Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!hellgate.utah.edu!hamblin.math.byu.edu!sol.ctr.columbia.edu!lll-winken!uunet!ogicse!milton!seymour
From: seymour@milton.u.washington.edu (Richard Seymour)
Newsgroups: comp.lang.fortran
Subject: Re: randu?
Message-ID: <17397@milton.u.washington.edu>
Date: 28 Feb 91 21:10:43 GMT
References: <62349@eerie.acsu.Buffalo.EDU>
Organization: University of Washington, Seattle
Lines: 28

What about RANDU?
<my comments in "<...>" brackets>
source: vax fortran user's guide, aa-d035c-te, for vms v3.0, fort v3.0)
page D-12  "Compatibility: VAX-11 Fortran and PDP-11  Fortran"

D.3.9  RANDU subroutine
The RANDU subroutine computes a pseudorandom numvber as a single-precision
value uniformly distributed in the range:  0.0 .LE. value .LT. 1.0

<thus from 0 to just-less-than One>

a call to randu has the form:   CALL RANDU (i1,i2,x)
where:  i1, i2 are INTEGER*2 variables or array elements that contain the
  seed for computing the <next> random number
        x is a real variable or array element where the computed random
           number is stored.

notes: 1) the values of i1 and i2 are updated during the computation
   2) the algorithm is:
       if I1=I2=0 set generator base:  x(n+1)=2**16+3
       otherwise:  x(n+1)=(2**16+3)*x(n) mod 2**32
      store generator base x(n=1) in i1,i2

     result is x(n+1) scaled to a real value y(n=1) for 
         0.0 .LE. y(n=1) .LT. 1.0

<p.s. to people copying this routine... the 2**16+3 is MACHINE DEPENDANT!
   other values make better distributions for IBM Pc's etc.>