Date   : Mon, 08 Apr 1985 08:50:46-EST
From   : Lee.Sailer@CMU-CS-C.ARPA
Subject: More randomness

Cuccia's (non) solution is an example of my problem.  I call it a non-solution
because (1) it defines a constant M >> the upper limit allowed in most Z80
Pascals, and (2) the statement x := (x*A) mod M is eventually gonna overflow.

Another problem: A and M relatively prime is NOT sufficient to guarantee good
statistical properties of a rn generator.  

So, in more direct form, here is my problem:  I am writing a journal article
that includes a monte-carlo statistical simulation.  I know that the journal
@i(requires) a "good" rn generator--most people use IMSL.  A good rn generator
is one that passes a wide variety of statistical tests for randomness. 
Therefore, I must now find a "good" generator for the Z80, or port my program
over to a mainframe with IMSL.

I have nothing against linear congruential generators as suggested by Cuccia,
but I gotta know what the right constants are for my z80, and I hafta have
a citation that says they are the right ones, else the journal will laugh
and say, "resubmit when you have duplicated your results on a real computer."

One last point--I have tested the built in random in Turbo Pascal, and it sure
isn't random.

lee