43

5. Week 5

Definition 5.1 (Support of a Continuous RV). Let X be a continuous RV with DF FX . The set

S := {x ∈ R : FX (x + h) − FX (x − h) > 0, ∀h > 0}

is defined to be the support of X.

Remark 5.2. The support S of a continuous RV X can be expressed in terms of the law/distribution
of X as follows.

S = {x ∈ R : P(x − h < X ≤ x + h) > 0, ∀h > 0} = {x ∈ R : PX ((x − h, x + h]) > 0, ∀h > 0}.

Remark 5.3. The support S of a continuous RV X can be expressed in terms of the p.d.f. fX as
follows. Z x+h
S = {x ∈ R : fX (t) dt > 0, ∀h > 0}.
x−h

Note 5.4. If x ∈
/ S, where S is the support of a continuous RV X, then there exists h > 0 such
that FX (x + h) = FX (x − h). By the non-decreasing property of FX , we conclude that FX remains
a constant on the interval [x − h, x + h]. In particular, fX (t) = FX′ (t) = 0, ∀t ∈ (x − h, x + h).

Example 5.5. Consider a continuous RV X with DF FX : R → [0, 1] and p.d.f. fX : R → [0, ∞)

given by 


 0, if x < 0, 
1, if 0 < x < 1,

 


FX (x) := x, if 0 ≤ x < 1, , fX (x) := 


 0,
 otherwise.
1, if x ≥ 1.


To identify the support S, we consider the following cases.

(a) Let x ∈ (−∞, 0). Then for all h with −x > h > 0, we have x − h < x + h < 0 and
consequently, FX (x + h) − FX (x − h) = 0 − 0 = 0. Therefore x ∈
/ S.
(b) Let x ∈ (1, ∞). Then for all 0 < h < x − 1, we have 1 < x − h < x + h and consequently,
FX (x + h) − FX (x − h) = 1 − 1 = 0. Therefore x ∈
/ S.
(c) Let x ∈ (0, 1). For any 0 < h < min{x, 1 − x}, we have 0 < x − h < x + h < 1 and
consequently, FX (x + h) − FX (x − h) = (x + h) − (x − h) = 2h > 0. For h ≥ min{x, 1 − x},
44

at least one of x − h, x + h is in (0, 1)c and hence FX (x + h) − FX (x − h) > 0. Therefore

x ∈ S.
(d) Let x = 0. Then for any h > 0, we have FX (0 + h) − FX (0 − h) = FX (0 + h) > 0. Then
0 ∈ S. By a similar argument, 1 ∈ S.

From the above discussion, we conclude that S = [0, 1].

Note 5.6. Cantor function (also known as the Devil’s Staircase) is an example of a continuous dis-
tribution function, which is not absolutely continuous. In this case, the DF F is not representable
Rx
as −∞ f (t) dt for any non-negative integrable function. In fact, in this example, F ′ (x) = 0 at
‘almost all points x’. We do not discuss these types of examples in this course.

Definition 5.7 (Quantiles and Median for an RV). Let X be an RV with DF FX . For any
p ∈ (0, 1), a number x ∈ R is called a quantile of order p if the following inequalities are satisfied,
viz.
p ≤ FX (x) ≤ p + P(X = x).
1
A quantile of order 2
is called a median.

Note 5.8. A quantile need not be unique.

Notation 5.9. We write zp (X) to denote a quantile of order p.

1 3
Notation 5.10. The quantiles of order 4
and 4
for an RV X are referred to as the lower and
upper quartiles of X, respectively.

Note 5.11. The inequalities mentioned in Definition 5.7 can be restated as

P(X ≤ x) ≥ p, P(X ≥ x) ≥ 1 − p.

Note 5.12. Let X be a continuous RV with DF FX . Then a quantile of order p is a solution to

the equation FX (x) = p, since P(X = x) = 0, ∀x ∈ R. Moreover, if FX is strictly increasing, then
zp (X) is unique for all p ∈ (0, 1).
45

Example 5.13. Consider a continuous RV X with DF FX : R → [0, 1] given by





0, if x < 0,


FX (x) := x, if 0 ≤ x < 1, .




1, if x ≥ 1.



For any p ∈ (0, 1), the solution to FX (x) = p is given by x = p, i.e. zp (X) = p. Moreover, the
median is z 1 (X) = 12 .
2

We now discuss functions of RVs and their law/distributions.

Remark 5.14 (Function of an RV is an RV). Let h : R → R be a function and let X be an RV

defined on a probability space (Ω, F, P). Since X : Ω → R is a function, we can consider the
composition of the functions h and X to obtain another function h ◦ X : Ω → R defined by
(h ◦ X)(ω) := h(X(ω)), ∀ω ∈ Ω. Since h ◦ X is a real valued function defined on Ω with (Ω, F, P),
h ◦ X is an RV defined on the probability space (Ω, F, P).

Notation 5.15. In the setting of the above remark, we shall write h(X) to denote h ◦ X.

Example 5.16. Let X be an RV defined on a probability space (Ω, F, P) and consider the function
h : R → R defined by h(x) = 3x2 + sin x + 1, ∀x ∈ R. Then h(X) = h ◦ X defined by (h ◦ X)(ω) :=
3X(ω)2 + sin(X(ω)) + 1, ∀ω ∈ Ω is an RV.

Remark 5.17 (DF of a function of an RV). We continue with the notations of Remark 5.14 and
are interested in computing the law/distribution of Y = h(X). Using Remark 3.20, we may
equivalently, compute the DF of Y and that will identify the required law. Then for any y ∈ R,
we have

FY (y) = P(Y ≤ y) = P(h(X) ≤ y) = P(h(X) ∈ (−∞, y]) = P(X ∈ h−1 ((−∞, y])),

where h−1 ((−∞, y]) denotes the pre-image of (−∞, y] under h (see Notation 2.39).
46

Example 5.18. Let X be a discrete RV with p.m.f.


 |x| if x ∈ {±1, ±2, . . . , ±10}
110
fX (x) :=
 0, otherwise

and take h : R → R as h(x) := |x|, ∀x ∈ R. Note that





 ∅, if y < 0,


h−1 ((−∞, y]) = {0}, if y = 0,




[−y, y], if y > 0.



Then the DF of Y = h(X) = |X| is given by

FY (y) = P(X ∈ h−1 ((−∞, y]))





 P(X ∈ ∅), if y < 0,


= P(X ∈ {0}), if y = 0,




P(X ∈ [−y, y]), if y > 0.






 0, if y < 0,


= P(X = 0), if y = 0,




P

fX (t), if y > 0.
t∈[−y,y]∩{±1,±2,...,±10}

0, if y ≤ 0,


= P
|t|

 t∈[−y,y]∩{±1,±2,...,±10} 110 , if y > 0.

From the structure of the DF we conclude that the RV is discrete. The p.m.f. may be computed
using the techniques discussed in earlier lectures.
47

Example 5.19. Let X be a continuous RV with p.d.f.


|x|
 , if − 1 < x < 1
 2


fX (x) = x
 3
, if 1 ≤ x < 2



0, otherwise

and take h : R → R as h(x) := x2 , ∀x ∈ R. Note that





∅, if y < 0,


h−1 ((−∞, y]) = {0}, if y = 0,


 √ √


[− y, y], if y > 0.


Then the DF of Y = h(X) = X 2 is given by

FY (y) = P(X ∈ h−1 ((−∞, y]))





 P(X ∈ ∅), if y < 0,


= P(X ∈ {0}), if y = 0,

P(X ∈ [−√y, √y]), if y > 0.








 0, if y < 0,


= P(X = 0), if y = 0,



 √ √
P({− y ≤ X ≤ y}), if y > 0.






 0, if y < 0,


= 0, if y = 0,

√


R y

√ fX (x) dx, if y > 0.

− y
48




 0, if y < 0,







 0, if y = 0,
 √

R y |x|
=  − y 2 dx, if 0 ≤ y < 1
√

R √y x


 R 1 |x|
dx + dx, if 1 ≤y<4


−1 2 1



 3


1, if y ≥ 4






 0, if y ≤ 0,



y, if 0 ≤ y < 1


2
=
y+2
, if 1 ≤ y < 4





 6


1, if y ≥ 4.



From the structure of the DF we conclude that the RV is continuous. The p.d.f. may be computed
using the techniques discussed in earlier lectures.

Note 5.20. We continue the discussion in Remark 5.17. In general, we may not be able to
reduce/simplify the expression h−1 ((−∞, y]) further, without additional information about h or
X. In what follows, we shall consider the cases where X is discrete or continuous and then attempt
to obtain the DF of h(X).

Theorem 5.21. Let X be a discrete RV with DF FX , p.m.f. fX and support SX . Let h : R → R

be a function. Then Y = h(X) is a discrete RV with support SY = h(SX ) := {h(x) : x ∈ SX },
p.m.f. fY given by 
fX (x), if y ∈ SY ,

P
x∈h−1 ({y})

fY (y) =
0, otherwise



and DF FY given by
X X X
FY (y) = P(Y ≤ y) = fY (t) = fX (x) = fX (x).
t∈SY ∩(−∞,y] x∈SX x∈SX ∩h−1 ((−∞,y])
h(x)≤y
49

Proof. Since SX is a finite or a countably infinite set, the set h(SX ) is also finite or countably
infinite. Now,
P(h(X) ∈ h(SX )) = P(X ∈ h−1 (h(SX ))) ≥ P(X ∈ SX ) = 1

and hence P(h(X) ∈ h(SX )) = 1. Here, we have used the fact that h−1 (h(SX )) ⊇ SX . Moreover,
for any x ∈ SX ,

P(h(X) = h(x)) = P(X ∈ h−1 ({h(x)})) ≥ P(X ∈ {x}) = fX (x) > 0

and hence Y = h(X) is discrete with support SY = h(SX ). The expressions for fY and FY follows
from standard arguments. □

Note 5.22. As a consequence of Theorem 5.21, we conclude that the functions of discrete RVs are
also discrete RVs.

Note 5.23. In Theorem 5.21, the function h need not be one-to-one or onto and therefore need
not have an inverse. This was the same problem encountered in Remark 5.17, which stops us in
computing the DF of h(X) for a general RV X.

As a special case of Remark 5.17, we get the next result. We do not give a separate proof, for
brevity.

Corollary 5.24. Continue with the notations of Theorem 5.21. Assume that h : SX → R is
one-to-one. Then we have 
−1
X (h (y)), if y ∈ SY ,

f

fY (y) =
0, otherwise



where h−1 : h(SX ) → SX denotes the inverse function of h : SX → R.

Example 5.25. Let X be a discrete RV with p.m.f.


1
 , if x ∈ {−2, −1, 0, 1}
 7


fX (x) = 3
 14
, if x ∈ {2, 3}



0, otherwise.
50

Consider the RV Y = X 2 . Here SX = {−2, −1, 0, 1, 2, 3} and SY = {0, 1, 4, 9}. Observe that,

P(Y = 0) = P (X 2 = 0) = P(X = 0) = 71 ,
1 1
P(Y = 1) = P (X 2 = 1) = P(X ∈ {−1, 1}) = 7
+ 7
= 27 ,
1 3 5
P(Y = 4) = P (X 2 = 4) = P(X ∈ {−2, 2}) = 7
+ 14
= 14
3 3
P(Y = 9) = P (X 2 = 9) = P(X ∈ {−3, 3}) = 0 + 14
= 14 .

Therefore, the p.m.f. of Y is 

1



 7
, if y = 0

 2



 7
, if y = 1

fY (y) =  5
14
, if y = 4

 3



 14
, if y = 9


0, otherwise,



and the DF of Y is 




0, if y < 0

1
if 0 ≤ y < 1


 ,
 7


FY (y) = 3
 7
, if 1 ≤ y < 4

11
if 4 ≤ y < 9




 14
,


1, if y ≥ 9.



In fact, after identifying SY , we could have directly computed the DF FY as follows:





 0, if y < 0,



0 ≤ y < 1,

P(Y = 0), if





FY (y) = P(Y ≤ y) = P(Y = 0) + P(Y = 1), if 1 ≤ y < 4,




= 0) + P(Y = 1) + P(Y = 4), if 4 ≤ y < 9,




 P(Y



1, if y ≥ 9.



and the p.m.f. fY from FY using standard techniques discussed in earlier lectures.
51

Example 5.26. Let X be a discrete RV with p.m.f.


 x if x ∈ {1, 2, . . . , 10}
55
fX (x) =
 0, otherwise.

Now consider the RV Y = X 2 . Note that the function h : R → R defined by h(x) := x2 , ∀x ∈ R

is one-to-one on the support SX . Here, Y is discrete with support SY = {1, 4, 9, · · · , 100} and by
Corollary 5.24, the p.m.f. fY is given by
 
√ √
 y,
X ( y), if y ∈ SY , if y ∈ SY ,
 
f
 
55
fY (y) = = .
0, otherwise 0, otherwise

 


The DF FY can now be computed from the p.m.f. fY using standard techniques.

Now we look at functions of continuous RVs.

Example 5.27. Let X be a continuous RV with p.d.f.


 e−x , if x > 0
fX (x) =
 0, otherwise

and let Y = [X], where [x] denotes the largest integer not exceeding x for x ∈ R. Note that
SX = [0, ∞). Moreover,
P(Y ∈ {0, 1, 2, . . .}) = P(X ∈ SX ) = 1

and hence Y is a discrete RV. Now, for y ∈ {0, 1, 2, . . .}

Z y+1 Z y+1
fY (y) = P(Y = y) = P(y ≤ X < y + 1) = fX (x) dx = e−x dx = 1 − e−1 e−y > 0.
y y

hence Y is a discrete RV with support SY = {0, 1, 2, . . .} and the above p.m.f. fY . Therefore, a
function of a continuous RV need not be a continuous RV.

Remark 5.28. Given any continuous RV X and a constant function h : R → R given by h(x) :=
c, ∀x ∈ R for some c ∈ R, the RV h(X) is discrete. Together with the above example, we
may conclude that additional information on h is required before we can conclude that h(X) is
continuous.