Introduction to Stochastic Models GSLM 54100 Outline counting process Poisson process definition: interarrival ~ i.i.d.
Download
Report
Transcript Introduction to Stochastic Models GSLM 54100 Outline counting process Poisson process definition: interarrival ~ i.i.d.
Introduction to Stochastic Models
GSLM 54100
1
Outline
counting process
Poisson process
definition: interarrival ~ i.i.d. exp
properties
independent increments
stationary increments
P(N(t+h) N(t) = 1) h for small h
P(N(t+h) N(t) 2) 0 for small h
composition of independent Poisson processes
random partitioning of Poisson process
conditional distribution of (Si|N(t) = n)
2
Counting Process
{N(t)}
is a counting process if N(t) = the total
number of occurrences of events on or
before t
N(0)
=0
N(t)
is a non-negative integer
N(t)
is increasing (i.e., non-decreasing) in t
for
s < t, N(s, t] = the number of events
occurring in the interval (s, t]
3
Poisson Process
a
Poisson process {N(t)} is a counting
process of rate (per unit time) if the interarrival times are i.i.d. exponential of mean
1/
I ~ i.i.d. exp()
3
for each sample
point, a Poisson
process is a graph
2
N(t)
1
S1
0
1
S2
2
S3
t
3
4
Distribution of Arrival Epochs
Sn
= X1 + … + Xn, Xi ~ i.i.d. exp()
ex (x)n1
f Sn ( x )
, x0
(n 1)!
5
Distribution of N(t)
P(N(t)
= 0) = P( X1 t ) et , t 0
P(N(t)
= 1) P( X1 t , X1 X 2 t )
t
E[ P( X1 t , X1 X 2 t )] 0 P( X 2 t y) f X1 ( y)dy
tet
t (t y ) y
e dy
0 e
3
2
N(t)
1
0
S1
S2
t
S3
6
Distribution of N(t)
P( N (t ) n)
P( Sn t , Sn X n1 t )
E[ P(Sn t , Sn X n1 t | Sn )]
0t P( X n1 t y) f Sn ( y)dy
y
(y )n1
t (t y ) e
0 e
dy
(n 1)!
n
(
t
)
e t
3
n!
ex (x)n1
f Sn ( x )
, x0
(n 1)!
2
N(t)
1
0
S1
S2
t
S3
7
Increments
N(s,
t] = N(t) N(s)
= number of arrivals in (s, t]
= increments in (s, t]
8
Properties of the Poisson Process
independent
dependent
stationary
increments
increments
increments
non-stationary
increments
P(N(t+h)
N(t) = 1) h for small h
P(N(t+h)
N(t) 2) 0 for small h
9
Property of the Poisson Process:
Independent Increments
number
of increments in disjoint intervals
are independent random variables
< t2 t3 < t4, N(t1, t2] = N(t2) - N(t1) and
N(t3, t4] = N(t4) - N(t3) are independent random
variables
for t1
dependent
increments (Example 7.2.2)
10
Property of the Poisson Process:
Stationary Increments
number
of increments in an interval of
length h ~ Poisson(h)
a
Poisson variable of mean h
N(s,
t] ~ Poisson((ts)) for any s < t
non-stationary
increments (Example 7.2.4)
11
Property of the Poisson Process
P(N(t+h)
N(t) = 1) h for small h
P(N(t+h)
N(t) 2) 0 for small h
12
Example 7.2.7 & Example 7.2.8
13
Properties of the Poisson Process
composition of independent Poisson processes
summation
of independent Poisson random
variables
random partitioning of Poisson process
random
partitioning of Poisson random variables
Example
7.2.12
conditional distribution of (Si|N(t) = n)
14
Summation of Independent
Poisson Random Variables
X
~ Poisson(), Y ~ Poisson() , independent
Z
=X+Y
distribution
of Z?
15
Summation of Independent
Poisson Random Variables
n
P( X n k , Y k )
P ( Z n) P ( X Y n)
k 0
P( X n k ) P (Y k )
e nk e k
k!
k 0 (n k )!
nk
k
e()
k 0 (n k )! k !
e() n n nk k
Ck
n ! k 0
n!
nk k
e
n!
k 0 k !(n k )!
e( ) ( ) n
n!
n
n
k 0
n
( )
n
Z ~ Poisson(+)
16
Composition of Independent
Poisson Processes
{X(t)}
~ Poisson
process of rate
X(t)
t
Y(t)
{Y(t)}
~ Poisson
process of rate
Z(t)
t
= X(t) +Y(t)
distribution
of
Z(t)?
Z(t)
type
t
of {Z(t)}?
17
Composition of Independent
Poisson Processes
Xi ~ i.i.d. exp()
X(t)
Yi ~ i.i.d. exp()
t
X1
Y(t)
X’1 ~ i.i.d. exp()
X’1=(X1Y1|X1>Y1)
t
Y1
Y2
Z1 ~ exp(+)
Z(t)
Z2 ~ exp(+),
independent of Z1
t
Z2 = min(X’1, Y2)
Z1 = min(X1, Y1)
by the same argument, Zi ~
exp(+), i.e., {Z(t)} is a
Poisson process of rate +
18
Random Partitioning of
Poisson Random Variables
X
items, X ~ Poisson()
each
item, if available, is type 1 with
probability p, 0 < p < 1
Y
= # of type 1 items in X
distribution
of Y?
19
Random Partitioning of
Poisson Random Variables
P(Y k ) E[ P(Y k | X )]
Ckn p k (1 p)nk
nk
e n
n!
P(Y k | X n) P ( X n)
nk
nk n k
(1
p
)
e ( p)k
(n k )!k !
nk
e ( p)k (1 p)nk nk
k!
(n k )!
nk
e ( p)k (1 p)m m
k!
m!
m 0
e ( p)k (1 p )
e
k!
e p ( p) k
k!
Y
~ Poisson(p)
20
Independent Partitioned
Random Variables
Y
~ Poisson(p), and XY ~ Poisson((1p))
surprising
fact: Y and XY being
independent
P(Y k , X Y n k ) P( X n, Y k )
e n
p)
P(Y k | X n) P( X n)
n!
e p ( p)k e(1 p ) [(1 p)]nk
n ! p k (1 p)nk e n
k!
(n k )!
k !(n k )!
n!
Ckn p k (1
n k
P(Y k ) P( X Y n k )
21
Random Partitioning
of Poisson Processes
{X(t)}
~ Poisson
process of rate
type 1
type 2
X(t)
type 2
each
t
X1
X2
item is type 1
with probability p
X3
Y(t)
distribution
t
Y1
of Yi?
type
of process of
{Y(t)}
22
Random Partitioning
of Poisson Processes
M
Y X i , X i ~ i.i.d. exp(), M ~ Geo( p)
i 1
have
argued that Y ~ exp(p) before, or
fY ( s)
fM
Xi
(s)
i 1
(1 p)m1 pes
m1
P( M m) f m
(s)m1
(m 1)!
k
[(1
p
)
s
]
pes
k!
k 0
Xi
m1
pe
(s)
i 1
m1
[(1
p
)
s
]
pes
(m 1)!
m1
s (1 p )s
e
pe ps
can argue that Yi ~ i.i.d. exp(p), i.e.,
{Y(t)} is a Poisson process of rate p
23
Random Partitioning
of Poisson Processes
{Y(t)}
is a Poisson process of rate p
{X(t)Y(t)}
is a Poisson process of rate
(1p)
no
dependence of interarrival times among
{Y(t)} and {X(t)Y(t)}
{Y(t)}
and {X(t)Y(t)} are independent
Poisson processes
24
Uniform Distributions
U, Ui ~ i.i.d. uniform[0, t]
for 0 < s < t, P(U > s) = (ts)/t
for 0 < s1 < s2 < t, one of U1 and U2 in (s1, s2] and
the other in (s2, t]
P(one
of U1, U2 in (s1, s2] & the other in (s2, t] )
= P(U1 (s1, s2], U2 (s2, t])
+ P(U2 (s1, s2], U1 (s2, t])
2( s2 s1 )(t s2 )
=
t2
25
Conditional Distribution of Si
P(S 1 s | N (t ) 1)
P( X 1 ( s, t ] | N (t ) 1)
(
S1|N(t) = 1) ~
uniform(0, t)
P( N (0, s] 0, N ( s, t ) 1)
P( N (t ) 1)
s
(t s )
e (t s)e
tet
ts
t
26
Conditional Distribution of Si
P(S1 (s1, s2 ], S2 (s2 , t ] | N (t ) 2)
P( N (0, s1 ] 0, N ( s1, s2 ] 1, N ( s2 , t ] 1)
P( N (t ) 2)
es1( s2 s1 )e( s2 s1 )(t s2 )e(t s2 )
2( s2 s1 )(t s2 )
t2
2t 2et
2!
it can be shown that
( S1, S2 |N(t) = 2)
~ ( U[1], U[2] |N(t) = 2)
27
Conditional Distribution of Si
N(t) = n, S1, …, Sn distribute as the
ordered statistics of i.i.d. U1, …, Un
Given
28
Example 7.2.15
29
Equivalent Definition
of the Poisson Process
a
counting process {N( t)} is a Poisson
process of rate (> 0) if
(i)
N(0) = 0
(ii)
{N( t)} has independent increments
(iii)
for any s, t 0,
P( N (t s ) N ( s) n) e t
( t ) n
, n 0,1,...
n!
30
Equivalent Definition
of the Poisson Process
a
counting process {N( t)} is a Poisson
process of rate (> 0) if
(i)
N(0) = 0
(ii)
{N( t)} has stationary and independent
increments
(iii)
P(N(h) = 1) h for small h
(iv)
P(N(h) 2) 0 for small h
31