PHYSICAL REVIEW ACCELERATORS AND BEAMS 24, 080401 (2021)

Pushing the capture limit of thermionic gun linacs

Sadiq Setiniyaz ,1,2,* Alejandro Castilla ,1,2 Julian McKenzie ,2,3 Robert Apsimon ,1,2
Boris Militsyn,2,3 Deepa Angal-Kalinin,2,3 and Graeme Burt1,2
1
Engineering Department, Lancaster University, LA1 4YW, United Kingdom
2
Cockcroft Institute, Daresbury Laboratory, Warrington, WA4 4AD, United Kingdom
3
ASTeC, STFC Daresbury Laboratory, Warrington, WA4 4AD, United Kingdom

(Received 20 May 2021; accepted 2 August 2021; published 16 August 2021)

Although accelerator technology has matured sufficiently, state-of-the-art x-ray linacs for radiotherapy
and cargo-scanning capture merely 30%–50% of the electrons from a thermionic cathode, requiring a
higher cathode current and leaving uncaptured electrons to cause problems such as back bombardment on
the cathode leading to a shortening of cathode life. Any solution to increase capture should be effective,
simple, reliable, compact, and low cost in order to be adopted by the industry. To address this, we present
the design of a 6 MeV high capture efficiency S-band electron linac that captures 90% of the initial dc
beam. This linac does not require any extra parts that would increase the cost as the high efficiency is
achieved via a low-field amplitude in the first bunching cell to decrease the number of back-streaming
electrons, to velocity bunch the electron beam, and recapture back-streaming electrons. Under the low field
amplitude, any electrons launched at decelerating phases travel backward with low speeds, thus most of
them can catch the next rf cycle, and get reaccelerated/recaptured. As the electron speed is low, the cell
length is also shorter than existing linacs. Such a short field is achieved by the use of asymmetric cells with
differential coupling to the side-coupled cells. Our novel design has implications for all commercial high
current thermionic gun linacs for increasing beam current and increasing cathode lifetime.

DOI: 10.1103/PhysRevAccelBeams.24.080401

I. INTRODUCTION regarded as quasi-dc. While half the electrons enter the

cavity at accelerating phases of the rf, the other half enters
Commercial linear accelerators have a wide range of
the cavity at decelerating rf phases, and get accelerated in
applications with the main application being as a MeV-level
the opposite direction, with most traveling backward and
x-ray source for the medical field of radiotherapy [1]. thus end up hitting the cathode. This phenomenon is known
Radiotherapy is a key component in the treatment of many as the back bombardment (BB) [7–15] effect. The back-
cancers, but a major barrier to scaling up the number of streaming electrons deposit their kinetic energy (KE) to the
commercial linacs is the maintenance cost, both in terms of cathode and heat it up, which causes the cathode to generate
manpower and the financial cost, due to the regular more electrons and increase the beam current. As the beam
replacement of components with a limited lifetime com- current increases, more rf energy is taken away by the
pared to the operating life of the machine. Increasing the beam, and the cavity voltage drops, which causes the final
mean time between failures of these components could beam energy to become unstable. In practice, the beam
significantly decrease the maintenance cost as well as current and pulse length are limited to mitigate BB. BB
reducing the amount of time the machine is down for shortens the cathode lifetime by heating it up so it is
repair [2]. One such component is the cathode in the poisoned easily and degrades faster [15–17].
thermionic electron gun [3]. Some work has been done to reduce the BB effect by
The linacs utilized for x-ray radiotherapy and cargo- external magnetic fields [7,9]; hollow cathodes [9,12];
scanning typically use thermionic cathodes that generate cooling [11]; and improvements to the cathode material
long macropulses of several microseconds [4–6]. Such [18]. These methods would add extra cost and complicate
macropulses cover many rf cycles and are therefore the system. Here, we introduce a novel method to directly
lower BB by increasing the capture efficiency and con-
*
[email protected] sequently reducing the number of electrons traveling
backward. The capture efficiency is the percentage of
Published by the American Physical Society under the terms of electrons that reach the exit of the linac from those emitted
the Creative Commons Attribution 4.0 International license.
Further distribution of this work must maintain attribution to from the cathode. In our design, we increased the capture
the author(s) and the published article’s title, journal citation, efficiency from the 30%–50% of most standard existing dc
and DOI. thermionic gun linacs to over 90%. Of the 10% uncaptured

2469-9888=21=24(8)=080401(12) 080401-1 Published by the American Physical Society

SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

electrons, around 3% are lost by hitting the cavity aperture,

and 7% are lost by back-streaming. In existing standard
thermionic gun linacs, aperture loss and back-streaming
loss are at least 15% and 36%, respectively. Thus, in our
design BB electrons are reduced by more than 80% for a
given gun current. The cathode lifetime can be further
increased as due to the high capture efficiency; less current
needs to be extracted from the cathode to supply the same
amount of current to the target after acceleration.
Consequently, the cathode can operate at a lower current
density that will slow down wear out and increase lifetime
[19] as well as further halving the number of BB electrons.
As the cathode current density is lowered, the cathode
temperature can be lowered as well [20], which again FIG. 1. Longitudinal E-field profile of linac from Ref. [29].
increases the lifetime by abating the aforementioned
poisoning.
Another advantage of high capture efficiency is it The Varian Clinac-1800 accelerator captures phases
reduces unwanted harmful radiation generated by electrons between −110° and 20°, which is capture efficiency of
hitting the cavity walls. Electrons hitting the cavity walls 36% [4]. A study by Aubin et al. [25] reported 37%  2%
will create x rays and require shielding around the cavity capture efficiency measured from a Varian 600C linac.
[21], and/or the use of solenoid magnets to focus these Their simulation predicted that the capture efficiency can
electrons [4]. This will increase gantry size, complicate the be increased up to 45% with a converging electron beam.
system, and eventually increase the cost. As the aperture Baillie et al. [26] designed a variable-energy linac and
loss is decreased by more than 90% in our cavity design, reported capturing 133.7 mA out of 373 mA gun current,
such a linac requires significantly less shielding. It also which is about 36% efficiency.
does not require a solenoid, although the addition of one ASTRA (a space charge tracking algorithm) is a 2.5D
would further reduce aperture loss, thus increasing capture. tracking code with space charge [27,28]. ASTRA simula-
While other high capture efficiency designs are reported tions were performed to study the capture and loss of an
in the literature, they tend to be long and need a large existing S-band linac structure with six cells by tracking
gantry. The associated high manufacturing and mainte- electrons through an E-field map shown in Fig. 1 [29]. The
nance cost would render them unfavorable options, espe- blue dots are measurements made by the bead-pull tech-
cially for radiotherapy. In our design, the linac is a π=2- nique, and the red line is the fit. Initial dc beams with 9.5,
mode standing wave (SW) side-coupled accelerating struc- 15, and 25 keV kinetic energy (KE), and 100 mA current
ture, similar to the structures in Refs. [22–24], with a length were simulated, with the results given in Table I. The
shorter than 30 cm and beam energy of 6 MeV. This is ASTRA simulations agree well with the prediction of Baillie
achieved by having a step in the gradient between the first et al. [25]. When the Ez -field profile is scaled up so at its
and second cell, allowing a low field capture cell, to allow max Ez;max ¼ 50 MV=m the final beam energy at the target
low energy velocity bunching, before being captured and (located at the exit of the linac) is 5.8 MeV.
accelerated in a higher gradient section. Literature reviews and our own simulations confirm that
the capture efficiency of existing commercial linacs is close
to 50% at best. 37% of this is the back-streaming electron
II. CAPTURE EFFICIENCY OF EXISTING loss and 15% are the electrons lost on the aperture. Existing
COMMERCIAL LINACS commercial linacs in operation capture only 36%–37% of
In existing commercial linacs, the first cavity cell is electrons, which indicates higher back-streaming loss and
roughly half the length of the other cells, so that in-phase aperture loss. A major cause of this loss is that the
electrons gain the maximum acceleration from the rf wave
[4], while increasing the number of electrons that receive
this acceleration. The accelerating gradient is constant TABLE I. Simulated capture efficiency and loss of commercial
along the standing wave cavity and ranges from linac.
10–25 MeV=m depending on the design. To match the Initial KE
phase velocity to the electron velocity, the cell length is a
9.5 keV 15 keV 25 keV
function of velocity, L ¼ βλ2 , where β is the ratio of particle
velocity to speed of light, and λ is the rf wavelength. More Capture (%) 41.6 45.4 48.0
Back-streaming loss (%) 34.6 35.5 36.6
than 50% of the beam is lost inside linacs as the linac can
Aperture loss (%) 23.8 19.1 15.4
only capture electrons at certain phases.

080401-2
PUSHING THE CAPTURE LIMIT OF THERMIONIC GUN … PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

accelerating/decelerating voltage of the first cell is much each cell. An output file is generated for each cell length
larger than the electron beam energy, causing almost half of and field amplitude, which contains information such as the
the electrons to be lost immediately. Significant improve- particle launch phase, exit phase, exit time, exit KE, and
ment can be achieved if the gradient and lengths of the first flag for whether the particle is lost or captured. The code
few cells could be optimized. can also be run in tracking mode, where we can investigate
the beam dynamics for specific field settings.
III. 1D TRACKING CODE FOR CAVITY
OPTIMIZATION B. Beam dynamics study of existing
commercial linacs with 1D code
A. 1D tracking code
We have studied the performance and beam dynamics of
A 1D tracking code was developed to optimize the length
a commercial linac with the field profile of Fig. 1 using the
and E-field amplitude of each cell by assessing the exit/
1D code. The tracking results are shown in Fig. 2. The exit
arrival phases and energies of electrons at each cell as a
phase, and KE, are given for all captured phases. The
function of the launch phase. This is a method originally
launched electrons covered a full rf cycle, i.e., 360°, but
proposed for the design of electron bunching for high-
only those launched within a phase range of −131° and 59°
efficiency klystrons [30]. This code does not include space
successfully exited the first cell. Electrons are further lost in
charge and is only used for a fast and efficient scan of
later cells, and only those launched at phases between
parameters, and the optimized result from this initial
−131° and 36° successfully exited the last cell. This gives a
optimization will be verified and further optimized by
capture efficiency of 167°=360° ¼ 46.4%. This is similar
ASTRA simulations that include transverse plane and space
to the results of ASTRA simulations shown in Sec. II. In
charge effects. The exit phase is the rf phase in which an
Fig. 2(a), we observe a slight narrowing of the exit phase
electron exits a cell. The launch phase is the rf phase an
range as electrons travel through each cell, which indicates
electron enters the first cell. The aim is to maximize the
range of launch phases which result in the same exit phase
after the first few cells. The code needs as inputs: the
E-field profile, rf frequency, particle charge, mass, and
initial KE. Users also need to specify the maximum
tracking time T max and number of time steps N steps . The
code computes the distance traveled by the particle dz
during the time interval of dt ¼ T max =N steps . The velocity
of the particle is obtained by using its KE. The change in
the KE during this time interval is

dðKEÞ ¼ Eðz; tÞdz; ð1Þ

where Eðz; tÞ is the longitudinal electric field seen by the

particle at time t and location z, which can be given as

Eðz; tÞ ¼ Ez cosð2πft þ ϕlaunch Þ; ð2Þ

where Ez is the longitudinal electric field profile, f is the

cavity frequency, and ϕlaunch is the launch phase. The
electrons are launched at a fixed phase/time interval and
tracked in the longitudinal direction until they exit from
either end of the field profile. Electrons are launched at a
range of phases to cover at least one full rf cycle. If an
electron exits from the beginning or end of the field profile,
it will be counted as lost or captured respectively. Electrons
that do not reach the exit within the maximum tracking time
are also counted as lost. Presently, the code neglects space
charge and so has only been used for initial parameter FIG. 2. Results of 1D tracking code simulations of a commer-
scans, as its speed and advanced methods of optimization cial linac with 25 keV initial electrons. Exit phase (a) and exit KE
have provided approximate global optimum values. (b) vs launch phase for cells c1 through c6 of the linac. The KE
For optimization purposes, the code can be run in a scan energy in (b) is projected to the y axis to show the KE spectrum in
mode, in which it varies the lengths and field amplitudes of arbitrary units at the corresponding cell exits.

080401-3
SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

beam, of duration equal to applied rf pulse length. This is

usually several microseconds long. The electron macro-
pulse splits into many electron bunches under the rf field as
some electrons are captured by accelerating phases and
some are lost. For S-band, the rf wavelength is 333 ps long
and so bunches are repeated every 333 ps. The captured
electrons experience different acceleration and some travel
faster than the others. In Fig. 3, we see early electrons (i.e.,
electrons from −130° to −70° phases) travel slower, and
later, faster electrons catch up with them, which leads to an
overall shortening of the bunch length. This process is
called velocity bunching.
Velocity bunching takes place in the early cells of
accelerators while the electron energy is subrelativistic,
and the bunch has a significant velocity gradient. As the
beam energy increases, the electrons in the bunch have
similar velocity, therefore velocity bunching cannot occur.
The field gradients in the early cells are important as they
FIG. 3. Applegate diagram of an existing standard commercial determine the amount of bunching. A too high or low field
linac. gradient results in suboptimal bunch length, and particle loss.
In existing commercial linacs, the field in the first cell is
too high for optimal bunching, as half of the phases are
bunching. In Fig. 2(b), we can see cell by cell energy gain. decelerated quickly and hit the cathode. Therefore, low-
It takes six cells for most electrons to reach 6 MeV. The ering the field amplitude in the first cell is the key to
electrons launched around ½40°; −60° phase range have optimizing bunching and capture. Having a low field
very low energy. All this information is useful to under- reduces the acceleration in the first cell, hence the relativ-
stand the beam dynamics of the linac. istic β is smaller. This implies the cell length should be
The 1D code can also generate an Applegate diagram, as shorter. However, we can make the cell longer than the one
shown in Fig. 3. Applegate diagrams are used in klystron expected from the beta so that electrons that do not escape
design to show how electrons bunch as they travel along the in the first rf cycle can be reaccelerated in the second rf
cavity and are useful here to study the dynamics of cycle. They will then arrive at the second cell at the correct
bunching. The x axis is the position of the particle inside phase to be accelerated. In other words, the first cell should
the cavity and the y axis is its phase. If the particle’s be longer than the synchronous beta-cell length to give
position increases or decreases as the phase increases, then additional delay for the recaptured electrons so they enter
the particle is traveling forwards or backward respectively. the second cell at an accelerating phase. This will cause
If an electron travels backward and passes z ¼ 0, then this them to form a bunch with electrons from the second rf
particle is lost. The range of phases occupied by the cycle. The electrons that enter the second cell are not fully
electrons indicates the amount of bunching. If the phase relativistic yet hence the length of the second cell needs to
range gets smaller as the position increases, then electrons be shorter than that of later cells, so electrons enter the third
are coming closer together and bunching. The diagram in cell at the right phase for acceleration. Electrons that exit
Fig. 3 shows that bunching mostly happened in the first the second cell are fully relativistic, therefore later cells
cell; how far an electron travels before it is lost; and which need no adjustment for the beta.
phases are captured. From electrons emitted across the full The first scan in 1D code is coarse as we want to find a
360° phase range, about half are launched at accelerating global optimum initially, in which four parameters (field
phases. The other half launched decelerating phases, travel length and amplitude of the first two cells) are scanned
backward, and thus become back-streaming electrons. As simultaneously. Once the global optimums are found, we
the field amplitude is high in the first cell, most of the back- performed fine scans near them in 1D code. Lastly, we
streaming electrons cannot catch the next rf cycle, will performed even finer scans in ASTRA to fine-tune param-
continue traveling backward, and finally will be lost by eters. By the end of the fine-tuning, the capture efficiency
hitting the cathode, other parts of the gun, and beam pipe. changes by less than 0.1% between parameters around the
This causes heating and damage to the cathode. optimums.

IV. VELOCITY BUNCHING AND RECAPTURE A. Optimization by using 1D code

In commercial accelerators, the electron source is a The optimization was performed by conducting a 4D
thermionic cathode that generates a continuous electron grid scan, where four parameters of the first two cells are

080401-4
PUSHING THE CAPTURE LIMIT OF THERMIONIC GUN … PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

TABLE II. Selected representative 1D scan results.

Scan parameters Results

Lc1 (mm) Ez;max;c1 (MV/m) Lc2 (mm) Ez;max;c2 (MV/m) Capture (%) KEave (MeV) σ KE (MeV) KEmax (MeV)
Highest capture 16 8 14 35 95.6 4.4 1.2 5.4
Highest KEave 32 30 38 60 56.9 5.9 1.1 6.7
High capture and 19.2 7.4 30 54 90.8 5.2 1.3 6.1
high KEave

scanned. A field profile similar to the one in Fig. 1 was used highest KEave has KEave ¼ 5.9, but the capture is only
as the starting point in optimization. The initial KE used in 56.9%. Scan results around these highest capture and
the scan was 25 keV. The ranges scanned over for each highest KEave are shown in Fig. 5. It shows we cannot
parameter are: the first cell’s length Lc1 from 10 to 46 mm have the highest capture and highest KEave at the same
and amplitude Ez;max;c1 from 4 to 50 MV=m; and the time. However, we can achieve reasonable high capture
second cell’s length Lc2 from 10 to 82 mm and amplitude (> 85%) and high KEave (> 5.2 MeV) at the same, as given
Ez;max;c2 from 30 to 100 MV=m. the Table II (third and fourth columns) and shown in Fig. 6.
The highest capture of 95.6% is achieved and the We used the optimized results in high-capture and high-
parameters are given in the first column of Table II. The KEave cases to generate the optimized field profile shown in
part of the scans around the highest capture are plotted in Fig. 7. The Applegate diagram shown in Fig. 8 and Ez vs z
Fig. 4. In each subfigure, two variables are varied while the relations shown in Fig. 9 are generated by using the
other two parameters are set to the optimum. The capture is optimized field profile given. In Fig. 8, we see that capture
more sensitive to the first cells’ parameters than the second is increased significantly over that of the standard commer-
cells’. It is least sensitive to Ez;max;c2 and most sensitive to cial linac shown in Fig. 3. 320° out of 360° phases are
Ez;max;c1 . This information is useful in guiding rf design as captured. The beam is well bunched as well. Bunching
they provide parameter sensitivities and tolerances. begins in the first cell and continues in the second cell until at
In the optimization process, one has to keep in mind that the exit of the second cell 320° of the launch phases are
the goal of the optimization is not solely to achieve the compressed to 105° at the exit. This is a threefold
highest capture, but rather to achieve high capture with a compression.
compact structure. In other words, the capture needs to be The most novel aspect of using this field profile is the
reasonably high and the linac cannot be long. To keep the recapture of all the back-streaming electrons in the first cell,
linac compact and simple, we would like to limit the as shown in Figs. 8 and 9. As the field amplitude is low in
number of cells to 6. The beam should have an energy peak the first cell, back-streaming electrons were able to catch
around 6 MeV, so we need at least four accelerating cells. the next rf cycle and get accelerated. The only losses are
Consequently, we can only use two cells for capturing and from a few back-streaming electrons from the second cell.
bunching. As the second cell has a high field amplitude, these back-
Selected representative 1D scan results are given in streaming electrons have a larger KE, and thus cannot be
Table II. While the highest-capture case has a capture of recaptured.
95.6%, the average KE KEave is only 4.4 MeV and one In addition to increasing the capture efficiency, the
would need to add extra cells to reach around 6 MeV. The output energy spread is also reduced as a result of the

FIG. 4. 1D code coarse scan results.

080401-5
SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

FIG. 8. Applegate diagram generated with the optimized field

FIG. 5. 1D code coarse scan results around highest capture and profile of Fig. 7.
highest KEave .

FIG. 6. 1D code fine scan results around high capture and high
KEave .

FIG. 9. Recapture of the back-streaming electrons in the first

cell. The dashed black line is the optimized field profile. The solid
lines are the fields seen by the electrons launched at phases
indicated in the parentheses.

improved bunching. The electrons are tracked through the

optimized field shown in Fig. 7 and results are given in
Fig. 10. The maximum of the Ez of all the cells is set to
54 MV=m, except for the first cell, which is set to
7.5 MV=m. The lengths of the first and second cells are
19 and 33 mm, respectively, and other cells are 50 mm
long. The captured phases are 325° out of the 360° launch
phases—which gives a capture efficiency of about 90%. A
simulation performed with more electrons yielded a more
FIG. 7. Optimized field profile from the 1D tracking code precise estimate of 92.5% capture. As can be seen, only a
optimization. few electrons launched at early and late phases are lost. The

080401-6
PUSHING THE CAPTURE LIMIT OF THERMIONIC GUN … PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

FIG. 11. Comparison of 1D tracking code and ASTRA tracking

results at the exit of the linac. The solid lines are KE spectrum and
the dots are KE as a function of launch phases.

The exit KE spectrum and KE as a function of the launch

phase, obtained by simulations in ASTRA and the 1D
tracking code, are shown in Fig. 11. As can be seen, the
two codes agree well. ASTRA simulations were performed
both with and without space charge, and no significant
difference was observed, thus indicating that space charge
is not dominant. Both codes captured and rejected similar
launch phases and produced a similar KE spectrum.

FIG. 10. 1D tracking results with optimized field profile: (a) exit C. Fine-tuning of linac parameters in ASTRA
phase vs launch phase and (b) exit KE vs launch phase. The inset
of (a) is given to show a larger range of exit phases. The y The next step is fine-tuning the linac parameters in
projection of the curves shows the KE spectrum (in arbitrary ASTRA. As with the optimization using the 1D code, the
units) at the corresponding cell exits. four parameters scanned are (a) the field amplitude of
the first cell Ez;max;c1 ; (b) the length of the first cell Lc1 ; (c) the
length of the second cell Lc2 ; and (d) the field amplitude of
bunching is greatly improved compared to the existing the second cell Ez;max;c2 . Based on the global optimal results
commercial linac, as most of the exit phases after the third of 1D code, the new scans are conducted in much smaller
cell are within the 60° to 120° range. Figure 10(b) shows steps for fine-tuning. As changing later parameters would
that the exit KE spread is smaller compared to the existing have an impact on the earlier parameters, multiple rounds of
linac. It is also shown that there is little acceleration gain in scans are performed, where the new scans were performed
the first cell, rather it is used solely for bunching. with optimum results of earlier scans to make sure the final
results are optimum for all the parameters simultaneously.
The results are shown in Fig. 12. The optimal values were
B. Verification of optimized field in ASTRA found to be Lc1 ¼ 15 mm, Lc2 ¼ 31 mm, Ez;max;c1 ¼
The results of the 1D tracking code were then bench- 8 MV=m, and Ez;max;c2 ¼ 54 MV=m. The amount of cap-
marked using ASTRA and the results showed good agree- ture was found to be more sensitive to the parameters of the
ment. In ASTRA simulations, 25 keV electrons were tracked first cell than of the second cell. The capture decreases
through the optimized field profile shown in Fig. 7. To strongly when Ez;max;c1 is greater than 8 MV=m.
create a 100 mA dc beam, electrons with a total charge of Note that in earlier simulations, the longitudinal position
0.1 nC are uniformly generated over a 1 ns time period, of the initial beam was at the inner wall of the cavity (noted
which covers three rf periods. The capture efficiency from as z ¼ 0), which is not realistic. In an actual cavity, there
the ASTRA simulations is 92%, which is close to the results needs to be at least a 1 cm distance between cavity entrance
from using the 1D code. In these initial simulations, the and initial beam position to account for the cavity wall and
cavity apertures are not included because we want to cathode flange thicknesses. Therefore, in the fine-tuning
compare and verify the results of the 1D tracking code process, we shifted the initial beam position from z ¼ 0 to
with ASTRA. Further simulations using ASTRA including the z ¼ −1 cm and it caused a slight decrease of the capture to
apertures are described in later sections. 87.4%. However, the fine-tuning improved this to over

080401-7
SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

point to the right to minimize the required input power,

giving a shunt impedance of just under 85 MΩ=m.
Our coupling cells have a racetrack cross section to
minimize the transverse size while keeping the capacitive
gap large enough (11 mm) to reduce the sensitivity to
mechanical tolerances. The slot geometry was optimized to
provide a 3.5% coupling factor, to reduce sensitivity
to tolerances, while keeping to peak magnetic flux density
to 214 mT at the design gradient.
The 1D cavity study showed that we need to have a short
first cell with a low accelerating field which is difficult to
realize in a real cavity. Studying the field minimization is
complex as it is difficult to work out the frequency of each
cell individually with side coupled cells and accelerating
cells. To study minimizing the field in one cell while
maintaining field flatness in the other cells, we performed a
FIG. 12. Results of fine-tuning the cavity parameters using study of a five cell cavity where the coupling cells are
ASTRA. identical to the main cells. We initially start with the cavity
tuned for field flatness then modify individual cells. First, we
89%. It can, however, be improved further to over 90% by studied varying the radius of the first (accelerating) cell and
optimizing the initial beam parameter as we will show in the second (coupling) cell. Varying only the first cell does
Sec. VII. indeed vary the field in that cell but at the expense of creating
a finite field amplitude in the coupling cells which would be
V. RF CAVITY DESIGN problematic in a real cavity due to the risk of multipactor and
a need to minimize peak fields in the compact side coupling
In order to keep manufacturing tolerances achievable cavity shape. Varying only the second cell has no effect as
while keeping the shunt impedance high, we have chosen a there are no fields in the coupling cells if all other cells are
reentrant side coupled π=2 mode cavity geometry. Due to tuned correctly, hence we next investigated varying the
the complexity of the behavior for some of the figures of frequency of both the first and second cells. This small
merit in these multidimensional parameter spaces, we detuning of the first cell then allows the amplitude in the first
concluded that the use of multiobjective genetic algorithms cell to be varied by changing the frequency of the second cell.
(MOGA) optimization and the Pareto plots [31] to select While this does indeed minimize the field in the coupling
the best combinations from a large set of variations of the cells it was found to be impossible to have a lower field in the
geometry was the best approach. Figure 13 presents first cell without some field in the second cell. The reason for
the results of a large number of simulations performed this is down to the reason that a π=2 has zero fields in the
in the vast parameter space, highlighting the benefits of coupling cells. The accelerating cells on either side of a side
using the Pareto fronts. As previously discussed we require coupling cell couple fields into the coupling cell that are
an aperture radius of 5 mm for high beam capture. If we 180 degrees different in phase and hence cancel each other
scale the results to a six cell cavity capable of achieving out. To have zero fields in a coupling cell with a different field
6 MeV accelerating voltage with a 4 MW input power while amplitude in the cells, either side requires that the coupling
constraining the peak surface electric field to 100 MV=m between each cell is modified inversely proportional to the
we obtain a narrow parameter space of geometries that meet field amplitude. We achieve this in our model cavity by
our requirements. We have chosen the furthest acceptable having a larger aperture between cells 1 and 2 than the
aperture between cells 2 and 3. It is found that this indeed
does allow us to vary the field in the first cell to any amplitude
without inducing a field in the coupling cells as long as the
aperture is varied to match the coupling.
Our next issue is the need to have a short field in the first
cell. Initially, we can reduce the cell length and the nose-
cone gap to shorten the field profile. However, it is found
this is limited to around 10 mm with a 5 mm aperture
radius. Minimizing the gap also has the added advantage of
reducing the accelerating field in that cell. This is due to the
decay of the evanescent field in the input beam pipe, hence
FIG. 13. Pareto front for a middle cell of a periodic structure as to minimize the length of the field profile we need to reduce
a function of aperture radius. the aperture of the input beam pipe. We find to minimize

080401-8
PUSHING THE CAPTURE LIMIT OF THERMIONIC GUN … PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

the field profile to the length given in the 1D studies we To feed the rf power into the linac, a waveguide coupler
require a 3 mm aperture radius on the input beam pipe. This was integrated to the third cell, shown in Fig. 14(b). The
is not dependant on the aperture radius between the cell and third cell was chosen as the field levels in the first two cells
the adjacent cell and hence this can be kept at 5 mm to are critical and hence to reduce variation the coupler is
minimize beam loss. placed close to these cells. It is envisioned that this rf
The results of these studies were translated into a side- coupler will have a side vacuum port (not drawn), which is
coupled cavity with six accelerating cells and five coupling similar to the setup of the Ref. [32]. The vacuum port will
cells. The input beam pipe has an aperture radius of 3 mm split into two ports: one used for rough pumping and the
while all other aperture radii are 5 mm. The coupling slots other for an ion pump. The rf coupler will be separated from
between the accelerating cells and coupling cells were each the waveguide by an rf vacuum window.
optimized to minimize the field in the coupling cells. This
proved difficult in practice due to the complexity of the VI. BEAM DYNAMICS SIMULATIONS WITH
model, the need to retune both cells when the coupling slot REALISTIC CAVITY MODEL
is modified, and the fact that the side coupled cells are Beam dynamics simulations of the realistic cavity model
highly reentrant and hence a small stored energy results in a are implemented by taking the field profile of the entire
large electric field. Despite this, it was possible to minimize cavity from CST, given by the red curve in Fig. 15, and
the field in the coupling cells to 6.5 MV=m at the design using this as input into ASTRA. Apertures (cavity irises) and
gradient. Several iterations were performed with the ASTRA space charge are also included in the simulations. The field
simulations to optimize the field profile, as described in the amplitude is scaled up so the maximum longitudinal
following section. For the final design, we require a peak electrical field Ez;max is 54 MV=m. The field from CST
input power of 2.65 MW to reach the design goal of has smaller field amplitudes in some cells than the idealized
54 MV=m maximum on-axis electric field. This gives peak field profile, which lowers the final KE. Several iterations
surface fields of 89 MV=m and 214 mT for the electric and of slightly modifying the CST model and tracking in ASTRA
magnetic fields respectively, which are well below the were performed to optimize the final cell lengths and field
maximum limits. The electric field in the final cavity design amplitudes. The resultant cell lengths are mostly the same
is shown in Fig. 14. as they were for the 1D model. In the ASTRA simulations,
the beam was started at −1 cm as the minimum required
distance between cavity inner wall and cathode exit is 1 cm.
The linac aperture is 5 mm in diameter, and the iris
thicknesses are also 5 mm. The initial beam longitudinal
profile is created as a 1 ns long flat distribution with 0.1 nC
charge, which is equivalent to 100 mA quasi-dc beam
covering more than three rf cycles. The other initial beam
parameters are given in Table III.
Due to the differences between the optimal field profile
and the field profile simulated in CST, particularly in terms
of field flatness, the capture efficiency is slightly less than
optimal. Scaling the field profile from CST to a max Ez;max

FIG. 14. The final cavity geometry showing the electric field
map at 1 J of stored energy. Part (a) features side coupled cells FIG. 15. Field profile of entire cavity generated in CST
and (b) features rf coupler. simulation.

080401-9
SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

TABLE III. Initial electron beam parameters. TABLE V. Capture efficiency for different Ez;max with opti-
mized initial beam parameters.
Parameter Unit Value
Capture (%)
Beam current mA 100
Pulse length ns 1 Ez;max (MV/m) KE ¼ 21.5 keV KE ¼ 25 keV
Initial beam position mm −10
54 88.82 88.27
rms transverse beam size mm 0.2
55 89.23 88.67
Correlated divergence mrad −12
58 90.26 89.76
Transverse emittance π · mm · mrad 0.158
64 84.32 91.37
Average KE keV 25
66 82.18 87.60
KE energy spread eV 6.7

parameters. However, the initial beam parameters can also

TABLE IV. Capture efficiency for different Ez;max . be tuned to increase the amount of capture. Further ASTRA
simulations were performed scanning the following param-
Ez;max (MV/m) 54 55 58 64 66 eters in order: transverse rms beam size, beam divergence,
Capture (%) 86.0 86.6 88.1 90.5 86.6 KE, and emittance. The results are given in Fig. 16. The
Back-streaming loss (%) 9.4 9.1 8.4 7.1 9.3 highest capture increased from 86% to 89% when the initial
Aperture loss (%) 4.6 4.3 3.5 2.4 4.1 rms beam size σ x;y is 0.3 mm. The correlated divergence
scan showed 0.3 mm and −13 mrad beam has the highest
captures of 88.30%. A lower KE of 21.5 keV pushed the
of 54 MV=m, the capture is 86%, with loss due to back- capture to 88.82%. Although the ASTRA scan shows a
streaming of 9.4%, and loss at apertures of 4.6%. However, smaller emittance increases the capture, generating a beam
when the Ez;max is increased to 64 MV=m, the capture is with emittance smaller than 0.1π mm mrad may not be
90.5% as shown in Table. IV. For radiotherapy applications, realistic. Simulations of a thermionic gun show that
we want to follow the standard for commercial linac emittance of around 0.1π mm mrad should be achievable.
operations and limit the maximum peak surface field, Epk , Overall, if the initial beam parameters are optimized, we
to well below 100 MV=m to reduce the risk of breakdowns. can increase the capture by 3% to 89%.
In our cavity design, we estimate Ez;max =Epk ¼ 0.607. This The capture can be increased further by combining the
gives Epk of 89 MV=m for Ez;max of 54 MV=m, which is optimized initial beam parameters (0.3 mm and −13 mrad)
well below the industrial safe operation limit. with a higher Ez;max, as shown in Table. V. The emittance
was set to a realistic achievable value of 0.158π mm mrad.
The scan was performed for beams with initial KEs of 21.5
VII. OPTIMIZING INITIAL BEAM PARAMETERS and 25 keV. For commercial applications, safe operation is
So far we have optimized capture efficiency by varying of paramount importance and the linac should be operated
the cavity parameters but used fixed initial electron beam with lower Ez;max, in the range of 54–55 MV=m. This gives
capture is around 89% for an initial KE of 21.5 keV.
Scientific research linacs, however, are often run at a higher
gradient compared to commercial counterparts as there are
sufficient monitoring, safety interlocks, and diagnostic
systems, and the cavities manufactured with higher quality.
Thus at the higher Ez;max of 65 MV=m, a capture of 91.4%
can be achieved with a beam with an initial KE of 25 keV.

VIII. DISCUSSION
In Ref. [33], we see it is difficult to achieve compactness
and high capture at the same time. While it is technically
possible to achieve above 90% capture by using long
bunchers (> 10 cells) that are operated under low rf
powers, the beam would experience other issues like space
charge (as the beam is accelerated slowly) and rf defocus-
ing effects, which would require external focusing.
FIG. 16. ASTRA simulations showing scans of the initial beam Besides, these long bunchers are low-β structures, which
parameters: (a) transverse rms beam size; (b) correlated diver- are very low in rf power consumption efficiency. All these
gence; (c) kinetic energy; (d) emittance. cons will render such cavities with long bunchers nonviable

080401-10
PUSHING THE CAPTURE LIMIT OF THERMIONIC GUN … PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

TABLE VI. Comparison of existing linacs and designs [33] to novel high capture linac.

Linacs Mean/Peak energy (MeV) Length (m) Gradient (MV/m) Capture (%)
S-band with prebuncher and 9.15-cm drift 6.8=NA 0.7 9.7 77
S-band TW option (a) 7.8=9.0 1.18 6.6 66
S-band TW option (b) 9.0=9.3 1.25 7.2 65
S-band SW side coupled medical linac 6.5=7.4 0.33 19.7 30
X-band SW side coupled medical linac 3.6=5.3 0.22 16.4 30
Ku-band split linac 0.14=0.18 0.09 1.6 27
C-band TW with amplitude ¼ 2 4.5=6.0 0.45 10.0 40
C-band TW with amplitude ¼ 3 5.3=6.0 0.32 16.6 43
C-band TW with amplitude ¼ 4 5.4=6.0 0.31 17.4 51
Novel S-band SW high capture linac 6.0=6.4 0.27 22.2 90

for commercial applications, and hence they are only used performance and lifetime. A new linac based on this design
in scientific facilities. Liu et al. [34] also designed an will be manufactured and tested in due course.
S-band traveling wave (TW) linac with 90% capture
efficiency, but the cavity uses 59 cells over 2 m to reach ACKNOWLEDGMENTS
10 MeV energy. This corresponds to 5 MV=m gradient,
The following project was supported by STFC (GCRF)
which is less than 14 of the design. Yurov et al. [35] were
Grants No. ST/S002081/1 and No. ST/S001190/1. The
able to design a 1–3 MeV cw linac with a capture efficiency author would like to thank the staff at CERN and the
of 50%, which is just slightly higher than existing industrial International Cancer Expert Corps (ICEC) for support and
linacs. advice during this project, as well as Taofeeq Ige, National
We summarized the linacs and linac designs in Ref. [33] Hospital Abuja, Hubert Foy, Africsis, and Surbhi Grover,
and compared them to our design as shown in Table VI. We Hospital of University of Pennsylvania, for providing
find our linac design is not only the most compact one but information on requirements for radiotherapy linacs in
has the highest capture as well. Unlike other high capture Africa and encouragement on this project.
designs, our design does not require extra components and
does not add complexity to the system, yet the beam energy
spread and size are kept small.
[1] M. Martins and T. Silva, Radiat. Phys. Chem. 95, 78 (2014).
IX. CONCLUSION [2] T. A. Ige, A. Jenkins, G. Burt, D. Angal-Kalinin, P.
McIntosh, C. N. Coleman, D. A. Pistenmaa, D. O’Brien,
Increasing the average lifetime of the key components of and M. Dosanjh, Clinical Oncology 31, 352 (2019).
commercial medical and security linacs is one of the main [3] S. Korenev, Radiat. Phys. Chem. 71, 537 (2004).
drivers for reducing linac downtime. One component [4] C. J. Karzmark, C. S. Nunan, and E. Tanabe, Medical
known to regularly fail is the electron gun [2]. To address Electron Accelerators (McGraw-Hill, New York, 1993).
this we have designed an S-band thermionic gun-based [5] C. Hernandez-Garcia, P. G. O’Shea, and M. L. Stutzman,
Phys. Today 61, No. 2, 44 (2008).
linac with a high capture efficiency of 90%, achieved using
[6] P. Hawkes and E. Kasper, Principles of Electron Optics
low energy velocity bunching with the inclusion of a low (Academic Press, New York, 2017).
gradient short bunching cell at the entrance of the linac. As [7] C. McKee and M. M. John, Nucl. Instrum. Methods Phys.
back-streaming and aperture losses are reduced by more Res., Sect. A 304, 386 (1991).
than 80% compared to existing standard linacs, the back- [8] M. Bakr, R. Kinjo, Y. W. Choi, M. Omer, K. Yoshida, S.
bombardment effect is reduced, beam quality will be Ueda, M. Takasaki, K. Ishida, N. Kimura, T. Sonobe, T.
improved, and the cathode can be operated at a lower Kii, K. Masuda, H. Ohgaki, and H. Zen, Phys. Rev. ST
current, therefore the cathode lifetime will increase. Accel. Beams 14, 060708 (2011).
Additionally, as there is less parasitic radiation, the linac [9] Y. Huang and J. Xie, in Conference Record of the 1991
will need less radiation shielding reducing the space around IEEE Particle Accelerator Conference (IEEE, New York,
1991), Vol. 4, p. 2017, https://doi.org/10.1109/
the linac on the gantry. The linac design can be easily
PAC.1991.164855.
implemented to existing commercial and research therm- [10] T. Kii, T. Yamaguchi, R. Ikeda, Z.-W. Dong, K. Masuda, H.
ionic gun linacs without extra parts and cost, which in the Toku, K. Yoshikawa, and T. Yamazaki, Nucl. Instrum.
long term can reduce operation and maintenance costs. This Methods Phys. Res., Sect. A 475, 588 (2001).
cavity design has a great potential to be implemented in [11] J. M. D. Kowalczyk and J. M. J. Madey, Phys. Rev. ST
commercial and research linacs to improve overall machine Accel. Beams 17, 120402 (2014).

080401-11
SADIQ SETINIYAZ et al. PHYS. REV. ACCEL. BEAMS 24, 080401 (2021)

[12] T. Kii, K. Masuda, S. Amazaki, T. Horii, H. Toku, K. [25] J. S. Aubin, S. Steciw, C. Kirkby, and B. G. Fallone, Med.
Yoshikawa, H. Ohgaki, and T. Yamazaki, Nucl. Instrum. Phys. 37, 2279 (2010).
Methods Phys. Res., Sect. A 483, 310 (2002). [26] D. Baillie, B. G. Fallone, and S. Steciw, Med. Phys. 44,
[13] J. P. Edelen, S. G. Biedron, J. R. Harris, J. W. Lewellen, 2124 (2017).
and S. V. Milton, IEEE Trans. Nucl. Sci. 61, 830 (2014). [27] K. Floettmann, ASTRA—A space charge tracking
[14] C. McKee and J. M. Madey, Nucl. Instrum. Methods Phys. algorithm, http://www.desy.de/∼mpyflo/, 2014.
Res., Sect. A 296, 716 (1990). [28] G. Pöplau, U. van Rienen, and K. Flottmann, in Proc.
[15] M. Borland, A high-brightness thermionic microwave EPAC 2006, Edinburgh, UK (2006), Vol. 060626, p. 2203.
electron gun, thesis/dissertation, 1991, https://doi.org/ [29] S. Rimjaem, E. Kongmon, M. W. Rhodes, J. Saisut, and C.
10.2172/10164508. Thongbai, Nucl. Instrum. Methods Phys. Res., Sect. B 406,
[16] A. M. Capece, J. E. Polk, and J. E. Shepherd, IEEE Trans. 233 (2017).
Plasma Sci. 43, 3249 (2015). [30] A. Y. Baikov, C. Marrelli, and I. Syratchev, IEEE Trans.
[17] B. Whelan, S. Gierman, L. Holloway, J. Schmerge, P. Electron Devices 62, 3406 (2015).
Keall, and R. Fahrig, Med. Phys. 43, 1285 (2016). [31] T. Luo, H. Feng, D. Filippetto, M. Johnson, A. Lambert, D.
[18] M. Bakr and H. Ohgaki, IEEE Trans. Electron Devices 65, Li, C. Mitchell, F. Sannibale, J. Staples, S. Virostek, and R.
5053 (2018). Wells, Nucl. Instrum. Methods Phys. Res., Sect. A 940, 12
[19] P. Palluel and A. M. Shroff, J. Appl. Phys. 51, 2894 (1980). (2019).
[20] W. B. Nottingham, Thermionic Emission (MIT, Cambridge, [32] K. Yaqub, S. Akbar, S. Javeed, F. Siddique, A. Rahman, N.
MA, 1956). Nisar, M. F. Khalid, N. U. Saqib, and M. A. Khan, Vacuum
[21] W. P. Swanson, Radiological Safety Aspects of the Oper- 187, 110151 (2021).
ation of Electron Linear Accelerators (International [33] S. V. Kutsaev, Eur. Phys. J. Plus 136, 446 (2021).
Atomic Energy Agency, Vienna, 1979). [34] H. Liu, X. Wang, and S. Fu, Chin. Phys. C 30, 581
[22] E. A. Knapp, B. Knapp, and J. Potter, Rev. Sci. Instrum. 39, (2006), http://hepnp.ihep.ac.cn/article/id/2b673073-f114-
979 (1968). 4a4a-a247-7ae9b8b7af3c.
[23] V. A. Vaguine, Rev. Sci. Instrum. 48, 1658 (1977). [35] D. S. Yurov, A. S. Alimov, B. S. Ishkhanov, and V. I.
[24] T. Wangler, RF Linear Accelerators (Wiley, New York, Shvedunov, Phys. Rev. Accel. Beams 20, 044702 (2017).
2008).

080401-12