New algorithms for generating Poisson variates

摘要

The paper discusses three recently-developed algorithms for generating Poisson(λ) random variates in situations where λ may vary between calls. KPFRAC is designed for λ < 1, KPLOW for 0 < λ ⩽ 30, and KEMPOIS for 10 ⩽ λ ⩽ 5000. All three algorithms use sequential searching, but each has novel features which speed up generation. Timing comparisons with two state-of-the-art routines indicate that a routine which uses KPLOW when λ ⩽ 30 and KEMPOIS when λ > 30 is much faster over most of the range 0 < λ < 700.