5.

Discrete Distributions continued…

5.6 Poisson distribution

A Poisson distribution is a discrete probability distribution that measures the probability of a certain number
of events occurring within a given period, assuming these events happen at a constant average rate and
independently of prior events. It is a distribution for count data where the experiment's outcome is a random
variable that takes on positive integer values, and the only other information available is a measurement of its
average value.

A Poisson process describes a situation where events occur at a constant rate, yet randomly and independently.
The Poisson process models events where the average time between events is known, but the exact timing of
events is random e.g.: modelling the number of phone calls received at a call centre. It assumes that:
• the calls occur randomly and independently of each other (the arrival of one call does not affect the
arrival of the next call) and,
• the number of calls per period of interest is constant e.g., calls could occur at an average rate of 15
calls per hour.

Early applications of the Poisson distribution included modelling alpha particles emitted from a radioactive
source within a given period. It could be used in analysing traffic flow, predicting faults in electric cables,
modelling the number of births per hour at a hospital, or modelling the calls that come into an exchange system
during a specific period, provided that the exchange services customers who act independently of each other.

We will study the Poisson distribution in two ways:

• in section 5.6.1, we derive the distribution as the limit of a Binomial distribution when the number of
trials 𝑛 approach infinity and the probability 𝑝, of success on each trial approaches zero, and
• in section 5.6.2, we define it as a distribution with its parameters.

5.1
5.6.1 Poisson Approximation to the Binomial distribution
This section shows how the Poisson distribution can be derived as the limit of a Binomial distribution when
the number of trials 𝑛 approaches infinity and the probability 𝑝, of success on each trial approaches zero.

For example, suppose we want the probability distribution of the number of vehicle breakdowns in a
designated stretch of road over some time period of length 𝑇. We could think of the time period of length 𝑇
𝑇
being split up into 𝑛 sub-intervals with each sub-interval being of length where 𝑛 is large. Suppose that
𝑛

these sub-intervals are so small that at most 1 breakdown occurs within this sub-interval with probability 𝑝,
different from zero.
Thus, we can state the following:
a) P(one breakdown occurring in a sub-interval) = 𝑝
b) P(no breakdown occurs in a sub-interval) = 1 − 𝑝
c) P(more than one breakdown occurring in the sub-interval) = 0
d) Total no. of breakdowns in the time period of interest = total no. of sub-intervals that contain one
breakdown
e) If breakdowns are regarded as independent from sub-interval to sub-interval, then the total number of
breakdowns can be described as following a Binomial distribution
f) There is no unique way of choosing 𝑛 and 𝑝 therefore, we can divide the week into a larger number of
sub-intervals (resulting in us having many shorter intervals) so that the probability, 𝑝 of a breakdown
occurring in a shorter sub-interval is much smaller

5.2
Proposition: Suppose that in the binomial pmf 𝑏(𝑥; 𝑛, 𝑝), we let 𝑛 → ∞ and 𝑝 → 0 in such a way that 𝑛𝑝
approaches 𝜆 > 0, then taking lim of the binomial probability function 𝑏(𝑥; 𝑛, 𝑝) = (𝑛𝑥)𝑝 𝑥 𝑞 𝑛−𝑥 will result
𝑛→∞
𝜆𝑥 𝑒 −𝜆
in the Poisson pmf .
𝑥!

Proof:
lim 𝑏(𝑥; 𝑛, 𝑝) = lim (𝑛𝑥)𝑝 𝑥 𝑞 𝑛−𝑥
𝑛→∞ 𝑛→∞

𝑛!
= lim 𝑝 𝑥 (1 − 𝑝)𝑛−𝑥
𝑛→∞ 𝑥!(𝑛−𝑥)!

𝑛! 𝜆 𝑥 𝜆 𝑛−𝑥
= lim (𝑛) (1 − 𝑛)
𝑛→∞ 𝑥!(𝑛−𝑥)!

𝑛×(𝑛−1)×…(𝑛−𝑥+1) 𝜆𝑥 𝜆 𝑛−𝑥
= lim (1 − 𝑛)
𝑛→∞ 𝑥! 𝑛𝑥

𝜆𝑥 𝑛×(𝑛−1)×…(𝑛−𝑥+1) 𝜆 𝑛−𝑥
= lim (1 − 𝑛)
𝑥! 𝑛→∞ 𝑛𝑥

𝜆𝑥 𝑛 (𝑛−1) (𝑛−𝑥+1) 𝜆 𝑛 𝜆 −𝑥
= 𝑥! lim ∙ ∙ …∙ (1 − 𝑛) ∙ (1 − 𝑛)
𝑛→∞ 𝑛 𝑛 𝑛

𝜆𝑥 𝑒 −𝜆 𝜆
= since lim → 0,
𝑥! 𝑛→∞ 𝑛

𝜆 𝑛
lim (1 − 𝑛) → 𝑒 −𝜆 *** https://www.youtube.com/watch?v=vxQu9oAmoR0
𝑛→∞

𝜆 −𝑥
lim (1 − ) → 1
𝑛→∞ 𝑛
and
𝑛 (𝑛 − 1) (𝑛 − 𝑥 + 1)
lim ∙ ∙ …∙ →1
𝑛→∞ 𝑛 𝑛 𝑛

Thus, we have shown that when 𝑛 is large, and 𝑝 is small and 𝑛𝑝 is constant, then
𝜆𝑥 𝑒 −𝜆
lim 𝑏(𝑥; 𝑛, 𝑝) = lim (𝑛𝑥)𝑝 𝑥 𝑞 𝑛−𝑥 = = 𝑝(𝑥; 𝜆) ∎
𝑛→∞ 𝑛→∞ 𝑥!

Note: When approximating Binomial probabilities for large n and small p using the Poisson frequency
function, as a rule of thumb, the approximation can be safely applied for 𝑛 > 50 and 𝑛𝑝 < 5.

5.3
Example 5.6.1.1
Consider the following combinations of 𝑛 and 𝑝 under the constraint that 𝑛𝑝 = 1 and the calculations of
𝑃(𝑋 = 𝑥) = (𝑛𝑥)𝑝 𝑥 𝑞 𝑛−𝑥 for the Binomial distribution.
Binomial
𝑛 𝑝 𝑋=0 𝑋=1 𝑋=2 𝑋=3 𝑋=4 𝑋=5 𝑋=6 𝑋=7 𝑋=⋯
5 0.2 0.3277 0.4096 0.2048 0.0512 0.0064 0.0003
10 0.1 0.3487 0.3874 0.1937 0.0574 0.0112 0.0015 0.0001 0.0000
20 0.05 0.3585 0.3774 0.1887 0.0596 0.0133 0.0022 0.0003 0.0000
100 0.01 0.3660 0.3697 0.1849 0.0610 0.0149 0.0029 0.0005 0.0001
1000 0.001 0.3677 0.3681 0.1840 0.0613 0.0153 0.0030 0.0005 0.0001
10000 0.0001 0.3679 0.3679 0.1839 0.0613 0.0153 0.0031 0.0005 0.0001
100000 0.00001 0.3679 0.3679 0.1839 0.0613 0.0153 0.0031 0.0005 0.0001

Notice that the probabilities that 𝑋 = 0, 1, 2, 3, 4, . .. approach the values 0.3679, 0.3679 0.1839, 0.0613 . . .
respectively as 𝑛 increases (and 𝑝 decreases).
Students are to insert calculations by hand here.

𝜆𝑥 𝑒 −𝜆
Compare the above probabilities to those below calculated using the Poisson distribution 𝑃(𝑋 = 𝑥)= 𝑥!

with 𝜆 = 𝑛𝑝 = 1
Poisson
𝑋=0 𝑋=1 𝑋=2 𝑋=3 𝑋=4 𝑋=5 𝑋=6 𝑋=7 𝑋=⋯
0.3679 0.3679 0.1839 0.0613 0.0153 0.0031 0.0005 0.0001

What do you notice?

5.4
5.6.2 Poisson probability distribution with its parameters
Definition 5.6.2.1 A random variable 𝑋 follows a Poisson distribution with parameter lambda 𝜆 , (𝜆 > 0) if
the pmf of X is
𝑒 −𝜆 𝜆𝑥
𝑝(𝑥; 𝜆) = { 𝑥! 𝑥 = 0,1,2 …
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
where:
𝜆 : is the rate of occurrence in some period (either time, area, distance etc) and it is the mean number of
successes in a specific interval
𝑥 : number of successes
𝑒 : mathematical constant approximately equal to 2.718..

The Poisson distribution is characterised by the rate of occurrence, lambda (λ), and the pmf has been expressed
in a general form. As the Poisson distribution assumes a constant rate of occurrence, the rate for a different
period can be calculated directly from the given λ. For example, if λ = 2 per hour it follows that λ = 1 per half
hour, and λ = 10 per five hours, etc.

One can show that 𝑝(𝑥; 𝜆) is a pmf. The first property 𝑝(𝑥; 𝜆) > 0 is satisfied for all 𝑥 since 𝜆 is positive.
The second property that ∑∞
𝑥=0 𝑝(𝑥; 𝜆) = 1 can be shown as follows:

We know that the Maclaurin infinite series expansion of 𝑒 𝜆 is given by

∞
𝜆
𝜆1 𝜆2 𝜆𝑥
𝑒 = 1+ + +⋯= ∑
1! 2! 𝑥!
𝑥=0

Multiplying the first and last term in the above equation by 𝑒 −𝜆 gives:
∞
𝜆 −𝜆
𝜆𝑥 −𝜆
𝑒 (𝑒 )= 1=∑ 𝑒
𝑥!
𝑥=0

Example 5.6.2.1
Suppose that a person receives 18 text messages on average in a 12-hour day. Assume that the number of
text messages received can be modelled by a Poisson distribution.
1. Calculate the probability that the person gets exactly 9 text messages in a day
2. Calculate the probability that the person gets at most 3 text messages in a day
3. Calculate the probability that the person gets more than 2 text messages in a day
4. Calculate the probability that the person gets 2 text messages in 3 hours
5. Calculate the probability that the person gets at most 2 text messages in 2 hours

5.5
Example 5.6.2.2
An early application of the Poisson distribution dates back to Ladislaus Bortkiewicz, a Polish mathematician
who applied the Poisson distribution to a question about deaths resulting from being kicked by a horse in the
Prussian army. The number of deaths was recorded for 10 corps or troops of Prussian calvary over 20 years
(1875-1895) resulting in 200 corps-years of data. The number of deaths per year is recorded in the following
table. The first column gives the number of deaths per year. The second column tells us how many times that
number of deaths was observed e.g., in 22 of the corps years, there were two deaths observed. The third column
gives us the relative frequency after dividing the observed frequency by 200. The fourth column gives the
∑ 𝑥𝑖 𝑓 𝑖
Poisson probabilities calculated using the average (𝜆) obtained from = 0.61.
𝑛

One can see that the relative frequencies and Poisson probabilities - after calculating probabilities using
𝜆𝑥 𝑒 −𝜆
, are quite close to each other.
𝑥!

Number of deaths Observed frequency Relative Frequency Poisson Probability

0 109 0.545 0.543

1 65 0.325 0.331
2 22 0.110 0.101
3 3 0.015 0.021
4 1 0.005 0.003

Example 5.6.2.3
There are two entrances to a parking lot. Cars arrive independently at both entrances according to a Poisson
distribution. At entrance I, they arrive at a rate of 4 per 5 minutes and at entrance II, they arrive at a rate of 3
per 5 minutes. What is the probability that 3 cars will arrive at the parking lot in 5 minutes?

Insert working here

Note If 𝑋~𝑃(𝜆)and 𝑌~𝑃(𝜇) and 𝑋 and 𝑌are independent then 𝑋 + 𝑌~𝑃(𝜆 + µ) - We will discuss this in
more detail after we look at joint random variables in the next chapter

5.6
5.6.3 Calculating Poisson probabilities in R
The above example’s calculations can be done in R by using the probability distribution function prefixed
with a ‘d’ for density:
Recalling the work covered in the Binomial distribution, let us consider the calculation of the binomial
probabilities in R for the following cases as 𝑛 increases and 𝑝 decreases.
𝑛 = 10 and 𝑝 = 0.1,
𝑛 = 100 and 𝑝 = 0.01
𝑛 = 10000 and 𝑝 = 0.0001

#Calculate the binomial probabilities for 𝑋 = 0 to 𝑋 = 7 (we are stopping at 7 by choice)

when 𝑛 = 10 and 𝑝 = 0.1
> dbinom(0:7,10,0.1)

This results in the output below:

[1] 0.348678440 0.387420489 0.193710244 0.057395628 0.011160261 0.001488035
0.000137781 0.000008748

#Calculate the binomial probabilities for 𝑋 = 0 to 𝑋 = 7 (we are stopping at 7 by choice)

when 𝑛 = 100 and 𝑝 = 0.01
> dbinom(0:7,100,0.01)

This results in the output below:

[1] 3.660323e-01 3.697296e-01 1.848648e-01 6.099917e-02 1.494171e-02 2.897787e-03
4.634508e-04 6.286346e-05

#Calculate the binomial probabilities for 𝑋 = 0 to 𝑋 = 7 (we are stopping at 7 by choice)

when 𝑛 = 10 000 and 𝑝 = 0.0001
> dbinom(0:7,10000,0.0001)

This results in the output below:

[1] 3.678610e-01 3.678978e-01 1.839489e-01 6.131017e-02 1.532448e-02 3.063976e-03
5.104584e-04 7.288616e-05

Now, we could calculate the Poisson probability for 𝜆 = 𝑛𝑝 = 1 and compare these probabilities with those
calculated for the Binomial distribution above.

5.7
To calculate the Poisson probabilities, we require the following syntax:
𝑑𝑝𝑜𝑖𝑠(x, λ, 𝑙𝑜𝑔)
The function 𝑑𝑝𝑜𝑖𝑠() creates the probability mass function by calculating the probability of a random
variable that is available within a certain range.

Syntax:
dpois(x, λ, 𝑙𝑜𝑔)

where,
x: number of successful events happened in an interval
𝜆: mean per interval
log: If TRUE then the function returns probability in form of log
*can exclude the log part as well

Calculation of the Poisson probabilities in R

Doing this in R gives us:
#Calculate the Poisson probabilities for 𝑋 = 0 to 𝑋 = 7 (we are stopping at 7 by choice) for 𝜆 = 𝑛𝑝 = 1
> dpois(0:7,1)

This results in the output below:

[1] 3.678794e-01 3.678794e-01 1.839397e-01 6.131324e-02 1.532831e-02 3.065662e-03 5.10
9437e-04 7.299195e-05

Comparing these probabilities, one can see that the Poisson approximation is very close to the exact Binomial
probabilities.

We can also calculate cumulative probabilities using the following syntax:

𝑝𝑝𝑜𝑖𝑠(x, λ, lower. tail, 𝑙𝑜𝑔)
This function is used for the cumulative probability function in an R plot. 𝑝𝑝𝑜𝑖𝑠(x, λ, lower. tail =
TRUE, 𝑙𝑜𝑔) calculates the probability of a random variable that will be equal to or less than a number.

Syntax:
ppois(x, λ, lower. tail, 𝑙𝑜𝑔)

where,
x: number of successful events that happened in an interval
𝜆: mean per interval
lower.tail: If TRUE then the left tail is considered otherwise if the FALSE right tail is considered
log: If TRUE then the function returns probability in the form of log

5.8
Answering Example 5.6.2.1 using R can be done as follows:

1. Calculate the probability that the person gets exactly 9 text messages in a day
𝑥 = 9; 𝜆 = 18
R code:
# Calculates P(X=x)
> dpois(9,18) # Or dpois(9,18,log=FALSE)

Output:
[1] 0.008325088

2. Calculate the probability that the person gets at most 3 text messages in a day
𝑥 = 0, 1, 2, 3; 𝜆 = 18

R code:
# Calculates P(X≤x)
> ppois(3,18)

Output:
[1] 1.756017e-05

Note:

3. Calculate the probability that the person gets more than 2 text messages in a day
𝑥 = 2 𝑜𝑟 𝑚𝑜𝑟𝑒; 𝜆 = 18
𝑃(𝑋 > 2) = 1 − 𝑃(𝑋 ≤ 2)

R code:
> 1-ppois(2,18)

Output:
[1] 0.9999972

4. Calculate the probability that the person gets 2 text messages in 3 hours

𝑥 = 2; 𝜆 = 18 𝑝𝑒𝑟 12 ℎ𝑜𝑢𝑟 𝑑𝑎𝑦 𝑠𝑜 𝜆 =? 𝑝𝑒𝑟 3 ℎ𝑜𝑢𝑟𝑠; 𝑠𝑜 𝜆 = 4.5

R code:
> dpois(2,4.5)

Output:
[1] 0.1124786

5. Calculate the probability that the person gets at most 2 text messages in 2 hours
𝑃(𝑋 ≤ 2); 𝜆 = 18 𝑝𝑒𝑟 12 ℎ𝑜𝑢𝑟 𝑑𝑎𝑦 𝑠𝑜 𝜆 =? 𝑝𝑒𝑟 2 ℎ𝑜𝑢𝑟𝑠; 𝑠𝑜 𝜆 = 3
R code:
> ppois(2,3)

Output:
[1] 0.4231901

5.9
Homework Example – do calculations manually and code your answer in R
Suppose that the number of customers arriving at the bank follows a Poisson distribution with an average of
𝜆 = 10 per hour. Find
i) 𝑃(more than 2 customers arrive per hour) – answer = 0.9972
ii) 𝑃(at least 2 customers arrive per hour) – answer = 0.9995
iii) 𝑃(6 customers arriving in 24 minutes) – answer = 0.1042

5.10
5.6.4 Derivation of the Expectation and Variance
The first and second moments, the mean and variance, describe the central tendency and spread measures,
respectively.
If 𝑋 is a discrete random variable that follows a Poisson distribution with parameter 𝜆 then
i) Mean =𝐸[𝑋] = 𝜆
ii) Variance = 𝑉𝑎𝑟(𝑋) = 𝜆
Students to do (ii) as a homework exercise using the property that 𝑉𝑎𝑟(𝑋) = 𝐸[𝑋 2 ] − [𝐸(𝑋)]2 and by
expressing 𝐸[𝑋 2 ] by 𝐸[𝑋(𝑋 − 1)] + 𝐸[𝑋]

i) Proof to derive the mean or expectation = E(𝑿) = 𝝀

∞

𝐸[𝑋] = ∑ 𝑥 𝑝(𝑥)
𝑥=0

∞
𝜆𝑥 𝑒 −𝜆
= ∑𝑥
𝑥!
𝑥=0

∞
𝜆𝑥 𝑒 −𝜆
= ∑𝑥
𝑥!
𝑥=1

∞
𝜆𝑥 𝑒 −𝜆
=∑
(𝑥 − 1)!
𝑥=1

∞
𝜆𝑥−1 𝑒 −𝜆
= 𝜆 ∑
(𝑥 − 1)!
𝑥=1

Letting 𝑥 − 1 = 𝑧, gives us

∞
𝜆𝑧 𝑒 −𝜆
= 𝜆 ∑
𝑧!
𝑧=0

= 𝜆 , since

∞
𝜆𝑧 𝜆1 𝜆2
∑ = 1+ + + ⋯ = 𝑒𝜆
𝑧! 1! 2!
𝑧=0
5.11
ii) To derive the variance = 𝑽𝒂𝒓(𝑿) = 𝝀

Start with 𝑉𝑎𝑟(𝑋) = 𝐸[𝑋 2 ] − [𝐸(𝑋)]2

Rewrite 𝐸[𝑋 2 ] as 𝐸[𝑋(𝑋 − 1)] + 𝐸[𝑋]

Thus,`
𝑉𝑎𝑟(𝑋) = 𝐸[𝑋(𝑋 − 1)] + 𝐸[𝑋] −[𝐸(𝑋)]2
= 𝐸[𝑋(𝑋 − 1)] + 𝜆 − 𝜆2

Now,
𝐸[𝑋(𝑋 − 1)]
∞

= ∑ 𝑥 (𝑥 − 1) 𝑝(𝑥)
𝑥=0

∞
𝜆𝑥 𝑒 −𝜆
= ∑ 𝑥 (𝑥 − 1)
𝑥!
𝑥=0

∞
−𝜆
𝜆𝑥
=𝑒 ∑ 𝑥 (𝑥 − 1)
𝑥!
𝑥=0

When 𝑥 = 0 and 𝑥 = 1, the first two terms will be zero, thus we can write the above equation as
∞
−𝜆
𝜆𝑥
=𝑒 ∑ 𝑥 (𝑥 − 1)
𝑥!
𝑥=2

∞
−𝜆
𝜆𝑥
=𝑒 ∑
(𝑥 − 2)!
𝑥=2

∞
−𝜆
𝜆𝑥−2 ∙ 𝜆2
=𝑒 ∑
(𝑥 − 2)!
𝑥=2

∞
2 −𝜆
𝜆𝑥−2
=𝜆 𝑒 ∑
(𝑥 − 2)!
𝑥−2=0

Letting 𝑧 = 𝑥 − 2 gives
5.12
 ∞
2 −𝜆
𝜆𝑧
𝜆 𝑒 ∑
(𝑧)!
𝑧=0

= 𝜆2 𝑒 −𝜆 ∙ 𝑒 λ

=𝜆2

Since

∞
𝜆𝑧 𝜆1 𝜆2
∑ = 1+ + + ⋯ = 𝑒𝜆
𝑧! 1! 2!
𝑧=0

𝑉𝑎𝑟(𝑋) = 𝐸[𝑋(𝑋 − 1)] + 𝐸[𝑋] −[𝐸(𝑋)]2

= 𝐸[𝑋(𝑋 − 1)] + 𝜆 − 𝜆2
= 𝜆2 + 𝜆 − 𝜆2
=𝜆

5.13
A reminder of some mathematical results used in the proofs and derivations:

The exponential function 𝑒 𝑥 , where 𝑒 is a constant, is that function which has its own derivative. Thus
𝑑
(𝑒 𝑥 ) = 𝑒 𝑥 (1)
𝑑𝑥

This function can be expressed as a power series in 𝑥, with coefficients 𝑎0 , 𝑎1 , 𝑎2 , … ., in the form
𝑒 𝑥 = 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥 2 + 𝑎3 𝑥 3 + 𝑎4 𝑥 4 + ⋯ + 𝑎𝑛 𝑥 𝑛 + ⋯ (2)

This series can be differentiated term by term. Thus,

𝑑
(𝑒 𝑥 ) = 𝑎1 + 2𝑎2 𝑥 + 3𝑎3 𝑥 2 + 4𝑎4 𝑥 3 + ⋯ + 𝑛𝑎𝑛 𝑥 𝑛−1 + ⋯ (3)
𝑑𝑥

𝑑
But because 𝑑𝑥 (𝑒 𝑥 ) = 𝑒 𝑥 by definition, we have, on equating the coefficients of 𝑥 in (2) and (3) above that

𝑎0 = 𝑎1 𝑎1 = 2𝑎2 𝑎2 = 3𝑎3 𝑎3 = 4𝑎4 … 𝑎𝑛−1 = 𝑛𝑎𝑛

When 𝑥 = 0, the series is equal to 𝑎0 . But any number raised to the power 0 is one so 𝑎0 = 𝑒 0 = 1.
Hence
𝑎1 = 𝑎0 = 1

1 1 1 1
𝑎2 = 2 𝑎1 = ×1= = 2! since 𝑎1 = 2𝑎2
2 2

1 1 1 1
𝑎3 = 3 𝑎2 = 3 × 2 = since 𝑎2 = 3𝑎3
3!

1 1 1 1 1
𝑎4 = 4 𝑎3 = 4 × 3 × 2 = since 𝑎3 = 4𝑎3
4!

.
.
.
1
𝑎𝑛 = 𝑛!

Thus,

5.14
Theorem A1: the series expansion of the exponential function 𝑒 𝑥 is
𝑥2 𝑥3 𝑥𝑟
1+𝑥+ + + ⋯.+ + ⋯.
2! 3! 𝑟!
and this is valid for all values of 𝑥.

The constant
12 13 1𝑟
𝑒 = 1+1+ + + ⋯.+ + ⋯.
2! 3! 𝑟!
This is an irrational number which equals approximately 2.718

Theorem A2:
𝑥 𝑛
lim (1 + 𝑛) = 𝑒 𝑥
𝑛→∞

Proof:
By the Binomial series expansion,
𝑛 𝑛 𝑛
(𝑎 + 𝑏)𝑛 = 𝑎𝑛 + ( ) 𝑎𝑛−1 𝑏 + ( ) 𝑎𝑛−2 𝑏 2 +….+ ( ) 𝑎𝑛−1 + 𝑏 𝑛
1 2 𝑛−1

we have (for 𝑛 an integer)

𝑥 𝑛 𝑥 𝑛(𝑛−1) 𝑥2 𝑛(𝑛−1)(𝑛−2) 𝑥3 𝑛(𝑛−1)…(𝑛−𝑟+1) 𝑥𝑟
(1 + 𝑛) = 1 + (𝑛 × 𝑛 ) + ( × 𝑛2 )+ ( × 𝑛3 ) + ⋯ + ( × 𝑛𝑟 ) + ⋯
2! 3! 𝑟!
𝑥2 1 𝑥3 1 2 𝑥𝑟 1 2 𝑟−1
=1+𝑥+ (1 − )+ (1 − ) (1 − ) + ⋯ + (1 − ) (1 − ) … (1 − )+⋯
2! 𝑛 3! 𝑛 𝑛 𝑟! 𝑛 𝑛 𝑛

As 𝑛 → ∞
1 2 𝑟−1
(1 − 𝑛) → 1, (1 − 𝑛) → 1,…, (1 − )→1
𝑛

Hence
𝑥 𝑛 𝑥2 𝑥3 𝑥𝑟
lim (1 + 𝑛) = 1 + 𝑥 + (1) + (1.1) + ⋯ + (1.1.1 … 1) + ⋯
𝑛→∞ 2! 3! 𝑟!

𝑥2 𝑥3 𝑥𝑟
=1 + 𝑥 + + + ⋯+ + ⋯ = 𝑒𝑥
2! 3! 𝑟!

By replacing 𝑥 by −𝑥 in the series expansion, we have

𝑥2 𝑥3 𝑥4 𝑥𝑛
𝑒 −𝑥 = 1 − 𝑥 + − + − ⋯ + (−1)𝑛 + ⋯,
2! 3! 4! 𝑛!

Thus
𝑥 𝑛
lim (1 − ) = 𝑒 −𝑥
𝑛→∞ 𝑛
5.15