Sie sind auf Seite 1von 12

Exam P CONTENTS

Exam P
Transformations and Order Statistics
of Continuous Random Variables
©2013 Yeng M. Chang (2015 update)

From my experience in tutoring for exam P/1, it seems that these topics consistently are the most
confusing topics for students. This document is meant to try to make these topics as clear as possible
for students as they study for this exam.
I assume that the reader knows everything about expected values, probability density functions,
(cumulative) distribution functions, and independence of continuous random variables in the exam
P syllabus. Although it is not very well emphasized in later exams, these two topics are very impor-
tant in exams MLC and C. Using the concepts presented here as it pertains to the material on these
courses can help further understanding of the material and even speed up calculations (especially
on exam MLC on multi-life models).
Unless stated otherwise, assume all random variables in this document are continuous random vari-
ables.
We go by the convention that if we write an equation for fX within a set of intervals, fX is 0
elsewhere. Similarly, if we write an equation for FX in an interval [a, b], for example, then FX (x) =
0 for x < a and FX (x) = 1 for x > b.
Numbered equations are essential for understanding the topics.

Contents
1 Transformations 2
1.1 The Cumulative Distribution Function (CDF) Method . . . . . . . . . . . . . . . . 2
1.2 Method of Transformations (Univariate) . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Method of Jacobians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Order Statistics 11
2.1 Minimum of a Set of Random Variables . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Maximum of a Set of Random Variables . . . . . . . . . . . . . . . . . . . . . . . 12

©2013 Yeng M. Chang (2015 update) 1


Exam P

1 Transformations
Let X be a random variable (either discrete or continuous). A transformation of X is a function of
X, which is also a random variable. We will also refer to any such transformation of X as simply
a “transformation.” For example, Y = 3X, Z = 3X + 4, W = |X| are all transformations of
X and are also random variables. We are specifically interested in finding the probability density
functions (PDFs) of transformations of continuous random variables.

1.1 The Cumulative Distribution Function (CDF) Method


Let X be a continuous random variable. Recall by definition that the (probability) density function
of X, denoted fX , is given by
d
fX (x) = [FX (x)] (1.1.1)
dx
where
FX (x) = P (X ≤ x) , (1.1.2)
which FX is known as the (cumulative) distribution function of X.
Now consider Y = g(X), where g is a continuous function. Then Y is a continuous random vari-
able as well, and we can use (1.1.1) and (1.1.2) to find the the PDF of Y , assuming we have the
distribution function of X. This method of finding the density function of Y is known as the CDF
method.

How to do the CDF Method for a Transformation Y = g(X):

1. Find the domain of X.

2. Find FX .

3. Find the domain of Y .

4. Write FY so that you get a function of FX .

5. Take the derivative of FY to get fY .

Example 1.1.1. Suppose X follows the exponential distribution with mean 2.

1. Then X > 0 and FX (x) = 1 − e−x/2 for x > 0.

Suppose Y = 5X + 4.

2. Then Y > 4 since Y = 5X + 4 > 5(0) + 4 = 4 (since X > 0).

3. We have
( ) ( )
y−4 y−4
FY (y) = P (5X + 4 ≤ y) = P X ≤ = FX = 1 − e−(y−4)/10 .
5 5

2 ©2013 Yeng M. Chang (2015 update)


Exam P 1.1 The Cumulative Distribution Function (CDF) Method

4. Therefore,
( )
d [ ] −1 1 −(y−4)/10
fY (y) = 1 − e−(y−4)/10 = −e−(y−4)/10 = e , y > 4. 
dy 10 10

Example 1.1.2. Suppose X follows the exponential distribution with mean 1, with

FX (x) = 1 − e−x , x > 0.

Consider W = X 2 , which has domain W = X 2 > 02 = 0.


For a fixed value w > 0, consider the set of values of X such that W = X 2 ≤ w.

W = X2

W ≤w

√ X
0 w

Looking at the graph above, if X 2 ≤ w, then 0 < X ≤ w (notice that X > 0 by definition!).
Thus,
( )
FW (w) = P (W ≤ w) = P X 2 ≤ w
( √ )
=P X≤ w
(√ )
= FX w

= 1 − e− w
, w > 0.

This gives
d [ √ ] √ −1 1 √
fW (w) = 1 − e− w = −e− w √ = √ e− w , w > 0. 
dw 2 w 2 w

Example 1.1.3. Suppose X follows the uniform distribution in the interval (−5, 5) and W = X 2 .
Then
x+5
FX (x) = , − 5 < x < 5.
10
It can be shown that 0 ≤ W < 25. Now for a fixed value 0 ≤ w < 25, consider the set of all values
of X such that W = X 2 ≤ w.

©2013 Yeng M. Chang (2015 update) 3


1.1 The Cumulative Distribution Function (CDF) Method Exam P

W = X2

W ≤w

√ √ X
− w 0 w

√ values of X, as opposed to Example 1.1.2. Looking


Notice here that we have to work√with negative
at this graph, if W ≤ w, then − w ≤ X ≤ w. So,

FW (w) = P (W ≤ w)
( √ √ )
=P − w≤X≤ w
( √ ) ( √ )
=P X ≤ w −P X ≤− w
√ √
w+5 − w+5
= −
√10 10
2 w
= , 0 ≤ w < 25.
10
Therefore, ( )
2 1 1
fW (w) = √ = √ , 0 < w < 25. 
10 2 w 10 w

Exercise 1.1.4. Repeat Example 1.1.3, except set W = |X|. Show that W follows a uniform
distribution in the interval [0, 5).

Exercise 1.1.5. Suppose X follows an exponential distribution with mean 2. Let Y = 3X + 5.


Show that
1
fY (y) = e−(y−5)/6 , y > 5.
6

Exercise 1.1.6 (Discrete Random Variables). Suppose X is a discrete random variable where

pX (x) = P (X = x) = p(1 − p)x−1 , 0 < p < 1, x = 1, 2, . . . .

Let W = X 2 . Show that



pW (w) = p(1 − p) w−1
, 0 < p < 1, w = 1, 4, . . . .

4 ©2013 Yeng M. Chang (2015 update)


Exam P 1.1 The Cumulative Distribution Function (CDF) Method

Notice that there is no need to use the CDF at all in the discrete case to get the probability mass
function of W . Furthermore, verify that

pW (w) = 1.
all w

You may use, without proof, that ∑


pX (x) = 1.
all x

Exercise 1.1.7 (Challenging). Suppose X follows the exponential distribution with mean 1, and
{
X 2, 0 < X ≤ 2
Y = √
X, X > 2.
The graph of Y is below.
Y

4 Y = X2


Y = X

2

X
0 2

Show that
 √ √
 − y
 1 − e , 0<y≤ 2
√ √ √
FY (y) = g(y) = 1 − e− 2 + e−2 − e− y + e−2 − e−y , 2<y≤4
1/4 2



g(4) + e−16 − e−y
2
y > 4.
(Thanks to the Math StackExchange for assistance.)
Exercise 1.1.8. Suppose Z follows the standard normal distribution, so that
1
fZ (z) = √ e−z /2 ,
2
− ∞ < z < ∞.

The symbol Φ denotes the CDF of Z. Consider the transformation Y = ceil(Z). Y is known as the
ceiling of Z; for every non-integer value of Z, Y is the rounded-up value to the nearest integer. If
Z is an integer, its value remains the same. For example, if {−4.7, −3, 1, 5.3} are values of Z, then
{−4, −3, 1, 6} are the corresponding values of Y .
Show that the probability mass function of Y , pY (y) ≡ Φ(y) − Φ(y − 1) for any integer y. Fur-
thermore, using Φ(2) = 0.9772, find the probability that Y = 2.

©2013 Yeng M. Chang (2015 update) 5


1.2 Method of Transformations (Univariate) Exam P

Exercise 1.1.9. Consider what is given in Exercise 1.1.8. Let W = floor(Z) be the floor of Z; for
every non-integer value of Z, W is the rounded-down value to the nearest integer. If Z is an integer,
its value remains the same. For example, if {−4.7, −3, 1, 5.3} are values of Z, then {−5, −3, 1, 5}
are the corresponding values of W .
Using Φ(1) = 0.8413, find the probability that W = 1.

1.2 Method of Transformations (Univariate)


Let X be a (continuous) random variable, and Y = g(X) be a monotonic (strictly increasing or
decreasing) function of X. Then g has an inverse (say g −1 ) and when fY exists,

( −1 ) d [ −1 ]
fY (y) = fX g (y) · g (y) .
dy
(Why I have decided not to number this equation will be explained soon.)
Proof. Suppose Y = g(X) is strictly increasing. Then
FY (y) = P (Y ≤ y) = P (g(X) ≤ y) .

Y = g(X)

y = g(x)

g(X) ≤ y
Y = g −1 (X)

X
x = g −1 (y)

Remember: our goal is to write P (g(X) ≤ y) in terms of FX . If g(X) ≤ y, what does this say
about the values of X? If you look at the graph above, if g(X) ≤ y, then X ≤ g −1 (y). This is
because all of the values of X in the shaded region above are those which are less than x = g −1 (y),
so that X ≤ x = g −1 (y). Hence
( ) ( )
FY (y) = P (g(X) ≤ y) = P X ≤ g −1 (y) = FX g −1 (y) .
Taking derivatives on both sides, we have
( ) d [ −1 ]
fY (y) = fX g −1 (y) · g (y) ,
dy
d −1
and [g (y)] > 0 since g −1 is increasing, so we don’t need the absolute value symbols.
dy

6 ©2013 Yeng M. Chang (2015 update)


Exam P 1.2 Method of Transformations (Univariate)

Now suppose Y = h(X) is strictly decreasing. Then, as before, FY (y) = P (g (X) ≤ y).

Y = g(X)
y = g(x)

g(
X
)≤
y
Y = g −1 (X)

X
x = g −1 (y)

The set of all values where g(X) ≤ y are where X ≥ g −1 (y), so


( ) ( )
FY (y) = P (g (X) ≤ y) = P X ≥ g −1 (y) = 1 − FX g −1 (y) .

Taking derivatives on both sides, we have

( ) d [ −1 ] ( ) d [ −1 ]
fY (y) = −fX g −1 (y) · g (y) = fX g −1 (y) · − g (y) ,
dy dy

d −1 d
but notice that [g (y)] < 0 (as you can see in the graph above). So − [g −1 (y)] > 0. Thus,
dy dy
in both cases (i.e., when Y is either increasing or decreasing), we can write

( −1 ) d [ −1 ]
fY (y) = fX g (y) · g (y) .
dy

Now: an explanation on why I didn’t number the equation previously. Notice that if Y = g(X), we
can say that X = g −1 (Y ). So, in fact, you don’t need the g in the notation at all. I prefer writing
instead Y = Y (X) and X = X(Y ) (with the understanding that X = Y −1 on the functions),
and then switching to lowercase letters when referring to particular values of X and Y . By writing
Y (X) and X(Y ), it makes it clear (in my opinion) that you want Y in terms of X, and X in terms
of Y respectively. So, I prefer the equation written as

d
fY (y) = fX (x(y)) · [x(y)] .
(1.2.1)
dy

where x(y) means “x in terms of y.”

©2013 Yeng M. Chang (2015 update) 7


1.2 Method of Transformations (Univariate) Exam P

How to do the Method of Transformations:

1. Find the domain of X.

2. Solve for X with respect to the transformation Y . If one cannot


find an equation for Y in the domain of X, then the method of
transformations cannot be applied.

3. Find the domain of Y .

4. Find fX .

5. Apply the formula



d

fY (y) = fX (x(y)) · [x(y)] .
dy

Example 1.2.1 (Example 1.1.1 with Method of Transformations). Suppose X follows the ex-
ponential distribution with mean 2, and Y = 5X + 4.
1. Then X > 0.
Y −4
2. We have Y = 5X + 4, so that X = .
5
3. Since X > 0, Y = 5X + 4 > 5(0) + 4 = 4, so Y > 4.
1
4. fX (x) = e−x/2 , x > 0.
2
y−4
5. We have x(y) = , as demonstrated in step 2. So,
5
( ) [ ] ( )
y−4 d y − 4 1 −(y−4)/10 1 1
fY (y) = fX · = e
= e−(y−4)/10 , y > 4.
5 dy 5 2 5 10
Notice how this answer matches with the answer in Example 1.1.1.

Example 1.2.2 (Example 1.1.2 with Method of Transformations). Suppose X follows the ex-
ponential distribution with mean 1, and W = X 2 .
1. Then X > 0.

2. We have W = X 2 , so that X = ± W . This may seem to be a problem, since we should√
only have one
√ function for X. But only one of these functions applies
√ when X > 0: W.
We ignore − W since none of the values of X are negative. So X = W for our purposes.

3. Since X > 0, W = X 2 > 02 = 0, so that W > 0.

4. fX (x) = e−x , x > 0.

8 ©2013 Yeng M. Chang (2015 update)


Exam P 1.3 Method of Jacobians


5. We have x(w) = w, so

(√ ) d [√ ] 1 √
fW (w) = fX
w · w = √ e− w , w > 0.
dw 2 w

Exercise 1.2.3. Explain why the Method of Transformations cannot be applied to Example 1.1.3.

Exercise 1.2.4. Do Exercise 1.1.5, except with the Method of Transformations.

Exercise 1.2.5. Suppose Z follows the standard normal distribution, so that

1
fZ (z) = √ e−z /2 ,
2
− ∞ < z < ∞.

Consider Y = eZ . Y is said to have a log-normal distribution. Find the equation for fY (y), y > 0.

1.3 Method of Jacobians


In order to understand this section, we will need to use some linear algebra. If we consider a two-
by-two matrix
[ ]
a b
A= ,
c d

the determinant of A, denoted det(A), is given by ad − bc.


Let X and Y be random variables with joint density fX,Y , and W1 = W1 (X, Y ), W2 = W2 (X, Y )
be one-to-one functions of X and Y , so that we can write X = X(W1 , W2 ) and Y = Y (W1 , W2 ).
The two-by-two Jacobian matrix, denoted J, of W1 and W2 is
 
∂ ∂
 ∂w1 [x(w1 , w2 )] ∂w2 [x(w1 , w2 )]
 
J =

.

 ∂ ∂ 
[y(w1 , w2 )] [y(w1 , w2 )]
∂w1 ∂w2

It can be shown that

fW1 ,W2 (w1 , w2 ) = fX,Y [x(w1 , w2 ), y(w1 , w2 )] |det(J)|

where | · | denotes the absolute value.

©2013 Yeng M. Chang (2015 update) 9


1.3 Method of Jacobians Exam P

How to do the Method of Jacobians:

1. Find fX,Y .

2. Find the domains of W1 and W2 .

3. Solve for X and Y with respect to W1 and W2 (using substitution


as necessary). If this cannot be done, then the method of Jacobians
cannot be applied.

4. Calculate
 
∂ ∂
 ∂w1 [x(w1 , w2 )] ∂w2 [x(w1 , w2 )]
 
J =

.

 ∂ ∂ 
[y(w1 , w2 )] [y(w1 , w2 )]
∂w1 ∂w2

5. Calculate det(J).

6. Apply the formula

fW1 ,W2 (w1 , w2 ) = fX,Y [x(w1 , w2 ), y(w1 , w2 )] |det(J)| .

Exercise 1.3.1. Suppose X and Y are independent exponential random variables with means 1 and
2, respectively. Let W = X + Y and Z = 2Y . Find the joint density function of W and Z.

1. fX,Y (x,y) ≡ fX (x)fY (y) due to independence of X and Y , so

1
fX,Y (x,y) = e−x e−y/2 , x > 0, y > 0.
2

2. W = X + Y > 0 + 0 = 0, so W > 0. Z = 2Y > 2(0) = 0, so Z > 0.

3. We have
Z
X =W −Y =W −
2
and
Z
Y = .
2
4. We have  ∂  
∂ 1
[x(w, z)] [x(w, z)] 1
 ∂w ∂z   2
   
J = = .
   
∂ ∂ 1
[y(w, z)] [y(w, z)] 0
∂w ∂z 2

10 ©2013 Yeng M. Chang (2015 update)


Exam P

( )
1 1 1
5. det(J) = 1 − (0) = . We don’t need to worry about the absolute value, since the
2 2 2
determinant is already positive.

6. Therefore,
1 1
fW1 ,W2 (w1 , w2 ) = fX,Y (w − z/2, z/2) (1/2) = e−(w−z/2) e−(z/2)/2 = e−w ez/4 ,
4 4
where w > 0 and z > 0.

2 Order Statistics
Let X = {X1 , X2 , . . . , Xn } be a set of independent random variables. We are interested in the order
statistics X(1) = min (X) and X(n) = max (X).
It is our task to find the PDF and CDF of both X(1) and X(n) .

2.1 Minimum of a Set of Random Variables


Note that
FX(1) (x) = P (min (X) ≤ x)
gives us no information (by itself) whatsoever about all of the random variables in the set X. What
if we looked at
P (min (X) > x) ?
If min (X) is greater than a fixed value x, then it follows that all variables in X must exceed x as
well.
So, to find the CDF of X(1) , we write

FX(1) (x) = P (min (X) ≤ x)


= 1 − P (min (X) > x)
= 1 − P (all of X1 , X2 , . . . , Xn > x)
= 1 − P (X1 > x) P (X2 > x) · · · P (Xn > x) by independence. (2.1.1)

If all of the random variables in X are identically distributed as well with CDF F , then P (Xi > x) =
1 − F (x) for all i, and
FX(1) (x) = 1 − [1 − F (x)]n . (2.1.2)
You can take derivatives of either (2.1.1) or (2.1.2) to get the PDFs. I do not recommend mem-
orizing the equation (2.1.2), but to memorize the method to get to (2.1.1), since given random
variables will not always be identically distributed.

Example 2.1.1. Let X1 , X2 , and X3 be exponentially distributed with means 1, 2, and 3 respec-
tively. Find the PDF of X(1) = min{X1 , X2 , X3 }.

©2013 Yeng M. Chang (2015 update) 11


2.2 Maximum of a Set of Random Variables Exam P

We have, as shown in (2.1.1):

FX(1) (x) = 1−P (X1 > x) P (X2 > x) P (X3 > x) = 1−e−x e−x/2 e−x/3 = 1−e−11x/6 , x > 0.

So, the PDF of X(1) is given by

d [ ] 11
fX(1) (x) = FX(1) (x) = e−11x/6 , x > 0.
dx 6
6
This is the PDF of an exponential distribution with mean .
11

2.2 Maximum of a Set of Random Variables


Now we find the CDF of X(n) . We don’t run into the same problem with X(n) as with X(1) : if the
maximum of all random variables in X is less or equal to some fixed value x, then all random
variables in X must be less than or equal to x as well. Thus,

FX(n) (x) = P (max (X) ≤ x)


= P (all of X1 , X2 , . . . , Xn ≤ x)
= P (X1 ≤ x) P (X2 ≤ x) · · · P (Xn ≤ x) by independence. (2.2.1)

If all of the random variables in X are identically distributed as well with CDF F , then
P (Xi ≤ x) = F (x) for all i, and
FX(n) (x) = [F (x)]n . (2.2.2)
As with min (X), I recommend memorizing the methods to get (2.2.1) rather than the formula
(2.2.2).

Example 2.2.1. Let X1 , X2 , and X3 be exponentially distributed with means 1, 2, and 3 respec-
tively. Find the CDF of X(3) = max{X1 , X2 , X3 }.
We have, as shown in (2.2.1):
( )( )( )
FX(3) (x) = P (X1 ≤ x) P (X2 ≤ x) P (X3 ≤ x) = 1 − e−x 1 − e−x/2 1 − e−x/3 , x > 0.

Order Statistics:

1. Find the domain of the order statistic.

2. Find the CDF of the order statistic you need by using the methods
to get (2.1.1) or (2.2.1).

3. Take the derivative of this function to get the PDF of the order
statistic.

12 ©2013 Yeng M. Chang (2015 update)

Das könnte Ihnen auch gefallen