A Random Question

[Paul Dale]

(06-11-2017 01:39 AM)DavidM Wrote:  Does this have a long enough cycle for a practical use on calculator-based apps?

I think that pretty much any RNG has a long enough cycle. How many random numbers is a calculator likely to need? A few hundred? Thousand? Million? Most modern RNG have periods many orders of magnitude beyond what's required.

It is more important to provide a good RNG that generates uniform numbers that pass the various statistical test suites. This doesn't mean the results are random of course, but it is indicative that they aren't needlessly biassed.

Quote:I have no data to back this up, but I'd be willing to bet that the "RAND n * CEIL" approach is used far more frequently than any other in calculator programs.

I certainly wouldn't be accepting this bet

Quote:Has this actually caused problems for anyone? I suspect we'll never really know.

It is impossible to know. Quite a few of the early RNGs were likely to have been poorly distributed and non-uniform. I'm guilty of this when I've had to reduce the step count in programs.


I've gone back and looked at the 34S RNG again. It looks fine in integer mode and double precision real mode. I suspect it will show some bias in single precision real mode.


Pauli