A Note on Generating Random Variables with T-concave Densities with the Ratio-of-Uniforms Method

Activity: Talk or presentationScience to science

Description

Devroye (2012) proposes an acceptance-rejection algorithm for distributions with given log-concave density f. It requires the exact location of the mode and has a uniformly bounded rejection constant but does not require the normalization constant for f. In this talk we show that the same idea also works for the ratio-of-uniforms method. Thus we get an acceptance-rejection algorithm with uniformly bounded rejection constant that works for the larger class of all T_{-1/2}-concave densities, a generalisation of log-concave densities, that includes unimodal densities with subquadratic tails. The derivation of the algorithm is simpler than the proof by Devroye (2012). Moreover, the method can also be extended to densities where the mode is only known approximately.
Period3 Jul 20177 Jul 2017
Event title11th International Conference on Monte Carlo Methods and Applications (MCM 2017)
Event typeUnknown
Degree of RecognitionInternational