To this end a superconducting cyclotron in addition to a
compact gantry is attractive. In addition the running costs are reduced when using a
superconducting coil (total power in the magnet about 30 kW).

7 . 3 Controls and operation: easy and cheap

A fixed frequency isochronous cyclotron, giving a fixed energy to the accelerated ions, is a
simple accelerator to operate: a PC is sufficient to tune and control the cyclotron.
REFERENCES

[1] J. Lawrence and E. Lawrence, "The biological action of neutron rays" Proc. National
Acad. Sciences 22: 124-133 (1936)
[2] R. Stone, J. Lawrence and P. Aebersold, "Preleminary report on use of fast neutrons in
treatment of malignant disease" Radiology 35: 322-327 (1940)
[3] M. Catteral, "Fast neutrons in oncology", British Journal of Hospital Medicine, 12,853-
860 (1974)

[4] R. Wilson "Radiological use of fast protons "Radiology 47, 487 (1946).
[5] C. Tobias, H. Anger and J. Lawrence, "Radiobiological use of high energy deuterons and
alpha", American Journal of Roentgenology 67: 1(1952).

[6] C. Tobias, J. Roberts and J. Lawrence, "Irradiation hypophysectomy and related studies
using 340 MeV p and 190 MeV d" Proceedings of the Int. Conf. on Peaceful uses of
Atomic Energy held in Geneva in the United Nations.

[7] F. Farley, Oral presentation : "Supertwist gantry and dynamic beam spreading".
XIXth PTCOG, October 31 to November 2, 1993, Cambridge, Mass., USA.

[8] B. Gottschalk and M.S. Wagner, "Contoured scatterer for proton dose flattening",
Harvard Cyclotron Laboratory, a preliminary report 3/29/89 (1989)

[9] F. Farley, private communication.

[10] E. Pedroni et al., The 200 MeV proton therapy project at PSI: conceptual design and
practical realisation, to be published in Medical Physics, NS 94022

[11] P. Mandrillon et al., Commissionong and implementation of the MEDICYC Cyclotron

programme, 12th Int. Conf. on Cyclotrons and their applications, Berlin, May 89.

[12] H. G. Blosser, 1988 Annual report of the National Superconducting Cyclotron

Laboratory, Michigan State University, p. 107.

[13] Y. Jongen et al., "Cyclotron based proton therapy system including a new design of large
throw gantries", to be published in the Proc. of the Como Conference on hadron therapy.
[14] P. Mandrillon et al., A compact facility for high energy proton therapy based on a
superconducting cyclotron, to be presented at EPAC 94 in London.

[15] H. G. Blosser et al., Proposal for a manufacturing prototype superconducting cyclotron

for advanced cancer therapy, NSCL-MSU, February 1993.

[16] H.N. Jungwirth and J.G. de Villiers, 13th Int. Conf. on Cyclotrons and their
Applications, Vancouver 1992, 709-712.

[17] P. Mandrillon and F. Farley, Cyclotrons as booster for high energy proton therapy, to be
published.
ISOTOPE PRODUCTION WITH CYCLOTRONS
V. Bechtold
Kernforschungszentrum Karlsruhe GmbH, Hauptabteilung Zyklotron, Karlsruhe, Germany

Abstract
In the last two decades cyclotron produced isotopes have gained
increasing importance in nuclear medicine diagnosis. The lecture will
cover aspects on target technology with regard to production efficency,
quality, reliability and automation. The production methods for the
most commonly used radionuclides will be described. An example for
two routine production systems based on gas target technology for I-
123 and Rb-81/Kr-81 are presented.

1. INTRODUCTION

Modern isotope production facilities consist of a compact H- cyclotron (Fig. 1) in the

energy range between 10 MeV and 30 MeV with extracted currents up to 350 A [1] and a
highly sophisticated target technology and chemistry [2].

Fig. 1 Compact H- cyclotron built by Cyclotron Cooperation, USA

The radioisotopes produced for medical applications are used in nuclear medicine for
diagnosis. In contrary to X-ray studies where only static information can be obtained, nuclear
medicine provides dynamic information and hence allows time-dependent studies of the
function of organs. The labelled biomolecules and the radioisotopes involved should have the
following ideal characteristics:

no - particle
short effective half-life
-energy in the range between 100 and 300 keV.
 The above characteristics result in a maximum efficiency in the diagnosis and a minimum
radiation dose to the patient. For example, Fig. 2 shows a comparison between I-131 and
I-123. The physical properties show that the cyclotron produced I-123 (-energy 159 keV; no
-; short half-life) is more favourable than the reactor produced I-131 (-energy 364, 637; --
emission, long half-life) for use in nuclear medical diagnosis.

Fig. 2 Comparison of the physical properties between I-131 and I-123

The production of radioisotopes used in nuclear medicine can be made with solid,
gaseous and liquid targets. Figure 3 is an example for a solid target for 30 MeV protons and a
current of 200 A. The target material which is electrochemically plated on a solid copper
backing is hit directly by the beam. The dissipated beam power of 6 kW can be removed by
enforced cooling from the back-side of the target. To keep the power density low the beam
strikes the target under an angle of 57. Thus the temperature on the target surface is kept
below 150 C.

Fig. 3 Layout of a solid target

Figure 4 shows a target layout for gaseous or liquid target material. In contrary to the
solid target the beam cannot hit the target material without entering the target vessel through a
thin metal window. The target vessel itself has to be water-cooled as well to cool down the
liquid or the gas.
 Fig. 4 Layout of liquid or gas target

2. TARGET BODIES AND WINDOWS

The choice of material for both the target body and window is dependent on the particular
nuclide production process. Although a general rule does not exist, there are some aspects of
the target bodies like activation, contamination, corrosion and cooling, which have to be
considered [3, 4]. The parameters are influenced by the choice of bombarding particle, beam
energy, beam current, material and solvent. The criteria for target-window materials should be
the following:

thickness of 1 200 m
pin-hole free
high mechanical strength
good thermal conductivity
high melting point
chemical resistance to oxidation

The target windows can be sealed either by 0-ring or welding. This depends on the
particular boundary conditions of cooling, temperature, pressure and radiation effects.

3. ACTIVITY CALCULATIONS

When irradiating a target material with charged particles from a cyclotron, the
disintegration rate D of a produced radionuclide is:

(
D = IN 1 e t )
where

I = intensity of the irradiating particles (number of particles/cm2 s)

N = number of target atoms
N = W.K/A w 6.02 . 1023.
W = weight of target material
K = natural abundance of target element
Aw = atomic weight
= formation cross section in barn (10-24 cm2)
= decay constant given by 0.693/t1/2 (s-1)
t = duration of irradiation (sec)
1 Ci = 3.7.1010 disintegrations/sec
From the equation it can be seen, that the amount of radioactivity produced, depends on
the intensity and energy (cross section is related to energy) of the bombarding particles, the
amount of target material, the half-life of the produced radionuclide and the duration of
( )
irradiation. The term 1 e t reaches unity when t is approximately 4-5 half-lives of the
radionuclide. At that time the yield of the produced nuclide becomes maximum. Production
rates and decay become equal. Figure 5 shows this effect.

Fig. 5 Production of radionuclides. The activity reaches saturation in 4-5 half-lives.

4. ACTIVITY TRANSFER

After irradiation the produced isotope is still in the target matrix (solid) or inside the target
vessel (gas or liquid). In both cases the activity has to be transferred from the irradiation station
into a hot cell for further processing. This transfer must be safe, reliable and sometimes fast
and can be done either manually or automatically. For dose considerations and for safety
reasons an automatic transfer is recommended. The commonly used transport systems are
rabbit or conveyer devices (for solid targets) and pipelines (for liquids and gases).

5. AUTOMATION

For routine (daily) production a microprocessor controlled system is absolutely

necessary. Either a semiautomatic mode or a fully automatic mode guarantees a high reliability
of the production process and a constant product quality. Moreover a remote handling of the
target and the chemistry prevents personnel from obtaining a too-high radiation dose.

6. MAIN RADIOISOTOPES

Table 1 summarizes the most important cyclotron-produced isotopes for medical

applications. Common features for this radionuclide production are:

production via (p,xn)-reaction

proton energy ranges between 10 MeV and 30 MeV
enriched isotopes for target material
recovery of target material
commercially available
Table 1
Most common cyclotron produced radioisotopes for medical diagnostics
 Radioisotope Half-life Nuclear reaction Bombarding energy
(h) (MeV)
Gallium-67 78.3 Zn-68(p,2n)Ga-67 25
Bromine-77 57 Kr-78(p,2n)Rb-77 30
Rb-77 Kr-77 Br-77
Rubidium-81 4.6 Kr-82(p,2n)Rb-81 30
Indium-111 67.2 Cd-112(p,2n)In-111 22
Iodine-123 13.2 Xe-124(p,2n)Cs-123 30
Cs-123 Xe-123 I-123
Thallium-201 73.1 Tl-203(p,3n)Pb-201 20
Pb=201 Tl-201
Fluorine-18 1.8 0-18(p,n)F-18 18

Methods of preparations for these cyclotron-produced radionuclides are described below.

6 . 1 Thallium-201

Half-life: 73.1 h
Nuclear reaction: Tl-203(p,3n)Pb-201
Proton-energy: 29 MeV
Type of target: solid target, Tl-203 electroplated on target backing
Target material: Tl-203 > 90% enriched
Activity transfer: rabbit or conveyer (irradiated target)
Separation: chemical (Tl-203 from Pb-201; Pb-201 from Tl-201)

Production process

Immediately after the end of bombardment (EOB) the enriched Tl-203 is dissolved in acid
and Pb-201 is isolated by the ion-exchanger method. The Tl-203 is retained on the column for
the further recovery process. Pb-201 is then absorbed on another ion-exchange column and
enough time is allowed for Pb-201 to decay to Tl-201. Tl-201 is then eluted as thallous
chloride in carrier-free form.

Tl-203, Pb-201 at EOB

target dissolved
separation by ion-exchange column

Pb-201 in acid ion-exchange column Tl-203 on column

Pb-201 decay of Pb201 and recovery

growth of Tl-201
Tl-201 eluted from column

Tl-201 as thallous chloride

carrier free
6 . 2 Indium-111

Half-life: 67.2 h
Nuclear reaction: Cd-112(p,2n)In-111
Proton energy: 22 MeV
Type of target: solid target, Cd-112 electroplated on target backing
Target material: Cd-112 > 90% enriched
 Activity transfer: rabbit or conveyer (irradiated target)
Separation: chemical (Cd-112 (extremely toxic) from In-111)

Production process

After irradiation the enriched Cd-112 target is dissolved in acid. The solution is passed
through an ion-exchange resin. In-111 is removed by elution. Cd-112 stays on the column
and is used for further recovery.

Cd-112, In-111
target dissolved
separation by ion-exchange column

In-111 Cd-112 on column

recovery

6 . 3 Gallium-67

Half-life: 78.3 h
Nuclear reaction: Zn-68(p,2n)Ga-67
Proton energy: 25 MeV
Type of target: solid target, Zn-68 electroplated on target backing
Target material: Zn-68 98 % enriched
Activity transfer: rabbit or conveyer (irradiated target)
Separation: chemical (Zn-68 from Ga-67)
Production process

When the irradiation of the enriched Zn-68 target is finished the target is dissolved in
acid. The solution passes an ion-exchange column, where Zn-68 is kept back on the resin
while the Ga-67 passes through.

Zn-68, Ga-67 at EOB

dissolved in acid
separation by ion-exchange column

Ga-67 in acid evaporation Zn-68 on column

Ga-67 recovery
6 . 4 Fluorine-18

Half-life: 110 min

Nuclear reaction: 0-18(p,n)F-18
Proton energy: 18 MeV
Type of target: liquid (water target)
Target material: 0-18 90 % enriched
Activity transfer: pipe line (F-18 in water)
Separation: chemical (fluorine from water)
Production process

The F-18 produced during irradiation is dissolved in the target water. Separation of F-18
from water is done via ion exchange column.

O-18 water, F-18 at EOB

separation by ion-exchange column

F-18 (F-form) recovery of O-18 water

6 . 5 Iodine-123

Half-life: 13.2 h
Nuclear reaction: Xe-124(p,2n)Cs-123 Xe-123 I-123
Proton energy: 30 MeV
Type of target: high pressure gas target
Target material: Xe-124>99.8 % enriched
Activity transfer: pipeline (I-123 activity dissolved in water)
Separation: physical (Xe-gas from solid I-123)

Production process

For a period of 6 hours after the end of bombardment the Xe-123 is still kept in the target
to allow Xe-123 to decay to I-123. In a next step the Xe gas is cryogenically pumped out of the
target into a storage vessel for reuse. The I-123 is dissolved in water and rinsed out of the
target.
Xe-124, Cs-123, Xe-123, I-123 at EOB
6 h decay in target

Xe-124, Xe-123, I-123

pump out Xe-123, Xe-124

rinse I-123 out of target

I-123 (I-form)

7. ROUTINE PRODUCTION SYSTEMS

7.1 Ultra-Pure Iodine-123 production via KIPROS

An example for a gas target arrangement for routine production is the Karlsruhe Isotope
Production System KIPROS (Fig. 6). This system provides ultra pure iodine-123 in large
batch sizes (1 - 4 Ci in 6 h). Iodine-123 has begun to replace the reactor isotope iodine-131
(half-life: 8.05 days) used until now.

Iodine-123 has the following important advantages:

 lower radiation dose for the patient (a factor 60-100 less than for iodine-131)
lower radiation dose for the hospital personnel,
markedly better scintigraph quality and hence a higher diagnostic power,
lower environmental contamination (a factor 5900 lower than for iodine-131).

Up to now, all the known iodine-123 production methods produce large amounts of
undesirable by-products in the form of iodine-124 or iodine-125. Both these isotopes posses
disadvantages comparable to those of iodine-131. With the aim to produce iodine-123 with as
little impurity as possible, a new production method via the Xe-124(p,2n)Cs-123 reaction was
developed:

5.9 min 2.08 h

Cs-123 Xe-123 I-123

Fig. 6 High pressure gas target system KIPROS for the production of ultra-pure I-123. A
sophisticated beam diagnostic system in front of the target assures a proper alignment
of the beam. KIPROS is completely operated with a Simatic S-135 microprocessor
system.
The production of the ultra-pure iodine-123 requires a very highly enriched Xe-124 gas
as target material. In the procedure described here, Xe-124 with an enrichment > 99.8%
(natural abundance 0.1%) is used, with the consequence that the end product iodine-123 has
only a contamination for I-125 of 4 x 10-5 and I-124 of 5 x 10-7 at 24 h after EOB.
As the cost of a litre of enriched Xe-124 is very high it is important to keep the target
volume as small as possible and at the same time be able to control the beam power,
50 A x 30 MeV = 1.5 kW, generated during irradiation.

KIPROS consists of a high pressure gas target, a sophisticated diagnostic system for
proper adjustment of the beam onto the target and a chemistry unit for the delivery of I-123 as
iodide in 0.02 N Na0H. The target body is made out of nickel-plated aluminium to reduce
activation and radiochemical contamination, and to assure proper cooling. The target is
operated at a pressure of 14 bars with beam currents up to 50 microamperes. The production
yield for I-123 with 30 MeV protons is 10 mCi/Ah.

To prevent gas losses in case of target window ruptures a dedicated safety system is
provided. It consists of a safety chamber in front of the target and a fast closing valve (12 ms
closing time) about 8 m upstream in the beam line. The target window can be changed
remotely with a robot (Fig. 7) so that a foil rupture during irradiation does not considerably
influence the production schedule.

Fig. 7 Nickel-plated aluminium target with window flange in front. A robot changes
automatically the target window in case of foil rupture. As part of the target
volume there is a cold finger for cryogenic gas transfer from the storage vessel
into the target (left side).
After an irradiation cycle the xenon gas is cryopumped out of the target and subsequently
the iodine-123 deposited on the cooled walls is washed out by means of 40 ml of distilled
water.

The essential advantages of the KIPROS method are:

highest purity of iodine-123 produced

high production efficiency (300 - 500 mCi/accelerator hour),
low production costs by the operation of a relatively small accelerator facility and
microprocessor-controlled production.

7 . 2 Chemistry and quality control

The hardware configuration of the chemistry unit for the production of I-123 is shown in
Fig. 8 while the flow diagram which illustrates the function of the concentration process is
shown in Fig. 9. At the end of the target wash-out procedure the active solution is purged from
the target into the glass vessel inside the hot cell. An additional vacuum vessel traps the small
amounts of radioactive gas coming from the target. The active solution is pumped through the
ion-exchanger column where Bio Rex 5 is used as the column resin with an absorption
efficiency > 99%. After the loading procedure the I-123 is eluted with 0.02 N NaOH from the
column into the glass vial. The flow direction through the column during the elution process is
opposite to the loading direction.

Fig. 8 Chemistry unit for the concentration of the iodine washout from 40 ml to 1.6 ml
(0.02 N NaOH). An ion-exchanger with a special resin is used. The loading
and reloading of the ion exchanger is carried out by HPLC pumps which allow
precise control of the flowrates. In process measurement of the iodine activity is
possible with small GM counters.

The chromatography pump keeps the flow rate constant, which can be preset before the
loading or elution process starts. At the end of elution the main part of the activity (> 90%) is
dissolved in a final process volume of < 1,6 ml of 0.02 N NaOH. The whole process from the
beginning up to the point when the final product is inside the glass vial can be controlled either
manually or fully automatically. Valves, pumps, cylinders, positions switches and sensors are
connected to the processor (see Section 7.3). The activity is monitored with small GM sensors
at a few positions to guarantee a reliable automatic process, which can be stopped in case of a
system failure (for example: an accidental column breakthrough during the loading procedure).
 Fig. 9 Flow diagram of the I-123 concentration unit

The configuration of the multiport valves enables different samples to be obtained during the
automatic run. The samples are used for the quality control.

The quality control includes tests for the product identity, the radionuclide purity, the
radiochemical purity and the pyrogen concentration. The radionuclide purity is measured with a
Ge detector while the radiochemical purity is determined by the thin-layer-chromatography
method. The pyrogen concentration is checked with the Limulus-test.

7 . 3 Control system

The control system consists of the programmable controller type Simatic S5-135 U, the
communication station with monitor and keyboard and the printer station (Fig. 10). The
controller crate contains the following modules necessary for the operation of the production
process:

Central processing unit (CPU) with a 16-bit microprocessor, internal memories, data
organisation and an external memory for the user software
Communication processor 525 for the process documentation which enables the data
transfer of all the various process parameters to a printer station
Communication processor 526 for the communication between the controller and the
terminal. Process commands can be entered via a keyboard, while status information (for
example: position of valve) and system messages are displayed on the monitor
Fig. 11
Analog input modules for the measurement of analog process values
- Digital input modules for the connection of status indication signals
Digital output modules for the connection of switching components.

The user software for the sequential and logical run of the process is written in a special
language, called "STEP 5", which is optimized for the S5 Simatic controllers. The user
software for the communication processors (CP 525 and CP 526) is written by entering values
in the different masks which are created by the COM 525 and COM 526 system programs.
 Fig. 10 Configuration of gas target control system

If RAMs are used as the external storage medium, the memory is buffered by a back-up
battery in the central crate of the S5-135 U controller. The typical cycle time for the CPU is
3-100 ms, which depends on the user program volume. The edition and the transfer of the user
program to the processor modules are performed with the programmer unit S5-PG 685 which
can also be used for an on-line test of a running program. The status picture (Fig. 11) indicates
the number of the current running program step, the start condition for the next step, a general
description of what is carried out in each step, the activated components and the current value of
a running timer. If the program is automatically stopped by an interlock, the reason is
displayed in the lower part of the status picture, which is reserved for error messages and
system information. At the end of the program the <Program End> information is displayed
and the processor automatically switches back to the initial state. The processor allows a
simultaneous run of two programs. The processor control enables the operation of the target
system with a minimum of manpower and guarantees a reproducible production process and
finally a reproducible quality of the final product.

7 . 4 The Rb-81/Kr-81m generator system

The radionuclide Kr-81m (half-life: 13 s), which is the daughter of Rb-81 (half-life
4.65 h), has gained increasing importance for lung function diagnosis over the last years. The
essential advantages of this radioisotope compared to the other radioisotopes used in this field
today are:
minimal radiation dose for the patient and hospital personnel,
static and dynamic ventilation studies,
no exhaust problems and also no risk of contamination.

The Kr-82 (p,2n) Rb-81 reaction is used for the production. During irradiation the Kr-82
gas (pressure: about 20 bars) is contained in a pressure vessel, which is sealed by a thin
molybdenum-steel entrance window.
 Fig. 11 Colour monitor with keyboard for the communication with the control processor

The target arrangement is similar to the Xe-target. The Rb-81 produced during
irradiation, deposits on the cooled walls of the target vessel and is removed after the end of
bombardment by means of sterile water. The Rb-81 is then pumped 15 m via a stainless steel
pipeline to a hot cell located outside the irradiation room. There the Rb-81 dissolved in water is
purged through a special ion exchanger which traps the Rb-81. This so-called generator is then
fitted into a transportable lead shield and a complete quality test is carried out. The KfK
possesses a pharmaceutical manufacturing permit for the Rb-81/Kr-81m generator according to
paragraph 13 AMG. For the application of the generator in the hospitals it was imperative to
develop a simple application system. The "Kryptovent II" system, (Fig. 12) which has been
developed at KfK, is offered with the generators as a complete inhalation system.

The high pressure gas target system KIPROS is usable for highly-enriched target gases in
particular, where it is necessary to minimize gas inventory and gas losses. This system is used
for the production of ultra-pure I-123, Rb-81 and Br-77 at different production facilities in
Europe, USA and Japan. The appropriate choice of nuclear reaction adapted to a compact
cyclotron (maximum energy 30 MeV protons) combined with a sophisticated technique and a
microprocessor controlled process, guarantees a reliable and cost-effective production of
radioisotopes with high yields and constant product quality.
 Fig. 12 Rubidium-81/Krypton-81 m generator and Kryptovent II for ventilation diagnostic

REFERENCES

[1] V. Bechtold, Proceedings of the 13th Int. Conf. on Cyclotrons and their Applications
Vancouver, July 6-10 (1992) p. 110

[2] F. Helus (Ed.), Radionuclides Production Vol I+II, CRC Press, Inc. Boca Raton Florida

[3] F. Helus and T.J. Ruth, Proceedings of the 1st Workshop on Targetry and Target
Chemistry, Heidelberg 4-7 October 1985, DKFZ press Dept.

[4] F. Helus, S.A. McQuarrie, T.J. Ruth, Proceedings of the 2nd Workshop on Targetry
and Target Chemistry, Heidelberg, 22-25 September, 1987, DKFZ press Dept.
DESIGNING A RADIOISOTOPE FACILITY

Dewi M Lewis
Amersham International, Amersham, UK

Abstract
The physics concepts and methodology of designing a radioisotope
production facility based on a cyclotron is discussed with particular
reference to manufacturing commercially useful isotopes known as
radiopharmaceuticals. Commercially available cyclotron systems will
be discussed and a conceptual facility will be described; the isotope
most used in nuclear medicine, thallium 201, a heart imaging agent,
will be used as a case study isotope. The requirements of Good
Manufacturing Practice for the isotope products and the necessary
compliance with regulatory approvals will be outlined.

1. INTRODUCTION

Cyclotron-produced radioisotopes were first developed for medical purposes at the

University of California, Berkeley, in 1936. For the next 30 years cyclotron construction
proliferated amongst universities and national research laboratories with increasing interest from
commercial organisations for processing the isotopes as radiochemicals. However, in the early
1960's genuine medical applications for several isotopes were identified, e.g. fluorine-18,
rubidium-81 etc. and the first cyclotron project totally dedicated to commercial isotopes
production was established in the UK by The Radiochemical Centre (now known as Amersham
International). For the period 1965-78 several 'first generation' cyclotrons were constructed,
such as the early Cyclotron Corporation cyclotrons. These were essentially research cyclotrons
simplified for industrial operations. However, from approximately 1975 onwards well-
engineered, user-friendly or 'second generation' compact machines became available, e.g.
Scanditronix MC-40 series. A significant technical leap was taken in 1988 when IBA in
Belgium developed a high intensity, low cost cyclotron with negative ion acceleration, high
efficiency extraction and with fully automated controls. Concurrently with the development of
these 30 to 40 MeV cyclotron models, several medically useful cyclotron isotopes were also
being licensed, e.g. gallium-67, iodine-123 etc.; by 1980 thallium-201 was available for
cardiac imaging and has now become the most important commercial cyclotron isotope In
addition to the growth in the medical applications of cyclotron isotopes, legislation for
pharmaceuticals was being extended to include these cyclotron produced materials, i.e. they
became radiopharmaceuticals.

Therefore, a present-day production facility will consist of a cyclotron, several beam lines
and mechanical transport systems and a target arrangement which is custom built to be handled
in the different chemical extraction processes for producing pure radiochemicals and then in the
pharmaceutical production systems for producing sterile, injectable materials which comply
with the pharmaceutical or drug licensing requirements of many countries. The underlying
physics and chemistry of producing cyclotron radiopharmaceuticals will be discussed and a
design of a conceptual facility will be developed within the text.

2. ISOTOPES AVAILABLE FROM CYCLOTRONS

Present-day radiopharmaceutical grade cyclotron isotopes are used for imaging and
diagnosis at nuclear medicine centres worldwide. A typical three-dimensional tomographic
heart image section is shown in Fig. 1 using thallium-201. Some 2-3 million patient
procedures are undertaken each year using materials produced from cyclotrons. The most
common cyclotron produced isotopes are given below in Table 1.

Fig. 1 Typical three-dimensional tomographic heart image

Table 1
Applications of cyclotron produced isotopes
Isotope Half Life Application
(Days)

Kr-81m 0.25 Lung function studies

I-123 0.55 Thyroid studies

In-111 2.80 Monoclonal antibody / infection imaging

etc.
Tl-201 3.05 Myocardial imaging (heart)

Ga-67 3.26 Tumour location (soft tissue)

Co-57 271.7 Calibration devices for cameras

Although even shorter-lived isotopes are also produced by a cyclotron, for the purpose of
this discussion these very short lived isotope (PET or Positron Emission Tomography isotopes,
Table 2) will not be discussed since their commercial usage and distribution have not yet been
realised. In general most of the commercially useful radiopharmaceutical isotopes have half
lives of the order of 1-7 days and their production reactions are given in Table 3.
Table 2
Common PET isotopes

Isotope Half Life (min.)

C-11 20
N-13 10
O-15 2.1
F-18 110

Table 3
Isotope production reactions
 Tl-203 (p,3n) Pb-201 Tl-201
Zn-68 (p,2n) Ga-67
Cd-112 (p,2n) In-111
Xe-124 (p,2n) Cs-123 Xe-123 I-123
Kr-82 (p,2n) Rb-81 Kr-81m
Ni-58 (p,2n) Co-57

The most commonly used commercial cyclotron isotope is thallium-201 which as a

potassium analogue, is readily absorbed within the muscle tissue of the heart, i.e.. the
myocardium. Thallium imaging is used to demonstrate blood flow within the heart muscle by
imaging a patient at rest and under stress; ischaemic tissue and infarctions can be identified
using 3D tomographic techniques. Thallium-201 is also one of the most complicated cyclotron
isotopes to manufacture and the specification of a licensed thallium radiopharmaceutical or drug
is summarised in Table 4.
Table 4
Specification of thallous (Tl-201) chloride injection

Half life 73.1 hours

Energy of gamma rays 0.068 to 0.082 (Hg K X rays)
Radionuclidic impurity Tl-202< 1.9%
Tl-200< 1.0%
Pb-203 < 0.25%
Specific activity > 3.7 GBq.g-1 (or 100 mCi per gm)
Chemical form Sterile, isotonic, aqueous solution of thallous chloride
plus 0.9% (v/v) benzyl alcohol and 0.95% (w/v) sodium
chloride
Reference day Average 6 days post bombardment
Expiry day Reference day + 3
Administration:
Normal dosage 55.5 MBq or 1.5 mCi
Whole body absorbed dose 2.4 MBq or 0.06 rad
Upper large intestine wall 27.7 MBq or 0.74 rad
The features of this pharmaceutical isotope material are:
the energy of the gamma radiation is high enough to penetrate tissue but still low enough
for efficient gamma camera detection, i.e.. around 50250 keV energy range.

radionuclidic impurities must usually be less than 1-2% since any longer half-life isotopes
contribute to increasing patient radiation dose and to blurring of images.
half lives are approximately 3 days, which is a convenient time period for the
manufacture, delivery and administration of the material to the patient.

the specific activity of the isotope must be high enough to produce sufficient events or
 counts in the required organ using a gamma camera imaging system.

the radiopharmaceutical must be manufactured according to the protocols of "Good

Manufacturing Practice".

the specification and manufacturing protocols must comply with approved drug licence
conditions issued by the appropriate Government bodies.

3. BOMBARDMENT AND ISOTOPE PRODUCTION PROCESS

For the purpose of this section the heart imaging agent thallium-201 will be used as a
'case study'; the overall production of this isotope will include the following steps:

Preparation of a target using as a starting material the enriched isotope thallium-203,

Bombardment by a 25-30 MeV cyclotron proton beam,

First chemical separation process to produce the intermediate or daughter isotope

lead-201,

Secondary chemical separation of the radioactive isotope thallium-201,

Pharmaceutical production of the sterile, radioactive, injectable thallous chloride.

Other important isotopes such as gallium-67, iodine-123 and indium-111 are produced by
very similar manufacturing processes. The preferred production reaction for producing
thallium-201 is shown below; in fact lead-201 is generated first.

decay
Thallium-203 (p,3n) Lead-201 Thallium-201

Naturally occurring thallium consists of two main isotopes - thallium-203 (natural

abundance 29.5%) and thallium-205 (natural abundance 70.5%); the initial raw material for
cyclotron targets is the enriched stable isotope - thallium-203. Enriched isotopes are produced
usually by large electromagnetic separators located within Government controlled
establishments, producing mainly uranium-235 for fuel and weapons applications. The original
'calutron devices' at Oak Ridge are now 40 years old and are based on the concept of a 180
magnetic spectrometer and use large dipole magnets. Radioisotope production consumes a few
kilograms of these materials annually worldwide and the enrichment requirements for the
different isotopes are given below in Table 5.

The excitation cross sections for proton bombardment of thallium-203 are given in Fig. 2,
indicating peak energies for the reactions (p,n), (p,2n) and (p,3n) etc. In addition, similar
production cross sections exist for protons bombarding thallium-205 indicating that even a trace
impurity level of thallium-205 within the target material thallium-203 will lead to radioactive
impurity species in the final product. The competing reactions are given below in Table 6.
Table 5
Enrichment of isotopes
 Isotope Other % naturally % radio- Electro- Approx. cost
naturally occurring pharmaceutical magnetic (M$/kg)
occurring specification separator
isotopes production
(hours per kg)
Tl-203 Tl-205 29.5% 98% 20,000 2.5
Zn-68 Zn-64, Zn- 18.6% 98% 65,000 4.5
66, Zn-67,
Zn-68, Zn-70
Cd-112 Cd-106, Cd- 24.1% 96% 55,000 3.5
108, Cd-110,
Cd-114, Cd-
116
Xe-124 Xe-126, Xe- 0.095% >20% Gaseous 0.1
128, Xe-130, diffusion per litre
Xe-132, Xe-
134, Xe-136

Fig. 2 Excitation cross sections for Thallium 203

From the graphs in Fig. 2 it is possible to determine:

The maximum bombardment energy for the proton beam; in this case it is determined by
the impurity radionuclide lead-200 created via the reaction (p,4n), resulting in a daughter
product Tl-200 which has a half life of 26 hours and a high energy gamma ray. This would
increase the patient dose considerably if the percentage impurity was allowed to be greater than
1%. Consequently this fixes the maximum proton energy for thallium production at 28.5 MeV.

The minimum bombardment energy or exit energy of the proton beam is similarly
constrained by the impurity radionuclide lead-202m created by the reaction (p,2n), the resultant
daughter product Tl-202 has a half life of 12 days and a -emission of 1.0 MeV; the minimum
energy at the exit of the enriched Tl-203 material is similarly fixed at 22 MeV.
Table 6
Competing reactions - thallium bombardment
Reaction Initial material Radioactive products Half life
 (p,4n) Tl-203 Pb-200 21.58 hours

Tl-200 26.1 hours
(p,3n) Tl-203 Pb-201 9.4 hours

(p,5n) Tl-205 Tl-201 73.1 hours
(p,2n) Tl-203 Pb-202m 3.62 hours

(p,4n) Tl-205 Tl-202 12.23 days
(p,n) Tl-203 Pb-203 51.9 hours

(p,3n) Tl-205 Tl-203 stable
(p,2n) Tl-205 Pb-204m 1.1 hour

Pb-204 17
>1.4x10 yrs
(p,n) Tl-205 Pb-205 7
1.5x10 yrs
The energy of the proton beam is reduced during passage through the target and energy is
lost by elastic collisions, atomic excitation and ionisation. Using the Bethe model for
relativistic velocities the expression for energy loss can be shown to be:
dE NZ
= 4.58 10 12 ln(1.82 10 30 v 2 / I ) ln(1 2 ) 2
dx v2
in units of MeV.mm-1 where:
N the number of atoms per unit volume of target material
Z the atomic number of the target material
v proton velocity
I effective ionisation potential of target material
(v/c)

The stopping power in thallium is shown in Fig. 3 and the computed range of protons in
thallium as a function of energy can be derived as in Fig. 4.

Fig. 3 Stopping power in thallium Fig. 4 Computed range of protons in thallium

as a function of energy
Another parameter used in isotope production is the yield rate Ao, which represents the
macro-production cross section for a particle beam with given incident and exit energies and
represents the instantaneous production rate of the isotope (i.e. assuming no radioactive decay).
Ao takes account of target density, molecular weight of the target material, enrichment or
abundance factor of the target isotope etc. and is expressed in Curies per mA.hr. A
comprehensive computer model and methodology has been developed by Forrest for any
production isotope and allows for energy loss and the reactions of all the isotopes involved. It
is assumed that a thin layer of the enriched isotope i.e. Tl-203, is deposited on a thicker cooled
target base (usually copper) and that the target assembly can be inclined at a given angle to the
direction of the proton beam. Table 7 shows a report print out for thallium, from which the

Table 7
Summary table of yield values (Ao) protons on thallium

design requirements for thallium bombardment can be derived. The model also predicts the
impurity levels for the other radionuclides and also the production level of Zn-65 (half life 265
days) produced by the reaction which creates a waste disposal problem due to its 1.12 MeV
gamma ray.
Zn-65 is produced by the reaction:
Cu-65 (p,n) Zn-65
This model assumes that the remainder of the energy is deposited within the target base; the
stopping power calculation can also be used to estimate the required thickness of this copper
base to ensure full proton absorption and no penetration into the cooling liquid which is usually
water.
For a given set of irradiation conditions the activity produced is:
At = AoIt I-et
t
where At is the activity in Curies at time t
I is the beam current in mA
t is the bombardment time in hours
is ln(2)/T
T is the half life of the isotope produced

Tpical Ao values is given in Table 8.

Table 8
Table of yield rates Ao
-1 -1
Isotope Ao (Ci mA hr ) Ein (MeV) Eout (MeV)

Tl-201 21.6 28.5 22.0

Ga-67 4.3 25.5 0.0
In-111 6.3 29.0 17.0
Co-57 0.032 22.0 0.0
The design requirements for thallium-201 bombardment can then be determined and are
given below in Table 9.
Table 9
Design requirement for thallium bombardment

Maximum energy 28.0 MeV Pb-200 production limitation

Minimum energy 22.0 MeV Pb-202 production limitation
dE/dx 11.9 -1 Derived from 28.0 MeV incident
MeV.mm beam energy and 22.0 MeV exit
energy

Effective thickness 0.566 mm Calculated from above

Angle of inclination 6 Mechanical design considerations
Thickness of electroplating 0.06 mm

In general, for solid targets the target materials are electroplated onto metal bases, usually
having good conduction characteristics, such as copper, silver or aluminium and proprietary
electroplating methods are often used ensuring that the starting raw materials comply with the
conditions laid down in the Drug Master File, i.e. the manufacturing protocols authorised
according to the drug licence. Criteria for electroplating include:

Plating to completion of the electrolyte or otherwise,

Electrical programming of the plating current cycle,
 Mechanical agitation of the target material,
The addition of additives or shiners to enhance electro-mechanical deposition,
Maintaining the minimum plating thickness consistent with the bombardment
requirements.

A typical electroplating station is shown in Fig. 5.

Fig. 5 A typical electroplating station

4. DESIGN CRITERIA FOR TARGETS

In commercial manufacturing of cyclotron isotopes, the key parameter for target design
is the beam current acceptance. Historically, production targets have been located either inside
cyclotrons as internal, glancing-angle targets on the external orbit of the correct energy or as
external targets bombarded at the end of beam lines. For industrial operation, the heat transfer
problem of this large energy density deposition has been addressed by either adjusting the
glancing angle in the range 4-15 for incidence on a static target or presenting circular
Table 10
Heating calculations for a stationary target
 BEAM PARAMETERS COPPER TO WATER FINAL TARGET SURFACE
HEAT TRANSFER PARAMETERS TEMPERATURE CALCULATION

a. 1/2 beam height(m) = 0.0094 Copper/water heat transfer The peak surface temperature
2 is calculated from the cooling water
b. 1/2 beam width(m) = 0.0009 coefficient (W/m K) = 60800 temperature plus three additional
temperature rises.
Beam current (A) = 150 cp. Specific heat of water
o o
at 20 C (J/kg K) = 4186 1. Water-copper ( C) = 44.8
Energy (MeV) = 29
nb. Viscosity of water at o
Q. power of beam (W) = 4350 o 2. Copper ( C) = 16.5
20 C (J/kg K) = 0.001
A. projected area of beam o
2 lb. Thermal conductivity of water 3. Plated target layer ( C)
on target (m ) = 0.0025 o = 6.7
at 20 C (W/m) = 0.602
q(m). Mean heat flux o
2 Pr. Prandtl number = 6.95 Total temperature rise ( C)
(W/m ) = 1710817 = 68.0
q. Peak heat flux d. Depth of water channel (m)
2
(W/m ) = 3934879 = 0.00375 Chilled cooling water temperatur
o
= 13.0 C
TARGET PLATING PARAMETERS w. Width of water channel (m)
= 0.025 Therefore, final peak target
l3. Thermal conductivity of surface temperature
plating material = 44 M. Water flow rate (kg/s) o
(W/m K) = 1.65 = 81.0 C
= 44
m. = M/(d*w) mass flux of cooling
X3. Plating thickness (m) 2
= 0.00010 water (kg.m s) = 17600

Re. Reynolds Number

= 2md/nb = 132000
COPPER BACKING PARAMETERS
nw. Viscosity of water at the temperature
l. Thermal conductivity of the Cu/water surface
of Cu (Wm/K) = 390 o
= 0.000548 at 50 C
x. Thickness of Cu (m) o
0.000283 at 100 C
= 0.002 o
0.000183 at 150 C
d. Approximate correction
factor to allow for heat
"spread out" in copper
= 0.44

geometry rotating targets at high speed to ensure optimum heat transfer. Current commercially
available cyclotrons are capable of producing external beams with good emittance control for
the current range 150 - 400 A per beam on target. For the purposes of this 'case study' it will
be assumed that a beam current of up to 200 A external beam will be incident on a static target
at a glancing angle. Typical beam emittance is of the order of 5 to 10 mm.mrad which
translates to available beam spot dimensions of the order of 3 - 20 mm in either dimension.
Early experimental measurement and heat conduction calculations published by Pinto (Table 10)
indicated that a temperature rise of 150C would occur with an incident proton beam of
30 MeV, 200 A on a target with no electroplated material but designed for optimum cooling.
He concluded that for production targets, temperature rises could be limited to about 150C
even with electroplated target material; this temperature is lower than the melting point of
thallium metal and would guarantee integrity of the targetry and retention of the expensive raw
material on the target surface. This calculation was substantially improved by Forrest to allow
for the heat transfer across the interface between the electroplated layer and the base material for
several different materials. The model also allowed for defining the optimum target inclination
angle for a given beam spot size and Table 11 indicates computed surface temperature increases
for different glancing angles based on the following assumptions:
Optimum target mechanical design to stimulate turbulent flow with water cooling,
Uniform electroplated material,
Table 11
 Target Heating calculations for different metals and beam incident angles

dimensions of approximately 6 mm high by 20 mm wide,

 Nominal beam energy 30 MeV,
A maximum possible proton beam current of 400 A.

The computed surface temperatures for different incidence angles are given for typical
initial raw materials used in radiopharmaceutical production with the requirement that the
surface temperature never exceeds the melting point of the material at the reduced vapour
pressure within the target enclosure. In general, optimisation of computed models and practical
experimentation result in the 'typical' target specification, shown in Table 12.
Table 12
Target specification

Beam current 400 A, 30 MeV

Beam size 20 mm x 6 mm
Target inclination 6
Maximum power density deposition 12 kW
Maximum temperature rise of Tl-203 128C

Due to the relatively high radioactivity activation levels of production targets, well
designed, remote controlled mechanical transport systems are necessary to ensure that target
integrity is maintained otherwise local contamination of radioactivity and loss of valuable
enriched isotopes could occur. Two approaches are now used:

Compact target transfer by pneumatically operated rabbit systems the system designed
by TRIUMF is shown in Fig. 6.

Railway transfer systems which remove the target assembly from the cyclotron into a
demounting station.

Fig 6 Target transfer showing rabbit (1) and target (2)

5. CYCLOTRON SPECIFICATION
Requirements of cyclotrons for commercial isotope production are:
Proton beams as indicated in Table 3; the large majority of useful isotopes can be
produced by proton reactions (p,xn).
High beam currents, since the isotope production rate is proportional to beam current.-
Variable energy is required to accommodate all the reactions in Table 3.
Extracted beams are preferred to allow optimisation of target design and ease of handling
High reliability operation is essential for the production of these short-lived isotopes
which have to be delivered several times a week.
Low radiation activation levels are essential at the cyclotron and around the target system
to allow for regular maintenance, equipment upgrades etc.
Low operating costs are desirable for a viable, competitive commercial operation.
The present day 'third generation' compact cyclotrons have been designed specifically for
this type of commercial operation and satisfy the above requirements. Two excellent models
are available as off-the-shelf packages, viz
IBA, Cyclone-30 designed by Yves Jongen, Louvain La Neuve, Belgium and
EBCO, TRS-30 designed by TRIUMF for EBCO Industries, Vancouver, Canada.
It is also worth mentioning Scanditronix's MC-32 NE which is a negative ion cyclotron but
does have a lower output level. For the purposes of this paper, the Cyclone-30 is assumed.
The performance specification of the Cyclone-30 is shown in Table 13 and a drawing of the
Cyclone-30 cyclotron is shown in Fig. 7.
Table 13
IBA Cyclone-30 specification

Beam current 350-500 A

Energy range 15-30 MeV
Extraction beam 2
Beam emittance 5-10 mm mrad
Power consumption 150 kW
Ion source External
-7
Base vacuum 3 x 10 torr

Clearly the above requirements for commercial isotope production are well met by the
general system specifications of the Cyclone-30, viz

Proton or deuteron beams can be produced efficiently.

High extracted beam currents of 350 to 500 A which are significantly higher than
previous 'second generation' cyclotrons, cf. 60 to 100 A.
The extraction foil mechanism together with a 'combination' magnet provide an adequate
range of extraction energies.
High reliability operation is achieved by simple yet elegant magnet and RF design, well-
engineered extraction equipment, high reliability sub systems, comprehensive diagnostic
facilities and stable high beam extraction performance.
Low radiation levels are maintained by efficient (negative ion) extraction, selection of low
activation material for construction of the cyclotron vacuum chamber etc., and low neutral
beam activation by high vacuum operation by using an external ion source.
Low operating costs are achieved by very efficient power consumption of the main
 magnet and RF systems, a high reliability operation as well as good maintainability and
access.

Fig. 7 Cyclone-30 cyclotron

The analysis in Section 4 indicated the following target demands for the extracted beam:

stable, high current extraction

low beam loss at extraction
accurate reproducible extraction energy with low dispersion
low emittance in both planes, preferably with balanced divergence.

Again, the Cyclone-30 performance meets these requirements although it is essential to ensure
that the beam size at the target location can be controlled within the "footprint" of the material on
the target thereby minimising the usage of the expensive enriched isotope material. Figure 8
shows a typical TRANSPORT generated computer calculation of beam line optimisation for a
possible Cyclone-30 beam to give a minimum 'image' size of 20 mm x 4 mm onto an inclined
target footprint of 20 mm x 40 mm, and using the minimum length of beam line yet allowing
sufficient space for shielding between cyclotron vault and target vault. For this particular
example the beam optics show a very large vertical beam dimension. Beam line design for
commercial operation has to satisfy two conflicting demands:

a small beam size at the target to minimise the usage of expensive enriched isotopes
a short beam line length to limit the construction costs of the facility.

The final design is a compromise ensuring that beam loss in the beam lines does not lead
to excessive radiation and activation of equipment.
 Fig. 8 Cyclone-30 beam optics calculated with the TRANSPORT program

6. RADIOCHEMISTRY

In cyclotron systems, the amount of isotope generated in the targets is determined by the
sum effect of the physical processes described earlier. Extraction of the radioisotope from the
targets is achieved by chemical or radiochemical methods. For example, in order to achieve the
correct radiochemical formulation for thallium-201, the following processes are required:

First chemical separation: Acid extraction of the primary product lead-201 from the initial
thallium-203 and the target base material, e.g. copper.
Second chemical separation: Ion exchange to remove the secondary product thallium-201
from the primary product lead-201.
Chemical recovery of the original thallium-203 from the aliquot of lead and thallium
isotopes as well as the copper impurities.
Possible removal of the radioactive zinc-65 from the copper material in the target base
since this radionuclide constitutes a severe radioactive waste disposal issue.

The graph of Fig. 9 shows the change in radioactivity during the manufacture of Tl-201
during bombardment, the first and second chemical separations, pharmaceutical production, the
decay of Tl-201 to the reference date of the product and to the expiry date of the material.

The majority of other isotopes using solid targets, e.g. Ga-67, In-111, etc. are produced
by similar techniques although some processes may employ organic solvent extraction
chemistry. The main exception is I-123 which is usually produced by dedicated bombardment
and processing equipment such as the gas system using Xe-124 as developed at KF Karlsruhe.

These chemical processes are performed within a facility dedicated for the production of
radiochemicals and containing heavily shielded production plant or cells with shielding
equivalent to 10 or 20 cm lead. Each plant is usually equipped with:
a set of manipulators,
leaded-glass window,
in-plant chemical equipment,
front-panel controls for electrics and pneumatics,
transfer ports for product flow, waste materials,
air flow into filtered air handling unit.
The radiochemical produced has the correct chemical ingredients but not necessarily the correct
concentration, and is not sterile enough for drug purposes.
 Fig. 9 Activity of thallium-201 during production, delivery and use

7. RADIOPHARMACEUTICAL PRODUCTION

As the usage of cyclotron isotopes increased, pharmaceutical licensing requirements for

manufacturing became more demanding. Today, cyclotron isotopes used in nuclear medicine
are not just radiochemicals with medical applications but genuine pharmaceutical materials
incorporating a radioisotope component. These isotope products are usually supplied in a
shielded vial container or syringe ready for injection.

The manufacture of radiopharmaceuticals include the following steps:

container preparation and cleaning, i.e. for vials or syringes

transfer of the radiochemical material to a pharmaceutical grade 'clean room',
quality control of parameters such as chemical purity, radionuclidic purity, specific
activity,
sterile dispensing of the radiochemical into its final container,
terminal sterilisation of the final product,
batch release of the product by the Qualified Person,
labelling, packing, distribution, etc.

The detailed steps in this manufacturing procedure have to comply with the principles of
"Good Manufacturing Practice" (GMP). All steps are defined in extensive documented
procedures within a Drug Master File. This is the legal document binding the manufacturers'
operations to the conditions of the pharmaceutical licence which is awarded by Government
Regulatory bodies such as CPMP (Europe), FDA (USA). In this way Regulatory Authorities
strive to guarantee the safety of the product and the reproducibility of the manufacturing
operation.

In order to comply with the regulatory requirements for these small volume 'parenterals',
special attention must be paid to ensure that the products are:
pyrogen free,
sterile,
free of particulate material etc.
To achieve this careful control is required of:
the manufacturing environment,
the manufacturing processes,
the product containers.
In addition there is also a requirement that manufacturing operations are performed in a manner
that minimises the risk of:
 cross contamination of one product by another,
incorrect labelling of product containers etc.
A system of Quality Assurance must be in place to ensure that all processes are
documented, validated and performed repeatably in accordance with written instructions. All
batches of materials used in processes must be tested to ensure conformance with specification;
traceability of materials through to the finished product must be maintained and batches of final
product for sale must be tested/inspected to demonstrate conformance with specification. All
production facilities must be regularly monitored to ensure continuing performance to
specification, e.g.
water quality in purified supplies
air quality in clean rooms
performance of sterilisers
integrity of air filters
calibration of measuring instruments
During the development phase of a new product, all processes must be systematically
validated to show that the process will produce the specified output for all inputs at the upper
and lower limits of their specification ranges. This will include a demonstration that the product
remains stable throughout its shelf life for specified storage conditions. All experimental and
test data must be retained and is liable to inspection by the Regulatory Authority. The finished
product must be stored and shipped in a manner that ensures that product deterioration does not
occur. Finally, the selection and training of staff is vital to the maintenance of GMP.

8. DESIGN OF THE FACILITY

For the purpose of this discussion, the construction of a complete radiopharmaceutical

grade facility is assumed on a green field site.
According to the principles described in Section 5, a cyclotron and associated beam lines
are selected and purchased from a specialist vendor. Design features of the facility for
accommodating this accelerator hardware should include:
construction of a solid base or plinth,
construction of a complete radiological shield or vault for the cyclotron, beam lines and
target system,
subterranean access to the cyclotron for maintenance,
adequate headroom for access to the ion source and for lifting the upper pole of the
magnet for maintenance.
The design of a conventional radiological shield or vault should allow for:
attenuation of the neutron flux (produced at the isotope target) to a level of
< 5 Sv.hr at the external surface of the shield
-1

selection of appropriate material (concrete) to ensure correct shielding efficiency

without voids and without high Z materials which result in high activation levels,
correct design to avoid neutron (and gamma) short paths in access ducts and channels
correct design of the movable access doors,
fail safe interlock system for access to each vault
adequate shielding and protection around the beamline penetrations into target vaults.
Target station design should allow for:
remote mechanical control of the target load/unload function,
 diagnostic and monitoring of target performance parameters,
protection against loose material contamination from the target assembler,
installation of pneumatic controlled rabbit systems (or equivalent) with shielding to
transfer targets to their respective handling stations,
mechanical back-up for systems 'disaster' recovery of radioactive targets,
material specification for radiation hard materials, low-Z materials and wherever
possible, the absence of semiconductor control devices.

Support services for the accelerator/cyclotron system are:

correctly fused, interference-free, electrical power supplies including uninterruptible

power supplies for key equipment, i.e. computers, safety systems,
high pressure, de-ionized water supply for cooling of the targets and the cyclotron
system,
air handling of cool, filtered air to avoid humidity and the accumulation of gaseous
radionuclides such as argon 41 with appropriate ? cycles,

compressed air for servicing much of the mechanical (pneumatic) controls,

mechanical engineering shop capable of machining radioactive components.

A validated waste disposal system with facility monitoring for unacceptable levels of
background radiation and contamination is necessary which includes:

solid waste disposal and compaction of target and ancillary cyclotron components
removal of effluent generated in the radiochemical and pharmaceutical areas

An overall facility design is shown in Fig. 10 with

segregated areas for cyclotron operations,

high level radioactive radiochemical processing area,
clean room environment for radiopharmaceutical finishing operations,
ground floor laboratories for quality control etc.,
appropriate personnel change rooms and environmental monitoring support services
upper storey accommodation for mechanical and electrical plant, non-radioactive
workshops, offices, etc.
warehouse, packing and product despatch areas for the final product.

9. CONCLUSION

The design principles of a cyclotron facility for the commercial production of

radiopharmaceuticals has been described, explaining the underlying physical principles of
design, the requirements of radiochemical production and final pharmaceutical finishing. The
design of a "green field" facility with all the support functions at today's prices would cost
approximately 8M sterling. However, in order to generate an adequate return on this financial
investment, sales of cyclotron isotope products of the order of 4M to 5M per year are
necessary. This can only be achieved by:

The large manufacturers who have existing approved drug licences and large sales
organisations capable of distributing the products too numerous hospitals and medical
centres, or
Government sponsored laboratories who usually have their capital expenditure covered
by central funds, and are allowed to operate on a cost recovery basis by producing small
volume niche products or radiochemical materials which are used by the larger
commercial manufacturers.
 Although cyclotron isotopes production is essentially a commercial operation, the total
manufacturing process encompasses fundamental accelerator physics, radiochemistry and
industrial pharmaceutical technology. This presents a very rewarding occupation to the people
working in the industry producing very valuable materials which are used by clinicians to
diagnose illness and to alleviate the suffering of many millions of people every year.

Fig. 10 Conceptual cyclotron facility

BIBLIOGRAPHY

A A Fleischer, G O Hendry, C G Smith, J L Tom and D K Wells, Design and performance of a

compact multi-particle cyclotron; Proc. Fifth Int. Cyclotron Conf., Oxford, Sept 1969
(Butterworths) p.573.
G Seaborg, Remarks at the Dedication of the Biomedical Cyclotron at UCLA, June 30 1971,
USAEC S-14-71.

J P Mamacos, P J Kinny and J S Laughlen, A Compact Cyclotron Installation for Biomedical

Uses, J.Nucl.Med.8 (1969) 330.

Y Jongen, J L Bol, A Chevalier, M Lacroix and G Ryckewaert, Construction of the Louvain-

la-Neuve 30 MeV 500 A H - Cyclotron, Proc. 11th Conf. on Cyclotrons & Applications,
Tokyo, 1987, p 275.
M Lagunas Solar, Cyclotron production of no carrier added medical radionuclides, IEE Trans.
on Nucl.Science, Vol. NS-30, No.2, p.1805 (1983).

Qaim, Weinrich, Ollig, Int. J. Appl. Rad. Isotopes, 30, 85 (1979).

T J Ruth, The Production of Radionuclides for the Biosciences, Proc. 13th Int. Conf.
Cyclotrons & Applications, Vancouver (World Scientific, 1992) p.206.

J Lamb, Commercial production of radioisotopes for nuclear medicine 1970-1980, IEEE Trans.
Nucl. Science (April 1981), Vol NS-28 No.2, p.1916.

K Lebourtz, S Greene and T Fairchild, J. Nucl. Med., 16, 151 (1975).

J Tracy, Isotope separation program - present and future, Nucl. Inst. & Methods in Physics
Research, 282 (1989) p.261.

G Pinto, M Straatman, D Schyler, S Curris and G O Hendry, Target systems for radioisotope
production, IEEE Trans. Nucl. Science (April 1983), Vol NS-30 No.2, p.1797.

J J Burgerjon, Z Gelbart, G O Hendry, J C Lofvendahl, L McIlwraith and G A Pinto, A high

power target system for radioisotope production, Proc. 11th Int. Conf. on Cyclotrons &
Applications, Tokyo, 1986.

T R F Forrest, A method for calculation of cyclotron product yields, Amersham International

internal report, November 1987, PTDS/R8.

T R F Forrest, Demonstration of a practical stationary target system, Amersham International

internal report, November 1987, PTDS/R10.

IBA, Ion Beam Applications SA, Chemin du Cyclotron 2, B-1348 Louvain-la-Neuve,

Belgium.

EBCO Industries Inc., 7851 Aldebridge Way, Richmond, BC, Canada V6X2A4.

Scanditronix AB, Jusbyborg, S-75229, Uppsala, Sweden.

European Communities Commission, "The Rules Governing Medicinal Products in the

European Community", Vol. IV (1989), ISBN 92-825-9572-2.
COMPUTATIONAL TOOLS FOR ACCELERATOR DESIGN

L. Rinolfi
CERN, Geneva, Switzerland

Abstract
Based on the method generally followed for an accelerator design, this
contribution describes some of the computational tools which can be
used. A brief review is given for simulation codes related to the design
of accelerator components and those related to the beam dynamics
with emphasis on electron linacs.

1. INTRODUCTION
The title of this contribution is rather ambiguous. Indeed it could imply that all
computational tools available for accelerator design are described. Although the purpose is to
describe the codes running on various computers, and probably not the machines themselves, it
would be hard to cover all the codes currently available, even if we restrict ourselves to the
field of linacs and cyclotrons.
Therefore we will focus on the method which can be applied in the design of any
accelerator and after that on the codes which have been found to be particularly useful in the
analysis and design of accelerators. If some readers do not find mention of their favorite codes,
this is possibly due to limitations of knowledge or of room to describe them.
A deliberate choice is also made to avoid detailed description of the physics of the codes.
In the references and bibliography given at the end, the reader will find more information about
existing and known codes and the related physics.

2. METHOD AND TOOLS

During the design phase of a whole accelerator project or a small test facility, or even
improvements to an existing accelerator, many discussions and calculations are involved. Once
the guide lines (physics) and the goals (beam performance) of the project are defined, the
exercise starts. The different steps, with many interchanges between them, are:

i) submit the project to discussion by presenting the basic ideas

ii) check the orders of magnitude for the main parameters and see if the beam
performance is achievable

iii) elaborate a first model, generally based on analytical calculations

iv) realise the next model either from existing simulation codes or from a new one to
be written in order to model the real machine as well as possible

v) improve the model by including new factors and new effects if necessary.
These general rules have to be taken as advice and will avoid disappointment to the
accelerator designer tempted to jump directly to calculations with a 3-D code!
 To illustrate the different phases, the following tool list is proposed:
i) pencil + paper
ii) pocket calculator
iii) desktop computer
iv) large computer.

Emphasis will be made on some existing simulation codes for linacs. Among the pro-
posals for the future linear accelerators, there is the CLIC (CERN Linear Collider) which is
foreseen beyond the year 2000. A small linear accelerator CTF (CLIC Test Facility) has been
installed at CERN and has been running for several years. It will be used, as an example, to
illustrate this contribution.

3. DESIGN ACCORDING TO THE MACHINE

For a given type of particle (leptons, ions,...), two main topics should be considered:

1) the accelerator structure

2) the beam dynamics.

Three main families of accelerators and a particular case are presented.

3.1 Linear accelerators
A linear accelerator is a machine in which particles are accelerated on a linear path, a
charged particle q crossing an electric field E being submitted to a force F = q E. An
electrostatic field can be used to accelerate such particles. The dc accelerators (type Van de
Graaff) are still used for low energy ion acceleration. Due to the limitations to high continuous
voltage, the acceleration in linear accelerators is produced by time-varying electric fields. The
two methods used are:
i) the RF (Radio Frequency) linac
ii) the induction linac.

The RF linacs are composed of resonant cavities (standing-wave structures) or

waveguides (travelling-wave structures) with sinusoidally varying electromagnetic fields in the
range of some MHz up to 30 GHz [1,2,3]. Induction linacs [4] are composed of an array of
modules where a pulsed current passing through a toroid produces a rapidly changing
magnetic flux and according to Faraday's law a voltage is induced across an accelerating gap.
The beam dynamics concern the longitudinal and the transverse planes. They will normally
differ according to the type of particles. Electrons are light particles and they are quickly
relativistic. In a path length less than 1 m, they can approach the velocity of light and they
travel inside the linac at this essentially constant velocity. The consequence in the longitudinal
phase plane is that the classical stability diagram of non-relativistic particles opens up and no
phase oscillations can occur. As the beam charge increases some additional beam-dynamics
effects begin to degrade the beam performance (space charge, beam loading, wake fields,...).
Solutions for reducing or cancelling these effects can significantly influence the parameter
choice for the accelerator design.
3.2 Circular accelerators
If linacs can be considered as RF-oriented machines, circular ring accelerators are rather
magnet-oriented machines. The accelerator structure consists basically of a pattern of
successive dipoles, quadrupoles and drift spaces. For the acceleration process or for particular
RF requirements, one or more RF cavities are installed. Again, the beam dynamics handle both
transverse and longitudinal planes. For the latter the behaviour is similar to the linac for non-
relativistic particles. For leptons, synchrotron radiation cannot be neglected and should be
taken into account from the beginning even for the first model. In order to get a correct closed
orbit, limitations on periodicity are preponderant in designing the lattice. Since a particle
passes the same place several times, there is a repetitive exposure to the errors and
perturbations existing in the machine. This can lead to destructive resonances.
An uncommon use for a circular machine is to decelerate particles. An example is given
by the CERN Proton Synchrotron (PS) which receives antiprotons at 3.5 GeV/c and after
deceleration down to 0.6 GeV/c ejects the particles towards LEAR (Low Energy Antiproton
Ring). A storage ring represents also a special design and application.
Some particular beam-dynamics effects related to circular machines (closed-orbit
distortion, intra-beam scattering, instabilities due to resonances, transition energy,...) and the
computational tools related to them, will not be described in this contribution.
3.3 Cyclotrons
As in the other accelerators the two main topics, the accelerator structure and the beam
dynamics should be considered. However for such machines, the magnet system and the RF
system are extremely interdependent. The cyclotron magnet design and the RF design require a
complete view of the cyclotron [5]. The two main issues are:
i) the isochronous aspect where the particles and the accelerating RF remain in phase
and where the magnetic field follows the relativistic mass increase during the
acceleration process.

ii) the resonances: during the acceleration it is unavoidable to cross resonances, as in

synchrotron machines. The time that the charged particles remain on the resonance
and the amplitude of the driving term determine the beam behaviour.
The analytical equations are given in Ref. [5, 6] and can be treated with a pocket
calculator or desktop computer. The available simulation codes will be mentioned in section 9.
3.4 Transfer lines
Although they are not accelerators, great care still needs to be taken in order to preserve
the beam quality. Here the transfer structure is composed of drift spaces, dipoles and
quadrupoles. For linacs the structures are open at each end and therefore they do not require
specific techniques to inject and extract beams. Only at the injection should the beam be
bunched. For circular machines, the beam characteristics at the end of the transfer line will
determine the injection process, while the beam characteristics at the end of the acceleration
will determine the optics of the transfer line. Devices like septa and kickers should be included
in the design. The goal is to get 100% transmission in the transfer line and no degradation of
the beam performance at the entrance of the next machine or in the detector. Many
computational tools used for beam transport in transfer lines are those used for linear or
circular accelerators. Therefore, no specific development is required and the names of the
programs dealing with such issues are mentioned later on in the text.
4. AN EXAMPLE: THE CLIC TEST FACILITY (CTF)
In 1985, a first proposal [7] for the CLIC project was made. For this new accelerator
project, where an electron beam collides with a positron beam, the choice for the RF frequency
is 30 GHz. At this frequency, there is no device available which is able to deliver enough
power to accelerate the beams. For the collision energy (500 GeV), an electric field of
80 MV/m is required in order to keep the length of the linear accelerators reasonable (in the
range of 4 km).
Therefore, the guide lines are: " invent a device (type klystron) which is able to deliver
several tens of MW at 30 GHz and prove that is possible to accelerate an electron beam with
such a device". The goals are:
i) production of an accelerating gradient of 80 MV/m
ii) prove that the RF structure is able to sustain such a high electric field.
We will see how the proposed steps for an accelerator design, with the corresponding
computational tools, can be applied.

5. BASIC IDEAS USING A PENCIL AND PAPER

When it was decided to produce a 30 GHz RF power source from an electron beam, a rough
investigation showed that a short pulse (10 ps) and high charge (20 nC) were required. After
having analyzed almost all classical machines, it became clear that both parameters were not
achievable simultaneously. Therefore it was necessary to find a new source producing such
electron pulses. The choice was made in the direction of photo-electrons: a photo-cathode
illuminated by a laser pulse would be able to produce such a beam. Figure 1 shows the first
layout of the CTF.

LAS: LIL Accelerating Structure: it is a 4.5 m long travelling-wave structure (3 GHz)

TRS: TRansfer Structure: it is a CLIC decelerating structure (30 GHz)
CAS: CLIC Accelerating Structure (30 GHz)

Fig. 1 First layout of CTF

6. ANALYTICAL CALCULATIONS USING A POCKET CALCULATOR
A pocket calculator is a common computational tool but very useful in the accelerator
design. Based on the CTF example, an exercise of rough estimations, for the order of
magnitude, is proposed below.
The electric field generated by a single bunch crossing a RF structure is given by the
following formula:
r0'
E1 = 0
q F ( ) (1)
2
where 0 is the RF angular frequency, r0' the normalised shunt impedance, q the charge and
F() the form factor of the bunch.
The peak power at the exit coupler of the structure is given by:
r0'
P10 = 0
v g q 2 F 2 ( ) (2)
4

where vg is the group velocity.

Now (see Fig. 1) if the exit of this first structure (TRS) is connected to a second
structure (CAS), then the accelerating field, E2, in the second structure, is given by:
0 r0' T
E2= q F() exp ( 0 f ) (3)
2 2Q
where is the transfer efficiency between the two structures, Tf is the filling time and Q is the
quality factor. The input data necessary for the calculations are the following:

0 = 2 30 GHz
r0' = R' Q = 26 . 2 k m
vg c = 0 . 082
Q = 4 224
l = 0 . 287 m
Tf = l v g
= 0 . 92

By virtue of (3):
E 2 = 1.755 x 1015 q F().

The goal being E2 = 80 MV/m, one finds:

q F() 46 nC.

Table 1 gives the results of the rough estimations if the form factor is equal to 1.
 Table 1
Electric fields and power for CTF

Parameters Values Unit

Charge q 46 nC

Power P10 63 MW

Peak decelerating field E1 on axis 113 MV/m

From these rough estimations, one can imagine that a single bunch of 46 nC with a form
factor close to 1 (i.e. bunch length [FWHH] of 5 ps) will not be easy to produce and transport.
Also a peak electric field of more than 100 MV/m on axis is not a current value for an RF
structure and will require a careful design. A pocket calculator is sufficient to estimate the
main parameters at this step of the accelerator design.

7. COMPUTATIONAL FACILITIES FROM A DESKTOP COMPUTER

7.1 First model of accelerator design from analytical calculations

It is often possible to give analytical expressions to calculate the beam dynamics in a
given accelerator as a first model. For example the paraxial equation is used as a first
approximation in many designs. References [1,2,3] provide basic equations for linacs and
References [5,6] for cyclotrons.
Coming back to the CTF, it was found after experiments that the space-charge effects
combined with large dispersion of the first optics had an effect too significant to reach the
expected beam performance. Therefore it was decided to simplify the optics for the electron
beam (i.e. no bends between the gun and the subsequent acceleration) and shorten the path
length of the beam at low energy. But as a consequence the optics for the laser became more
complicated to correctly illuminate the photo-cathode. A new RF structure (Booster) was
added to decrease the space charge-effects at low energy. Figure 2 shows the layout of the
new design.

4 C EL L
B OO STE R

Fig. 2 Layout of the present CTF

 An analytical model was developed on a desktop computer for this new layout. Figure 3
gives the beam envelope between the photo-cathode and the RF transfer structure (TRS) at
30 GHz.

Fig. 3 Horizontal beam envelopes

It is worthwhile to mention another problem for this facility. Although the different steps
were respected in the design of such a linear accelerator, it was found experimentally that it is
still not straightforward to reach the goals. Several effects were not taken into account in the
various simulation codes (high current density on the photo-cathode, beam loading, wake
fields, misalignments, ...). Therefore, a new idea was proposed, to produce long pulses at the
source with high charges, accelerate them, and compress afterwards in a bunch compressor.
After some studies of existing devices (three dipoles, magnet, wiggler) and after rough
calculations, the conclusion for the best solution seems to be the magnetic bunch compressor
with three dipoles. An analytical model was developed. Figure 4 gives the trajectory of the
reference particle inside the bunch compressor.

Fig. 4 Central trajectory for an electron at 10 MeV/c in the bunch compressor

 The main characteristics are derived from a matrix product of three dipoles and two drift
spaces. The beam parameters at the output of the bunch compressor are given as a function of
the beam parameters at the input by the following expression:

x 1 4 tan + 2 d / cos2 0 0 x
x' x'
0 1 0 0
l = 0 0 1 4 (tan ) + 2 d tan
2 l
p p
p 0 0 0 1 p

output input

The final equation which gives the compression factor is:

p
l = (4 (tan ) + 2 d tan 2 )
p
where l is the path length difference from the reference particle at the exit of the bunch
compressor,
p
is the relative energy spread,
p
is the curvature radius of the dipoles,
is the deflection angle of the dipole,
d is the drift space between each dipole.

For this sub-system, after having developed the analytical model, one can use a tracking
program to optimise it (see 8.1.3).

7.2 Exchange of software sources and data

One of the advantages of working on a desktop computer is the possibility to develop
ones own software. It is also helpful to receive or to send data files without having to
recreate them. An input file which already worked correctly for a given accelerator (linac,
cyclotron) is very beneficial if one only has to change some input data in order to adapt it to
another specific design. The possibility to receive files containing a magnetic or electric field
map from the person who made the measurements is a big advantage. The accelerator design
will be improved if the measured fields are used rather than an ideal or a simplified distribution.
Last but not least the program written on a desktop computer will always be available in
the form preferred by the user and will not be modified by the Computer Central Facilities.
7.3 Graphic plots and schematic drawings
Although millimetric paper is still useful, the possibility to plot various graphics from a
desktop computer represents an enormous advantage. It is almost impossible to mention all
existing software with facilities to plot graphics (EXCEL, Cricket graph, Canvas,...). A tool
like MathCad allows interactive calculations from mathematical formulae and enables
simultaneous plotting of any corresponding graphics related to the formulae.
7.4 Desktop computer versus large computer
For the purpose of this contribution, both computers have been mentioned separately.
However, with the increase of the computational power of the desktop computers, the limits
between them tend to nullify each other. Several computational tools can be run with almost
the same facilities in both types of computers, at least those concerned by the accelerator
design.

8. SIMULATION CODES FROM LARGE COMPUTERS

One can split into two main families the sophisticated simulation codes running on large
computers; those related to the beam dynamics and those related to the design of particle
accelerator devices. Often they contain more than 15000 lines of standard FORTRAN-77. The
minimum time of execution is an important feature for the big codes.
8.1 Beam dynamics simulation codes
8.1.1 Example of an electron linac
An electron linac can be modelled with the three following codes: EGUN for the
thermionic gun, PARMELA for the bunching system and TRANSPORT for the accelerating
sections.
EGUN [8] computes trajectories of ions and electrons in electrostatic and magnetostatic
fields. It includes space-charge effects and self magnetic fields. It assumes either a rectangular
or a cylindrical symmetry. Magnetic fields may be used from POISSON output or from an
externally calculated array of the axial fields. New versions of EGUN have been implemented
and are also available for desktop computers. EGUN is distributed by W.B. Herrmannfeldt
(SLAC) for a fee. Figure 5 gives an example of EGUN output for electron trajectories coming
from a thermionic cathode capable of delivering up to 15 A with pulses of 30 ns. An external
magnetic field of 0.2 T is applied at the exit of the anode.

Fig. 5 Beam trajectories from a thermionic cathode calculated using EGUN

 PARMELA (Phase And Radial Motion in Electron Linear Accelerator) is a multiparticle
tracking program with space-charge forces. It computes the particle distribution in six-
dimensional phase space at the exit of each RF cell for any bunching system, the independent
variable being time. The code generates several types of initial electron distributions but the
contour of the chosen distribution should be as close as possible to the beam dimensions
derived from EGUN. The electric and magnetic fields in the RF structure can be provided by
Fourier coefficients for ideal sine waves or from a table calculated by a code such as
SUPERFISH. External magnetic fields can be superimposed on any element (RF structure or
magnet) in the model. Although the main uses of PARMELA are for bunching systems, such
codes can be used to simulate the beam dynamics in a whole linac composed of RF structures
but also of magnets (dipoles, quadrupoles, solenoids, ....). From the output data it is possible
to plot the transverse and the longitudinal phase spaces at any place of the bunching system. In
addition, the beam envelopes and the energy gain can be plotted all along the line. PARMELA
is distributed by Lloyd Young (Los Alamos) free of charge. This code has proven to be a user-
friendly tool and rather powerful in predicting the real behaviour of the beam. However it is
poorly documented and some laboratories have written specific user guides [9]. Figure 6 gives
the transverse phase space at the output of the bunching system. From such a distribution, it
will be possible to fit an ellipse from which parameters can be used as input data for
TRANSPORT. These ellipses provide a good estimation of the transverse beam emittances.

Fig. 6 Transverse phase space at the bunching system output calculated with PARMELA

TRANSPORT [10] computes the transfer matrix of a beam line (first order up to the
third order). Elements of the line could be dipoles, quadrupoles, sextupoles, octupoles,
solenoids and accelerating cavities. Starting from an ellipse in the transverse planes, one can
perform the calculations all along the line for a given energy. As soon as particles are ultra-
relativistic one can forget the space-charge effects. This is the case, in general, at the bunching
system exit. However if the charge is high in the accelerating structures, another phenomenon
appears: beam loading. By simulation of the extreme energies, TRANSPORT can simulate the
beam loading effects in the electron linacs. From the output data one can plot the beam
envelopes, the energy gain, the beta functions, the dispersion function. Figure 7 shows an
example of half-beam envelopes. The initial conditions are taken from PARMELA output and
an ellipse of emittance is defined for 90% and 50% of particles and for a given distribution.
Then the optimisation of the optics is made in such a way to obtain 100% transmission.
TRANSPORT includes fitting conditions which, for example, allow either to obtain a given
beam radius (in the RF structure) or a minimum spot size (on the positron target). It is also
possible to obtain an ellipse of emittance for the following transfer line or the following
accelerator.

Fig. 7 Beam envelopes along the linac calculated with TRANSPORT

8.1.2 Example of a cyclotron

As already mentioned for an electron linac, the design starts with a crude and simple
model. The design process will then be carried on by iterative steps. The big computers are
necessary to perform detailed and time-consuming calculations for a final optimization.
Although the cyclotron is a circular machine, the beam dynamics differs from that on
synchrotrons because the particle trajectory is not a closed orbit. It is a spiral starting from the
center of the magnet and ending up outside the magnetic field. The design of the magnet is the
most important point. After the simple and analytical calculations, detailed magnet
calculations, using a finite elements method, are performed. First, a 2-D code solves x - y
problems where the third dimension is long enough to be seen as infinite. Axisymetric 3-D
problems can be solved by a 2-D code.
Secondly, a 3-D code is used to be as close as possible to the real device.
Finally magnetic maps, from 2-D and 3-D codes are elaborated in the cyclotron median
plane. Then, they are used as input data to specific programs which calculate the particle
trajectories. The main output parameters are the frequency error, the integrated phase shift and
the betatron frequencies.
 8.1.3 Example of a sub-system
When it is necessary to handle high charges where the space-charge effects are not
negligible, a sophisticated code like PARMELA becomes indispensable. The example of the
longitudinal phase space for the bunch compressor in the CTF is given below.

Fig. 8 Longitudinal phase space at the buncher input for low charge

Figure 8 gives the distribution of 100 particles before the bunch compressor (Fig. 4).
o
One can derive the phase extension (12 ) and the energy spread ( 1 %) of the bunch. Figure 9
o
shows the same distribution after the bunch compressor. The bunch length becomes (2 ) and
the energy spread remains the same as predicted by the analytical calculations. Both figures are
plotted from PARMELA output for a very low charge. With high charge, the lower energy
particles in the head of the bunch are accelerated by the space-charge forces and therefore gain
energy while the higher energy particles in the tail of the bunch are decelerated and lose
energy. The global result is a decrease in energy spread and can be evaluated only with a
sophisticated code.

Fig. 9 Longitudinal phase space at the buncher output for low charge
8.2 Component design simulation codes
In all accelerators, there are several types of components and devices which perform
various functions: acceleration, bunching, focussing, injection, extraction, monitoring. As for
the beam dynamics one can divide the components into two main families; one deals with the
longitudinal plane and the other with the transverse planes. To illustrate such simulation codes,
one example of each family will be considered.
8.2.1 RF cavity design: SUPERFISH code [11]
This simulation code evaluates the eigenfrequencies and the fields for arbitrarily-shaped
two-dimensional waveguides in Cartesian coordinates, and three-dimensional axially-
symmetric RF cavities in cylindrical coordinates. SUPERFISH calculates time varying fields
and spatial RF fields. It can also calculate quantities which are relevant for the design of drift-
tube linacs (transit-time factor, power losses, ...). It contains subroutines to generate the mesh,
to define the lattice, to solve the equations and to plot the fields. Figure 10 shows the electric
field calculated by SUPERFISH for the RF gun of the CTF. It is a standing-wave structure
with cylindrical symmetry. The density of the field lines can be changed according to the needs.

Fig. 10 Electric field in the RF gun from SUPERFISH calculations

From the output of SUPERFISH, the following parameters are deduced: the quality
factor Q, the shunt impedance Rs, the power P, corresponding to the requested electric field
Ez, and the stored energy. Also the ratio of the peak electric field to the accelerating electric
field is given. For example, the RF gun shown in the Figure 10, has a shunt impedance of
1.6 M. For the nominal peak electric field on the photo-cathode of 100 MV/m, the requested
RF power is 5.5 MW. From this data, one can calculate the necessary power that a klystron
should deliver according to the loss along the RF line.
 8.2.2 Magnet design: POISSON code [12]
This simulation code is a finite-element triangular-mesh which performs two-dimensional
calculations for electrostatic, magnetostatic and temperature distribution problems. The code
solves Poissons or Laplaces equations in 2-D regions and handles problems with finite
permeability iron but not with permanent magnets. Several laboratories have made
improvements (LANL, BNL, LBL, Texas Accelerator Center, CERN, ...). The code contains
sub-routines like AUTOMESH, LATTICE, which give to the user more control over the type
of mesh to be generated. Different mesh sizes can be used at any number of places. Figure 11
gives an example of a mesh for a solenoid. The mesh is finer where the beam passes. Figure 12
shows a plot of the magnetic flux lines.

Fig. 11 Mesh line for a solenoid from POISSON

Fig. 12 Magnetic flux lines inside the solenoid of Fig. 11

9. COMPUTER CODES FOR PARTICLE ACCELERATOR DESIGN
Many useful codes are widely distributed but several users make their own improvements
and the original author loses control. With each new generation of high energy accelerators,
limitations in the existing codes are found and therefore it is necessary either to improve an
existing code or to write a new one. The consequence is often a duplication of the effort and a
proliferation of different codes. When a simulation code provides an output which is
inconsistent with the experimental results or when the data are suspicious, one should make
comparisons with another similar code. But then arises the problem of different types of input
and output and the different theoretical approaches. Table 2 lists some of the simulation codes
presently available (alphabetic order).

Table 2
Some current simulation codes

Use Codes
Beam Dynamics:
- Linacs DYNAC, FLUX 2D, ITACA, MAPRO, PARMILA,
PARMELA, ...
- Cyclotrons NAJO, RELAX 3D, ...
- Circular machines ABCI, MAD, PATRAC, PATRICIA, SAD, ...
- Transfer lines TRANSPORT, TURTLE, ...

Component design
- RF cavities MAFIA, SUPERFISH, ...
- Magnets PE2D, POISSON, TOSCA, ...
- Ions and e- sources AXCEL, EGUN, ...
- Klystrons RMKT, ...

For the design of the next generation of high energy accelerators, there is a strong wish
to establish an international collaboration and even an international center for accelerator
codes. However it is not obvious how such collaborations can deal with the various types of
machines and with the different types of particles. For the time being, each laboratory involved
in the design of accelerators uses the widely distributed codes like TRANSPORT,
SUPERFISH, POISSON, and develops its own codes for the specific applications. A
compendium has been published by the Los Alamos Accelerator Code group [13] which gives
a description of 203 codes used in 10 laboratories around the world. Though it is not an
exhaustive list, it represents a useful document for those who are looking for existing
simulation codes which could cover the specific problem they have to deal with. Some new
improvements are not reported in [13]. For example, starting with the same approach as
MAPRO and PARMILA (protons and ions), the code DYNAC [1] provides a better accuracy
and can also be applied to electrons.
10. CONCLUSION
Whatever accelerator design one has to elaborate, it will always be necessary to use a
pencil and paper to express the basic ideas to be submitted to discussion and critics.
It is fundamental to develop analytical models in order to understand the simple physics
principles and for that a pocket calculator will always be useful and sufficient.
Finally when the physics is sufficiently understood, it is time to move to a desktop
computer or a large computer to run sophisticated codes which, when they are correctly used,
allow a rather good description of the new accelerator design and save time since it is not
necessary to build a prototype (magnet for example).
The door is still open to develop new simulation codes to take into account the very high
charges with their related effects, the superconductivity in linacs, cyclotrons and other circular
machines, and the high energies of the future accelerators.

ACKNOWLEDGEMENTS
My thanks go to F. Chautard, D.J. Warner and M. Weiss for their valuable comments on
this paper and to S. Turner for the careful reading of the text.

REFERENCES

[1] P. Lapostolle, Introduction to RF linear accelerators, these proceedings.

[2] D. Warner, Fundamentals of electron linacs, these proceedings.

[3] M. Weiss, Fundamentals of ion linacs, these proceedings.

[4] J. Mascureau, Induction linacs, these proceedings.

[5] Y. Jongen, Cyclotron magnet calculations, these proceedings.

[6] T. Stammbach, Introduction to cyclotrons, these proceedings.

[7] W. Schnell, Consideration of a two beam RF scheme for powering an RF linear collider,
CLIC note 7, November 1985.

[8] W.B. Herrmannsfeldt, EGUN-An electron optics and gun design program, Stanford
Linear Accelerator Center, report 331 (1988).

[9] B. Mouton, The PARMELA program, Version 4.3, LAL/Sera/93-455.

[10] K.L. Brown, F. Rothacker, D.C. Carey and C. Iselin, TRANSPORT: A computer pro-
gram for designing charged particle beam transport systems, CERN report 80-04.

[11] K. Halbach and R.F. Holsinger, SUPERFISH-A computer program for evaluation of RF
cavities with cylindrical symmetry, Particle Accelerators 7 (4), 213-222 (1976).
[12] R.C. Iselin, Solution of Poisson's or Laplace's equation in two-dimensional regions,
CERN long write up T604

[13] H. Deaven, and K. Chan, Computer codes for particle accelerator design and analysis,
Los-Alamos, LA-UR-90-1766.

BIBLIOGRAPHY

P. Lapostolle and A. Septier, Linear accelerators, North Holland Publishing Company,

Amsterdam, The Netherlands (1970).

S. Turner, ed., RF engineering for particle accelerators CAS, Oxford 1991, CERN 92-03
(1992).

S. Turner, ed., Magnetic measurement and alignment, CAS, Montreux 1992, CERN 92-05
(1992).
MATCHING ACCELERATORS TO THE TASK
M. Weiss
CERN, Geneva, Switzerland

Abstract
When a task or a project calls for the use of accelerators, these
have to be chosen in such a way as to match best the requirements.
The procedure applied in such cases is outlined in an example of a
medical facility for cancer treatment.

1. INTRODUCTION
The choice of an accelerator for a given facility is not a simple affair, it requires a good
knowledge of several types of accelerator, which can be quite different in the way they
operate. In a course like the present Applications of Cyclotrons and Linacs, it would have
been natural to try to compare these two types, and figure out which one is better suited for a
specified task. But, even among cyclotrons, various types exist, and the same is true for
linacs. In practice, one would require different specialists to analyse, compute, and optimize
different types of accelerators and, in the end, one would have to choose the best solution,
which is usually possible only after long discussions. To simplify things, this paper deals only
with linear accelerators, but nevertheless establishes procedures by which different types of
linac are chosen to form an accelerator complex best suited to the task.
In Section 2 the task is specified, in this case a compact accelerator for medical
purposes. Section 3 gives some preliminary design considerations, while Section 4 outlines
the design method and gives some results. In Section 5 the solution is discussed, and
Section 6 gives the concluding remarks.

2. DESCRIPTION OF THE TASK

The task we have chosen to demonstrate, the design procedure of an accelerator facility,
is really under study. It is an Italian project, aimed at establishing a network of hospitals
equipped with accelerators for cancer treatment. The project is called TERA, the Italian
acronym for hadrontherapy (terapia con adroni). Protons and light ions, like carbon or
oxygen, seem well suited to the treatment of tumours: the distructive action of a beam of such
particles can be directed on the cancerous cells, leaving the healthy tissue practically
unharmed. This is due to the fact that the destructive action of the beam is mostly
concentrated on the part of the body in which the particles are brought to a stop (Bragg peak).
By varying the energy of the beam and scanning it sidewise, one can expose only the region
of the cancerous cells of a tumour to destruction, and spare the healthy tissue.
In the TERA project, one of the hospitals in the network is the Centre, being the one
best equipped with accelerators and with beam delivery systems to the various treatment
rooms. Other hospitals are to be equipped with compact accelerators, i.e. accelerators of
limited size and weight. Among the candidates for a compact accelerator are the cyclotrons
(normal and superconducting) as well as linacs. In the latter case, in order to make the
accelerator small, one is tempted to go to high frequencies, and compose the compact
accelerator of various linac structures, each being the best choice in its energy range.
The specifications for a compact accelerator determine the particle type (proton), its
maximum energy (200 MeV), the number of particles accelerated per second (2 10 10) and,
if the beam is pulsed, a minimum repetition rate (100 Hz). In addition, the requirements are to
construct a reliable machine, easy to operate and to maintain, which means that one should
rely on proven technologies. The accelerator should fit in a room of 100 square metres
maximum, all equipment included. Of course, the cost efficiency of the proposed
accelerator could, in the end, be the determining factor for the choice.
This paper deals with the design of a compact accelerator composed of linacs operating
in the standing wave mode.

3. PRELIMINARY DESIGN CONSIDERATIONS

Before starting to design linacs as elements of the compact accelerator, one has to
define the compact accelerator itself in general terms. The following preliminary
considerations are of interest:

Peak intensity in pulsed beams: the average current corresponding to the required
2 10 particles per second is very small, approximately 3 nA (linacs can handle
100 mA and more); therefore, to save RF power, one can operate the linacs in a
pulsed regime. For pulsed beams, a minimum repetition rate of 100 Hz is required by
the physicians; if the beam pulse is 3 s long the peak current in the pulse is:
3nA
= 10 A ,
100 Hz3 s
which is again very small, even if one includes a safety factor in the peak current.

Approximate compact linac length: at this stage, we do not distinguish between the
various types of linacs used in the accelerator and apply the simple formula:
200 MeV
L= ,
qE0 T cos s
where q is the charge of the proton, E 0 T is the effective accelerating field which we
assume to be 10 MV/m, s is the synchronous phase, 20o , and L is the
approximate accelerator length, for which we get 22 m. Adding 20% to this length for
drift spaces between linac tanks, etc., we see that the accelerator will be about 27 m
long.

RF and mains power: again we do not distinguish between the various parts of the
compact accelerator and apply the following formula for the peak RF power required
to cover dissipation losses in the structure

Pp =
( E0 T )
2
L ,
Zseff
where Zseff is the effective shunt impedance per unit length [1], assumed here to be
about 50 M/m; one obtains Pp = 45 MW. (The additional power which is delivered
to the beam being accelerated is neglected because of the very small beam current.)
The average RF power is, of course, much lower, and is obtained by multiplying Pp
by the RF duty factor (product of the repetition rate, 100 Hz, and the RF pulse length,
10 s [beam pulse length, 3 s, + RF field build up, 7 s]):
Pa = 45 kW .
Assuming that the efficiency of the RF power production is 40%, we would require a
mains power of
Pmains 115 kW .
 Elements of the accelerator complex: the first linac in the complex is the radio-
frequency quadrupole (RFQ) [24], which accepts protons with input energies as low
as a few tens of keV, and then accelerates them to a few MeV. Protons gain their
input energy in an electrostatic accelerator. At high energies (from about 100 MeV
on) the side-coupled linac (SCL) [5] is a very suitable structure, operating efficiently
at very high frequencies, such as 3 GHz. For intermediate energies, between for
example 5 and 100 MeV, one uses the drift-tube linac (DTL) [6], which usually
operates in the frequency range of a few hundred MHz. Figure 1 shows schematically
the layout of the linacs, also including the ion source with the electrostatic accelerator
and the beam transport and matching lines.

Fig. 1 Elements of the accelerator complex.

Choice of frequencies: high frequencies are preferred for several reasons:
i) the accelerator is smaller;
ii) less RF power is required as Zseff f1/2 [7], f being the RF frequency;
iii) RF power generation is cheaper;
iv) the breakdown limit is higher. According to Kilpatricks law [8], the
breakdown field increases approximately as f 0.4.
An interesting frequency is 3 GHz, where the RF power generation technique is very
well developed. However, as we shall see, such a high frequency can be used only at
higher energies, approaching 100 MeV. At lower energies one has to use a
submultiple of 3 GHz, and then pass to the higher frequency when possible.

Phase acceptance and phase damping: a linear accelerator, apart from the RFQ,
accepts bunches with a phase spread of up to three times the synchronous phase angle
with which the linac operates. In linacs the phase extension of the bunch is damped
proportionally to ( )3/4. When passing to a higher frequency linac in a chain of
accelerators, one must make sure that the phase spread has been sufficiently damped
beforehand, because it will be increased by the ratio of the higher to the lower
frequency.

4. DESIGN METHOD
Our design method is to analyse separately the various linacs composing the compact
accelerator, and then try to match them to each other. It is convenient to start with the most
important linac, which in our case is the SCL. The SCL is the last accelerator in the chain,
which will bring the particles to 200 MeV; we shall operate it at the high frequency of 3 GHz.
The design of a linac is always an iterative process, as the optimization depends on
many parameters, which are interrelated.

4.1 Side-Coupled Linac (SCL)

 The SCL contains accelerating and coupling cells, which form a tank; a tank contains a
limited number of cells (for example, about 15 accelerating cells), and between tanks one
places magnetic quadrupoles in order to focus the beam. The electromagnetic energy passes
from one accelerating cell to the next via the coupling cell and from tank to tank via the
bridge coupler, as shown in Fig. 2.

Fig. 2 SCL tanks connected via a bridge coupler (not to scale).

The effective shunt impedance, Z seff, of a linac depends greatly on the size of the
aperture for the passage of the beam. This aperture is a function of the beam emittance, of the
margin we leave for alignment tolerance, and of the distance between focusing elements. For
a tank containing 15 accelerating cells and for a normalized beam emittance (geometrical
emittance multiplied by relativistic factors ) of 0.2 mm mrad, an aperture radius of 3 mm
is found to be sufficient. All the cells in a tank are the same, corresponding to the average
beam velocity in this tank. With the structure design program SUPERFISH [9] we shape the
SCL cells in order to obtain an optimum Zseff. Figure 3 shows the Zseff of the SCL tanks for
different particle energies.
The graph in Fig. 3 is very instructive: it shows how the SCL becomes less efficient at
lower energies. The reason for this becomes clear in Fig. 4: at low energies the cells are very
short and there is an important RF energy loss per unit length on the walls.

Fig. 3 Zseff of SCL tanks as a function of energy.

 Fig. 4 Shape of an SCL tank at different energies.
We decide, for the time being, that the SCL should start at about 70 MeV. We still have
to choose the average accelerating field in the tanks, E0. A high E 0 makes the accelerator
short (interesting for a hospital), but the required RF power increases, being proportional to
E02 . The Kilpatrick breakdown limit Ek is, at 3 GHz, 47 MV/m (see Fig. 5), and as this limit
is very conservative, one usually allows for the maximum surface field Es 2 Ek. In SCL
structures, due to their geometry, Es 6 E0; from this a reasonable value for E0 is about
15 MV/m.

Fig. 5 Kilpatrick breakdown limit as function of frequency.

To design an SCL linac the following procedure is used:

i) A structure program like SUPERFISH determines the geometry of linac cells

corresponding to the frequency of 3 GHz and to different particle velocities; the
 cells have a bore radius of 3 mm. The program also computes the field
distribution in the cell, the transit-time factor T, the wall losses, etc. The program
SUPERFISH is well known and used in many laboratories.
ii) The cells are grouped in tanks with a linac design program; once the synchronous
phase angle is specified (in our case s = 15o), the design program computes the
energy gain of each tank, and determines the required focusing strength of
magnetic quadrupoles. Linac design programs are more specific and differ from
laboratory to laboratory.
iii) The last step in the calculation is made with a beam simulation program, which
simulates the beam by a few thousand particles and follows them through the
linac; if the result is not satisfactory, one starts the whole procedure again, after
having changed some design parameters. The simulation program for the SCL
differs again from laboratory to laboratory.
In Fig. 6 accelerating cells for different energies are presented, computed by
SUPERFISH. To keep the resonant frequency always at 3 GHz, the ratio g/L (gap to cell
length) had to be varied from 0.255 (at 70 MeV) to 0.386 (at 200 MeV). Table 1 gives some
results of the linac design program: the SCL accelerator is about 12 m long (including
intertank distances), contains 30 tanks (the tank length increases from 26 cm at the beginning
to 39.5 cm at the end) with 14 accelerating cells each, and consumes 26 MW of peak RF
power. As 3 GHz klystrons with Pp = 30 MW are available on the market, only one klystron
suffices for the whole SCL, which is interesting as far as costs are concerned (one big
klystron is cheaper than several smaller ones giving the same power). The magnetic
quadrupoles placed between the tanks (the intertank distance is 1.5 ) are 38 mm long and
their gradients start with 117 T/m (tesla/m) and are reduced gradually to 88 T/m (these
gradients correspond to the specified phase advances per period of 45 o at 70 MeV, falling
gradually to 30o at 200 MeV); the bore radius is 3 mm. The results of the beam simulation
program are shown in Figs. 7 and 8. Figure 7 presents the various beam profiles along the
SCL, while Fig. 8 gives the input and output phase space projections. No particles are lost in
the SCL, as the beam is well inside the aperture.

Fig. 6 Accelerating cells of the SCL as computed by SUPERFISH (presented with and without the triangular
mesh of variable size).
Fig. 7 Beam profiles plotted against tank numbers along the TERA SCL. The display at the top shows the
horizontal (x) coordinates of the beam; the centre display shows the phase spread of the beam; and the
bottom display shows the energy spread.

Fig. 8 Input (upper three displays) and output (lower three displays) phasespace projections of the beam in the
TERA SCL: xx', yy', and ddW.
 Table 1
Design parameters of the SCL

Tank Tank Total Energy Z seff Power/tank Total Quad

number length length power gradient
[mm] [mm] [MeV] [M/m] [MW] [T/m]
[MW]
1 259.75 314.8 73.80 61.7 0.877 0.88 116.4
2 265.20 636.3 77.27 63.3 0.874 1.75 115.3
3 270.60 964.3 80.82 64.9 0.871 2.62 114.3
4 275.94 1298.8 84.43 66.4 0.869 3.49 113.3
5 281.23 1639.7 88.12 67.9 0.867 4.36 112.2
6 286.46 1987.0 91.87 69.3 0.866 5.22 111.3
7 291.64 2340.6 95.70 70.7 0.865 6.09 110.3
8 296.76 2700.4 99.59 72.0 0.864 6.95 109.3
9 301.82 3066.4 103.55 73.3 0.864 7.82 108.3
10 306.83 3438.4 107.57 74.5 0.864 8.68 107.4
11 311.78 3816.5 111.66 75.7 0.864 9.55 106.4
12 316.67 4200.5 115.81 76.9 0.865 10.41 105.5
13 321.50 4590.4 120.03 78.0 0.865 11.28 104.5
14 326.28 4986.1 124.31 79.1 0.866 12.14 103.6
15 330.99 5387.5 128.65 80.1 0.867 13.01 102.7
16 335.65 5794.6 133.04 81.1 0.868 13.88 101.7
17 340.25 6207.3 137.50 82.0 0.869 14.75 100.8
18 344.79 6625.5 142.01 82.9 0.871 15.62 99.8
19 349.27 7049.1 146.59 83.8 0.872 16.49 98.9
20 353.69 7478.1 151.21 84.7 0.874 17.36 98.0
21 358.06 7912.5 155.90 85.5 0.875 18.24 97.0
22 362.37 8352.0 160.63 86.2 0.877 19.12 96.1
23 366.62 8796.8 165.42 87.0 0.879 19.99 95.1
24 370.81 9246.6 170.26 87.7 0.881 20.88 94.2
25 374.94 9701.4 175.15 88.4 0.883 21.76 93.2
26 379.02 10161.3 180.09 89.0 0.885 22.64 92.2
27 383.04 10626.0 185.08 89.7 0.887 23.53 91.3
28 387.01 11095.5 190.12 90.3 0.889 24.42 90.3
29 390.92 11569.8 195.21 90.8 0.891 25.31 89.3
30 394.77 12048.7 200.34 91.4 0.893 26.20 88.3
Total peak RF power: 26.20 MW

4.2 Radio-Frequency Quadrupole (RFQ)

Having determined the high-energy end of the compact accelerator (SCL from 70 to
200 MeV), we now consider the low-energy part and proceed by designing an RFQ. To
accept the beam, the average aperture radius of the RFQ is 2 mm. For the correct functioning
of the RFQ, the cell length ( /2) should be at least as long as the aperture radius. Assuming
an injection energy of 80 keV ( = 0.013), we find that > 0.3 m or f < 1 GHz. So, for the
operating frequency of the RFQ we choose the highest acceptable submultiple of 3 GHz,
i.e. 750 MHz.
To design the RFQ one proceeds essentially in the same way as with the SCL. Having
decided to use the four-vane RFQ, it suffices, due to symmetry, to compute only an octant
of it by SUPERFISH, as shown in Fig. 9. The inner diameter of the RFQ is 10 cm. To
facilitate the design, the RFQ is usually divided into several sections, each section performing
a specified action on the beam. Most of the RFQ design programs now in use were started at
Los Alamos National Laboratories (LANL), USA. In our case, the beam intensity is rather
low and consequently the transmission efficiency is of minor importance. It matters more to
maintain the beam emittances, as one has to inject into other accelerators and even jump in
frequency. Figures 10 and 11 show outputs of the beam simulation program. Figure 10 shows
the various beam envelopes along the RFQ, while Fig. 11 presents the evolution of the beam
in the longitudinal phase plane: one sees how the beam gets progressively bunched, and how
it is well inside the separatrix. The RFQ accelerates to 4 MeV, an energy sufficiently high to
inject into the following accelerator, the DTL. The RFQ is 2.6 m long and its transmission
efficiency is nearly 90%. The peak RF power to cover structure losses is about 0.4 MW, the
average one is about 1000 times smaller. As it is well known, the RFQ does not need an outer
focusing, because all the actions on the beam (bunching, acceleration and focusing) are
performed by the RF electromagnetic fields.

Fig. 9 An octant of the RFQ computed by SUPERFISH.

Fig. 10 Beam profiles along the RFQ; the top two displays show the horizontal and vertical beam envelope; the
third display shows the phase spread and the fourth one the energy spread.
Fig. 11 The evolution of the longitudinal phase plane in the RFQ: the separatrix and the longitudinal beam
emittance.

4.3 Drift Tube Linac (DTL)

The DTL, operating at 750 MHz, has to accelerate the beam from 4 to 70 MeV; at its
output, the beam must be sufficiently damped in phase to allow a frequency jump of 4 in the
SCL.

Fig. 12 A cell of the DTL computed by SUPERFISH (variable mesh size).

Proceeding again in the usual way, we compute some cells using SUPERFISH (an
example of a cell is shown in Fig. 12) and then use a design program such as PARMILA
(developed at LANL) which interpolates between SUPERFISH results and creates a linac
according to our specifications. We have specified the average field on the axis,
E0 = 6.6 MV/m (which keeps the maximum surface field Es below 2 E k ), the synchronous
phase angle s = 30o , the focusing and FFDD with all quadrupole gradients being 20 T/m
except the first four which will serve for matching. PARMILA has calculated a DTL with 145
cells and a length of 13.5 m. The peak RF power consumption is about 9 MW. The outputs of
the beam simulation program are shown in Figs. 13 and 14. Figure 13 presents the beam
envelopes and we see that they perform betatron and synchrotron oscillations: we have
deliberately slightly mismatched the beam at the input to the DTL to see the effects. Figure 14
shows some characteristics of the output beam. Of particular interest is the phase spread, as
we have to inject into the SCL, which has a four times higher frequency and operates with a
synchronous phase angle of 15o . The phase spread of 5o (which will become 20o with
respect to the frequency of 3 GHz) is too high. We will either have to match better into the
DTL or increase the synchronous phase angle in the first few tanks of the SCL.

Fig. 13 Beam sizes along the DTL in the horizontal, vertical, and longitudinal direction.

5. DISCUSSION OF THE RESULTS

In Section 4 we have shown the main lines of an approach to the design of the
accelerator complex intended for medical purposes. Very common types of accelerators have
been chosen: the RFQ in the low energy range, where a beam bunching is also necessary; the
DTL in the middle energy range, where a quadrupole focusing in each cell is available; and
the SCL in the high-energy range, where one could apply the high frequency of 3 GHz. The
emphasis in our study has been given to the design of cavities and to the beam dynamics.
Normally a thorough matching study between the accelerators would follow with a beam
simulation through the whole machine. A reconsideration of input and output energies of
various accelerators (for example, as a function of their respective Z seff) could also be
undertaken in the final optimization.
Fig. 14 Output beam characteristics of the DTL: the upper displays show the phase spectrum and the cross-
section of the beam; the lower displays show the longitudinal beam emittance and the energy spectrum.

The accelerator complex, including the ion source and the matching lines, is about 30 m
long. To place the accelerator more conveniently in a room, one would envisage a 180o bend
after the DTL. In this case, the two parts of the compact accelerator, about three to four
metres from each other, would leave enough space to place all the electronics and additional
equipment between them. The peak RF power consumption at 3 GHz is about 26 MW, and at
750 MHz it is about 10 MW. The bigger power consumption at the higher frequency is due to
the much higher field E0 in the SCL, chosen in order to shorten the accelerator complex.
The RF power generation at 750 MHz is much less developed than at 3 GHz, and in
addition it is expensive. It would be advantageous for the medical accelerator to find means to
switch to the higher frequency at a lower energy and drastically reduce the power
consumption at 750 MHz. This has in fact been done for the TERA project... but that is
another story.

6. CONCLUDING REMARKS
The design of an accelerator complex, apart from the electrodynamics and beam
dynamics problems treated in this paper, of course involves other subjects which have to be
properly dealt with. The mechanics has to provide a rigid structure, easy to align, and
thermally stable (water cooled, if necessary); the RF has to be properly fed into the cavity, so
that the latter is matched to the RF line, and the amplitudes and phases of the fields in the
cavities have to be kept in tight tolerances. The vacuum system has to provide the required
operational pressure (in the range of 107 Torr) and the control system must be capable of
handling remotely all of the major parameters. The diagnostics equipment plays an important
role in the commissioning of the accelerator, as well as in its operation.

7. ACKNOWLEDGMENT
A. Lombardi's help with the computations was greatly appreciated.

REFERENCES

[1] M. Weiss, Fundamentals of ion linacs, these proceedings.

[2] R.H. Stokes et al., The radio frequency quadrupole, Proc. 11th Int. Conf. on High
Energy Accelerators, Geneva, 1980 (Birkhauser, Basle, 1980).

[3] M. Weiss, Radio frequency quadrupole, Proc. CAS, Aarhus, Denmark, 1986; CERN
8710 (1987).

[4] A. Schempp et al., Zero-mode RFQ development in Frankfurt, Proc. Linear Accel.
Conf., Seeheim, Germany, 1984, ed. N. Angert (ISSN: 01714546, GSI8411, 1984),
p. 338.

[5] E.A. Knapp, High energy structures, in Linear Accelerators, eds. P.M. Lapostolle and
A.L. Septier (North Holland, Amsterdam, 1970), pp. 601616.

[6] P. Lapostolle, Introduction la thorie des acclrateurs linaires, CERN 8709 (1987).

[7] G.A. Loew and R.B. Neal, Accelerating structures, in Linear Accelerators (see
Ref. [5]), pp. 4752.

[8] W.D. Kilpatrick, Criterion for vacuum sparking designed to include both RF and DC,
in The Review of Scientific Instr., Vol. 28, number 10, October 1957, pp. 824826.

[9] K. Halbach and R.F. Holsinger, SUPERFISH a computer program, Part. Accel. 7 (4)
(1976) pp. 213222.
COMMERCIALLY AVAILABLE ACCELERATORS
O. Barbalat
CERN, Geneva, Switzerland

1. INTRODUCTION

Particle accelerators originated from research for nuclear physics but they have now
found many applications in research, industry and medicine. The extent of some of these
applications is now wide enough to have a number of firms producing commercial machines
adapted to the needs of this growing market. This report briefly reviews the machines made or
commercialized by European companies.

As the parameters and performance figures may sometimes vary substantially it has
seemed preferable to advise the interested reader to consult directly the firms concerned. Their
addresses are given in Appendix.

2. RADIATION THERAPY

The bulk of radiation therapy for cancer treatment is today achieved through the use of
electron accelerators. Because of their higher cost, the use of ions is still limited to a very small
number of facilities.

Linear electron accelerators for cancer therapy are machines mounted in a mobile gantry
which also includes a beam transport system to deliver with precision the radiation to the area
requiring treatment. Depending on the type of treatment the system can work either in the
electron beam mode or in the photon mode (photons are produced by bremsstrahlung on a
target).

The machines can work at different energies, adjustable from 6 to 25 MeV depending on
the depth of the tumour.

Manufacturers such as General Electric Medical Systems* (FR), Philips (NL) and
Siemens (DE) offer not only the accelerator but complete diagnostics and therapy systems
combining, in a single computer assisted facility, the diagnostics device (X-ray scanner or
Magnetic Resonance Imager) and the accelerator for radiation therapy.

Scanditronix (SE) is also manufacturing medical electron accelerators but they are based
on the microtron principle. They offer machines capable of reaching 50 MeV.

3. RADIATION PROCESSING ELECTRON ACCELERATORS

The applications of these machines range from the treatment of polymer cable insulation
to increase the heat resistance to food processing to improve preservation, sterilization of agro-
industrial effluents or inactivation of medical waste and many others.

The present market for radiation beam processing is however still rather limited and while
capable of adapting their medical machines, major manufacturers like Philips and Siemens do
not mention this application in their catalogues. However General Electric has created in France

* The company is the successor of CGR (Compagnie Gnrale de Radiologie), a long established
French manufacturer of medical equipment
a subsidiary with former CGR people, called MeV Industrie SA, for this particular market.
Their standard machine has an energy of 10 MeV with a beam power of 20 kW.

The Budker Institute of Nuclear Physics in Russia has commercialized two lines of
machines. One is based on a high voltage rectifier for energies up to 2 MeV and another uses
radio frequency resonators for energies up to 3 MeV.

Scanditronix (SE) is also producing electron microtrons for these types of industrial
processing.

Ion Beam Applications (IBA) in Belgium is developing an electron machine for high
power applications (100 kW), the Rhodotron, on the basis of a novel design from CEA Saclay.

4. ION IMPLANTATION

The implantation of ions for semiconductor manufacturing is with radiation therapy the
largest application of particle accelerators. However there does not seem to be any European
manufacturer.

Ion implanters for the modification of properties of the materials, in particular metals, are
proposed by Danfysik (DK) and Whickham Ion Beam Systems (UK).

These machines are limited to the few hundreds of keV range and use high voltage
rectifiers.

5. CYCLOTRONS FOR ISOTOPE PRODUCTION AND PROTON THERAPY

While in the past a number of companies (Philips, CGR, Scanditronix) did manufacture
cyclotrons, the only present producer is IBA (BE) which has developed several lines of
compact and efficient ion accelerators ranging from 5 to 30 MeV, both for general purpose
medical products and for short- lived positron emitting isotopes used for tomography (PET
imaging). IBA is also offering higher energy cyclotrons (200/500 MeV) for proton therapy.
Oxford Instruments (UK) has developed a compact 12 MeV cyclotron for PET applications.

6. SYNCHROTRON RADIATION

Oxford Instruments has also designed and built a compact 700 MeV electron synchrotron
(using superconducting magnets) to produce synchrotron radiation for micro-lithography.

7. RESEARCH ACCELERATORS

Medium and high energy research accelerators have traditionally been designed and built
by the laboratory which intends to use them. However there are now a few commercial firms
which can take this activity over at least for machines having parameters (energy/beam power)
which do not require substantial developments.

One can mention in particular Euro-MeV (FR) and Scanditronix (SE) for electron
accelerators and IBA (BE) for cyclotrons.

Low energy accelerators and the associate systems (ion sources, beam lines and
monitors) used in particular for research in material science through various ion beam analysis
techniques are manufactured by High Voltage Engineering Europa (NL). The machines, called
Tandetrons, use a high voltage multiplier-rectifier stack with negative ion sources combined
with charge exchange. They can reach energies up to 15 MeV.

8. CONCLUSION AND OUTLOOK

The present commercial market for accelerators is dominated by medical applications:

electron linacs for therapy and ion cyclotrons for isotope production mainly for medical
diagnostics.

If the prospects mentioned elsewhere in these proceedings to use accelerators in the

energy field for nuclear waste incineration and energy generation materialize, it could open a
wide new domain for ion linacs or cyclotrons and lead to the development of a new industrial
sector.

Medical machines are characterized by energies in the tens of MeV range and beam power
levels of a few kW. Present radiation processing applications require beam power of a few tens
of kW. The new applications will require energies in the GeV range and megawatts of beam
power. They will need high efficiencies, both in term of power conversion from electric mains
to beam and in terms of beam loss to keep induced radioactivity to a manageable level. It will
be a challenge both for the accelerator designers as well as the industrialists, but it is also a
highly promising perspective.

ACKNOWLEDGEMENTS

Accelerator manufacturers have kindly supplied information and data sheet concerning
their products.

BIBLIOGRAPHY

O. Barbalat, Applications of Particle Accelerators, CERN/AC/93-04 (BLIT) rev. February,

1994.

John Gordon et al., Atom smashers for profit and pleasure, Physics World, June 1994.

Franois Savatier, Dcontamination. Nouvelles applications pour relancer l'ionisation, L'Usine

Nouvelle, 5 mai 1994.

See also specific contributions to the School included in these Proceedings.

APPENDIX: LIST OF FIRMS MANUFACTURING ACCELERATORS

Danfysik A/S Oxford Instruments Ltd.

DK-4040 JYLLINGE Osney Mead
Tel. + 45.46.78.81.50 GB-OXFORD OX2 0DX
Fax. + 45.46.73.15.51 Tel. + 44.865.24.14.56
Fax. + 44.865.72.32.64
EuroMeV
M. J.Ch. Malglaive Philips Nederland B.V.
49 rue Lamartine Philips Medical Systems
F-78000 VERSAILLES Mr. G.H.M. Coolen (Sales Support)
Tel. + 33.1.30.21.82.06 P.O. Box 90050
Fax. + 33.1.39.53.65.17 NL-5600 PB EINDHOVEN
Tel. + 31.40.788.326
GE Medical Systems Fax. + 31.40.785.827
100 rue Camille Desmoulins
F-92137 ISSY-les-MOULINEAUX Cedex Scanditronix AB
Tel. + 33.1.40.93.30.00 Husbyborg
Fax. + 33.1.40.93.33.33 S-75229 UPPSALA
Tel. + 46.18.18.07.00
High Voltage Engineering Europe Fax. + 46.18.53.72.76
B.V.
Mr. H. van Oosterhout (Sales Manager) Siemens Aktiengesellschaft
P.O. Box 99 Medical Engineering Group
NL-3800 AB AMERSFOORT Henkestrasse 127
Tel. + 31.33.619741 D-8520 ERLANGEN
Fax. + 31.33.615291 Tel. + 49.9131.84.0
Fax. + 49.9131.84.37.54
Ion Beam Applications S.A.
Chemin du Cyclotron, 2 Whickham Ion Beam Systems Ltd.
B-1348 LOUVAIN-la-NEUVE DARCHEM
Tel. + 32.10.47.58.11 West Auckland Road
Fax. + 32.10.47.58.10 Darlington
GB-Co DURHAM DL3OUP
Budker Institute of Nuclear Physics Tel. + 44.325.38.23.33
Professor A.N. Skrinsky Fax. + 44.325.28.21.94
Director
Siberian Division
Academy of Science of the Russian
Federation
630090 NOVOSIBIRSK
Russia
Tel. + 7 095.65.60.31/65.09.10
Fax. + 7.095.237.46.21

MeV Industrie S.A.

Parc Technologique Gemini II
Route de l'Orme
F-91195 St. AUBIN
Tel. +33.1.60.19.13.29
Fax. +33.1.60.19.10.32