Automatic nonuniform random variate generation. (Q4435446)

The main topic of the book are universal generators, i.e. the generators usable for a broad class of distributions (e.g. not normal distribution specific). In many cases the distribution can be empirical. The book therefore does not discusses some well known distribution specific generators. It benefits from the long term research of the authors. All known important methods for the construction of universal nonuniform generators are presented together with the algorithms of corresponding generators. Generators of random vectors with various properties are also given. The last three chapter discuss the problems of the generation of time series (stationary Gausian, non-Gausian, Markov chains and Monte Carlo methods) and give some simulation examples from finance and Bayesian statistics. The algorithms are given mainly in an ad hoc algorithmic language, sometimes in C. There are more than 100 of such algorithms. The versions of almost all of the generators in ISO C (and some further information) can be downloaded from http://statistik.wu-wien.ac.at/arvag/. Yet more algorithms can be downloaded from UNU.RAN library http://statistik.wu-wien.ac.at/unuran/. The FORTRAN and Java source codes are available (via an autommatic generation)from http://statistik.wu-wien.ac.at/anuran/. The effectivity of the generators is measured relative to a generator of exponentially distributed numbers. This metric is almost platform independent. The values of the metric are given for almost all the presented generators. The book presents for all the algorithms the necessary theoretical background, the algorithms in an algorithmic language, and the values of the above metric for the generator. All chapters contain excercises. The issue how to test the statistical quality of the generatos is not discussed. The book can serve as a monography and as a textbook. The book is a good choice for everybody interested in nonuniform random number generators.