An object-oriented high-performance pseudorandom number generator package for analysis simulation

Analysis simulation poses rigid requirements on pseudorandom number generators (PRNGs) in the following aspects, such as stochastics, robustness, period, and efficiency. The algorithms of typical Prime Modulus Multiplicative Linear Congruential Generators (PMMLCGs), both individual and combined, are briefly introduced. Various methods, Viz. modulus operator, fmod function, and modulo decomposition are discussed in detail. An object-oriented high-performance package of the Combined PMMLCG (CPMMLCG) has been implemented in C++ language using the component technology, which ensures its good portability between different platforms. Numerical experiments show the package is also efficient and has good figure of merit. From experiment results, it concludes that modulo decomposition method outperforms functionalities provided by standard C/C++ runtime (CRT) library, that whether floating-point arithmetic show the superiority over integer arithmetic depends on the complexity of CPMMLCGs and that the type of coefficients (variables or constants) has negligible effects on CPMMLCGs. These research results lay a solid foundation for the parallel simulation of the multiple samples, and the achievement of the confidence interval of simulation results.