Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!munnari!natmlab!mikeb
From: mikeb@natmlab.OZ (Michael Buckley)
Newsgroups: sci.math
Subject: Re: Need formula for Normal Distribution Simulation
Message-ID: <692@natmlab.OZ>
Date: Thu, 30-Oct-86 17:48:59 EST
Article-I.D.: natmlab.692
Posted: Thu Oct 30 17:48:59 1986
Date-Received: Fri, 31-Oct-86 20:26:20 EST
References: <146@helm.UUCP> <5270001@hpfcpf.HP.COM>
Reply-To: mikeb@natmlab.UUCP (Michael Buckley)
Organization: CSIRO Maths and Stats & Applied Physics, Sydney, Australia
Lines: 24


John Best has written an article titled 'Some easily programmed pseudo-
random normal generators' (Aust. Computer J., 11, 1979, pp.60-62) in which
he compares six different methods. His comparison is based on the average
number of uniform variates required to get one normal variate, and on
the average number of mathematical function (e.g. sin, log) evaluations
required, since these things take time to do -- much more than
arithmetic operations.

The Box-Muller method requires two mathematical function evaluations
per normal variate, which makes it slow compared to its competitors.
The Marsaglia-Bray polar method is quite a lot better, though John
finds that (given the relative speeds of uniform variate generation
and mathematical function evaluation on the machine he was then using)
two of the other four methods were slightly faster still.

I can send more detail by email if required, or a photocopy of John's
paper by paper mail.

--
Michael Buckley, CSIRO Div Maths & Stats, PO Box 218, Lindfield, NSW, Australia.
PHONE:	+61 2 467 6066			ACSNET:	mikeb@natmlab.oz		
ARPA: mikeb%natmlab.oz@seismo.css.gov	JANET:	natmlab.oz!mikeb
UUCP:	....{seismo,hplabs,mcvax,ukc,nttlab}!munnari!natmlab.oz!mikeb