Re: Mersenne Twister range in rand.oct and randn.oct

[Mike Miller]

On Sat, 3 Dec 2005, Paul Kienzle wrote:

> On Dec 1, 2005, at 9:58 AM, Mike Miller wrote:
>
> > On Thu, 1 Dec 2005, Mike Miller wrote:
> >
> > > In one Mersenne Twister FAQ I see this:
> > >
> > >   2002-version mt19937ar.c has versions [0,1],[0,1),[0,1),(0,1) (32-bit
> > >   precision), and [0,1) with 53-bit precision.
> > >
> > > Which am I getting when I use rand.oct in Octave forge?  What I would 
> > > really like to know are the largest and smallest values that can be 
> > > produced both by rand.oct and by randn.oct.
> >
> >
> > I should say that it's really one .oct file:  rand.oct.
> > randn is derived from rand.oct.
> >
> > I took a look at Octave Forge's rand.cc but I can't find in it what I need 
> > to know (because of my incompetence, obviously).
>
> The answer is in Makefile rather than rand.cc.  This sets the ALL_BITS flag 
> for building randmtzig.c.  Here's the relevant code:
>
> /* generates a random number on (0,1) with 53-bit resolution */
> USE_INLINE double randu53(void)
> {
>    const uint32_t a=randi32()>>5;
>    const uint32_t b=randi32()>>6;
>    return(a*67108864.0+b+0.4) * (1.0/9007199254740992.0);
> }


OK.  Thanks.  I guess "(0,1)" means that neither 0 nor 1 will ever be 
generated as a random uniform, but does that last line of the code tell me 
the largest and smallest possible values for a random uniform?  I can't 
figure out what the largest and smallest numbers would be.

Is it difficult to determine the range of possible values for the normal 
deviates (randn with mean 0 and variance 1 in Octave Forge)?

Thanks.

Mike



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------