Ziggurat Revisited

Random numbers following a Standard Normal distribution are of great importance when using simulations as a means for investigation. The Ziggurat method (Marsaglia and Tsang, 2000; Leong, Zhang, Lee, Luk, and Villasenor, 2005) is one of the fastest methods to generate normally distributed random numbers while also providing excellent statistical properties. This note provides an updated implementations of the Ziggurat generator suitable for 32and 64-bit operating system. It compares the original implementations to several popular Open Source implementations. A new implementation embeds the generator into an appropriate C++ class structure. The performance of the different generator is investigated both via extended timing and through a series of statistical tests, including a suggested new test for testing Normal deviates directly. Integration into other systems such as R is discussed as well.