RF Theory and Design - Notes

Jeremiah Holzbauer Ph.D.

USPAS - Grand Rapids

June 2012

1 Introduction
This document is a short summary of the theory covered in the USPAS class Applied
Electromagnetism for Accelerator Design. This is a living document, and will be expanded
at a future date, please excuse small errors and sudden transitions. This document is
Version 1.

2 Resonator Theory
This section will provide insights for use in further sections by producing a generic formal-
ism to describe resonators with the explicit goal of treating resonating cavities with the
same mathematical treatment. Starting from the simple harmonic oscillator, damping and
driving terms will be added, and their effects derived. The treatment of electromagnetic
resonators by this formalism with then be justified, and a special case of interest will be
presented.

2.1 The Ideal Harmonic Oscillator

The ideal, linear oscillator is the most basic starting point for solving many problems in
physics. This is especially true in accelerator physics because so many of the phenomena of
accelerator physics can be treated as purely classical and having few confounding effects.
The harmonic oscillator, without damping, has the form

d2 x
+ ω02 x = 0 (1)
dt2
where x(t) is the oscillating position at time t, and ω0 is the oscillation frequency. The
general solution for this is characterized by an amplitude A and a phase ϕ

x(t) = A cos (ω0 t + ϕ). (2)

1
2.2 The Driven, Damped Harmonic Oscillator
For this to be a useful model we must consider damping. The form of damping that is of
interest to us (because it is the form of the losses in a resonator), is damping proportional
to the change in “position”, which has a form of

d2 x dx
+γ + ω02 x = 0, (3)
dt2 dt
where the damping coefficient γ has the dimension of frequency. This type of equation has
different forms of solution depending on the strength of the damping. If we choose the
damping to be weak, the general solution to this equation has the form

x(t) = Ae−γ1 t cos (ω1 t + ϕ) (4)

where √
γ2
ω1 = ω0 1− (5)
4ω02
and
1
γ1 = γ. (6)
2
Note that adding damping shifts the resonator frequency based on how strong the damp-
ing is. From this we can define a “quality factor” Q which is related to the rate at which
the resonator loses energy. Q is defined by the equation Q = ω0 /2γ with weak damping
characterized by Q ≫ 1. For a typical superconducting cavity, Q is generally ∼ 5 × 109 ,
justifying the assumption of weak damping.

The effects of a driving term must also be considered. For an arbitrary driving term of
the form f (t), the differential equation becomes

d2 x dx
2
+γ + ω02 x = f (t), (7)
dt dt
with the solution (for γ = 0)
∫
ẋ0 1 t ( ( ))
x(t) = x0 cos ω0 t + sin ω0 t + sin ω0 t − t′ f (t′ )dt′ (8)
ω0 ω0 0

where x0 and ẋ0 are the initial position and velocity. The third term of this solution gives
the contribution from the driving term, and it is worthwhile to notice that a harmonic
driving term with frequency equal to the resonant frequency of the oscillator will produce
the largest oscillations, as expected.

2
2.3 Oscillator Behavior Near Resonance
It is most useful to examine the response of the weakly damped oscillator to driving near
its resonant frequency because this is the desired case for resonator operation. Assuming
a driving term that is purely harmonic (f (t) = f0 cos ωt) and ignoring transient behav-
ior yields a solution with a relatively simple form. These assumptions are very justified
in almost every accelerator application; most changes made to the driving term of the
cavity have a time scale much larger than the RF period (which is usually on the order
of nanoseconds). This type of differential equation is solved quite simply by assuming a
complex solution and writing

d2 Ξ dΞ
+γ + ω02 Ξ = f0 e−iωt (9)
dt2 dt
with x(t) = ℜΞ(t). We are seeking a solution of the form Ξ(t) = Ξ0 e−iωt where Ξ0 is also a
complex number of the form Ξ0 = |Ξ0 |eiϕ . Solving for the real variable of interest x gives
x(t) = ℜΞ(t) = ℜ(|Ξ0 |e−iωt+iϕ ) = |Ξ0 | cos (ωt − ϕ). Plugging this form of the solution into
the differential equation gives
( 2 )
−ω − iωγ + ω02 Ξ0 = f0 (10)

simplifying to
f0
Ξ0 = (11)
ω02 − ω 2 − iωγ
with squared amplitude of
f02
|Ξ0 |2 = . (12)
(ω02 − ω 2 )2 + ω 2 γ 2
The resulting behavior can be see in Figure 1 for a variety of damping coefficients γ [1].
Again, it is easy to see that the maximum response will be shifted slightly depending on
the strength of the damping. This shift can be neglected for γ ≪ ω0 , and as demonstrated
earlier, this is a very good approximation for superconducting resonators. Another impor-
tant feature of these curves is the characteristic width of each curve. This width (∆ω),
defined as the width at the level that is 3 dB below the maximum response, is equal to 2γ
where γ is the damping parameter. An alternative and equivalent definition of the Quality
Factor is Q = ω/∆ω.
The phase response of Ξ is also of interest. This phase ϕ can be interpreted as the
difference in phase between the driving term and the response of the resonator, and is an
important quantity for resonator control.
( ) ( )
ℑΞ0 ωγ ω 1 γ
tan (ϕ) = = 2 = ( )2 (13)
ℜΞ0 ω0 − ω 2 ω0 ω0
1 − ωω0

3
Figure 1: Resonant curves for damped, driven harmonic oscillator with γ = 0.5, 0.2, 0.1, 0

A plot of ϕ can be seen in Figure 2. The most important feature of this behavior is the
nearly linear region near resonance. Most cavity control systems treat the cavity response
as linear and must operate in this region to remain stable.

2.4 Special Case: The Duffing Equation

As we shall see, certain non-linear behavior in superconducting resonators in operation can
be well modeled by the Duffing Equation. The Duffing Equation adds a weakly non-linear
term to the restoring force such that

d2 x dx
+γ + ω02 x + αx3 = f0 cos ωt. (14)
dt2 dt
For the purposes of this application, it can be assumed both the damping, driving, and
non-linear terms are small compared to the frequency ω0 . Additionally, we will only look
for solutions where ω ≃ ω0 . Using the standard van der Pol transformation [2], seen in
Figure 3, we transform into a rotating coordinate frame. Using the transformations
ẋ
u = x cos ωt − sin ωt (15)
ω

4
Figure 2: Oscillator phase shift compared to driving term versus detuning. Note the nearly
linear region near resonance.

and
ẋ
v = −x sin ωt − cos ωt, (16)
ω
we arrive at the following differential equations for u and v:

1
u̇ = [−(ω 2 − ω02 )(u cos ωt − v sin ωt) − ωγ(u sin ωt + v cos ωt)
ω
+ α(u cos ωt − v sin ωt)3 − f0 cosωt] sin ωt (17)

1
v̇ = [−(ω 2 − ω02 )(u cos ωt − v sin ωt) − ωγ(u sin ωt + v cos ωt)
ω
+ α(u cos ωt − v sin ωt)3 − f0 cosωt] cos ωt. (18)

Because we are assuming small non-linearities and constant frequency (ω), we are only
interested in the average behavior of these functions. Averaging over a period of 2π/ω, we
get ( )
1 3
u̇ = −ωγu + (ω − ω0 )v − α(u + v )v
2 2 2 2
(19)
2ω 4
and ( )
1 3
v̇ = ωγv − (ω − ω0 )u + α(u + v )u − f0 .
2 2 2 2
(20)
2ω 4

5
 Figure 3: Van der Pol coordinate transformation into a frame rotating at ω.

In reality, these are not the values of interest for our resonator. For √ more meaningful
results, we again transform these quantities into the magnitude (r = u2 + v 2 ) and phase
(ϕ = arctan (v/u)) in the rotating frame. This results in the following differential equations:
1
ṙ = (−ωγr − f0 sin ϕ) (21)
2ω
and ( )
1 3 3
ϕ̇ = −(ω − ω0 )r + αr − f0 cos ϕ .
2 2
(22)
2ω 4
Again, we can simplify by assuming a steady state solution: ṙ = ϕ̇ = 0. The resulting
amplitude and phase plots can be seen in Figures 4 and 5, and the non-linear response
is evident. Because of this detuning, both the amplitude and phase become multi-valued
for certain ranges of detuning, and if the effect is severe enough, controlling the resonator
behavior can become prohibitively challenging.

6
Figure 4: Oscillator amplitude versus detuning. Circles represent α = −0.002, crosses
represent α = 0.

2.5 Electromagnetic Fields as a Harmonic Oscillator

To this point, I have focused on the formalism for a generic harmonic oscillator with damp-
ing and driving terms. Occasionally, I have motivated certain approximations and simplifi-
cations as relevant to accelerating resonators without proof, but now I wish to demonstrate
the applicability of this formalism to electromagnetic resonators in preparation for the rest
of the document. For a (much) more rigorous treatment of this material, the reader is
directed to Chapter 7 of J.D. Jackson’s excellent book “Classical Electrodynamics” [3].
Maxwell’s Equations govern all electromagnetic behavior. In free space, these equations
are

∇·B =0 (23)
∂B
∇×E+ =0 (24)
∂t
∇·D =0 (25)
∂D
∇×H − =0 (26)
∂t

7
Figure 5: Oscillator phase versus detuning. Circles represent α = −0.002, crosses represent
α = 0.

where E is the electric field, B is the magnetic field, and D (equal to ϵ0 E in this case) is
the electric displacement field. Assuming a harmonic time dependance of the form e−iωt
gives

∇·B =0 (27)
∇ × E − iωB = 0 (28)
∇·D =0 (29)
∇ × H + iωD = 0 . (30)

From this form, we can construct an arbitrary solution using Fourier superposition. As-
suming that we are in a region where there are no losses (µ and ϵ are real and positive),
Maxwell’s Equations reduce to

∇ × E − iωB = 0 (31)
∇ × B + iωµϵE = 0 (32)

8
where the divergence equations can be found from taking the divergences of these two
equations. Combining these two equations we get the Helmholtz wave equations:

∇2 E + µϵω 2 E = 0 (33)
∇ B + µϵω B = 0 .
2 2
(34)
√
A solution to equations of this form is a plane wave of the form eikx−iωt , where k = µϵω is
the wave number. While solutions of this form are discussed many places at great length,
the harmonic behavior allows us to use the same differential formulations to treat the
resonator. For the remainder of this document, we will only consider harmonic behavior
in vacuum where ϵ = ϵ0 and µ = µ0 with c2 ϵ0 µ0 = 1.

3 A Simple Electromagnetic Resonator - The Pillbox Cavity

This section will provide a treatment of the most basic resonator shape, the pillbox cavity.
The geometry and electromagnetic mode structure will be motivated and the performance
of this cavity as an accelerating resonator will be quantified. Derivation and description of
generic cavity figures of merit will be provided and derived for this geometry.

3.1 Cylindrical Waveguide Modes

Electromagnetic fields assume a certain form near a perfect conductor. Because a perfect
conductor allows no field penetration into the surface and can sustain any currents required
to expel the field lines, the fields at the surface of the perfect conductor must satisfy:

n̂ × E
⃗ = 0, n̂ · H
⃗ =0 (35)

where n̂ is the conductor surface’s normal vector. Let us consider a cylindrical waveguide of
radius R that is infinite in length. Defining our cylindrical coordinate system in the usual
way, we can use the generalized plane wave solution to the Helmholtz equation derived in
the previous section, we can assume the fields take the form
⃗ x, t) = E(ρ,
E(⃗ ⃗ ϕ)eikz−iωt (36)
⃗ x, t) = H(ρ,
H(⃗ ⃗ ϕ)eikz−iωt . (37)
For the purposes of an accelerating cavity, it is desired to have longitudinal electric fields
with no longitudinal magnetic fields, i.e., Bz = 0. Additionally, we know that the depen-
dence of the fields on ϕ must be periodic, so we assume a dependence of the form eimϕ
where m is an integer. This gives us a form for the electric field of

Ez (ρ, ϕ, z, t) = f (ρ)eikz−iωt−imϕ (38)

9
which we can plug into the wave equation:
( )
1 ∂2z 1 d df m2 ω2
= ∇2
z =⇒ ρ − f+ −k f =0
2
(39)
c2 ∂t2 ρ dρ dρ r2 c2

to get the form of f (ρ). The solution of this equation is given by standard Bessel functions
f (ρ) = E0 Jm (k⊥ ρ) where m is the order of the Bessel function and
1√ 2
k⊥ = ω − c2 k 2 . (40)
c
To satisfy the boundary conditions at ρ = R, Ez must be zero. This requires that k⊥ R
is equal to one of the Bessel zeros, jm,n (where n is an integer indicating the zero), giving
k⊥ = jm,n /R. Eliminating k⊥ from our equations gives our definition of the wave number
k for this geometry:
( )1/2
2
ω 2 jm,n
km,n = ± − 2 . (41)
c2 R
It is of interest to note that this gives a lower bound on the frequency for a traveling
wave solution. For ωjm,n c/R, the wave number becomes imaginary, resulting in a solution
known as an “evanescent” wave. This frequency is known as the “cut-off” frequency, with
larger diameter waveguides able to support lower frequencies. This wave exponentially
decays/increases as it propagates, and is of interest in the study of resonators only for
study of mode propagation into and out of the resonator through beam pipes or other
ports. For the remainder of this section, we will only consider traveling wave solutions.
Now that the Ez has been defined and remembering that Bz = 0, we can use ∇ × E = iωB
and c2 ∇ × B = −iωE to find the remaining components of both fields: Eϕ , Eρ , Bϕ , Bρ .
These component are
ikm,n R ′ ( ρ ) ikm,n z−iωt−imϕ
Eρ = E0 Jm jm,n e (42)
jm,n R

mkm,n R2 ( ρ ) ikm,n z−iωt−imϕ

Eϕ = −E0 2
J m jm,n e (43)
ρjm,n R
mωR2 ( ρ ) ikm,n z−iωt−imϕ
Bρ = E0 2 2 Jm jm,n e (44)
c ρjm,n R
iωR ′ ( ρ ) ikm,n z−iωt−imϕ
Bϕ = E0 2 Jm jm,n e (45)
c jm,n R
′
where Jm is the derivative of the mth order Bessel function. Standard convention designates
these modes as T Mmn for T ransverse M agnetic modes where the index m is the azimuthal
order and index n is the radial order. It is worth noting that the boundary condition

10
Eϕ (ρ = R) = 0 is automatically satisfied because this component has the same radial
dependence as Ez .
This derivation for T M modes is based on the condition that Bz = 0. One can follow
the same approach starting with the condition of Ez = 0 to obtain another family of modes,
T ransverse Electric (T E) modes. As these are not useful for accelerator applications, they
are beyond the scope of this work.

3.2 From Traveling Waves to Standing Waves

The goal of this section is to describe a cavity, not a waveguide. To this end, we will take a
section of waveguide and terminate it with conducting faces at z = 0 and z = L. In order
to satisfy these additional boundary conditions with T M modes, we take two modes with
equal m and n but with opposite values of k. This is equivalent to combining identical
waves traveling in opposite directions. The equations
1 ikm,n z
(e + e−ikm,n z ) = cos (km,n z) (46)
2
1
(km,n eikm,n z − km,n e−ikm,n z ) = ikm,n sin (km,n z) (47)
2
can be plugged into the full field components, and both Eρ and Eϕ gain a factor of
sin (km,n z). This means that the boundary condition requiring these components to be
zero at the z = 0 plane is automatically satisfied, and the boundary condition at z = L can
be satisfied by requiring km,n L = lπ where l is an integer index that gives the longitudinal
mode order. It is of interest to note that these new boundary conditions are now also
satisfied for the magnetic fields. The resulting field components are
( ( )
ρ) lπz −iωt−imϕ
Ez = E0 Jm jm,n cos e (48)
R L
( )
lπR ′ ( ρ) lπz −iωt−imϕ
Eρ = −E0 J jm,n sin e (49)
jm,n L m R L
( ( )
imlπR2 ρ) lπz −iωt−imϕ
Eϕ = −E0 2 Jm jm,n sin e (50)
ρjm,n L R L
( ( )
mωR2 ρ) lπz −iωt−imϕ
Bρ = E0 2 2 Jm jm,n cos e (51)
c ρjm,n R L
( )
iωR ′ ( ρ) lπz −iωt−imϕ
Bϕ = E0 2 J jm,n cos e . (52)
c jm,n m R L
Note that this also changes the form of the wave number, resulting in a discrete dispersion
relation instead of a continuous curve:
[( ) ]1/2
2
ωm,n,l lπ 2 jm,n 2
=± + 2 . (53)
c2 L R

11
This shift to a discrete set of modes (conventionally referred to as T Mmnl modes) can be
repeated for T E modes as well.

3.3 The “Fundamental Mode” and its Properties

For use as an accelerating cavity, the T M010 mode (called the “fundamental mode”) gives
the best performance. The justification for this is simple when remembering that our desire
was to have only the accelerating electric field component be non-zero near the beam axis.
The field components for this mode are very simple:
( )
2.405ρ −iωt
Ez = E0 J0 e (54)
R
( )
E0 2.405ρ −iωt+i3π/2
Hϕ = J1 e (55)
η R
where √
µ0
η= u 376.7 Ω (56)
ϵ0
is the impedance of free space. The frequency for this mode does not depend on the length
of the cavity L, reducing to
2.405c
ω010 = . (57)
R
For this mode to successfully accelerate beam, ports must be added for the beam to enter
and exit the cavity. The rest of this section will not include these ports, as they are ex-
tremely challenging to model analytically. This is a perturbation best studied by computer
simulations.

3.4 The Interaction of Conductors and Electromagnetic Fields

There is no material that will not dissipate power when an RF field is applied to its surface.
The losses in a cavity are given by
dP 1
= Rs |H|2 (58)
dA 2
where dP/dA is the dissipated power per unit area and H is the local magnetic field. The
remaining quantity is the surface resistivity. For a normal conducting material, this can
be on the order of mΩ, and for a high quality superconducting surface, this can be on the
order of nΩ. These losses are generated by the surface currents induced by the changing
magnetic fields in the resonator. Remembering that, for an RF cavity,

E = E0 e−iωt (59)

12
and defining the electrical conductivity σ such that
⃗
⃗j = σ E (60)

where ⃗j is the surface current, we can show that, for a good conductor,

∇2 E
⃗ = τn2 E
⃗ (61)
√
where τn2 = iωσµ0 . For a simple geometry, such as an infinite half-plane of conductor
at x > 0 with an applied electric field that is only in the z direction, the solution to this
equation is
−x −ix
Ez = E0 e−τn x = E0 e δ e δ (62)
where
1
δ=√ . (63)
πf µ0 σ
This quantity, δ, is called the “skin depth”, and gives a characteristic length for how deeply
the electromagnetic fields penetrate the conductor. As an example of this, a copper surface
with σ = 6 × 107 S/m at a frequency of 1 GHz has a skin depth of approximately 2 µm.
Also note that as the frequency increases, this skin depth decreases. Because this skin
depth is so small compared to the wavelength of the resonator, it can be ignored in the
treatment of the cavity properties. In the geometry above, the current density and losses
can also be quantified. The current and field of this configuration are

jz = j0 e−τn x , Hy = H0 e−τn x . (64)

Using these, the surface impedance Z0 can be calculated where

E0
Z0 = (65)
I
with I being the current. The losses come from the real part of the surface impedance,
where the imaginary part of this is the reactance. Calculation yields
√
πf µ0 1
Rs = = . (66)
σ σδ
Note that these calculations are done for normal conducting materials.

3.5 Cavity Figures of Merit - Accelerating Voltage

All cavities are judged by a small set of electromagnetic figures of merit. These figures of
merit fall into two different categories: efficiency and scaling. Efficiency figures of merit
are measures of how efficiently the cavity performs and do not scale with stored energy.
The scaling figures of merit are all figures of merit that do scale with the stored energy

13
of the cavity. There are three major scaling figures of merit. First, and perhaps most
important, is the Accelerating Voltage (Vacc ). This is the measure of how much energy a
charged particle passing through a cavity will gain, and it is defined as
∫ +∞ ∫ +∞ ( )
ωz
Vacc = Eacc (z) cos (ωt + ϕ)dz = Eacc (z) cos + ϕ dz (67)
−∞ −∞ βc
where z axis is the beam axis, β = v/c, ϕ is a phase offset that maximizes Vacc and
assuming that β doesn’t change. This last approximation is equivalent to saying that the
energy gained in the cavity is small compared to its totally energy. This approximation
is not valid if a high degree of accuracy is required, say, for particle tracking. Practically,
this can be calculated as
√( )2 (∫ +∞ )2
∫ +∞
Vacc = Eacc sin (ωt)dz + Eacc cos (ωt)dz . (68)
−∞ −∞

For the pillbox cavity, this integral has an analytic solution because the accelerating electric
field is constant in z.
( )

 ∫ +∞ 
 
 ∫ L 
 ω0 L

 
 
 
 sin 2βc
Vacc = 
 Eacc (z)eiωt+iϕ dt

 = E0 

 eiω0 z/βc dz 

 = LE0 ( ) (69)
 −∞   0  ω0 L
2βc

The length of the cavity, at this stage, remains unspecified. An important consideration of
the length is that this type of cavity is often operated in series as can be seen in Figure 6,
with each cavity being π ahead of the previous one in phase. It is obvious to see that it is
desirable to have the time it takes the particle to transit the cavity be equal to the time it
takes the fields to reverse:
L π βλ
= ;L = . (70)
βc ω0 2

3.6 Cavity Figures of Merit - Transit Time Factor

Given a cavity length that is appropriate for the velocity of particle being accelerated, it is
clear that there is still some loss in accelerating efficiency because the fields are varying in
time. The maximum acceleration that could be expected from a cavity of this type would
be ∫ +∞

 

V0 = 
Eacc (z) dz = LE0 . (71)
−∞
The ratio of Vacc to V0 gives the efficiency figure of merit called the Transit Time Factor
(T T F ). For the pillbox cavity with a length of βλ/2, the T T F is
( )
ω0 L ( )
Vacc sin 2βc sin π2 2
TTF = = ( ) = (π) = . (72)
V0 ω0 L
2
π
2βc

14
 Figure 6: An example of pillbox type cavities coupled together.

It is important to note that, in reality, a given length of cavity L gives an optimum particle
velocity βopt . This means that the T T F varies with β for a given cavity geometry:
( ) ( )
πβ
sin ω2βc
0L
sin 2βopt
T T F (β) = ( ) = ( ) . (73)
ω0 L πβopt
2βc 2β

Figure 7 shows this curve for a cavity designed for βopt = 0.50 for 2, 4, and 8 cavities in
series. Notice that as the number of synchronized cavities (“gaps”) increases, the range
of velocities that can be efficiently accelerated shrinks. This is of considerable importance
for an accelerator that has to accelerate a beam with a velocity that changes significantly.
While larger numbers of synchronized gaps mean a higher amount of acceleration per
meter of constructed accelerator, this may require more cavities designed for a variety of
different βopt . Two extreme examples of this are heavy ion and electron accelerators. An
electron is such a light particle that it very rapidly approaches the speed of light, thus only
requiring one resonator geometry. To give an example of how fast this happens, remember
that the International Linear Collider has a resonator designed to give roughly 30 MeV of
acceleration with each resonator having 9 cells. This means that, from rest, an electron
would be going 99%(!!!) the speed of light at the exit of the first resonator (even after
including the reduction in acceleration efficiency when the electron’s speed is mismatched
relative to the optimium beta). For a heavy ion accelerator, the situation is much different.
For a beam of uranium (common in nuclear physics), over 500,000 times as much energy
(almost 2 TeV!!!) is required to reach the same velocity, even in its highest charge state.
Even using the most flexible accelerating resonators (2-gap structures), this would take at

15
Figure 7: The Transit Time Factor (TTF) for 2, 4, and 8 cavities in series with βopt = 0.50.

least 5 different cavity designs to achieve (and it would be ∼ 4 km long!).

3.7 Cavity Figures of Merit - Peak Surface Electric and Magnetic Field
The remaining scaling figures of merit are the peak surface electric and magnetic fields.
For the pillbox cavities, these values

Epk = E0 (74)

E0 E0
Hpk = J1 (1.84) = (75)
η 647 Ω
are located on the end walls of the cavity.

16
3.8 Cavity Figures of Merit - Geometry and Quality Factor
The efficiency with which the resonator stores energy is of principal importance. For
superconducting cavities, this is especially true because all dissipated power is exhausted
into liquid helium. The figure of merit for this efficiency is the Quality Factor:
ω0 U
Q= . (76)
Pd
From our study of the harmonic oscillator, remember that we defined the quality factor as
the ratio of the energy stored in the resonator to the energy lost per radian of oscillation.
Inspection shows that this definition is consistent with the resonator definition. In addition
to the Quality Factor, there is a figure of merit that does not depend on the surface
resistance (Rs ) called the Geometry Factor. This is defined as

ωU
G = Rs Q = ( ) (77)
Pd
Rs

and is solely dependent on the geometry. This is a more useful figure of merit for cavity
design because the Rs depends on many factors outside of the electromagnetic design such
as material purity, surface quality, heat treatment of the cavity, and many more. This
is also a useful figure to compare different geometries and cavities of different frequencies
because it is independent of frequency [4, pp. 43-44]. To calculate the Geometry Factor
for a given geometry, we must first calculate the stored energy (U ) and dissipated power
(Pd ). Remembering that
dPd 1  
H⃗
2

= Rs  
 (78)
dA 2
⃗
 
where  H 
  is the magnetic field at dA, and that the Geometry Factor requires P /R
 d s
the equation for the total dissipated power becomes
∫  
Pd 1 
 ⃗
 2
= H
  dA
 (79)
Rs 2 S

where S includes all RF surfaces in the cavity. Note that we are assuming that Rs is
constant over the surface, which is unlikely to be strictly true. While it is expected that its
value will not vary much over the surface, accurate measurements of Rs and its variation
over the surface of the cavity are extremely difficult. In any case, the variation will not
likely be predictable, and should not affect cavity design. In addition to Pd , we need the
stored energy in the cavity. In general, the time averaged stored energy density in the
cavity is [5] ( )
1 ⃗ 2 + 1 |B|
⃗ 2 .
u= ϵ|E| (80)
2 µ

17
At any given time, the energy is stored in the electric and magnetic fields of the cav-
ity. However, the total energy is fixed, and in this time-averaged form, both electric and
magnetic contributions are equal. This means we can simplify this density to
ϵ 2 1  2
u= E 
 = 
B 
 (81)
2 2µ
which yields ∫ ∫
ϵ0 
 2 1  2
U= E0 
 dV = 
B0 
 dV. (82)
2 V 2µ0 V
This gives us
∫ 
⃗ 
ω0 µ0 V 

H 
2 dV

G = Rs Q = ∫   . (83)
S
H  dA
2

With some calculation, these integrals can be done for the pillbox cavity as described [4,
p. 46]:
πϵ0 E02 2
U= J1 (2.405)LR2 (84)
2
πRs E02 2
Pd = J1 (2.405)R(R + L) (85)
η2
L
ω0 µ0 LR2 2.405L 453 R
G= = η = L
Ω. (86)
2(R2 + RL) 2(R + L) 1+ R
By remembering that we desire
βλ
L= (87)
2
we find that, for a pillbox cavity,
L βπ
= (88)
R 2.405
which gives a Geometry Factor for ALL optimized pillbox cavities of

G = 257β Ω. (89)

As a comparison, the Facility for Rare Isotope Beams (FRIB) calls for a cavity (of a
totally different geometry) designed for β = 0.29. An equivalent pillbox cavity would have
G ≈ 74 Ω with the current FRIB cavity design having G = 76 Ω. While a pillbox cavity is
impractical at these velocities (it would require a cavity that has a diameter over 5 times
its length), this demonstrates the usefulness of this figure of merit for comparison between
cavities.

18
3.9 Cavity Figures of Merit - R/Q
The third major efficiency figure of merit is the R/Q. This is a measure of how efficiently
the cavity transfers its stored energy to the beam passing through it. This is the ratio of
the effective shunt impedance to the cavity’s quality factor, and this is a point that needs
to be stressed. There are many definitions of the shunt impedance depending on context
or even convention, and I will be using the common accelerator conventional definition [4,
p. 47]:
V2
Ra = acc (90)
Pd
in units of Ω per cell. This is further complicated by the fact the the majority of this
document will be spent discussing cavities that, while one “cell”, have two accelerating
gaps. From all this, the definition of the R/Q is
2
Vacc
R/Q = . (91)
ω0 U
One final note is that this is the effective shunt impedance. While I will omit the eff
subscript from R, I will be using Vacc instead of V0 , which is implied by “shunt impedance”
in some circles. For a pillbox cavity, this can be readily calculated
L
R/Q = 150 Ω = 196β Ω. (92)
R

3.10 Pillbox Cavities for Low Velocity Particles

The relationship between R/Q and β indicates that pillbox type cavities would perform
best for high velocity beams, and would perform quite poorly for very low velocity beams.
This can be seen by comparing the performance of a FRIB cavity design for β = 0.29,
a pillbox R/Q ≈ 57 Ω compares poorly to the R/Q = 110 Ω/gap of the current design.
In addition to efficiency, there are mechanical drawbacks to using a pillbox cavity at low
velocity. As has been shown, the optimum length of the accelerating gap for a pillbox
cavity is βλ/2, and I will now posit that what this actually means is that the distance
between adjacent, synchronized gaps should be
( )
1 βλ
Lgap−to−gap = n + (93)
2 2
where n is an integer. Pillbox cavities require very little space between synchronized gaps,
thus for them (and a vast majority of resonators) n will be 1. This gap-to-gap distance
is a more general principle of accelerating resonators, and thus can be used to compare
different geometries. This scaling gives interesting results for pillbox cavities designed for
very low velocity beams. FRIB calls for the first resonator to be optimized for β = 0.041.
With a conservative gap-to-gap distance of 0.1 m, this gives a cavity with f = 61.5 MHz,

19
and a cavity radius of R = 0.534 m. This cavity has very poor mechanical qualities from
such a large ratio of radius to gap size, in addition to the loss in accelerating efficiency due
to the small gap size.

4 Coaxial Resonators
In the previous section, we derived the mode structures and properties of resonators con-
structed from a terminated section of cylindrical waveguide. The scaling of these structures
means they become undesirable for acceleration of very low velocity particles. Coaxial res-
onators originate from the mode structure of coaxial transmission lines. These resonators
offer a practical option for lower velocity particle acceleration because they can achieve the
lower fundamental frequency required for a lower optimum velocity of beam while main-
taining efficiency of acceleration. In payment for this, these cavities have more complex
geometries that require more material, construction, and processing cost and time. Their
mode structures, scaling parameters, and figures of merit will be derived and compared
with the pillbox cavity type, and their use in heavy ion accelerators will be justified.

4.1 Coaxial Transmission Line Geometry

The addition of a cylindrical inner conductor to a cylindrical waveguide significantly alters
the mode structure. First, modes with longitudinal electric fields are no longer supported,
and the modes of interest have no longitudinal magnetic fields, giving these modes their
designation as T EM , or T ransverse Electric and M agnetic. There are two types of termi-
nation that are of interest for this type of geometry, short and open. A short termination is
the same type that was used in the previous section for pillbox cavities, a conducting plane
in the ρ − ϕ plane. This gives a magnetic field maximum and a zero in the electric fields at
the boundary. The other type, an open, is the addition of a perfect magnetic conducting
plane in ρ − ϕ. This provides a zero in the magnetic field and a maximum in the electric
field at the boundary. While this is, technically, non-physical, it is an excellent starting
place for analytical calculations. An easy justification for the contained nature of modes
using an effective open boundary condition can be seen from considering a length of coaxial
line, shorted at one end, and connected to a cylindrical waveguide with the same radius
as the outer conductor at the other end. As long as the frequency of the coaxial mode
was below the cutoff frequency of the cylindrical waveguide, it would be reflected off the
transition and bound in the coaxial section. Clearly, there would be some non-ideal field
profile on the length scale of the evanescent decay length, but this is only a perturbation of
the mode shape. It should be noted that while you can solve for pillbox modes with open
terminations, these structures will give, in reality, unbound modes that are completely
impractical.

20
4.2 The Ideal Quarter Wave Resonator
First, we will discuss a coaxial resonator with one short termination (at z = L) and one
ideal open termination (at z = 0), seen in Figure 8. Coaxial lines are typified by a radial
electric field and an azimuthal magnetic field, with a 1/ρ dependence in amplitude. The
traveling wave formulation of this gives

Figure 8: Schematic of the ideal QWR.

I0 µ0 c iωt+ikz
Eρ = − e (94)
2πρ
I0 µ0 iωt+ikz
Bϕ = e . (95)
2πr

21
Again, summing two identical traveling waves with opposite directions of propagation
eikz + e−ikz = 2 cos (kz) (96)
eikz − e−ikz = 2i sin (kz) (97)
we can satisfy the boundary conditions to get
E0 a ( pπz )
Eρ = cos sin (ωt) (98)
ρ 2L
E0 a ( pπz )
Bϕ = − sin cos (ωt) (99)
ρc 2L
where p is an integer giving the order of the mode in z, E0 is the peak surface electric
field (located at z = 0 and ρ = a), a and b are the inner and outer conductor radii, and
ω = pcπ/2L. From this, it is trivial to see that the peak surface magnetic fields is E0 /c
and is located at z = L and ρ = a. Rearranging this last equation gives us L = λ/4 for
the lowest mode, p = 1, giving the cavity its name: Quarter Wave Resonator (QWR). The
accelerating axis for this cavity is at z = ϕ = 0, giving an ideal QWR accelerating field seen
in Figure 9. Using these field distributions, it is straight-forward to find an ideal QWRs
figures of merit. The stored energy, U , is
∫
ϵ0 
 2
U = E  dV (100)
2 V
∫ ∫ 2π ∫ L ( πz )
ϵ0 E02 a2 b 1
= dρ dϕ cos2 dz (101)
2 a ρ 0 0 2L
( )
ϵπE02 λ b
= ln (102)
8 a
remembering that L = λ/4. The dissipated power, Pd , is also straight-forward to calculate:
∫ [ ( ) ( )]
Pd 1 
 
 πE02 a2 ϵ0 λ 1 1 b
= 2  B 2
dA = + + 2 ln (103)
Rs 2µ0 S 2µ 0 4 a b a
remembering to sum the contributions from the inner conductor, outer conductor, and the
short plate. The dissipated power on the open termination is zero because the magnetic
field there is zero. These numbers allow us to calculate the geometry factor G:
( )
cπµ0 ln ab
G = [λ (1 1) ( b )] . (104)
2 a + b + 4 ln a
The calculation of the optimum beta is less straight-forward for this geometry compared to
the pillbox cavity. The integral is made complicated by the form of the accelerating field
which gives an equation for the accelerating voltage
∫ b ( )
E0 a ωx
Vacc = 2 sin dx (105)
a x βc

22
Figure 9: An example of the accelerating electric field for an ideal coaxial resonator. The
beam axis passes radially through the cavity at z = 0.

that does not have an analytic solution. We can sidestep this problem by remembering
that we can find the optimum velocity by solving for when the derivative of Vacc is zero.
( )
∫ b cos ωx ωx [∫ a ( ) ]
∂Vacc βc β 2 c 2E0 aω ωx
= 2E0 a − dx = 2
cos dx (106)
∂β a x β c b βc

which gives [ ( ) ( )]
∂Vacc 2E0 a ωa ωb
= sin − sin = 0. (107)
∂β β βc βc
Using the trigonometric identity
( ) ( )
u+v u−v
sin u − sin v = 2 cos sin (108)
2 2

23
we obtain ( ) ( )
ω(a + b) ω(b − a)
cos sin = 0. (109)
2βc 2βc
This gives conditions of
nβλ
nβλ = b − a; =b+a (110)
2
where n is an integer. The first equation give solutions where the gap size (b − a) is
extremely large, large enough for the particle to see at least one full RF period. Thus,
these solutions are likely to be minima, and will be disregarded. The second equation gives
the relationship that we expected for n = 1, that
βλ
=b+a (111)
2
which is identical to the relationship assumed for the pillbox cavity (remembering that
b + a is the gap-to-gap distance). For an ideal QWR, knowing the frequency and optimum
particle velocity desired specifies two of the three variables for the cavity, with the gap size
b − a being the only unspecified value. A plot of R/Q and G versus gap size (b − a) can
be seen in Figure 10. In this plot, it is easy to see that a large gap size results in a more
efficient cavity. This is tempered, however, with the knowledge that the larger the gap
for a given βopt , the smaller the inner conductor diameter. The remaining figures of merit
for the ideal QWR are the peak surface electric and magnetic fields. These are both on
the inner conductor, and their magnitudes scale like 1/a. The limit to how much we can
reduce a as we increase R/Q and G is driven by the requirement of minimizing these peak
surface fields.
From Figure 11, we can see that the ideal open termination is not truly representative
of the electric field region of a practical QWR. While a detailed treatment of QWR design
is outside the scope of this document, an issue of some importance is the asymmetry of this
geometry not present in the ideal QWR fields. This asymmetry causes steering electric and
magnetic fields on the beam axis, especially when the frequency of the cavity is increased
above ∼ 200 MHz. This steering can be corrected to a degree [6], but QWR usage is
generally limited to applications at lower frequencies, and therefore lower velocities.

4.3 The Ideal Half Wave Resonator

If, instead of a short termination at (z = L) and an ideal open termination at (z = 0), you
have a short termination at (z = L/2) and another short termination at (z = −L/2) (seen
in Figure 12), the mode structure changes, giving you a higher frequency cavity for a given
length. The field profiles change to
E0 a ( pπz )
Eρ = cos sin (ωt) (112)
ρ L

24
 Figure 10: R/Q and G versus gap size for an ideal 80.5 MHz, β = 0.041 QWR.

E0 a ( pπz )
Bϕ = − sin cos (ωt) (113)
ρc L
where p is an integer giving the order of the mode in z, E0 is the peak surface electric
field, a is the inner conductor radius, and ω = pcπ/L. Rearranging this last equation gives
us L = λ/2 for the lowest mode, p = 1, giving the cavity its name: Half Wave Resonator
(HWR). These equations are all very similar to a QWR, and in fact are only different
because of the change in definition of the length, L. An ideal HWR is effectively an ideal
QWR, but reflected about the QWR’s ideal open boundary condition. This means that
almost all of the derived quantities for the QWR can be used here with minor modification.

25
 Figure 11: FRIB Quarter Wave Resonator designs.

These modifications give

∫
ϵ0 
 2
U = E 
 dV (114)
2 V
∫ b ∫ 2π ∫ L/2 ( πz )
ϵ0 E02 a2 1
= dρ dϕ cos2 dz (115)
2 a ρ 0 −L/2 L
( )
ϵπE02 a2 λ b
= ln . (116)
4 a
∫ [ ( ) ( )]
Pd 1 
 2 πE02 a2 ϵ0 λ 1 1 b
= 2 B 
 dA = + + 2 ln (117)
Rs 2µ0 S µ0 4 a b a
(b)
cπµ0 ln a
G = [λ (1 1) ( b )] (118)
2 a + b + 4 ln a

26
 Figure 12: Schematic of an ideal HWR.

remembering that 2L = λ. Additionally,

∫ b ( )
E0 a ωx
Vacc = 2 sin dx (119)
a x βc

βλ
= b + a. (120)
2
remain identically valid for the HWR. It should be noted that, for a given length, the
frequency of the structure has doubled. In addition, because the losses are driven by the
magnetic fields, the losses for a given acceleration will double going from a QWR to a
HWR. This will double the HWR’s R/Q, but because the stored energy also doubles, G
remains the same. In addition, the location of the peak surface electric and magnetic fields
remain the same, with the caveat that there are not two peak surface magnetic fields, at

27
z = L/2 and z = −L/2. The benefit of the HWR geometry is the symmetry along the long
axis, meaning that steering fields are only generated from manufacturing errors as opposed
to the inherent geometry itself. This means that the frequency limitations of the QWR do
not apply to HWR, allowing acceleration of higher velocity particles.

4.4 Coaxial Cavity Efficiency at Low Velocity

While Vacc for coaxial cavities is not analytic, the possible acceleration (V0 ) is. For a QWR,
this leads to ( )
b
Vacc = V0 × T T F = 2E0 a ln × TTF (121)
a
which leads to ( ) ( )
R 8 ln ab b
= TTF × 2
2
≈ T T F × ln
2
× 300 Ω (122)
Q π ϵ0 c a
which does not scale with β. The T T F at optimum velocity for most designs will be
0.85 < T T F < 0.90, and does not scale strongly with β. For most coaxial designs, b/a
is between 2.5 and 3.5, although the exact definitions of a and b break down for realistic
cavity designs. This means that coaxial cavities maintain their acceleration efficiency for
very low velocity particles while pillbox type cavities become increasingly less efficient.
To directly compare an ideal HWR to an ideal pillbox cavity, the HWR’s R/Q must be
halved to account for the HWR having two gaps compared to a single pillbox. This is
an interesting comparison because it can be used to indicate the β where the R/Q of the
pillbox cavity is equal to the R/Q of a coaxial cavity, giving some idea of the velocity where
one geometry should replace the other:
0.5 × R/QHW R 150 Ω × F
βtransition = ≈ π ≈ 0.75 × F (123)
R/Qpillbox 150 2.405 Ω
where ( )
b
F = T T F × ln
2
. (124)
a
F is dependent on the design of the cavity, but for most HWR designs will be between
0.8 and 1, giving a transition β of ∼ 0.7. This scaling is born out in the parameters of
realistic cavity designs for applications around these velocities. While the R/Q scales like
β for a pillbox and is roughly constant for coaxial cavities, the Geometry Factor is more
complex. For an ideal pillbox, G = 257β Ω. For a HWR, assuming a reasonable b/a of 3,
the dependence is more complex:
1300β
GHW R ≈ Ω (125)
5.3 + 4.4β
which goes to zero for small β, and is only ∼ 120 Ω for β = 1. A plot of this behavior can
be seen in Figure 13. For comparison, an ideal HWR with β = 0.5 would have a Geometry

28
Factor of ∼ 80 Ω, where the realistic FRIB design for a similar cavity has a Geometry
Factor of ∼ 100 Ω. This is because the ideal HWR geometry is not very efficient; a more
complex inner and outer conductor geometry provides this increase in efficiency.

Figure 13: Geometry Factor plotted versus β for the ideal cavity geometries.

4.5 An Aside: The Three-Quarter Wave Resonator

An extension of the logic used to get the parameters for the HWR can be used for a more
novel type of cavity geometry. It has been proposed [7] to operate a QWR-like geometry
(open and short boundary conditions) in the p = 2 mode, giving L = 3λ/4. The particles
would be accelerated through the electric field maximum that is closer to the short plate,
away from the open boundary. This would combine the benefits of the QWR geometry
(open end for processing and inspection after fabrication) while avoiding the steering fields
present in a traditional QWR. This cavity type is called a Three-Quarter Wave Resonator,
or 3QWR. Comparing a HWR and 3QWR at the same frequency and beta, it is clear that

29
the 3QWR will have the same G while the R/Q will be 1/3 lower.

4.6 Another Aside: Power Transmission Choices Based on Frequency

The frequency scalings we’ve derived for the cylindrical and coaxial waveguides give some
idea for the type of transmission line that would be required to transmit energy to these
cavities. For the cylindrical waveguide, we derived a cut-off frequency of
2.405c
ωcutoff = (126)
R
where R is the radius of the pipe. For the FRIB cavities, operating at 80.5 and 322
MHz, this would require a pipe radius of 1.43 and 0.36 m, far too large to be practical.
For the coaxial waveguide, there is no cutoff frequency, making this geometry the desired
transmission line for this application.

5 Superconducting Material Properties

Oxygen-Free, High Conductivity (OFHC) copper and niobium have been the materials
of choice for the vast majority of accelerators because of the efficiency offered. Copper
is an excellent conductor of both electricity and heat, allowing for low losses and easy
evacuation of the joule heat generated. Niobium is a poor conductor at room temperature,
but becomes superconducting below 9.2 K. The operational implications of using these
materials as well as the theory behind their losses will be discussed below.

5.1 Superconducting Material Theory

As has been shown, the skin depth δ and the surface resistance Rs for a normal conducting
material are
1
δ=√ (127)
πf µ0 σ
1
Rs = . (128)
σδ
Superconducting materials behave very differently from normal conducting materials, and
require a very different treatment to derive equivalent quantities. A superconducting ma-
terial is typified by a sudden transition between standard normal conducting physics to a
DC resistivity of zero at a certain temperature (below 9.2 K for bulk niobium). Below this
critical temperature (Tc ), it becomes energetically favorable for electrons to be combined
into “Cooper pairs” which act as a superfluid, allowing current to flow through the mate-
rial with no losses [8]. The current theory of this behavior is BCS theory [9, 10], and its
predictions agree well with measured results [11, 12]:
( )2
−4 CRRR f
e(− T )
17.67
RBCS = 2 × 10 (129)
T 1.5

30
as a good fit for the BCS component of the surface resistance of niobium where RBCS
is in ohms, T is the temperature in Kelvin, and f is the frequency in GHz. CRRR is
a factor based on the material purity, 1 for reactor grade niobium, and approximately
1.5 for RRR = 300 niobium. From this, it is easy to see that it is desirable to operate
superconducting cavities at the lowest temperature possible. In practice, RBCS isn’t the
only contribution to the surface resistance. The remaining resistance is a material/cavity
processing dependent resistance called the “residual resistance” Rres that does not depend
on temperature or frequency. While residual resistances of below 1 nΩ have been achieved,
10 nΩ is more reliably reproducible for large scale cavity fabrication [13]. If the cavity is
operated at or below 2 K, the residual losses tend to dominate performance. An approach
for estimating the skin depth of a superconducting material can be derived [4, pp. 85-88]
by using a two-fluid model for the electrons in the material, one fluid composed of the
Cooper pairs, and the other the unbound electrons. This gives a penetration depth (called
the London Length [14]) of
m
λ2L = (130)
ns e2 µ0
where m and e are the mass and charge of the electron, and ns is the density of the
superconducting electron pairs. For niobium, this penetration depth is on the order of tens
of nanometers, meaning that the RF performance is dominated by the quality of a very
thin layer on the surface of the material. The temperature dependence for λL has been
empirically found as [15]
[ ( )4 ]−1/2
T
λL (T ) ≈ λL (0) 1 − . (131)
Tc

While it becomes energetically favorable for electrons to form Cooper pairs below Tc , this is
offset by the application of a magnetic field. Magnetic flux is expelled in the transition from
normal to superconducting (called the Meissner Effect [16]), but the fields still penetrate
the material on the order of the London Length. BCS theory approximates the temperature
dependence of this critical magnetic field as
[ ( )2 ]
T
Hc (T ) = Hc (0) 1 − . (132)
Tc

Finally, niobium is a Type-2 superconductor. This means that the transition from su-
perconducting to normal conducting in the presence of a magnetic field actually has two
critical magnetic field levels, Hc1 and Hc2 , in contrast to a Type-1 superconductor in which
the superconductivity breaks completely at one field level. This behavior can be see in Fig-
ure 14. One of the reasons niobium is used for superconducting accelerators is the relatively
high Tc1 , one of the highest for any Type-2 superconductor.

31
Figure 14: Magnetic field just inside superconducting material around transition for Type
1 and 2 materials. Note the sudden penetration for Type 1 and the gradual break down
for Type 2.

5.2 The Advantages of Superconductivity

Assuming that it is operated at a temperature well below 9.2 K (not challenging with
liquid helium’s boiling point of 4.2 K at 1 bar), superconducting cavities can be operated
with very low losses. The surface resistance for a well prepared bulk niobium can be as
low as single nano-ohms, as compared to micro-ohms for a well prepared copper cavity.
This means that for the same cavity and field level, the losses are lowered by 6 (!) orders
of magnitude. Some of this is lost because of the requirement to operate at very low
temperature. The maximum achievable efficiency for heat transfer depends on the input
and output temperature and is given by the Carnot Efficiency:
TC
ηc = (133)
TH − TC
where TH and TC are the temperature at which the heat is exhausted and extracted in
Kelvin. For 4 K operation, this gives an optimum operation efficiency of .013. This
should be combined with the technical efficiency of cryogenics systems (ν ≈ 0.2 [4, p. 48]),
resulting in an increase in required wall power by two to three orders of magnitude. This
means that the six orders of magnitude decrease in material losses is, in practical terms,
three orders of magnitude gain in wall power efficiency. For applications which call for the
cavities to be operated in continuous wave (CW) mode, the cost savings of superconducting
operation can be substantial, despite the complex technology involved.

32
5.3 The Challenges of Superconducting Materials
While the decrease in required power is substantial, there are drawbacks to using super-
conducting cavities. The cryogenic engineering talent and facilities required is a dramatic
investment, and can be a significant portion of the overall cost of an accelerator. Addi-
tionally, the ultimate performance and achievable accelerating gradient is limited by the
peak surface magnetic field. Even for an extremely well prepared cavity, the peak sur-
face magnetic field is ∼ 220 mT, and for low velocity cavities, even with state of the art
processing, this number is closer to 120 mT. Because the losses in a superconductor also
increase strongly with frequency, this means that superconducting cavities are limited to
operation under 2 GHz for practical reasons. Even the International Linear Collider which
is designed to operate at 1.3 GHz and 2 K has moved to a pulsed design to limit the load
on the cryogenic systems. Additionally, there are many processing and fabrication consid-
erations for these cavities to achieve a low surface resistance with little to no radiation.
These considerations will be discussed later in this document.

5.4 Practical Concerns: RRR and Impurities

Procurement of the appropriate material for an accelerating cavity is an important step in
obtaining good performance. While the RF losses in a cavity are small, the specific heat
of niobium at those temperatures is also quite low. If the thermal conductivity is small,
local heating can occur at high fields, breaking the superconductivity in what is known as
a thermal quench. The thermal conductivity of the niobium is dominated by impurities
and crystal structure, which is measured by the RRR:

σ(4.2K)
RRR = ∝κ (134)
σ(300K)

where σ is the electrical conductivity and κ is the thermal conductivity. Material can be
readily obtained from suppliers with RRR of over 250, considered good enough to give
good cavity performance. This material is produced by repeated electron beam melting in
vacuum to drive off impurities, and can also be improved by baking over 1200◦ C in vacuum
with either titanium or Yttrium as a getter to remove impurities.

6 Coupler Theory and Design

Cavities are ultimately powered by a power plant, like all other devices that require elec-
tricity. This wall power is used to power both the low power RF signal generators and the
amplifiers (klystrons, solid state, or other types) that generate the large amounts of power
needed to fill the cavity. This power must be effectively transformed from traveling wave
power in the transmission lines to the standing wave mode of the cavity. In this section,
the basic design and some considerations of these couplers will be discussed in the context

33
of different applications. These topics will be expanded upon with more specific design and
theory for couplers used for low beta superconducting resonators like those being designed
for the FRIB driver linac.

6.1 Coupler Figures of Merit

Types of transmission lines have already been discussed, and coaxial waveguide has been
demonstrated to be the only practical solution at low frequency. From this, we can design
two types of couplers, probe (capacitive) and loop (inductive). Probe couplers use the
transmission line’s inner conductor to excite the cavity electric fields in the local region.
There are different ways to quantify the coupling strength, but the overlap between the
excited fields from the inner conductor of the coupler and the fields in the cavity drive the
coupling strength. Thus, the probe type coupler is more effectively used in the high electric
field region of the cavity. The desired coupling strength is then achieved by changing the
length of the coupler inner conductor. Magnetic field coupling is achieved by attaching the
coupler inner conductor to its outer conductor, making a loop. Magnetic fields are excited
normal to the loop, and the strength of the coupler is determined by the confluence of these
excited fields and the cavity mode’s magnetic field distribution. The desired coupling
strength is achieved by varying the area of the loop exposed to the cavity (moving the
coupler in or out of the cavity) or changing the angle between the cavity field and the
normal of the loop (rotating the coupler). The coupler strength can be quantified by the
rate at which energy enters or leaves the cavity through the coupler. This emitted power,
Pe , can be normalized to the stored energy in the cavity and used to form an effective
qualify factor for the coupler called the “external Q”:
ωU
Qext = . (135)
Pe
The ratio of the external Q to the cavity Q is called the coupling factor, β (not to be
confused with the relativistic β = v/c)
Q0 Pe
β= = . (136)
Qext Pd
In reality, what will be measured if a cavity is connected to a Vector Network Analyzer
(VNA) will be a quality factor given by the total losses in the cavity. The total losses in
the cavity are given by
PT = Pd + Pe + Pt (137)
where Pe is the power emitted from the input coupler (port 1), Pt is the power emitted
from the transmitted power coupler (port 2), and Pd is the power dissipated in the cavity
itself. This gives us a “loaded” quality factor of
1 1 1 1
= + + (138)
QL Q0 Qext,1 Qext,2

34
which gives
Q0 = QL (1 + β1 + β2 ). (139)
It is desired to measure Q0 directly, but in steady state, only the scattering parameters
(S11 and S21 ) can be measured. The scattering parameters are defined as

Pr
|S11 |2 = (140)
Pf

Pt
|S21 |2 = (141)
Pf
with Pf being the forward power and Pr being the total reverse power from the cavity, the
vector sum of the reflected and emitted traveling waves. β1 and β2 can be expressed in
terms of these scattering parameters [4, p. 48]

|S21 |2
β2 = (142)
1 − |S11 |2 − |S21 |2

1 − |S11 |
β1 = (143)
1 + |S11 |
or
1 + |S11 |
β1 = (144)
1 − |S11 |
depending on coupling strength. For β1 < 1 (Equation 143), the coupling is considered
“undercoupled,” for β1 > 1 (Equation 144), the coupling is considered “overcoupled.” The
remaining case, β = 1, is the “matched” condition. In this configuration, there is no reverse
power because the emitted and reflected power from port 1 are equal in magnitude and
opposite in phase. These equations are derived assuming the measurement is being done in
a steady-state and that β1 ≫ β2 so that Pt can be dropped while considering the behavior
of Port 1. It can be see from these equations that some additional knowledge is required
for accurate interpretation of the measured values, any measurement of |S11 | can result
in two different β1 results. This information can be found by changing the parameters of
the measurement. Observing the change in |S11 | because of adjusting the coupling is a
known way to determine if the cavity is over or under coupled. For example, increasing
the length of a probe coupler should increase the coupling strength β. If |S11 | increases
because of this change, the coupler is now overcoupled. If |S11 | drops, the coupler started
as undercoupled. This type of change must be done with care to avoid confusion from
crossing β = 1. Additionally, the Pf can be modulated, and the behavior of Pr and Pt can
be used to determine the coupling. This method is only realistic when the time constant
of the cavity
QL
τ= (145)
ω

35
is large enough to be measured easily. Because QL is so small when the cavity is normal
conducting (∼ 1000), it is only possible to measure the coupling by modulation of the
forward power when the cavity is superconducting, with QL ≈ 1 × 107 or higher. For a
realistic cavity, two couplers are used. The “input” coupler is a high power coupler designed
to be either close to matched or overcoupled depending on the situation. The second
coupler, called the “pickup” coupler, is used as a field probe, allowing simple measurement
of the cavity stored energy. This coupler is significantly undercoupled so that the power
radiating out of this coupler (Pt ) is very low, so it can be treated as a perturbation, greatly
simplifying the mathematical analysis of the measured results. A useful rule of thumb
when choosing the pickup coupling is to ensure that Pt is always below 1 mW, even at the
extreme limit of cavity performance.

6.2 Bandwidth Considerations

The bandwidth of the cavity is determined by the total energy loss in operation, giving the
equation
f0
∆fBW = . (146)
QL
Because Q0 is so large for superconducting cavities, this means that the cavity bandwidth is
often dominated by the choice of Qext,1 . When testing a cavity in controlled environment,
the cavity frequency can be allowed to drift as needed. This means that the bandwidth
of the cavity can be kept quite small, often as close to matched as possible, because the
RF circuitry is flexible enough in the configuration to not require a large bandwidth. This
keeps the power requirements as small (and inexpensive) as possible, simplifying the testing
hardware required. For operation in a real machine, the cavity frequency must be tightly
controlled for stable operation. This means that the RF control systems must react to
the changing cavity conditions while keeping the amplitude, phase, and frequency of the
cavity within tight, preset parameters. This is a much more difficult task, and the cavity
bandwidth must be large enough to allow this stable operation of the cavity. This means
heavily overcoupling the cavity, meaning that Pe dominates cavity losses and Pr ≈ Pf .
How overcoupled the cavity must be is determined by size and time scale of the frequency
perturbations the cavity is likely to experience as well as the control systems used.

6.3 High Power Concerns

When the cavity is heavily overcoupled, Pf is much larger than Pd . This means that
the input couplers must be able to handle full power in the forward and reverse direction
simultaneously, and this leads to stringent cleaning standards for couplers, much like the
cavities. Couplers are often cleaned and conditioned much like normal conducting cavities
are, and their cooling can be a serious issue. Couplers designed for such high power and so
heavily overcoupled will often be a significant perturbation to the cavity fields, and their

36
effect on cavity frequency and performance must be taken into account during standard
cavity design. Additionally, the coupling is chosen based on an assumption of Q0 , but
in reality, this can vary from cavity to cavity. The ability to vary the input coupling in
situ adds mechanical complexity to coupler design, but allows fine tuning of each cavity
to either minimize power requirements on an overperforming cavity or allow operation of
a poorly performing cavity. This benefit extends to Dewar testing, where Q0 is varying
more widely than in real operation. A variable coupler allows for testing or conditioning
at a higher field by keeping the cavity close to matched for a much wider range of cavity
performance.

7 Electromagnetic Cavity Simulation

The ideal HWR geometry is an instructive tool for estimating cavity performance because
it is analytically solvable. This geometry is, however, not practical for use in a real accel-
erator. There are many features that must be added and optimized that cannot be treated
analytically, and must be simulated. Modern electromagnetic simulation software can sim-
ulate the parameters of almost any geometry and evaluate its figures of merit, often only
limited by the amount of computing power available. In this section, the basic methods
that these solvers use to simulate these complex geometries will be discussed. In addition,
more practical considerations and methods used to perform the design simulations will be
presented for use in the rest of this document.

7.1 Finite Element Solvers

In order to solve a complex geometry like that of a realistic cavity, the volume of the cavity
is broken into a discrete set of interlocking simple geometries, called elements. Tetrahedral
elements are the most commonly used for cavity design because they often give the most
accurate representation of the complex surfaces involved in cavity design. An example of
this can be seen in Figure 15, generated in ANSYS-APDL. The fields in the cavity are then
assumed to have a simple representation in the volume of the element, expressed as a low-
order (often 1st or 2nd) function of space. An example of the structure of these elements
can be seen in Figure 16. Because the fields in the cavity are assumed to be monochromatic,
the electric fields in the cavity can be described by the Helmholtz Equation:
⃗ = 0.
(∇2 + k 2 )E (147)

This equation is then discretized and applied to each element. Elements “interact” at fixed
places, generally along the outside edges, called nodes. The parameters of each element are
iterated with the goal of minimizing the global stored energy. The study and development
of these solvers is an active and growing field, but for the basic methods and mathematical
foundations of these solvers, “Electro-magnetics and Calculations of Fields” by Ida and

37
Figure 15: A cavity vacuum space meshed with tetrahedral elements, a coupler probe is in
the foreground with a beam port cup visible in the background.

Bastos [26] or “Finite-element Methods for Electromagnetics” by Humphries [27] are rec-
ommended as starting places. What this procedure allows is the accurate representation
of the fields in a complex geometry with evaluation of the electromagnetic figures of merit.

7.2 Geometry Creation

The first step of any simulation is defining the problem. Because the geometries involved
are very complex, it was chosen to use a Computer-Aided Design (CAD) program, Solid-
Works [28], to create the needed geometry. This is because the native geometry creation
in the electromagnetic solver used for this project (and most other solvers) works by the
creation and Boolean combination of primitive objects (cone, spheres, rectangular boxes)
or swept two dimensional shapes. After the geometry is completed, it must be exported
in a form that can be read by the simulation software. The format of choice for this was
SAT, for ease of both export and import. Depending on the complexity of the geometry, it
may be necessary to use different SAT [29] file settings or versions (as of publication, there
were 18 versions) for the simulation software to properly interpret the geometry.

38
Figure 16: Structure of a first order (linear) tetrahedral element. Image credit to F.
Krawczyk.

7.3 Simulation Procedure

Assuming that the geometry has been successfully created and imported into simulation
software, the basic electromagnetic simulation occurs in four major stages: Setup, Mesh-
ing, Solving, and Post Processing. Each of these stages carries its own importance and
subtleties, but in the following subsections, the work flow for each stage in the context of
the simulation software used for the work presented later will be outlined. The simulations
software used for the vast majority of the work presented is ANSYS-APDL [30]. Complete
scripts detailing the execution of the described procedures can be found in the Appendices.
It should be noted that, while the general process can be applied, the details of repeating
this work in different simulation software (CST-Microwave Studio [31], ANALYST [32],
HFSS [33], ACE3P [34]) will very likely be quite different, i.e., Caveat Lector.

39
7.4 An Aside - Computer Resource Management
Computer resource management is an important setup for large scale simulations. This
includes storing solutions files on the fastest available disk drive available. Using solid
state drives in place of traditional palette drives can decrease simulation time by 30 -
40%, depending on the type of solution being done, assuming a large enough solid state
drive can be acquired. Eigenmode simulations tend to be memory limited, as opposed to
processor speed limited, meaning that investments in upgrading available memory speed
and size can speed simulations considerably. Additionally, the temporary files generated
during simulation and results files can be extremely large. If the disk drive you are using
is 32-bit formatted, it will be unable to store these files. Some simulation software have
a feature that ensures no files are larger than the 4 Gigabyte limit for FAT32 formatted
drives. A 32-bit architecture also limits the amount of memory that can be allocated to
a single program to 2 Gigabytes as well, so moving to 64-bit architecture removes both
of these limitations. Depending on the software in use, multiple processor cores can be
used in parallel to increase solving speed. When starting a simulation, ensure that the
maximum number of processors, largest amount of memory, and fastest hard drive space
available are allocated for the job.

7.5 Simulation - Setup

Given that the skin depth for high conductivity materials (and especially superconducting
materials) is so small, the fields in the conductor and the losses generated can be treated
as a perturbation. That leads to two standard types of boundary conditions, Perfect
Electric Conductors (PEC), and Perfect Magnetic Conductors (PMC). The PEC boundary
condition indicates that E∥ = 0, and H⊥ = 0, and is appropriate for conductor surfaces and
some boundary conditions depending on the mode structure you are trying to simulate.
The PMC boundary condition indicates that E⊥ = 0 and H∥ = 0, and is used for a majority
of symmetry planes. All exterior surfaces of the model must be defined as one of these.
Starting from a basic imported vacuum space, it is very useful to define groups of important
surfaces for later reference. These groups can include symmetry planes, conducting planes
of the same conductivity, or areas to be used for special types of analysis. For the HWR
model seen in Figure 17, generated in SolidWorks, there are two groups of critical surfaces,
the symmetry planes seen in blue and the cavity RF surface in gray. The symmetry planes
are defined as PMC, and the RF surfaces are defined as PEC. Additionally, any surface
can be assigned a conductivity. This will be used to perturbatively calculate losses based
on the magnetic fields on that surface. While the cavity RF surfaces are generally assigned
the same conductivity, surfaces of interest can be assigned different conductivities based
on their material type. This can be very useful for specialized investigations. Numerical
constants can also be set for convenient use later. Grouping their declarations early in a
script eases later modifications. The type of element must be defined as a high-frequency

40
eigenmode type element, and the order of this element must be set. Material properties
are also set here, but for a simple high-frequency eigenmode solution, ϵ = ϵ0 and µ = µ0
are the only required definitions.

Figure 17: HWR vacuum model. Symmetry planes can be seen in blue, Perfect Electric
Boundaries can be seen in gray, the cavity origin is marked in red.

7.6 Simulation - Meshing

Finite element simulations require the cavity volume to be represented by a collection
of smaller volumes called elements. The process of defining the size and shape of these
elements is called meshing. These elements should be, at largest, one tenth the wavelength
you expect to simulate to improve accuracy [26]. The accuracy of the remaining figures of
merit are strongly influenced by the quality of the mesh. Curved surfaces are especially
sensitive to meshing as the elements on the surface must be small enough to accurately
represent the curvature. A rule of thumb for small curved features is the element size should
be smaller than one third the radius of curvature. An example of inadequate meshing can

41
be seen in Figure 18.

Figure 18: Comparison of different meshing on a curved surface. Left: Poor meshing on a
curved surface. Right: Same surface with improved mesh density. Images generated using
ANSYS.

While most software has advanced automatic meshing algorithms, additional manual
refinement may be required based on the geometry being simulated. The first, most com-
mon, mesh optimization step is the setting of a maximum element edge length. This can
be done for both whole volumes and on areas of known high surface fields. For the vol-
ume sizing, it is important to set a global maximum value for an element edge length to
prevent drastic gradients in element size. The standard rule of thumb for electromagnetic
simulation calls for a maximum element edge length of λ/10 ≈ 9 cm for 322 MHz. For
complex geometries like the ones used for HWRs, additional surface refinement may be
required to accurately calculate surface losses and peak surface fields. For accurate surface
fields on the small curved surfaces, elements as small as 1 mm were found to be needed in
some locations of the cavity. If the transition between these element sizes is not carefully
controlled, poor mesh quality can result. Setting a global maximum on the element size
is a powerful tool for regulating this transition. A major factor in mesh quality tolerance
is the order of element chosen (i.e., what order function is used to represent the fields in
an element). As can be seen in Figure 19, increasing the mesh density reduces frequency
error down to a baseline inaccuracy that is solver dependent. For ANSYS-APDL, this
error was found to be roughly 1 part in 105 , or ∼ ±50 kHz out of 322 MHz. This level of
accuracy was achieved with second order elements at a much lower density than for first
order elements. While the solve times for the same accuracy is roughly equivalent, the
use of second order elements greatly reduces the meshing tolerances while giving similar or

42
Figure 19: Fractional frequency error for different mesh densities in identical geometries
for first and second order elements.

improved surface field accuracy. There are sometimes limitations from the meshing that
can help drive the decision to strive for first or second order meshing. Some geometries
are so complex that their mesh densities must be high for geometric fidelity. Depending on
the available computing hardware, this type of geometry can often require far more mesh
elements than could be solved with second order elements, but can be solved with first
order elements at a loss of accuracy. In this situation, all results, especially surface fields,
must be interpreted with this in mind. When the computing hardware allows, this type
of complex geometry should be solved using second order elements if possible, especially
if surface fields are of particular interest. Meshing must also be approached carefully with
geometries that are very simple. Meshing software will often generate very large elements
in these geometries if allowed, and care must be taken to ensure they do not grow large
enough to sacrifice accuracy. This is an ideal situation to use a global element edge length
limit.
Additional care must be taken when generating geometry to avoid causing meshing
problems. While this is a more serious issue for mechanical simulations, and will be dis-
cussed in greater detail in the relevant section, there are certain features that should be
avoided if possible, and these are places of optimization to generally reduce the size of simu-
lations without great sacrifice for accuracy. CAD geometries are represented in simulation

43
software as a continuous and closed collection of surface, each bounded by lines. Mesh
structure must respect the shape and boundary of these surfaces, and care should be taken
to not create very small areas unnecessarily, or otherwise collect many boundaries together
without purpose. An example of a mesh that is overly dense at surface intersections can be
see in Figure 20. This type of feature can be tolerated if the solution time isn’t excessive,
but when working on a very large simulation, it may be beneficial to find instances like
this and correct the geometry to allow for more optimal meshing.

Figure 20: The intersection of four surfaces can force excess mesh density, indicated by the
red arrow. Images generated using ANSYS.

Automatic meshing algorithms often make meshing decisions based strictly on the ge-
ometry, not on the structure of the mode being simulated. This is understandable because
the information is not available to the program before the simulation is complete. Some
programs like ANALYST have automatic iteration algorithms that use solution data to
adjust and optimize the meshing until the solution has converged to the accuracy desired.
For most simulations software, foresight can dramatically improve the quality of the sim-
ulation results. We’ve touched on this for surface fields and especially external surfaces

44
with a large curvature, but this is also quite important for the accelerating field profile.
Large elements on the beam axis result in a non-physical accelerating field because the
fields cannot be properly interpolated between elements to give a smooth shape. Improv-
ing the density of the meshing around the beam axis gives a more accurate accelerating
field profile. Both of these behaviors can be seen in Figure 21.

Figure 21: Top Left: Poorly meshed cavity, beam axis marked in red. Top Right: Resulting
accelerating electric field. Note the non-physical roughness that results from the large
element size. Bottom Left: Cavity with improved meshing on beam axis which is marked
in red. Bottom Right: Resulting accelerating electric field. Note the improved field profile
that results. Images generated in ANSYS.

7.7 Simulation - Solving

Once the cavity has been meshed and appropriate boundary conditions (PEC, PMC, con-
ductivities, etc.) have been specified, most solvers require some options set to calculate an
eigenmode. For ANSYS-APDL, the solver needs to be set to modal, and a frequency range
given for the mode(s) desired. The solver will search for eigenvalues (resonant frequencies)

45
for the cavity element matrix from the low end to the high end of the range until the re-
quested number of modes are found. For most purposes, this range should only be as large
as needed to encompass all the variation in the single frequency anticipated. Searching for
a range that is ∼ ±10% of the desired resonant frequency is often large enough to still
find the resonance even after relatively large geometry changes. For complex structures,
this range may also include unintended trapped modes in small features like couplers or
tuners, so it is often advisable to scan a larger frequency range for modes before focusing
exclusively on the mode of interest. After launching the solution process, the solver will
often list the amount of memory needed for the solution. “In core” solutions are done
with the entire simulation resident in memory. This greatly speeds simulation time and
is the desired mode of operation. “Out of core” solutions require the solver to transfer
data between memory and the hard disk, a much slower process. This is one of the prime
reasons simulation programs should be allowed access to as much memory as possible.

7.8 Simulation - Post Processing

The first step after a new simulation finishes is verifying that the solve has completed
properly. While there are various intuitive methods for this, the most effective is often
plotting the fields in the cavity. For most geometries, the mode structure is known, at least
in concept. Qualitative comparison of the solved fields with this general form (comparing
a HWR simulations to the general structure of the ideal HWR mode, for instance) can
give a good indication of the quality of the solve. This will also identify surfaces requiring
additional mesh refinement, an example of which can be seen in Figure 22. In this example,
the coarse meshing resulted in a non-physical roughness to the surface fields, with one
element having artificially high fields. Improved meshing removes a great deal of this
roughness, and gives a more accurate peak surface field. Once the solution has been
evaluated and found to be sufficient, extraction of the basic cavity figures of merit can
begin. This starts with extraction of bulk properties of the mode and cavity, such as total
dissipated power, total stored energy, and cavity volume. These values must be scaled
depending on the amount of symmetry used in the cavity. Some solvers automatically
do this, but in ANSYS-APDL, it must be done by hand. It should be noted that all
stored energy dependent quantities like dissipated power and peak surface fields must be
normalized to the stored energy. ANSYS-APDL solves the cavity mode at some nominally
small amount of stored energy, and this value must be used to scale the cavity figures
of merit. Once the dissipated power, frequency, and stored energy are known, the cavity
quality factor can be calculated:
2πf U
Q= . (148)
Pd
This can then be multiplied by the surface resistance given in the setup phase of the
simulation to calculate the geometry factor:
G = Rs × Q. (149)

46
Figure 22: Left: Magnitude of the surface electric field on a beam port cup with inadequate
meshing for surface field accuracy. Right: Identical geometry with improved meshing and
more realistic surface fields.

The peak surface fields can be found in ANSYS-APDL by making an element table of the
field in question. This table can be sorted by field magnitude and the maximum value
can be extracted. Again,
√ this is the
√ field value for the nominally low stored energy of the
simulation, so Epk / U and Bpk / U should be calculated for later scaling when the proper
stored energy is known. The remaining figures of merit require Vacc , and the bulk of the
remaining post processing will revolve around its calculation. The first step is to extract
the accelerating electric field component along the beam axis. In ANSYS-APDL, this is
done via the PATH command, specifying the beginning, end, and field component desired.
This gives the value of the accelerating field at a number of discrete points along the line
specified, and this number must be large enough for this to be a good approximation of
the actual (continuous) accelerating field profile. The first calculation is
∫ +∞  ∑
 

N
 

V0 = Eacc (z)
  dz = Eacc,i 
  ∆z (150)
−∞ i

which gives the maximal acceleration possible, V0 . The next step is calculating the acceler-
ating voltage, which includes the field’s variation in time. This calculation uses the general
form √
(∫ +∞ )2 (∫ +∞ )2
Vacc = Eacc sin (ωt)dz + Eacc cos (ωt)dz (151)
−∞ −∞

which does not depend on the choice of (z = 0). This becomes

v
u( N ( ) )2 (∑ ( ) )2
u ∑ N
Vacc (β) = t Eacc,i sin
ωzi
∆z + Eacc,i cos
ωzi
∆z (152)
βc βc
i i

47
 Figure 23: Plot of transit time factor versus particle beta.

in practical terms, although this can be simplified sometimes by the choice of (z = 0) and
cavity symmetry. For instance, a HWR with the origin placed in the geometric center of the
cavity can neglect the cos term because of symmetry. This formula depends on β, although
does not directly give the optimum β. To find this, Vacc (β) is calculated for β ranging from
0.001 to 1 by steps of 0.001. Taking the maximum of this set gives the optimum β with
enough accuracy for essentially all applications. This data can be normalized to V0 to
give the transit time factor curve, an√example of which can be seen in Figure 23. Once
the optimum β has been found, Vacc / U is calculated, giving the last of the desired field
quantities. This is used to calculate R/Q:
( )
1 Vacc 2
R/Q = × √ . (153)
ω U
To find the figures of merit for the cavity at full field, some normalization must be chosen.
For the design of the FRIB cavities, it was convention to choose a Vacc desired
√ for a cavity
and normalize to that. This determined the stored energy (U ) from Vacc / U , and therefore
the peak surface fields from the equivalent ratios already calculated. A detailed ANSYS-

48
APDL script used for all of this work can be seen in Appendix A. To this point, the basic
electromagnetic figures of merit have been calculated. Sometimes, more details or complex
problems needed to be solved, and these often can be solved in post processing. This type
of problem often involves field levels or losses on certain parts of the cavity (RF joint,
plunger, flange). These problems are solved by selecting the areas, elements, and nodes
related to the area in question, and repeating the same process used to extract the fields
or losses for the full cavity. If necessary, the losses can be scaled to a different resistivity to
simulate a small normal conducting part of the cavity and the quality factor recalculated
for the new losses.

7.9 Coupling Power to the Cavity

An obvious but non-trivial feature of any cavity is a port for coupling power from an
external generator into the cavity. Coupler design drives the size and positioning of the RF
port. For FRIB, it was decided to use electric probe coupling for both input and pickup
couplers, ninety degrees rotated from the beam axis, as can be seen in Figure ??. While
each of these couplers will perturb the cavity fields, couplers designed for high power are
often much larger than testing couplers, and will perturb the cavity more to achieve the
desired stronger coupling. These disturbances will shift the resonant frequency of the cavity
and must be taken into account during design, although the shift itself is often measured
experimentally. Good coupler and cavity interface design also ensures that regions of high
surface fields are not created around the coupler port, although the location of the RF
port on the outer conductor means this is not generally challenging. Additionally, the
coupling strength as a function of probe length can be simulated with a standing wave
simulation. A typical geometry for this type of simulation can be seen in Figure 24. The
coupling strength of the port can be simulated by solving the model with the termination
of the coupler port (indicated by a red arrow in Figure 24) as both PEC and PMC, and
extracting the fields on the boundary. This is called the Current-Voltage method [35, 36],
and assumes that the fields in the coupler port are purely TEM at the termination. At the
PEC termination, the electric field is zero because the TEM only allows fields transverse
to the surface, but these are forced to zero by the PEC condition. The distribution of the
magnetic fields can be seen in Figure 25, and it has the characteristic 1/ρ dependence of a
coaxial TEM mode. An identical distribution for the electric field is seen when the cavity
is simulated with a PMC termination. The Current-Voltage method uses a lumped circuit
for the coaxial transmission line, and calls for the calculation of the “voltage” between the
inner and outer conductor on a PMC boundary and the “current” across the termination
for a PEC boundary. Using this model, we can calculate Qext as
Qext = Qi + Qv (154)
where ∫
ωU 1
Pi = = |E|
⃗ 2 ds (155)
Qi 2η S

49
Figure 24: The coupler probe can be seen in the foreground connected to the RF port
by the surface indicated by the red arrow. This is a quarter of the geometry cut on the
symmetry planes.

and ∫
ωU η
Pv = = |H|
⃗ 2 ds (156)
Qv 2 S
following the standard definition of a quality factor. The form of a TEM wave in a coaxial
transmission line gives a relatively simple form for Qext :
( )
ωU η 1
Qext = ( ) + (157)
π ln ab (aEpk )2 η(aHpk )2

where a and b are the inner and outer conductor radii for the coupler port, η = 377 Ω is
the impedance of free space, and Epk and Hpk are the peak fields on the termination for
the PMC and PEC boundary conditions, respectively.
Comparison of simulation and measured data for a prototype β = 0.53 HWR can be
seen in Figure 26. The details of measuring the Qext on a real cavity will be discussed
later. The most error-prone part of this measurement is the transformation of a given
coupler tip length to penetration into the cavity volume. While this is easy to do in a CAD
model, errors in manufacturing and positioning add significant error to this part of the

50
Figure 25: Plot of the magnitude of the magnetic field on the PEC termination of a coupler
port. Simulated with planes of symmetry in x and y.

measurement. This can be see in the data plotted in Figure 26, where the measured values
(thin black trend lines) and the simulated data (large black squares) are offset from each
other by as much as 5 mm. This can even be seen in the shift between different measurement
runs which are a similar distance apart. The slope of the measured and simulated data
agrees well, however. This is because the errors in fabricating the port diameter and
measuring the coupler probe diameter are much smaller, and can be simulated accurately.
It is important to notice that at very negative penetration into the cavity (very small
probe lengths), the trend line changes significantly from both the measured and simulated
data. This is because the approximation used in the simulation breaks down significantly
for very small probe lengths, the modes no longer have enough space to transform into a
pure TEM shape before reaching the termination plane of the RF flange. To accurately
model this behavior, a more advanced theoretical method or a traveling wave simulation
is required. These additionally require a much more accurate description of the geometry
involved, including the coupler feed through and flange shape.

51
Figure 26: Plot of measured and simulated Q-externals for a 0.53 HWR prototype. The
simulated data is represented by the large black squares. The rest are repeated rounds of
measured data.

7.10 Etching Coefficient Estimation

A major step in cavity processing is a bulk etching of the RF surfaces. The goal of this
etching is to remove around 150 µm of material from the RF surface that has been damaged
by cavity fabrication, improving cavity performance. This etch generally changes the cavity
frequency enough so as to require that this shift be well understood so that the correct
final frequency can be repeatably achieved. Using Slater’s Theorem (see Section 8.4)
∫
∆f 1 ( )
= ϵ0 E 2 − µ0 H 2 dV (158)
f0 4U ∆V

this shift can be perturbatively estimated. Because the simulation procedure requires the
volume be broken into discrete elements, this formula requires modification to:
∆f f0 ∑ ( )
= ϵ0 Ei2 − µ0 Hi2 dAi (159)
∆T 4U
i

where ∆T is the thickness of the etched layer, Ai is the area of the external face of element
i, and the fields Ei and Hi are the fields from surface element i. For an ideal HWR, this
can be calculated directly from Slater’s theorem: −1383 Hz/µm. This large negative shift
is intuitive because etching of an ideal HWR effectively makes the cavity longer, directly

52
lowering the cavity frequency. In the process of cavity optimization, the changes to the
electric field region increase its contribution to this shift greatly, producing a less negative
coefficient. For the HWR design being developed here, the simulated etching coefficient is
−206 Hz/µm. In reality, this calculation makes assumptions that are unlikely to be true.
The most important one is that the etching is uniform over the cavity surface, but etching is
a quite complex and dynamic process, involving both fluid dynamics and thermodynamical
issues. What can be said is that, in an ideal etching setup with very good temperature
and flow control, the achieved frequency shift should approach the simulated value. Of
additional interest is using etching as a way to fine-tune the frequency of a cavity. This
process is called differential etching, etching shift versus the acid fill level can be seen in
Figure 27 for an example HWR. This process has been demonstrated and used for cavity
production for QWRS at TRIUMF [37] and elsewhere. From this plot, it can be seen that

Figure 27: Plot of simulated etching coefficient for a prototype β = 0.53 HWR filled with
acid to different levels. The blue diamonds represent simulation data, and the red squares
are this data mirrored assuming symmetry.

a much more positive frequency shift can be achieved if the cavity is filled such that the
full electric field region is etched while only one of the magnetic field regions are etched.
Experience gained from fabricating a cavity design is required to determine if such a cavity
tuning process would need to be developed.

53
8 Coupled Electromagnetic and Mechanical Simulations
A mature cavity design starts with an electromagnetic model, but must account for various
practical concerns. During normal operation, the cavity walls may be subjected to small
forces, which will result in small perturbations to the shape of the cavity. This deformation
will change the cavity vacuum space, and therefore the resonant frequency. This can
have significant controls implications and must be evaluated and optimized before cavity
production. Because matching the simulation setup to the exact conditions the cavity will
see in operation is difficult, several rounds of simulation and prototyping can sometimes be
required to achieve the accuracy and performance desired. In this section, the procedure for
this type of simulation, the types of pressures simulated, and the results of these simulations
for the HWR developed in the last section will be discussed in detail.

8.1 Generalized Procedure

The mechanical simulation process has many parallels to the electromagnetic simulation
procedure. The major steps remain Geometry Creation/Setup, Meshing, Solving, and
Post-Processing. From the cavity vacuum geometry, the SHELL command in SolidWorks
will create the niobium space that represents the cavity walls. This command has the
option to shell with multiple thicknesses in different parts of the cavity, and this can be
used to model cavities made from materials of various thicknesses. From this, the helium
vessel can be constructed around the cavity, using different bodies for different materials.
Once the material space has been generated, it is exported as a SAT (or other appropriate
file type) file for importation into ANSYS. Once in ANSYS, the different bodies must be
“glued” together. This command (“VGLUE”) takes different volumes that are next to each
other and joins them for simulations purposes. Before this command is given, two volumes
may be touching, but where they touch they both have their individual surfaces with no
reference between them. After the command, the redundant surfaces are removed so that
the different volumes share a common surface, meaning that the mesh and mechanical
forces will be shared across the boundary, effectively binding them together. From the
material space of the cavity, seen in Figure 28, the vacuum space must be recreated. This
is done by using the existing geometry as the base for the remaining areas needed to create
the continuous set of areas that define the vacuum space: symmetry faces and ends of
ports. Once these are created and the vacuum space has been created, all volumes must be
meshed. Because there are volumes that share the same areas, most notably the vacuum
space and the cavity space, special care must be taken to mesh in the correct order with the
correct refinement to give a satisfactory mechanical and electromagnetic solution. For this
purpose, it is highly recommended to use second order elements for the electromagnetic
simulation. Using the very fine mesh required for a satisfactory electromagnetic solution
will almost certainly force there to be an extremely high number of elements in the mechan-
ical simulations, unnecessarily stretching the limits of even a very high powered computer.

54
Other than this caveat, the procedure for performing the electromagnetic simulation is the
same as described earlier in this document. Once the frequency is simulated, the next step

Figure 28: Material space after importation into ANSYS-APDL.

is the setup of the mechanical simulation. This includes giving all volumes the appropriate
material properties, including the vacuum space. This may seem counter-intuitive, but
treating the vacuum space like a material is essential for accuracy. This is because this
process preserves the meshing in the vacuum space, allowing it to be perturbed by the
mechanical deformation, providing a far more accurate result than remeshing after the me-
chanical simulation. The material properties used for the various mechanical simulations
can be seen in Table 1, all of which are for cryogenic temperatures.
The remaining steps before starting the simulation are applying the pressure to be
studied and the boundary conditions. The forms of the pressures will be discussed later in
this section. The boundary conditions are of critical importance for this type of simulation.

55
 Table 1: Material Properties used for Simulation.

Elastic Moduli Major Poisson’s Ratio Mass Density

Pa - kg/m3
“Vacuum-ium” 1 0.38 0.01
titanium 1.23E11 0.338 4510
niobium 1.25E11 0.301 8570
304 Stainless Steel 2.00E11 0.33 7860
niobium titanium 1.24E11 0.32 5700

First, any faces on symmetry planes must be constrained to stay in that plane, i.e. fixing
the variable defining the plane while letting the two other degrees of freedom float. Then
the simulation must be fixed in space. If you are using three symmetry planes, then this
is automatically done. If two or less planes of symmetry are being used, at least one part
of the cavity (a point that is on all symmetry planes used) must be fixed. While this will
not change the simulation result, it will prevent the solution from diverging if the applied
force isn’t symmetric. Additionally, other parts of the geometry can be fixed, based on
the situation you are trying to simulate. An example of this is the relationship between a
HWR and its tuner. While simulating how much force the tuning will require, the beam
ports cannot be fixed, but when doing some other simulation that assumes the tuner is in
place, fixing the beam port flanges in their normal direction is a good approximation. If
more detailed studies are required an effective spring boundary condition can be applied
to simulate, for instance, a weakly compliant tuner. After solving for the deformation, the
meshing in the material space can be discarded, and the vacuum space material should
be reset to true vacuum. The deformation should be kept by updating the coordinates
of the mesh nodes. Once the simulation is switched back to modal from structural, the
standard electromagnetic simulation procedure should be used to get the new frequency.
The difference between the first and second frequency is the frequency shift for the applied
pressure.

8.2 Geometry Modification for Simulation

Without careful consideration of the geometry, the number of elements in a mechanical
simulation can become needlessly large. Therefore, serious thought must be put into a
cavity solid model before simulations can proceed. A mechanically designed model will in-
clude many small surfaces/features that are required for fabrication, but will make accurate
simulation effectively impossible. It is necessary to make certain geometrical changes to
the model before it is simulated to ensure both accuracy as well as efficiency. Some of these
geometry features will not exist in the finished cavity, but are required for fabrication. An

56
example of this would be a small cut in preparation for welding of two subassemblies that
will be obliterated by the welding process, leaving a much smoother surface. This type
of feature must be modified to accurately represent the final geometry as well as possible.
In non-sensitive areas, this type of feature can be safely removed entirely, which will give
great benefit in simulation performance with almost no sacrifice in the result’s accuracy.
The second type of geometry feature that can be changed are small details that are totally
unrelated to electromagnetic performance and have little to no impact on mechanical per-
formance. Excellent examples of this type of feature are bolt holes and chamfers on parts
of the cavity vessel. These features will exist in the finished cavity, but do not affect the
cavity vacuum and do not change the mechanical properties of the cavity assembly. These
features are often quite small as well, leading to dense local mesh that is not required.
Removing these two types of features can drastically improve simulation efficiency by re-
ducing meshing requirements, an example of this process can be seenb in Figure 29. It
is, indeed, even common for these types of features to forestall any simulation because of
the complexity of meshing them. Many of these features can include overlapping volumes
or other small/non-physical features that will need to be fixed before any work can be
done. Additionally, even if they can be meshed, the density required may be so high as
to exceed the limits of the available simulation hardware. Additionally, simplifications can
be made to allow estimates of less critical figures of merit. An excellent example of this is
the removal of bellows from a mechanical simulation. Their removal greatly simplifies the
model by removing all of the associated small/delicate features while allowing relatively
accurate simulations. This is achieved by replacing the bellows with a contact body with
an effectively equivalent spring constant, or depending on the level of deformation under
study, they can be safely removed all together. This modification must include more careful
deliberation of the consequences involved, and the increased error of this modification. Ad-
ditionally, small gaps between volumes must be removed to allow “gluing” during meshing.

8.3 Accuracy Level

Because the mesh is kept during the mechanical simulation, the simulated frequency shift
is extremely accurate. This is crucial because some of the effects being simulated only shift
the frequency by tens to hundreds of hertz. As can be seen in Figure 30, these types of
shifts can be accurately simulated to and below this level.

8.4 An Aside: Slater’s Theorem

Cavity perturbation analysis is an extremely powerful technique for cavity design and
simulation. An analytical description of perturbations in resonant frequency due to per-
turbations in cavity geometry was first developed by Slater [40]. His description is built on
classical resonator invariants which are constant under adiabatic changes to the resonator.

57
Figure 29: Left: Flange before simplification for simulation. Bolt holes are indicated by
red arrows, fillets and chamfers by green arrows, and a complex feature (vacuum seal
knife edge) is indicated by a blue arrow. Right: Simplified flange ready for inclusion in a
simulation.

This assumption is valid under almost all cavity design considerations where the time scale
of the changes to be studied are at least five orders of magnitude longer than the RF period.
For the classical harmonic oscillator, the stored energy at any time is given by

ω 2 mx2 p2
U= + (160)
2 2m
where p and x are canonical variables. This means that the area of the ellipse described
by this equation is an adiabatic invariant. Modifying this equation into the standard form
of the ellipse
x2 p2 x2 p2
1 = 2 + 2 = 2U + (161)
a b ω2 m
2mU
we find the area of the ellipse is
U
A = πab = (162)
f
where f is the resonant frequency. Given that A is constant, we find that
U df
dU = (A)df = . (163)
f

58
Figure 30: Frequency shift versus applied pressure for a mechanical simulation. Note the
linearity down to very small frequency shifts.

Using the known form of dU from Maxwell’s Equations and assuming a finite (if small)
change in volume (∆V ), we get Slater’s Theorem:
∫
∆f ∆U 1 ( )
= = ϵ0 E 2 − µ0 H 2 dV. (164)
f0 U 4U ∆V

8.5 Pressure Sensitivity

During operation, the cavity will have a pressure differential across its walls. This is
because, while the vacuum space is at extremely low pressure, the helium space is at ∼ 28
torr (the pressure of 2 K liquid helium). This pressure differential deforms the cavity,
shifting the cavity frequency. The deformation can be seen in Figure 31. As expected,
the largest deformation occurs in the weakest places, the outer conductor between the RF
and beam ports and the inner conductor just above the beam port. Immediately applying
Slater’s Theorem we can see that deformation in the electric field region will give an
opposite contribution to the frequency shift to the shift from deformation in the magnetic
field region. The LC circuit model of the cavity gives an intuitive reason for this scaling.

59
Figure 31: Left: Deformation of the inner conductor due to pressure differential. Right:
Deformation of the outer conductor due to a pressure differential (5000 Pa).

In this model, the frequency of the cavity is given by

1
ω2 = (165)
LC
where L is the inductance and C is the capacitance. Removing a small volume in the
high magnetic field region will decrease the inductance, increasing the resonant frequency.
Removing a small volume in the high electric field region will increase the capacitance,
reducing the cavity frequency.
This behavior means that, if modification is necessary, stiffening can be used to balance
these two contributions, resulting in a near-zero frequency shift coefficient. While a fixed
frequency shift can be accounted for, the helium pressure can only be maintained to ∼ ±1
torr at a frequency of 1 Hz. This fast varying pressure must not vary the cavity frequency
too much, or controlling the cavity in operation becomes quite difficult. This is why the
FRIB specifications require that the final coefficient |df /dP | be below 2 Hz/torr. The bare
cavity gives a shift of −4.07 Hz/torr. This agrees with what would be predicted by Slater’s
Theorem because most of the loss in volume (−dV ) is in the electric field region.

60
8.6 Lorentz Force Detuning
The fields in the cavity will interact with the induced charges and currents on the RF sur-
faces. The detuning from these forces is therefore called Lorentz Force Detuning (LFD) [41].
Again using Slater’s Theorem, we find that
∫ ∫
∆f 1 ( ) 1
= ϵ0 E − µ0 H dV = −
2 2
(P )dV (166)
f0 4U ∆V U ∆V

where P is the LFD pressure. It is worth noting that the term (P )dV is always a positive
quantity, a change in sign of the pressure P changes the direction of deformation, changing
the sign of dV . This means that any contribution to this frequency shift will be negative.
This means that the compensation techniques that could be used for correcting the pressure
sensitivity do not apply here, so we must rely on straightforward stiffening of the cavity.
The LFD frequency shift scales with the stored energy in the cavity, and therefore the field
level squared. Thus, the LFD coefficient, KL , is defined as

∆f Vacc,βopt
KL = ; Eacc = . (167)
(∆Eacc )2 βopt λ

This type of detuning is well modeled by the Duffing Equation, discussed earlier along
with the basic controls implications of this effect. Because the FRIB HWRs are to be
run with a duty factor of 1, this effect is not the dominant concern it can be for pulsed
machines [42, 43], so the use of advanced compensation methods and control systems is
not anticipated for FRIB. FRIB specifications do, however, call for |KL | to be below 3
Hz/(MV/m)2 for the HWRs. The bare cavity, as designed earlier, gives an initial KL =
−3.06 Hz/(MV/m)2 . The deformation resulting from this force can be seen in Figure 32.

8.7 Cavity Tuning

For the β = 0.29 and 0.53 HWRs, it was decided to tune the cavity in operation by
deforming the beam port cups by application of pressure on the beam ports themselves.
The design of this tuner requires an estimate of the physical range and force required to
give the FRIB specified tuning range of 150 kHz. For tuning measurements, all distances
and forces given are for movement of one port while the frequency shift is for both ports
moves symmetrically. Simulation of this effect is done by applying force to the beam port
flange, and gives coefficients of df /dF = −126 kHz/kN and df /dx = 599 kHz/mm for
full tuning range parameters of 150 kHz, 0.25 mm, and 1.185 kN. While this does provide
a potentially large tuning range, the resolution of the tuner is also of great importance.
For the controls, the desired frequency resolution of the tuner is 1 Hz, meaning that the
tuner must have a resolution of ∼ 2 nm or ∼ 8 mN. It should be noted that this frequency
sensitivity could have been estimated from the earlier beam port cup sensitivity study.
This predicted a frequency sensitivity of 820 kHz/mm, and it is expected that a deformed,

61
Figure 32: Deformation of the cavity with the Lorentz Force applied at full accelerating
voltage.

as opposed to fully moved, beam port cup would be slightly less sensitive. The deformation
from the applied tuning pressure can be seen in Figure 33.

9 Cavity Testing Theory and Setup

A critical part of cavity design is feedback from prototyping and testing of realistic cavities.
This is because cavity performance is a combination of RF design, material selection and
handling, fabrication technique, cavity processing, and other, more subtle effects. There
are a many steps in the fabrication process that must be taken into account in the design
process that can only be quantified on a real cavity, such as weld shrinkage, coupling
strengths, superconducting surface resistance (and therefore quality of cavity processing),

62
 Figure 33: Simulation of cavity deformation for an applied tuning force.

and many others. Cavity testing measures the cumulative effect of all of these factors. In
this section, the theory and setup of a cavity test will be discussed. Additionally the data
measured during a specific HWR test will be examined along with the details of translating
this data into relevant cavity parameters.

9.1 Testing Theory

The basic theory of a cavity test centers around measuring the quality factor of the cavity
at a variety of stored energies. While the cavity is designed to optimize many figures
of merit, most of them cannot be directly measured during a cavity test. These include
peak surface fields, R/Q, βopt , and the Geometry Factor, which must be inferred using
simulation data. A cavity test is performed with two power couplers, one close to matched
to minimize the required forward power (“input” coupler) and one very weakly coupled

63
(“pickup” coupler). Power is coupled to the cavity with the input coupler, while the pickup
coupler is used strictly as a diagnostic tool. Cavity tests are done with a phase lock loop
matching the driving frequency to the cavity resonance. The basic testing phase lock loop
arrangement can be seen in Figures 37 and 38.
This RF circuit allows a signal generator to track the cavity resonant frequency by
mixing the transmitted power signal with the forward power signal, giving a DC signal
whose amplitude is based on the difference between the driving frequency and the cavity
resonant frequency. With the cavity driven in steady-state on resonance, only the cavity
frequency (f ), forward power (Pf ), reverse power (Pr , vector sum of the reflected and
emitted traveling waves from the input coupler), and the transmitted power (Pt , from the
pickup coupler) can be measured. Using equations from earlier

Pr
|S11 |2 = (168)
Pf

Pt
|S21 |2 = (169)
Pf
|S21 |2 Q0 Pt
β2 = = = (170)
1 − |S11 | − |S21 |
2 2 Qext,2 Pd
1 ± |S11 | Q0 Pr
β1 = = = (171)
1 ∓ |S11 | Qext,1 Pd
and
Q0 = QL (1 + β1 + β2 ) (172)
we see that we don’t have enough information to calculate Q0 and U . To do this, we must
determine the QL and the coupling (whether β1 is greater or less than 1). Both of these
can be obtained from modulating the forward power. The most straightforward method is
to suddenly shut off the forward power, so that the equation for the stored energy in the
cavity becomes
dU ωU
= −PL = − (173)
dt QL
which gives ( )
ωt
U (t) = U0 exp − . (174)
QL
For a sufficiently large QL (true for most Dewar tests), a decay time τL can be measured
for a given U/U0 (usually about a −6 dB fall), giving QL . Additionally, the transitory
response of Pr determines whether the cavity is over or under coupled. This effect can
be seen in Figure 34. In this way, we have enough information to calculate not only β1
and β2 , but also Q0 . Given this, we can calculate Qext,2 , allowing us to transform our
measurement of Pt into a measurement of U . While it is often assumed that Qext,1 and

64
Figure 34: Transient behavior of the reverse power for different coupling strengths. All
curves are in response to the shown forward power behavior. Image credit to Tom Powers,
TJNAF.

Qext,2 do not change dramatically over the course of a Dewar test, it is recommended to
periodically repeat the modulated measurements to ensure these values have not drifted
during testing. With knowledge of Qext,2 , every set of power measurements can be used to
directly calculate Q0 and U :
Qext,2
Pd = Pf − Pr − Pt ; U= Pt (175)
ω
where the first equation is the cavity energy balance, and
ωU
Q0 = . (176)
Pd
This method, known as the “direct” method, has the disadvantage of using Pr extensively.
Especially far from unity coupling, Pf − Pr is likely to be quite small, and error prone.
Fortunately, there is another method for calculating Q0 and U . While the first method
does not require the use of Qext,1 , this method uses this previously measured number in
place of Pr . This method is known as the “indirect” method, because, while measuring
Qext,1 involves Pr , this usage makes it less sensitive to the uncertainty of its measurement.
Plugging the definition of Qext,1 into our equations, and making the usual assumption that
β2 ≪ 1, we find that
Qext,1
Q0 ≈ √ . (177)
Qext,1 Pf
2 Qext,2 · Pt − 1

65
It should be noted that once Qext,2 and Qext,1 are calculated with modulated measurements,
they are assumed to be constant for following measurements. This, however, makes each
CW measurement over-specified, and this can be used to check the measurement’s accuracy.
In principle, both the direct and indirect methods should give the same answer. Starting
from
Qext,2 β1
= . (178)
Qext,1 β2
it is straight-forward to derive (assuming, as usual, that Pt is small compared to Pf )
√
Qext,2
|S21 | = 1 ± |S11 |. (179)
Qext,1

Plotting the left side of this equation versus the right side of the equation using the data
(|S21 | and |S11 |) and the assumed values for Qext,2 and Qext,1 gives an easy graphical
method of checking the consistancy of the measured valueds. Depending on the coupling,
|S11 | will vary from 1 to −1, so the plotted data should, ideally, lay on a simple triangle.
Thus, the plot of these quantities is an excellent way to test whether the values of Qext,1 ,
Qext,2 , Pr , Pf , and Pt are consistent. Any systematic deviation between the plotted data
and this “Duality Triangle” indicates a problem with the data. This generally happens
when the assumption of constant coupling is invalid, which is the case for the test presented
in Section 4.4, a dramatic shift in coupling between test day 1 and test day 2 had to be
accounted for.

9.2 Measurement Procedure

For a given cavity configuration, there are generally two phases to measuring the cavity’s
parameters. The first consists of the modulated measurements required to calibrate the CW
measurement, and this is generally done at low field levels to avoid dynamic behavior that
can occur at higher fields. After the field level is set, generally around Epk ∼ 1 MV/m,
the phase of the transmitted power signal going to the mixer is adjusted to maximize
Pt . This is important because, while there is a phase shift from driving the cavity off
resonance, there is also a phase shift from the cable and devices used to do the measurement.
Having a mechanism for shifting this phase (physical or electronic) is crucial not only for
compensating the overall phase difference between Pt and the signal sampled from Pf
(modulo 2π), but also for fine-tuning before every measurement. This is because the phase
of the circuit can (and will) drift during the course of the cavity test because of thermal
or mechanical changes in the components and transmission lines. For a perfectly adjusted
circuit, the cavity is driven directly on resonance, maximizing U and therefore Pt for a
given Pf . Having properly adjusted the phase, the forward power is shut off, and the
dissipated power is plotted versus time, generally on a spectrum analyser. The decay time
is recorded for a given power difference (generally −6 dB), and the initial Pt is recorded.

66
It is important to accurately correlate this measurement with a set of CW measurements
taken immediately after the modulated measurement. Combining these modulated and
CW measurements allows the calculation of Qext,1 and Qext,2 as detailed in the previous
subsection, and these are used to interpret the following CW measurements.
The second phase of measurement is to perform CW measurements at different Pf . The
forward power is increased in steps of 1 − 3 dB, and several CW data points are recorded at
every step, after adjusting the phase shift. The higher the cavity fields, the smaller the steps
in Pf should be, to get the best resolution in this region. This is because the most dynamic
behavior happens here, and sudden shifts in Q0 can occur. This is generally continued until
limited either by cavity quench, maximum available Pf , or high radiation levels from field
emission. It should be noted that many values are recorded with each data point, not
just power levels. This includes data such as date and time, readings from temperature
and pressure sensors, liquid helium level, and more. Any interesting phenomena are then
investigated, time permitting, such as the detailed behavior of sudden drops in Q0 , field
emission onset levels, quench field level, etc. If desired, time can be spent conditioning
the cavity at high fields in an attempt to improve performance. This is most generally
done when there is high radiation levels in an attempt to destroy or otherwise shut off the
serious field emission sites causing the radiation. In cases of high radiation, staying at high
power or pulsing to even higher power can be successful in improving performance, allowing
access to higher field levels. Once all of the desired data are collected for a given cavity
configuration, then the configuration is changed and the process is repeated. The nature
of the cavity configurations depends strongly on the exact phenomenon being studied, but
a common example of this would be measurement at different temperatures as the cavity
is pumped down to 2 K.

9.3 Calibration of the RF Network

It is important to remember that the RF circuit has losses at almost every point. This
means that the powers measured by the power meters (Cal Points E, F, and G, seen in
Figure 35) have additional losses in them from the RF circuit compared to the values we
need to calculate the cavity parameters. These power values would, ideally, be recorded
at the cavity (Cal Points B and C). Calibrating the RF circuit is the process of measuring
the effective losses between these locations. Calibration measurements are all done using
a Vector Network Analyzer (VNA), carefully calibrating the measurement equipment at
each step. All measurements are done at four frequencies: the design frequency, the design
frequency ±0.5 MHz, and the cavity frequency as measured during the cavity test. It
should also be noted that all losses should be measured in decibels, this allows direct
addition of losses in series. This allows interpolation of the calibration for the test data
as required. This calibration takes place in two phases for logistical reasons. While all
parts of the RF circuit outside the Dewar are room temperature and can be considered
relatively stable if left in place, the transmission lines and connections inside the Dewar

67
Figure 35: Basic RF circuit for a phase lock loop for cavity testing with the calibration
points marked.

must be calibrated while the cavity is cold because of the significant thermal shifts. This
is generally measured directly after the completion of a cavity test, while the circuit is as
thermally stable as possible. This is done by connecting Cal Point A to Port 1 of a VNA
though a calibrated cable, then measuring the round-trip losses (S11 ), repeating for Cal
Point D. These values are called the “Cold Calibration”. The rest of the process is called
the “Warm Calibration” because it involves measurement of the rest of the RF circuit,
which is at room temperature. The first step for this is calibration of the Pt and Pr power
sensors. Because we have already measured the losses inside the Dewar, all that remains is
measuring the losses from Cal Point D to Cal Point E for Pt and from Cal Point A to Cal
Point F for Pr . These losses will include the cable losses as well as the attenuation from the
directional couplers used to sample these signals. This measurement is done by connecting
the Dewar end to Port 1 of the VNA and the power sensor end to Port 2 of the VNA, and
measuring the losses (S21). The final calibration for these power sensors is the warm losses
plus half of the appropriate round-trip cold losses. The most complex calibration is for
Pf . This is because what we desire is the forward power at Cal Point B (again, what the
cavity actually sees) as measured at Cal Point G for a given power at Cal Point H (right

68
out of the amplifier). The first step accounts for the losses from the amplifier to the cavity
by connecting Cal Point H to Port 1 of the VNA and Cal Port A to Port 2 of the VNA,
then calibrating out the through losses. Port 2 of the VNA is then moved without changing
the calibration to Cal Point G while terminating Cal Point A with a 50 Ω load, where the
power loss (S21) is measured. The power measured, though, is the forward power at Cal
Point A seen at Cal Point G. This means that, for the final calibration of Pf , the warm
calibration number must have half the appropriate cold calibration subtracted to properly
reflect the forward power at the cavity.

10 Challenges for Cavities in Testing and Operation

There are many practical concerns to take into account when designing a cavity besides
the standard figures of merit. Based on the geometry, it is possible (almost impossible to
suppress entirely, in practice) to have stable resonant trajectories for stray electrons in the
cavity (themselves almost impossible to suppress). This behavior can severely complicate
machine operation and even limit achievable field and quality factor. Beyond this, pro-
cessing and fabrication are imperfect processes, and can introduce their own operational
complications. This section will discuss these more common challenges that are often
encountered while testing or operating a cavity and their mitigation.

10.1 Multipacting
The name multipacting is a compound word used to shorten “multiple impacting”, and this
name is a good description of this phenomenon. Stray electrons in the cavity vacuum space
interact with the cavity fields and strike the cavity walls. The impact can generate more
electrons emitted from the surface, which will also have some trajectory based on their
energy, angle of emission, and RF phase. The number of electrons emitted is quantified by
the Secondary Emission Yield (SEY), which depends on many factors including electron
energy, surface preparation, and impact angle. The electron’s trajectory also extracts
energy from the cavity, depositing it into the cavity walls. Based on the surface it strikes,
the energy of impact, and the angle of impact, the average number of ejected electrons
varies, and can be over 1.4 [46], even in a well processed cavity. The energy this cascade
extracts is trivial, in principle, until it is sustained for many generations. If the trajectories
of the electrons becomes closed, i.e. return to near the original emission point (and RF
phase), the cascade continues, eventually consuming all additional RF power injected into
the cavity.

10.2 Avoiding Multipacting Before Testing/Operation

There are multipacting considerations both in design and operation. In the design, care
must be taken not to introduce the conditions for strong multipacting, mostly characterized

69
by regions of parallel surfaces, especially items like couplers, plungers, and stub tuners.
The strongest multipacting occurs when there is a wide range of trajectories in location,
energy, and RF phase that satisfy the resonant condition. These barriers are called “hard”
barriers because they are extremely difficult to get beyond in testing/operation. Particle
tracking simulations [46, 47] for predicting multipacting have been increasing in accuracy
in recent years, and have become reliable for predicting and avoiding strong multipacting
barriers in cavity designs. However, because of the perfection of the models used, these
simulations often find far more barriers than will present themselves in practice. Fabrication
errors are often beneficial, in this case. The random errors added to the surface often
break the resonant condition for the weaker multipacting bands, although this is not a
strong enough benefit to loosen manufacturing tolerances. Additionally, careful surface
preparation during cavity fabrication and processing can reduce the SEY, automatically
narrowing multipacting bands. For HWRs of this geometrical type, multipacting studies
have been done [46] and found no significant barriers (besides the rinse port plungers,
which were later removed from the design). This has been born out in testing where no
significant multipacting has been encountered.

10.3 Dealing with Multipacting During Testing/Operation

Once encountered in operation, there are two common strategies to overcome this limita-
tion. The first involves rapidly adjusting the RF phase, frequency, or forward power to try
to “jump” past the barrier. Because multipacting is a resonant behavior, perturbations to
the conditions in the cavity can disturb the electron cascade long enough for the cavity to
be brought to a higher power, ending the resonant condition. This is easier in operation
than in testing because the much higher coupling available means that the cavity field can
be ramped much more rapidly. It is uncommon to have significant multipacting barriers at
high fields, so this method will often allow testing of the cavity high field behavior. Care
must be taken to not reduce the cavity fields below the barrier, or risk becoming trapped
in the barrier again. The second strategy for multipacting is reducing the local SEY. This
is achieved by allowing the multipacting barrier to condition itself. The electron bom-
bardment will cause gasses to desorb from the surface, generally decreasing the strength
of the multipacting band. While this process has obvious limits, strong barriers encoun-
tered at the beginning of testing can be significantly weakened to allow testing to proceed.
Any venting of the cavity or further processing includes the danger of reintroducing these
barriers.

10.4 Field Emission

Field emission is caused by a stray electron in the cavity gaining energy from the cavity
fields and impacting the cavity wall with enough energy to generate an X-Ray instead
of more electrons. Because this is not a resonant behavior, it requires a different source

70
of electrons, usually particulate (“field emitters”) on the cavity surface not removed by
processing. These field emitters locally enhance the electric field, causing a stream of
electrons to flow into the cavity at high fields. Because of this behavior, field emission
is a problem at high fields, unlike multipacting. At low fields, the fields on the emitters
aren’t enough to generate large numbers of electrons, and they don’t gain enough energy
from their trajectory to generate significant numbers of X-Rays. As the fields in the cavity
increase, both the number of electrons and the energy they gain increase until cavity
performance begins to suffer. The energy extracted by the electrons during their transit
can significantly reduce the cavity quality factor, and can limit achievable fields by requiring
more RF power, both from extracted energy and increased coupler mismatch to the cavity.
If the field emission is strong enough, the energy deposited by the electrons on the cavity
wall can be enough to cause a local thermal quench. Once the superconductivity is broken
locally, the cavity stored energy is rapidly dissipated there, turning the field emission off.
Field in the cavity can be restored once the wall has cooled and become superconducting
again. There are several common ways of mitigating this problem. During processing,
high-pressure rinsing with ultra-pure water forceably removes larger particulate from the
surface, significantly reducing the sources for electrons at high fields, while clean room
preparation and assembly reduces reintroduction of these particulate after cleaning. During
operation, the field emitters can be used to process themselves. By operating close to the
limit of operation (limited by available power or radiation limits), often pulsing to higher
fields as well, enough power can sometimes be introduced to strong emitters to cause
their destruction. This process works best for small numbers of strong emitters, and has
obvious limits. While strong emitters can be suppressed by careful fabrication, processing,
and testing, stray electrons can never be eliminated from the cavity. With a non-zero
population of electrons in the cavity from things such as residual gas, cosmic rays, and
weak field emitters, it is common to see minor field emission at very high fields. Care must
be taken during design of testing facilities to include adequate radiation shielding to allow
testing of even the worst cavities.

10.5 Q-Disease
Hydrogen is introduced into the cavity material at many points along the cavity processing.
The strongest source is etching, especially if the temperature of the acid blend used isn’t well
controlled. The hydrogen itself is not thought to degrade cavity performance. However,
between 40 K and 120 K [48], niobium reacts with hydrogen to form several kinds of
hydrides on the surface of the cavity. These hydrides are dielectric, and extremely lossy.
In operation, the quality factor of the cavity is severely reduced, and degrades further
as the field is increased. The formation process of these hydrides is relatively slow; if
the dangerous temperature region is passed through in under an hour [48], little to no
degradation is seen. This results in interesting requirements for the design of cryomodule
cryogenic systems, a balance between efficient use of cryogenics and rapid cooling of the

71
cavities through the dangerous temperature region. Additionally, a degassing bake can
be used to drive off the hydrogen from the material, significantly reducing the danger of
Q-Disease. This bake is generally done at 600◦ C for roughly 10 hours in high vacuum.
Once observed in testing or operation, there is no known remedy for Q-Disease besides a
warming to 200 K and more rapid cooling cycle.

11 Electromagnetic and Mechanical Simulations

Three dimensional computational simulation provides a powerful tool during the design of
accelerating resonators. Properly done, both electromagnetic and mechanical simulations
can be performed to great accuracy, allowing exploration of different designs and their
implications before prototyping. Most problems can even be solved on a modern, powerful
desktop computer, although more powerful codes and computer infrastructure exists when
needed. However, the interconnection of simulation and prototyping in the design of a
resonator cannot be overlooked, nor overstated. As the interface between powerful three
dimensional geometry creation programs and flexible simulation software is improved, the
design of these complex structures will become more rapid and accurate. The strength of
simulations lies in the rapid design optimization, the cost-effective analysis of many options.
This relies, however, on the relative simplicity and exactness of the computerized model.
Many real life considerations cannot be simulated, and prototyping must inform these
decisions. This interaction becomes even more important during the design of complex
geometries like HWRs. This is because the analytical formulas and simple two-dimensional
simulations traditionally used for cavity design are not powerful enough to guide this type
of design. The new generation of powerful simulation software must be matched with the
human experience of prototyping and testing these new, complex geometries for future
projects that will rely on them to succeed.

12 On Simulation Scripting and Optimization

Cavity design requires a solid understanding of both the electromagnetic behavior of a
design and the mechanical complexities of fabrication. This knowledge can be used to
design the cavity geometry with the proper variables and features required to efficiently
optimize the design while allowing practical and reliable fabrication. While prototyping al-
most invariably uncover unexpected factors and problems that require further cavity design
iterations, proper parameterization greatly speeds cavity optimization. The optimization
time itself can be greatly reduced with simulation scripting and automation. The scripts
seen in the Appendices required significant time and effort to create, but once completed
they greatly facilitated rapid simulation and optimization. This scripting is possible be-
cause ANSYS-APDL is designed to make geometric references consistant and allows very
flexible geometric reference operations. An excellent example of this flexibility can be seen

72
in this section from Appendix A:
!Calculates Volume parameters
vsum
!Select volume 1
vsel,s,,,1
!Define cavity vacuum as "cavity"
cm,cavity,volu
!Select "cavity"
cmsel,s,cavity
!Select areas associated with cavity vacuum
aslv,s
!Unselect symmetry planes
asel,u,loc,x,0
!asel,u,loc,y,0
asel,u,loc,z,0
!asel,u,,,4
!Define PEC walls
cm,cavwall,area
aplot
This section of code defines two objects for use later in the script: the cavity vacuum volume
cavity and the cavity RF surfaces cavwall. These references are used extensively through
the rest of the scripts to set boundary conditions and for post-processing. It is important to
note the extreme flexibility of the code used to do this. For essentially any cavity type, this
code will work properly if the correct volume number is used in line 4 and the appropriate
symmetry planes are selected in lines 12-14, an almost trivial amount of work. This is
just one example of this scripting philosophy that can be seen in the Appendices, and
it allows rapid generation of the appropriate scripts for a wide variety of cavity designs,
greatly speeding simulation. This means that the time required for parameter scans used
to optimize cavity performance can be reduced to the time it takes for the computation.
Proper scripting of the meshing can also reliably reduce this time as well. Some code
systems like ACE3P are being designed to be fully scriptable, from geometry creation to
post-processing, using high-power computing clusters to speed simulation time. This suite
of software shows great promise for allowing rapid simulation and optimization of even the
most complex geometries. Simulation software that allows scripting and optimization like
this is understandably preferable for serious optimization projects.

References
[1] G. Stupakov, “Lecture notes on Classical Mechanics and Electromagnetism in Accel-
erator Physics”, United States Particle Accelerator School, Revision 20, 2011

73
Figure 36: Both days of testing data plotted on the Duality Triangle to verify self consis-
tency.

[2] B. Van der Pol, J. Van der Mark, “Frequency demultiplication”, Nature, 120, 363-364,
(1927)

[3] J.D. Jackson (1999), “Classical Electrodynamics, 3rd Edition”, John Wiley and Sons
Inc.

[4] H. Padamsee, J. Knobloch, T. Hayes (2008), “RF Superconductivity for Accelerators,

2nd Edition”, John Wiley and Sons Inc.

[5] J.D. Jackson (1999), “Classical Electrodynamics, 3rd Edition”, John Wiley and Sons
Inc., p. 298

[6] A. Facco, V. Zvyagintsev, “Beam Steering in Superconducting Quarter-Wave Res-

onators: An Analytical Approach”, Phys. Rev. ST Accel. Beams 14, 070101 (2011)

[7] E. Zaplatin et al., “Low-Beta Superconducting RF Three-Quarter Wavelength Cavi-

ties”, ESS 02-125-L, April 2002, Juelich

[8] M. Tinkham, “Introduction to Superconductivity, 2nd Edition”, Dover Publications,

Inc., p. 44

74
 [9] M. Tinkham, “Introduction to Superconductivity, 2nd Edition”, Dover Publications,
Inc., Chapter 3

[10] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957)

[11] H. Padamsee, J. Knobloch, T. Hayes (2008), “RF Superconductivity for Accelerators,

2nd Edition”, John Wiley and Sons Inc., p. 88

[12] W. Hartung, “Introduction to Superconducting Cavities: Lecture Notes for USPAS

Course on Beam Measurements and Experiments at NSCL”, Lecture notes, US Par-
ticle Accelerator School, Michigan State University, June 2007

[13] A. Facco, “SRF Low-beta Accelerating Cavities for FRIB”, Lecture notes, Phys 905 -
The Accelerator Physics of FRIB, Michigan State University, October 2011

[14] F. and H. London, Proc. Roy. Soc. (London) A149, 71 (1935)

[15] M. Tinkham, “Introduction to Superconductivity, 2nd Edition”, Dover Publications,

Inc., Chapter 1

[16] W. Meissner and R. Ochsenfeld, Naturwissenschaften 21, 787 (1933)

[17] The Argonne Accelerator History Document Collection,

www.aai.anl.gov/history/project pages/atlas.html

[18] ATLAS Home Page, www.phy.anl.gov/atlas/

[19] Laboratori Nazional de Legnaro Home Page, www.lnl.infn.it

[20] TRIUMF National Laboratory Accelerator Physics,

www.triumf.ca/research/research-topics/accelerator-physics

[21] SPIRAL-2 at GANIL homepage, http://www.ganil-spiral2.eu/spiral2

[22] Stopped and reaccelerated beams at NSCL, www.nscl.msu.edu/exp/sr

[23] Soreq Applied Research Accelerator Facility, http://www.soreq.gov.il/default EN.asp

[24] C. Piel et. al., “Phase 1 Commissioning Status of the 40 MeV Proton/Deuteron Ac-
celerator SARAF”, Proceedings of EPAC08, Genoa, Italy, THPP038

[25] H. Padamsee, J. Knobloch, T. Hayes (2008), “RF Superconductivity for Accelerators,

2nd Edition”, John Wiley and Sons Inc., p. 48

[26] N. Ida, J.P.A. Bastos, “Electromagnetics and calculation of fields”, Second Edition,
Springer Verlag, N.Y., 1996

75
[27] S. Humphries, “Finite-Element Methods for Electromagnetics”,
www.fieldp.com/femethods.html

[28] SolidWorks Corp., Dassault Systems, www.solidworks.com/default.htm

[29] Developed by Spatial Corporation, www.spatial.com

[30] ANSYS Inc., www.ansys.com

[31] CST Computer Simulation Technology AG, www.cst.com

[32] AWR Corp., www.awrcorp.com

[33] ANSYS HFSS, www.ansoft.com/products/hf/hfss

[34] Advanced Computations Department, SLAC National Accelerator Laboratory,

https://slacportal.slac.stanford.edu/sites/ard public/acd/Pages/Default.aspx

[35] W. Hartung, “LEP 1500 Plans Collapse and Other Stories or What I Learned on my
CERN Vacation,” Report NS/RF-92-1701, National Superconductor, 160000 Lincoln
Avenue, Brentwood, Colorado 80523 (1992).

[36] W. Hartung and E. Haebel, “In Search of Trapped Modes in the Single-Cell Cavity
Prototype for CESR-B,” Proceedings of the 1993 Particle Accelerator Conference,
Washington, D.C., May 1993, Vol. 2, pp. 898-900.

[37] V. Zvyagintsev, et. al., “SCRF Development at TRIUMF”, RUPAC 2010, Protvino,
Russia

[38] H. Padamsee (2009), “RF Superconductivity: Volume II: Science, Technology, and
Applications”, John Wiley and Sons Inc.

[39] SRF2011, conferences.fnal.gov/srf2011, 15th International Conference of RF Super-

conductivity, July 25-29, 2011, Chicago IL, USA

[40] J.C. Slater, “Microwave Electronics”, D. Van Nostrand Company, New York, 1950,
pp. 80-81.

[41] “Handbook of Accelerator Physics and Engineering”, edited by A. Chao and M.

Tigner, 3rd Printing (2006), World Scientific Publishing Co. Pte. Ltd., p. 580

[42] M. Grecki et. al., “Compensation of Lorentz Force Detuning for SC Linacs (with Piezo
Tuners)”, Proceedings of EPAC08, Genoa, Italy, MOPP129

[43] R. Carcagno et. al., “First Fermilab Results of SRF Cavity Lorentz Force Detuning
Compensation Using a Piezo Tuner”, Proceedings of SRF2007, Peking University,
Beijing, China, TUP57

76
[44] H. Padamsee, J. Knobloch, T. Hayes (2008), “RF Superconductivity for Accelerators,
2nd Edition”, John Wiley and Sons Inc., p. 120

[45] T. Powers, “Theory and Practice of Cavity RF Test Systems”, Technical Document,
Thomas Jefferson National Accelerator Facility

[46] L. Ge et. al., “Multipacting Simulation and Analysis for the FRIB Superconducting
Resonators Using TRACK-3P”, Proceeding of LINAC10, Tsukuba, Japan, September
12 - 17, 2010, THP092

[47] R.L. Geng, “Multipacting Simulations for Superconducting Cavities and RF Coupler
Waveguides”, Proceedings of the 2003 Particle Accelerator Conference, Portland, OR,
May 12-16, 2003

[48] K. Saito, “Cavity Preparation”, Lecture notes from ILC School 2009

[49] J. Popielarski, FRIB, Private Communications, November, 2011

77
Figure 37: One half of the RF circuit for a phase lock loop for cavity testing (circulator,
isolators, and RF diodes excluded). Letters in Red connect to the other half of the RF
circuit, seen on the next page.

78
Figure 38: One half of the RF circuit for a 79
phase lock loop for cavity testing (circulator,
isolators, and RF diodes excluded). Letters in Red connect to the other half of the RF
circuit, seen on the previous page.