1
Let X represent a Binomial r.v as in (3-42). Then from (2-30)
Since the binomial coefficient grows quite rapidly with n, it is difficult to compute (4-1) for large n. In this context, two approximations are extremely useful.
4.1 The Normal Approximation (Demoivre-Laplace Theorem) Suppose with p held fixed. Then for k in the neighborhood of np, we can approximate
(4-1)
PILLAI
4. Binomial Random Variable Approximations,
Conditional Probability Density Functions
and Stirling’s Formula
2
(4-2)
Thus if and in (4-1) are within or around the neighborhood of the interval we can approximate the summation in (4-1) by an integration. In that case (4-1) reduces to
where
We can express (4-3) in terms of the normalized integral
that has been tabulated extensively (See Table 4.1).
(4-3)
(4-4)
PILLAI
3
For example, if and are both positive ,we obtain
Example 4.1: A fair coin is tossed 5,000 times. Find the probability that the number of heads is between 2,475 to 2,525. Solution: We need Here n is large so that we can use the normal approximation. In this case so that and Since and the approximation is valid for and Thus
Here
(4-5)
PILLAI
4
Table 4.1
PILLAI
5
Since from Fig. 4.1(b), the above probability is given by
where we have used Table 4.1
4.2. The Poisson Approximation As we have mentioned earlier, for large n, the Gaussian approximation of a binomial r.v is valid only if p is fixed, i.e., only if and what if np is small, or if it does not increase with n?
Fig. 4.1
(a)
(b)
PILLAI
6
Obviously that is the case if, for example, as such that is a fixed number.
Many random phenomena in nature in fact follow this pattern. Total number of calls on a telephone line, claims in an insurance company etc. tend to follow this type of behavior. Consider random arrivals such as telephone calls over a line. Let n represent the total number of calls in the interval From our experience, as we have so that we may assume Consider a small interval of duration Δ as in Fig. 4.2. If there is only a single call coming in, the probability p of that single call occurring in that interval must depend on its relative size with respect to T.
Fig. 4.2
PILLAI
7
Hence we may assume Note that as However in this case is a constant, and the normal approximation is invalid here.
Suppose the interval Δ in Fig. 4.2 is of interest to us. A call inside that interval is a “success” (H), whereas one outside is a “failure” (T ). This is equivalent to the coin tossing situation, and hence the probability of obtaining k calls (in any order) in an interval of duration Δ is given by the binomial p.m.f. Thus
and here as such that It is easy to obtain an excellent approximation to (4-6) in that situation. To see this, rewrite (4-6) as
(4-6)
PILLAI
8
(4-7)
(4-8)
since the finite products as well as tend to unity as and
The right side of (4-8) represents the Poisson p.m.f and the Poisson approximation to the binomial r.v is valid in situations where the binomial r.v parameters n and p diverge to two extremes such that their product np is a constant.
Thus
PILLAI
9
Example 4.2: Winning a Lottery: Suppose two million lottery tickets are issued with 100 winning tickets among them. (a) If a person purchases 100 tickets, what is the probability of winning? (b) How many tickets should one buy to be 95% confident of having a winning ticket? Solution: The probability of buying a winning ticket
Here and the number of winning tickets X in the n purchased tickets has an approximate Poisson distribution with parameter Thus
and (a) Probability of winning
PILLAI
10
(b) In this case we need
But or Thus one needs to buy about 60,000 tickets to be 95% confident of having a winning ticket!
Example 4.3: A space craft has 100,000 components The probability of any one component being defective is The mission will be in danger if five or more components become defective. Find the probability of such an event. Solution: Here n is large and p is small, and hence Poisson approximation is valid. Thus and the desired probability is given by
PILLAI
11
Conditional Probability Density Function
For any two events A and B, we have defined the conditional probability of A given B as
Noting that the probability distribution function is given by
we may define the conditional distribution of the r.v X given the event B as
(4-9)
(4-10)
PILLAI
12
(4-11)
Thus the definition of the conditional distribution depends on conditional probability, and since it obeys all probability axioms, it follows that the conditional distribution has the same properties as any distribution function. In particular
Further
(4-12)
(4-13)
PILLAI
13
Since for
The conditional density function is the derivative of the conditional distribution function. Thus
and proceeding as in (3-26) we obtain
Using (4-16), we can also rewrite (4-13) as
(4-14)
(4-15)
(4-16)
(4-17)
PILLAI
14
Fig. 4.3
(a)
1
1
(b)
1
1
Example 4.4: Refer to example 3.2. Toss a coin and X(T)=0, X(H)=1. Suppose Determine
Solution: From Example 3.2, has the following form. We need for all x.
For so that and
PILLAI
15
For so that
For and
(see Fig. 4.3(b)).
Example 4.5: Given suppose Find Solution: We will first determine From (4-11) and B as given above, we have
(4-18)
PILLAI
16
(4-19)
(4-20)
(4-21)
For so that
For so that Thus
and hence
PILLAI
17
(a)
Fig. 4.4
Example 4.6: Let B represent the event with For a given determine and Solution:
For we have and hence
(4-22)
(4-23)
(b)
PILLAI
18
For we have and hence
For we have so that Using (4-23)-(4-25), we get (see Fig. 4.5)
(4-24)
(4-25)
(4-26)
Fig. 4.5
PILLAI
19
We can use the conditional p.d.f together with the Bayes’ theorem to update our a-priori knowledge about the probability of events in presence of new observations. Ideally, any new information should be used to update our knowledge. As we see in the next example, conditional p.d.f together with Bayes’ theorem allow systematic updating. For any two events A and B, Bayes’ theorem gives
Let so that (4-27) becomes (see (4-13) and (4-17))
(4-27)
(4-28)
PILLAI
20
Further, let so that in the limit as
or
From (4-30), we also get
or
and using this in (4-30), we get the desired result
(4-29)
(4-30)
(4-31)
(4-32)
(4-33)
PILLAI
21
To illustrate the usefulness of this formulation, let us reexamine the coin tossing problem.
Example 4.7: Let represent the probability of obtaining a head in a toss. For a given coin, a-priori p can possess any value in the interval (0,1). In the absence of any additional information, we may assume the a-priori p.d.f to be a uniform distribution in that interval. Now suppose we actually perform an experiment of tossing the coin n times, and k heads are observed. This is new information. How can we update Solution: Let A= “k heads in n specific tosses”. Since these tosses result in a specific sequence,
Fig.4.6
(4-34)
PILLAI
22
and using (4-32) we get
The a-posteriori p.d.f represents the updated information given the event A, and from (4-30)
Notice that the a-posteriori p.d.f of p in (4-36) is not a uniform distribution, but a beta distribution. We can use this a-posteriori p.d.f to make further predictions, For example, in the light of the above experiment, what can we say about the probability of a head occurring in the next (n+1)th toss?
(4-35)
(4-36)
Fig. 4.7
~
PILLAI
23
Let B= “head occurring in the (n+1)th toss, given that k heads have occurred in n previous tosses”. Clearly and from (4-32)
Notice that unlike (4-32), we have used the a-posteriori p.d.f in (4-37) to reflect our knowledge about the experiment already performed. Using (4-36) in (4-37), we get
Thus, if n =10, and k = 6, then
which is more realistic compare to p = 0.5.
(4-37)
(4-38)
PILLAI
24
To summarize, if the probability of an event X is unknown, one should make noncommittal judgement about its a-priori probability density function Usually the uniform distribution is a reasonable assumption in the absence of any other information. Then experimental results (A) are obtained, and out knowledge about X must be updated reflecting this new information. Bayes’ rule helps to obtain the a-posteriori p.d.f of X given A. From that point on, this a-posteriori p.d.f should be used to make further predictions and calculations.
PILLAI
25
Stirling’s Formula : What is it?
Stirling’s formula gives an accurate approximation for n!
as follows:
in the sense that the ratio of the two sides in (4-39) is near
to one; i.e., their relative error is small, or the percentage
error decreases steadily as n increases. The approximation
is remarkably accurate even for small n. Thus 1! = 1 is
approximated as and is
approximated as 5.836.
Prior to Stirling’s work, DeMoivre had established
the same formula in (4-39) in connection with binomial
distributions in probability theory. However DeMoivre
did not establish the constant
(4-39)
PILLAI
26
term in (4-39); that was done by James Stirling
( 1730).
How to prove it?
We start with a simple observation: The function
log x is a monotone increasing function, and hence we have
Summing over
we get
or
(4-40)
PILLAI
27
The double inequality in (4-40) clearly suggests that log n!
is close to the arithmetic mean of the two extreme numbers
there. However the actual arithmetic mean is complicated
and it involves several terms. Since (n + )log n – n is
quite close to the above arithmetic mean, we consider
the difference1
This gives
1According to W. Feller this clever idea to use the approximate mean (n + )log n – n is due
to H.E. Robbins, and it leads to an elementary proof.
(4-41)
PILLAI
28
PILLAI
(4-42)
Thus {an} is a monotone decreasing sequence and let c
represent its limit, i.e.,
From (4-41), as this is equivalent to
To find the constant term c in (4-44), we can make use of
a formula due to Wallis ( 1655).
1By Taylor series expansion
(4-44)
(4-43)
Hence1
29
PILLAI
The well known function
goes to zero at
moreover these are the only
zeros of this function. Also
has no finite poles.
(All poles are at infinity). As a result we can write
or [for a proof of this formula, see chapter 4 of Dienes,
The Taylor Series]
which for gives the Wallis’ formula
30
Thus as this gives
Thus as
But from (4-41) and (4-43)
and hence letting in (4-45) and making use
(4-45)
(4-46)
PILLAI
or
31
PILLAI
of (4-46) we get
which gives
With (4-47) in (4-44) we obtain (4-39), and this proves
the Stirling’s formula.
Upper and Lower Bounds
It is possible to obtain reasonably good upper and
lower bounds for n! by elementary reasoning as well.
To see this, note that from (4-42) we get
so that {an – 1/12n} is a monotonically increasing
(4-47)
32
PILLAI
sequence whose limit is also c. Hence for any finite n
and together with (4-41) and (4-47) this gives
Similarly from (4-42) we also have
so that {an – 1/(12n+1)} is a monotone decreasing sequence
whose limit also equals c. Hence
or
Together with (4-48)-(4-49) we obtain
(4-48)
(4-49)
33
PILLAI
Stirling’s formula also follows from the asymptotic
expansion for the Gamma function given by
Together with the above expansion can
be used to compute numerical values for real x.
For a derivation of (4-51), one may look into Chapter 2
of the classic text by Whittaker and Watson (Modern
Analysis).
We can use Stirling’s formula to obtain yet another
approximation to the binomial probability mass
(4-50)
(4-51)
34
function. Since
using (4-50) on the right side of (4-52) we obtain
and
where
and
Notice that the constants c1 and c2 are quite close
to each other.
(4-52)
PILLAI