Thread Subject: randn algorithm

Hi,

I currently convert a Matlab simulation code to Fortran and
need to generate normally distributed random numbers in the
process. Unfortunately, the random numbers provided by the
generators I have at my disposal in Fortran seem to have
slightly different properties than the ones generated by
Matlab's "randn"

The question is, which generate in Intel MKL would provide
random numbers most similar to those provided by randn
(which, I take from "doc randn", uses the ziggurat algorithm)?

I am currently using VSL_BRNG_MCG31 as the base generator
and BOXMULLER as the distribution generator. There are a
bunch of different possibilities as for both the base and
the distribution generator, so I am clueless here.

It would be most useful to know if Matlab uses some external
library (and which one) in generating normal random numbers.

Thanks!

I shall qualify the question:

- does randn use the same base generator as rand (i.e.,
Mersenne Twister in 7.4)?
- what would be a good newsgroup to ask what differences to
expect between using the Box-Muller transform as opposed to
ziggurat?

"Joachim " <a@b.com> wrote in message
<g0a8rf$obt$1@fred.mathworks.com>...
> I shall qualify the question:
>
> - does randn use the same base generator as rand (i.e.,
> Mersenne Twister in 7.4)?
> - what would be a good newsgroup to ask what differences to
> expect between using the Box-Muller transform as opposed to
> ziggurat?

You might try:

sci.math.num-analysis

James Tursa

Joachim wrote:
> I shall qualify the question:
>
> - does randn use the same base generator as rand (i.e.,
> Mersenne Twister in 7.4)?

No, it uses something that is simpler, and in this context faster, It's related
to but not the same as the one that Marsaglia's original paper described -- it
has a longer period, for one thing.


> - what would be a good newsgroup to ask what differences to
> expect between using the Box-Muller transform as opposed to
> ziggurat?

sci.stat.math perhaps? Statistically, I don't think you need to worry about the
differences between those two transformation methods. The uniform generator
that you use underneath them is probably more important.