HW7 Solutions
HW7 Solutions
HW7 Solutions
submitted
via Canvas in pdf format
2 Reading assignment
Class notes: Read the class notes up to and including Chapter 37 on
random vectors (Chapters 1 through 36 were previously assigned.).
Weiss textbook: Sections 6.5, 6.6 (up to and including Proposition 6.20
only), 7.4, 8.6, 8.7, 9.5, 9.6, 10.3, 11.3, 11.4
Schaum’s book (Hwei Hsu): Skip the “notes” (use our notes and Weiss’s
textbook). However, carefully work through exercises 3.2, 3.8, 3.9, 3.12 –
3.23 (all), 3.43, 3.44, 4.1, 4.3 –4.12 (all), 4.17, 4.18, 4.20, 4.21, 4.22, 4.24,
4.29, 4.30. Do not turn these in, but make sure you understand the details
of the calculations and how to solve these exercises. You would need to
show reasoning as given in our class notes, but his examples are excellent
for the calculations.
1
3 Written Assignment (to be turned in)
3.1 Exercise 1 (random vectors) (20 points –5 points per
part)
3.1.1 Exercise part a.
!
Suppose X = (X; Y ) is a random vector on a probability space ( ; F; P ) and
that F!X
is its joint cdf. For which set B (draw a picture of B; or describe it
clearly in words and/or mathematical notation) is it true that
n! o
F!
X
(2; 5) = P X 2 B ?
F!
X
(x; y) = P (fX xg \ fY yg) :
2
irrelevant as we take the limit as x goes to in…nity. Consider these parts one at
a time.
1st part: suppose y < 0: Then F! X
(x; y) = 0 for all x; and so
FY (y) = lim F!
X
(x; y) = 0:
x!1
FY (y) = lim F!
X
(x; y) = 1=2:
x!1
FY (y) = lim F!
X
(x; y) = 1:
x!1
Thus, 8
< 0; if y < 0
FY (y) = 1=2; if 0 y < 1 :
:
1; if y 1
From this, we can see that P (fY 2 Ag) = 1 for the set A = f0; 1g and that
P (fY = 0g) = 1=2 and P (fY = 1g) = 1=2: Thus, Y is a simple random variable
that takes the values 0 and 1 each with probability 1/2.
In general, the limits above are not simply limits of constants, but the pro-
cedure given above will work anyway.
3.1.5 Solutions
Solution to a):
F!
X
(2; 5) = P (f(X; Y ) (2; 5)g)
= P (fX 2g \ fY 5g)
= P (fX 2 Bg) ;
where
B = (x; y) 2 R2 : x 2 and y 5 :
Geometrically, B is the southwest region corresponding to the point (2; 5) ;
the set of all points on or to the left of the vertical line x = 2 and simultaneously
on or below the horizontal line y = 5:
Here is another way to describe B: Consider the point (2; 5) 2 R2 : Imagine
the vertical and horizontal lines passing through and meeting at (2; 5) : These
lines partition R2 into four quadrants. The set B is the bottom-left quadrant,
including its boundary edges.
3
Solution to b): Suppose W Exp ( ) for some > 0: For each c 0; we
have Z c
FW (c) = P (fW cg) = 0 dw = 0:
1
c
For each c > 0; the tail probability formula gives P (fW > cg) = e ; and thus
c
FW (c) = P (fW cg) = 1 P (fW > cg) = 1 e :
0; if a 0
FX (a) = 5a
1 e ; if a > 0
0; if b 0
FY (b) = 8b
1 e ; if b > 0
Since X and Y are independent, the joint cdf is the product of the marginal
cdfs, so
F!X
(a; b) = FX (a) FY (b) for all a; b 2 R:
Therefore,
5a 8b
1 e 1 e ; if a > 0 and b > 0
F! (a; b) = :
X 0; otherwise
FX (x) = lim F!
X
(x; y) = 0:
y!1
FX (x) = lim F!
X
(x; y) = 1=2:
y!1
FX (x) = lim F!
X
(x; y) = 1:
y!1
Thus, 8
< 0; if x < 0
FX (x) = 1=2; if 0 x < 1 :
:
1; if x 1
Solution to d): From FX (from the solution to c)), we can see that P (fX = 0g) =
1=2 and P (fX = 1g) = 1=2 (since the cdf jumps up by 1=2 at each of these
4
points). Thus, X is a simple random variable that takes the values 0 and 1 each
with probability 1/2.
We have
1=2; if x 2 f0; 1g
pX (x) = :
0; otherwise
denote the set of all points x at which FX is discontinuous. Since the only
possible discontinuities for a cdf are jump discontinuities, JX is precisely the set
of points at which the graph of FX jumps. For this reason, we will call JX the
jump set of X:
Since FX is monotonically increasing, it can have only countably many points
of discontinuity, and so JX is always countable, for any X:
Recall that a random variable X is discrete provided there exists a countable
c
set A of real numbers such that P (fX 2 Ag) = 1: Since each point x in JX
satis…es P (fX = xg) = 0; it can be shown that a random variable X is discrete
if and only if P (fX 2 JX g) = 1: Thus, when X is a discrete random variable,
JX is the smallest countable set which contains all the probability of X; if A is
any countable set for which P (fX 2 Ag) = 1; then JX A:
x=0 x=3
y= 1 1=40 6=40
y=1 3=40 13=40
y=4 8=40 9=40
5
First,
X
pX (0) = p(X;Y ) (0; b)
b2R
X
= p(X;Y ) (0; b)
b2f 1;1;4g
X
pX (3) = p(X;Y ) (3; b)
b2R
X
= p(X;Y ) (3; b)
b2f 1;1;4g
6
!
is a joint pmf for X :
It is a valid joint pmf since (1 = 246) (x + 3y) 0 whenever x 2 f0; 1; 2g
and y 2 f5; 6; 7; 8g ; and since 0 0 (so that each part of the pmf is always
non-negative), and since
X X2 X 8
1
p!
X
(x; y) = (x + 3y) = 1:
x=0 y=5
246
(x;y)2R2
fX = 0g \ fY 2 f5; 6; 7; 8gg
0 1
[
= fX = 0g \ @ fY = ygA
y2f5;6;7;8g
[
= (fX = 0g \ fY = yg)
y2f5;6;7;8g
8
[
= (fX = 0g \ fY = yg) :
y=5
Similarly,
8
X
pX (1) = p!
X
(1; y) ;
y=5
and so on.
Thus, to …nd pX (a) we …x the …rst component equal to a and we sum the
joint pmf over all y values in f5; 6; 7; 8g (which is JY ). Similarly, to …nd pY (b)
we …x the second component equal to b and we sum the joint pmf over all x
values in f0; 1; 2g (which is JX ).
For instance,
X2
pY (5) = p!
X
(x; 5) ;
x=0
and so on.
7
These formulas follow from the 3rd Kolmogorov axiom, as shown above, but
we can also consider them to be applications of the Fundamental Probability
Formula for discrete random vectors. For instance, pY (5) is P (fY = 5g) ; and
the Fundamental Probability Formula tells us to …nd this by summing the joint
pmf over all values where the y component is set equal to 5:
3.2.4 Exercises
!
For each of these exercises, suppose that X = (X; Y ) is as in the “more back-
ground for the assigned exercise” section above.
3.2.5 Solutions
Solution to a):
8
X X8
1 39
pX (0) = p!
X
(0; y) = (0 + 3y) = ;
y=5 y=5
246 123
8
X X8
1 41
pX (1) = p!
X
(1; y) = (1 + 3y) = ;
y=5 y=5
246 123
8
X X8
1 43
pX (2) = p!
X
(2; y) = (2 + 3y) = ;
y=5 y=5
246 123
8
X 8
X
pX (a) = p!
X
(a; y) = 0 = 0 if a 2
= f0; 1; 2g :
y=5 y=5
8
Therefore, 8
>
> 39=123; if x = 0
<
41=123; if x = 1
pX (x) = :
>
> 43=123; if x = 3
:
0; otherwise
Solution to b):
2
X X2
1 16
pY (5) = p!
X
(x; 5) = (x + 15) = ;
x=0 x=0
246 82
2
X X2
1 19
pY (6) = p!
X
(x; 6) = (x + 18) = ;
x=0 x=0
246 82
2
X X2
1 22
pY (7) = p!
X
(x; 7) = (x + 21) = ;
x=0 x=0
246 82
2
X X2
1 25
pY (8) = p!
X
(x; 8) = (x + 24) = ;
x=0 x=0
246 82
2
X 2
X
pY (b) = p!
X
(x; b) = 0 = 0 if b 2
= f5; 6; 7; 8g :
x=0 x=0
Therefore, 8
>
> 16=82; if y = 5
>
>
< 19=82; if y = 6
pY (y) = 22=82; if y = 7 :
>
>
>
> 25=82; if y = 8
:
0; otherwise
Solution to c): X and Y are not independent random variables. For example,
15
p!
X
(0; 5) = 0:06097561;
246
while
13 8
pX (0) pY (5) = 0:06186794;
41 41
so it is not the case that the joint pmf is the product of the marginal pmfs at
each point.
9
Solution to d): By the Fundamental Probability Formula,
X
P (fY = 3X + 2g) = p!X
(a; b)
(a;b)2JX JY ; b=3a+2
X
= p!
X
(a; b)
(a;b)2f0;1;2g f5;6;7;8g; b=3a+2
= p!
X
(1; 5) + p!
X
(2; 8)
16 26 42 7
= + = = :
246 246 246 41
= p!
X
(0; 7) + p!
X
(0; 8)
7 8 15
= + = :
82 82 82
10
3.3.2 Exercises
!
a. Suppose X = (X; Y ) is a discrete random vector (an arbitrary one, not one
of the speci…c ones introduced in this assignment).
Consider an arbitrary number a not in JX : Suppose b is any real num-
ber (possibly in JY ; possibly not in JY ). Fill in the missing reasons (labeled
“why?”):
p(X;Y ) (a; b) = P (fX = ag \ fY = bg)
P (fX = ag) (why?)
= 0 (why?).
!
b. Suppose X = (X; Y ) is a discrete random vector (an arbitrary one, not one
of the speci…c ones introduced in this assignment). Using the idea from part a.,
show that if (a; b) 2
= JX JY then p(X;Y ) (a; b) = 0:
Remark 1 The contrapositive of this statement is quite important. It is that if
p(X;Y ) (a; b) > 0 then (a; b) 2 JX JY : This shows why JX JY is so useful.
It contains all points (a; b) for which p(X;Y ) (a; b) > 0 (as well as possibly some
points for which p(X;Y ) (a; b) = 0), and it is also countable since JX and JY
are countable. The converse of the statement in part b. is false. I.e., it is not
true that if p(X;Y ) (a; b) = 0 then (a; b) 2
= JX JY ; there may indeed be points
(a; b) 2 JX JY for which p(X;Y ) (a; b) = 0:
Instructions for part c) and all later parts of this exercise: For part c)
!
and for all later parts of this exercise, suppose that X = (X; Y ) is a discrete
random vector whose joint pmf’s values are as given by the following table,
which gives p(X;Y ) (a; b) for all (a; b) 2 f0; 3g f 1; 1; 4g :
x=0 x=3
y= 1 1=40 6=40
y=1 3=40 13=40
y=4 8=40 9=40
This random vector was introduced earlier in this assignment. You may use
all results derived about it above without proof and without having to repeat
the same steps. Simply state any results you need which were proven above,
and, as your reasoning, cite the relevant part of the assignment speci…cally.
c. Find
P (fX 1g \ fY > 0g) + P ln (XY + 2) < X 2 Y :
Background for part d): I will now demonstrate the easy process for …nding
the pmf for W = X + Y: The key is to identify the jump set of W and to use
11
the Fundamental Probability Formula to …nd P (fW = wg) for each w 2 JW :
Once we verify that X
P (fW = wg) = 1;
w2JW
it will follow (since that sum equals P (fW 2 JW g)) that W is discrete, and we
will get its pmf right away by setting pW (w) = P (fW = wg) for all w 2 JW
and pW (w) = 0 for all w 2 = JW :
We …rst make a table of all (x; y; w) values where (x; y) 2 JX JY and
w = x + y:
x y w =x+y
0 1 1
0 1 1
0 4 4
3 1 2
3 1 4
3 4 7
The only values of w resulting from (x; y) pairs in JX JY are 1; 1; 2; 4
(which occurs for two di¤erent pairs), and 7; so JW will be f 1; 1; 2; 4; 7g (as
we will soon con…rm).
We now …nd P (fW = wg) for each w 2 f 1; 1; 2; 4; 7g as follows:
2XY 2 + X
Z= :
jY j
12
Hint: Proceed exactly as in my example above. Note that you will not deduce
that Z is discrete until late in your proof (as I did).
Background for part e): When …nding the expected value of a continuous
function Y = g (X) of a discrete random variable, we proceed by examining the
sum X
E (jg (X)j) = jg (x)j pX (x) :
x
I.e., we must check that the sum representing E (g (X)) is absolutely con-
vergent, since in that case we can conclude that E (g (X)) is well-de…ned.
Another –generally much slower –way to …nd this expected value is to …nd
the pmf for Y and use that to …nd E (Y ) by checking to see whether
X
E (jY j) = jyj pY (y) < 1:
y
The original way –what we’ve done primarily –is generally better, but both
methods work.
The same approach works for the expected value of a continuous function
of several discrete random variables. We will sum the function times the joint
pmf and examine it for absolute convergence. We can restrict the sum to values
in the Cartesian product of the jump sets of the component random variables,
since, as we have seen above, the joint pmf is zero otherwise (and adding zeroes
does not a¤ect a sum).
For a discrete random vector with two components, for instance, we proceed
by checking that
X
E (jg (X; Y )j) = jg (x; y)j p(X;Y ) (x; y) < 1:
(x;y)2JX JY
13
We can proceed this way with continuous functions of any …nite number of
component random variables, not just two.
Important: In every case, and for discrete random vectors having any number
of constituent discrete random variables, E is linear since it is de…ned in terms
of sums, which are linear.
I will demonstrate the process for the random variable X as above.
X is a simple random variable, so X has moments and central moments of
all orders. In particular, E (X) is well-de…ned, and we can calculate it quickly
as follows (see above for pX ):
X 12 28 21
E (X) = x pX (x) = 0 +3 = :
40 40 10
x2JX
We could also have checked by direct calculation that E (jXj) < 1; but I
wanted to demonstrate this alternative, faster method.
We can also try to …nd E (X) by using the formula for E (g (X; Y )) with
g (x; y) = x; which is continuous. We …rst check for absolute convergence:
14
calculate
X
E (X) = x p(X;Y ) (x; y)
(x;y)2JX JY
X
= x p(X;Y ) (x; y)
(x;y)2f(0; 1);(0;1);(0;4);(3; 1);(3;1);(3;4)g
Exercise for part e): Proceeding as I did for E (X) ; …nd E (Y ) in two ways:
…rst by using pY (above) and then by using p(X;Y ) (above).
h. Calculate
2XY 2 + X
E
jY j
(i.e., E (Z)) in two ways: by using pZ (calculated above) and then by using
p(X;Y ) and proceeding as in part e.
1 There is an easy, quick proof that these two expressions for E (X) are equal:
X
E (X) = x p(X;Y ) (x; y)
(x;y)2JX JY
X X
= x p(X;Y ) (x; y)
x2JX y2JY
2 3
X X
= x 4 p(X;Y ) (x; y)5
x2JX y2JY
X
= x pX (x) :
x2JX
15
i. Calculate
2XY 2 + X
V AR ;
jY j
speci…cally making use of p(X;Y ) and the shortcut formula for variance. Do not
make use of pZ (which also would work –and you can use that method to check
your answer –but is not what is being assigned).
Important: You are speci…cally required to use p(X;Y ) here. It is true that
using pZ would be easier in this particular case, but that is only because we
found pZ above already. In general, when you need to calculate the expected
value or variance of a function of several discrete random variables, you do not
want to start by …nding the pmf for the function of the variables; doing so is
too slow. The fastest way is to use the joint pmf directly, and that’s why I’ve
instructed you to proceed in that way here – to ensure that you practice this
valuable method, which is typically the best method.
Hint: Calculating the variance of a function of several variables is not compli-
cated. The quickest way is usually to use the shortcut formula. In this exercise,
you are to calculate
2XY 2 + X
V AR ;
jY j
2
which is simply V AR (Z) (which equals E Z 2 E (Z) ). Thus, our desired
variance is equal to
2
! 2
2XY 2 + X 2XY 2 + X
E E :
jY j jY j
3.3.3 Solutions
Solution to a): Whenever A B; we have P (A) P (B) : The missing
reasons are as follows:
16
Solution to b): Suppose that (a; b) 2= JX JY : Then either a 2
= JX ; in which
case
p(X;Y ) (a; b) = P (fX = ag \ fY = bg) P (fX = ag) = 0;
or b 2
= JY ; in which case
Solution to c): The joint pmf is positive only for the points (0; 1) ; (0; 1) ; (0; 4) ; (3; 1) ; (3; 1) ; (3; 4) :
Of those six points, only the points (3; 1) and (3; 4) satisfy X 1 and Y > 0;
so the Fundamental Probability Formula implies that
22
P (fX 1g \ fY > 0g) = p(X;Y ) (3; 1) + p(X;Y ) (3; 4) = :
40
Of those six points, only the points (3; 1) and (3; 4) satisfy ln (XY + 2) <
X 2 Y; so the Fundamental Probability Formula implies that
22
P ln (XY + 2) < X 2 Y = p(X;Y ) (3; 1) + p(X;Y ) (3; 4) = :
40
Therefore,
44
P (fX 1g \ fY > 0g) + P ln (XY + 2) < X 2 Y = :
40
Solution to d): We …rst make a table of all (x; y; z) values where (x; y) 2
JX JY and
2xy 2 + x
z= :
jyj
2
x y z = 2xyjyj+x
0 1 0
0 1 0
0 4 0
3 1 9
3 1 9
3 4 99=4
The only values of z resulting from (x; y) pairs in JX JY are 0; 9; and 99=4;
so JZ will be f0; 9; 99=4g (as we will soon con…rm).
17
We now …nd P (fZ = zg) for each z 2 f0; 9; 99=4g as follows:
Since we already veri…ed that Y has …nite expectation, we can also use p(X;Y )
to calculate
X
E (Y ) = y p(X;Y ) (x; y)
(x;y)2JX JY
X
= y p(X;Y ) (x; y)
(x;y)2f(0; 1);(0;1);(0;4);(3; 1);(3;1);(3;4)g
18
Notice that these two calculations for E (Y ) – one using pY and the other
using p(X;Y ) –yield the same result.
Solution to h): In each of these cases, the sums are …nite, hence absolutely
convergent, so E (Z) is well-de…ned and Z has …nite expectation. We …rst
calculate E (Z) using its pmf pZ ; which we found above:
X 12 19 99 9 315
E (Z) = z pZ (z) = 0 +9 + = :
40 40 4 40 32
z2JZ
2XY 2 + X
E
jY j
X 2xy 2 + x
= p(X;Y ) (x; y)
jyj
(x;y)2JX JY
X 2xy 2 + x
= p(X;Y ) (x; y)
jyj
(x;y)2f(0; 1);(0;1);(0;4);(3; 1);(3;1);(3;4)g
1 3 8 6 13 99 9
= (0) + (0) + (0) + (9) + (9) +
40 40 40 40 40 4 40
315
= :
32
19
Solution to i): Using the joint pmf,
2
!
2XY 2 + X
E
jY j
X 2
2xy 2 + x
= p(X;Y ) (x; y)
jyj
(x;y)2JX JY
X 2
2xy 2 + x
= p(X;Y ) (x; y)
jyj
(x;y)2f(0; 1);(0;1);(0;4);(3; 1);(3;1);(3;4)g
1 3 8 6 13 9801 9
= (0) + (0) + (0) + (81) + (81) +
40 40 40 40 40 16 40
112 833
= = 176:3015625:
640
2 2
Since 2XYjY j+X is non-negative everywhere, it equals its own absolute
value, and so
2
! 2
!
2XY 2 + X 2XY 2 + X 112 833
E =E = < 1;
jY j jY j 640
2 2 2
which implies that 2XYjY j+X has a …rst moment, and hence that 2XYjY j+X
has second moments, hence also second central moments (i.e., variance):
2
! 2
2XY 2 + X 2XY 2 + X 2XY 2 + X
V AR = E E
jY j jY j jY j
2
112 833 315
=
640 32
406 539
= = 79:402148:::
5120
20
Starting from a joint pdf, we can get a joint cdf by using the Fundamental
Probability Formula.
3.4.2 Exercises
a. Suppose (X; Y ) is an absolutely continuous random vector with joint cdf
8
> 0;1 2
>
>
if x < 0 or y < 1
>
< 10 x y 10 1 2
x + 401 1
xy 2 40 x; if 0 x < 1 and 1 y < 5
2 2 3
F(X;Y ) (x; y) =
> 5 x + 5 x; if 0 x < 1 and y 5 :
>
> 1 2
y + 1
y 1
; if x 1 and 1 y < 5
> 40
: 10 8
1; if x 1 and y 5
(1)
Find a joint pdf of (X; Y ) :
Hint: When the time comes to di¤erentiate (after checking that all the hy-
potheses for our procedure are satis…ed), be sure to di¤erentiate on open regions
(excluding all boundary points!) only, setting f (x; y) = 0 along all boundary
points.
The …nal answer will be
1 1
f (x; y) = 5x + 20 y; if 0 < x < 1 and 1 < y < 5
:
0; otherwise
Of course, you’ll need to do many steps to explain how to get to this point.
b. We now consider the reverse problem: going from a joint pdf to the joint
cdf.
Remember that
is the probability that our random vector (X; Y ) takes a value in the southwest
region determined by the point (a; b) :
To …nd this probability, we use the Fundamental Probability Formula. Specif-
ically, we intersect the southwest region determined by the point (a; b) with the
set ff > 0g of points (x; y) where f (x; y) > 0; and then we integrate the joint
pdf over this intersection. Remember that, whether it is with one or more vari-
ables, there is never a reason to integrate a function over a region where the
function equals zero (the integral over that region will just be zero).
Here is a sample calculation. Suppose 0 a < 1 and b 5: For sake of
de…niteness, imagine the point to be (2=3; 7) ; but it’s really an arbitrary (a; b)
with a 2 [0; 1) and b 2 [5; 1):
On a sheet of paper, we can draw axes and then shade the southwest region
corresponding to the point (a; b) : We can also (in a di¤erent color) shade the
rectangle consisting of all (x; y) pairs for which 0 < x < 1 and 1 < y < 5: The
intersection of these regions is the set of all (x; y) pairs for which 0 < x a and
21
1 < y < 5: Draw a picture to see this. Thus, by the Fundamental Probability
Formula we have
Z aZ 5
1 1 2 3
F (a; b) = x + y dydx = a2 + a;
0 1 5 20 5 5
which agrees with the 3rd line of the joint cdf above (except that I used an x in
that formula above instead of an a; I used an a here so that we could integrate
with respect to x: We cannot use the same letter for two inconsistent purposes
within the same problem).
In the same way, the other formulas can all be derived.
Exercise for part b): …nd a formula for F (a; b) when a 1 and 1 b < 5:
Do so by proceeding as I indicated above. Be sure to include the graph showing
the two shaded regions intersecting to form the region of integration, and write
what the double integral is (showing the function to be integrated and all the
limits of integration), as well as its value (which you may calculate in one step
with your calculator).
3.4.3 Solutions
Solution to a):
Step 1: We will …nd FX (x) and show it is continuous for all real x:
Since
FX (x) = lim F(X;Y ) (x; y) ;
y!1
22
Each piece of this multi-part function is a polynomial, so FX is continuous
on the open intervals ( 1; 0) ; (0; 1) ; and (1; 1) :
23
Step 3: Let N be the portions of the lines x = 0; x = 1; y = 1; and y = 5
which are in the closed …rst quadrant. At each (x; y) not in N; we are in one of
the open regions determined by replacing each inclusive inequality ( ; ) with
the corresponding strict inequality (<; >). In each such open region, F(X;Y ) is
given by a polynomial expression (a di¤erent one in each region). Polynomials
have continuous partial derivatives of all orders, and the partial derivatives are
polynomials too. It follows that all four of the second-order partial derivatives
of F(X;Y ) exist and are continuous at each point of R2 n N:
Step 4: We may now apply our theorem from class to deduce that (X; Y ) is
absolutely continuous and to …nd a joint pdf of (X; Y ) : The latter task involves
setting f equal to the mixed second order partial derivatives @ 2 =@y@x of F(X;Y )
in each open piece and setting f equal to 0 on N:
Note that
@2 @ @ @
(r (x)) = r (x) = (0) = 0
@y@x @x @y @x
and that
@2 @ @ @
(s (y)) = s (y) = (0) = 0:
@y@x @y @x @y
Thus, the mixed second partials will be 0 unless the function depends on
both x and y; and so we quickly get
(
@2 1 2 1 2 1 2 1
f (x; y) = @y@x 10 x y 10 x + 40 xy 40 x ; if 0 < x < 1 and 1 < y < 5
0; otherwise
1 1
= 5x + 20 y; if 0 < x < 1 and 1 < y < 5
:
0; otherwise
24
3.5 Exercise 5 (absolutely continuous random vectors) (45
points –5 points per part)
3.5.1 Background
The theory of functions g (X) of an absolutely continuous random variable X
!
extends in a straight-forward way to functions g (X1 ; X2 ; : : : ; Xn ) (or g X ) of
several absolutely continuous random variables X1 ; X2 ; : : : ; and Xn which are
!
part of an absolutely continuous random vector X :
!
Warning: as in the discrete case, if X = (X1 ; X2 ; : : : ; Xn ) is absolutely con-
tinuous, then each of the Xi s is absolutely continuous. However, unlike in the
discrete case, it is possible for each of the Xi s to be absolutely continuous
!
without X being absolutely continuous; thus, it is possible for there to exist
marginal pdfs fXi for each of the Xi s but for there not to exist a joint pdf for
!
X : A detailed example is given in the class notes.
In light of this warning, we will often restrict attention to functions of ab-
solutely continuous random vectors, or, equivalently, to functions of several
absolutely continuous random variables which together happen to also form an
absolutely continuous random vector (which, as noted in the warning, is not
always true for a collection of absolutely continuous random variables).
Many familiar functions of several absolutely continuous random variables
forming an absolutely continuous random vector will themselves represent ran-
dom variables.
In this exercise, you will end up needing to evaluate several double integrals.
Do not evaluate them by hand unless you have to. Your calculator can do several
of these, and you may use it for the evaluations. However, you must …rst write
out what integration you are going to carry out – indicating the function, the
exact limits of integration, and even the order of integration.
For instance, you might write something like this:
“
(something)
Z 8 Z 20
= ((1=5) x + cy) dxdy
5 10
= 195c + 90:”
Do not actually write that, since it is not part of the solution to the assigned
homework. However, at least that much detail must be given.
3.5.2 Exercises
For each part below, suppose (X; Y ) is an absolutely continuous random vector
with joint pdf given by
25
for some constant c:
a. What is the value of c? I.e., what value of c makes f a legitimate joint pdf?
Hint: In order for f to be a legitimate joint pdf, we need f (x; y) 0 for all
x 2 R and y 2 R; and we also need
Z 1Z 1
f (x; y) dydx = 1:
1 1
When evaluating Z Z
1 1
f (x; y) dydx;
1 1
or with the dxdy integral, we can restrict our limits of integration signi…cantly
by removing regions where the joint pdf is 0: You will …nd that c = 1=20: I am
giving you this answer since, without this, you would get all of the remaining
parts of this problem incorrect. You of course need to show all the work used
to …nd it.
b. I will give an example of the process of going from a joint pdf to a marginal
pdf by …nding a pdf for the random variable X above.
From part a. (this result was given in the hint), we have
If x 2
= (0; 1) ; we have f (x; y) = 0 for all y; and so
Z 1 Z 1
fX (x) = f (x; y) dy = 0 dy = 0:
1 1
26
Proceeding similarly (and considering cases, as I demonstrated), …nd a mar-
ginal pdf of Y:
Hint: Remember that we integrate functions only over regions where the func-
tion is non-zero; the integral of a function over a region where the function
equals zero is zero.
Another –generally much slower –way to …nd this expected value is to …nd
a pdf for Y and use that see whether
Z 1
E (jY j) = jyj fY (y) dy
1
The original way – what we’ve done all along – is generally much easier, but
both methods work.
The same approach works for the expected value of a continuous function of
several absolutely continuous random variables which also form an absolutely
continuous random vector.
For an absolutely continuous random vector with two components, for in-
stance, we proceed by checking whether or not
Z 1Z 1
E (jg (X; Y )j) = jg (x; y)j f(X;Y ) (x; y) dydx
1 1
27
is …nite. If it is, then E (g (X; Y )) is well-de…ned and is given by
Z 1Z 1
E (g (X; Y )) = g (x; y) f(X;Y ) (x; y) dydx:
1 1
We can proceed this way with functions of any …nite number of component
random variables from an absolutely continuous random vector.
Important: In every case, and for absolutely continuous random vectors having
any number of constituent absolutely continuous random variables, E is linear
since it is de…ned in terms of integrals, which are linear.
I will demonstrate the process for the random variable X as above.
Since fx : fX (x) > 0g is contained inside a bounded interval, it follows that
X has moments and central moments of all orders. In particular, E (X) is
well-de…ned, and we can use fX (see above) to calculate
Z 1
4 3 17
E (X) = x x+ dx = :
0 5 5 30
In this case, we could also have checked by direct calculation that E (jXj) <
1; but I wanted to demonstrate this alternative, faster method.
We can also …nd E (X) by …nding E (g (X; Y )) with g (x; y) = x: This
method has the advantage of allowing us to work directly with the joint pdf; it
is not necessary to …rst …nd a marginal pdf for X:
As always, we start by checking for absolute convergence:
which is …nite, and so E (g (X; Y )) (i.e., E (X) in this case) is well-de…ned and
is calculated as follows:
Z 1Z 1
E (X) = x f(X;Y ) (x; y) dydx
1 1
Z 1Z 5
1 1
= x x + y dydx
0 1 5 20
17
= :
30
Using elementary calculus, it is not di¢ cult to see that these two methods for
28
…nding E (X) –one method using fX ; the other using f(X;Y ) –are equivalent2 .
Important: We could also have replaced jxj by x since jxj = x for all x 2 (0; 1) :
However, the method I used is more general, since it would have worked even
if our integral were, say, from x = 10 to x = 4; in which case we would have
used the fact that jxj 10 for all x 2 ( 10; 4) :
Exercise for part d): Proceeding as I did for E (X) ; …nd E (Y ) in two ways:
…rst by using fY (above) and then by using f(X;Y ) (above).
e. Find E 2XY Y2 :
f. Find V AR 2XY Y 2 :
Hint: Use the shortcut formula for variance.
g. Let
P (f(X; Y ) 2 Bg)
and also
P (fX 2 [0; 1=2]g) P (fY 2 [1; 2]g) :
Use a calculator to calculate the integrals involved. What, if anything, do these
numbers tell us about the independence of X and Y ?
Hint: If X and Y are independent, then
29
h. Whenever X is a random variable and B is a Borel subset of R; fX 2 Bg
is an event, and therefore we can do everything with it that we could do with
events. For instance,
P (fX > 2g \ fY 5g)
P (fX > 2g j fY 5g) = ;
P (fY 5g)
and so on. More generally,
P (fX 2 Eg \ fY 2 F g)
P (fX 2 Eg j fY 2 F g) =
P (fY 2 F g)
as long as P (fY 2 F g) 6= 0:
Find
P (fX + Y < 3g j fY 2g) :
Hint: draw a picture! Be sure to shade the area over which you will integrate
the joint pdf above (for one of the steps of the solution). It’s a fairly simple
region, so if you get something truly bizarre then you did something wrong.
i. Here, we will study the independence (or lack of it) of X and Y by looking
at pdfs. Since (X; Y ) is an absolutely continuous random vector, X and Y are
independent if and only if
f (a; b) = fX (a) fY (b) (2)
except possibly for a 2-negligible set (i.e., a set having area 0) of (a; b) values.
If (and only if) we can …nd a set of (a; b) values having positive area (i.e., a set
which is not 2-negligible) for which equation (2) fails, it will follow that X and
Y must be dependent. In this problem, you will do precisely that.
If x 2
= (0; 1) then f (x; y) = 0 and also fX (x) = 0: Therefore, f (x; y) =
fX (x) fY (y) ; since both sides are 0:
If y 2
= (1; 5) then f (x; y) = 0 and also fY (y) = 0: Therefore, f (x; y) =
fX (x) fY (y) ; since both sides are 0:
So, we cannot conclude from our work so far that X and Y are independent,
nor can we conclude they are dependent. We must consider the remaining case,
when 0 < x < 1 and 1 < y < 5:
Let T be the rectangle
T = (0; 1) (1; 5) = (x; y) 2 R2 : 0 < x < 1; 1 < y < 5 :
Find a subset T of T having positive area and for which equation (2) fails,
and draw the appropriate conclusion about the independence of X and Y:
Be very clear about what T is, describing it algebraically and also graphing
it. Show that T has positive area, and explain why f (x; y) 6= fX (x) fY (y)
for all (x; y) 2 T :
Hint: For each (x; y) 2 T; we have x 2 (0; 1) and y 2 (1; 5) ; so you can
…nd simple expressions for f (x; y) ; fX (x) ; and fY (y) : Substitute those into
equation (2) to derive the equation
2xy y 6x + 3 = 0:
30
Let A be the set of all (x; y) 2 T for which this equation holds. Then T = T nA
is the set of all (x; y) 2 T for which this equation fails. You will need to graph
T and to …nd its area.
Solving the polynomial equation above for y would be useful, but it cannot
necessarily be done! Doing so would require that we divide both sides of the
equation by 2x 1; which we cannot do unless 2x 1 6= 0: Therefore, consider
the case where 2x 1 = 0 …rst, and then consider the case where 2x 1 6= 0:
This will let you graph A; and then you can easily graph T : Find the area
of T exactly, and be sure to explain why f (x; y) 6= fX (x) fY (y) for each
(x; y) 2 T : Then draw the appropriate conclusion about the independence or
dependence of X and Y:
3.5.3 Solutions
Solution to a): Setting
Z 5 Z 1
x 2
1= + cy dxdy = 12c +
1 0 5 5
gives c = 1=20: We also note that, when c = 1=20; the function f is always
non-negative, since 0 is non-negative, and since
x x y
+ cy = + >0
5 5 20
when 0 < x < 1 and 1 < y < 5: Therefore c = 1=20 (and only this value) makes
f a legitimate pdf.
31
Solution to c): By the Fundamental Probability Formula, we will integrate
the joint pdf over the set B \ ffX;Y > 0g ; where
B = (x; y) 2 R2 : x + y > 3 ;
and ffX;Y > 0g is the rectangle (0; 1) (1; 5) : My suggestion is to draw the
rectangle and shade it with horizontal lines, and then draw the open half-plane
x + y > 3 and shade it with vertical lines, and see where the two regions overlap.
Whether we proceed graphically or algebraically, our region of integration is
and therefore
Z 1 Z 5
x y 11
P (fX + Y > 3g) = + dydx = :
0 3 x 5 20 15
Solution to d): ffY > 0g = (1; 5) ; which is bounded, so Y has moments and
central moments of all orders. We may therefore calculate
Z 1 Z 5
1 1 49
E (Y ) = y fY (y) dy = y y+ dy =
1 1 20 10 15
and also Z 1Z 5
x y 49
E (Y ) = y + dydx = :
0 1 5 20 15
Solution to e): Note that 2xy y 2 < 106 when 0 < x < 1 and 1 < y < 5; so
Z 1Z 5
x y
E 2XY Y2 106 + dydx = 106 < 1;
0 1 5 20
32
2
Since 2XY Y2 is always non-negative, we have
2 2 1609
E 2XY Y2 =E 2XY Y2 = < 1;
15
2
and so 2XY Y 2 has …nite expectation. In particular, 2XY Y 2 has a sec-
ond moment, hence a second central moment (i.e., variance), which we compute
quickly as follows:
2 2
V AR 2XY Y2 = E 2XY Y2 E 2XY Y2
2
1609 124 8759
= = :
15 15 225
Solution to g):
We note that
2 7 7 1
P (fX 2 [0; 1=2]g) P (fY 2 [1; 2]g) = = 6= = P (f(X; Y ) 2 [0; 1=2] [1; 2]g) :
5 40 100 16
Thus, X and Y cannot be independent.
Solution to h): We are given that X and Y are absolutely continuous random
variables with joint pdf given by
Then
P (fX + Y < 3g \ fY 2g)
P (fX + Y < 3g j fY 2g) = :
P (fY 2g)
33
We can calculate P (fY 2g) in several ways. For instance, we can use the
joint pdf above:
Z 1Z 5
x 1 33
P (fY 2g) = + y dydx = :
0 2 5 20 40
Finally,
Since A and B are disjoint, and since their union is all of T; A and B form a
partition of T:
34
For each (x; y) 2 T; we have
1 1
f (x; y) = x + y;
5 20
4 3
fX (x) = x+ ;
5 5
1 1
fY (y) = y+ :
20 10
Thus, (x; y) 2 A if and only if
1 1 4 3 1 1
x+ y = x+ y+
5 20 5 5 20 10
2 3 1 3
= x+ y + xy + :
25 100 25 50
We can rewrite this so as to have a common denominator of 100 :
20 5 8 3 4 6
x+ y= x+ y+ xy + ;
100 100 100 100 100 100
which is true if and only if
20x + 5y = 8x + 3y + 4xy + 6;
12x + 2y = 4xy + 6;
6x + y = 2xy + 3;
2xy y = 6x 3;
35
and so
The set A consists of two line segments in the rectangle T; and so Area (A) = 0;
which implies that
Area (B) = Area (T ) = 4:
T = B is the desired set (it was given the notation T in the problem
statement). It consists of all ordered pairs (x; y) with x 2 (0; 1=2) [ (1=2; 1) and
y 2 (1; 3) [ (3; 5) :
Since the equation f (x; y) = fX (x) fY (y) is false in a set (i.e., T ) having
positive area, we conclude that X and Y are dependent.
3.6.2 Exercises
a. Suppose that X and Y are independent random variables on the same
probability space, each having second moments. Let X 0 = X X and Y =
0
Y Y:
Show that X 0 and Y 0 are independent, and use that fact to give a quick
proof that Cov (X; Y ) = 0:
36
Remark 2 The contrapositive of this result – that if Cov (X; Y ) 6= 0 then X
and Y are dependent – is also extremely important.
37
3.6.3 Solutions
Solution to a): Since X and Y are independent, and since X 0 and Y 0 are
functions of pairwise disjoint subcollections of X and Y; it follows that X 0 and
Y 0 are also independent.
X 0 and Y 0 must also have second moments, since their second moments are
the central moments of X and Y; respectively, which are de…ned since we are
given that X and Y have second moments.
Thus, since X 0 and Y 0 are independent, we have
E (X 0 Y 0 ) = E (X 0 ) E (Y 0 ) ;
E ((X X ) (Y Y )) = E (X X) E (Y Y ):
The left side of this expression is Cov (X; Y ) : The right side is zero, since
the linearity of E implies that the …rst central moments of X and of Y are each
zero.
Thus, Cov (X; Y ) = 0:
as claimed.
38
those random variables ensures that the expected value of a product of any
three of them will be the product of the expected values, hence 5 5 5 = 125:
Thus,
E (H) = E (X1 X2 X3 )
= E (X1 ) E (X2 ) E (X3 ) :
Here,
and thus
E ( 8W 3X + 5Y 3Z + 5)
= 8E (W ) 3E (X) + 5E (Y ) 3E (Z) + 5
= 8 (20) 3 (2) + 5 ( 5) 3 (9) + 5
= 213:
39