Re: Randomness as a computational resource

From: none (_at_(none))
Date: 07/03/05


Date: Mon, 04 Jul 2005 08:06:43 +1200

Douglas A. Gwyn wrote:
> A source of "randomness" (entropy) *is* useful at times,
> for example in stochastic simulations or generating crypto
> keys. (The former application can usually get by with a
> "pseudo"random source, i.e. a completely deterministic one.)

In fact we prefer a pseudo random number generator. Why? Well one reason
is if--heaven forbid, there is a bug that comes up in a test. If you use
real random numbers duplication the bug is rather difficult... With a
pseudo rng its trivial.

But some simulations are sensitive to the lack of randomness esp in LCC
generators. I usually test with a non-linear rng as well to show that it
makes no difference. In fact lately i use the full 2^48 length of the
Java LCC, and have needed to use the non-linear one anyways. I use AES
in counter mode. Its only five times slower than a LCC, which is fast
for non-linear. Some use the BBS generator, but i find it far to slow.

Some stochastic Algo's use the unrandomness of pseudo RNG to improve
there convergent properties. In particular stochastic integration can
improve on the base performance by using a LFSR m-sequence to speed it
up quite a bit (numerical recipes in C has some details).

greg