dx50azlm
September 11th, 2002, 07:07
I still like probablistic algorithms because if you run them for 20 or so iterations (assuming CPU cycles are cheap these days), an error of 1/(2^1000000) is so close to zero, it might as well be zero

The ECPP test is one such test that works like this, but I'm very interested in the new polytime algorithm by Agrawal, Kayal and Saxena (the authors of the paper from the first post).
Their result lies on using Fermat's Little Theorem. A quick rundown of the proof is at
h**p://w*w.utm.edu/research/primes/prove/prove4_3.ht*l