🚀 go-pugleaf

In article <1992Jul25.191638.16109@samba.oit.unc.edu>,
Jim.Buddenhagen@bbs.oit.unc.edu (Jim Buddenhagen) writes:
 
|> Portable pseudo-random number generator desired, for uniform and/or normal
|> deviates.  What is desired is code (C preferred) or algorithm that is
|> machine independent (i.e. same seed, same sequence regardless of computing
|> platform).  Purpose will be to generate random datasets (or pointers into
|> databases) for benchmark tests accross several platforms.  Must be able to
|> generate several independent sequences 'simultaneously' (ok to have several
|> copies of routine with different names).
|> 
|> I am aware of routines in Numerical Recipies, and in Knuth, but unsure
|> of quality.  Would welcome any comments, suggestions, pointers to code,
|> or references.

There are NO generators currently available that can produce multiple
sequences with any reasonable degree of independence, and almost every
expert (including Knuth) warns against misusing single generators in this
way.  Almost the best that you can do is to get non-overlapping sequences,
which is NOT the same thing!  Most of the papers that purport to describe
generators for independent sequences use definitions of independence that
are completely unheard of by statisticians.

The main exceptions that I know of are various papers by A. De Matteis and
others (which are well worth reading) and N.M. Maclaren "The Generation of
Multiple Independent Sequences of Pseudorandom Numbers" Appl.Statist. (1989)
38 pp 351-359.  But PLEASE note the provisos in the latter about the
criterion used for independence - there is more justification than there
was room for in the paper, but little is certain in this area.


Nick Maclaren
University of Cambridge Computer Laboratory,
New Museums Site, Pembroke Street,
Cambridge CB2 3QG, England.
Email:  nmm@cl.cam.ac.uk
Tel.:   +44 223 334761
Fax:    +44 223 334679