Stochastic Modelling T G Venkatesh

Poisson Stochastic Process:

Figure 8: Arrival Process

A stochastic process in which the inter arrival times between the events follow exponential distribution is said to be
a Poisson Stochastic Process. Poisson stochastic process is the basic ingredient for the queuing theory. In Figure 8, we
show a sample arrival process. Here X1 is the time of arrival of the first customer (occurrence of the first event). Xi is
the time between the arrival of i − 1th and ith customers. In a Poisson stochastic process, X1 , X2 , . . . are independent
and identical random variables following an exponential distribution. N (t) is the number of arrivals until time t. N (t)
is a non-decreasing function.
Examples of Poisson Stochastic Processes:
• In a large city, homing millions of people, the passengers in a railway station gets service to book train tickets.
The arrival of passengers to the booking counter is a Poisson stochastic process.Here each customer arrives to
the counter independently.

• Birth statistics of a large country like India

• Cosmic particles from stars
• Electrons from thermionic emissions.
Few Definitions:

• The occurrence instance of an event is called epoch.

• Time difference between two epoch is called inter-arrival time.
• N (t) is a non-negative integer valued function denoting the number of events occurred till time t. N (t) has a
discrete state space (non-negative integers) where the time t is continuous. N (t) is a non-decreasing function.

N (t) = Number of events occurred in (0, t]

• N (t) is a discrete state and continuous time stochastic process. It is also known as a counting process.
The number of events occurred in the interval (s, t], where s ≤ t, is given by N (t) − N (s).
In addition, we can keep more structure to it. The number of events in the non-overlapping intervals (t1 , t2 ] and
(t3 , t4 ]. where t1 ≤ t2 ≤ t3 ≤ t4 , are independent to each other.

P (N (t4 ) − N (t3 ) = n|N (t2 ) − N (t1 ) = m) = P (N (t4 ) − N (t3 ) = n)

45
July, 2021 EE 6150: Stochastic Processes and Theory of Queues Instructor: TG Venkatesh
Independent Increment Process and Poisson Stochastic Process Scribe: Navya

Independent Increment Process

If a counting process {N (t), t ≥ 0} satisfies the independent increment property, then the process is known as inde-
pendent increment process. It is a discrete state continuous time process satisfying N (t) ≥ N (s) for the continuous
time variable t ≥ s.
If we fix the interval size to a fixed value, say τ . Then irrespective of the time instant at which an event occurs,
the number of arrivals in each of these intervals depends only on the interval’s duration τ , rather than its exact time-
instants. Therefore the nature of such processes doesn’t change with time. Such a process is known as a time stationary
process.
The counting process which satisfies both the time stationary and independent increment property is known as
stationary increment process and it satisfies the property P[N (t + τ ) − N (t) = n] = P[N (s + τ ) − N (s) = n] for
any given time -instants s, t.

Poisson Stochastic Process

A discrete state continuous time counting process {N (t), t ≥ 0} is said to be a Poisson process if it satisfies the
following properties:
1. N (0) = 0
2. The process N (t) has independent increments.
3. The process N (t) has stationary increments.
The pmf of a Poisson process is given as,
n
e−λt (λt)
P(N (t) = n) = (7)
n!
which models the number of arrivals in an interval (0, t) and λ is the Poisson rate. If we fix the value of the time
variable t, then the values taken by N (t) process is a random variable whose range is {0, 1, 2, 3 . . . }. The probability
that the process takes a given value say n is given by its pmf as shown in (7).
�∞ −λt n
The above given pmf satisfies the property n=0 e n!(λt) =1 and thus it can be considered as a valid pmf.
Example: If it is given that customers arrive at a particular queue at a mean rate of 0.5 customers per minute. Then
what is the probability that 7 customers could have arrived in a duration of 3 minutes?
Solution: Here it is given λ = 0.5, t = 3, n = 7. Therefore the average rate of customers arrival in a duration of 3
minutes is given as λ × t = 0.5 × 3 = 1.5. Thus,
7
e−1.5 (1.5)
P(N (0, 3] = 7) = (8)
7!
Now, consider the case of ∞ population, where each customer arrivals is independent of the arrivals of the other
customers. Further, consider a small interval of time (h → 0), such that there can be either 1 arrival or no arrivals in
that small time interval.
The value of P(N (t + h) − N (t) = 1) is proportional to the interval length h as per the independent stationary
increment property. By introducing the proportionality constant as λ which denotes the rate of occurrence of an event,
we get P(N (t + h) − N (t) = 1) = λh. As the value of the interval size increases, i.e., as h increases, probability
of a single event occurring in that interval increases with λh. Determine how small the value of h should be, so that
exactly only 1 event occurs in that interval length h? As the value of h decreases, the error also decreases where error
is a function of O(h). Therefore,
Stochastic Modelling T G Venkatesh

P(N (t + h) − N (t) = 1) = λh + error =⇒ P(N (t + h) − N (t) = 1) = λh + O(h) (9)

Therefore we requires limh→0 O(h) h = 0. Any function which satisfies this rule is known as order of h function.
Examples: h2 , h3 , 2h2 + 3h3 , ... are all order-h functions. It means that these functions decays faster than h. If O(h) is
a order-h function, than cO(h) is also a order-h function where c is a constant. Similarly if f (h), g(h) are both order-h
functions, than f (h) + g(h) is also a order-h function.

P(N (t + h) − N (t) = 1) = λh + O(h)

P(N (t + h) − N (t) = 2) = O(h)
P(N (t + h) − N (t) = 3) = O(h) (10)
...
P(N (t + h) − N (t) = 0) = 1 − λh + O(h)

Let Pn (t + h) = P(N (t + h) = n). i.e., in an interval (t+h) they are n events. The different ways in which this
can occur is,
1. There can be n events in (0, t) and no events in (t, t + h)
2. There can be (n − 1) events in (0, t) and 1 event in (t, t + h)
3. There can be (n − 2) events in (0, t) and 2 events in (t, t + h)
4. There can be (n − 3) events in (0, t) and 3 events in (t, t + h)
5. . . .
Therefore by including all the above mentioned different cases, we get,

Pn (t + h) = Pn (t)[1 − λh + O(h)] + Pn−1 (t)(λh + O(h)) + Pn−2 (h)(O(h)) + Pn−3 (h)(O(h)) + . . . (11)

Re-arranging the above terms we get,

Pn (t + h) − Pn (t) = −λhPn (t) + λPn−1 (t).h + O(h) (12)

Dividing by h, and taking limits as h → 0, we get,

� P (t + h) − P (t) � O(h)
n n
lim = −λPn (t) + λPn−1 (t) + lim (13)
h→0 h h→0 h
The above equation can be re-written in the form of a differential equation as,
�
Pn (t) = −λPn (t) + λPn−1 (t) (14)

However, the above equation is not true for the case of n = 0, as n − 1 is invalid while computing the above equation.
Therefore assuming Pn−1 (t) = 0 for the initial condition, we get,
�
P0 (t + h) = P0 (t)[1 − λh + O(h)] =⇒ P0 (t) = −λP0 (t) (15)

The solution for the above ordinary differential equations given in (14) with the initial conditions of (15) can be
−λt n
found out as P(N (t) = n) = e n!(λt) .

Poisson Process: Properties

In this sub-section, we explore the key properties that a Poisson process possesses.

47
Stochastic Modelling T G Venkatesh
1. Sum of two independent Poisson processes:
Let N1 (t) and N2 (t) be two independent Poisson processes with rates λ1 and λ2 . Now we want to model the combined
process N (t) = N1 (t) + N2 (t) which captures the number of events in interval (0, t) and these events can be either
of N1 (t) or N2 (t) type. Example:Router in a communication network which multiplexes two incoming links which
are joined at the router to form a single outgoing link. In this case, the router acts as a statistical multiplexer for these
two independent incoming links. Further the packet arrivals at each of these links are assumed to be Poisson arrivals.
Therefore,
�n r n−r
e−λ1 t (λ1 t) e−λ2 t (λ2 t)
P(N (t) = n) = (16)
r=0
r! (n − r)!
Further simplifying the above expression, we get,

e−(λ1 +λ2 )t n
P(N (t) = n) = [(λ1 + λ2 )t] (17)
n!
Thus from the above equation, we observe that the combined process is also a Poisson process with rate λ = λ1 + λ2
Another way to prove the above property is by using the Moment generating Functions(MGF). The MGF of the
Poisson process G(s) is calculated as,
∞
� n
e−λt (λt)
G(S) = E[sN (t) ] = sn P(N (t) = n)where P(N (t) = n) =
n=0
n!
∞
(18)
� n
e−λt (λt) n
=⇒ G(s) = E[sN (t) ] = s = e−λt(s+1)
n=0
n!

Applying MGF to the two independent Poisson processes we get, G1 (s) = e−λ1 t(s+1) and G2 (s) = e−λ2 t(s+1) .
Further the MGF of the combined process N (t) = N1 (t) + N2 (t) can be found out as,

G(s) = E[sN (t) ] = E[sN1 (t)+N2 (t) ] = E[sN1 (t) ]E[sN2 (t) ] = G1 (s)G2 (s) = e−(λ1 +λ2 )t(1+s) (19)

We can observe that G(s) is the MGF of the combined Poisson process whose pmf can be calculated as P(N (t) =
−(λ1 +λ2 )t n
n) = e n! [(λ1 + λ2 )t] which is same as (17).

2. Thinning of Poisson Process:

Often in the real world, we might be interested only in a subset of the events generated by the Poisson process.
Example 1: Assume that the arrival of a products in a manufacturing company is modeled by using a Poisson process.
These products can be either faulty with a probability p or a right one with a probability of (1 − p). Thus on arrival,
the product is dropped ( or thinned down) with probability p if it is a faulty one or retained with probability (1 − p)
if the product is a right-one. Example 2: There are two separate screens for doing the security checks of males and
females in a market. Thus assuming that the arrival of the customers to the market is Poisson with rate parameter λ
and further, the customers are split into two queues that are separate for the males and females. Assuming that the
customer is a male with a probability of p, we can model the arrival process at the male security check as another
Poisson process with a rate parameter of λp. Similarly, the arrival process at the female security check is modeled
using another independent Poisson process with rate parameter λ(1 − p). Similarly if the Poisson events (with rate
�i
parameter λ )are divided into i classes with probabilities p1 , p2 , ...pi such that n=1 pn = 1. Then the arrivals to each
of these classes can be modeled as independent Poisson processes with arrival rates of λp1 , λp2 , ..λpi .
Lets assume that the process {N (t)|t ≥ 0} is sampled into another process {M (t)|t ≥ 0} where with a probability
p the events of the original process N (t) are retained in the sampled process M (t). Then what is the P(M (t) =
n|N (t) = n + r)? That is only n events of the total n + r events are retained in the sampled process. It is calculated
as,
�∞ � � −λt n+r
n+r n re (λt)
P(M (t) = n|N (t) = n + r) = p (1 − p) (20)
r=0
n (n + r)!

48
Stochastic Modelling T G Venkatesh
Further simplifying the above equation, we get,
n
(λpt) −λpt
P(M (t) = n|N (t) = n + r) = e (21)
n!

49
July, 2021 EE 6150: Stochastic Processes and Theory of Queues Instructor: TG Venkatesh
Poisson Process Scribe: Aparna

Recall of Poisson Process

A stochastic process {N (t), t ≥ 0} is said to be a counting process if the process counts the number of events that occur
by the time t. One of the most important counting process is a Poisson process. The counting process {N (t), t ≥ 0}
is defined as a Poisson process with rate λ > 0, if the following axioms hold good,
1. N (0) = 0
2. {N (t), t ≥ 0} has independent increments
3. P(N (t + h) − N (t) = 1) = λh + O(h)
4. P(N (t + h) − N (t) > 1) = O(h)
It means that the number of events in any interval of length t is a Poisson random variable with mean λt.

Poisson Process: Properties

The key properties possessed by a Poisson process that we studied in the previous lecture are :
1. The sum of two Poisson processes is again a Poisson process with the rate parameter of the compound process
as the sum of the rates of the individual processes.
2. A Poisson process with rate parameter λ can be split into individual queues with probabilities p and (1 − p),
then the arrivals at each of these queues can also be modeled using a Poisson process with rates λp and λ(1 − p)
In this lecture, we study the other key properties of a Poisson process, such as
1. Inter arrival times of the Poisson process are exponentially distributed.
2. The time for the nth event of a Poisson process to occur denoted as P(Sn ≤ t) = P(N (t) ≥ n) is modelled
using a Gamma distribution .
3. Uniformity Property for the conditional distribution of arrival times

Inter Arrival Times

The time between two successive Poisson events {N (t)|t ≥ 0} is known as inter-arrival time and is denoted as
{Ti |i = 1, 2, . . . }. A Poisson process is modeled as a discrete state continuous-time stochastic process. Starting
from time 0, we look for the time of occurrence for the 1st event of the Poisson process to occur, and we denote this
inter-arrival time as T1 . Similarly, starting from T1 , we look for the time instant of the second event of the Poisson
process and denote this inter-arrival time as T2 . Thus each of these inter-arrival times can be considered as random
variables, and a stochastic process can be constructed comprising of these random variables as {Ti |i = 1, 2, . . . }.
Here we observe that i is a discrete-valued parameter while Ti is a continuous-valued random variable. Therefore the
CDF of the inter-arrival times is given as P(Ti ≤ t) = FTi (t).
The successive inter-arrival times are found out to be IID, that follows from the independent increment nature
of the Poisson process. Each of these inter-arrival time random variables Ti are of exponential distribution of mean
1/λ and represented as Ti = Exp(λ). P (Ti > t) is just the probability of zero Poisson arrivals in time ”t”, and is
exp(−λt). The CDF of Ti is 1 − exp(−λt), showing that Ti are exponentially distributed. Suppose we start looking
at a random time and wait for the occurrences of an event. What is the time of observance of the third event, given the
second event has already occurred? This is known as residual time. The residual time is also exponentially distributed
with the same parameter λ.
Stochastic Modelling T G Venkatesh
Example 1: Say the event is counting the number of customer arrivals at a shop. Given that the 10th customer has
already entered the shop, What is the time for the 11th customer to enter?
Example 2: A bus arrives at a bus stop. Assume that the bus will not start unless there are 10 customers. Assume
the arrival of customers is Poisson distributed. Let the time of occurrence of the nth event be denoted ad Sn . therefore
the time of occurrence of the 1st event be S1 = T1� , the time of occurrence for the second event be S2 = T1 + T2 . Thus
n
the time of occurrence of the nth event be Sn = i=1 Ti . The CDF of Sn is given as P(Sn ≤ t). Based on this, we
can tell that the Poisson process satisfies N (t) ≥ n. In other words, it means that at time t, the number of occurrences
≥ n then the time of occurrence of the nth event < t. Therefore the CDF of Sn (t) is given as,
∞
� j
(λt)
FSn (t) = P(sN ≤ t) = P(N (t) ≥ n) = e−λt (22)
j=n
j!

Therefore the density function of Sn (t) is calculated by differentiating FSn (t) w.r.t t
n−1
(λt)
fSn (t) = λe−tλ (23)
(n − 1)!
Thus from the above, we observe that the density function for the random variable Sn , which denotes the time of
occurrence of the nth Poisson event is nothing but the Gamma distribution which is the sum of n exponential random
variables.

Uniformity Property of Poisson Stochastic process

Assume that the doctor assistant has booked an appointment for a patient at 8 : 00 and the doctor came at 8 : 30 and
finds the patient waiting. Given that the doctor knows only that one patient has arrived in between 8 : 00 and 8 : 30,
what is the distribution of the time of arrival of the patient? It is a random variable between 0 and 30. More generally
we can show that given that there is a Poisson event between 0 and t, the time of occurrence of the event is uniformly
distributed between 0 and t. Given that N (t) = 1. What is the distribution of Ti ? i.e.,
P(T1 ≤ s; N (t) = 1)
P(T1 ≤ s|N (t) = 1) = (24)
P(N (t) = 1)
The numerator in above expression means that 1 event should occur between 0 and s and no event to occur between
(s, t). Thus further simplifying the above expression, we get,
(λs)1 e−λs −λ(t−s) λ(t−s)0
e
1! 0!
= s/t (25)
λte−λt
Therefore the CDF of random variable FT1 (s) = s/t. Thus the PDF is given as,
�
1/t; o ≤ s ≤ t
fT1 (s) = (26)
0; else

The above result can be generalized further, but before proceeding, we first discuss important results regarding the
order statistics.

Order Statistics
Assume X1 , X2 , . . . , Xn as uniformly distributed random variables distributed between (0, 1). The first minimum
of X1 , X2 , . . . , Xn be represented as X(1) . Similarly the ith minimum of X1 , X2 , . . . , Xn be represented as X(i) .
We say that X(1) , X(2) , X(3) , . . . , X(n) are the order statistics of X1 , X2 , . . . , Xn . Then the joint density function of
X(1) , X(2) , X(3) , . . . , X(n) for uniformly distributed random variables is given as,
n!
FX(1) ,X(2) ,X(3) ,...,X(n) (x1 , x2 , . . . , xn ) = P(X(1) ≤ x1 , X(2) ≤ x2 , . . . , X(n) ≤ xn ) = (27)
tn

51
Stochastic Modelling T G Venkatesh
Generalisation for the Conditional Distribution of arrival times
Now assume that by the time doctor has arrived at 8 : 30, they are 5 patients waiting. i.e., N (t) = 5. We want to find
the joint PDF of the waiting time of these 5 patients?
Let S1 , S2 , . . . , Sn be the arrival times of the n Poisson events. Now we need to find fS1 ,S2 ,...,Sn (s1 , s2 , . . . , sn |N (t) =
n) i.e., at time t we know only n events have occurred i.e., N (t) = n and we also know that s1 ≤ s2 ≤ s3 ≤ . . . sn ≤ t.
The above equations are further solved by assuming that between the time 0 and s1 , an event occurs with a rate of λ
and the probability of such an event to occur is λe−λs1 . Similarly the probability for another event to occur between
the time instants s1 and s2 is given as, λe−λ(s2 −s1 ) . Further all the above events are assumed to be independent events
occurring over non-overlapping intervals. Therefore,

P(s1 , s2 , s3 , . . . , sn ; N (t) = n)
fS1 ,S2 ,S3 ,...,Sn (s1 , s2 , s3 , . . . , sn |N (t) = n) =
P(N (t) = n)
(28)
λs1 e−λs1 λ(s2 − s1 )e−λ(s2 −s1 ) . . . λ(sn − sn−1 )e−λ(sn −sn−1 ) e−λ(t−sn )
= n = n! /tn
e−λt (λt) /n!

Therefore given that N (t) = n, the n arrival times S1 , ..., Sn have the same distribution as the order statistics
corresponding to n independent random variables that are uniformly distributed on the interval (0, t).
Now lets analyse 2 independent Poisson processes {N1 (t)|t ≥ 0} with parameter λ1 and the other process be
{N2 (t)|t ≥ 0} with parameter λ2 . Lets consider few examples based on these independent Poisson processes.

1. What is the probability that event 1 (E1 ) occur before event 2 (E2 )?
Solution: Let T1 and T2 be the time of occurrence of the two events.
� ∞ � ∞
λ1
P(E1 < E2 ) = P(T1 < T2 ) = P(T1 < T2 |T1 = x)λ1 e−λ1 x dx = P{x < T2 }λ1 e−λ1 x dx =
0 0 λ1 + λ2
(29)
2. Similarly, what is the probability that two events of the first process occur before the second process?
Solution: Let Sji represents theith event of the j th process. Therefore using memory-less property, we need to
calculate,
λ1 λ2
P(S21 < S12 ) = P(S11 < S12 ).P(S21 < S12 |S11 < S12 ) = (30)
λ1 + λ2 λ 1 + λ2
λ2
3. Similarly, the probability that the second event occurs before the first event is given by λ1 +λ2 .
This is analogous to the Bernoulli trial. Let the occurrence of an event from process 1 behead (H) and event
from process 2 be tail (T). Then what is the probability of HTHHT?

PH PT PH PH PT
(31)
PH + PT PH + PT PH + PT PH + PT PH + PT
λ1 λ2
where PH = λ1 +λ2 and PT = λ1 +λ2

4. What is the probability of getting a minimum of n heads in (m+n-1) trails?

� �
m+n−1 �
m + n − 1 � λ1 �r � λ2 �m+n−r−1
(32)
r=n
r λ1 + λ2 λ1 + λ2

52
Stochastic Modelling T G Venkatesh
Generalisation of Poisson Process to Higher dimensional
A process consisting of randomly occurring points in the plane is said to constitute a two-dimensional Poisson process
having rate λ. There can be two indices i, j as {Ni,j |i, j = 1, 2, . . . } is known as Bidirectional Stochastic Process
where i, j can be either continuous or discrete.

P(Ni,j (t + h) − Ni,j (t) = 1) = λh + O(h)

P(Ni,j (t + h) − Ni,j (t) = 0) = 1 − λh + O(h) (33)
P(Ni,j (t + h) − Ni,j (t) > 1) = O(h)

Non-homogeneous Poisson process

Here the rate of the Poisson process λ(t) is not constant over a given time period

P[N (t + h) − N (t) = 1] = λ(t)h + O(h)

P[N (t + h) − N (t) = 0] = 1 − λ(t)h + O(h) (34)
P[N (t + h) − N (t) > 1] = O(h)

Example Problems
Now let us consider few more examples.
Example 1: Customers arrive at a store at the rate of 10 per hour. Each customer is either male or female with a
probability of 1/2 . Assume that we know that exactly 10 women entered within a given hour (say, 10 to 11 am). (a)
Compute the probability that exactly 10 men also entered.

Solution: Male and female arrivals are independent Poisson processes, with parameter is (1/2)10 = 5, therefore the
10
probability that exactly 10 men also entered is e−5 510!

(b) Compute the probability that at least 20 customers have entered?

�∞ �∞ k �9 k
Solution: k=10 P(k men entered) = k=10 e−5 5k! = 1 − k=0 e−5 5k!

Example 2: Assume that cars arrive at a rate 10 per hour. Assume each car will pick up a hitchhiker with a probability
of 1/10. You are second in line. What is the probability that you’ll have to wait for more than 2 hours?
Solution: Cars that pick up hitchhikers are a Poisson process with rate 10
1
10 = 1. For this process P(T1 + T2 > 2) =
P(N (2) ≤ 1) = 3e−2

Example 3: Order of events in independent Poisson processes: Assume that you have two independent Poisson
processes, N1(t) with rate λ1 and N2(t) with rate λ2 . What is the probability that n events occur in the first process
before m events occur in the second process?
Solution: The probability that n events occur in the first process before m events occur in the second process is,

� �
n+m−1 �
n + m − 1 � λ1 �k � λ2 �n+m−k−1
(35)
k λ1 + λ2 λ1 + λ2
k=n

5
Example 4: Assume that λ1 = 5, λ2 = 1. (a) Then P(5 events in the first process before 1 in the second) = ( 56 ) .
�6 � �� �k � 1 �6−k 11.55
(b) P(5 events in the first process before 2 in the second) = k=5 k6 56 6 = 66

53
Stochastic Modelling T G Venkatesh
Example 5: You have three friends, A, B, and C. Each will call you after an Exponential amount of time with the
expectation of 30 minutes, 1 hour, and 2.5 hours respectively. You will go out with the first friend that calls. What is
the probability that you go out with A?
Solution: Interpret each call as the first event in the appropriate one of three Poisson processes with rates 2, 1, and 2/5,
assuming the time unit to be one hour. (Recall that the rates are inverses of the expectations.)Probability of A calling
first is then clearly λ1λ+λ
1
2
= 2+1+(2/5)
2
17 .
= 10

Example 6: An office has two clerks, and three people, A, B, and C, enter simultaneously. A and B begin service at the
two clerks while C waits for the first available clerk. Assume that the service time is Exponential(λ). (a) Compute the
probability that A is the last to finish the service. (b) Compute the expected time before C is done (i.e., C’s combined
waiting and service time).
Solution: This is the probability that two events happen in a rate Poisson process before a single one in an indepen-
dent rate process, that is, 1/4
(b) First C has to wait for the first event in two combined Poisson processes, which is a single process with rate 2λ,
and then for the service time;
Solution: The answer is 2λ 1
+ λ1 = 2λ3

Example 7: A car wash has two stations, 1 and 2, with Exponential(λ1 ) and Exponential(λ2 ) service times. A car
enters station 1. Upon completing the service at station 1, the car then proceeds to station 2, provided station 2 is free;
otherwise, the car has to wait at station 1, blocking the entrance of other cars. The car exits the wash after service at
station 2 is completed. When you arrive at the wash, there is a single car at station 1. Compute your expected time
before you exit?
Solution: Your total time is (time the other car spends at station 1) + (time you spend at the station 2)+(maximum of
the time the other car spends at station 2 and the time you spend at the station 1). If T1 and T2 are Exponential(λ1 )
and Exponential(λ2 ), then we need to compute

E(T1 ) + E(T2 ) + E(max{T1 , T2 }) (36)

where
max{T1 , T2 } = T1 + T2 − min{T1 , T2 } (37)
and min{T1 , T2 } is Exponential(λ1 + λ2 ). Therefore,
2 2 1
E(T1 ) + E(T2 ) + E(max{T1 , T2 }) = + − (38)
λ1 λ2 λ1 + λ2

54