You are on page 1of 11

Poisson distribution as a limit theorem.

In a group of 2000 people, is there any chance of finding a day in the


year which is not a birthday?

Figure: The chance of finding a day in the year which is not a birthday.
A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 1 / 11
Frame Title

Under the assumption of homogeneity, what is the probability that k


insurance claims or credit defaults occur in a year?
If 10, 000 balls are distributed in 1000 cells, is there any chance of
finding an empty cell? exactly m empty cells?
We can develop a very nice expression as a limiting approximation to
answer questions of this kind.

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 2 / 11
Classical occupancy problem.

Consider the problem of a random distribution of r balls in n cells,


assuming that each arrangement has probability n−r . We seek the
probability pm (r , n) of finding eactly m cells empty.
Let Ak be the event that cell number k is empty (k = 1, 2, . . . , n). In
this event all r balls are placed in the remaining (n − 1) cells, and this
can be done in (n − 1)r distinguishable ways. Similarly, there are
(n − 2)r arrangements, leaving two pre-assigned cells empty, etc.

r
2 r 3 r
    
1
pi = 1− , pij = 1 − , pijk = 1 − ,... (1)
n n n

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 3 / 11
Classical occupancy problem.

Or,

 
n ν r
Sν = 1− (2)
ν n

The probability that at least one cell is empty is given by


P1 = S1 − S2 + S3 − . . . and hence the probability that all cells are
occupied is 1 − S1 + S2 − . . . or

n  
X n ν ν r
p0 (r , n) = (−1) 1− (3)
ν n
ν=0

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 4 / 11
Exactly m cells empty.

Consider now a distribution in which exactly m cells are empty. These
n
m cells can be chosen in m ways. The r balls are distributed among
the remaining (m − m) cells so that each of these cells is occupied;
the number of such distributions is (n − m)r p0 (r , n − m). Dividing
throughout by nr we find for the probability that exactly m cells
remain empty-
n
r
1− m

pm (r , n) = m n p0 (r , n − m)
hP r i
n r n−m ν n−m

= m 1− m ν
 
n ν=0 (−1) ν 1 − n−m
n
 hPn−m ν n−m
 n−m r  n−m−ν r i
= m ν=0 (−1) ν
n
 Pn−m ν n−m
 n m+ν n−m r
= m ν=0 (−1) ν 1− n

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 5 / 11
The limit approximation to pm (r , n).

We begin by estimating Sν of the formula (2). Since :

(n − ν)ν < n · (n − 1) · (n − ν + 1) < nν


 ν r  ν r  ν r
(n − ν)ν 1 − < (n)ν 1 − < nν 1 −
n n n
 ν ν+r
  ν r
nν 1 − < ν!Sν < nν 1 −
n n

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 6 / 11
The limit approximation to pm (r , n).
We have the infinite series :

1
= 1 + t + t2 + t3 + . . .
1−t
t2 t3 t4
−log (1 − t) = t + + + ...
2 3 4
t 2 3 4
= t + t + t + t + ...
1−t

We can therefore write the double inequality:

t
t < − log(1 − t) <
1−t
t
− < log(1 − t) < −t
1−t
e −t/(1−t) < (1 − t) < e −t
A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 7 / 11
The limit approximation to pm (r , n).

Using the double inequality with t = ν/n, we get:

 ν+r ir
−ν/n h
n e ν
1−ν/n < ν!Sν < nν e −ν/n

ν+r ν
h i h iν
ne − n−ν < ν!Sν < ne −r /n

Our purpose is to discuss the limiting form of the above formula.


That is, what happens as n → ∞ and r → ∞? The relation between
n and r is arbitrary in principal. However, the ratio r /n represents the
average number of things per cell. Let’s assume that this quantity is
bounded.

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 8 / 11
The limit approximation to pm (r , n).

Substitute :

ne −r /n = λ

and suppose that r and n increase in such a way that λ remains


bounded.
Then, for each fixed ν, the raio of the extreme members in the above
equation tends to unity (both the left hand side and right hand side
limits equal λν ) and we conclude that:

λν λν
Sν < and − Sν → 0.
ν! ν!

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 9 / 11
The limit approximation to pm (r , n).

It follows that :


X λν
p0 (r , n) − (−1)ν →0
ν!
ν=0

p0 (r , n) = e −λ

The factor of p0 (r , n − m) may be rewritten as Sm in the equation (4)


and we therefore for each fixed m

λm −λ
pm (r , n) = e
m!

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 10 / 11
Poisson Distribution

Theorem.
If n and r tend to infinity, so that λ = ne −r /n remains bounded, then the
expression below, called the Poisson pmf holds for each fixed m:

λm −λ
p(m; λ) = e
m!

A review of basic probability theory Poisson distribution as a limit theorem Sep 2020 11 / 11

You might also like