October 2010
S 0 = 0,
Sn =
i =1
Xi , n 0.
A renewal process is the process N = {N (t ), t R+ } dened as N (t ) = max{n : Sn t }, t 0. Distribution of N (t ): Since N (t ) n Sn t then P {N (t ) = n} = P {N (t ) n} P {N (t ) n + 1} = P {Sn t } P {Sn+1 t } = Fn (t ) Fn+1 (t ).
Theorem
For all t 0 it holds m(t ) = Proof:
i =1
Fn (t ).
E [N (t )] =
i =1
nP {N (t ) = n} n [Fn (t ) Fn+1 (t )]
i =1
= =
i =1
nFn (t )
i =1
nFn+1 (t ) =
i =1
Fn (t ).
Under assumption F (0) = P {Xi = 0} < 1 then m(t ) < . From now on we will always assume this property.
Renewal equation
Theorem
The following holds:
t
m(t ) = F (t ) +
0
m(t x )dF (x ) t 0.
Proof:
E [N (t )] = E [E [N (t )| X1 ]] =
0 t
E [N (t )| X1 = x ]dF1 (x )
=
0 t
E [N (t )| X1 = x ]dF1 (x ) +
t
E [N (t )| X1 = x ]dF1 (x )
t
=
0
E [N (t x ) + 1]dF1 (x ) = F (t ) +
0
m(t x )dF (x )
The distribution of SN (t )
Theorem
Let F denotes the survival function of the Xi . For 0 s t it holds
s
P {SN (t ) s } = F (t ) + Proof:
F (t y )dm(y ).
0
P {SN (t ) s } =
n=0
= F (t ) +
n=1
= F (t ) +
n=1 s 0
= F (t ) +
0
F (t y )d
n=1
Fn (y ) = F (t ) +
0
F (t y )dm(y )
Limit theorems
Theorem
With probability 1, it holds lim N (t ) 1 = t
SN (t ) N (t )
t N (t )
<
SN (t )+1 N (t )
t . Moreover
for t . Thus
The following three results are stated without proof. Elementary Renewal Theorem
Theorem
It holds m(t ) 1 = t t lim
Blackwells Theorem. Recall that a nonnegative r.v. X (or its distribution F ) is said to be a lattice if there exist d 0 such that i =0 P {X = nd } = 1.
Theorem
If F is not a lattice, then for all a 0 m(t + a) m(t ) If F is a lattice with period d, then E [number of renewals at nd ] d as n a as t
Theorem
If F is not a lattice, and h is a Riemann integrable function, then
t t 0
lim
h(t x )dm(x ) =
0
h (t ) dt .
Theorem
Let P (t ) = P [system is ON at time t ]. If E [Xi ] < and F is nonlattice, then E [Zi ] lim P (t ) = t E [Zi + Yi ] Proof:
= H (t ) +
0
H (t y )dm(y ) t
H (t )dt E [Z ] = E [X ] E [Z + Y ]
Equilibrium distributions
Consider: Y (t ) = SN (t )+1 t : residual life at t A(t ) = t SN (t ) : age at t
Theorem
If the Xi are nonlattice and < then lim P {Y (t ) x } = lim P {A(t ) x } =
t x 0
F (y )dy
Proof:
t
lim P {A(t ) x } = =
E [min(X , x )] E [X ]
0
x 0
F (y )dy
is called equilibrium distribution of X The distribution Fe (x ) = 0 (or Palm distribution). Its relevance in the applications is obvious.
F (y )dy
0 t 0 t
P {XN (t )+1 > x |SN (t ) = y } dFSN (t ) (y ) P {XN (t )+1 > x |XN (t )+1 > t y } dFSN (t ) (y ) F (max(x , t y ) dFSN (t ) (y ) F (t y )
t
=
0
t x
=
0 t x
dFSN (t ) (y ) +
t x
F (x ) dFSN (t ) (y ) F (t y )
t t x
F (x ) dFSN (t ) (y ) +
F (x ) dFSN (t ) (y ) F (x )
A renewal process N that possesses stationary increments is called stationary point process. Particularly important among the stationary point processes is the equilibrium renewal process, denoted here as Ne , that is dened as the renewal processes but where the rst epoch time X1 = S1 is distributed according to the equilibrium distribution (the distribution of Y (t )).
Theorem
For the equilibrium renewal process Ne : t me (t ) = ; P {Ye (t ) x } = Fe (x ) for all t 0; Ne has stationary increments. Proof: see Theorem 3.5.2 in Ross.