Random number generation speed

Millar, B and Broughton, R. A Preliminary PK Experiment with a Novel Computer-Linked High-Speed Random Number Generator. In J. Morris, W. Roll, and R. Morris (eds.). Research in Parapsychology, 1975. Metuchen, N.J. Scarecrow Press, 1976, pp. 83-84. 

PK tests with a high-speed random number generator. 

Random number generation, which occurs in prekey and main key generation, has been omitted, because real random numbers are needed, and the speed depends very much on the type of the implementation (physical or involving the user). Furthermore, Remark 9.20 has been used. 

The initial population in aU three algorithms was generated by rand2 (Press, Flannery, Teukolsky, Vet-terlingrandom, 1992) random number generating subroutine. The same initial populations in all three algorithms allowed compare of the speed and final function value. 

Because d Gaussian random numbers per particle are required at every iteration, where d is the spatial dimension, the speed of the random number generator is the limiting factor for MPC-AT. In contrast, the efficiency of SRD is rather insensitive to the speed of the random number generator since only d — uniformly distributed random nnmbers are needed in every box per iteration, and even a low quality random nnmber generator is sufficient, because the dynamics is self-averaging. A comparison for two-dimensional systems shows that MPC-AT—a is about a factor 2-3 times slower than SRD, and that MPC-AT-l-u is about a factor 1.3-1.5 slower than MPC-AT-fl [40]. 

---