Prove that there are at most finitely many random (Kolmogorov random) prime numbers

92 Views Asked by At

Prove that there are at most finitely many random (Kolmogorov random) prime numbers. A number is Kolmogorov random if

$$K(n) ≥ K(\operatorname{bin}(n)) ≥ \lceil \log_2(n+1)\rceil - 1.$$

Hint: You can use the prime number theorem.

I tried to prove it by contradiction but I couldn’t connect the prime number theorem and the definition of Kolmogorov randomness.

1

There are 1 best solutions below

0
On

You need to use both the prime number theorem and the fact that there exists a Turing machine computing the function $p_n$ given by the $n^{th}$ prime. This means the Kolmogorov complexity of a prime $p = p_n$ is bounded by the Kolmogorov complexity of $n$ plus a constant (for the size of this Turing machine). The PNT gives that $n \approx \frac{p}{\log p}$ from which it follows that $K(p)$ is approximately bounded by

$$K \left( \frac{p}{\log p} \right) + C \le \left\lceil \log_2 \frac{p}{\log p} \right\rceil + C$$

This grows like $\log_2 p - \log_2 \log p + C$ and as $p \to \infty$ the $\log_2 \log p$ term eventually (if very slowly) overtakes the constant term. So we conclude that sufficiently large primes are not Kolmogorov random, because they can be described by the Turing machine computing $p_n$ + the index $n$, which eventually becomes small enough compared to $p_n$ to compress it nontrivially.

This argument is very general and shows more generally that there are only finitely many Kolmogorov random elements of any increasing computable sequence $a_n$ such that $\lim_{n \to \infty} \frac{a_n}{n} = \infty$, or something like that.