Count the successes in n independent trials with success probability p: the count is binomial, B(n, p). When n is large and p small it is close to the Poisson distribution with the same mean λ = np. The top panel shows both probability mass functions (bars: binomial; red points: Poisson) and, after Simulate, the observed frequencies of simulated counts. The bottom panel keeps λ fixed and lets n grow: the exact total variation distance between the two distributions falls like 1/n and always stays below the Le Cam bound np² and the sharper Barbour–Hall bound (1 − e−λ)p.
Probabilities are computed exactly from the formulas (log-gamma for the binomial coefficients). The total variation distance is half the sum of |b(k) − π(k)| over all k, including the Poisson tail beyond n. Simulated counts use a seeded generator and the inverse of the binomial CDF (Algorithm 1.1 in the notes), so everyone sees the same run.
code/ch01_generative_discrete.R (Figures 1.1, 1.3 and 1.5).