Handout4 Transformation of Rvs
Handout4 Transformation of Rvs
Suppose we know the probability distribution of X but our interest lies in a transformation, say,
X 2 , or in general in Z := h(X). As a result, we want to know the distribution of Z. Similarly,
maybe, the joint distribution of (X, Y ) is known but we want to know the distribution of a
transformation, say, X + Y , or in general, W := g(X, Y ).
Recall that “probability distribution” of a r.v. is a mathematical object from which one should
be able to calculate any probability related to the r.v.
One way to do that is to provide the so called probability mass function (pmf ) if the desired r.v.
is discrete or the probability distribution function (pdf ) if the r.v. is continuous. Consequently,
we need to provide (i) the possible values the r.v. X can take and (ii) the corresponding
probability mass/density.
This leads to the following logical steps:
Step 1 Figure out all possible values that the new random variable Z or W can take.
This will, of course, depend on the values X, Y can take and the nature of the functions
h(·) and g(·, ·). So, a graph/plot of the functions h(·) and g(·, ·) will help.
Step 2 Derive the formula for pdf or pmf. This step differs for the discrete and continuous
case.
You should make a distinction between small letters and capital letters. The capital letter
X stands for the r.v. (which is the word description of a numerical quantity related to an
experiment). The small letter x represents a possible value X can take. In particular, x
is a number.
1
Example: Suppose X is a (continuous) random variable with the pdf
Solution:
Step 1: [Deciding the possible values: The transformation here is x 7→ x1 . If we think about it (and/or
draw a graph), we realize/see that for positive input, the output is also positive.]
1
Since X takes only positive values, Y = X also takes positive values.
Step 2: Since the transformed r.v. Y is continuous, we need to derive the pdf. However, we start with
the cdf. This means we need to provide/calculate FY (y) := P (Y ≤ y) for every y ∈ (−∞, ∞).
Note that, if y ≤ 0, then FY (y) = P (Y ≤ y) = 0. [This is because Y takes only positive values. Hence
the probability that Y takes a negative value or even lower is zero.]
For y > 0, we have
defn. Y 1 (⋆) 1
FY (y) := P (Y ≤ y) ====== P ≤y === P X≥ .
X y
[In (⋆) we have used the fact that 1/X is smaller than (or =) y if and only if X is larger than (or =) 1/y]
λ −λ/y
= e , for y > 0.
y2