Chapter 3 - Poisson process

Chapter 3 - Poisson process

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Counting processes
Counting processes are increasing processes {Xt , t R+ } with
Xt N.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Counting processes
Counting processes are increasing processes {Xt , t R+ } with
Xt N.
Notation
The idea is to count something over time (eg, the arrivals of
customers, ...) : Xt is the number of arrivals between time 0
and time t so naturally we set X0 = 0.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Counting processes
Counting processes are increasing processes {Xt , t R+ } with
Xt N.
Notation
The idea is to count something over time (eg, the arrivals of
customers, ...) : Xt is the number of arrivals between time 0
and time t so naturally we set X0 = 0.
Let N(s,t] be the number of arrivals during the interval (s, t] :
N(s,t] = Xt Xs .
Note that Xt = N(0,t] .
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Definition of a Poisson process

Poisson process
A Poisson process with intensity > 0 is a counting process
{Xt } with
1
independent increments ;
2
P(Xt+h Xt = 1) = h + o(h) when h 0 ;
3
P(Xt+h Xt > 1) = o(h) when h 0.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Some remarks about (1)

Recall that by independent increments we mean that, for any
t1 < t1 t3 < t4 ,
Xt2 Xt1 indep. of Xt4 Xt3 .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Some remarks about (1)

Recall that by independent increments we mean that, for any
t1 < t1 t3 < t4 ,
Xt2 Xt1 indep. of Xt4 Xt3 .
In other words :
N(t1 ,t2 ] is indep. of N(t3 ,t4 ] .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Some remarks about (1)

Recall that by independent increments we mean that, for any
t1 < t1 t3 < t4 ,
Xt2 Xt1 indep. of Xt4 Xt3 .
In other words :
N(t1 ,t2 ] is indep. of N(t3 ,t4 ] .
This is sometimes referred as the loss of memory property.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Remarks on (2) and (3)

Recall (2) and (3) :
P(Xt+h Xt = 1) = h + o(h),
P(Xt+h Xt > 1) = o(h).

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Remarks on (2) and (3)

Recall (2) and (3) :
P(Xt+h Xt = 1) = h + o(h),
P(Xt+h Xt > 1) = o(h).
We can write (3) in a different way :
P(Xt+h Xt > 1)
0.
h0
h

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Remarks on (2) and (3)

Recall (2) and (3) :
P(Xt+h Xt = 1) = h + o(h),
P(Xt+h Xt > 1) = o(h).
We can write (3) in a different way :
P(Xt+h Xt > 1)
0.
h0
h
Two customers will never arrive exactly at the same time, and
the probability that two customers arrive at a very small period
of time is negligible.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Remarks on (2) and (3)

Recall (2) and (3) :
P(Xt+h Xt = 1) = h + o(h),
P(Xt+h Xt > 1) = o(h).
We can write (3) in a different way :
P(Xt+h Xt > 1)
0.
h0
h
Two customers will never arrive exactly at the same time, and
the probability that two customers arrive at a very small period
of time is negligible. Remark that (2) and (3) imply that :
P(Xt+h Xt = 0) = 1 h + o(h).
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

About Poisson
Simon Denis Poisson (1781-1840), the French mathematician
and physicist, who received the Copley medal from the Royal
Society of London in 1832.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Fundamental theorem
A reminder : the Poisson distribution
U P(`) for any k N,

`k `
P(U = k) = e .
k!

Then :
E(U) = Var(U) = `.


s
U P(`) for any s > 0, E e sU = e (e 1) .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Fundamental theorem
A reminder : the Poisson distribution
U P(`) for any k N,

`k `
P(U = k) = e .
k!

Then :
E(U) = Var(U) = `.


s
U P(`) for any s > 0, E e sU = e (e 1) .
Theorem
Let {Xt } be a Poisson process with intensity and
N(s,t] = Xt Xs for s < t. Then
N(s,t] P((t s)).
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

An immediate consequence
The arrival rate
The arrival rate between times s and t is :
Xt Xs
.
t s

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

An immediate consequence
The arrival rate
The arrival rate between times s and t is :
Xt Xs
.
t s
Consequence of the theorem
For a Poisson process, the expected arrival rate is constant,
equal to :


Xt Xs
E
= .
t s

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Arrival times
Definition
Let S1 , S2 , ... be the arrival times :
Sk = inf{t 0 :

Xt = k}

(and by convention S0 = 0).

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Definition of a Poisson process

Main properties
Some extensions

Chapter 3 - Poisson process

Arrival times
Definition
Let S1 , S2 , ... be the arrival times :
Sk = inf{t 0 :

Xt = k}

(and by convention S0 = 0).

Note that :
Xt =

1Si t .

i=1

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Distribution of the arrival times

Theorem
For any k 1, Sk (k, ) where we recall that the (k, )
distribution has density :
f (x) =

k x k1
e x .
(k)

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Gap between arrival times

Definition
For any k 1, Tk = Sk Sk1 .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Gap between arrival times

Definition
For any k 1, Tk = Sk Sk1 .
Theorem
The Tk are iid with distribution E(), where we recall that the
exponential distribution E() has density
f (x) = e x .
In particular, E(Tk ) = 1/ and Var(Tk ) = 1/2 .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Subdividing an interval with a given count

Question : suppose we know that Xt = n. How are the arrival
times S1 , S2 , ..., Sn distributed within the interval [0, t] ?

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Subdividing an interval with a given count

Question : suppose we know that Xt = n. How are the arrival
times S1 , S2 , ..., Sn distributed within the interval [0, t] ?
Example : Xt = 1. Then for 0 < s < t,
P (S1 < s, Xt = 1)
P (Xt = 1)


P N(0,s] = 1, N(s,t] = 0


=
P N(0,t] = 1

P (S1 < s|Xt = 1) =

se s e (ts)
s
=
=
t
te
t

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Subdividing an interval with a given count

Question : suppose we know that Xt = n. How are the arrival
times S1 , S2 , ..., Sn distributed within the interval [0, t] ?
Example : Xt = 1. Then for 0 < s < t,
P (S1 < s, Xt = 1)
P (Xt = 1)


P N(0,s] = 1, N(s,t] = 0


=
P N(0,t] = 1

P (S1 < s|Xt = 1) =

se s e (ts)
s
=
=
t
te
t
in other words, L(S1 |Xt = 1) = U[0, t] the uniform
distribution on [0, t].
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Subdividing an interval with a given count

Theorem
Let U1 , U2 , ..., Un be an iid sample with a uniform distribution
U[0, t]. Let U(1) , ..., U(n) be the corresponding order statistic,
ie :
{U1 , . . . , Un } = {U(1) , . . . , U(n) } ;
U(1) U(n) .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Subdividing an interval with a given count

Theorem
Let U1 , U2 , ..., Un be an iid sample with a uniform distribution
U[0, t]. Let U(1) , ..., U(n) be the corresponding order statistic,
ie :
{U1 , . . . , Un } = {U(1) , . . . , U(n) } ;
U(1) U(n) .
Then L(S1 , . . . , Sn |Xt = n) = L(U(1) , . . . , U(n) ).
In other words, conditional on Xt = n, the n arrivals are
uniformly distributed on [0, t].

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Marked Poisson process

Defintion - marked Poisson process
A marked Poisson process is defined as :
a Poisson process {Xt , t 0} with arrival times S1 , S2 , ...
a collection of iid random variables M1 , M2 , ... indep. of
{Xt }.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Marked Poisson process

Defintion - marked Poisson process
A marked Poisson process is defined as :
a Poisson process {Xt , t 0} with arrival times S1 , S2 , ...
a collection of iid random variables M1 , M2 , ... indep. of
{Xt }.
For example :
Sk is the arrival time of the k-th customer.
Mk is the money spent by the k-th customer.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Marked Poisson process

Defintion - marked Poisson process
A marked Poisson process is defined as :
a Poisson process {Xt , t 0} with arrival times S1 , S2 , ...
a collection of iid random variables M1 , M2 , ... indep. of
{Xt }.
For example :
Sk is the arrival time of the k-th customer.
Mk is the money spent by the k-th customer.
Or :
Sk is the time of occurence of the k-th earthquake ;
Mk is its magnitude.
...
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Definition of a Poisson process

Main properties
Some extensions

Chapter 3 - Poisson process

Thinned Poisson process

Definition - thinned Poisson process
Let {Xt } be a marked Poisson process with arrival times S1 ,
S2 , ... and marks M1 , M2 , ... Be(p). We define the thinned
Poisson process {Yt } by
Yt =

1Si t,Mi =1 .

i=1

In other words, {Yt } is obtained from {Xt } by erasing the

arrivals Si with Mi = 0.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Definition of a Poisson process

Main properties
Some extensions

Chapter 3 - Poisson process

Thinned Poisson process

Definition - thinned Poisson process
Let {Xt } be a marked Poisson process with arrival times S1 ,
S2 , ... and marks M1 , M2 , ... Be(p). We define the thinned
Poisson process {Yt } by
Yt =

1Si t,Mi =1 .

i=1

In other words, {Yt } is obtained from {Xt } by erasing the

arrivals Si with Mi = 0.
Theorem
The process {Yt } is actually a Poisson process with parameter
p.
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Superposition of Poisson process

Theorem
Let {Xt } and {Yt } be two Poisson processes independent of
each other, with intensity given by and , respectively. Let
us put Zt = Xt + Yt .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Superposition of Poisson process

Theorem
Let {Xt } and {Yt } be two Poisson processes independent of
each other, with intensity given by and , respectively. Let
us put Zt = Xt + Yt . Then {Zt } is a Poisson process with
intensity + .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Campbells Theorem

Campbells Theorem
For a Poisson process {Xt } with arrival times S1 , S2 , ... and
for any function f integrable on [0, t], we have
"X
#
Z t
t
X
f (Si ) =
f (s)ds.
E
0

i=1

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Campbells Theorem for marked processes

Campbells Theorem for marked processes

For a marked Poisson process {Xt } with arrival times S1 , S2 ,
..., marks M1 , M2 , ... M and for any function f integrable
on [0, t] M, we have
#
"X
Z t
t
X
f (Si , Mi ) =
E
E [f (s, M1 )] ds.
0

i=1

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Application of Campbells Theorem

A squirrel drinks in a river at random times given by a Poisson
process : S1 , S2 ...

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Application of Campbells Theorem

A squirrel drinks in a river at random times given by a Poisson
process : S1 , S2 ... Unfortunately, the river is contaminated by a
dangerous substance, and at each time he absorbs a random
quantity Mi of poison.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Application of Campbells Theorem

A squirrel drinks in a river at random times given by a Poisson
process : S1 , S2 ... Unfortunately, the river is contaminated by a
dangerous substance, and at each time he absorbs a random
quantity Mi of poison. Each dose of poison is eliminated by his
body at an exponential rate.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Application of Campbells Theorem

A squirrel drinks in a river at random times given by a Poisson
process : S1 , S2 ... Unfortunately, the river is contaminated by a
dangerous substance, and at each time he absorbs a random
quantity Mi of poison. Each dose of poison is eliminated by his
body at an exponential rate.
In other words, the quantity of poison in his body at time t is :
Q=

Xt
X

Mi e (tSi ) .

i=1

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Application of Campbells Theorem

A squirrel drinks in a river at random times given by a Poisson
process : S1 , S2 ... Unfortunately, the river is contaminated by a
dangerous substance, and at each time he absorbs a random
quantity Mi of poison. Each dose of poison is eliminated by his
body at an exponential rate.
In other words, the quantity of poison in his body at time t is :
Q=

Xt
X

Mi e (tSi ) .

i=1

Thanks to Campbells Theorem one can compute

Z t


E(M1 )
E(Q) =
E(Mi e (ts) )ds =
1 e t .

0
Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Inhomogeneous Poisson process : definition

In some cases, the assumption that the expected arrival rate is
constant is not realistic. For example, customers are more
likely to come to a shop after 5pm than before.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Inhomogeneous Poisson process : definition

In some cases, the assumption that the expected arrival rate is
constant is not realistic. For example, customers are more
likely to come to a shop after 5pm than before.
Definition
Let (t) > 0 be a function of the time t. An inhomogeneous
Poisson process with intensity (t) is a counting process {Xt }
with
1
independent increments ;
2
P(Xt+h Xt = 1) = (t)h + o(h) when h 0 ;
3
P(Xt+h Xt > 1) = o(h) when h 0.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Inhomogeneous Poisson process : properties

Let {Xt } be an inhomogeneous Poisson process with intensity
(t). Then we can prove that
Z t

N(s,t] = Xt Xs P
(x)dx .
s

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Inhomogeneous Poisson process : properties

Let {Xt } be an inhomogeneous Poisson process with intensity
(t). Then we can prove that
Z t

N(s,t] = Xt Xs P
(x)dx .
s

In particular, when (t) = is constant, {Xt } is a Poisson

process and we have :
N(s,t] = Xt Xs P ((t s)) .

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Planar Poisson point process

It is also possible to extend the Poisson process to the case
where the index t does not belong to R but to R2 . In this case,
t does not refer to the time but to a position in the plane.

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Planar Poisson point process

It is also possible to extend the Poisson process to the case
where the index t does not belong to R but to R2 . In this case,
t does not refer to the time but to a position in the plane.
It is more convenient to extend the definition of the arrival
times S1 , S2 , ...

Nial Friel

Stochastic Models - STAT30090 - STAT40680

Chapter 3 - Poisson process

Definition of a Poisson process

Main properties
Some extensions

Planar Poisson point process

It is also possible to extend the Poisson process to the case
where the index t does not belong to R but to R2 . In this case,
t does not refer to the time but to a position in the plane.
It is more convenient to extend the definition of the arrival
times S1 , S2 , ...
A Poisson process on the plane is a random family of points S
such that
the number of each point in the region K is distributed
according to P(.area(K )) ;
inside a region K , the points are uniformly distributed ;
the number of points in two disjoints regions K1 and K2 is
independent.
Nial Friel

Stochastic Models - STAT30090 - STAT40680