R devel - Feb 2013 - limitations to random number generator in 64-bits machines

Dear List,

Recently I got the comment that the implementation of the random number 
generator used by default in R (Mersenne-Twister) could not be "safe" 
for 64-bits machines, so I decided to put the question here because I do 
not have expertise in that topic, and because this question could be 
"too technical for R-help's audience". I apologise if this is not
the case.

The period 2^19937 - 1 mentioned in the help page of 'RNG' for the 
Mersenne-Twister generator, is it the same for 32-bits machines and 
64-bits ones ?

In addition:

-) If I want to generate two consecutive sequences s_1 and s_2 of n 
pseudo-random numbers each, and knowing how the Random number generator 
is coded, can we estimate in advance the correlation coefficient rho 
between s1 and s2?

-) Let us say that we compute the correlation coefficient rho between 
s_1 and s_2 and find it is not null. How small should it be so that we 
can reasonably use  a statistical analysis that does suppose that the 
sequences are independent ?


Thank in advance for any help you can provide,

Mauricio Zambrano-Bigiarini

-- 
================================================Water Resources Unit
Institute for Environment and Sustainability (IES)
Joint Research Centre (JRC), European Commission
TP 261, Via Enrico Fermi 2749, 21027 Ispra (VA), IT
webinfo    : http://floods.jrc.ec.europa.eu/
================================================DISCLAIMER:\ "The views
expressed are purely those of th...{{dropped:10}}

AFAICS, these are statistics, not R, issues, and are completely off
topic here. You should post on a statistics list, such as
stats.stackexchange.com, instead.

Cheers,
Bert

On Thu, Feb 21, 2013 at 5:20 AM, Mauricio Zambrano-Bigiarini
<mauricio.zambrano at jrc.ec.europa.eu> wrote:
> Dear List,
>
> Recently I got the comment that the implementation of the random number
> generator used by default in R (Mersenne-Twister) could not be
"safe" for
> 64-bits machines, so I decided to put the question here because I do not
> have expertise in that topic, and because this question could be "too
> technical for R-help's audience". I apologise if this is not the
case.
>
> The period 2^19937 - 1 mentioned in the help page of 'RNG' for the
> Mersenne-Twister generator, is it the same for 32-bits machines and 64-bits
> ones ?
>
> In addition:
>
> -) If I want to generate two consecutive sequences s_1 and s_2 of n
> pseudo-random numbers each, and knowing how the Random number generator is
> coded, can we estimate in advance the correlation coefficient rho between
s1
> and s2?
>
> -) Let us say that we compute the correlation coefficient rho between s_1
> and s_2 and find it is not null. How small should it be so that we can
> reasonably use  a statistical analysis that does suppose that the sequences
> are independent ?
>
>
> Thank in advance for any help you can provide,
>
> Mauricio Zambrano-Bigiarini
>
> --
> ================================================> Water Resources Unit
> Institute for Environment and Sustainability (IES)
> Joint Research Centre (JRC), European Commission
> TP 261, Via Enrico Fermi 2749, 21027 Ispra (VA), IT
> webinfo    : http://floods.jrc.ec.europa.eu/
> ================================================> DISCLAIMER:\ "The
views expressed are purely those of th...{{dropped:10}}
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel


-- 

Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: 467-7374
Website:
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm