(j3.2006) (SC22WG5.4116) [ukfortran] Parallel random numbers

[N.M. Maclaren]

On Oct 22 2009, Bill Long wrote:
>Aleksandar Donev wrote:
>> 
>>> Specifying what is defined and what is not is very hard when
>>> you need to allow for two different models.
>>
>> Perhaps then we should pick one of the models, namely, the independent 
>> sequence one.
>
>Yes, I'm convinced this is the only practical option.

But that's precisely what we CAN'T do!  Requiring independence is just
far too much to demand of implementors.  What we could require is complete
DEPENDENCE, which would make the generators useless for many purposes.

I think that what you actually mean is separate copies of probably
identical generators, perhaps seeded differently.  We certainly shouldn't
forbid that.

>> 3) No restrictions on calling RANDOM_SEED and RANDOM_NUMBER ala ordering 
>> of segments.
>
>This approach is certainly the easiest to implement. It also makes the 
>most sense when trying to write portable code.

I will draft something along these lines, that leaves maximum flexibility.

Regards,
Nick.