2 О псевдослучайных числах
Генераторы псевдослуч. числе -- вещь крайне коварная (великий Джон фон Нейман лажанулся в самом первом своем -- не сумел (!) подойти "математически" (!), занимался рукосуйством).
А. Качество "случайности" (главный спец: George Marsaglia; серия тестов "Diehard")
Б. Длина периода.
Касательно того, что выше, из письма (2004):
Цитата:
.........
2 - Very many 'Platform supplied' random number generators are rubbish.
3 - BlackBox does provide one (as an example) in ObxRandom.
It is due to Lewis, Goodman, & Miller, and is widely cited.
It has a small (31 bit) state vector which, in my opinion,
makes it marginal for large simulations, and inadequate for very large simulations.
It is better that many, and among the best of its type.
For performance details refer to Donald Knuth,
The Art of Computer Programming, Volume 2, Table 1, page 106, line 19.
He describes it as 'adeguate but less outstanding', and '... have known defects'.
4 - There are two other random number generators in module LibRandom
(available from CPC). These are recommended by Knuth, have large
state vectors (upto over 6000 bits), some good proven properties
(eg virtually infinite cycle length), have passed many
empirical tests, and are widely available (any platform with
either c or FORTRAN).
...
Robert Campbell