New PRNG algorithms for calculators

[Namir]

(02-18-2023 01:29 AM)robve Wrote:  

(02-17-2023 08:21 PM)robve Wrote:  Thanks for sharing your in-depth study. Very interesting from a historical perspective!
I didn't see the "983" algorithm results in the links, so perhaps something to add in the future.

In addition, perhaps it should be mentioned whether these PRNG pass the spectral test? And other tests such as the Chi-squared test, as mentioned in an earlier post on how bad some PRNG are (e.g. Sharp PC-1350) when plotting random (x,y) points.

A colleague of mine is an expert on this subject and has written and published about the challenges. Even more challenging is to design Parallel PRNG and quasi-RNG, an important subject in high-performance parallel Monte Carlo simulations.

- Rob

Part 1 lists the Matlab code that calculates the penalty function. This function performs several statistical tests and obtains results (see HHC2017 video). for each test, it compares the absolute difference of theoretic value minus the calculated result. It then multiplies that result by a weight factor. The penalty factor returns the sum of the weighted absolute differences between the theoretical values and the calculated statistics. The small the value, the better is the quality of the random number s generated.

Namir