Re: Dig. Cash Question.

[smb@research.att.com]

	 > Anyway -- suppose that in some group, you know that a^n=b, where a
	 > and b are members of the group, and n is an integer.  a^n indicates
	 > the group operation iterated n times.  The discrete log problem is
	 > recovering n, given ``a'' and a^n=b.
	 > 
	 > In some groups, this is a very hard problem.  The group most commonl
	y
	 > used in cryptography is the field GF(p), i.e., the field of integers
	 > modulo p, where p is some large number, preferably a prime, and ``a'
	'

	 If I understand this correctly, if p is not a prime, then n may not be
	 unique.

Well, n isn't unique even if p is prime.  Consider a=10,p=11.
10^2=10^4=10^6=10^8=10^10=1 mod 11.  You only get a maximum-length
cycle if ``a'' is a primitive root, hence the restriction I stated
in the part I deleted...

It doesn't matter that n isn't unique, though you do want a good
distribution.  Primitive roots have a maximal distribution, which is
why they're good.  But a reduction by, say, a factor of 2 doesn't
matter in practice.  (For p=11, try a=3.)  The implementation of
secure RPC in SunOS uses Diffie-Hellman (which relies on the
difficulty of the discrete log problem) with base that's not a
primitive root.  To be sure, their key exchange was cryptanalyzed,
but that's because they picked a 192-bit modulus, not because of
the exponentiation base.

If I recall correctly, if p=kq+1, for q a prime and k a small integer,
there are (q-1)/k primitive roots in GF(p).  That suggests generating
p=2q+1, p and q prime, which gives a very good density.  And checking
if a number is a primitive root is easy (again, to my recollection;
I'm not a number theorist) if you know the factorization of p-1, which
of course we do in this case.