My question is: Is there a (possibly probabilistic) prime test with sub-logarithmic runtime complexity? Is it possible to construct one?
I have found the following complexities for the most common tests.
- fermat test has O(k*log2n *log log n * log log log n)
- elliptic curve tests with O($log^6$)
- cyclotomy test with O($log^{(c * log log log)}$)
- Miller-Rabin has O($log^4$)
- AKS has O($log^{12}$) improved to O($log^{7.5}$) and then to O($log^6$)
thank you.
For generic integers, it's not even possible to test whether it's a multiple of $3$ in sub-logarithmic time. (If you've given the base-$10$ or base-$2$ digits of the integer $p$, for example, you need to examine every single digit to answer the question, and there are $\asymp \log p$ digits.) So the answer is definitely no.