WAVEGUIDES

Waveguides are basically metal pipes carrying electromagnetic waves, usually the mi-
crowaves. In these notes, we shall start with the idealized waveguides without any dissi-
pation of the EM energy — hence no attenuation of the waves, — and then consider the
attenuation in a later section.

To avoid the attenuation, we need two things: (1) The material inside the waveguide —
if any — should be linear and have have real permittivity ǫ(ω), real permeability µ(ω), and
hence real refraction index n(ω). In these notes, I allow for general real ǫ and µ, as long as
they are uniform inside the waveguide. Although in real life, most waveguides are filled with
air, thus ǫ ≈ 1 and µ ≈ 1, while the rest are filled with non-magnetic dielectrics, thus ǫ > 1
but µ ≈ 1. (2) No dissipation of EM energy by the electric currents in the waveguide’s walls,
so we assume the walls to be perfect conductors. As a consequence of perfect conductivity,
the walls have negligibly small skin depth, hence the boundary conditions on the EM fields
at the inner sides of the walls are

Etangent = 0 and Bnormal = 0. (1)

For simplicity, let’s also assume a straight-pipe geometry of the waveguide, although the
pipe’s cross-section can of any shape — round, rectangular, or whatever, — for example

(2)

Such waveguides have a translational symmetry in z direction, so they can carry EM waves

1
running in z direction, the general form of such EM waves being

~ y) eikz−iωt,
E(x, y, z, t) = E(x, ~
H(x, y, z, t) = H(x, y) eikz−iωt. (3)

The (x, y)–dependence of the amplitudes E~ and H

~ here obtains from the Maxwell equations
in which we replace
∂ ∂
→ ik, → −iω, (4)
∂z ∂t
∂ ∂
but keep the transverse derivatives ∂x and ∂y as they are. Such 2D residuals of Maxwell
equations are best written down in 2D vector notations separating the transverse x and y
components
∂ ∂
Et = (Ex , Ey ), Ht = (Hx , Hy ), ∇t = ( , ) (5)
∂x ∂y
∂
from their longitudinal components Ez , Hz , and ∂z → ik. Thus

(∇ · E)3d = ∇t · Et + ikEz (6)

and likewise for the magnetic fields, so the time-independent Maxwell equations

∇ · E = 0 and ∇ · B = 0 =⇒ ∇·H = 0 (7)

become

∇t · Et + ikEz = 0, (M1)
∇t · Ht + ikHz = 0. (M2)

As to the curl (∇ × E)3d , it has longitudinal component

(∇ × E)z = (∇t × Et )2d (8)

and transverse components

(∇ × E)t = ikẑ × Et + ∇t Ez × ẑ = ẑ × ikEt − ∇t Ez (9)

where ẑ is the unit vector in z direction, thus ẑ × (a 2d vector) rotates that 2d vector 90◦ to

2
the left. Consequently, the Induction Law
∂B ∂H
∇×E = − = −µµ0 → +iωµµ0 H (10)
∂t ∂t
becomes in 2D vector notations

(∇t × Et )2d = iωµµ0 Hz , (M3)

ẑ × ikEt − ∇t Ez = iωµµ0 Ht . (M4)

Likewise, the Maxwell–Ampere Law

∂D ∂E
∇×H = = ǫǫ0 → −iωǫǫ0 E (11)
∂t ∂t
(in the absence of conduction currents inside the waveguide) becomes

(∇t × Ht )2d = −iωǫǫ0 Ez , (M5)

ẑ × ikHt − ∇t Hz = −iωǫǫ0 Et . (M6)

The general solutions of the 2D equations (M1–M6) are linear combinations of discrete
modes, each mode having its own relation between the wave number k and the frequency ω.
Specifically — as we shall see later in these notes, — for each mode#ν we have
ω 2 n2
kν2 (ω) = − Γ2ν (12)
c2
where Γ2ν is an eigenvalue of a 2D differential operator, thus
k

(13)

ω
k0 k1 k2 k3 k4 k5

3
Note each mode with Γν 6= 0 has a cutoff frequency

c
ωmin (ν) = Γν (14)
n

below which the mode cannot propagate through the waveguide. Or rather, at frequencies
below the cutoff, instead of a propagating wave

E, H ∝ eikz e−iωt (15)

the mode becomes an evanescent wave

E, H ∝ e−κz e−iωt (16)

exponentially attenuating at the rate

2n
q q
2 2
2κ = 2 Γν − (ωn/c) = 2 (ν) − ω 2 .
ωmin (17)
c

On the other hand, at frequencies above the cutoff, the mode#ν is a propagating wave with
dispersion relation (12), hence phase velocity

s
c ω2 c
vphase = × 2 2 > (18)
n ω − ωmin (ν) n

and group velocity

s
c 2 (ν)
ω 2 − ωmin c
vgroup = × 2
< (19)
n ω n

(assuming n(ω) = const).

Multiple wave modes propagating at different group velocities would mess up any signal
transmitted by the waves. Consequently, when designing a waveguide for the microwaves
carrying signals in a particular frequency range, only one mode — say mode#1 — can

4
propagate through the waveguide while all the other modes quickly attenuate down. In
terms of the cutoff frequencies (14), this means

ωmin (1) < desired range of ω’s < ωmin (ν = 2, 3, . . .). (20)

We shall return to this issue later in these notes once we learn how to calculate the Γ pa-
rameters — and hence the cutoff frequencies — for the modes of rectangular and circular
waveguides. But before we can get there, we need to understand how the dispersion equa-
tion (12) arises in the first place.

So let’s go back to the Maxwell equations in the 2D form:

∇t · Et + ikEz = 0, (M1)
∇t · Ht + ikHz = 0, (M2)
(∇t × Et )2d = iωµµ0 Hz , (M3)


ẑ × ikEt − ∇t Ez = iωµµ0 Ht , (M4)
(∇t × Ht )2d = −iωǫǫ0 Ez , (M5)


ẑ × ikHt − ∇t Hz = −iωǫǫ0 Et . (M6)

Together, eqs. (M4) and (M6) can be solved for the transverse components Et and Ht of
the EM fields in terms of the longitudinal components Ez and Hz . Indeed, let’s move the
transverse components to the LHS of the equations and the longitudinal components to the
RHS, thus
ωµµ0Ht − kẑ × Et = iẑ × ∇t Ez ,
(21)
kẑ × Ht + ωǫǫ0 Et = −iẑ × ∇t Hz .

Next, let’s form linear combinations of these two equations of the form

ωǫǫ0 (first) + kẑ × (second) (22)

and

−kẑ × (first) + ωµµ0(second). (23)

5
Using ẑ × (ẑ × vt ) = −vt for any transverse vector vt , we get

ωǫǫ0 ωµµ0Ht − ωǫǫ0 k ẑ × Et + k 2 (−Ht ) + kωǫǫ0 ẑ × Et

= iωǫǫ0 ẑ × ∇t Ez + ik ∇t Hz ,
(24)
− kωµµ0 ẑ × Ht + k 2 (−Et ) + ωµµ0k ẑ × Ht + ωµµ0ωǫǫ0 Et
= ik ∇t Ez − iωµµ0 ẑ × ∇t Hz ,

where

ω 2 n2
 
2 2

1
ωµµ0 × ωǫǫ0 = ω × ǫµ = n × ǫ0 µ0 = 2 = , (25)
c c2

and therefore
ω 2 n2
 
2
− k Ht = iωǫǫ0 ẑ × ∇t Ez + ik∇t Hz ,
c2
 2 2  (26)
ω n 2
− k Et = ik∇t Ez − iωµµ0 ẑ × ∇t Hz .
c2

The physical consequences of eqs. (26) depend on whether k = ωn/c or k 6= ωn/c, so let’s
work them out case by case.

Transverse Electromagnetic (TEM) Waves

For k = ωn/c — exactly as for a plane wave — the LH sides of eqs. (26) vanish for any
transverse fields Et and Ht , so on the RHS we should have ∇t Ez = 0 and ∇t Hz = 0, hence
Ez (x, y) = const and Hz (x, y) = const. Moreover, at the boundary wall we should have Ez =
0, hence Ez ≡ 0 throughout the waveguide. Likewise, Hz ≡ 0 throughout the waveguide
because any Hz (x, y) = const 6= 0 would lead to a magnetic flux F = µµ0 Hz × area 6= 0
and hence to EMF = iωF 6= 0 in the walls surrounding the waveguide. Thus, the wave in
question is purely transverse, Ez = 0 and Hz = 0, exactly as for a plane wave, and that’s
why it’s called a transverse electromagnetic (TEM) wave.

In the absence of Ez and Hz , eqs. (M4) and (M6) relate the transverse electric and

6
magnetic fields of a TEM wave to each other exactly as in a plane wave,

ẑ
Ht (x, y) = × Et (x, y), (27)
Z

where
r
ωµµ0 µµ0
Z = = (28)
k ǫǫ0

is the wave impedance of the medium filling up the waveguide. Also, in the absence of Ez
and Hz , Maxwell eqs. (M1) and (M3) become

(∇t · Et )2d = 0 and (∇t × Et )2d = 0, (29)

hence

Et (x, y) = −∇t Φ(x, y) (30)

for some scalar potential Φ(x, y) obeying the 2D Laplace equation

∂2 ∂2
 
△2d Φ ≡ ∇t2 Φ ≡ + Φ(x, y) = 0. (31)
∂x2 ∂y 2

The boundary condition for this Laplace equation follows from E⊥ = 0 at the waveguide’s
wall. In 2D terms, this means that at the boundary of the waveguide’s cross-section in the
(x, y) plane, both Ez and the tangential component of Et must vanish. For the TEM wave,
Ez = 0 anyway, while zero tangential component of Et = −∇t Φ means Φ = const along the
boundary of the cross-section.

The solutions to the Laplace equation (31) subject to this Dirichlet-like boundary condi-
tion depend on the cross-section’s topology. Most commonly, the waveguide is topologically
a cylinder — it has an outer conducting wall of some shape but no inner conductors dis-
connected from the outer wall, — so its cross-section completely fills its outer boundary —
which can be a circle, or a rectangle, or whatever, — but has no inner boundaries. In this
case, the Laplace equation has no solutions besides the trivial Φ(x, y) = const, Et (x, y) = 0,
and there are no TEM waves.

7
 On the other hand, the waveguides with both outer and inner conducting walls (discon-
nected from each other so that we may have Φ(inner wall) 6= Φ(outer wall)) do allow for the
TEM waves. For example, in a coaxial waveguide we may have

V0
Φ(ρ, φ) = −V0 log ρ + const, Et (ρ, φ) = ρ̂ , (32)
ρ

and hence a non-trivial TEM wave

V0 ikz−iωt V0 ikz−iωt
E(ρ, φ, z, t) = e ρ̂, H(ρ, φ, z, t) = e φ̂. (33)
ρ Zρ

Transverse Electric (TE) and Transverse Magnetic (TM) Waves

Now consider the wave modes with k 6= ωn/c and hence

ω 2 n2
Γ2 = − k 2 6= 0. (34)
c2

For waves like these, eqs. (26) determine the transverse field components in terms of the
longitudinal components,

i  
Ht = ωǫǫ0 ẑ × ∇t Ez + k∇t H z ,
Γ2 (35)
i  
Et = 2 k∇t Ez − ωµµ0 ẑ × ∇t Hz .
Γ

Plugging in these transverse components into the Maxwell equations (M3) and (M5) leads
to the eigenvalue equations for the longitudinal fields.

△2d + Γ2 Ez (x, y) = 0,


(36)
△2d + Γ2 Hz (x, y) = 0.

Indeed, combining Maxwell eq. (M5) with the first eq. (35), we arrive at

i  
−iωǫǫ0 Ez = ∇t × Ht 2d
= ωǫǫ ∇
0 t × (ẑ × ∇ E
t z ) + k∇t × ∇ H
t z (37)
Γ2 2d

8
where ∇t × ∇t Hz = 0 (because ∇t × ∇t = 0) while

∇t × (ẑ × ∇t Ez ) = ẑ ∇t2 Ez − ∇t (ẑ × ∇t = 0)Ez −→ ∇t2 Ez in 2d sense. (38)

Thus,
i
−iωǫǫ0 Ez = ωǫǫ0 ∇t2 Ez (39)
Γ2

and therefore

∇t2 + Γ2 Ez (x, y) = 0.


(40)

Likewise, combining the second eq. (35) with Maxwell eq. (M3) gives us

i  
iωµµ0Hz = ∇t × Et 2d
= k∇t × ∇ E
t z − ωµµ ∇
0 t × (ẑ × ∇ H
t z ) (41)
Γ2 2d

where ∇t × ∇t Ez = 0 while

= ∇t2 Hz .


∇t × (ẑ × ∇t Hz ) 2d
(42)

Thus,
I
iωµµ0 Hz = − 2
ωµµ0 ∇t2 Hz (43)
Γ

and consequently

∇t2 + Γ2 Hz (x, y) = 0.


(44)

However, while the longitudinal electric and magnetic fields obey similar-looking equa-
tions (40) and (44), they are subject to different boundary conditions (which we shall see in
a moment), so in general they have different eigenvalues Γ2 . Thus, for any particular mode ν
with eigenvalue Γ2ν , we generally have a solution of eq. (40) or a solution of eq. (44), but not
both of them. Consequently, the wave running down the waveguide is either a transverse
magnetic (TM) wave with Ez 6= 0 but Hz = 0, or a transverse electric (TE) wave with
Hz 6= 0 but Ez = 0.

9
Transverse Magnetic (TM) Waves

In a TM wave Hz ≡ 0, the Ez obeys the eigenstate equation

△2d + Γ2 Ez (x, y) = 0,


(40)

while the transverse electric and magnetic fields follow from the Ez (x, y) — or rather, from
its 2d gradient ∇t Ez (x, y) — as

ik iωǫǫ0
Et (x, y) = ∇t Ez (x, y), Ht (x, y) = ẑ × ∇t Ez (x, y), (45)
Γ2 Γ2

cf. eqs. (35) for Hz ≡ 0. In particular, these transverse fields are related to each other as

ẑ
Ht (x, y) = × Et (x, y), (46)
Z

similar to the transverse fields in a plane wave, but for a different wave impedance

s
(ωn/c)2 − Γ2
r
k µµ0 ck
ZTM = = × = Zplane × < Zplane . (47)
ωǫǫ0 ǫǫ0 ωn (ωn/c)2

Now consider the boundary conditions for the eigenvalue equation (40). Let n be a unit
vector ⊥ to the waveguide’s wall while t̂ is a unit vector along the wall, both n and t̂ lying
in the (x, y) plane. In term of these vectors, the boundary conditions

Ek = 0, H⊥ = 0 (48)

at a perfectly conducting wall become

Ez = 0, t̂ · Et = 0, n · Ht = 0. (49)

Fortunately, the last two conditions automatically follow from the Dirichlet condition Ez = 0
and eqs. (45). Indeed, if Ez = 0 all along the boundary, then Ez has zero gradient along the

10
boundary, t̂ · ∇t Ez = 0, and therefore

ik
t̂ · Et = t̂ · ∇t Ez = 0,
Γ2
iωǫǫ0 iωǫǫ0
n · Ht = n · (ẑ × ∇t Ez ) = ∇t Ez · (n × ẑ = t̂) = 0.
Γ2 Γ2

Thus, all of the boundary conditions at a perfectly conducting wall reduce to the Dirichlet
condition Ez = 0.

Consequently, the Γ2 parameters of the TM waves are eigenvalues of the 2D Laplace

operator (or rather −△2d ) subject to the Dirichlet boundary condition,

−△2d Ez (x, y) = Γ2 Ez (x, y), Ez (@boundary) = 0. (50)

Transverse Electric (TE) Waves

In a TE wave the Ez ≡ 0, the Hz obeys the eigenstate equation

△2d + Γ2 Hz (x, y) = 0,


(40)

while the transverse electric and magnetic fields follow from the Hz (x, y) — or rather its 2D
gradient ∇t Hz (x, y) — as

ik −iωµµ0
Ht (x, y) = ∇t Hz (x, y), Et (x, y) = ẑ × ∇t Hz (x, y), (51)
Γ2 Γ2

cf. eqs. (35) for Ez ≡ 0. In particular, these transverse fields are related to each other as

ẑ
Et (x, y) = −Zẑ × H(x, y) =⇒ Ht (x, y) = × Et (x, y), (52)
Z

similar to the transverse fields in a plane wave or a TM wave, but for a different wave

11
impedance

s
(ωn/c)2
r
ωµµ0 µµ0 ωn
ZTE = = × = Zplane × > Zplane . (53)
k ǫǫ0 ck (ωn/c)2 − Γ2

As to the boundary conditions at a perfectly conducting wall,

Ez = 0, t̂ · Et = 0, n · Ht = 0, (49)

the first condition is trivial for a TE wave while the other two conditions follow from eqs. (51)
and the Neumann boundary condition

n · ∇t Hz = 0 (54)

for the longitudinal magnetic field. Indeed, eqs. (54) and (51) immediately imply

ik
n · Ht = n · ∇t Hz = 0,
Γ2 (55)
−iωµµ0 −iωµµ0
t̂ · Et = 2
t̂ · (ẑ × ∇t Hz ) = ∇t Hz · (t̂ × ẑ = −n) = 0.
Γ Γ2

Consequently, the Γ2 parameters of the TM waves are eigenvalues of the 2D Laplace

operator (or rather −△2d ) subject to the Neumann boundary condition,

−△2d Hz (x, y) = Γ2 Hz (x, y), n · ∇t Hz (@boundary) = 0. (56)

12
 Wave Energy and Power

Consider the power carried by the EM waves down the waveguide. Locally, the (time-
averaged) power density is given by the Poynting vector

1
Re E∗ × H 1
Re E × H∗ .



hSi = 2 = 2 (57)

Let’s calculate this Poynting vector for the TM and TE waves. For a TM wave

ik iωǫǫ0
E = Ez ẑ + ∇t Ez , H = ẑ × ∇t Ez , (58)
Γ2 Γ2

hence

iωǫǫ0 ∗ kωǫǫ0
E∗ × H = 2
Ez ẑ × (ẑ × ∇t Ez ) + 4
∇t Ez∗ × (ẑ × ∇t Ez ) (59)
Γ Γ

where

ẑ × (ẑ × ∇t Ez ) = −∇t Ez (60)

and

∇t Ez∗ × (ẑ × ∇t Ez ) = ∇t Ez∗ · ∇t Ez ẑ.

(61)

Altogether,

ωǫǫ0 kωǫǫ0
E∗ × H = −i ∗
|∇t Ez |2 ẑ


2
Ez ∇t Ez + 4
(62)
Γ Γ

where the first term on RHS is imaginary and points in a transverse direction while the
second term is real and points along the waveguide. The imaginary term here describes
oscillations of the EM energy across the waveguide, but they do not contribute to the time-
averaged power flow. Instead, the time-averaged Poynting vector stems from the second

13
term only, thus
kωǫǫ0
hSi = 4
|∇t Ez |2 ẑ, (63)
2Γ
so the net EM power flowing down the waveguide is

kωǫǫ0
ZZ ZZ
Pnet = dx dy hSz i = dx dy |∇t Ez |2 . (64)
2Γ4
cross cross
section section

Likewise, for a TE wave

ik ωµµ0
H = Hz ẑ + ∇t Hz , E = −i ẑ × ∇t Hz , (65)
Γ2 Γ2

hence
ωµµ0 kωµµ0
E∗ × H = i 4
(ẑ × ∇t Hz∗ ) × (Hz ẑ) − 4
(ẑ × ∇t Hz∗ ) × ∇t Hz (66)
Γ Γ
where

(ẑ × ∇t Hz∗ ) × ẑ = +∇t Hz∗ , (67)

(ẑ × ∇t Hz∗ ) × ∇t Hz = − ∇t Hz∗ · ∇t Hz ẑ,


(68)

and therefore
ωµµ0 kωµµ0
E∗ × H = i 4
Hz ∇t Hz∗ + 4
|∇t Hz |2 ẑ. (69)
Γ Γ
Similar to the TM case, the first term in this formula is imaginary and transverse while the
second term is real and longitudinal, so only the second term contributes to the time-averaged
Poynting vector. Thus,
kωµµ0
hSi = |∇t Hz |2 ẑ (70)
2Γ4
and therefore the net EM power flowing down the waveguide

kωµµ0
ZZ ZZ
Pnet = dx dy hSz i = dx dy |∇t Hz |2 . (71)
2Γ4
cross cross
section section

14
 Note similar form of eqs. (64) and (71) for the two kinds of waves,

kω
ZZ 
dx dy ǫǫ0 |∇t Ez |2 or µµ0 |∇t Hz |2 .


Pnet = (72)
2Γ4
cross
section

Moreover, in both cases we have an integral of the form

ZZ
dx dy |∇t ψ(x, y)|2 (73)

where ψ(x, y) — being either Ez (x, y) or Hz (x, y), depending on the wave type — obeys the
eigenstate equation (∇t2 + Γ2 )ψ(x, y) = 0. Consequently, we may simplify the integral (73)
by integrating by parts:

ZZ ZZ
2
dx dy |∇t ψ(x, y)| = dx dy ∇t ψ ∗ · ∇t ψ
(74)
Z ZZ
∗
= dℓ ψ ∇t ψ · n − dx dy ψ ∗ ∇t2 ψ.
boundary

Moreover, the boundary integral here vanishes by the boundary conditions for the Ez or
Hz : Dirichlet for the Ez and hence ψ ∗ = Ez∗ = 0, or Neumann for the Hz and hence
∇t ψ · n = ∇t Hz · n = 0. Thus, we are left with

ZZ ZZ ZZ
2 ∗
dx dy |∇t ψ(x, y)| = − dx dy ψ ∇t2 ψ = +Γ 2
dx dy ψ ∗ ψ, (75)

where the second equality follows from eigenstate equation ∇t2 ψ = −Γ2 ψ. And thanks to
this formula, we may simplify the integral (72) for the net EM power as

kω
ZZ 
dx dy ǫǫ0 |Ez |2 or µµ0 |Hz |2 .


Pnet = (76)
2Γ2
cross
section

15
 This EM power flows along the waveguide at the group velocity
s
2 (ν)
ω 2 − ωmin
c c
vgroup = √ × 2
< (19)
ǫµ ω n

To see that, let’s calculate the net energy of the waves per unit length of the waveguide. The
volume density of (time-averaged) EM energy us

ǫǫ0 µν0
hui = |E|2 + |H|2, (77)
4 4

hence for a TM wave

ǫǫ0  2 2
 µµ0
hui = |Ez | + |Et | + |Ht |2
4  4
ǫǫ0 k 2 µµ0 (ωǫǫ0 )2  
= |Ez |2 + 4 |∇t Ez |2 + × |ẑ × ∇ E
t z | 2
= |∇ E
t z | 2
(78)
4 Γ 4 Γ4
ǫǫ0 ǫǫ0 |∇t Ez |2  2 
= |Ez |2 + × k + ω 2
ǫǫ0 µµ 0
4 4 Γ4
where
ω 2n2
k 2 + ω 2 ǫǫ0 µµ0 = k 2 +
c2
ω 2 n2 2 ω 2 n2
= − Γ + (79)
c2 c2
ω 2n2
= 2 2 − Γ2 ,
c
therefore
  2 2  
ǫǫ0 2 1 ω n 2 2
hui = |Ez | + 4 2 2 − Γ |∇t Ez | . (80)
4 Γ c
Likewise, for a TE wave
µµ0  2 2
 ǫǫ0
hui = |Hz | + |Ht | + |Et |2
4  4
k2 ǫǫ0 (ωµµ0 )2 

µµ0 2 2 2 2

= |Hz | + 4 |∇t Hz | + × |ẑ × ∇ H
t z | = |∇ H
t z |
4 Γ 4 Γ4
(81)
µµ0 µµ0 |∇t Hz |2  2 
= |Hz |2 + × k + ω 2
µµ 0 ǫǫ0
4  4  Γ4
ω 2 n2
 
µµ0 2 1 2 2
= |Hz | + 4 2 2 − Γ |∇t Hz | .
4 Γ c

16
Thus, for both kinds of waves, the time-averaged EM energy density has form

  2 2  
1 2 1 ω n 2 2
hui = |ψ| + 4 2 2 − Γ |∇t ψ| (82)
4 Γ c

√ √
where ψ = ǫǫ0 Ez for a TM wave or ψ = µµ0 Hz for a TE wave.

Integrating this volume energy density over the waveguide’s cross-section, we get the line
energy density

U
ZZ
= hui dx dy
length
cross
section
 2 2  ZZ
1 1 ω n
ZZ
2 2
= |ψ| dx dy + 4
2 2 −Γ × |∇t ψ|2 dx dy
4 4Γ c
hh using eq. (75) for the second integral ii (83)
 2 2 
1 1 ω n
ZZ ZZ
2 2 2
= |ψ| dx dy + 2 2 −Γ ×Γ |ψ|2 dx dy
4 4Γ4 c
 ZZ 
1 2
= |ψ| dx dy × F
4

where

ω 2n2 ω 2 n2 2ω 2n2
 
1 2
F = 1 + 4 2 2 − Γ2 ×Γ 2
= 1 + 2× 2 − 1 = 2 2 . (84)
Γ c Γ c c Γ

Thus altogether,

U ω 2 n2
ZZ
= |ψ|2 dx dy
length 2c2 Γ2
cross
section
(85)
ω 2 n2
ZZ  
= ǫǫ0 |Ez |2 or µµ0 |Hz |2 dx dy.
2c2 Γ2
cross
section

When this EM energy flows down the waveguide with velocity venergy , it transmits the

17
net power

U
Pnet = venergy × . (86)
length

Indeed, the formula for the net power

kω
ZZ 
dx dy ǫǫ0 |Ez |2 or µµ0 |Hz |2 ,


Pnet = (76)
2Γ2
cross
section

we have derived earlier in this section involves exactly the same integral as eq. (85) for the
line energy density, the only difference being the pre-integral factors. Consequently, we may
obtain the velocity of the energy flow from the ratio of those factors:

ω 2 n2 kc2

net power kω
venergy = = = . (87)
energy/length 2Γ2 2c2 Γ2 ωn2

Hence, for

2 ω 2 n2 2 n2 2 2


k = − Γ = ω − ω min (88)
c2 c2

we have
s
k n 2
ωmin
= 1− (89)
ω c ω2

and therefore
s
kc2 c 2
ωmin
venergy = = 1− = vgroup . (90)
ωn2 n ω2

Thus, the EM energy in the waveguide flows with the same velocity as the information carried
by the EM waves.

18
 Rectangular and Circular Waveguides
Rectangular Waveguides

Many waveguides have rectangular a × b cross-sections

b ǫ, µ

(91)

For this geometry, the eigenstates of the (minus) Laplacian operators obtain via the separa-
tion of variables method: We look for

Ez (x, y) = f (x) × g(y) or Hz (x, y) = f (x) × g(y), (92)

hence
(∇t2 + Γ2 )Ez (∇t2 + Γ2 )HZ f ′′ (x) g ′′ (y)
or = + + Γ2 = 0, (93)
Ez Hz f (x) g(y)

and therefore
f ′′ (x) = −αf (x) for a constant α,
g ′′ (y) = −βg(y) for a constant β, (94)
α + β = Γ2 .

For the TM waves, the Dirichlet boundary conditions Ez = 0 on all 4 sides of the rectangle
translate to

f (x = 0) = f (x = a) = 0, g(y = 0) = g(y = b) = 0, (95)

19
so the solutions are

mπx
f (x) = sin for an integer m = 1, 2, 3, . . . ,
a
nπy
g(y) = sin for an integer n = 1, 2, 3, . . . , (96)
b
 mπ 2  nπ 2
2
Γm,n = + .
a b

Similarly, for the TE waves, the Neumann boundary conditions n · ∇t Hz on all 4 sides
translate to

f ′ (x = 0) = f ′ (x = a) = 0, g ′ (y = 0) = g ′ (y = b) = 0, (97)

so the solutions are

mπx
f (x) = cos for an integer m = 0, 1, 2, 3, . . . ,
a
nπy
g(y) = cos for an integer n = 0, 1, 2, 3, . . . , (98)
b
 mπ 2  nπ 2
Γ2m,n = + .
a b

Note that for similar m ≥ 1 and n ≥ 1, the TMm,n and the TEm,n wave modes have exactly
the same Γm,n . However, only the TE waves — but not the TM waves — may have m = 0
or n = 0. Also, for m = n = 0, even the TE wave does not exist: Although

0πx 0πy
Hz (x, y) = H0 cos × cos = H0 = const (99)
z b

is an eigenstate of the −∇t2 operator with Neumann boundary conditions for the eigenvalue
Γ20,0 = 0, it violates other Maxwell equations and boundary conditions. Specifically, such
Hz (x, y) = const 6= 0 would lead to magnetic flux F = µµ0 Hz × ab 6= 0 and hence EMF =
iωF in the walls surrounding the waveguide, in disagreement with the boundary condition
Etangent = 0.

20
 Altogether, the rectangular waveguide has wave modes with cutoff frequencies

r
min c πc m2 n2
ωm,n = √ Γm,n = √ × + 2
ǫµ ǫµ a2 b
πc (100)
q
= √ × m2 + n2 (a/b)2
ǫµ a
for integer m, n = 0, 1, 2, 3, . . . , (m, n) 6= (0, 0).

For b < a, the lowest cutoff frequency belongs to the TE1,0 mode, the next lowest to the
TE2,0 or TE0,1 mode, depending on the aspect ratio a/b, and then we start getting both TE
and TM modes. For a typical aspect ratio a : b = 2 : 1, the first dozen modes in the order
of their cutoff frequencies are

πc
mode cutoff frequency in units of Ω1 = √
ǫµ a
TE1,0 1.000
TE2,0 and TE0,1 2.000
TE1,1 and TM1,1 2.236
(101)
TE2,1 and TM2,1 2.828
TE3,0 3.000
TE3,1 and TM3,1 3.606
TE4,0 and TE0,2 4.000

Consequently:

• For ω < Ω1 , all the wave modes in the waveguide are attenuating rather than propa-
gating, so it cannot carry the waves of that frequency.

• For ω > Ω1 but ω < 2Ω1 , the wave guide has a single propagating mode, namely TE1,0 ,
which is good for carrying signals down the waveguide.

• For ω > 2Ω1 the waveguide has 2 or more propagating modes with different group
velocities. This is bad for carrying signals, but OK for sending a steady MW power
down the waveguide.

21
 To conclude this section, let me write down all the EM field components for the lowest
wave mode TE1,0 :

πx
Hz (x, y, z, t) = H0 × cos × eikz−iωt ,
a
ika πx
Hx (x, y, z, t) = − H0 × sin × eikz−iωt ,
π a (102)
iωµµ0 a πx
Ey (x, y, z, t) = + H0 × sin × eikz−iωt ,
π a
Hy ≡ Ex ≡ Ez ≡ 0.

Circular Waveguides

Now consider a waveguide with a circular cross-section

R
(103)

ǫ, µ

To find the eigenvalues of the (minus) Laplacian operator for this geometry, we separate the
variables in the 2d polar coordinates (ρ, φ), thus

Ez (ρ, φ) = f (ρ) × g(φ) or Hz (ρ, φ) = f (ρ) × g(φ), (104)

hence

ρ2 ρ2 f ′′ (ρ) + ρf ′ (ρ) g ′′ (φ)

0 = × (∇t2 + Γ2 )(f g) = + ρ2 Γ2 + = 0 (105)
fg f (ρ) g(φ)

and therefore

g ′′ (φ) + m2 g(φ) = 0,
1 ′ m2 (106)
f ′′ (ρ) + f (ρ) − 2 f (ρ) + Γ2 f (ρ) = 0 for the same m2 .
ρ ρ

22
Solving the g equation gives us

g(φ) = e±imφ for an integer m = 0, 1, 2, 3, . . . , (107)

and then the solution of the f equation which does not blow up at the center ρ = 0 is the
Bessel function of order m,

f (ρ) = Jm (Γρ) (108)

a special function defined via the integral

Zπ
1

Jm (x) = cos mτ − x cos τ dτ, (109)
π
0

see these notes from an applied math class at the Brown University for more detail.

The value of the Γ in eq. (108) follows from the boundary condition at the waveguide’s
wall at ρ = R: For the TM waves we want Ez = 0 and hence f (ρ = R) = 0, while for the
TE waves we want ∂Hz /∂ρ = 0 and hence f ′ (ρ = R) = 0. Let jm,n be the nth positive root
′
of the Bessel function Jm (x) while jm,n is the nth positive root of its derivative dJm (x)/dx.
Then the TM waves have
jm,n
Γm,n = , (110)
R

while the TE modes have

′
jm,n
Γm,n = , (111)
R

with the corresponding cutoff frequencies

c c ′
ωmin (TMm,n ) = √ × jm,n , ωmin (TEm,n ) = √ × jm,n . (112)
µǫ R µǫ R

′
Unfortunately, there are no closed formulae for the Bessel roots jm,n and jm,n , but you can
find them numerically using Mathematica; here are the tables for m ≤ 5 and n ≤ 5 taken

23
from the Wolfram MathWorls:

n m 0 1 2 3 4 5
1 2.4048 3.8317 5.1356 6.3802 7.5883 8.7715
2 5.5201 7.0156 8.4172 9.7610 11.0647 12.3386
jm,n = (113)
3 8.6537 10.1735 11.6198 13.0152 14.3725 15.7002
4 11.7915 13.3237 14.7960 16.2235 17.6160 18.9801
5 14.9309 16.4706 17.9598 19.4094 20.8269 22.2178

n m 0 1 2 3 4 5
1 3.8317 1.8412 3.0542 4.2012 5.3175 6.4156
′ 2 7.0156 5.3314 6.7061 8.0152 9.2824 10.5199
jm,n = (114)
3 10.1735 8.5363 9.9695 11.3459 12.6819 13.9872
4 13.3237 11.7060 13.1704 14.5858 15.9641 17.3128
5 16.4706 14.8636 16.3475 17.7887 19.1960 20.5755

We see that the mode with the lowest cutoff frequency is TE1,1 , the next lowest being TM0,1 .
Consequently,
c
• For ω < 1.84 × √ǫµR , all the wave modes in the waveguide are attenuating rather than
propagating, so it cannot carry the waves of that frequency.

• For ω > 1.84 × √c but ω < 2.40 × √c , the wave guide has a single propagating
ǫµR ǫµR
mode, namely TE1,1 , which is good for carrying signals down the waveguide.

• For ω > 2.40 × √c , the waveguide has 2 or more propagating modes with different
ǫµR
group velocities. This is bad for carrying signals, but OK for sending a steady MW
power down the waveguide.

Note that for a circular waveguide, the range of frequencies for which only one mode
can propagate is relatively narrow, from Ωmin to about 1.36 × Ωmin . By comparison, the
a
rectangular waveguides with aspect ratios b ≥ 2 have relatively wider ranges, from Ωmin to
2 × Ωmin .

24
 Walls of Finite Conductivity
Thus far, we have assumed that the waveguide’s walls have perfect conductivity. Now
suppose the conductivity σ is finite but high enough that the skin depth

r
2
δ = (115)
µmetal µ0 ωσ

of the wall’s material is much smaller that the waveguide’s diameter or the wavelength λ =
2π/k. This assumption holds true for most waveguides: for example, a WR42 rectangular
waveguide of size 10.7 × 4.32 mm carrying a wave of frequency ω = 2π × 25 GHz has
λ = 14.5 mm, while the skin depth in copper at that frequency is only 0.4 µm.

Boundary Conditions.

At the edge of a perfectly conducting wall, the EM fields inside the waveguide obey
boundary conditions

Ek = 0, H⊥ = 0. (116)

For the imperfectly conducting walls, these conditions are no longer exact, but for the walls
of good conductivity they remain approximately true. Specifically, the boundary conditions
become

H⊥ = O(kδ) × Hk ≪ Hk (117)

and

Ek = O(Zkδ) × Hk ≪ ZHk ∼ E⊥ . (118)

Fortunately, for the purpose of solving the eigenvalue equations for the Γ2 of various wave
modes, these boundary conditions may be approximated by the idealized conditions (116),
so we do not have to redo the analysis of the previous sections.

To derive the boundary conditions (117) and (118), we start by noting that in a good
conductor, the conductance current Jc = σE is much stronger than the displacement current

25
Jd = −iωǫǫ0 E, hence

∇ × H = Jc + Jd = (σ − iωǫǫ0 )E ≈ σE (119)

and therefore

−2i
∇2 H = ∇(∇ · H) − ∇ × (∇ × H) = 0 − σ∇ × E = −σ(iωµµ0 )H = H. (120)
δ2

In the coordinates (ξ, η, ζ) where ξ is ⊥ to the wall (and ξ = 0 at the inner surface) while η
and ζ = z are tangent to the wall, the magnetic field changes with the tangent coordinates
(η, ζ) on the scale O(λ) ≫ δ, so eq. (120) implies

H(ξ, η, ζ) ≈ H0 (η) exp(ikζ) exp((i − 1)ξ/δ). (121)

The space derivatives applied to this magnetic field act as

∂ (i − 1) ∂ ∂
= while = ik and = O(k), (122)
∂ξ δ ∂ζ ∂η

so the magnetic Gauss law

∂Hξ ∂Hη ∂Hζ

∇·H = + + = 0 (123)
∂ξ ∂η ∂ζ

requires Hξ to be much smaller that Hη or Hζ . Specifically,

H⊥
= O(kδ) ≪ 1. (124)
Hk

As written, this relation applies inside the metal wall of the waveguide, but it can be extended
inside the wave channel itself as a boundary condition at the metal’s edge. Indeed, the

26
magnetic field continues across the metal’s edge as

µmetal
Hk (channel) = Hk (metal) but H⊥ (channel) = H⊥ (metal), (125)
µchannel

hence at the boundary of the wave channel

H⊥ µmetal
= × O(kδ). (126)
Hk µchannel

The waveguides are usually made from good electric conductors such as copper, brass, alu-
minum, or silver, and all these metals have µ ≈ 1, hence

H⊥
at the boundary of the wave channel = O(kδ) ≪ 1, (127)
Hk

as promised in eq. (117). In principle, one can break this boundary condition by making the
waveguide wall out of steel or other high-µ alloy, but nobody in his/her right mind ever does
it, so let’s not consider this possibility any further. Instead, we take eq. (127) as generally
true, which justifies the idealized approximation H⊥ ≈ 0 (in comparison to the Hk ).

Next, consider the electric field inside the metal wall. Inverting eq. (119), we have

1
E ≈ ∇ × H, (128)
σ

hence for the magnetic field as in eq. (122),

σEξ = O(k)Hζ − ikHη ,

i−1
σEη = ikHξ − Hζ , (129)
δ
i−1
σEζ = Hη − O(k)Hξ .
δ

In magnitude, the tangent Eη , Eζ components of this electric field are

Hk Hk
   
σEk = O + O(k × H⊥ ) = O , (130)
δ δ

27
hence
Ek 1
∼ ∼ δkZ ≪ Z (131)
Hk δσ
where Z is the wave impedance of the waveguide, and the second relation stems from
  
1 1 ωµmetal µ0 σ 1 ωµmetal µ0 µmetal
δkZ = 2
= × = = ∼ 1. (132)
δσ δ 2 σkZ 2kZ 2µchannel

The relation (131) holds inside the metal wall, but since the tangent components Ek and Hk
of both electric and magnetic fields are continuous across the metal’s surface, we see that
the boundary values of the tangent fields inside the wave channels also obey
Ek
∼ δkZ ≪ Z, (133)
Hk

exactly as promised in eq. (118).

Finally, the normal component E⊥ of electric field is discontinuous across the metal’s
boundary, so it is not subject to any boundary conditions stemming from the skin effect.
However, everywhere inside the waveguide E ∼ ZH, so the dominant components of the
electric and the magnetic fields at the boundary should also have a similar relation E⊥ ∼
ZHk . Comparing this relation to eq. (133) for the Ek , we immediately see that

Ek δkZHk
∼ = δk ≪ 1, (134)
E⊥ ZHk

which justifies the Ek ≈ 0 approximation.

Attenuation

Another consequence of the imperfectly conducting waveguide walls is the wave energy
loss to the Ohmic resistance and hence slow attenuation of the wave power,

power ∝ e−αz (135)

for
(power loss)/length
α = . (136)
(net power)
The industry-standard copper waveguides for centimeter-range microwaves) have attenuation

28
rates ranging from 0.11 dB/m (1/40 m) for WR-90 at 10 GHz to 2.7 dB/m (1/60 cm) for
WR-10 at 90 GHz.

So let’s calculate the power loss due to Ohmic resistance in the walls and hence the
attenuation rate (136). In my notes on the skin effect, I have calculated the power loss in
terms of the tangent electric field Ek on the metal’s surface, but that is not very useful for
the current purposes since in the previous sections of the current notes we have solved the
wave equations using the Ek ≈ 0 approximation. Instead, let’s calculate the power loss in
terms of the tangent magnetic field Hk which does not even approximately vanish at the
metal’s surface.

Given the tangent magnetic field just outside the metal’s surface, — and hence also just
inside the surface, — we can find the conduction current inside the metal as

i−1 ˆ i − 1 (i−1)ξ/δ ˆ
J = ∇×H ≈ ξ × Hk = e ξ × Hsurface
k , (137)
δ δ

hence density of the dissipated power

power |J|2 1 −2ξ/δ surface

2
= = e H k . (138)
volume 2σ σδ 2

Integrating this power density over the metal’s depth ξ yields

large×δ Z∞
δ
Z
−2ξ/δ
e dξ ≈ e−2ξ/δ dξ = , (139)
2
0 0

hence
power loss 1 2
= Hsurface
k (140)
wall area 2σδ
and therefore
power loss 1
I
dη |Hη |2 + |Hζ |2 ,


= (141)
waveguide length 2δσ
C

where the integral is over the boundary C of the waveguide’s cross-section and η is the
coordinate along that boundary.

29
 In the Appendix to these notes, I calculate the integral (141) for all the TMm,n and TEm,n
waves in a rectangular waveguide. But for the moment, let’s not assume any particular cross-
sectional geometry or a particular TM or TE mode, so instead of calculating the integral (141)
let’s simply estimate its value up to unknown O(1) numeric constants which do not depend
on the wave’s amplitude or frequency. For a TM wave of amplitude E0 — that is,

Ez (x, y) ∼ E0 , Hz (x, y) ≡ 0, (142)

we generally have

∇t Ez (x, y) ∼ ΓE0 , (143)

iωǫǫ0
Ht (x, t) = − 2 ẑ × ∇t Ez (x, y)
Γ
ωǫǫ0
∼ × ΓE0 , (144)
Γ2

hence at the boundary

ωǫǫ0
Hζ = 0, Hη ∼ E0 , (145)
Γ

and therefore

I  ωǫǫ 2
0
dη |Hη |2 + |Hζ |2 × |E0 |2 × perimeter.


∼ (146)
Γ
C

For a waveguide of a particular geometry and characteristic width a — such as the larger
width of a rectangular waveguide, or a diameter of a circular waveguide, — the integral (146)
becomes
I  ωǫǫ 2
0
dη |Hη |2 + |Hζ |2 × |E0 |2


= A(mode) × a (147)
Γ
C

where A(mode) is an O(1) dimensionless number depending on a particular TM wave mode,

and also on the waveguide’s cross-sectional shape, but independent on the wave’s frequency

30
or amplitude. Consequently,

(power loss) Aa  ωǫǫ0 2

= × |E0 |2 . (148)
length 2σδ Γ

At the same time, the net power transmitted down the waveguide follows from the
integral
ZZ
|Ez |2 dx dy ∼ |E0 |2 × area, (149)
cross
section

so we may parametrize this integral as

ZZ
|Ez |2 dx dy = B(mode) × a2 |E0 |2 (150)
cross
section

where B(mode) is another O(1) dimensionless number depending on a specific TM mode and
a specific waveguide geometry. In terms of this number, the net power transmitted down
the waveguide is
kωǫǫ0
ZZ
Pnet = × |Ez |2 dx dy
2Γ2
cross
section (151)
kωǫǫ0
= × Ba2 × |E0 |2 .
2Γ2
Comparing this net power to the power loss (148) per unit length, we find the attenuation
rate
(power loss)/length
α =
(net power)
(Aa/2σδ) (ωǫǫ0 /Γ)2 × |E0 |2
= (152)
(Ba2 /2) (kωǫǫ0/Γ2 ) × |E0 |2
A/B ωǫǫ0
= × .
aσδ k

Nest consider a TE wave of amplitude H0 , that is

Ez (x, y) ≡ 0, Hz (x, y) ∼ H0 , (153)

31
hence

∇t Hz (x, y) ∼ ΓH0 , (154)

and therefore
ik k
Ht (x, y) = 2
∇t Hz (x, y) ∼ H0 . (155)
Γ Γ
In particular, at the boundary

k
Hζ = Hz ∼ H0 , Hη ∼ H0 , (156)
Γ

and therefore

k2
I I
dη |Hζ |2 ∼ (perimeter) × |H0 |2 , dη |Hη |2 ∼ (perimeter) × × |H0 |2 , (157)
Γ2
C C

which we may parametrize as

I
dη |Hζ |2 = C(mode) × a × |H0 |2 ,
C
(158)
k2
I
2
dη |Hη | = D(mode) × a × 2 × |H0 |2 ,
Γ
C

for some mode-dependent — but frequency-independent and amplitude-independent — O(1)

numbers C(mode) and D(mode). In terms of these parameters, the EM power loss per unit
length becomes

(power loss) 1
I
dη |Hη |2 + |Hζ |2


=
length 2σδ
C (159)
k2
 
a
= C(mode) + 2 × D(mode) × |H0 |2 .
2σδ Γ

At the same time,

ZZ
|Hz |2 dx dy = E(mode) × a2 × |H0 |2 (160)
cross
section

for a yet another mode-dependent dimensionless O(1) number E, hence the net power of the

32
wave is related to its amplitude as

kωµµ0
Pnet = × E(mode) × a2 × |H0 |2 . (161)
2Γ2

Consequently, comparing the power loss rate (159) to this net power, we get the attenuation
rate
(power loss)/length
α =
(net power)
(a/2σδ) × C + D(k/Γ)2 × |H0 |2


= (162)
(Ea2 /2) (kωµµ0/Γ2 ) × |H0 |2
1 CΓ2 + Dk 2
= × .
aσδ Ekωµµ0

Now let’s bring eq. (162) and (152) for the attenuation rate to a common form

F (mode) ω 2 + G(mode)Ω2
α = × √ (163)
aZpw σδ ω ω 2 − Ω2

where Ω = ωmin = (c/n)Γ is the cutoff frequency for the mode in question, while
r
µµ0
Zpw = (164)
ǫǫ0

is the wave impedance of a plane wave in the material filling the waveguide’s channel; for
the evacuated or air-filled waveguides, Zpw ≈ 377 Ω. To bring eq. (162) for the TE waves to
the form (163), we use
r
ωn 2 np 2
k = − Γ2 = ω − Ω2 , (165)
c c

hence

n2 Dn2
   
2 2 C
= 2 CΩ2 + D(ω 2 − Ω2 ) = 2
− 1 Ω2


CΓ + Dk ω + (166)
c c2 D

while
Enµµ0 p
Ekωµµ0 = × ω ω 2 − Ω2 , (167)
c

33
and therefore
C
D/E (n/c)2 ω2 + ( D − 1)
α = × × √ . (168)
aσδ (n/c)µµ0 ω ω − Ω2
2

The second factor here amounts to

√
(n/c)2 n = ǫµ 1
= = , (169)
(n/c)µµ0 cµµ0 Zpw

so eq. (168) for the attenuation rate indeed takes form

F (mode) ω 2 + G(mode)Ω2
α = × √ (163)
aZpw σδ ω ω 2 − Ω2

where we identify

D(mode) C(mode)
F (TE mode) = , G(TE mode) = − 1. (170)
E(mode) D(mode)

Similarly, for the TM waves we have

A/B ωǫǫ0
α = × (152)
aσδ k

where
ωǫǫ0 ǫǫ0 ω 1 ω2 + 0
= ×√ = × √ , (171)
k n/c ω 2 − Ω2 Zpw ω ω 2 − Ω2
hence
A/B ω2 + 0
α = × √ , (172)
aZpw σδ ω ω 2 − Ω2
which also has form
F (mode) ω 2 + G(mode)Ω2
α = × √ (163)
aZpw σδ ω ω 2 − Ω2
where we identify

A(mode)
F (TM mode) = , G(TM mode) = 0. (173)
B(mode)

Now consider the frequency dependence of the attenuation rate. Beside the explicitly
ω–dependent second factor in eq. (163), the skin depth δ (in the denominator of the first

34
factor) depends on frequency as δ ∝ ω −1/2 , thus

r
Ω
σ(ω) = σ(Ω) × (174)
ω

and hence

F (mode) ω 2 + G(mode)Ω2
α(ω) = × p . (175)
aZpw σδ(Ω) Ωω(ω 2 − Ω2 )

In this form, the first factor depends on the waveguide’s design and on the specific mode but
does not depend on the wave’s frequency, while the second factor depends on the frequency
or rather on the ratio ω/Ω of the wave’s frequency to the cutoff frequency for the mode in
question. Graphically, α as a function of ω/Ω behaves as

4 TM mode, G = 0
TE mode, G = 0.5
3
TE mode, G = 1.5
2

0
0 1 2 3 4 5 (176)
Note that for all the modes, we have strong attenuation for frequencies just above the
cutoff frequency Ω. Physically, this is caused by the slow velocity of the energy flow
p
venergy = (c/n) 1 − (Ω/ω)2 , so the energy does not move very far while it’s dissipated
by the conduction currents. At higher frequencies, the attenuation becomes weaker, reaches
a minimum, and then starts slowly growing with the frequency due to shrinking skin depth.
The optimal frequency which minimizes the attenuation rate depends on the G parameter

35
of the mode in question, specifically

ω
optimal 2

= larger root of x2 − 3(G + 1)x + G = 0. (177)
Ω
√
For all the TM modes — which all have G = 0 — the optimal frequency is ωopt = 3 × Ω,
while the TE modes have higher ωopt /Ω ratios; for example, the dominant TE1,0 mode of a
rectangular waveguide with a : b = 2 : 1 has ωopt ≈ 2.1 × Ω.

As to the first factor in the equation

F (mode) ω 2 + G(mode)Ω2
α(ω) = × p , (175)
aZpw σδ(Ω) Ωω(ω 2 − Ω2 )

it does not depend on the frequency of the actual wave, but it does depend on the frequency
band for which the waveguide is designed. If one wants only one mode — such as TE1,0 —
to propagate down the waveguide, then one uses a waveguide with

ωdesign
1 < < 2 (178)
Ω(TE1,0 )

and hence
π(c/n) (1 to 2)π(c/n)
a ∼ ∼ . (179)
Ω(TE1,0 ) ωdesign

1/2
Thus (1/a) ∝ ωdesign , and also (1/δ(Ω)) ∝ Ω1/2 ∝ ωdesign , and consequently

 3/2
α ∝ ωdesign . (180)

And that’s why the WR-10 waveguide used for the 90 GHz microwaves has 27 times larger
attenuation rate than the WR-90 waveguide used for the 10 GHz microwaves, 2.7 dB/m
versus 0.1 dB/m.

PS: Eq. (175) gives the attenuation rate due to electric resistivity of the waveguide walls,
but there could be additional attenuation due to other causes. For example, for a waveguide

36
filled with a dielectric, the dielectric constant ǫ(ω) may develop an imaginary part at high fre-
quencies, which would make the dielectric absorb some of the microwave power and dissipate
it as heat. Similarly, a poor dielectric which has a small but non-zero conductivity would
absorb some of the microwave power and dissipate it as heat. The air-filled waveguides do
not have these kinds of attenuation, but they are vulnerable to corrosion — especially when
the air inside them gets dirty and humid — which would drastically decrease the surface
resistivity of the metal walls and hence drastically increase the attenuation rate.

Finally, a real-life waveguide may have extra attenuation of the wave power due to
scattering by any extraneous objects of size >
∼ 0.1λ inside the waveguide, or any wall defor-
> 0.1λ. Also, a waveguide which bends too sharply may case wave scattering
mations of size ∼
and hence extra attenuation.

Appendix: F and G Parameters for a Rectangular Waveguide

In this Appendix, I calculate the A, B, C, D, E parameters — and hence F and G — for

all the TE and TM modes in a rectangular waveguide. Let’s start with a TMm,n mode for
some integer m and n:

mπx nπy
Ez = E0 sin sin ,
a b
inπωǫǫ0 mπx nπy
Hx = 2
E0 sin cos ,
bΓ a b
imπωǫǫ0 mπx nπy (181)
Hy = − 2
E0 cos sin ,
aΓ a b
π 2 m2 π 2 n2
Γ2 = + .
a2 b2

For this mode

Za Zb
2 2 def ab
Bm,n × a |E0 | = dx dy |Ez (x, y)|2 = |E0 |2 , (182)
4
0 0

hence
ab/4 b
Bm,n = = ∀m, n. (183)
a2 4a

37
At the same time, along the cross-section’s perimeter we have
(
2
 πωǫǫ 
0 2 (n/b)2 sin2 (mπx/a) at the long sides,
|Hk | = |E0 | × (184)
Γ2 (m/a)2 sin2 (nπy/b) at the short sides,

hence
 πωǫǫ   
a b
I
2 0 2 2 2
|Hk | dℓ = |E0 | × 2 × × (n/b) + 2 × × (m/a)
Γ2 2 2
perimeter
an2 bm2
 πωǫǫ   
0 2
= |E0 | × + 2 .
Γ2 b2 a

Interpreting this integral as

 ωǫǫ 2
0
Am,n × a |E0 |2 , (185)
Γ
we find
π2 an2 bm2 (b/a)m2 + (a/b)2 n2
 
Am,n = + 2 = . (186)
aΓ2 b2 a m2 + (a/b)2 n2

Altogether, for the TM waves we have

Am,n m2 + (a/b)3 n2
F (TMm,n ) = = 4 2 , (187)
Bm,n m + (a/b)2 n2

which varies in the range

4 < F (TM) < 4(a/b). (188)

Now consider a TEm,n mode

mπx nπy
Hz = H0 cos cos ,
a b
imπk mπx nπy
Hx = − H 0 sin cos ,
aΓ2 a b
inπk mπx nπy (189)
Hy = − 2 H0 cos sin ,
bΓ a b
π 2 m2 π 2 n2
Γ2 = + .
a2 b2

38
Similarly to what we had earlier for the TM modes, for this TE mode we have

Za Zb
2 def ab
Em,n × a2 |H0 | = dx dy |Hz (x, y)|2 = |H0 |2 , (190)
4
0 0

provided both m > 0 and n2 so that both cos2 (mπx/a) and cos2 (nπy/b) average to 12 ; other
wise, for m = 0 or n = 0 we have one of the cos2 factors being 1 for all x or all y, hence

Za Zb
2 def ab
Em,n × a2 |H0 | = dx dy |Hz (x, y)|2 = |H0 |2 . (191)
2
0 0

Thus,
1
(
ab/(4 or 2) b 4 for m > 0 and n > 0,
Em,n = 2
= × 1
(192)
a a for m = 0 or n = 0.
2

As to the magnetic field at the cross-section’s perimeter, we now have

along a long side

mπx m2 π 2 k 2 mπx
|Hk |2 = |H0 |2 × cos2 + 2 4
|H0 |2 × sin2 ,
a a Γ a
(193)
along a short side
nπy n2 π 2 k 2 nπy
|Hk |2 = |H0 |2 × cos2 + 2 4
|H0 |2 × sin2 ,
b b Γ b
hence for m > 0 and n > 0

m2 π 2 k 2 n2 π 2 k 2
   
a b
I
2 2 2 2 2
|Hk | dη = 2 × |H0 | + |H0 | + 2× |H0 | + 2 4 |H0 |
2 a2 Γ4 2 b Γ
perimeter
m2 n2 π2 k2
 
2
= (a + b) × |H0 | + + × 4
|H0 |2 ,
a b Γ
(194)
but for m = 0

n2 π 2 k 2
 
b
I
2 2 2 2
|Hk| dη = 2 × a × |h0 | + 2 × |H0 | + 2 4 |H0 |
2 b Γ
perimeter (195)
n2 π2k2
= (2a + b) × |H0 |2 + × |H0 |2 ,
b Γ4

39
and likewise for n = 0
m2 π 2 k 2
I
|Hk |2 dη = (a + 2b) × |H0 |2 + × 4 |H0 |2 . (196)
a Γ
perimeter

Interpreting these integrals as

ak 2
Cm,n × a × |H0 |2 + Dm,n × 2
× |H0 |2 , (197)
Γ
we find
 a+b
 a
 for m > 0 and n > 0,
Cm,n = 2a+b (198)
a for m = 0 but n > 0,

 a+2b
a for n = 0 but m > 0,
while
π2 m2 n2 m2 + (a/b)n2
 
Dm,n = + = . (199)
aΓ2 a b m2 + (a/b)2 n2
Consequently, the TE waves have

m2 + (a/b)n2 4 for m > 0 and n > 0,


Dm,n a
F (TEm,n ) = = × ×
Em,n m2 + (a/b)2 n2 b 2 for m = 0 or n = 0,

(a/b)m2 + (a/b)2 n2 (200)


 4 m2 + (a/b)2 n2 for m > 0 and n > 0,
= 2 for m = 0 but n > 0,


2(a/b) for n = 0 but m > 0,


which varies in the range

2 ≤ F (TE) < 4(a/b), (201)

while

(b/a)m2 + (a/b)2 n2

 m2 + (a/b)n2
 for m > 0 and n > 0,
Cm,n
G(TEm,n ) = − 1 = 2(a/b) for m = 0 but n > 0, (202)
Dm,n 

2(b/a) for n = 0 but m > 0,


which varies in the range

b 2a
< G(TE) ≤ . (203)
a b
In particular, the dominant TE1,0 wave — which is the mode with the lowest cutoff frequency

40
— has
2a 2b
F (TE1,0 ) = , G(TE1,0 ) = . (204)
b a

Resonator Cavities
• Wikipedia article about microwave cavities.

In principle, a microwave cavity can have any kind of geometry. In practice, most
cavities are cylinders of radius R and length d of comparable magnitudes; less common are
rectangular q × b × d cavities. Either way, the cavity can be thought as a finite length d
of a waveguide in which some kind of a TM or TE wave travels along the z axis and gets
reflected back and forth of the two conducting end-caps at z = 0 and z = d, thus

E, B ∝ Aeikz−iωt + Be−ikz−iωt . (205)

Approximating the end-caps — as well as the its walls — as perfectly conducting, we also
make them perfectly reflecting, thus |B| = |A|, while the relative phase of the forward and
backward amplitudes A and B is different for the TM and the TE waves.

Indeed, consider a TM wave

Hz (x, y, z, t) ≡ 0, (206)
 
Ez (x, y, z, t) = ψ(x, y) Ae+ikz−iωt + Be−ikz−iωt , (207)
ik 
+ikz−iωt −ikz−iωt

Et (x, y, z, t) = 2 ∇t ψ(x, y) Ae − Be , (208)
γ
iωµµ0
 +ikz−iωt −ikz−iωt

Ht (x, y, z, t) = ẑ × ∇t ψ(x, y) Ae + Be . (209)
Γ2

Note the relative signs (marked in red) between the forward and the backward waves here.
In particular, note the opposite sign for the Et components, which stems from the opposite
∂
signs of ∂z → ±ik.

41
 At the perfectly conducting endcap at z = 0, we have boundary conditions

Et = 0, Hz = 0. (210)

The Hz = 0 condition here is automatic for the TM waves, while the Et = 0 condition calls
for A − B = 0. Likewise, similar boundary conditions at the other endcap at z = d call for
Ae+ikd = Be−ikd . Altogether, this requires
pπ
k = for a non-negative integer p = 0, 1, 2, 3, . . . (211)
d
while
pπx
Ez = A cos × ψ(x, y)e−iωt . (212)
d

On the other hand, a TE wave has

Ez (x, y, z, t) ≡ 0, (213)
 
Hz (x, y, z, t) = ψ(x, y) Ae+ikz−iωt + Be−ikz−iωt , (214)
ik  
Ht (x, y, z, t) = 2 ∇t ψ(x, y) Ae+ikz−iωt − Be−ikz−iωt , (215)
γ
−iωǫǫ0
 +ikz−iωt −ikz−iωt

Et (x, y, z, t) = ẑ × ∇ t ψ(x, y) Ae + Be . (216)
Γ2
For this wave, both Hz = 0 and Et = 0 boundary conditions at each endcap are non-trivial;
however, both conditions lead to the same relations between the forward and backward
amplitudes. Specifically

A + B = 0 to keep Hz = 0 and Et = 0 at z = 0,
(217)
Ae+ikd + Be−ikd = 0 to keep Hz = 0 and Et = 0 at z = d,

which together require

pπ
k = for a positive integer p = 1, 2, 3, 4, . . .
d
while
pπx
Hz = iA sin × ψ(x, y)e−iωt . (218)
d

Either way, we have a discrete set of wave numbers k and hence discrete set of resonance

42
frequencies:
ǫµω 2 2 2 2
 πp 2
= Γ + k = Γ + (219)
c2 d

and hence
r  πp 2
c
ωm,n,p = √ Γ2m,n + . (220)
ǫµ d

For example, a rectangular a × b × d cavity resonates at frequencies

r
c m2 π 2 n2 π 2 p2 π 2
ωm,n,p = √ + + (221)
ǫµ a2 b2 d2

for integer m, n, p. Moreover, at least two of this integers must be positive while the third
may vanish. Specifically, the TM waves must have positive m and n while p may vanish,
while the TE waves may have either m = 0 or n = 0 (but not both) while p must be positive.

The lowest frequency resonance — called the fundamental mode — of the rectangular
cavity with d > a > b is the TE1,0,1 wave with m = 1, n = 0, and p = 1. For this mode,

π 2 c2
 
2 1 1
ω = + 2 , (222)
ǫµ a2 d
πx πz −iωt
Hz = H0 cos sin e , (223)
a d
a πx πz −iωt
Hx = − H0 sin cos e , (224)
rd a d
a2 πx πz
Ey = i 1 + 2 Zpw H0 sin sin . (225)
d a d
Hy = Ex = Ez = 0. (226)

For a cylindrical cavity, the TM and the TE waves have

′
jm,n
jm,n
Γ(TMm,n ) = , Γ(TEm,n ) = , (227)
R R

where jm,n is the nth positive root of the mth Bessel function Jm (x) and jm,n
′ is the nth
positive root of its derivative dJm (x)/dx. Consequently, the resonant frequencies for these

43
waves are

c2 (jm,n )2 p2 π 2
 
2
ωm,n,p (TM) = +
ǫµ R2 d2
for m = 0, 1, 2, . . . , n = 1, 2, 3, . . . , p = 0, 1, 2, . . . , (228)
!
′ )2
2 c2 (jm,n p2 π 2
ωm,n,p (TE) = +
ǫµ R2 d2
for m = 0, 1, 2, . . . , n = 1, 2, 3, . . . , p = 1, 2, 3 . . . . (229)

For a stubby cavity with d < 2.03R, the fundamental mode is TM0,1,0 with

c j1,0 ≈ 2.40
ω = √ × ,
ǫµ R
Ez (ρ, φ, z) = E0 J0 (j0,1 ρ/R), (230)
iE0 ′
Hφ (ρ, φ, z) = J (j0,1 ρ/R), (231)
Zpw 0
Eρ = Eφ = 0, (232)
Hz = Hρ = 0. (233)

But for a longer cavity with d > 2.03R, the fundamental mode becomes TE1,1,1 with

s
′ ≈ 1.84)2
(j1,1
c π2
ω = √ + , (234)
ǫµ R2 d2

′ ρ
!
j1,1 πz
H z = H 0 × J1 × cos(φ) × sin , (235)
R d
!
′ ρ
j1,1
πR πz
Hρ = ′ H0 × J1′ × cos(φ) × cos , (236)
j1,1 d R d
!
′ ρ
j1,1
πR R πz
Hφ = − ′ H0 × ′ J1 × sin(φ) × cos , (237)
j1,1 d j1,1 ρ R d

44
 Ez = 0, (238)
v
!2 !
′ ρ
u
u πR R j1,1 πz
Eρ = −it1 + ′ d Zpw H0 × ′ ρ J1 × sin(φ) × sin , (239)
j1,1 j1,1 R d
v
!2 !
′ ρ
u
u πR j1,1 πz
Eφ = it1 + ′ d Zpw H0 × J1′ × cos(φ) × sin . (240)
j1,1 R d

Quality of a Resonator Cavity

The quality of any kind of a resonator — be it a mechanical pendulum, an LC circuit,

or a microwave cavity, — is defined as

stored energy
Q = ω0 × (241)
power loss

where ω0 is the central frequency of the resonance. In class, I am going to explain this
subject off-the-notes.

In these notes, I am going to focus on the microwave cavities and estimate their qualities
as resonators. In general, a microwave cavity looses power to 3 mechanisms: (1) Ohmic
losses in the cavity walls, (2) losses in the dielectric filling the cavity, and (3) losses through
the hole in the wall or the antenna connecting the cavity to the outside world, thus

net
Ploss = P1 + P2 + P3 (242)

and hence
1 1 1 1
= + + . (243)
Qnet Q1 Q2 Q3
In these notes, I am going to focus on the first mechanism and estimate Q1 , but in real life
the other 2 mechanisms might reduce the net quality of the microwave cavity.

Similarly to the waveguides, the power loss due to Ohmic resistance in the walls is

Rs
ZZ
P = |H|2 d2 area (244)
2

where the area integral is over the entire surface of the cavity — including both the side

45
walls and the end caps, — while

r
1 ωµmetal µ0
Rs = = (245)
σδ 2σ

is the surface resistivity of the metal; for a copper wall and ω = 2π × 10 GHz, Rs ≈ 26 mΩ.
At the same time, the energy stored in the cavity is

ǫǫ0 µµ0 µµ0

ZZZ   ZZZ
2
U = |E| + |H| d3 x =
2
|H|2 d3 x, (246)
4 4 2

hence cavity quality

|H|2 d3 volume
RRR
ω0 U ωµµ0
Q = = × RR . (247)
P Rs |H|2 d2 area

Sometimes, this formula is written down as

G
Q = (248)
Rs

where
|H|2 d3 volume
RRR
G = ω0 µµ0 × RR (249)
|H|2 d2 area

is often called the geometry factor because it depends on the cavity’s geometry but not on
the conductivity of it’s walls. But just as often, the name geometry factor refers to the
dimensionless geometry factor

√
|H|2 d3 volume
RRR
G ω ǫµ
Ĝ = = × RR ; (250)
Zpw c |H|2 d2 area

it depends on a particular resonating mode of the cavity as well as on the ratios of its
dimensions — for example on the d/R ratio of a cylindrical cavity, — but does not depends
on its overall size. In terms of this dimensionless geometry factor, the quality factor of the

46
microwave cavity is
Zpw
Q = × Q̂. (251)
Rs

As an order-of-magnitude estimate,

|H|2 d3 volume cavity′ s volume

RRR
∼ ∼ (cavity′ s size), (252)
cavity′ s area
RR
|H|2 d2 area

while
√
ω ǫµ 1
∼ . (253)
c cavity′ s size

hence

Ĝ ∼ 1. (254)

Thus, a typical microwave cavity has

Zpw
Q ∼ ∼ 104 . (255)
Rs

To conclude this section, let me actually calculate the geometry factor Ĝ for the funda-
mental TM0,1,0 mode of a cylindrical cavity with d < 2.03R. In this mode, the magnetic
field points in the φ direction throughout the cavity while its magnitude depends only on
the radial coordinate ρ but on on φ or z,

j0,1 ≈ 2.40
Hφ = H0 J0′ (Γρ) for Γ = . (256)
R

Consequently,
ZZ
|H|2 d2 area = 2πRd × |H0 |2 × C (257)
sidewall

where

C = |J0′ (ΓR)|2 = |J0′ (j0,1 )|2 ≈ 0.270, (258)

47
while

ZR Zj0,1
1
ZZ
|H|2 d2 area = 2π|H0|2 × |J0′ (Γρ)|2 ρ dρ = 2πR2 × |H0 |2 × (J0′ (x))2 x dx
(j0,1 )2
endcap 0 0
(259)
where
Zj0,1
1 C
2
(J0′ (x))2 x dx = (260)
(j0,1 ) 2
0

for the same constant C as in eq. (258), — which really surprised me when I have calculated
this integral using Mathematica, — hence altogether

C
ZZ
|H|2 d2 area = 2πRd|H0|2 × C + 2 × 2πR2 |H0 |2 × = 2πC R(d + R) × |H0 |2 . (261)
2
whole
surface

At the same time,

ZR
CR2
ZZZ
|H|2 d3 volume = 2πd|H0 |2 × |J0′ (Γρ)|2 ρ dρ = 2πd|H0|2 × , (262)
2
0

so the ratio of the volume to surface integrals amounts to

πCdR2
 ZZ
Rd
ZZZ
2 3
|H| d volume |H|2 d2 area = = . (263)
2πCR(R + d) 2(R + d)

Meanwhile the resonating frequency ω0 of the TM0,1,0 mode is

 
c j0,1
ω0 = √ × Γ= , (264)
ǫµ R

hence
j0,1 Rd j0,1 d 1.20 d
Ĝ = × = × ≈ . (265)
R 2(R + d) 2 R+d R+d

For example, in a cavity with d = R the fundamental mode has Ĝ = 0.40. For a more specific
example, consider an air-filled cavity with copper walls and R = 1.15 cm which makes for

48
ω0 = 2π × 10.0 GHz. At this frequency, copper has rs ≈ 25.8 mΩ, hence quality factor

377 Ω
Q ≈ × 0.40 = 5850. (266)
26 mΩ

For a longer cylindrical cavity with d > 2.03R, the fundamental mode switches from the
TM0,1,0 to the TE1,1,1 , and the calculation becomes more complicated. So instead of doing
it here, let me put it on your next homework set#10 as problem#4.

49