Chapter 6: Counting Statistics

Chapter 6: Counting Statistics

1
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Table of Contents

F Statistical laws governing radioactive decay.

F Commonly used statistical distributions.
F Error and error propagation.
F Counting radioactive samples.
F Minimum significant measured activities – Type-I errors.
F Minimum detectable true activities – Type-II errors.
F Effect of instrument response
F Introduction to Monte Carlo methods

2
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Statistics of Radiation and Radiation Detection

Statistical nature of radiation and radiation interaction:

F How much energy will an 1 MeV photon lose in its next collision with an atomic
electron?
F Will a 400keV photon penetrate a 2 mm lead shielding without interaction?
F When we use measured count-rate to estimate the activity of a source, and
how certain are we on the estimation?

3
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Exponential Radioactive Decay

Sample activity (A)
F True sample activity is never known.
F The best we can do is to repeat the counting process for a number of times and
use the average as an indication of the sample activity – average number of
decays in the sample per second.

A = A0 e - lt
F The above equation can be interpreted by implying that the probability that an
atom survives a time t without disintegration is
q = probability of survival = e - lt
and
p = probability of decay = 1-q = 1-e -lt

The actual number of decay events is fluctuating around the average value
predicted by this equation.
4
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

Consider the radioactive disintegration process in a sample, it follows the following
four conditions:
F It consists of N trials.
F Each trial has a binary outcome: success or failure (decay or not).
F The probability of success (decay) is a constant from trial to trial – all atoms
have equal probability to decay.
F The trials are independent.
In statistics, these four conditions characterize a Bernoulli process.

5
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
Given, p, N and t, what is the probability of observing n disintegrations within a
time t?
F The number of ways to chose n atoms from a total of N atoms in the sample is

æNö N!
çç ÷÷ =
è n ø n!( N - n)!
F So the probability of the n atoms chosen decayed during the time span t is

æ N ö n N -n
Pn = çç ÷÷ p q
ènø
F The above equation describes the so-called Binomial distribution.

F What are the mean and standard deviation of a Binomial distribution?

6
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution

F What are the mean and standard deviation of a Binomial distribution?

7
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution

8
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution

9
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
For a binomial distribution, the mean or the expectation of the number of
disintegration in time t is given by

N
æ N ö n N -n
N
µ º å n × Pn = å n × çç ÷÷ p q = Np
n =0 n =0 ènø

and the fluctuation on the number of disintegrations is given by the variance or

the standard deviation of the

N
s º å (n - µ )2 × Pn = Npq
2

n =0

and
N
sº å (n - µ )2
× Pn = Npq
n =0

10
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
Considering a realistic case, in which we use a detector to measure the number of
counts and use the measured count rate to infer to the activity of the source.
Given (a) each disintegration yield one single particle and (b) the detection
efficiency of the detector is e, then

The prob. of detecting a count within a time t is

The probability of an atom disintegrates and results in a detected count is

p * = ep = e (1 - e -lt )
and the probability of an atom either does not disintegrate
or the resultant particle is not detected is
q * = 1 - ep = 1 - e + ee -lt

Therefore, we can use the binomial distribution to describe the counting statistics
as

11
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
The prob. of detecting n count within a time t is
æNö
P = çç ÷÷(ep ) (1 - ep )
* n N -n
n
ènø
p : prob. of an atom disintergrates within a time t
e : detection efficiency of the detector
The mean number of detected counts is

æNö *
( ) (q )
N N
µ º å n × Pn = å n × çç ÷÷ p * N -n
n
*
= eNp
n =0 n =0 ènø

and the variance on the number of detected counts is

(s ) º å (n - µ )
N
* 2 2
× Pn* = Np * × q *
n =0

12
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

An Example Binomial Distribution

13
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
For a binomial distribution, the mean or the expectation of the number of
disintegration in time t is given by

N
æ N ö n N -n
N
µ º å n × Pn = å n × çç ÷÷ p q = Np
n =0 n =0 ènø

and the fluctuation on the number of disintegrations is given by the variance or

the standard deviation of the

N
s º å (n - µ )2 × Pn = Npq
2

n =0

and
N
sº å (n - µ )2
× Pn = Npq
n =0

14
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

An Example Binomial Distribution

15
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

æNö
Pn = çç ÷÷ p n q N - n
ènø

16
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

(Revisited)
Consider the radioactive disintegration process in a sample, it follows the following
four conditions:
F It consists of N trials.
F Each trial has a binary outcome: success or failure (decay or not).
F The probability of success (decay) is a constant from trial to trial – all atoms
have equal probability to decay.
F The trials are independent.
In statistics, these four conditions characterize a Bernoulli process.

17
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution (Revisited)

F The probability of exactly n decays out of a total of N atoms in the source is

æ N ö n N -n
Pn = çç ÷÷ p q
ènø
For a binomial distribution, the mean or the expectation of the number of
disintegration within the measurement period is given by
N
æ N ö n N -n
N
µ º å n × Pn = å n × çç ÷÷ p q = Np
n =0 n =0 ènø

and the fluctuation on the number of disintegrations is quantified by the variance

N
2
σ ≡ ∑ ( n − µ) ⋅ Pn = Npq
2

n =0

18
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution – An Example

Consider a particle physics experiment – estimate the flux rate (number of particles per
second across a unit solid angle) of a certain type of particle. Suppose the particles are
coming from a point source.

Note that the detection efficiency of the detector is p=0.55, and the measurement has
S·T=1.9×102 cm2 · sec.

If measurement did not register a single count, how do we estimate and

report the flux rate (number of particles coming towards the detector
surface per cm2 per second) of the particle?

19
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

Consider the radioactive disintegration process in a sample, it follows the following
four conditions:
F It consists of N trials.
F Each trial has a binary outcome: success or failure (decay or not).
F The probability of success (decay) is a constant from trial to trial – all atoms
have equal probability to decay.
F The trials are independent.
In statistics, these four conditions characterize a Bernoulli process.

20
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution
Given, p, N and t, what is the probability of observing n disintegration within a time
t?
F The number of ways to chose n atoms from N atoms in the sample is

æNö N!
çç ÷÷ =
è n ø n!( N - n)!
F The probability of exactly n decays is

æ N ö n N -n
Pn = çç ÷÷ p q
ènø
F The above probability function characterizes the so-called Binomial
distribution.

21
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

F The Bernoulli Process and binomial distribution provide a nice statistic model
for the decay of radioactive substances.
F In reality, we often encounter situations, in which p is very small and N is very
large …

F In such cases, the statistical description of the decay process could be

simplified …

22
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Probability Distribution Function of Binomial Distribution

for 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0
For Binomial distribution,

æ N ö n N -n æNö N!
Pn = çç ÷÷ p q , where çç ÷÷ =
ènø è n ø n!( N - n)!
As in the previous example, for large N and small n, one can write
Binomial Expansion
0
0
𝑎+𝑏 = 2 𝑐03 ⋅ 𝑎 083 ⋅ 𝑏 3
345
0
𝑛
=2 ⋅ 𝑎 083 ⋅ 𝑏 3
𝑟
345
0
𝑛!
=; ⋅ 𝑎 083 ⋅ 𝑏 3
𝑛 − 𝑟 ! 𝑟!
<45

Tylor Expansion
D
𝑥0 𝑥5 𝑥G 𝑥I 𝑥K
𝒆C = ; = + + + …⋅
𝑛! 0! 1! 2! 3!
045
 Chapter 6: Counting Statistics

Mean of Binomial Distribution for 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0

The mean of binomial distribution is given by 𝜇 ≡ 2 𝑛 ⋅ 𝑃0

0

where

therefore

24
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Mean of Binomial Distribution for 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0

Binomial Distribution in Extreme Case

Therefore,

[1 + (- p)]N =

Using E.17 and letting p’=-p, we have the following equation

¥
( Np ' ) n ( Np ' ) 2
å0 n! = 1 + Np'+ 2! + ! = e Np '

25
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Mean of Binomial Distribution for 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0

Now go back to the mean of a Binomial distribution given by

µ º å n × Pn
n

Since
¥
( Np ) n ( Np ) 2
å0 n! = 1 + Np +
2!
+ ! = e Np
,

then

µ º å n × Pn
n

26
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Standard Deviation and Variance of Binomial Distribution

for 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0
The variance of binomial distribution is given by
D D

𝜎 I ≡ 2 (𝑛 − 𝜇)I ⋅ 𝑃0 = 2 𝑛I 𝑃0 − 𝜇 I
045 045

Since

D D D
𝑛𝜇 0 𝑛 ⋅ 𝜇0 𝜇0
; =; =;
𝑛! 𝑛! 𝑛−1 !
045 04Y 04𝟏
D D
𝜇 08Y 𝜇0
= 𝜇; = 𝜇; = 𝜇𝒆8\
𝑛−1 ! 𝑛!
04𝟏 04𝟎

D
𝜇0
; = 𝐞\
Substitute E.26 into the first equation, we have 𝑛!
045

And the standard deviation is given by

 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

Consider the radioactive disintegration process in a sample, it follows the following
four conditions:
F It consists of N trials.
F Each trial has a binary outcome: success or failure (decay or not).
F The probability of success (decay) is a constant from trial to trial – all atoms
have equal probability to decay.
F The trials are independent.
In statistics, these four conditions characterize a Bernoulli process.

What happens if 𝑁 >> 𝑛 𝑎𝑛𝑑 𝑝 → 0 ?

28
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Process

The counting statistics related to nuclear decay processes is often more

conveniently described by the Poisson distribution, is related to situations that
involves a collection of multiple trials that satisfy the following conditions:

1. The number of trials, N, is very large, e.g. N>>1.

2. Each trial is independent.
3. The probability that each single trial is successful is a constant and approaching
zero, p<<1. So the number of successful trials is fluctuating around a finite
number.

29
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution and Poisson Distribution

Binomial distribution
Poisson distribution when
The probability of observing n successful trails
N>>1, p<<1
out of a total of N independent trails:
µ n -µ
! N $ n N−n P(n | µ ) = e
Pn = # &p q n!
" n %
mean of the observed number of successful trails :
Mean of n:
N N
æNö
µ º å n × Pn = å n × çç ÷÷ p n q N -n = Np µ (n) = µ = N ⋅ p
n =0 n =0 ènø

Standard deviation:
Standard deviation :
N
2
Std(n) ≡ ∑(n − µ ) ⋅ Pn = Npq σ = µ = Np
n=0

Gaussian distribution, If N is further increased, and p is further decreased

( x-µ )2
1 -
p(x | µ , s ) = e 2s 2

2p s
30
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Distribution

Remember the conditions for Binomial distribution to be approximated by

Poisson Distribution:
1. The number of trials, N, is very large, e.g. N>>1.
2. Each trial is independent.
3. The probability that each single trial is successful is a constant and approaching
zero, p<<1. So the number of successful trials is fluctuating around a finite
number.
31
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Distribution
The probability of having n successful trials can be approximated with the Poisson
distribution.

µn
P(n | µ ) = e-µ
n!
and the mean and the variance of number of successful trial are given by

Mean(n) = µ = N × p

Std (n) º s = µ = Np

32
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Registered k
counts The detector covers 10% solid angle.

detection efficiency: 𝜆 =55%.

N particles
reached the Det.
detector The measurement takes T = 30 min.

Source N particles reached the detector.

mean 𝑁 = 𝑚
detected k=0 count.

The prob. of having N particles reaching the detector would follow the Poisson
distribution, _ ` 8_
𝑃 𝑁 = 𝑒 ,
a!
where m is the expected number of particles that reach the detector.

Once the N particles reached the detector, the number of particles detected would
follow the Binomial distribution, so that the probability of detecting k particles is

𝑁 f
𝑃 𝑘 |𝑁 = 𝜆 (1 − 𝜆)a8g .
𝑘

33
 Chapter 6: Counting Statistics

Poisson Distribution
Therefore, the total probability of detecting k counts is
D

𝑃 𝑘 = 2 𝑃 𝑘|𝑁 𝑃 𝑁 |𝑚
a4f
D
a! f a8f _ ` 8_
=2 𝜆 1−𝜆 ⋅ e
a4f a8f !f! a!

_j n 8(_j)
= e
f!

If we would like to have 90% chance of detecting at least 1 particle, then we could
write
1 − 𝑃 𝑘 = 0 = e8_j = 0.9,
then the mean number of particles reaching the detector during the 30 minute
measurement should be
𝑚 = 4.2.
34
 Chapter 6: Counting Statistics

The detector covers 10% solid angle with

respect to the point source.
detection efficiency: 𝜆 =55%.

The measurement takes T = 30 min.

N particles reached the detector.
detected k=0 count.

So finally, we can conclude that:

Because we did not record any count, we have 90% confidence to claim that
the source strength (average number of particles emitted per second)
should not exceed

4.2
𝐴≤ = 42 𝑝𝑎𝑟𝑡𝑖𝑐𝑙𝑒𝑠 𝑝𝑒𝑟 𝑠𝑒𝑐 = 42 𝐵𝑞
10%

35
 Chapter 6: Counting Statistics

Binomial Distribution and Poisson Distribution

Binomial distribution
Poisson distribution when
The probability of observing n successful trails
out of a total of N independent trails:
N>>1, p<<1
! N $ n N−n µ n -µ
Pn = # &p q P(n | µ ) = e
" n % n!

mean of the observed number of successful trails :

Mean of n:
N N
æNö
µ º å n × Pn = å n × çç ÷÷ p n q N -n = Np µ (n) = µ = N ⋅ p
n =0 n =0 ènø

Standard deviation:
Standard deviation :
N
2
Std(n) ≡ ∑(n − µ ) ⋅ Pn = Npq σ = µ = Np
n=0

36
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 5: Methods for Radiation Detection – Semiconductor Detectors

Effect of Fano Factor on Energy Resolution

à ~3eV for silicon à 100 keV gives 30000 e-h pairs à what is the standard
deviation associated with the number of e-h pairs per 100 keV energy
deposition?
à ~30eV for gas detectors à 100 keV gives 3000 e-ion pairs à ??
à ~1 keV for NaI(Tl) à 100 keV only gives 100 photoelectrons à ??

à The measured Fano factors: 0.143 for silicon, 0.129 for germanium, 0.1 for CdZnTe and
~0.1 for HgI2.
à In comparison, the Fano factors for gas and scintillators are ~1.

37
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 5: Methods for Radiation Detection – Semiconductor Detectors

Semiconductor Detector Configurations

High-purity germanium (HPGe) detectors

F Supper-pure material available, for example 1 part in 1012

F Depletion depth of >1cm à good detection volume.
F Requires cooling to liquid nitrogen temperature to reduce leakage current.

38
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 5: Methods for Radiation Detection – Semiconductor Detectors

Signal Generation by Ionizing Radiation

in Semiconductors
The Fano factor.

à For a given energy deposition E in the detector and a known energy ε required to
create an e-h pair, the observed fluctuation in the number of charge carriers created
is smaller than the one predicted by the Poisson statistics.
à The measured Fano factors: 0.143 for silicon, 0.129 for germanium, 0.1 for CdZnTe
and ~0.1 for HgI2.
à In comparison, the Fano factors for scintillators are ~1.

39
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

40
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Distribution
The probability of having n successful trials can be approximated with the Poisson
distribution.

µn
P(n | µ ) = e-µ
n!
and the mean and the variance of number of successful trial are given by

Mean(n) = µ = N × p

Std (n) º s = µ = Np

41
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

The Gaussian (Normal) Distribution

As p (the prob. of an atom decay within t) is getting even smaller and N is getting
larger, both Binomial and Poisson distributions are approaching an extremely
useful form of distribution – the Gaussian distribution.

Gaussian distribution is defined for a continuous variable x

( x-µ )2
1 -
p(x | µ , s ) = e 2s 2

2p s

but it is very useful for describing the counting fluctuation on discrete numbers.

42
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Distribution

43
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

The Gaussian (Normal) Distribution

Binomial and Poisson distributions practically match the normal distribution when
µ >=30.

44
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution and Poisson Distribution

Binomial distribution
Poisson distribution when
The probability of observing n successful trails
out of a total of N independent trails:
N>>1, p<<1
! N $ n N−n µ n -µ
Pn = # &p q P(n | µ ) = e
" n % n!

mean of the observed number of successful trails :

Mean of n:
N N
æNö
µ º å n × Pn = å n × çç ÷÷ p n q N -n = Np µ (n) = µ = N ⋅ p
n =0 n =0 ènø

Standard deviation:
Standard deviation :
N
2
Std(n) ≡ ∑(n − µ ) ⋅ Pn = Npq σ = µ = Np
n=0

Gaussian distribution, If N is further increased, and p is further decreased

( x-µ )2
1 -
p(x | µ , s ) = e 2s 2

2p s
45
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

The Gaussian (Normal) Distribution

Binomial and Poisson distributions practically match the normal distribution when
µ >=30.

46
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

The Gaussian (Normal) Distribution

For a variable, x, following the normal distribution, the probability that it takes a
value between x and x is equal to the area under under the curve p(x) between
these two values:

Many common manipulations when carried out on counting data that were
originally Gaussian distributed will produce derived values that also follow
Gaussian shape:
FMultiplying or dividing the data by a constant,
FCombining two Gaussian-distributed variables through addition, subtraction, or
multiplication or,
FCalculating the average of a series of independent measurements.

47
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

The Gaussian (Normal) Distribution

Turner, pp. 316-317.

48
NPRE 441, Principles of Radiation Protection, Spring 2020
 Central Limit Theorem
The sum or average of a large number of independent random variables follows
Gaussian (Normal) distribution.

The distribution of an average tends to be NORMAL, even when the distribution

of the underlying variables from which the average is computed is decidedly non-
Normal.

49
NPRE 441, Principles of Radiation Protection, Spring 2020
 Central Limit Theorem
Consider a series of independent and identically distributed (i.i.d.) random
variables, x1, x2, …, xn, whose probability density function are given by

ì 1, 0 £ x £ 1
pn ( x ) = í
î0 , otherwise

50
NPRE 441, Principles of Radiation Protection, Spring 2020
An Example of Central Limit Theorem

åx n
x= n
n

51
NPRE 441, Principles of Radiation Protection, Spring 2020
 Why Gaussian Random Variable is Important?

When a quantity is derived as the result of a large number of

accumulative effects, and each individual effect has a small
contribution to the final outcome, then the distribution of the
quantity tends to follow Gaussian distribution.

52
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Radioactive Disintegration – Bernoulli Process

Consider the radioactive disintegration process in a sample, it follows the following
four conditions:
F It consists of N trials.
F Each trial has a binary outcome: success or failure (decay or not).
F The probability of success (decay) is a constant from trial to trial – all atoms
have equal probability to decay.
F The trials are independent.
In statistics, these four conditions characterize a Bernoulli process.

53
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Poisson Process

The counting statistics related to nuclear decay processes is often more

conveniently described by the Poisson distribution, is related to situations that
involves a collection of multiple trials that satisfy the following conditions:

1. The number of trials, N, is very large, e.g. N>>1.

2. Each trial is independent.
3. The probability that each single trial is successful is a constant and approaching
zero, p<<1. So the number of successful trials is fluctuating around a finite
number.

54
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Binomial Distribution and Poisson Distribution

Binomial distribution
Poisson distribution when
The probability of observing n successful trails
out of a total of N independent trails:
N>>1, p<<1
! N $ n N−n µ n -µ
Pn = # &p q P(n | µ ) = e
" n % n!

mean of the observed number of successful trails :

Mean of n:
N N
æNö
µ º å n × Pn = å n × çç ÷÷ p n q N -n = Np µ (n) = µ = N ⋅ p
n =0 n =0 ènø

Standard deviation:
Standard deviation :
N
2
Std(n) ≡ ∑(n − µ ) ⋅ Pn = Npq σ = µ = Np
n=0

Gaussian distribution, If N is further increased, and p is further decreased

( x-µ )2
1 -
p(x | µ , s ) = e 2s 2

2p s
55
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error and Error Propagation

Two ways to express the error associated with a given measurement:

Probable error:
FThe symmetric range about the mean, within which there is 50% chance that a
measurement will fall.
FThe width of the range depends on the distribution of the variable. For example,
for Gaussian distributed error, the probable error is ±0.675 s.

Fractional standard deviation:

FThe ratio of the standard deviation and the mean of the distribution of the
random variable.
FFor Poisson distributed random variable, the fractional standard deviation is
simply
s 1
=
µ µ
56
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation
In some situations, the variable of interest (Q) is not measured directly, but derived
as a function of more than one independent random variable whose values are
directly measured. The error on the measured values is propagated into the
uncertainty on the resultant quantity Q.

Suppose a quantity Q(x,y) that depends on two independent random variables x

and y.
The sample mean and variance of variables x and y are derived as sx and sy, by
repeating measurements.
The standard deviation of the indirect quantity Q is approximately given by

2 2
æ ¶Q ö 2 æ ¶Q ö
s Q2 @ ç s +
÷ x ç ç ÷ s
÷ y
2

è ¶x ø è ¶y ø
2
æ ¶Q ö
sQ @ åçç ÷ s xi
2 2
÷
i è ¶xi ø
57
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error and Error Propagation

A Taylor series of a real function of a single variable, f(x), around a point x0 is given by

1 2 1 3
f (x0 + Δx) = f (x0 ) + f x (x0 )Δx + f xx (x0 ) ( Δx ) + f xxx (x0 ) ( Δx ) +......
2! 3!
where
"d d %
f xx (x0 ) = $ f (x)'
# dx dx & x=x0

A Taylor series of a real function of two variables, f(x,y), is given by

f ( x0 + Dx, y0 + Dy ) =
[
f ( x0 , y0 ) + f x ( x0 , y0 )Dx + f y ( x0 , y0 )Dy ]
+
1
2!
[
f xx ( x0 , y0 )(Dx ) + 2 f xy ( x0 , y0 )DxDy + f yy ( x0 , y0 )(Dy )
2 2
]
1
3!
[ 3 2 2
]
+ f xxx ( x0 , y0 )(Dx ) + 3 f xxy ( x0 , y0 )(Dx ) (Dy ) + 3 f xyy ( x0 , y0 )(Dx )(Dy ) + f yyy ( x0 , y0 )(Dy ) + ......
3

58
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation

59
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation

60
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error and Error Propagation

61
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error and Error Propagation

Assumptions ??

62
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation Formula

The error propagation formula is exact only when
• the two variables, x and y, are independent to each other,
• and when Q(x,y) could be approximated as a linear function of both x
and y.

2
æ ¶Q ö
s Q2 @ å çç ÷÷ s xi 2
i è ¶xi ø

2
æ ¶Q ö
s Q2 @ å çç ÷÷ s xi 2
i è ¶xi ø
 Chapter 6: Counting Statistics

Error Propagation Formula

The error propagation formula is exact only when
• the two variables, x and y, are independent to each other,
• and when Q(x,y) could be approximated as a linear function of both x
and y.

Note that the formula would break down when the second and third
and higher order partial derivatives are not negligible.
f ( x0 + Dx, y0 + Dy ) =
[
f ( x0 , y0 ) + f x ( x0 , y0 )Dx + f y ( x0 , y0 )Dy ]
+
1
2!
[
f xx ( x0 , y0 )(Dx ) + 2 f xy ( x0 , y0 )DxDy + f yy ( x0 , y0 )(Dy )
2 2
]
1
3!
[ 3 2 2
]
+ f xxx ( x0 , y0 )(Dx ) + 3 f xxy ( x0 , y0 )(Dx ) (Dy ) + 3 f xyy ( x0 , y0 )(Dx )(Dy ) + f yyy ( x0 , y0 )(Dy ) + ......
3

64
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation
Case 1: Sums or differences of counts – u is the sum or difference of two random
numbers representing counts measured in two independent experiments.

Example: estimating the net counts from a sample.

65
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation

Case 2: Multiplication or division by a constant

Example: estimating the count rate,

Assuming that the error in the measuring time is negligible, we get

66
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation
Case 3: Multiplication or division of counts

2
Using the equation æ ¶Q ö
sQ @ å çç ÷÷ s xi 2
2

i è ¶xi ø

One gets

67
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation in Net Count Rate Measurement

Assuming no error on t

I
where 𝜎03 ≡ 𝜎 I 𝑟0 , 𝜎zI ≡ 𝜎 I 𝑛z , 𝜎zI ≡ 𝜎 I 𝑛{ ,
I 0} I 0•
and 𝜎z3 = 𝜎I , 𝜎{3 = 𝜎I .
~} ~•
Turner, pp. 324.
68
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation in Net Count Rate Measurement

69
NPRE 441, Principles of Radiation Protection, Spring 2020
 Chapter 6: Counting Statistics

Error Propagation in Net Count Rate Measurement

Assuming no error on t

Turner, pp. 324.

70
NPRE 441, Principles of Radiation Protection, Spring 2020