1. The Gamma Function

Definition — Gamma Function

Theorem — Properties

  1. for
  2. for positive integers ;

Proof of 1. Integrate by parts with , :

Property 2 follows by repeating 1 down to . For 3, substitute and use the normal integral .


2. Gamma Distribution

Definition — Gamma Distribution


= shape, = scale.

Theorem — Mean and Variance

Proof (mean). Substituting :

Similarly , so .

Gamma pdfs, β = 1: α = 1 (black, exponential), α = 2 (blue), α = 4 (amber, dashed) 0 2 4 6 8 10

Theorem — Gamma–Poisson Relationship

In a Poisson process with rate , the waiting time until the -th event ( an integer) is Gamma with shape and scale .

Proof. iff fewer than events occur in :

Differentiating in gives the gamma density with .

Worked Example

Calls arrive at a switchboard at per minute. Probability that the 3rd call comes within 1 minute (, ):

Mean waiting time min, variance .

⚠️ Some books use a rate parameter instead of the scale . Check the density before plugging in.


3. Exponential Distribution

Definition — Exponential Distribution

Gamma with :



It models the time between events of a Poisson process (with ) and lifetimes of components that do not wear out.

Theorem — Memoryless Property

Proof. .

The exponential is the only continuous distribution with this property (the geometric is the discrete analogue).

Worked Example

An LED driver has lifetime .

  • Given it has already run 600 h, — the same!
  • Median lifetime: h (less than the mean, since the distribution is right-skewed)
Exponential(β = 500 h); shaded P(T > 200) = 0.6703 median 346.6 mean 500 0 200 500 1000 1500 2000

⚠️ Memorylessness means “a used component is as good as new” — fine for random shocks, wrong for parts that wear out (use Weibull with there).


4. Chi-Squared Distribution

Definition — Chi-Squared Distribution

Gamma with , :

(a positive integer) is the degrees of freedom.

Theorem — Origin of

If are independent , then

(proved with moment-generating functions in Chapter 9).

Notation — Critical Values

is the value with area to its right: .

Example. With : and . So .

⚠️ is not symmetric: you cannot get the left critical value by changing sign; look up both tails.

is used for inference about a variance (Chapter 10) and for goodness-of-fit and independence tests (Chapter 14).


5. Beta Distribution

Definition — Beta Distribution



Models proportions and probabilities (fraction of time a server is busy, a proportion defective). .

Worked Example

The fraction of CPU time used by a background task is : .

, .

Beta(2,3) (black; shaded P(X<0.5)=0.6875), Beta(2,2) blue, Beta(5,2) amber dashed 0 0.2 0.4 0.6 0.8 1

6. Lognormal Distribution

Definition — Lognormal Distribution

is lognormal if :

Theorem — Probabilities via the Normal Table

Worked Example

Repair times (hours) are lognormal with , .

Median h; mean h.

⚠️ and are the mean and SD of , not of .


7. Weibull Distribution

Definition — Weibull Distribution




With it reduces to the exponential.

Definition — Reliability and Failure Rate

  • Reliability:
  • Failure (hazard) rate:
Failure rateInterpretation
decreasing”infant mortality” (early defects)
constant random failures (exponential)
increasingwear-out / aging
Weibull failure rate Z(t) = αβt^(β−1), α = 1: β = 2 black (wear-out), β = 1 blue, β = 0.5 amber 0.5 1 1.5 2

Worked Example

Bearing life (in 1000 h) is Weibull with , .

  • (thousand hours)
  • Failure rate at : — increasing, so the bearing wears out.

Summary Table

DistributionSupportMeanVarianceTypical use
Gammawaiting time to -th event
Exponentialtime between events, memoryless lifetimes
sums of squared normals; variance tests
Betaproportions
Lognormalskewed positive data, repair times
Weibullsee abovereliability, failure times