13.3 Probability generating functions for discrete distributions

Probability generating functions (PGFs) are powerful tools for analyzing discrete random variables. They provide a compact way to represent a distribution's probabilities and moments, making it easier to work with complex distributions and perform calculations.

PGFs are closely related to moment generating functions (MGFs) but are particularly useful for discrete distributions. They allow for easy computation of probabilities, moments, and analysis of sums of independent random variables, making them invaluable in probability theory and statistical analysis.

Probability Generating Functions

Definition and Properties

Top images from around the web for Definition and Properties

• 

  Discrete Random Variables (3 of 5) | Concepts in Statistics View original

  Is this image relevant?

• 

  Discrete Random Variables (3 of 5) | Concepts in Statistics View original

  Is this image relevant?

1 of 1

Top images from around the web for Definition and Properties

• 

  Discrete Random Variables (3 of 5) | Concepts in Statistics View original

  Is this image relevant?

• 

  Discrete Random Variables (3 of 5) | Concepts in Statistics View original

  Is this image relevant?

1 of 1

• Probability generating function (PGF) for a discrete random variable X defined as $G_X(t) = E[t^X]$, where t represents a real number and E denotes expected value
• Power series representation of probability mass function (PMF) of discrete random variable
• For discrete random variable X with possible values 0, 1, 2, ..., PGF expressed as $G_X(t) = \sum_{k=0}^{\infty} p_k t^k$, where $p_k = P(X = k)$
• Domain of PGF typically $|t| \leq 1$ ensures convergence of power series
• Key properties include $G_X(1) = 1$ and $G_X(0) = P(X = 0)$ provide useful correctness checks
• PGFs uniquely determine probability distribution of discrete random variable allows recovery of probabilities and moments

Examples and Applications

• Bernoulli distribution with parameter p has PGF $G_X(t) = q + pt$, where $q = 1 - p$ (coin flip)
• Binomial distribution with parameters n and p has PGF $G_X(t) = (q + pt)^n$ (number of successes in n trials)
• Poisson distribution with parameter λ has PGF $G_X(t) = e^{\lambda(t-1)}$ (number of events in fixed time interval)
• Geometric distribution with parameter p has PGF $G_X(t) = \frac{pt}{1 - qt}$, where $q = 1 - p$ (number of trials until first success)
• Negative binomial distribution with parameters r and p has PGF $G_X(t) = (\frac{p}{1 - qt})^r$, where $q = 1 - p$ (number of failures before r successes)

Deriving Probability Generating Functions

Derivation Techniques

• Use definition of expected value $E[g(X)] = \sum_{x} g(x)P(X = x)$ to derive PGF
• Apply properties of expected values such as linearity and independence
• Recognize common series expansions (exponential, geometric, binomial)
• Utilize probability mass function (PMF) of distribution in derivation process
• Employ moment generating function (MGF) relationship $G_X(t) = M_X(\ln(t))$ when MGF known

Step-by-Step Examples

• Derive PGF for Bernoulli distribution:
  1. Start with PMF: $P(X = 0) = 1-p, P(X = 1) = p$
  2. Apply definition: $G_X(t) = E[t^X] = \sum_{x=0}^1 t^x P(X = x)$
  3. Simplify: $G_X(t) = t^0(1-p) + t^1p = (1-p) + pt$
• Derive PGF for Poisson distribution:
  1. Begin with PMF: $P(X = k) = \frac{e^{-\lambda}\lambda^k}{k!}$
  2. Use definition: $G_X(t) = E[t^X] = \sum_{k=0}^{\infty} t^k \frac{e^{-\lambda}\lambda^k}{k!}$
  3. Recognize series: $G_X(t) = e^{-\lambda} \sum_{k=0}^{\infty} \frac{(\lambda t)^k}{k!} = e^{-\lambda}e^{\lambda t} = e^{\lambda(t-1)}$

Probability Generating Functions vs Moment Generating Functions

Relationships and Conversions

• Moment generating function (MGF) relates to PGF by $M_X(t) = G_X(e^t)$
• PGF expressed in terms of MGF as $G_X(t) = M_X(\ln(t))$
• Both PGFs and MGFs uniquely determine probability distribution of random variable
• nth derivative of PGF evaluated at t = 1 gives nth factorial moment
• nth derivative of MGF at t = 0 gives nth raw moment
• Conversion between PGF and MGF properties allows application of results across functions

Comparative Advantages

• PGFs typically easier to work with for discrete distributions (integer-valued random variables)
• MGFs more commonly used for continuous distributions (real-valued random variables)
• PGFs always exist for non-negative integer-valued random variables
• MGFs may not exist for some heavy-tailed distributions (Cauchy distribution)
• PGFs useful for analyzing sums of independent random variables through multiplication
• MGFs facilitate moment calculations through differentiation

Applications of Probability Generating Functions

Probability and Moment Calculations

• Calculate probabilities using $P(X = k) = \frac{1}{k!} \frac{d^k}{dt^k} G_X(t)|_{t=0}$, where $\frac{d^k}{dt^k}$ denotes kth derivative
• Compute mean (first moment) of distribution using $E[X] = G'_X(1)$, where $G'_X$ denotes first derivative of PGF
• Calculate variance using $Var(X) = G''_X(1) + G'_X(1) - (G'_X(1))^2$, where $G''_X$ second derivative of PGF
• Obtain higher-order moments by evaluating higher-order derivatives of PGF at t = 1
• Example: For Poisson distribution with PGF $G_X(t) = e^{\lambda(t-1)}$, mean calculated as $G'_X(1) = \lambda e^{\lambda(t-1)}|_{t=1} = \lambda$

Distribution Analysis and Transformations

• Find distribution of sums of independent random variables by multiplying respective PGFs
• Analyze compound distributions using composition of PGFs (outer function PGF of number of events, inner function PGF of distribution being compounded)
• Example: Sum of independent Poisson random variables X and Y with parameters λ1 and λ2:
  1. $G_X(t) = e^{\lambda_1(t-1)}$ and $G_Y(t) = e^{\lambda_2(t-1)}$
  2. $G_{X+Y}(t) = G_X(t)G_Y(t) = e^{\lambda_1(t-1)}e^{\lambda_2(t-1)} = e^{(\lambda_1 + \lambda_2)(t-1)}$
  3. Resulting PGF corresponds to Poisson distribution with parameter λ1 + λ2