Basic Physics Behind

Operation of TPC

Part I
-- Fundamental Processes in the TPC --

Keisuke Fujii
ILC-TPC School
Beijing: 7-11 Jan., 2008, Revised on Feb.2, 2008

1
 Purpose
Being a non-expert, I will focus on very basic principles,
trying to introduce you, students excluding experts, to
more advanced topics containing more practical and
technical aspects to be covered by real experts in this
school.
Emphasis will be put on concepts and philosophy, and
hence practical examples will be minimum, for them
take a look at excellent text books such as

V.Palldino & B.Sadoulet 1974: LBL-3013

F.Sauli 1977: CERN 77-09
W.Blum & L.Rolandi 1993

2
 Fundamental Processes
Beam Ionizations
Liberation of Electrons
PI (N ; N̄ ) Normal incidence
(no angle effect)
No δ-ray

Drift Volume Drift and Diffusion

x2i
! "
1
PD (xi ; σd ) = √ exp − 2
2πσd 2σd
√
σd = Cd z

Amplification and
Amplification Gap further Diffusion
"θ
(θ + 1)θ+1
! ! ! ""
G G
PG (G/Ḡ; θ) = exp −(θ + 1)
Readout Pads Γ(θ + 1) Ḡ Ḡ

Pad Response

Coordinate
x

3
 Subjects to be Covered
Liberation of electrons by ionization (dE/dx)

Classical theory of electron transportation in a

chamber gas and its applications

Behaviors of electrons in E and B fields

Transport coefficients: Vd, Cd

Gas amplification

Creation of signals

Coordinate measurement

4
 Subjects Left Out
Electron attachment

CF4, O2 contamination, etc.

Transportation of ions

+ve ion feed back and gating

Ion exchange and aging

These are very important in practice, but simply beyond the

scope of my lectures.

5
Liberation of Electrons
Through Ionization of Gas Molecules
By a Fast Moving Charged Particle

This part will be brief, though it is a deep subject.

6
 Ionizing Collisions
Various Ionization Mechanisms
Primary Direct Ionization The probability distribution for the free
A charged particle going through a chamber flight path ( l ) is then given by
gas ionizes gas molecules along its path and dl
f (l; λ)dl = P (0; l/λ) P (1; dl/λ) = e−l/λ
leaves a track of ionization. This is called λ
primary ionization. The ionizing collisions The average number of ionizing collisions
are statistically independent, and hence the per 1cm is about 28 for a minimum ionizing
number of such collisions obeys the Poisson particle passing through an Ar gas at 1atm.
distribution: n This corresponds to an ionization cross
n̄ −n̄
P (n; n̄) = e section of
n!
where the average number of collisions is σI (Ar) ! 10−18 [cm2 ] ∼ πRAr
2 2
× ZAr αQED
given by the thickness ( L ) of the medium
and the mean free path ( λ ) as Secondary Ionization
n̄ = L/λ The electrons kicked out from molecules, if
The mean free path is of course related to energetic enough, will further ionize gas
the cross section per electron in the gas molecules. Some of the gas molecules might
and the electron density in the gas. be excited to some intermediate state that
N σI λ = 1 can lead to further ionizations through a

7
process like of a length “L” is given by
A∗ B → AB + e− ! "
dE L
!nI " =
where A* is an excited molecule and B is a dx W
molecule with an ionization potential that is
The “W” depends on the gas and the nature
lower than the excitation energy of A*.
of the incident particle, but it is known to
A* is often a metastable excited state of a
be independent of incident energy if E > a
noble gas used as the main gas component
few keV for electrons or if E > a few MeV
(e.g. Ar) and B is often a quencher added to
for alpha-particles.
stabilize the gas amplification process.
A* can also be an optical excitation with a
For a noble gas, “W” ranges from 46 eV for
long life time.
He to 22 eV for Xe. For Ar it is 26 eV.
A∗ = metastable → Penning Effect The “W” values are typically a factor of 1.5
A∗ = optical excitation → Jesse Effect to 2 larger than the ionization potentials.

Average Energy for Ionization In order to see how the average energy
Only a certain fraction of the energy loss loss depends on the particle’s speed or on
by the fast charged particle is used for the nature of the gas, and to understand
ionization. We define “W” as the average the distribution around the average, let us
energy required for the creation of a single review next the Allison-Cobb formulation
ionization electron. Then the average of the energy loss process.
number of ionization electrons along a track

8
 Allison-Cobb Formulation
dE/dx as a photo-absorption/ionization by a virtual photon
A Charged Particle in a Dielectric !dE/dx" = −eE(cβt, t) · β/β
The energy loss of a charged particle that
passes through a dielectric medium is due The Maxwell equations
to the negative work done by the E-field The Maxwell equations in a dielectric with a
created by the electrons and the nuclei of magnetic permeability of one (B = H ) read
the molecules making up the medium: ∇·B = 0
! #
" 1 ∂
dE/dx · dx = −e Ea · dx ∇×E = − B
c ∂t
a
∇·D = 4πρ
where e is the particle charge and the sum
1 ∂ 4π
is over all the charges in the medium. ∇×B =
c ∂t
D+
c
j
Averaging the both sides over a small tube The charge and current densities are
around the incident particle, we have
ρ(x, t) = e δ 3 (x − cβt)
!dE/dx" · dx = −eE · dx j(x, t) = c β ρ(x, t)
! #
with "
E = Ea where c β is the velocity of the charged
a particle that can be regarded as constant
If we can determine the E-field by solving during its passage through the dielectric
the Maxwell equations, we can get dE/dx medium.

9
In order to close the Maxwell equations we In the limit of ω → ∞ , we have
need a material equation for the dielectric iG̃(0) G̃! (0) G̃! (0)
!(k, ω) − 1 " − + ··· = − 2 + ···
medium: ! !
ω ω2 ω
D(x, t) = E(x, t) +
∞
dτ d3 ξ G (|ξ| , τ ) E(x − ξ, t − τ )
with
! !
|ξ |<cβτ
d3 ξ G (|ξ| , τ ) eik·ξ !
0
d3 ξ G (|ξ| , τ ) = G̃(τ )
|ξ |<cβτ |ξ |<cβτ
which expresses the dependence of the
since only the region near τ ! 0 contributes.
electric flux density on the electric fields
at causally connected space-time points The Solution
through a Green function G. With the scalar and vector potentials
Defining Fourier transform of f as B̃(k, ω) = ik × Ã(k, ω)
! 3
d x dt
f˜(k, ω) ≡ f (x, t) e−i(k·x−ωt) Ẽ(k, ω) =
iω
Ã(k, ω) − ikφ̃(k, ω)
(2π) 2
c
we have in the Coulomb gauge
D̃(k, ω) = "(k, ω) Ẽ(k, ω)
with B̃(k, ik · Ã(k, ω) = 0
! ∞ !
!(k, ω) = 1 + dτ d3 ξ G (|ξ| , τ ) ei(k·ξ −ωτ ) we can translate the Maxwell eqs. into
0 |ξ |<cβτ
k2 ! φ̃(k, ω)
= 2e δ(ω − cβ · k)
The dielectric constant being independent ! ω2 ! ωk
k2 Ã(k, ω) = φ̃(k, ω) + 2eβ δ(ω − cβ · k)
of the direction of the wave vector shows c2
Ã(k, ω) −
c
the isotropy of the dielectric medium. The solution is hence given by
2e
Notice that the E- and D-fields being real φ̃(k, ω) = 2 δ(ω − cβ · k)
#k
implies ωk/ck2 − β
∗
! (k, ω ) = !(k, −ω)
∗
Ã(k, ω) = 2e 2 2 δ(ω − cβ · k)
#ω /c − k2

10
The Energy Loss Formula The upper limit of the omega integral is of
course finite in reality as constrained by
Putting them together into the energy loss
kinematics:
formula after inverse Fourier transform, 2me c2 β 2 γ 2
! ωmax =
we obtain 1 + 2γ (me /m) + (me /m)2
!dE/dx" = −e (β/β) · E(cβt, t)
where m is the particle mass and me the
!
= −e
d3 k dω
(β/β) · Ẽ(k, ω) ei(k·cβ t−ωt) electron mass.
(2π)2

Allison-Cobb (PAI) Model

! ! ∞ " #
2e2 +∞
ω 2 /k 2 c2 − β 2 1
= − dω dk i (ωk) − 2
(2π)β 2 c2 −∞ |ω|/cβ
2 2
$ω /c − k 2 k $
All we need is the complex dielectric const.
or with !∗ (k, ω ∗ ) = !(k, −ω)
! ! ∞ ! := !1 + i!2
2e2 ∞
!dE/dx" = − 2
πβ 0
"
dω dk
# ω/cβ 2 $ # $ # $%
Allison and Cobb relate its imaginary part
1 1
to the photo-absorption cross section of
ω ω
(ωk) β − 2 2 Im
2
2 2 2
+ 2 Im
k c $ω − k c kc $
the medium with
The 1st term in the square bracket is from
N !c"
the vector potential (transverse) and !2 (ω) !
Z ω
σγ (ω)
vanishes at beta=0 while the 2nd term is
where N is the electron density of the
from the scalar potential (longitudinal) and
medium and Z the atomic number of the
stays finite at beta=0. The energy loss is
molecule. They then get the real part of
calculable when the complex dielectric
the dielectric constant with the Kramers-
constant is given. The formula suggests if
Kronig relation
the dielectric constant is real, there will be !
2 ∞
ω " !2 (ω " )
no energy loss. !1 (ω) − 1 = P dω "
π 0 ω "2 − ω 2

11
Indeed, we have The photo-absorption cross section for Ar
eikx = ei(Re k+i Im k)x is shown below.
= ei(Re k)x e−(Im k)x
1 102
= ei(Re k)x e− 2 (x/λ)

SG (hW) [Mb]
Ar
which together with 10
√
k= ! ω/c 1
implies the attenuation length “lambda” to
be given by 10-1
√
1/λ ≡ 2 Im k = 2 (ω/c) Im #
10-2
= 2 (ω/c) Im (#1 + i#2 )1/2
# (ω/c) #2
10-3
for a low density medium such as our TPC
gas mixtures. 10-4
G.V.Marr & J.B.West (1976)
On the other hand, the attenuation length
10-5
is related to the absorption cross section 102 103 104
h W [eV]
for photons:
1 The cross section is roughly consistent with
σγ (ω) =
(N/Z)λ what you expect from the geometrical
leading us to cross section of Ar:
N !c"
!2 (ω) ! σγ (ω) π(0.18 [nm])2 × 8 × αQED " 58 [Mb]
Z ω
# electrons in the outermost shell
12
The cross section can be translated into we obtain the real part as
the imaginary part of the dielectric const.
as shown in the next figure. 1 0.001

1 - 1 [s 103]
Ar
0.8 0.008
-2
10
2

M-shell Ar 0.6 0.006

-3
10
0.4 0.004
10-4
L-shell
-5 0.2 0.002
10

10-6 0 0

10-7 -0.2 -0.002

K-shell
10-8 -0.4 -0.004

10-9 102 103 104

h W [eV]

10-10
102 103 104 where the region near the L-shell peak is
h W [eV]
zoomed up to show the resonance effect.

Using the Kramers-Kronig relation: These are, however, for real photons. The
! ∞
2 " ω !2 (ω )
" "
crucial step taken by Allison and Cobb was
!1 (ω) − 1 = P dω
π 0 ω "2 − ω 2 to extend this to virtual photons.

13
 Kramers-Kronig Relation
Relation between Real and Imaginary Parts of Epsilon
Analyticity of Epsilon This shows that the integral of the epsilon
The complex dielectric constant is analytic over the upper semicircle vanishes. We can
in the upper half omega plane as is seen hence express the epsilon using Cauchy’s
from its definition integral:
! ! !
∞ 1 +∞
!(ω # ) − 1
!(k, ω) = 1 + dτ d ξ G (|ξ| , τ ) e
3 i(k·ξ −ωτ ) !(ω) − 1 = dω #
2πi −∞ ω # − ω − i(+0)
0 ξ
| |<cβτ
! +∞ " #
1 1
= dω P #
#
+ iπ δ(ω − ω) (!(ω # ) − 1)
#
It is real on the imaginary axis because of 2πi −∞ ω −ω
! +∞
!∗ (k, ω ∗ ) = !(k, −ω) =
1
P dω #
!(ω # ) − 1 1
+ (!(ω) − 1)
2πi −∞ ω# − ω 2
which is also easily derived from the above
Moving the last term of the R.H.S. to the
definition of the epsilon.
L.H.S. we get
Recall also its asymptotic behavior !
1 +∞
!(ω # ) − 1
!(ω) − 1 = P dω #
iG̃(0) G̃! (0) G̃! (0) πi −∞ ω# − ω
!(k, ω) − 1 " − + ··· = − 2 + ···
ω ω2 ω Taking the real parts of the both sides,
with
! ! dividing the integral path into -ve and +ve
d3 ξ G (|ξ| , τ ) eik·ξ ! d3 ξ G (|ξ| , τ ) = G̃(τ ) parts, and noting !2 (−ω) = −!2 (ω), we have
|ξ |<cβτ |ξ |<cβτ
!
The 1st derivative must vanish because of 2 ∞
ω " !2 (ω " )
!1 (ω) − 1 = P dω "
causality (G=0 for tau<0) and G’s continuity. π 0 ω "2 − ω 2

14
The Allison-Cobb Model Strictly speaking, the formula is valid only
We have shown that the evaluation of the for real photons. Allison and Cobb, however,
energy loss can be reduced to that of the assume that this holds approximately for
imaginary part of the epsilon. As mentioned even virtual photons, as long as they are
above, Allison and Cobb used the photo- below the free electron boundary.
absorption cross section to estimate the

h k [eV]
imaginary part. To show how, let us first Free Electron
introduce the general oscillator strength
104 Kinematic Limit
function “f” by |ω| =
! k2
2 me for B=0.9
2π 2 N e2
!2 (k, ω) = f (k, ω)
me ω
The general oscillator strength function is 103
Bound Electron |ω|
related to the dipole transition probability c
=k
q= 2meEI
for absorption or emission of a photon. Free Photon
Since the epsilon_2 is an odd function of 102
omega, the “f” is an even function of omega.
EI=30 [eV]
Since the imaginary part of the epsilon is
related to the absorption cross section we
can hence express the oscillator function in 102 103 104
h W [eV]
terms of the cross section
me c The assumption implies that the “f” does
f (k, ω) = σγ (|ω|)
2π 2 e2 Z not depend on “k” in the resonance region.

15
 ! " #
Allison and Cobb further assume that the !2 (k, ω) =
Nc
σγ (|ω|) θ |ω| −
!k 2
ωZ 2me
contribution from the region above the $ |ω| " #%
!k 2

free electron line can be approximated as a +

0
σγ (ω ! ) dω ! δ |ω| −
2me
contribution entirely coming from this line. Now all we need to do is to put this into the
The assumption is equivalent to ignoring the dE/dx formula
momentum of bound electrons. This seems a !
2e2 ∞
! ∞
!dE/dx" = − 2 dω dk
rather crude approximation, but it works as πβ 0
" # ω/cβ 2 $ # $ # $%
ω 1 ω 1
they respect the Bethe sum rule: (ωk) β − 2 2 Im + 2 Im
2
k c 2 2
$ω − k c2 kc $
!
and carry out integrations. Since the 1st
∞
f (k, ω) dω = 1
0
term in the square bracket is dominated by
and set ! " # |ω|
f (k, ω) =
me c
δ |ω| −
!k 2
σγ (ω ! ) dω !
the resonance region due to the pole of the
2π e Z 2me
propagator, we can set
2 2
0

From classical point of view, the oscillator Nc

!2 (k, ω) ≈ σγ (|ω|)
strength function counts the number of ωZ
bound electrons in a molecule contributing Since there is no k-dependence here, the k-
to a particular oscillation mode with the integral is now straightforward. As for the
frequency omega. The Bethe sum rule 2nd term in the square bracket, since it is
hence just dictates that the total number dominated by the nearly free electron
of bound electrons in the molecule is const. region, and since we are dealing with a low
density material (|!| ! 1 ), we can set
! "
Putting these together, we arrive at 1 !2
Im = − " − !2 (k, ω)
! |!|

16
The Allison-Cobb Formula Denoting the differential cross section for
With these approximations, we can carry the particle hitting the bound electron with
out the integrations in an elementary way the virtual photon by dσ , we can write
! "
and obtain dE
# ∞
dσ
! " # = N (!ω) d(!ω)
e2 ∞
Nc $ 2 %2 &− 21 dx 0 d(!ω)
!dE/dx" = dω σγ (ω) ln (1 − %1 β ) + %2 β
2 2
πβ 2 c2 0 Z ' (
+ω β − 2 atan
2 %1
)
%2 β 2
*
Since there is no fear for confusion, we put
|%| 1 − %1 β 2
) * E = !ω
Nc 2me β 2 c2
+
Z
σγ (ω) ln
!ω and compare this with the Allison-Cobb
formula. We then obtain
! +
N c ω σγ (ω # ) #
+ dω !
σγ (E) " # 2 $2 %− 12
ω 0 Z
dσ α
with Nc = ln (1 − %1 β ) + %2 β
2 2
!2 (ω) = σγ (|ω|) dE πβ 2 EZ
ωZ & ' ( )
and the Kramers-Kronig relation 1 %1 %2 β 2
+ β 2 − 2 atan
! ∞
N c! |%| 1 − %1 β 2
2 ω " !2 (ω " )
!1 (ω) − 1 = P dω "
π ω "2 − ω 2 ( )
0
σγ (E) 2me β 2 c2
+ ln
EZ E
The Differential Cross Section 1
* E
σγ (E )
"
+
+ dE "
We can reinterpret the average energy loss E2 0 Z
formula as from discrete collisions of the The 1st and 2nd lines on the R.H.S. are
particle with a bound electron exchanging a from the vector potential (transverse
virtual photon with an energy ! ω . photons) and the 3rd and 4th lines from
the scalar potential (longitudinal photons).

17
The Beta Dependence The rest of the cross section can be put
The transverse cross section vanishes in together and cast into the form

the beta->0 limit, while the longitudinal dσN C α  σγ (E) β2
=  ln $
cross section behaves as 1/beta^2. dE πβ 2 EZ % 12
2
(1 − %1 β 2 )2+ (%2 β2)
)
On the other hand, in the beta->1 limit, the
& ' ( E
σγ (E) 2me c2 1 σγ (E ! )
+ ln + 2 dE !
EZ E E Z
longitudinal cross section becomes const.  
0

a  β2 
=  ln $ % 12 + b 
β2
The 2nd term becomes important in this
2
(1 − %1 β 2 )2 + (%2 β 2 )
 
limit and is related to the emissions of 1 + (βγ) 
2
(βγ) 2

= a  ln $ % 21 + b 
Cherenkov photons. As a matter of fact, (βγ)2 2
(1 + (1 − %1 )(βγ)2 ) + (%2 (βγ)2 )
2

recalling that !2 ! 1 and |!1 − 1| " 1

The cross section hence behaves as
for a low density medium, we notice that
√ 1 + (βγ)2 ! "
below and above β = 1/ "1 dσN C
!a ln(βγ)2 + b
! " dE (βγ)2
!2 β 2 where |!1 − 1| (βγ)2 " 1 and !2 (βγ)2 ! 1 . The
atan " 0 → π
1 − !1 β 2
conditions hold over a rather wide range
Above, the 2nd term becomes for a low density medium with |!1 − 1| " 1.
! "
dσC α 1 In the relativistic region where
N ! 1−
dω c $1 (ω) β 2 1 ! (βγ)2 ! 1/|# − 1|
which is none other than the well known the X-section thus shows the relativistic
frequency distribution of the Cherenkov rise dσN C
! 2a ln(γ) + ab
radiation. dE

18
The relativistic rise comes from the fact

NE (dS/dE) (arbitrary unit)

that in the beta->1 limit the virtual photon
gets closer to the real photon pole of the Non-relativistic Region
propagator and hence acquires a longer 1/B2
range, resulting in a larger cross section. Saturation
by Density Effect
This rise, however, saturates where
(βγ)2 ! 1/|# − 1| Relativistic Rise
10
and reaches an asymptotic value
dσN C Minimum
! a ln |" − 1|−1 + ab
dE at BG ~ 4
The saturation due to a finite density of
the medium is called the density effect.
1 10 102 103 104
BG
The saturation sets in when
−1
2 ln γ ! ln |" − 1| It should also be noted that the saturation
or when depends on the photon energy. In the free
1
γ ! γ ∗ := ! electron region (high omega limit), we have
|1 − "|
4πN e2 ωp2
!=1− := 1 − 2 < 1
It is worth remembering that the minimum me ω 2 ω
occurs at around βγ ∼ 4 with almost no where the plasma frequency is given by
dependence on the medium. ωp2 := 4πN e2 /me

19
In this high frequency or high E region, the At high E values, only a quasi-free electron
gamma* is given by region will be kinematically allowed, and
ω ! m "1/2 E
e
hence the 4th term determines the trend.
γ ! ∗
= !
ωp 4πN e2 ! dσ α 1 E
σγ (E ! )
→ dE !
dE πβ 2 E 2 Z
The higher the energy, the larger the 0

gamma at which the saturation sets in. Recalling the relation between the cross
section and the oscillator strength and the
The Energy Transfer / Collision Bethe sum rule, we have
For a low E, the resonance region dominates dσ 2πe4 1 e2 ωp2
→ = 2 2
and hence the 1st and the 3rd terms of dE me c2 β 2 E 2 c β 2 N E2
!
dσ α σγ (E) " # 2 $2 %− 12 This is the Rutherford scattering formula.
= ln (1 − %1 β ) + %2 β
2 2
dE πβ 2 EZ
& ' ( )
The formula indicates that the delta-ray
1 %1 %2 β 2
production has a long tail characterized by
+ β 2 − 2 atan
N c! 1 − %1 β 2
|%| 1/E^2 behavior.
( )
σγ (E) 2me β 2 c2
+ ln
EZ E There is of course a kinematical limit to set
+
1
* E
σγ (E " ) the maximum energy transfer, but this
+ 2 dE "
E 0 Z limit is practically never reached since such
energetic collisions create delta electrons
give a major contribution. The E spectrum
which will make separate tracks.
hence reflects the resonance structures of
the photo-absorption cross section.

20
 The Bethe-Bloch Formula
Relation between Allison-Cobb and Bethe-Bloch
! " # $ % $ %
The Decomposition to T/L Parts 1 1
+∞
dE e2
= dω ω β 2 − ln
dx T (2π) c2 β 2 i −∞ $(ω) 1 − β 2 $(ω)
The dE/dx formula can be separated into
To proceed further, we will make full use of
the transverse and the longitudinal parts
! " ! " ! " the analyticity of the epsilon in the upper
dE dE dE
dx
=
dx T
+
dx L
half omega plane. We will try to move the
Let us now examine them separately. integration path to the upper semicircle,
since the epsilon reaches its asymptotic
The Transverse Part form there: 2
ωp
The transverse part is given by !(ω) → 1 −
! " # # ∞ $ % ω2
dE e2 +∞
β 2 k 2 c2 − ω 2 1
=
and hence we can carry out the integration.
2
dω ω dk
dx T (2π) c2 β 2 i −∞ (ω/cβ)2 k 2 c2 − $(ω) ω 2 k2
There is, however, a cut on the imaginary
Replacing the complex dielectric constant
axis if
by the Allison-Cobb model led us to the 1 − β 2 "(0) < 0

Allison Cobb formula. This time we will try In this case, the discontinuity across this
to carry out the integrations directly. We cut contributes to the integration along the
assume that the epsilon does not depend on real axis, too. We thus split the integral as
k following Allison and Cobb. The k-integral !
dE
" !
dE
" !
dE
"
= −
is then straightforward: dx T dx −C∞ dx Ccut

21
where we defined the integration paths as For a high beta value, the cut shows up and
in the following figure: we have
! " # $ % $ %
dE e2 1 1
= dω ω β − 2
ln
Cd dx Ccut (2π) c2 β 2 i Ccut $(ω) 1 − β 2 $(ω)
# $ %& $ %
e2 l
1 1
= − dζ ζ β − 2
ln
(2π) c2 β 2 i 0 $(i ζ) 1 − β 2 $(−δ + i ζ)
$ %'
1
− ln
1 − β 2 $(+δ + i ζ)
and hence ! " $ %
#
e2 l
1
R=d
dE
= 2 2 dζ ζ β − 2

Ccut il dx Ccut c β 0 #(i ζ)

Q
This represents the density effect.
d 0 d Putting these together, we arrive at
W
! " # $ % & ' ( % &)
dE e2 1 2 1 l
1
= 2 2 ln 2 2
The upper semicircle gives
ω −β − dζ ζ β −
dx T c β 2 p 1 − β2 0 $(i ζ)
! " # $ % $ %
1 1
The formula corresponds to the 1st and the
dE e2
= dω ω β 2 − ln
dx −C∞ (2π) c2 β 2 i −C∞ $(ω) 1 − β 2 $(ω)

=
e2
#
dω ω β −
$
1 2
% &
ln '
1
(
)
2nd terms of the Allison-Cobb formula.
(2π) c2 β 2 i −C∞ 1 − ωp2 /ω 2 1 − β 1 − ωp2 /ω 2
2

The Longitudinal Part

* $ % +
e2 1
= ω2 ln −β 2
2 c2 β 2 p 1 − β2
The longitudinal part is given by
In the low density limit, since there exits ! " # # $ %
no cut, this gives the total transverse
∞
dE e2 +∞
1
= dω ω dk 2
− 2
dx L (2π) c2 β 2 i −∞ (ω/cβ)2 k $(ω)
contribution, which is entirely specified by
Slightly going off the Allison-Cobb model
the plasma frequency of the medium.
we assume that epsilon is k-independent.

22
We carry out the k-integration to get Recalling the kinematic relation
! " # $ %#
dE e2 +∞
ω dω ∞
dk 2 2me c2 β 2 γ 2
=
(2π) c2 β 2 i
− ! ωmax =
dx L −∞ $(ω) (ω/cβ)2 k2 1 + 2γ (me /m) + (me /m)2
# $ % $ %
=
e2
Im
+∞
−
ω dω
ln
2
kmax c2 β 2 we have
π c2 β 2 0 $(ω) ω2
(!kmax )2 ! 2me !ωmax = 2me Tmax
We recall here that epsilon becomes real in
the omega->0 limit. This means that for the With these, we arrive at
! " # $
e2 1 2 2 me c2 β 2 Tmax
epsilon to acquire an imaginary part, there dE
= 2 2 ωp ln
dx c β 2 I2
must be photon absorption that in turn
L

requires some finite amount of energy to that corresponds to the 3rd and 4th terms
excite the lowest level. There must hence (longitudinal part ) of the Allison-Cobb
be a lower limit to the omega range of the formula.
integration as well as to the k-integration
The Bethe-Bloch Formula
range.
Noting that Putting the T and L parts together, we get
! " # $ % & % & '
! ∞
1
! +∞
1
! dE e2 1 2 2 me c2 β 2 Tmax 1
− Im
ω dω
=
i ω dω
=
i ω dω
=
π 2
ω = 2 2 ω ln + ln −β 2

"(ω) 2 "(ω) 2 1 − ωp /ω
2 2 2 p dx c β 2 p I2 1 − β2
−∞ −C∞
&)
0
( l %
1
we introduce such a limit on the k-integral
2
− dζ ζ β −
0 $(i ζ)
with a frequency ω̄ defined by with
ωp2 := 4πN e2 /me
! " #
and
+∞
ω dω π 2 π 2
Im − ln(!ω) ≡ ωp ln I ≡ ω ln(! ω̄)
"(ω) 2 2 p
0 Tmax = ! ωmax ! 2me c2 β 2 γ 2
where “I” is the effective binding energy.

23
 The dE/dx Fluctuation
Energy loss per a finite sample thickness
! " #
The General Formula
∞
∂F (x, ∆) dσ(E)
= N [F (x, ∆ − E) − F (x, ∆)] dE
∂x 0 dE
Let x be a finite sample thickness and ∆
By making Laplace transform
be the corresponding energy loss. ! ∞
What we need is a probability distribution F̄ (x, s) = d∆ F (x, ∆) e−s ∆
0
function: F (x, ∆) . The distribution must 1
! +i∞+σ

satisfy the following equation F (x, ∆) = ds F̄ (x, s) es ∆

2πi −i∞+σ
! " # $ %
F (x + δx, ∆) = F (x, ∆) 1 −
∞
N δx
dσ(E)
dE
we obtain
dE ! ∞ " #
0 ∂ F̄ (x, s) dσ(E) $ %
" ∞ # $ = −F̄ (x, s) dE N 1 − e−sE
dσ(E) ∂x dE
+ F (x, ∆ − E) N δx dE 0
dE
0
with the boundary condition
where the 1st term on the R.H.S. is the F̄ (x, s) = 1
probability of losing ∆ in the first x and corresponding to its original form:
then nothing happening in the next δx , F (0, ∆) = δ(∆)
while the 2nd term on the R.H.S. gives the
The equation is readily solved to yield
sum of the probabilities of losing ∆ − E in ! " ∞ # $ '
dσ(E) % &
the first x and E in the subsequent δx . F̄ (x, s) = exp −x dE N 1 − e−sE
dE
This leads us to the following equation
0

24
By inverse Laplace transformation, we get We will discuss physical meanings of this
!
F (x, ∆) =
1 +i∞+σ
ds exp [ s ∆
assumption later. The assumption ensures
2πi −i∞+σ ! " # & that we can always choose E1 in such a way
∞
dσ(E) $ %
−x
0
dE N
dE
1 − e−sE that E ! E ! 1/s 0 1

This is a general solution. In principle we Then we can separate the integral into two
can numerically calculate the probability parts (0, E1 ) and (E1 , ∞), and in the first we
distribution once a concrete expression is can make approximation
given for the differential X-section (e.g. 1 − esE " −sE
the Allison-Cobb). Landau analytically did to obtain
the integral with the Rutherford scattering ! ∞ "
dσ(E)
#
$ %
! E1 "
dσ(E)
#
1−e−sE
dE = s
cross section.
dE E
0 dE 0 dE
! ∞ " #
dσ(E) $ %
+ 1 − e−sE
Landau Distribution
dE
E1 dE

Let E0 be a characteristic energy (of the Recalling the Rutherford formula

order of average binding energy) of atoms dσ
=
2πe4 1
in the medium and Emax be the maximum dE me c2 β 2 E 2
energy transfer from the particle to an we obtain for the 1st term
electron in the medium. Landau assumed ! E1 "
dσ(E)
#
2πe4
"
E1
#
dE E= ln
that the integral in the exponent comes 0 dE me c2 β 2 E!

only from the region where we choose the lower limit of the
1/Emax ! s ! 1/E0 integral so as to reproduce the Bethe-Bloch

25
 ln E = ln
! I2
+ β2
The 2nd term now becomes
2me c β γ
2 2 2 ! ∞
1 − e−sE
dE = s (1 − γE − ln sE1 )
In this way, we can take the bound electron E1 E2

effects approximately. Putting the 1st and 2nd terms together, we

On the other hand, the 2nd term can be arrive at
! " #
partially integrated to give ∞
dσ $ %
! ! ∞ xN dE 1 − e−sE = ξ s (1 − γE − ln sE # )
∞
1 − e−sE 1 " # e−sE 0 dE
dE = 1−e−sE1
+s dE
E1 E2 E1 E1 E with 2πN e4
$ ! ∞ % ξ := x
e−sE me c2 β 2
" s 1+ dE
E1 E
Introducing a dimensionless scaling variable
Again using sE1 ! 1 , we have ! "
ξ
! ∞ ! ∞
∆ − ξ ln E! + 1 − γE
e−sE e−z η(ξ, ∆) :=
dE = dz ξ
E1 E sE1 z
! 1
dz
! 1
e−z − 1
! ∞ −z
e
we finally arrive at
= + dz + dz !
sE1 z sE1 z 1 z 1 1 +i∞+σ
1
! 1 ! 1 −z ! ∞ −z F (x, ∆) = du e u ln u + η u =: φ(η)
dz e −1 e ξ 2πi −i∞+σ ξ
" + dz + dz
sE1 z 0 z 1 z
The universal function φ(η) attains its
! 1
=
dz
− γE maximum of about 0.18 at
sE1 z
η = ηMPV ! −0.05
where the Euler constant is
γE ! 0.577 It has a FWHM of about 4, suggesting that
xi sets the scale of energy loss.

26
 0.2
Region of Applicability
F(H)
Let us now examine the approximation we
0.18
made to derive the Landau distribution and
0.16
clarify the region of applicability.
0.14
Inspection of !
0.12 1 +i∞+σ
φ(η) := du e u ln u + η u
0.1
2πi −i∞+σ
FWHM ! 4 tells us that the major contribution comes
0.08
from the region where ln u ~ 0 or u ~ 1. The
0.06
assumption for the important “s” region
0.04
1/Emax ! s ! 1/E0
0.02 ηMPV ! −0.05
0 -4
can hence be translated to
-2 0 2 4 6 8 10 12 14
H u E0 u Emax
sE0 =
ξ
E0 ∼
ξ
"1 and sEmax =
ξ
Emax ∼
ξ
"1

The Most Probable Value (MPV) position is or equivalently

2πN e4
!
ξ
" E0 ! ξ := x 2 2
! Emax " 2me c2 β 2 γ 2
∆ = ∆ MPV := ξ ln + 1 − γE + η MPV me c β
E!
! "# ! " $
It says that the medium must be thick
2πN e 4πN e4 γ 2 x enough for “xi” to be much larger than the
4
= x ln − β + 1 − γE + η MPV
2
me c2 β 2 I2
binding energy scale, while it must be thin
which shows roughly the same beta enough to be much less than the maximum
dependence as the Bethe-Bloch formula. energy transfer per collision.

27
 Comparison with Data
Allison-Cobb (1980)
The energy loss

∆ = nI W

# ionization
electrons

Average energy
for creation of 1
electron

Some numbers to
remember

WAr = 26 [eV]
nI (1 [cm] Ar) ! 100
# primary clusters
! 30
1 [cm] of Ar

28
 Cluster Size Distribution
The number of electrons per cluster
Charge Size for inclined tracks. What we need to know
Depending on the number of electrons made is the probability of the primary ionizing
by 2ndary ionizations, each primary cluster collision with an energy transfer “E” to yield
has different number of electrons in it. Its “k” 2ndary electrons: P2 (k; E). Once this is
distribution function is not easy to calculate known, we can calculate the cluster size
from the 1st principle, since the number of distribution as
! " #
electrons per cluster is typically a few and 1 dσ
Pcl (k) = dE P2 (k; E)
hence its statistical treatment as with the σ dE
number of ionization electrons per a finite This was done by Lapique and Piuz (1980)
sampling thickness is inadequate. for a pure Ar gas. As said above, however,
What we actually measure with a TPC is the calculation of P2 (k; E) is a complicated
usually the charge collected on a pad with a process, since 2ndary ionizing collisions are
finite size. Individual primary clusters seem no longer statistically independent. Their
not to be our concern. This is, however, work was hence only partially successful.
certainly wrong for pixel readout. Even for There is, however, a beautiful measurement
conventional pad readout, the cluster size by the Heidelberg group (Fischle et al 1991)
fluctuation might be a concern, since its that can be used in Monte Carlo simulations.
fluctuation affects the spatial resolution

29
 Cluster Size Data
The Heidelberg Group Experiment (Fischle, Heintze, and Schmidt 1991)

30
 Cluster Size Distribution
Geometrical Size of Cluster
Geometrical Size Valid for low and intermediate Z
The delta electrons have a finite range and
hence give a finite geometrical size to the
primary ionization cluster.
Since the electric field due to a relativistic
charged particle is perpendicular to its
trajectory, the delta electron tends to be
kicked out in the perpendicular direction.
If the delta ray has a finite range, it would A = 5.37 × 10−4 [g cm−2 keV−1 ]

result in a shift of the charge centroid in B = 0.9815

that direction, deteriorating the spatial C = 3.1230 × 10−3 [keV−1 ]

! "
resolution. In the case of a TPC operated in R(E) = A E 1 −
B
1+CE
a high magnetic field, the delta electron is
curled up and hence the transverse cluster
size can hardly be affected.
At B=0, however, the effect may be visible. Typical delta-ray range values in Ar (N.T.P.)
I quote here an empirical formula by R(1 [keV]) ! 30 [µm]
Kobetich and Katz (1968): R(10 [keV]) ! 1.5 [mm] : 0.05 % of collisions

31
 Cluster Size Distribution
Appendix (Delta-ray Kinematics)
Kinematics For a minimum ionizing particle ( βγ ! 4 ), we
Let the angle between the incident particle hence have
direction and that of the delta-ray emission E
cos2 ψ !
be ψ , and let the maximum kinetic energy 2me c2 + E
transfer be Emax , we have
This formula tells us that on a purely
2me c + Emax E
2
kinematical basis, delta-ray emission should
cos2 ψ =
2me c2 + E Emax approximately be perpendicular to the
with incident particle direction, as long as we are
2me c2 β 2 γ 2 talking about a delta-ray with an energy
Emax =
1 + 2γ (me /m) + (me /m)2
negligible compared to the electron mass.
for a free electron at rest.

This implies that

Emax ≈ 2me c2 (βγ)2 " (βγ)2 [MeV]

for a particle with m ! me .

32
Classical Theory of
Electrons in a Gas

33
 Why Classical Theory?
Is it OK to treat it classically instead of quantum mechanically?

Inter-molecular Distance
NA = 6.02 × 1023 [mol−1 ] V1 = (3.3 [nm])3 per molecule
V = 2.24 × 10 [cm /mol] 4 3 at 0 ◦ C, 1 atm
RAr = 0.18 [nm] Inter-molecular distance
Thermal Energies ! 20 × RAr
! " ! " !
1 1 3 !v 2 " = 0.39 × 10−3 c
m v2 = MV2 = KT = 0.039 [eV]
2 2 2 !" #
! V 2 = 1.4 × 10−6 c
! = 0.98 [nm]
2
m !v " !" #
2
! V
! = 3.6 × 10 −3
[nm] $ # 4 × 10−3
MAr !V 2 " !v 2 "
The de Broglie wave length of the electron is small enough compared to
the inter molecular distance implying that it is much smaller the mean free
path. The C.O.G. motion of the electron can hence be treated classically.

34
 Boltzmann Equation
Basic Equation Governing Electron Transportation
We often see formulae for electron drift and diffusion as derived
from it, so it must be useful, but itself is rarely discussed in usual
introductory text books.
So, what is it?
Where does it come from?
And how?

I can only show you a rough sketch, but I hope it will make you feel
a little bit more comfortable when you see it next time.
For (older) pragmatic people, it might become a little bit boring, but
maybe it’s OK even for them to recall their student time.

We will find that THE KEY WORD IS “PROJECTION”!

“PROJECTION” is a technique to forget about
unwanted details and make life easy!

35
 Phase Space
Stage where solutions dance, we only see their shadows
Our system of interest
Ionization electrons drift and diffuse independently
It suffices to consider a single electron in a gas consisting of N
gas molecules in a chamber (note: N is a huge number).

Molecules’
Microscopic Picture
Phase Space
Sub-Space Causal deterministic motion by
6(N + 1)-dim.
sample p2 !N
P 2b
point (a) H = +
Solution Lines 2m 2Mb
(X a , P a ) b=1
+Uext (x, p) : Lorentz force (E,B)
+UmM (x, X) : elec. + mol. collision
!
+UMM (X, X ) : mol. + mol. collision
Projection p = P0 Macroscopic Picture
Projection = Coarsification
A shadow trajectory = information loss
x = X0 Electron’s Sub-Space
Stochastic probabilistic motion

36
 Liouville’s Theorem
Solutions flow as perfect incompressible fluid
Hamiltonian Equation of Motion D(t) forms an Abelian group:
Motion of a phase space point: D(t1 )D(t2 ) = D(t1 + t2 )
! " D(0) = 1
X
Φ= D(−t)D(t) = D(t − t) = D(0) = 1
P
satisfies
! " Liouville’s Theorem
∂H P ! "
Φ̇ = with Φ̄ = ∂D(t)Φ
T −X J(t) = det =1
∂ Φ̄ ∂Φ
The map preserves phase space volume.
Time Evolution Operator
Write its formal solution as Liouville’s Equation
State density
Φ(t) = D(t) Φ(0) ρ(Φ; t) = ρ(X, P ; t) function

then this is a 1-to-1 map because ∂ ∂H ∂

of the uniqueness of solution. 0= ρ+ ρ
∂t ∂ Φ̄ ∂ΦT

37
 Proof of Liouville’s Eq.
Proof is easy enough to give here
Proof of Liouville’s Theorem Derivation of Liouville’s Equation
Equation of motion says In general, for any observable A:
∂H
D(dt)Φ = Φ + dt Φ̇ = Φ + dt T d ∂ ∂
∂ Φ̄ A(Φ; t) = Φ̇ A(Φ; t) + A(Φ; t)
resulting in dt ∂Φ T ∂t
∂D(dt)Φ ∂2H ∂H ∂ ∂
= 1 + dt T = A(Φ; t) + A(Φ; t)
∂Φ ∂ Φ̄ ∂Φ ∂ Φ̄ ∂ΦT ∂t
We hence have
! " ! "
∂D(dt)Φ ∂2H # $ Since Liouville’s theorem requires that
det = 1 + dt Tr + O (dt)2
the state density stays unchanged,
∂Φ T
∂ Φ̄ ∂Φ
# $
= 1 + O (dt)2 which implies
1 1 d
(J(dt) − 1) = (J(dt) − J(0)) = J(0) = 0 d ∂H ∂ ∂
dt dt dt 0= ρ= ρ + ρ
which leads us to dt ∂ Φ̄ ∂ΦT ∂t
d ∂
!"
∂ D(t − t1 )D(t1 )Φ
#"
∂ D(t1 )Φ
#$
This is actually a continuity equation in
J(t) = lim det
dt t1 →t ∂t ∂ D(t1 )Φ ∂Φ
the full phase space of the system or
! " #$ ! " #$
= lim
∂
det
∂ D(t − t1 )D(t1 )Φ
· det
∂ D(t1 )Φ
conservation of probability:
t1 →t ∂t ∂ D(t1 )Φ ∂Φ !
! $
=
d
J(0) · J(t1 ) = 0 d6(N +1) Φ ρ(Φ; t) = 1
dt ∴ J(t) = 1

38
 State Density Function
All we know about the ensemble
Microscopic Picture
Once an initial distribution is given, the state density function evolves
deterministically according to Liouville’s equation.
ρ(Φ; t) = ρ(D(−t)Φ; 0)
The bundle of solution lines forms a But how should we fix the initial
manifold consistent with constraints
distribution?
imposed upon the system such as
conservation of total energy and Ergodic hypothesis:
chamber volume boundaries
Probability is proportional to phase
{X b , P b } space volume
Dynamical
variables to Thermal equilibrium = Equal weight
be integrated Projection
out Macroscopic Picture
p = P0
Projection = Coarsification
= information loss
x = X0 Dynamical variables of interest
Projected volume decides probability

39
 Maxwellian Distribution
A detour which proves the power of ergodic hypothesis
State Density Function for Molecules
Ignore the electron, for the moment, and concentrate on the molecules, whose sate density function in thermal
equilibrium. Good approximation since we can safely assume that the molecules colliding with the electron never
met it in the past.
Interaction hamiltonian of the molecules has a nonzero
value only when the inter-molecule distance becomes
negligibly small compared to its average determined by
The strip is actually the the gas density.
surface of a 3(N-1)-dim. !N
1 !N

!N sphere: ! N Etot = Mb V 2b := Y 2b = R2
Y 2b = R2 − Y12 2
Y 2b = R2 b=1 b=1
b=1
b=2 The phase space points uniformly distribute over the
! surface of a 3N-dim. sphere of radius R=sqrt(Etot) x
Y1= Mb /2 |V 1 | 3N-dim. box with a volume L^{3N}. Note that the
projection of spatial dimension simply gives L^{3N}.
Projection !" #
S3(N −1) R2 − Y12
|Y 1 | = Y1 !" #3(N −1)−1 $ %3N/2
subspace 2
! ∝ 2
R − Y1 2 ≈R 3N −4
1 − (Y1 /R)
R
R2 − Y12 &
Y12 /((2/3)R2 /N )
'3N/2
≈R 3N −4
1−
3N/2
& 2
' & 2
'
Y Y
→ R3N −4 exp − 1
∝ exp − 1
(2/3)(R2 /N ) kB T

40
 Projection of Liouville’s Eq.
Electron distribution as the projection of the full state fun.

Liouville’s Equation Notice that ! "

p
State density F ext = e E + ×B (e < 0)
ρ(Φ; t) = ρ(X, P ; t) function
mc
The “molecule only” terms become
∂ ∂H ∂
0= ρ+ ρ surface integrals upon integration and
∂t ∂ Φ̄ ∂ΦT
vanish because rho has the same value
Separating the part containing the everywhere on the surface.
electron’s dynamical variables from the ! "
∂ p ∂ ∂
rest, we have 0= + · + F ext · f
! " ∂t m ∂x ∂p
∂ p ∂ ∂ N $ $ ! "
0= + · + F ext · ρ # ∂ ∂
∂t m ∂x ∂p + d3 X b d3 P b F b · − fb
N ! " ∂p ∂P b
# ∂ ∂ b=1
+ F b· − ρb where we have introduced
∂p ∂P b
b=1
1-body distribution" function:
$ #
+ Molecule Only Terms
N
!
f (x, p; t) = d6 Φb! ρ (x, p; {Φb! } ; t)

Now project the both sides to the

b! =1

2-body distribution function:

electron subspace by integrating out ! "# $
molecules’ dynamical variables. fb (x, p; X b , P b ; t) = d6 Φb! ρ (x, p; Φb , {Φb! } ; t)
b! !=b

41
 Collision Term
Time average over the collision period
! "# # $

Collision Term b! !=b

d 3 X b! d 3 P b! H(Φ)
H̄2 (x, p; X b , P b ) = ! "# # $
We move the 2-body term to the R.H.S. d 3 X b! d 3 P b!
and call it the collision term: b! !=b
p2 P 2b
! " N $
# $ % & = + + UmM (|x − X b |)
∂f ∂ ∂ 2m 2Mb
=− d3 X b d3 P b F b · − fb + const.
∂t coll ∂p ∂P b
which is none other than the 2-body
b=1

for obvious reason. Notice that if it Hamiltonian describing the collision.

were not for this term, the electron Then we have
would have behaved as a single particle in ! t+ ∆t " # ! t+ ∆t $ %
∂ ∂ ∂f
external E and B fields.
2 2
b
− dt" F b · − fb = dt" "
With this term, however, the projected t− ∆t
2 $
∂p ∂P b
% $ t− ∆t
2 %
∂t
trajectory of the electron will show a ∆t ∆t
= fb φ2 ; t + − fb φ2 ; t −
shaggy apparently random motion, though $
2
% $
2
%
the full trajectory should be smooth and = fb D2 (−∆t)φ2 ; t −
∆t
− fb φ2 ; t −
∆t
causal in the full phase space. 2 2

Noting that the 2-body system can be where D2 (t! ) is the 2-body time evolution
regarded as isolated during the short operator and
period of collision time and the collision φ2 = (x, p; X b , P b )
motion averaged using the projected H: is the 2-body phase space point in question.

42
 Collision Term (Continued)
Decomposition of 2-body fn. to products of 1-body fns.
Before and after the collision period of fb (φ2 ; t) = fb (x, p; X b , P b ; t)
the 2-body system, their space coordinates ≈ f (x, p; t) Fb (P b ; t)
don’t change macroscopically, but their Notice that the momentum transfer is
momenta may seem to jump by a finite determined by the relative momentum and
amount. the impact parameter. This replacement
Microscopically, however, the jump is a drops the information on the impact
function of the impact parameter and their parameter by throwing away the coordinate
relative momentum and should be causal in information of the molecule. This loss of
our classical mechanical treatment. information is the source of the stochastic
! t+ ∆t
2
"
∂ ∂
# nature of the collision process.
− dt" F b · − fb
t− ∆t ∂p ∂P b We hence make the replacement
2 $ % $ % ! ! ! "
∆t ∆t p = mv
= fb D2 (−∆t)φ2 ; t − − fb φ2 ; t − d3 X b → dσb |v − V b | ∆t Pb = Mb V b
2 2
Probabilistic view point enters upon since the volume integral should be taken
replacing the 2-body state density function over the region where
by the product of the 1-body state density D2 (−∆t) "= 1
functions for the electron and the molecule or over the X-section along the expected
in question. trajectory of the 2-body system.

43
 Collision Term (Continued)
Time average over the collision period
Time Averaged Collision Term Since the same kind of molecules should
Averaged over the collision time, we get contribute equally to the summation (rho
! t+ ∆t " #
1 2 ∂f should be symmetric under exchange of
dt" the same kind of molecules), we can rewrite
∆t t− ∆t
2
∂t coll
$N ! ! this to
! " $ $
= d3 P b dσb |v − V b | ∂f #
= Nk d3 P dσk |v − V |
b=1 ∂t coll
× [f (x, p + ∆q; t) Fb (P b − ∆q; t) k
× [f (x, p + ∆q; t) Fk (P − ∆q; t)
− f (x, p; t) Fb (P b ; t)]
− f (x, p; t) Fk (P ; t)]
In what follows we understand the time where Nk is the number of molecules of k-
derivative as appropriately averaged over th kind. Noting
the collision period as above, and simply ! ! !
write 1= d3 X d3 P Fk (P ; t) = L3 d3 P Fk (P ; t)
! " N $ $
∂f # we define the density of molecules of k-th
= d3 P b dσb |v − V b |
∂t coll
b=1
kind nk = Nk /L3 and
× [f (x, p + ∆q; t) Fb (P b − ∆q; t)
F̄k (P ; t) = L3 Fk (P ; t)
− f (x, p; t) Fb (P b ; t)]

44
 The Boltzmann Equation
The fundamental equation
Then we finally arrive at the Boltzmann
Flowing in P − ∆q
equation:
! "
∂ p ∂ ∂ P
+ · + F ext · f (x, p; t)
∂t m ∂x ∂p p
# $ $
= nk d3 P dσk |v − V |
p + ∆q p + ∆q
k%
× f (x, p + ∆q; t) F̄k (P − ∆q; t)
TP = C
&
− f (x, p; t) F̄k (P ; t) p
part flowing in P
part flowing out

where the external force is given by P − ∆q Flowing out

! v "
F ext =e E+ ×B
c
and the velocities are defined by
p = mv
P = Mk V

45
 Inelastic Scattering
A short comment in passing
So far, we have been assuming that the The Boltzmann equation, as it is, can hence
electron-molecule collisions are elastic as be applied to those more general cases.
described by a scattering potential.
In practice, however, the inclusion of
If we are to consider inelastic scattering inelastic processes complicates the
involving some change of internal degrees treatment significantly, since we can no
of freedom of the colliding molecule, we longer assume that the relative speed
need to expand the phase space to include stays the same before and after the
the internal degrees of freedom and then collision.
project out these internal degrees of
freedom as needed. After all, the physics that controls the
The resultant loss of information can again electron transport in a gas lies in the
be taken statistically into account as in the collision term, and that’s where all the
form of the inelastic cross section. complications come from. Calculating the
properties of complex molecules from the
We can hence regard the Boltzmann eq. as 1st principle (=Q.M.) is often impracticable.
the one after this extra projection.

46
 Transport Coefficients
Things you want to derive from the Boltzmann Equation

We often see formulae for electron drift and diffusion as derived

from the Boltzmann equation, but they are given almost always
without proof.
Where do they come from?
And how?

I can only show you a rough sketch, but I hope it will make you feel
a little bit more comfortable when you see them next time.
Some of you, pragmatic people might already have been pretty much
fed up, but be patient recalling your student time.

Again we will find THE KEY WORD IS “PROJECTION”!

“PROJECTION” makes life easy!

47
 The Boltzmann Equation
From now on we will work in velocity space
The Boltzmann equation in (x,v) space is
Flowing in V +∆V
readily read out from its (x,p) version:
! "
∂ ∂ F ext ∂
+v· + · f (x, v; t) V
∂t ∂x m ∂v v
# $ $
= nk d3 V dσk |v − V |
v + ∆v
k % v + ∆v TP = C
× f (x, v + ∆v; t) F̄k (V + ∆V ; t)
&
− f (x, v; t) F̄k (V ; t) v
part flowing in V
part flowing out

where the external force is given by V +∆V Flowing out

! v "
F ext =e E+ ×B Before moving on, it is worth noting that
c
the Boltzmann eq. implies a scaling between
and the velocity changes must satisfy
the gas density and the field strengths for
∆q = m∆v = −Mk ∆V steady state solutions having no (x;t)
dependence.

48
 Velocity Space
Decomposition of f(x,v;t) to n(x;t) fbar(v;x,t)
The probability density of finding the Putting this into the Boltzmann equation,
electron in the vicinity of x is given by we have
! ! "
∂ ∂ F ext ∂ # ¯$
n(x; t) = d3 v f (x, v; t) +v· + · nf
∂t ∂x m ∂v
% & &
With this, we can define the velocity = n(x; t) nk d3 V dσk |v − V |
distribution function by ' k
¯
× f (v + ∆v; x, t) F̄k (V + ∆V ; t)
f¯(v; x, t) := f (x, v; t) / n(x; t) (
¯
− f (v; x, t) F̄k (V ; t)
By definition this must satisfy the
Notice that on the R.H.S. (collision term),
normalization condition:
! n(x,;t) has been factored out, since the
collision is a very local phenomenon.
d3 v f¯(v; x; t) = 1

as is obvious by integrating both sides of It is tempting to assume that fbar will soon
the following over velocities become independent of position and time
due to random collisions with molecules, but
f (x, v; t) = n(x; t) f¯(v; x, t)
this turns out incorrect as we will see next.

49
Simple Minded Factorization
f(x,v;t) = n(x;t) fbar(v) does not work!
Integrating both sides of the B.Eq. over
!
the electron velocities, we have d3 x n(x; t) = 1
!
∂ ∂
"
F ext ∂ f ¯ # we have
+ · !v" + d3 v · n(x; t) !
∂ " # F ext ∂
∂t ∂x
"
m ∂v d3 x · v nf¯ + · f¯
$ " " ∂x m ∂v
= n(x; t) d3 v nk d3 V dσk |v − V | $ ! !
% k = nk d3 V dσk |v − V |
× f¯(v + ∆v; x, t) F̄k (V + ∆V ; t) k %
& × f¯(v + ∆v; x, t) F̄k (V + ∆V ; t)
− f¯(v; x, t) F̄k (V ; t) &
where ! − f¯(v; x, t) F̄k (V ; t)
!v" := d3 v f¯(v; x, t) v The 1st term on the R.H.S. is zero since it
becomes a surface integral where n=0.
is the local average velocity, which is in
Combining this with the eq. on the left page
general position dependent.
yields ! "
If we assume a simple minded factorization ∂ ∂
+ !v" · n(x; t) = 0
f (x, v; t) ≈ n(x; t) f¯(v) ∂t ∂x
and integrate the both sides of the B.Eq. which implies a simple drift w/o diffusion,
over the electron positions, noting possible only if n is uniformly distributed.

50
 Concept of Velocity Shell
Towards more realistic solutions to the B.Eq.
We will hence be forced to retain the time v1 The 3rd axis in the
and position dependence in fbar and think direction of the average
about another way of approximation. velocity of the shell

The Basic Idea

v
The motion of the electron is dominated by
random and almost isotropic velocity with a φ
small modulation (drift velocity) due to the θ v3
external E and B fields.
!v"shell
The Velocity Shell
We hence consider a fraction of the phase
space where the electron has speed in the v2
range (v, v+dv), a shell of a 3-dim sphere in
the velocity space. We will then consider the velocity space in
For the class of phase space points in the a spherical coordinate system:
velocity shell, the distribution should be Ω = (cos θ, φ)
almost isotropic with the small modulation.

51
 Harmonic Expansion
Expansion in terms of spherical harmonics
Harmonic Expansion Since we took the 3rd axis in the direction
In each velocity shell, we expand fbar in of the average velocity of the shell, this
terms of spherical harmonics as implies
!∞ m=+l
! f¯1−1 = f¯11 = 0
¯
f (v; x, t) = Ylm (θ, φ) f¯lm (v; x, t)
l=0 m=−l Ignoring l>1 terms, we can put
The distribution will then be dominated by f¯(v; x, t) ≈ f0 (v; x, t) + f1 (v; x, t) cos θ"
!v
low l spherical harmonics, = f0 (v; x, t) + f 1 (v; x, t) ·
v
l=0 (scalar=monopole) : dominant where  
0
l=1 (vector=dipole) : drift f 1 (v; x, t) :=  0 
f1
Average shell velocity
"! The average shell velocity then becomes
!
!v"Ωv = dΩv v f¯(v; x, t) dΩv f¯(v; x, t) vf 1
!v"Ωv =
  3f0
f¯1−1 − f¯11
v
=
6 f¯0
¯−1 + f¯1 )
√ 0 −i (f√1 1 Notation !
¯
2 f1
0
∗
!l m| [Object]" = dΩ (Ylm ) [Object]

52
 Harmonic Expansion
Projection of B.Eq. to harmonic components
Harmonic Expansion of B.Eq. The Scalar Equation (l=0)
! "
All we need to do is to put ∂ v ∂ 1 ∂ 4π 2 eE
(nf0 ) + ·(nf1 ) + v ·nf 1
∂t 3 ∂x 4πv 2 ∂v 3 m
f¯(v; x, t) ≈ f0 (v; x, t) + f1 (v; x, t) cos θ" #
!v 1 ∂
= f0 (v; x, t) + f 1 (v; x, t) · =n nk σ̄m,k (v; [f0 ])
v 4πv 2 ∂v
k
into the Boltzmann equation, and project where σ̄m,k is in general a complicated fn.
out l=0 (scalar) and l=1 (vector) components If collisions are all elastic, a concrete
formula is known (c.f. Huxley & Crompton)
!0 0| [B.E.]" = Scalar Eq.
n nk σ̄m,k (v; [f0 ]) ! $
!1 0| [B.E.]" = Vector Eq. " 2#
V ∂f0
m
= 4πv 2 n nk v σm,k (v) v f0 +
This projection is a tedious but doable Mk 3 ∂v
mathematical exercise, at least for the
L.H.S. of the Boltzmann equation. All you effective collision frequency
νm,k := nk v σm,k (v)
need to know is the composition rules of
the spherical harmonics, which you must The scalar equation can be interpreted as
have learned in a Q.M. course. the continuity equation expressing energy
conservation.
I just show the results of the exercise.

53
 Scalar Equation
Interpretation of Scalar Eq.
First recall that the total weight of the Putting these into the scalar equation
velocity shell (v,v+dv) is ! "
∂ v ∂ 1 ∂ 4π 2 eE
(nf0 ) + ·(nf 1 ) + v · (n f 1 )
ñ dv = (4π v 2 dv) (n f0 ) ∂t 3 ∂x 4πv 2 ∂v 3 m
# %
1 ∂ $
while the shell averaged velocity is given by = n nk σ̄m,k (v; [f0 ])
vf 1 4πv 2 ∂v
k
!v"Ωv =
3f0 and canceling out common factors, we get

net loss of the shell population net gain of the shell population
due to drift due to collisions

# $ % '
∂ ∂ ! " ∂ e E !v"Ωv ∂ &
ñ + · ñ !v"Ωv + · ñ = n nk σ̄m,k (v; [f0 ])
∂t ∂x ∂v m v ∂v
k

change rate of the net loss of the shell population due heating up due to external fields
shell population

The shell population times the mv^2/2 is the total energy of the shell, and hence the
conservation of population is equivalent to that of energy.

54
 Harmonic Expansion
Projection of B.Eq. to harmonic components (continued)
Momentum Transfer X-Section The Vector Equation (l=1)
The collision term is characterized by a ∂ ∂ eE ∂
(n f 1 ) + v (n f0 ) + (n f0 ) − ω × (n f 1 )
quantity called the momentum transfer ∂t ∂x m ∂v
cross section. = −ν̄m (v) (nf 1 )
where
It is defined in general by !
ν̄m := nk v σm,k (v) : effective coll. freq.
vr!
σm,k = σ0,k − σ1,k k
vr (−e)B
! ω :=
where vr and vr are relative speeds of mc : cyclotron freq. vec.
electrons in the molecule rest frame
Notice that the electron charge is -ve,
before and after the collision, and their
hence (-e) is +ve.
ratio is unity for elastic scattering, and
!
σ0,k = dσk The vector equation can be interpreted as
! the continuity equation expressing
σ1,k = dσk cos θ θ momentum conservation.

!
p σm,k = dσk p (1 − cos θ)

55
 Vector Equation
Interpretation of Vector Eq.
The effective collision frequency is related Multiplying the both sides of the vector eq.
to mean free time 1 by tau with this in mind makes the
τ=
ν̄m meanings of the vector eq. clearer.
On the other hand the total momentum of
the velocity shell (v,v+dv) is This part remains even after the steady
v f1 4π v 2 dv state is reached and hence should be kept
dptot = (4π v dv) (n f0 ) m
2
= m v (n f 1 )
3 f0 3 as significant.

Quickly become velocity increase velocity change by

small after injection
path length during tau a single collision
! "
∂ ∂ eE ∂
τ (n f 1 ) + (τ v) (n f0 ) + τ (n f0 ) − (ω τ ) × (n f 1 ) = − (nf 1 )
∂t ∂x m ∂v

change of rotation during tau

Can be large
distribution for a point
during tau source

56
 Vector Equation
Separation of Drift and Diffusion
The Vector Equation ν̄m (v) (nf E ) − ω × (n f E ) = −
eE ∂
(n f0 )
m ∂v
∂ ∂ eE ∂
(n f 1 ) + v (n f0 ) + (n f0 ) − ω × (n f 1 ) ∂
∂t ∂x m ∂v ν̄m (v) (nf G ) − ω × (n f G ) = −v (n f0 )
∂x
= −ν̄m (v) (nf 1 ) Notice that these are linear equations
We assume that the 1st term (t-derivative) of the form ! "
is negligible compared with the rest. This [ ν̄m (v) − ω× ] n f E/G = [fn. of f0 ]
assumption implies that the electron is in a that can be solved by matrix inversion,
quasi-equilibrium at least locally. ! "
−1
n f E/G = [ ν̄m (v) − ω× ] [fn. of f0 ]
Then we have
once f0 is given.
∂ eE ∂
v (n f0 ) + (n f0 ) − ω × (n f 1 ) Notice also that upon the integration over
∂x m ∂v
x the contribution from fG must vanish.
≈ −ν̄m (v) (nf 1 )
!
We now decompose f1 as [ ν̄m (v) − ω× ] d3 x (n f G )
f1 = fE + fG !
∂
= −v d3 x (n f G )
∂x
to separate the vector eq. into the
= Surf. int. = 0
following two:

57
 Vector Equation
Separation of Drift and Diffusion
Now recall that f1 is related to the drift This means that the fG and hence WG does
velocity of the shell through not contribute to the average velocity of
vf 1 the whole ensemble:
!v"Ωv =
3f0 !v" = !W E "
We can hence rewrite the average velocity and
of the shell as !W G " = 0
!v"Ωv =: W = W E + W G We can hence interpret WE as the drift
with velocity due to the external field and WG
vf E/G
W E/G := as the convection velocity due to diffusion
3f0
of the velocity shell at a given spatial point.
Notice that W is a function of the speed v
and the position of the electron, and the
We will hence concentrate on WE for our
average over the whole phase space sample
discussions on the drift velocity v_D, while
is given by ! ! for our discussions on the diffusion we will
!v" = d3 x (4π)v 2 dv (n f0 )W focus on WG, which is our next task.
! " # !
4π
= v 3 dv d3 x (n f 1 )
3

58
 Drift Velocity
Mobility Matrix
We start from the equation for fE, which Recall your linear algebra course, then the
can be rewritten with WE as reciprocal of the matrix M is given by
! "  
v ∂ eE ν 2 + ω12 ω1 ω2 − νω3 ω1 ω3 + νω2
f0 [ ν̄m (v) − ω× ] W E = − f0 −1
3 ∂v m [M ] = ω2 ω1 + νω3 ν 2 + ω22 ω2 ω3 − νω1 
ω3 ω1 − νω2 ω3 ω2 + νω1 ν 2 + ω32
Notice that n(x;t) does not depend on v and % 2 &
hence can be cancelled out. ÷ν ν +ω 2

We now introduce a matrix [M]: with det [M]

[M ] := [ ν̄m (v) − ω× ] ω := ω =
2 2
ω12 + ω22 + ω32
  and
ν ω3 −ω2 (−e)B
ω :=
= −ω3 ν ω1  mc
ω2 −ω1 ν
WE can now be written as
where use has been made of a shorthand: ! " ! "
v ∂ −1 eE
f0 W E = − f0 [M ]
ν̄m (v) → ν 3 ∂v m
For notational convenience, we will use this which can be averaged over v to give
!
abbreviation in what follows. !W E "v (x; t) := (4π)v 2 dv f0 W E
Now all we need to do is a matrix inversion.
= [µ] E

59
 Drift Velocity
Mobility Matrix (continued)
The Mobility Matrix If there is a B-field, the mobility matrix
We introduced the local mobility matrix: will acquire nonzero off-diagonal elements
! " #
4πe ∂ −1 and hence the direction of the E-field and
[µ] := − dv v 3 f0 [M ]
3m ∂v the direction of the drift velocity will
which is in general a function of (x;t). differ (so-called Lorentz angle effects).
To get the position-averaged mobility
suitable for the centroid motion, we define Special Case [1] (B=0)
!
The matrix [M] becomes “nu” and hence
f0∗ (v; t) := d3 x (n f0 )
the [mu*] becomes a single number:
! ∞ " #
and the (global) mobility matrix: 4πe v 3
d
! " # µ∗ = − dv f0∗
4πe d −1
3m 0 ν dv
[µ∗ ] := − dv v 3 f0∗ [M ]
3m dv The drift direction should be anti-parallel
With this, we can write with the E-field. This suggests that the
!W " = !W E " = [µ∗ ] E integral should be negative, since (e < 0).
Assuming that f*0 has a single peak, and
Notice that the mobility matrix is the integral weights more on the higher
proportional to a unit matrix if B=0. side of the peak, it is indeed so.

60
 Drift Velocity
Mobility Matrix (continued)
Special Case [2] (B//E) Special Case [3] (v-dist=delta fn.)
This is the case of our interest. Assuming If the velocity distribution can be taken as
that E and B are in the 3-axis direction, a delta function:
then   1
0 f0 =
∗
δ(v − v̄)
ω = 0 4πv 2
ω Putting this into the def. of the mobility
matrix, we have
and the inverse of [M] becomes ! " #
 2  4πe d −1
ν −νω 0 [µ∗ ] = − dv v 3 f0∗ [M ]
−1 1 3m dv
[M ] = νω ν2 0  ! $
ν (ν + ω )
2 2 4πe −1 d % 3 ∗&
0 0 ν 2 + ω2 = − [M ] (v̄) dv v f0
3m dv '
d % 3& ∗
But the E-field has no 1- or 2- components, − v f0
! dv
there will be no 1- or 2-components in the 4πe −1 e −1
= [M ] (v̄) dv v 2 f0∗ = [M ] (v̄)
drift velocity, either. Moreover, the 3rd m m
component coincides with the B=0 case.
The mobility matrix is thus parameterized
There is hence no B-field effect on the
by just two parameters, the collision freq.
drift velocity in the E//B case.
at vbar and the cyclotron frequency.

61
 Mean Free Time
Mobility Matrix (continued)
Usual Simplistic Arguments We now need to evaluate the time average
Case (3) formula is usually obtained by of the collision force:
!
time-averaging the Newtonian equation of 1 T
!F coll "t = lim dt F coll
T →∞ T
motion. 0
N !
1 # ti
dv ! v " = lim "N dt F coll (t)
m = e E + × B + F coll N →∞
i=1 ∆Ti i=1 ti−1
dt c #N ! ti + δt
1
1 2
= lim "N dt F coll (t)
We define the time average of a variable A N →∞
i=1 ∆Ti /N
N δt
i=1 ti − 2
to be
N
! 1 1 # 1
1 T = lim m ∆v = !m∆v"
!A"t := lim dt A(t) τ N →∞ N i=1 τ
T →∞ T 0
Notice that there appear the mean free
Upon this time average, the L.H.S. of the time and the average momentum transfer.
Newtonian eq. vanishes, since we are The momentum transfer averaged over all
considering a bounded motion for which the angles is easy to get for isotropic collisions
velocity stays finite. We hence have !
dΩ
! " !m∆v"Ω = m∆v
!v"t 4π
0=e E+ × B + !F coll "t !
c d cos θ
= − m v (1 − cos θ) = −m v
2

62
 Mean Free Time
Mobility Matrix (continued)

We can think of the average that appears in The Drift Velocity Formula
1 1 !
N
1 This is a simple linear equation, and can be
!F coll "t = lim m ∆v = !m∆v" solved by matrix inversion as we did, and
τ N →∞ N i=1 τ
yields the formula you often see in the
being first taken over scattering angles for
text book
each group with nearly the same momentum ! "# $ % # & &
µE
and then over such groups. Then we have !v" = Ê + (ωτ ) Ê × B̂ + (ωτ )2
Ê · B̂ B̂
1 + (ωτ )2
1 This formula can hence be regarded as the
!F coll "t = − m !v"
τ limiting case of the delta function like v
Collecting things together, we arrive at the distribution or of a single velocity shell.
time-averaged Langevin equation: We can also rewrite the Langevin equation
! "
1 (−e)B eE in the following form
− × #v$ = eτ
τ mc m [1 − ω τ ×] #v$ = E
m
Notice that 1/tau=nu and
This implies µ(B = 0)
(−e)B
ω := eτ
mc [1 − ω τ ×] #v$ · #v̂$ = |#v$| = E · #v̂$
tell us that the content of the square m
bracket is the same [M] we met before. which is known as Tonk’s theorem.

63
 The Inverse of [M]
Another Expression

The drift velocity formula for a single shell diffusion matrix is given by
1! 2 "
! "# $ % # & &
µE −1
!v" =
1 + (ωτ )2
Ê + (ωτ ) Ê × B̂ + (ωτ ) 2
Ê · B̂ B̂ [D] = v [M ]
3
can be rewritten as Twice integrating by parts the diffusion
! "# $ e
!v" =
τ
1 − (ωτ )B̂ × +(ωτ ) B̂ B̂·
2
E eq. in the comoving frame, we have
1 + (ωτ )2 m !
d 2
d3 x! n(x! ; t) (x! · e)
dt
This implies that the inverse of [M] can be ! "
∂
#T $
1 2 −1
%" ∂ #
2
= d x 3 !
v [M ] n(x! ; t) (x! · e)
cast into the form ∂x ! 3 ∂x !

! "# $ ! $ %
τ 3 ! 2 −1
[M ] −1
= 1 − (ωτ )B̂ × +(ωτ ) B̂ B̂· 2 = d x v e [M ] e n(x! ; t)
2 T
3
1 + (ωτ )2
2$ 2 T −1
%
= v e [M ] e
3
From this we have immediately
T This implies
B̂ [M ]−1 B̂ = τ
2! 2 T "
and d 2
σ = 2De =
−1
v e [M ] e
τ dt xe 3
êT⊥ [M ]−1 ê⊥ =
1 + (ωτ )2 and hence
! "
where ê⊥ is a unit vector perpendicular to 1! 2 " 1 τ
DL = v τ and DT = v2
the B-field. We will see later that the 3 3 1 + (ωτ )2

64
 Diffusion
Diffusion Matrix
 
So far we have been discussing WE (or ν +2
ω12 ω1 ω2 − νω3 ω1 ω3 + νω2
−1
equivalently fE), the drift due to the [M ] = ω2 ω1 + νω3 ν 2 + ω22 ω2 ω3 − νω1 
ω3 ω1 − νω2 ω3 ω2 + νω1 ν 2 + ω32
external fields. % &
We now turn our attention to the vector ÷ ν ν 2 + ω2
eq. for WG (fG), which can be cast into the with det [M]
form: ω := ω =
2 2
ω12 + ω22 + ω32
! "
(n f0 ) [ ν̄m (v) − ω× ] W G = −
v2 ∂
(n f0 )
and
3 ∂x (−e)B
ω :=
mc
Notice that this time, since n(x;t) depends
on x, we cannot cancel out n. Nevertheless, The solution is then
2
! "
there appears the same matrix [M]: v −1 ∂
(n f0 ) W G = − [M ] (n f0 )
3 ∂x
[M ] := [ ν̄m (v) − ω× ] W E
  which can be averaged over v to give
ν ω3 −ω2 !
= −ω3 ν ω1  n !W G "v (x; t) := (4π)v 2 dv (n f0 ) W G
ω2 −ω1 ν ! " #
4π −1 ∂
=− v 2 dv v 2 [M ] (n f0 )
and hence with the same inverse matrix: 3 ∂x

65
 Diffusion
Diffusion Matrix (Continued)
Crucial step is to replace f0 on the R.H.S. with
by f0*: ! ω 2 := ω 2 = ω12 + ω22 + ω32
f0∗ (v; t) := d3 x (n f0 ) and
(−e)B
ω :=
so that we can take out f0* out of the mc
spatial derivative and get The approximation
∂ ∂ ∂
n !W G "v = − [D] n (n f0 ) ≈ f0∗ (n)
∂x ∂x ∂x
with the diffusion matrix [D] given by
! allowed us to define the diffusion matrix
4π −1 [D] that satisfies the usual definition
[D] = v 2 dv v 2 [M ] f0∗
3 ∂
The inverse of [M] is as before: n !W G "v = − [D] n
  ∂x
ν 2 + ω12 ω1 ω2 − νω3 ω1 ω3 + νω2
−1
[M ] = ω2 ω1 + νω3 ν 2 + ω22 ω2 ω3 − νω1  current density
ω3 ω1 − νω2 ω3 ω2 + νω1 ν 2 + ω32
% & (w/o common drift ) grad (density)
÷ ν ν 2 + ω2
There is some subtlety in this approx. but
det [M] we will not get into it now.

66
 Diffusion
Diffusion Matrix (continued)
Special Case [1] (B=0) Special Case [2] (B//E)
The matrix [M] becomes “nu” and hence This is the case of our interest. Assuming
the [D] becomes a single number: that E and B are in the 3-axis direction,
then  
! 0
4π v4 ∗
D= dv f0 (v) ω = 0
3 ν ω
where the collision frequency is given by
and the inverse of [M] becomes
!  2 
ν= nk v σm,k (v) ν −νω 0
−1 1 νω
[M ] = ν2 0 
k
ν (ν + ω )
2 2
0 0 ν 2 + ω2
We can hence rewrite the diffusion
Then we have
constant as !
! 1 (4πv 2 )f0∗ (v) 2
1 (4πv 2 )f0∗ (v) 2 DL = dv v = D33
D= dv " v 3 ν
3 n vσ (v) !
k k m,k
1 (4πv 2 )f0∗ (v) ν 2
DT = dv v = D11,22
The diffusion is hence isotropic (as long as 3 ν +ω
2 2
!
the approximation is valid) and inversely 1 (4πv 2 )f0∗ (v) ω 2
D12 = −D21 = − dv v
proportional to gas density and X-section. 3 ν +ω
2 2
All the other components are zero.

67
 Diffusion
Diffusion Matrix (continued)

Notice that the longitudinal diffusion const. Special Case [3] (v-dist=delta fn.)
!
1 (4πv 2 )f0∗ (v) 2 If the velocity distribution can be taken as
DL = dv v = D33
3 ν a delta function:
1
is the same as with the B=0 case. f0∗ = δ(v − v̄)
4πv 2
On the other hand, the transverse one
! Putting this into the def. of the diffusion
1 (4πv 2 )f0∗ (v) ν 2 !
DT = dv v = D11,22 matrix: 4π −1
3 ν 2 + ω2 [D] = v 2 dv v 2 [M ] f0∗
3
is reduced by a factor
with the inverse of [M] given by
ν2 1 1  
= with τ = ν 2 + ω12 ω1 ω2 − νω3 ω1 ω3 + νω2
ν 2 + ω2 1 + (ωτ )2 ν −1
[M ] = ω2 ω1 + νω3 ν 2 + ω22 ω2 ω3 − νω1 
in the integrand. Where the tau, being the ω3 ω1 − νω2 ω3 ω2 + νω1 ν 2 + ω32
% &
inverse of the collision frequency, is the ÷ ν ν 2 + ω2
mean free time between collisions. we have 1 2 −1
! [D] = v̄ [M ] (v̄)
1 (4πv 2 )f0∗ (v) ω 2 3
D12 = −D21 = − dv v
3 ν 2 + ω2 If B=0, this implies a naive expectation
corresponds to rotation about the field axis 1 2 1 (v̄τ )2
but it is not of our interest.
[D] = v̄ τ =
3 3 τ

68
 Random Walk
Diffusion Matrix (continued)
Usual Simplistic Arguments The solution to this is a helix
 
Case (3) formula is usually obtained by the sin θ (cos(ωt + φ) − cos φ)
v̄ 
random walk theory with a fixed mean free x(t) = sin θ (sin(ωt + φ) − sin φ)  + x0
ω
time: ωt cos θ
1
τ̄ = If we have N collisions over the time t, the
ν(v̄)
probability of finding the electron at x is
The probability for the electron to fly over
!N "# ∞ # $
a time “t” and then get scattered by the 1 dΩi
P (x) = dti e−ti /τ̄
molecule by an angle “Omega” is given by i=1 % 0
τ̄ 4π
! " N
'
1 t dΩ &
P (t, Ω) = exp − dt 3
×δ x− ∆x(θi , φi , ti )
τ̄ τ̄ 4π
i=1
if the scattering is isotropic. For the sake with
of simplicity, let us further assume that ∆x = x − x0
there is no E and B along the 3rd axis, then The average position is apparently zero
Newton’s eq. reads because of the angular integrals which are
!
d isotropic.
v =ω×v !x" = d3 x P (x)x = 0
dt

69
 Random Walk
Diffusion Matrix (continued)

Now the mean square transverse distance Similarly the mean square longitudinal
is given by !
distance is given by
" # !
2
σ⊥ = σx21 + σx22 = d3 x P (x) x21 + x22 2
σL = σx23 = d3 x P (x) x23
N '! ∞
$ v̄ %2 & ! (
1 −ti /τ̄ dΩi N #! ! $%
N
= dti e " ∞
1 dΩ
ω i=1 0 τ̄ 4π = v̄ 2
dti e −ti /τ̄ i
t2i cos2 θi
N i=1! 0 τ̄ 4π i=1
) ∞ !
× 2 sin2 θi (1 − cos(ωti )) 1 ! dΩ #2
= N v̄ 2
dt# e−t /τ̄ t cos2 θ
! i=1 τ̄ 4π
$ v̄ %2 ! ∞ 1 dΩ ! 0∞
=N dt$ e−t /τ̄
!
2 sin2 θ (1 − cos(ωt$ )) t 1 ! 1
ω τ̄ 4π = v̄ 2 dt# e−t /τ̄ t#2
! 0 τ̄ 0 τ̄ 3
t $ v̄ %2 ∞ $ 1 −t! /τ̄ 4 2 (v̄τ̄ )2
= dt e (1 − cos(ωt$ )) =t = 2 DL t
τ̄ ω 0 τ̄ 3 3 τ̄
4 (v̄τ̄ )2
=t = 2 σ 2
= 2 · 2 DT t yielding
3 τ̄ (1 + (ωτ̄ )2 ) T 1 (v̄τ̄ )2
DL =
which leads us to the expression 3 τ̄
1 (v̄τ̄ )2 We reencounter the familiar result
DT =
3 τ̄ (1 + (ωτ̄ )2 ) DT 1
=
DL 1 + (ωτ̄ )2

70
 Diffusion
Diffusion Matrix (continued)
Case [2] (B//E) Revisited
Now go back to B//E, and rewrite the (b) ωτ ! 1 !
1
DT (B) ≈ dv (4πv 2 )f0∗ (v) τ (v)
diffusion constants 3 " #
! −2
× (ωτ (v)) − (ωτ (v))
−4
v2
1 (4πv 2 )f0∗ (v) τ (v)
DT = dv v2 $% 2 & % 2 &'
3 1 + (ωτ (v))
2
1 v v
= −
3
It is interesting to consider the following
2
ω τ ω4 τ 3
( 2)
1
two extreme cases:
τv
≈ ( ) ( ) ( 2 ) ( 2 )2
3 $τ v 2 % v 3 / v
2 2 2
+ ω 2 $τ v 2 % v / v
(a) ωτ ! 1
τ τ τ τ
!
1 DT (0)
DT (B) ≈ dv (4πv 2 )f0∗ (v) τ (v) = ( v 2 ) ( v 2 )2 ( 2 ) ( 2 )2
3 " # $τ v 2 % τ 3 / τ + ω 2 $τ v 2 % vτ / vτ
2
× 1 − (ωτ (v)) v 2
1 $% 2 & % &' This implies
= τ v − ω2 τ 3 v2
3 DT (0)
1
% 2& ≈ C + (ωτ2 )2
≈
τv DT (B)
3 1 + ω 2 $τ 3 v 2 % / $τ v 2 %
DT (0) with ! 2 " # v2 $ ! " # 2$
= τv τ3 τv 2 v
1 + ω 2 $τ 3 v 2 % / $τ v 2 %
and
τ
C= ! v 2 "2 τ22 = ! v 2 "2
This implies
! " τ τ
DT (0) τ v 3 2
≈ 1 + (ωτ1 )2 with τ1 =
2
valid for a high B-field.
DT (B) !τ v 2 "
valid for a low B-field.

71
 Amendolia et al. 1986

τ1 = (0.41 ± 0.02) × 10−10 [s]

τ2 = (0.266 ± 0.006) × 10−10 [s]

C = 2.8 ± 0.2

P9 (1atm, E=115 [V/cm])

72
 Scalar Equation
We need to solve the scalar equation, too
What We Have Done So Far c) We have also defined the diffusion matrix
!
a) We have shown that 4π −1
[D] = v 2 dv v 2 [M ] f0∗
!W G " = 0 and, hence !v" = !W E " 3
which means diffusion does not with which the convection current due to
contribute to the drift velocity of the diffusion is given by
centroid, as naively expected. ∂
n !W G "v = − [D] n
∂x
b) We have defined the mobility matrix for
These results came solely from the vector
the centroid
! " # equation, and f0* remains as unknown.
4πe d
Remaining Questions
−1
[µ∗ ] := − dv v 3 f0∗ [M ]
3m dv
  a) How should we relate [D] to the electron
ν 2 + ω12 ω1 ω2 − νω3 ω1 ω3 + νω2
−1
[M ] = ω2 ω1 + νω3 ν 2 + ω22 ω2 ω3 − νω1  cloud size? In other words, we need to
ω3 ω1 − νω2 ω3 ω2 + νω1 ν 2 + ω32 know the spatial distribution, n(x;t).
% & b) How can we determine f0*?
÷ ν ν 2 + ω2
with which, we can write
In order to answer these questions, we now
!W " = !W E " = [µ∗ ] E
need to look at the scalar equation.

73
 Scalar Equation
Derivation of Diffusion Equation
The Diffusion Equation we v-integrate the both sides to get
! " #
We hence restart from the scalar equation ∂ ∂ vf
! " n+ · n (4πv 2 ) dv f0 1 = 0
∂ v ∂ 1 ∂ 4π 2 eE ∂t ∂x 3f0
(nf0 ) + ·(nf1 ) + v ·nf 1
∂t 3 ∂x 4πv 2 ∂v 3 m where the R.H.S. is a surface integral.
=n
#
nk
1 ∂
σ̄m,k (v; [f0 ])
Recall also the shell averaged velocity
4πv 2 ∂v formula
k
By defining vf 1
!v"Ωv =
4π 2 eE 3f0
σE := v · nf 1
3 m
and then the quantity in the parentheses is the
!
σcoll := nk σ̄m,k current density at (x; t)
k
!
vf
we can rewrite it in the following form n (4πv 2 ) dv f0 1 = n !W "v
3f0
∂ v ∂ 1 ∂
(nf0 ) + ·(nf 1 ) = − (σE − σcoll )
∂t 3 ∂x 4πv 2 ∂v The above equation now becomes
Recalling ! ! ∂ ∂
n+ (n !W "v ) = 0
d3 v f0 = (4πv 2 )dv f0 = 1 ∂t ∂x
! which is none other than the usual equation
f0∗ (v; t) := d3 x (n f0 ) of continuity.

74
 Scalar Equation
Derivation of Diffusion Equation
In to this continuity equation: ! "2
∂ ∂ ∂
∂ ∂ n + !v" · n−D n=0
n+ (n !W "v ) = 0 ∂t ∂x ∂x
∂t ∂x
we can now put In the co-moving frame of the centroid
n !W "v = n !W E "v + n !W G "v ( x! = x − "v# t ), this becomes
! "2
recalling ∂ ∂
n−D n=0
!W E "v = [µ] E ≈ [µ ] E = !v"
∗ ∂t ∂x!
and ∂ The solution to this equation with the point
n !W G "v = − [D] n source initial condition is given by
∂x
! #3 $ %
We then obtain n= "
1
exp −
x!2
! "T ! " 2π(2Dt) 2(2Dt)
∂ ∂ ∂ ∂
n + !v" · n− [D] n=0
∂t ∂x ∂x ∂x This implies that the electron cloud will
have a Gaussian spread given by
which is none other than the diffusion eq.
as you transform this into a more familiar σx2 = 2Dt
form if [D] is a constant D times a unit after created as a point-like cluster.
matrix
OK, now the remaining task is f0*!

75
 Scalar Equation
Equation for f0*
Velocity Distribution Function Ignoring the time derivative assuming that
We again start from the scalar equation the electron’s velocity distribution reaches
∂ v ∂ 1 ∂ a steady state in a short time, this reads
(nf0 ) + ·(nf 1 ) = − (σE − σcoll )
∂t 3 ∂x 4πv 2 ∂v 1 ∂
0=− (σ ∗
− σ ∗
)
This time we integrate out x, since we are 4πv 2 ∂v E coll

now interested in the velocity distribution The equation expresses the balance
! between the external force and the
f0∗ (v; t) := d3 x (n f0 ) collision force. The concrete form of the
Upon this integration the 2nd term of the collision term depends on the nature of the
L.H.S. vanishes since it becomes a surface molecules in the gas in question and hence
integral where the electron is absent. the concrete form of the equation also
Noting that the R.H.S. is a function of f0 depends on it. When only elastic collision is
and this spatial integration replaces f0 by there, it is known (c.f. Huxley & Crompton)
f0*, we have that the equation becomes
! 2 " 2 #$ d ∗ 3 m v ∗
∂ ∗ 1 ∂ u + V f + f =0
f0 = − (σ ∗
− σ ∗
) dv 0 M 0
∂t 4πv 2 ∂v E coll
with eE eE
u := = τ
mν m

76
 Scalar Equation
Solution for f0* (elastic only case)
we end up with
Solution for f0* ! 2
"
m v /2
The equation for f0* for a monatomic gas f0∗ (v) = A exp −
kB T
and for elastic collisions only
! 2 " 2 #$ d ∗ 3 m v ∗ which is none other than the Maxwellian
u + V f + f =0 distribution as expected.
dv 0 M 0
with eE eE
u :=
mν
=
m
τ Special Case II (nu/v=const.)
has the solution ! " % When the collision frequency divided by v
3m v
v dv
f0∗ (v) = A exp − # 2$ or equivalently the cross section can be
M 0 u + V
2
regarded as constant within the range
Once the momentum transfer X-section is where f0* is significant, we have instead
! " # $
known, nu(v) is known, and hence we can v 4
f0 (v) = A exp −
∗
calculate f0*. That’s the recipe. α
with ! "2
Special Case I (E=0) α4 =
4 M e E/n M

In this case u=0, and since 3 m m σm

! "
1 3 This is called a Druyvesteyn distribution
M V 2 = kB T
2 2 and has a sharper peak than the Maxwellian.

77
 Cross Section Shape
How sigma_m behaves?
Order of Magnitude Estimate The orbit radius can then be approximated
The most popular chamber gas is Ar, so as exactly as with a hydrogen atom for n=3.
!
let’s try an order of magnitude estimation r(n) ! n
mc αQED
of the electron-Ar cross section.
! (0.5 × 10−8 [cm]) × 3
Ar has an atomic number A=18 with the
= 1.5 × 10−8 [cm]
first 3 shells filled up. It is a perfectly
σAr ! π r(3)2
symmetric molecule and hence the remnant
! π (1.5 × 10−8 [cm])2
electric field dies away very quickly. The = 7.1 × 10−16 [cm2 ]
scattering cross section is therefore But the life is not so simple, This gives a
largely determined by the size of the kind of upper limit, and the real X-section
outermost orbit. The electron in the can be much smaller because of the so
outermost orbit experiences an attractive called Ramsauer effect, a QM effect.
force from the nucleus largely shielded by !
!"
!2
!
the other electrons. ! 0 +A 1 !
σAr ∝ ! σAr !
Let’s assume that because of this shielding,
! " − "r + i Γ2r !

the electron only feels the net charge of 1 The interference makes a dip below the
unit of (-e). resonance peak!

78
 Ar Cross Section
That used in Magboltz

0 RAr = 1.8 × 10 −8
[cm] (exp.)
σAr ! π r(3)2
! π (1.5 × 10−8 [cm])2
= 7.1 × 10−16 [cm2 ]

1 ! c αQED "2
!dip ! m ! 1.5 [eV]
2 n

79
 Characteristic Energy
Mobility and Momentum Transfer Cross Section
For simplicity, we will assume B=0 here. We can further rewrite!the formula
"
as
In this case the mobility constant can be 2e e d
µ= !τ " + (vτ )
cast into the form 3m 3m dv

4πe ∞
! "
d
# Here we used a shorthand, mu*=mu, since
µ∗
= − dv v 3 τ (v) f0∗ (v) there is no fear for confusion.
3m 0 dv
! ∞ $" #
4πe d 3 Recalling that (v tau) is inverse of the gas
= − dv v τ (v)f0∗ (v)
3m 0 dv ' density times the effective momentum
d % 3 & ∗
− v τ (v) f0 (v) transfer X-section, the 2nd term vanishes
dv
! if the X-section change is negligible over
4πe ∞ d % 3 &
=
3m 0
dv
dv
v τ (v) f0∗ (v) the velocity distribution given by f0*. This
4πe
! ∞ $ "
d
#' is true near the X-section minimum. On the
= dv v 2 3τ (v) + v τ (v) f0∗ (v) falling edge it is +ve and on the rising edge
3m 0 dv
( )
e e d it is -ve.
= "τ # + v τ
m 3m dv On the other hand, (roughly speaking) the
where use has been made of the fact that 1st term attains its maximum near the X-
f0* vanishes at the boundaries. If tau is section minimum. Net effect is that the
constant, we recover our simple-minded mobility attains its maximum near the
formula. The mu* becomes max. with tau. Ramsauer dip.

80
 Characteristic Energy
De/mu as an estimate of the average electron energy
Characteristic Energy Nernst-Townsend Formula
The diffusion constant at B=0 is given by In the thermal limit, the characteristic
! energy is given by
1 ∞ " # 1$ 2 %
D= dv (4πv 2 ) f0∗ (v) v 2 τ (v) = v τ
3 0 3 De
= kB T = 0.025 [eV] (1 atm, 20◦ C)
which can be cast into the form µ
! "
2 1 and called the Nernst-Townsend formula.
D= mv 2 τ
3m 2 For a cool gas such as CO2, this formula
Recalling ! " holds up to about 1kV/cm, while for Ar,
e e d
µ= !τ " + v τ this breaks down at an E-field value as low
m 3m dv
as 1V/cm. The electrons in a pure Ar gas
and ignoring the variation of tau over the can be easily heated up to 1eV or higher.
velocity range determined by f0*, we have
! "
eD 2 1 In the thermal limit, Cd (the diffusion
!k := ≈ mv 2
µ 3 2 coefficient: the rms size of a cluster after
The quantity (eD/mu) is termed the a unit length of drift) is given by
! "
characteristic energy of the electron for 2D 2kB T
obvious reason. Cd := ≈
µE E

81
 Cd and Vd for CO2
Trying to understand Magboltz results (B//E)
It is interesting to test the expectation
with the simulation by Magboltz.
We can see that the diffusion coefficient
Cd behaves as 1/sqrt(E) up to 1kV/cm with
almost no dependence on the B-field. ! "
2D 2kB T
The almost no dependence of Cd on B can Cd := ≈
be understood as the consequence of the µE E
smallness of ωτ .

It is also worth noting that the mobility is

approximately constant and hence the drift
vdrift ≈ const. × E
velocity is proportional to the E-field in the
same region: E < 1kV/cm.

The fact that the electron velocity stays

thermal up to rather high E-fields is
attributable to many excitation modes of
CO2 to cool the electrons.

82
 Cd and Vd for P5
Trying to understand Magboltz results (B//E)
In the case of P5, a more suitable gas for a
TPC, the behavior is very different.
First notice the strong dependence of the DT = DL
diffusion constant on the B-field, which
suggests
ωτ ! 1
and hence DT (0)
≈ C + (ωτ2 )2
DT (B)
Assuming that tau is nearly constant in the
relevant velocity range, we have
! 2 " # v2 $
τv τ
τ2 =
2
! v 2 "2 ≈ τ̄
τ
DT minimum
The transverse diffusion is thus expected
to attain its minimum at the maximum tau,
or at around the Ramsauer dip. Notice also
that the drift velocity attains its maximum The DL being different from the DT is
near there as expected. unexpected and needs explanation.

83
 Diffusion Revisited
Electric Anisotropy
The fact that the diffusion in the E-field over the electron speed, we can translate
direction differs from those in the other the scalar equation into the following form
!
directions was first noticed by Wagner, ∂ ∂
(n !""v ) + (!"W "v ) = n !W "v · eE − dv mv σcoll
Davis, and Hurst (1967). This is called the ∂t ∂x
electric anisotropy. Since our previous where 1
! := mv 2
result indicated that the DL is equal to the 2
DT at B=0, some approximation we made to The equation expresses the conservation
some E-dependent terms must have been of energy. Recall that if collisions are all
inadequate to explain this phenomenon. We elastic and the speed of molecules can be
then need to go back to the scalar equation neglected, we have
! m
∂ v ∂ 1 ∂ σcoll ≈ 4πv 2 v (nf0 ) νm,k
(nf0 ) + ·(nf 1 ) = − (σE − σcoll ) Mk
∂t 3 ∂x 4πv 2 ∂v where k

with 4π 2 eE νm,k (v) := nk v σm,k (v)

σE := v · nf 1
3 m When a molecule is at rest, energy loss per
and
! collision is given by
σcoll := nk σ̄m,k Mk m m
∆! = λk ! = 2 ! ! 2 !
k (Mk + m)2 Mk
By multiplying both sides by the kinetic where lambda is called the fractional
energy of the electron and integrating energy loss.
84
 Diffusion Revisited
Electric Anisotropy
The collision term then becomes Setting our coordinate axes in such a way
! !
dv mv σcoll ≈ n dv(4πv )f0 (λν) % = n % (λν)
2
"
#
that the E-field points to the 3rd axis
v
direction and recalling that
with !
(λν) := λk νm,k !W E "v · eE = eµ E 2 ! "
k 1 ∂n
Putting these together, we get !W G "v · eE = −eE D
n ∂x3
∂ ∂ ! "
(n !""v ) + (!"W "v ) = n !W "v · eE − n " (λν) we can rewrite the energy balance eq. as
∂t ∂x v ! "
1 ∂n
Since we are dealing with an electron cloud eµ E 2 − eE D ≈ #"$v (λν) (#"$v )
n ∂x3
in a quasi steady state, the L.H.S. may be Denoting !¯ = !!"v and expanding this eq.
ignored in considering a change in the scale around the energy corresponding to the
of one mean free time. !The eq.
" then reads average energy at the peak of the electron
n !W "v · eE = n ! (λν)
v density where the derivative is zero, we
indicating local balance between the energy have # $
gain from the E-field and the loss due to e 1 2¯"0 1 1 ∂n
e ! ∂ν " E 2 − eE
collisions. Since the current density m ν0 + ∂#̄ 0 ∆¯
" 3m ν0 n ∂x3
% # $ &
n !W "v = n !W E "v + n !W G "v ∂ν
≈ λ (¯"0 + ∆¯
") ν̄0 + ∆¯
"
∂¯
" 0
includes the contribution from diffusion, so
where use has been made of
does the L.H.S. of the above equation.

85
 Diffusion Revisited
Electric Anisotropy
e e 1! 2 " 2¯
"0
µ= !τ " ≈ and D=
3
v τ ≈
3mν0
Depending on the sign of “gamma”, the
m mν
variation induces a bunching (+ve) or
The energy balance eq. can be solved for
debunching (-ve) effect and hence makes
the energy shift as
2 !¯2
'
1 ∂n
(
the longitudinal diffusion different from
∆¯
!=− ! " #$ % 0 " #" # &
3eE 1 + ν!¯00 ! 0+
∂ν
∂¯
!¯0
(λν)0
∂(λν)
∂¯
!
0
n ∂x3 the transverse one.
At the leading edge, where the density has Putting these into
∂
a -ve slope, the energy shift is +ve, and at n !W "v = n µE − D
∂x
n

the trailing edge, the energy shift is -ve. and taking the 3rd component, we have
The energy is hence higher at the leading n !W3 "v = n µE − D
∂
n
∂x3
edge and lower at the trailing edge than at e
!
γ
"
∂
= n E−D 1− n
the center. This energy shift induces the mν0 1 + γ + γ ! ∂x3
variation of mobility along the E-field: which implies ! "
! " # $ γ
e e 1 ∂ν DL = D 1 −
µ= ≈ 1− ∆#̄ 1 + γ + γ!
mν mν0 ν0 ∂#̄ 0
Denoting ! " ! " The formula shows that DL=DT(B=0) where
"¯0 ∂ν "¯0 ∂(λν)
γ= and γ = “gamma”=0, the collision freq. attains its
!
ν0 ∂¯
" 0 (λν)0 ∂¯"
we have minimum (=tau maximum) meaning near the
0

! " #$
e 2 γ #̄0 1 ∂n Ramsauer dip as we have seen for P5.
µ≈ 1+
mν0 3eE (1 + γ + γ ! ) n ∂x3

86
 Gas Amplification
in a Strong E-Field
This part will be very brief, since my understanding of
this subject is very much limited!

87
 Gas Amplification
Average Gas Gain
Twonsend Coefficient derivatives of the electron state density
The probability per unit length for a seed function on the R.H.S. of Boltzmann eq. can
electron in a strong E-field producing an be ignored.
additional ionization electron is called the Then the Townsend coefficient, having the
first Townsend coefficient ( α ). We can dimension of inverse length, must scale
write the average increase of electrons with the mean free path inverse and hence
( dN ) over a path ( ds) to be should be proportional to the gas density:
! "
E B ρ
dN = N α ds α = α0 , ·
ρ ρ ρ0
The Townsend coefficient is determined by unless E-field variation is so quick that the
the cross sections for ionizing collisions or f(v;x) changes significantly over a few mean
excitation collisions leading to secondary free paths.
ionizations through Penning effect or Jesse Taking this condition for granted we can
effect. These cross sections are a function write the average gas gain as a line integral:
!" #
of the electron’s speed or equivalently its N B
Ḡ := = exp ds α(E(s))
energy, which is in turn a function of two N0 A

scaling variables: “E/(gas density)” and which in general depends on the possible
“B/(gas density)”, as far as the t-and x- path along which the avalanche develops.

88
The formula allows one to calculate the with the E-field. If the E-field is constant,
average gas gain once the 1st Townsend the gas gain increases with the gap. The E-
coefficient is given as a function of the E- field, however, decreases when the gap is
field. Strictly speaking, the scaling holds increased. This suggests that the gas gain
only when we change both the E- and B- must attain a maximum for an appropriate
fields simultaneously. As far as I know gap value, around which the gas gain is
there is no analytic treatment of general E stable against gap variation. This is the
and B configurations. When the E- and B- operation principle of the micromegas.
fields are parallel, however, the longitudinal Paul Colas
motion will not be affected by the B-field
and hence we can ignore the B-field effect
on the Townsend coefficient (recall that
the electron energy is characterized by
eD/mu which is unaffected).
In the case of uniform E//B, we have

Ḡ(∆) = exp [α(V /∆) ∆]

where ∆ is the amplification gap and V is

the high voltage across it.
This should be a good approximation for a
GEM or micromegas in particular. Notice
that the Townsend coefficient increases

89
 Gas Amplification
Statistics of Avalanche Fluctuation
Alkhazov’s Theory (1970) Graphically we can represent this as in the
The avalanche formation involves various following figure:
mechanisms: impact ionization, Penning and
º
Jesse processes. We consider here the P
®
®
» N'
®
case where the impact ionization dominates. º N 1 ®¼

° dl ¤
® x
N 
® pi
P »
We further assume a uniform E-field in the ®
®¼
0
N '1
º
®
®
P » N N'
amplification region. A B-field, if there is x
®
®¼

any, should be parallel to the E-field. Now

l x l
let the probability of getting N electrons
at the point x from the beginning of the
We can define the avalanche fluctuation
amplification region be P(N; x), then P(N; x)
function as
must satisfy the following self-consistency
p(z, x) := N̄ (x) P (N̄ (x)z; x)
equation:
! x N
" −1 and its n-th moment as
P (N ; x) = dl pi (l) P (N ; x − l)P (N − N ; x − l)
" " ! ∞
0 N ! =1 Mn := dz z n p(z, x)
where pi(l) is the probability of 1st ionizing 0
∞ # $n
" 1 N
collision taking place at the distance l from = N̄ (x) P (N ; x)
N̄ (x) N̄ (x)
the origin of the seed electron. N =0

90
Because of the central limit theorem, we with
!
expect that the avalanche fluctuation fn. ∞
J(n) := dl pi (l) e−n α l
and hence its moments also are determined 0

by the early stage of the avalanche growth,

determined by the probability for the 1st
which implies that p(z, x) should become x-
ionizing collision.
independent
On the other hand, we have
p(z, x) → p(z)
M0 = M1 = 1
at large x where
by definition. M1=1 determines the 1st
N̄ (x) → eα (x−∆0 )
Townsend coefficient:
Keeping these in mind, we can derive from ! ∞
2J(1) = 2 dl pi (l) e−α l = 1
! x N
" −1 0
P (N ; x) = dl pi (l) P (N ; x − l)P (N − N ; x − l)
" "
0 N ! =1 Once pi(l) is given, we can hence calculate
an equation for Mn: Mn recursively.
"
The self-consistency equation also induces
n
! ∞
n!
Mn = Mk Mn−k dl pi (l) e−n α l
k=0
k!(n − k)! 0 an equation for p(z):
! ! z! " #
This leads us to a recurrence formula:
∞
1 1 z"
p(z) = dz " dz "" p(z "" ) p(z " − z "" ) pi ln
αz z 0 α z
n−1
! n! Mk Mn−k J(n)
Mn = which can be used to get an approximate
k!(n − k)! 1 − 2J(n)
k=1
solution by iterative substitutions.

91
The self-consistency equation for p(z)
! ∞ ! z! " # Snyder’s Model
1 1 z "
p(z) = dz " dz "" p(z "" ) p(z " − z "" ) pi ln If the ionization probability is constant as
αz z 0 α z
implies that the large l behavior of pi(l) given by the 1st Townsend coefficient:
controls the behavior of p(z) near z=0. pi (l) = α e−α l
Assuming the exponential shape for the we have an exponential distribution
large l limit:
p(z) = e−z
pi (l) → C e−a l as l → ∞
as the exact solution to the above equation.
where C is a constant, we have
! ! This can be easily checked by substituting
∞ z!
p(z) ! z
a
α −1 dz # dz ## p(z ## )p(z # − z ## )
C #a/α
z this in the self-consistency equation.
α
0 0
In this case we have
near z=0. Denoting
a Mn = n!
θ := − 1
α We thus have
we hence obtain M2 = 2
p(z) ! C ! z θ
in particular.
where C’ is a constant. In the case of Polya We will see the significance of this number
distribution, we have later when we discuss the effective number
of seed electrons (Neff). Experimentally
1
θ = θpol := 2
−1 we know that M2 is smaller than 2 for GEM
σ
and Mircomegas detectors.

92
 Derivations of Recurrence Formulae
! ∞ " #n $∞ N $ −1 " # #n
N̄ (x − l) N + (N − N # )
Mn = dl pi (l) P (N # ; x − l) P (N − N # ; x − l)
0 N̄ (x) N =1 N ! =1
N̄ (x − l)
! ∞ " #n $∞ N $ −1 $ n " #k " #n−k
N̄ (x − l) n! N# N − N#
= dl pi (l)
0 N̄ (x) N =1 N ! =1 k=0
k!(n − k)! N̄ (x − l) N̄ (x − l)
× P (N # ; x − l) P (N − N # ; x − l)
! ∞ n
$ n!
= dl pi (l) e−n α l Mk Mn−k
0 k!(n − k)!
k=0

! ∞ ! zeα l
p(z) = dl pi (l) dz "" eα l p(z "" ) p(zeα l − z "" )
0 0
! ∞ ! ∞ ! z!
= dl pi (l) eα l dz " δ(z " − zeα l ) dz "" p(z "" ) p(z " − z "" )
0 z 0
! ∞ ! z! ! ∞
= dz " dz "" p(z "" ) p(z " − z "" ) dl pi (l)δ(z " − zeα l ) eα l
z 0 0
! ∞ ! z! " #
1 1 z "
= dz " dz "" p(z "" ) p(z " − z "" ) pi ln
αz z 0 α z

93
Legler’s Model It is hence important to have a high E-field
in the early stage of the avalanche growth
Legler assumed that any ionizing collision
in order to suppress gain fluctuation.
may take place only after the seed electron
flying over a minimum distance: From
n−1
! n! Mk Mn−k J(n)
x0 := U0 /E Mn =
k!(n − k)! 1 − 2J(n)
so as to gain enough energy for ionization with
k=1
!
from the E-field. Legler further assumed ∞
J(n) := dl pi (l) e−n α l
the probability of ionizing collision being 0

constant after the seed electron having M0 = M1 = 1

reached the threshold. The probability of we have
2 J(2)
the 1st ionizing collision is then given by M2 =
1 − 2 J(2)
pi (l) = ai e−ai (l−x0 ) θ(l − x0 ) which leads us to
(2 − eα x0 )2
As mentioned before, 2J(1) = 1 gives σ = M2 − 1 =
2
2 − (2 − eα x0 )2
α
ai = −αx0 (0 ≤ α x0 ≤ ln 2) Denoting
2e −1 κ := (2 − eα x0 )2
Notice that in the low E-field limit, where we then obtain
αx0 → 0 as E/ρ → 0 κ
σ 2 = M2 − 1 = (0 ≤ κ ≤ 1)
and hence 2−κ
ai → α as E/ρ → 0 √ √
2(1 − κ) κ
θ= √ = √ θpol ≤ θpol
converging to Snyder’s model. κ 1+ κ

94
The theta parameter controls the behavior and define α UI
near z=0. The inequality χ :=
we have E
√ √
2(1 − κ) κ (2 − eχ )2
θ= √ = √ θpol ≤ θpol σ = M2 − 1 =
2
κ 1+ κ 2 − (2 − eχ )2

states that the turn over near z=0 is less Alkhazov 1970

prominent than that expected from the

variance assuming a Polya distribution
1
θpol := −1
σ2
or for the same theta, the variance is
smaller than that expected for the Polya.
Legrer’s model thus suggests a probability
distribution for the gas gain fluctuation,
p(z), being non-Polya. Nevertheless, we can
calculate the variance by
(2 − eα x0 )2
σ = M2 − 1 =
2
2 − (2 − eα x0 )2 The variance depends on the E-field. The
with data suggest
x0 := U0 /E
If we set σ 2 = M2 − 1 >
∼ 0.2
U0 = UI : ionization pot. for Ar.

95
Extension to a nonuniform E field Here we have assumed that N1 2nd stage
Consider first the avalanche development in avalanches develop independently.
a uniform E field. Dividing the amplification The average of N is then given by
region (0,x) into two parts (0,l), (l,x). N̄12 =
!
P (N12 ; x) N12
N12  
º
! ! N1
$
®
1
P
®
»
®
N 2,1 = N1 N2,j P (N1 ; l)  P (N2,j ; x − l)
®¼ N1 N2,1 ,··· ,N2,N1 j=1
º
®
2
P
®
» N 2,2 = N̄1 N̄2
®

which leads us to a functional equation

º ®¼
® ¥ N1 ´
N ¤ ¤ ¤N
®
P » D ¦¦ N 2, j µ
µ
®
N 2,1 ,L,N 2,N1 § ¶
P
N̄ (x) = N̄ (l) N̄ (x − l)
®¼ N1 j1

x
º
Noting that Nbar(0)=1, we have from this
®
N1 ®
P » N 2,N1
® dN̄ N̄ (x) − N̄ (x − l)
®¼
(x) = lim
dx l→0 l
l x l N̄ (l) − N̄ (0)
= lim N̄ (x − l)
l→0
! l
The self-consistency equation for this dN̄ !!
= N̄ (x)
division reads dx ! x=0
 
We find again the familiar equation
! ! N1
!
P (N ; x) = δ N − N2,j 
!
N1 N2,1 ,··· ,N2,N1 j=1 dN̄ dN̄ !!
  = α N̄ with α :=
N1
& dx dx !x=0
× P (N1 ; l)  P (N2,j ; x − l)
j=1
where α is the 1st Townsend coefficient.

96
This eq. allows us to extend our uniform E- which leads us to
! " %
field result to a nonuniform case (N12 )2 = P (N1 ; l) N1 (N2 )2
# $
+ (N1 ) − N1 (N̄2 )
2 2

!" x # N1
& '
Ḡ(x) := N̄ (x) = exp dl α(l) = N̄1 (N2 )2 − (N̄2 ) 2
+ (N1 )2 (N̄2 )2
0
Denoting
This is none other than the average gas N 2 − N̄ 2 := N̄ 2 f (N̄ )
gain formula we have derived before. we arrive at
f (N̄1 N̄2 ) = f (N̄1 ) + (N̄1 )−1 f (N̄2 )
or
Let us now consider the variance of the f (N̄ (x)) = f (N̄ (l)) + (N̄ (l))−1 f (N̄ (x − l))
avalanche fluctuations:
#
If the gain of the 1st stage is large, the
! "2 ! "2
(N12 )2 − N̄12 := P (N12 ; x) (N12 )2 − N̄12 fluctuation in the 2nd stage is negligible,
being consistent with naive expectation.
N1 2

Recalling the self-consistency equation

 
Differentiating both sides with respect to
! ! N1
! x and then taking l-to-x limit, we get
P (N ; x) = δ N − N2,j  ! " ! "
N1 N2,1 ,··· ,N2,N1 j=1 df dN̄ 1 df dN̄
=
  dN̄ dx N̄ dN̄ x=0 dx x=0
N1
&
× P (N1 ; l)  P (N2,j ; x − l) Recalling that
j=1 dN̄
= αN̄ and N̄ (0) = 1
we have   2
dx
! ! N1
$ N1
! we obtain
(N12 )2 = P (N1 ; l)  P (N2,j ; x − l)  N2,j  ! "
df 1 df C
N1 N2,1 ,··· ,N2,N1 j=1 j=1
= 2 =:
dN̄ N̄ dN̄ x=0 N̄ 2

97
General solution to this equation is Recalling
!" x #
C
f (N̄ ) = C ! − Ḡ(x) := N̄ (x) = exp dl α(l)
N̄ 0

We need to impose the boundary condition we arrive at

# x
! " f0 (x)
f (N̄ (0)) = f (1) = 0 f Ḡ(x) = dl α(x)
0 [Ḡ(x)]2
since
P (N ; 0) = δ(N − 1) From this and
1
which requires f0 ≡ σ 2 =:
θpol + 1
C! = C we can calculate the Polya parameter if the
Townsend coefficient and f0 are known.
Denoting
Notice that the avalanche fluctuation is in
f0 := f (∞) = C
general non-Polya. Nevertheless we may use
we can rewrite the equation for f in the the Polya parameter as an index.
following form:
In the case of Legler’s model, we have
df f0
= 2
dN̄ N̄ (2 − eχ(x) )2
f0 (x) =
2 − (2 − eχ(x) )2
This equation allows us to extend our with
α(x) U0
uniform field results to a nonuniform field. χ(x) := α(x) x0 (x) =
E(x)

98
 Central Limit Theorem
Sketch of Its Proof
Characteristic Function Once a characteristic function is given, we
The characteristic function of a probability can calculate these moments as
!
n d
n !
distribution function P(x) is defined by Mn = (−i) φ(s)!!
! ds n
s=0
φ(s) := dx eisx P (x)
Examples
which is essentially the Fourier transform For instance, the characteristic function of
of the p.d.f. and hence uniquely specifies it. a Gaussian distribution is
The characteristic function comes in handy ! +∞
1 (x−x̄)2
for calculations of moments:
− 2σ2
φG (s) = isx
dx e √ e
! −∞ 2πσ
Mn := dx xn P (x) 1
= e− 2 σ
2 2
s +ix̄s

It is easy to make sure that the first three

By definition, we have
moments obtained from this characteristic
M0 = 1 function indeed coincide the above.
M1 = x̄
M2 = σ 2 + x̄2 M1 = x̄
M2 = σ 2 + x̄2

99
For an exponential distribution, we have from which we have
! +∞
1 x̄ = 1
φE (s) = dx eisx e−x/λ 1
0 λ σ2 =
−1 1+θ
= (1 − isλ)
If theta=0, the Polya distribution becomes
and hence
an exponential one with lambda=1 as is
M1 = x̄ = λ clearly seen either from the definition or
M2 = σ 2 + x̄2 = 2λ2
from its characteristic function.

For a Polya distribution

(θ + 1)θ+1 θ −(θ+1)x The Polya distribution becomes a delta-
PP (x) = x e
Γ(θ + 1) function in the limit of theta going to
we have infinity as is easily seen from its
! ∞
(θ + 1)θ+1 θ −(θ+1)x characteristic function:
φP (s) = dx e isx
x e
0 Γ(θ + 1) !
θ+1
"θ+1
" #θ+1 φP (s) =
θ+1 θ + 1 − is
=
θ + 1 − is → eis as θ → ∞

The 1st and the 2nd moments obtained The asymptotic form coincides with the
from the characteristic function are characteristic function for a Gaussian with
M1 = x̄ = 1 a mean value of unity and with a zero width.
2+θ
M2 = σ 2 + x̄2 =
1+θ

100
Composition Rules Proof of Central Limit Theorem
A p.d.f. for a random variable x induces a We consider N variables x1, ..., xN, obeying
p.d.f. for a variable (ax). The characteristic the same p.d.f.: P(x), and consider the
function for (ax) is then given by distribution of
! 1 !
N
1 z := √ (xi − x̄)
φax (s) = d(ax) eis(ax) P (x) = φx (as)
a N σ i=1

The characteristic function for (x+a) is The characteristic function for this is
! √
φz (s) = [φx−x̄ (s/ N σ)]N
φx+a = d(x + a) eis(x+a) P (x) = eias φx (s)
Recall now that we can expand phi in terms
A p.d.f. for a variable x1 and another p.d.f. of moments as follows
√
for a variable x2 induce a p.d.f. for their !∞
√ (is/ N σ)k
φx−x̄ (s/ N σ) = Mk
sum (x1+x2). The characteristic function k=0
k!
" #
s2 1
for this reads = 1− +O
! ! ! 2N N 3/2
φ1+2 (s) = dx eisx dx1 dx2 P1 (x1 ) P2 (x2 ) In the large N limit, we hence have
√
× δ (x − (x1 + x2 )) φz (s) = [φx−x̄ (s/ N σ)]N
! "N
= φ1 (s) · φ2 (s) s2 1 2
→ lim 1 − = e− 2 s
N →∞ 2N

For N variables with the same p.d.f., we get implying that the p.d.f. for z is a Gaussian
φN (s) = [φ(s)]N centered at zero with a variance of 1.

101
102
 Subjects Covered
Liberation of electrons by ionization (dE/dx)

Classical theory of electron transportation in a

chamber gas and its applications

Behaviors of electrons in E and B fields

Transport coefficients: Vd, Cd

Gas amplification Paul

Creation of signals Yulan
Coordinate measurement Takeshi

103
 Subjects Left Out
Electron attachment

CF4, O2 contamination, etc. Paul

Transportation of ions

+ve ion feed back

Yulan
Ion exchange and aging

These are very important in practice, but simply beyond the

scope of my lectures.

104