Conventional random number generators provide the speed but not necessarilythe high quality output streams needed for large-scale stochastic simulations.Cryptographically-based generators, on the other hand, provide superior qualityoutput but are often deemed too slow to be practical for use in largesimulations. We combine these two approaches to construct a family of hybridgenerators that permit users to choose the desired tradeoff between quality andspeed for a given application. We demonstrate the effectiveness, performance,and practicality of this approach using a standard battery of tests, which showthat high quality streams of random numbers can be obtained at a costcomparable to that of fast conventional generators.
展开▼
