New Appoach for Random Number Generation???

[ttw]

I think (though it needs some analysis) that numeric noise is rather predictable. Given any two input floating point numbers, the roundoff error is easily obtainable. Knuth has a pretty good analysis in Volume 2.