Re: KISS 64-bit pseudo-random number generator

On Thu, 19 Sep 2024 0:28:33 +0000, Krishna Myneni wrote:

On 9/12/24 21:10, Krishna Myneni wrote:

[..]

A bit more involved test of the same 64-bit PRNGs starts to show the
possibility of defects in Marsaglia's 64-bit kiss prng (RAN-KISS)
compared to a simple 64-bit LCG prng (RANDOM).

[..]
I wonder if it is at all possible to really prove something
about the PRNG *with tests of this type*. Intuition wants us to
believe that the longer we run the simulation, the closer
the result must match the expected outcome. Shouldn't we
compute the probability that after a certain size run the
result does NOT match the known result (given an ideal PRNG),
or how unlikely it is that the result has a given error?
Example: say the result of PRNG-a consistently has one of
its bits (say bit 0) stuck at zero. Would the test under
consideration detect this specific problem at all?
-marcel