more number theorymore number theory

Mon Apr 11 19:56:13 PDT 1994

-----BEGIN PGP SIGNED MESSAGE-----

> What estimates exist for the density of large Carmichael numbers,
> say 1000 bits long?

I'm not sure off hand - maybe Ray can try to check the source of his
formula.

Carmichael numbers must be square free and the product of at least
three primes... I seem to remember a formula for the distribution of
square free integers, but can't quite remember it...

> test? Are other probability tests like Miller-Rabin any more
> provably likely to detect these?

Well Phil, you are in luck!  Miller-Rabin isn't fooled by Carmichael
numbers.  There still is a chance for failure, but it doesn't depend
on the input (i.e. there are no bad inputs for Miller-Rabin like there
are for pseudoprime testing).  Failure depends on how many iterations
you perform (n iterations = 2^-n chance of failure) and the values of
the base you choose.

For example, in Miller-Rabin, the Carmichael number 561 is exposed to
be composite by choosing a base of 7.

I'm familiar with two other primality testing algorithms (I'm no
number theory wiz so there are probably more): Lucas' and Lehmer's.
Well, Lehmer's method is a modification of Lucas' method.  They both
are slow, but have the advantage of being true.