Archived: 21 January 2009
University of Adelaide Masters by coursework thesis, June 2007.
Supervisor: Paul Coddington
Random number generators are widely used in many applications. Some scientific applications, such as Monte Carlo simulations, require a high quality random number generator since the correctness of their results rely heavily on the randomness of the numbers produced by the generator. However, since the random number generators in practice use deterministic algorithms to produce a sequence of numbers with particular distribution properties, these numbers produced are not truly random but deterministic as well. So some kinds of correlations may exist in the random number sequences. It is crucial to test comprehensively before use any of the random number generators, especially for the applications in which a high quality generator is an essential requirement.
This project improves the existing JAPARA library which provides several parallel random number generators in Java, and develops comprehensive random number generator tests for the library. Because they are parallel generators, apart from the correlations existed in each random number sequence they produced, we need to check the correlations between different sequences of random numbers produced on different threads. In addition, the project also investigates the effects of initialization methods of parallel generators on the quality of these generators.