1. Transformations of Discrete Random Variables

Theorem — One-to-One Transformation (discrete)

Let have pmf and let be one-to-one, with inverse . Then

Theorem — Not One-to-One

If several values map to the same , add their probabilities:

Worked Example

= heads in 3 tosses: .

  • (one-to-one): .
  • (not one-to-one): , , , .
014

2. Transformations of Continuous Random Variables

Theorem — Change of Variable (one variable)

Let have pdf and be strictly monotone with inverse . Then

Proof (increasing ). . Differentiate: . If is decreasing, and ; both cases are .

Worked Example — From Uniform to Exponential

, . Inverse , :

So , which is also . (This is how computers generate exponential random numbers from uniform ones.)

Theorem — Not One-to-One (continuous)

If has several inverse branches :

Worked Example — Square of a Standard Normal

, . Branches , each with :

using . This is exactly the density with : .

⚠️ Always transform the support too, and don’t forget the absolute value — a density cannot be negative.


3. Transformations of Two Variables (Jacobian)

Theorem — Two Variables

Let have joint pdf , and , be one-to-one with inverse , . Then

To get the distribution of a single function , introduce a convenient helper , find , then integrate out.

Worked Example

independent : . Let , .

Inverse: , .

It factors on a rectangle, so and are independent: and .

flowchart LR
    A["(X₁, X₂) joint pdf"] --> B["Choose Y₁, Y₂;<br/>solve for x₁, x₂"]
    B --> C["Jacobian J =<br/>det ∂(x₁,x₂)/∂(y₁,y₂)"]
    C --> D["g(y₁,y₂) = f(w₁,w₂)·|J|<br/>+ new support"]
    D --> E["Integrate out y₂<br/>if only Y₁ is needed"]

4. Moments

Definition — Moments

The -th moment about the origin of :

and .


5. Moment-Generating Functions

Definition — Moment-Generating Function (MGF)


(when it exists for in an interval around 0).

Theorem — MGF Generates Moments

Proof. , so . The coefficient of — i.e. the -th derivative at 0 — is .

Theorem — Key Properties

  1. Uniqueness: if for all near 0, then and have the same distribution.
  2. and .
  3. If are independent and : .

Proof of 3. by independence.

Table of MGFs

Distribution
Binomial
Poisson
Geometric,
Normal
Gamma,
Exponential

Worked Example — Binomial Moments via the MGF

(binomial theorem).

  • ✓

Worked Example — Normal MGF (completing the square)

For :

Then gives .


6. Applications of MGFs

Theorem — Sum of Independent Poissons

If independent, then .

Proof. , the Poisson MGF; apply uniqueness.

Example. Two independent links lose on average 3 and 5 packets per hour. Total losses per hour Poisson(8).

Theorem — Sum of Squared Standard Normals

If are independent , .

Proof. Each (Section 2) with MGF ; the product is .


7. Distributions of Linear Combinations of Random Variables

Theorem — Linear Combination of Independent Normals

If are independent, then

Proof.

which is a normal MGF.

Worked Example — Fit of a Shaft in a Bearing

Bearing diameter (in 0.01 mm above nominal), shaft diameter , independent. Clearance :

Interference (shaft too big) happens when :

Clearance Y = X₁ − X₂ ~ N(4, 2.5); shaded P(Y < 0) = 0.0548 Y = 0 μ = 4 -4 0 4 8 12

Theorem — Sums of Other Independent Variables

If independent then
Bernoulli, of themBinomial
Binomial (same )Binomial
PoissonPoisson
Exponential, of themGamma
NormalNormal

⚠️ These results need independence. Means always add, but variances add only for independent (or uncorrelated) terms.