Generating beta variates via patchwork rejection (Q2366302)

The authors suggest an improved version of the acceptance-rejection algorithm for generating beta distributed pseudorandom variables with parameters \(p>1\) and \(q>1\). Their new sampling method improves acceptance-rejection in the center by using the so-called patchwork rejection idea of \textit{Do Le Minh} [ACM Trans. Math. Software 14, No. 3, 261-266 (1988; Zbl 0648.65003)]. The expected number of uniforms needed is bounded by 3.26 for \(p\searrow 2\) and \(q\searrow 1\), or \(p\searrow 1\) and \(q\searrow 2\), and decreases to 1.45 as \(p,q\to\infty\). Experiments covering a wide range of pairs \((p,q)\) show improvement in speed over competing algorithms in most cases.