Re: FLUKA: FLUKA random numbers

From: Alfredo Ferrari (alfredo.ferrari@cern.ch)
Date: Mon Jun 17 2002 - 17:30:36 CEST


Hi all

Alberto is right but:

a) the random number generator period is not the only number to look for,
   indeed one can get repeated sequences even using much less random
   numbers than the period. In particular, let us consider a (simple)
   event type which uses just one random, and suppose we start two runs
   with seeds spaced by 1000000 calls, of course after 1000000 events
   the first run will go exactly on the steps of the first one. In real
   cases with many randoms per event (and also a random number of randoms
   per event) the situation is more complex, but still there is a
   non negligible chance to get the same events after relatively few
   events even for seed spaced by billion calls (I had a couple of
   examples)

b) The new feature explained by Alberto is there just to cure these
   problems, *UNFORTUNATELY* it is not yet in the version distributed
   on the web, but only in my development version for the time being.
   It will be surely in in the next release, but don't try it now,
   it will not work.

                         Ciao
                        Alfredo

On Mon, 17 Jun 2002, Alberto Fasso' wrote:

> The period of the FLRNDM random number generator is 2*144 = 10**43.
>
> Now there is no need anymore to use "spaced" random seeds, because it is
> possible to initiate up to 9.E8 independent sequences which will never overlap.
> To this purpose, in the RANDOMIZE card set WHAT(2) = any number < 9.E8.
> By default, this number is 54217137 (the number used as an example in the
> original paper:
> G. Marsaglia and A. Zaman
> A new class of random number generators
> Ann. Appl. Probab. 1, 462-480 (1991)
> but according to the authors, any number can be used (although I am
> superstitious and prefer using some large odd number...)
>
> Alberto
>
>
> On Fri, 14 Jun 2002 tap4@casino.PHYS.CWRU.Edu wrote:
>
> > Hi Stefan,
> >
> > Laura and I have been using the FLRNDM random number generator
> > together with the ranskip files that you sent, for our jobs on separate
> > processors. We are having a problem with several jobs replicating the
> > results of jobs started with earlier ranskip files. For instance, the
> > job started with ranskip005 replicates the results from the latter
> > part of the job started with ranskip004. Even if the random number
> > ranges of the two jobs start to overlap, it is strange that they would
> > be in phase. I don't think the random number generator has finished
> > one or more cycles when this happens. I have an idea about why this is
> > happening and am working to confirm it. However, I need some
> > information about the random number generator and the the ranskip
> > files. The questions I have are:
> >
> > 1. What is the period of the random number generator?
> >
> > 2. What is the spacing between the random seeds in the ranskip files?
> >
> > Please let me know if you have the answers.
> >
> > thank you,
> > Thushara
> >
>
>

-- 
+----------------------------------------------------------------------------+
|  Alfredo Ferrari                ||  Tel.: +41.22.767.6119                  |
|  CERN-SL                        ||  Fax.: +41.22.767.7555                  |
|  1211 Geneva 23                 ||  e-mail: Alfredo.Ferrari@cern.ch        |
|  Switzerland                    ||          Alfredo.Ferrari@mi.infn.it     |
+----------------------------------------------------------------------------+


This archive was generated by hypermail 2.1.6 : Tue Jun 10 2003 - 14:47:42 CEST