pgf_binom <- function(z, n, p) {
(z * p + 1 - p)^n
}
z <- seq(0, 1, length.out = 1000)
n <- 10
p <- 1/3
plot(z, pgf_binom(z, n, 3/4), type = "l")Probability Generating Functions
A probability generating function for a discrete random variable \(X\) taking values \(\{0,1,\ldots\}\) is defined as \[ G(z) = E[z^X] = \sum_{j = 0}^\infty z^j P(X = j) \]
Which is defined for all complex \(z\) in which the sum converges. It always converges for \(|z| < 1\) but the radius might be larger depending on our distribution. Clearly we have that \(G(0) = 0\). Also we can see that \[ G(1) = \sum_{j = 0}^\infty P(X = j) = 1 \] since we are summing over the entire sample space.
Probability of Even/Odd
This is problem 1.11 in Ross’s Stochastic Processes Second Edition.
We have that \[ \begin{aligned} G(-1) &= \sum_{j = 0}^\infty (-1)^j P(X = j)\\ &= P(X = 0) - P(X = 1) + P(X = 2) - \cdots\\ G(1) &= P(X = 0) + P(X = 1) + P(X = 2) + \cdots\\ G(-1) + G(1) &= 2P(X = 0) + 2P(X = 2) + 2 P(X = 4) + \cdots \end{aligned} \]
Assuming that 0 is considered even it follows that \[ \begin{aligned} P(X \text{ is even}) = \frac{G(1) + G(-1)}{2} = \frac{1 + G(-1)}{2} \end{aligned} \] We also have that \[ P(X \text{ is odd}) = 1 - \frac{1 + G(-1)}{2} = \frac{1 - G(-1)}{2} \]
Binomial
Assume that \(X\) is a binomial with parameters \(n\) and \(p\) \[ P(X = x) = {n \choose x} p^x (1 - p)^{n - x} \] Then we can compute the probability generating function
\[ \begin{aligned} G(z) = E[z^X] &= \sum_{j = 0}^\infty z^j P(X = j)\\ &= \sum_{j = 0}^\infty z^j {n \choose j} p^j (1 - p)^{n - j}\\ &= \sum_{j = 0}^n {n \choose j} (zp)^j (1 - p)^{n - j}\\ &= (zp + 1 - p)^n && \text{Binomial theorem} \end{aligned} \]
Note that the sum converges for all real \(z\). This is a good point to check that our result matches that \(G(1) = 1\). So \(G(-1) = (1 - 2p)^n\). Thus, \[ P(X \text{ is even}) = \frac{1 + (1 - 2p)^n}{2} \]
Poisson
Let \(X\) be a Poisson random variable with mean \(\lambda\) \[ P(X = k) = e^{-\lambda} \frac{\lambda^k}{k!} \]
Then \[ \begin{aligned} G(z) = E[z^X] &= \sum_{j = 0}^\infty z^j e^{-\lambda} \frac{\lambda^j}{j!}\\ &= e^{-\lambda} \sum_{j = 0}^\infty \frac{(\lambda z)^j}{j!}\\ &= e^{-\lambda} e^{\lambda z}\\ &= e^{-\lambda + \lambda z} \end{aligned} \] Note that the sum converges for all real \(z\). Again as an exercise check that \(G(1) = 1\). It follows that \[ G(-1) = e^{-2\lambda} \] So the probability of a Poisson random variable being even is \[ P(X \text{ is even}) = \frac{1 + e^{-2\lambda}}{2} \]
Geometric
Assume now that \(X\) is geometric with parameter \(p\) with \(X \in \{1,2,\ldots\}\). \[ P(X = k) = p (1 - p)^{k - 1} \] Then the probability generating function for \(X\) is
\[ \begin{aligned} G(z) = E[z^X] &= \sum_{j = 1}^\infty z^j P(X = j)\\ &= \sum_{j = 1}^\infty z^j p (1 - p)^{j - 1}\\ &= pz \sum_{j = 1}^\infty (z(1 - p))^{j - 1}\\ &= pz \sum_{j = 0}^\infty (z(1 - p))^{j}\\ &= \frac{pz}{1 - z(1 - p)} && \text{ for } |z| < \frac{1}{1 - p} \end{aligned} \] Since \(1 - p < 1\), then the sum converges for \(z = -1\). It follows that \[ G(-1) = \frac{- p}{2 - p} \] So the probability that the geometric random variable \(X\) is even is given by
\[ \begin{aligned} P(X \text{ is even}) &= \frac{1 + \frac{- p}{2 - p}}{2}\\ &= \frac{1 - p}{2 - p} \end{aligned} \]