Scribd
Upload
12 views3 pages

Construction of Random Variable Distribution Function

The document discusses the construction of a random variable with a specified distribution function, detailing the properties of distribution functions and the relationship between random variables and their distribution functions. It introduces the uniform [0, 1] random variable and demonstrates how to construct a random variable X such that its distribution function FX matches any given distribution function F. Theorems are presented to facilitate the simulation of random variables with any distribution function using uniform random variables, highlighting practical applications in computational statistics.

Uploaded by

Talha Farooq
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views3 pages

Construction of Random Variable Distribution Function

The document discusses the construction of a random variable with a specified distribution function, detailing the properties of distribution functions and the relationship between random variables and their distribution functions. It introduces the uniform [0, 1] random variable and demonstrates how to construct a random variable X such that its distribution function FX matches any given distribution function F. Theorems are presented to facilitate the simulation of random variables with any distribution function using uniform random variables, highlighting practical applications in computational statistics.

Uploaded by

Talha Farooq
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Stat 851 Fall 2024

Constructing a random variable with a given distribution function

We begin by recalling that a function F : R → [0, 1] is called a distribution function if

(i) lim F (x) = 0 and lim F (x) = 1,


x→−∞ x→∞

(ii) F is right-continuous, and


(iii) F is increasing.

We now suppose that (Ω, F, P) is a probability space and that X : (Ω, F, P) → (R, B) is a
random variable so that X −1 (B) ∈ B for all B ∈ B so that PX = P ◦ X −1 , the law of X, is
a probability on (R, B) and is characterized by the distribution function of X which is given
by FX (x) = PX ((−∞, x]) for x ∈ R.
The goal of these notes is to show a partial converse, namely that to every distribution
function F there exists a random variable X with FX = F . In particular, this random
variable will be constructed on the uniform probability space (Ω, F, P) = ([0, 1], B1 , unif).
As such, we often call ([0, 1], B1 , unif) the canonical probability space.
For the remainder of these notes, let (Ω, F, P) = ([0, 1], B1 , unif). Suppose that we define
the function U : Ω → Ω by setting U (ω) = ω so that U is the identity function on Ω.
However, since Ω = [0, 1], we can also view U as a function U : Ω → R. It now follows that
U : ([0, 1], B1 , unif) → (R, B) is a random variable since U −1 (B) = B ∩ [0, 1] ∈ B1 for any
B ∈ B. We call U a uniform [0, 1] random variable.
We will now compute FU (x), the distribution function of U . If 0 ≤ x ≤ 1, then
P ({ω ∈ Ω : U (ω) ≤ x}) = P ({ω ∈ [0, 1] : ω ≤ x}) = unif([0, x]) = x − 0 = x,
if x < 0, then
P ({ω ∈ Ω : U (ω) ≤ x}) = P ({ω ∈ [0, 1] : ω ≤ x}) = unif(∅) = 0,
and if x > 1, then
P ({ω ∈ Ω : U (ω) ≤ x}) = P ({ω ∈ [0, 1] : ω ≤ x}) = unif([0, 1]) = 1 − 0 = 1.
To summarize, if x ∈ R, then

0, if x < 0,

FU (x) = P ({ω ∈ Ω : U (ω) ≤ x}) = x, if 0 ≤ x ≤ 1, (1)

1, if x > 1.

Note that we sometimes call FU the uniform [0, 1] distribution function or the distribution
function of a uniform [0, 1] random variable.
Suppose now that F : R → [0, 1] is any distribution function. We will use U to construct
the random variable X on (Ω, F, P) with FX = F . In order to motivate the construction,
consider the particular case when F : R → [0, 1] is continuous and strictly increasing. For
such a function F , we know that F −1 is also continuous and strictly increasing.
Define

X = F −1 ◦ U so that X : Ω → R is given by X(ω) = F −1 (U (ω)).

The distribution function of X is given by

FX (x) = P ({ω ∈ Ω : X(ω) ≤ x})


= P {ω ∈ Ω : F −1 (U (ω)) ≤ x}


= P ({ω ∈ Ω : U (ω) ≤ F (x)})


= F (x)

for any x ∈ R where the third equality uses the fact that F is strictly increasing and
continuous, and the fourth equality follows from (1). In other words, FX = F as required.

Example 0.1. Suppose that


1 1
F (x) = arctan(x) +
π 2
for x ∈ R. It is easy to check that F is a distribution function and that F −1 (x) = tan(πx− π2 ).
If we define  π
X(ω) = tan πU (ω) − ,
2
then FX (x) = F (x). We say that X is a Cauchy random variable.

In the case that the distribution function F is neither strictly increasing nor continuous the
same type of result holds, provided that one is careful and chooses a single-valued inverse
of F .

Theorem 1. If F is a distribution function, then the random variable

X : ([0, 1], B1 , unif) → (R, B)

defined by
X(ω) = inf{x : F (x) ≥ ω}
for 0 ≤ ω ≤ 1 satisfies FX (x) = F (x).

Proof. Suppose that X(ω) = inf{x : F (x) ≥ ω} for 0 ≤ ω ≤ 1. Our goal is to compute

FX (x) = P ({ω ∈ Ω : X(ω) ≤ x})

for x ∈ R and show FX = F . Since X is an increasing function on [0, 1], we see that the
event A = Ax = {ω ∈ Ω : X(ω) ≤ x} is necessarily an interval with endpoints 0 and sup A.
Therefore, since P = unif so that the probability of an interval contained in [0, 1] is just its
length, we conclude

FX (x) = P ({ω ∈ Ω : X(ω) ≤ x}) = P (A) = unif(A) = sup(A) − 0 = sup(A).


Thus, to prove FX = F we must show that F (x) = sup(A). Since F is right-continuous, the
definition of X implies that F (X(ω)) ≥ ω. Thus, if ω ∈ A, then

F (x) ≥ F (X(ω)) ≥ ω

so that F (x) is an upper bound of A. However, we know that X(F (x)) ≤ x so that F (x) ∈ A.
This implies that F (x) = sup(A) as required.

A few remarks are in order. Note that since the uniform random variable U satisfies U (ω) = ω
an equivalent version of Theorem 1 is the following.

Theorem 2. Suppose that F is a distribution function, and let U : [0, 1] → R be given by


U (ω) = ω. The distribution function FX of the random variable X : ([0, 1], B1 , unif) → (R, B)
defined by
X(ω) = inf{x : F (x) ≥ U (ω)} (2)
for 0 ≤ ω ≤ 1 satisfies FX = F .

Also note that if F is continuous and strictly increasing, then

inf{x : F (x) ≥ U (ω)} = F −1 (U (ω))

as expected by our discussion preceding Theorem 1.

Remark. Theorem 2 is important in practice because it allows one to simulate on a computer


a random variable having any given distribution function. The only requirement is one be
able to simulate a uniform [0, 1] random variable U . Of course, there is the question of
how to simulate a uniform [0, 1] random variable, and this leads to the study of pseudo-
random numbers on a computer. Fortunately, every major programming language and piece
of mathematical software has facilities for simulating such uniform random variables. There
is also the question of how to compute (2), the inverse of F . Even if it cannot be done in
closed-form by hand, it can always be done numerically on the computer. Fortunately, there
are more efficient ways of simulating certain random variables through the use of various
transformations. Nonetheless, we have here a theoretical result with practical applications
which can always be applied.

Remark. Thanks to advances in computing power, this so-called inversion sampling method
is so fast that, for the software package R, this method is the most efficient way of generating
normal random variables. See

https://stat.ethz.ch/R-manual/R-devel/library/base/html/Random.html

for details.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy