New PRNG algorithms for calculators

[Namir]

In 2002 I took up a project, I called Project 997, to study random number generators for calculators. I started with algorithms that used floating points (limiting the random numbers to 10 digits using MATLAB for the calculations). I based my HHC 2022 presentation for PRNGs based on that work. When I returned home, I added more floating-point-based PRNGs and studied variants of other well-known integer-based PRNGs like Wichmann-Hill, ACORN, and MRG32a. The project required running several laptops daily and for many hours/day and for many weeks.

The approach I used the penalty factor that I had presented at HHC2017 and is a singly number that indicates the goodness of a batch of random numbers. The method I used included three stages:
1) Finding the least penalty factor for a relatively wide range of coefficients. This step determines the values of the optimum coefficients.
2) Finding the least penalty factor for a relatively narrow range of coefficients, based on the best coefficients obtained in step 1.
3)Generate batches of one million random numbers using random seeds and the coefficients obtained in stage 2.

I wrote a seven-part study (plus a summary of best results) and posted on my website. You can click here to access the page that lists all the parts of Project 997.

Enjoy!

Namir