Re: Uncorrelated random variables

[David Bateman]

Bill Denney wrote:

>On Thu, 29 Jun 2006, Steve C. Thompson wrote:
>
>  
>
>>Others will correct me if I'm wrong (I have only a
>>vague understanding of the exact algorithms used), but
>>I believe Octave uses the Mersenne Twister random
>>number generator to generate uniform random variables,
>>and the Marsalia-Tsang ziggurat method for generating
>>normal random variables.
>>    
>>
>
>I believe that you are correct about the generators, and the Mersenne 
>Twister is uncorrelated for 2^19937-1 integers (aka about 10^6000 
>integers) before it is correlated.  It's good enough for just about any 
>scientific work, but it's not good for cryptography, though, for reasons 
>I'm not qualified to explain.
>
>  
>
Its no good for cryptography as 624 consequetive values in the sequence
in fact define the state vector of the generator, and from then on you
can predict all future values the generator will produce. If you want to
use it for crypto work than it must be hashed... However, as for
uncorrelated-ness, as long as the initial state vectors are independent,
and they will always be if you have /dev/random on your machine, and
essentially always for other systems by using the lsb of the clock, you
have a sequence that should never repeat itself in the lifetime of the
universe.

D.

-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary