Processing math: 100%

Expansion in a prime divisibility graph

Joint with M. Radziwill

Let N=Z(N,2N] and P[1,H] a set of primes with Hexp(logN/2). Given any subset XN, define the linear operator (A|Xf)(n)=pP:p|n;n,n±pXf(n±p)pP;n,n±pXf(n±p)n on functions f:NC. Let L=pP1p .

We prove that, for any C>0, there exists a subset XN of density 1O(eCL) in N such that A|X has a strong expander property: every eigenvalue of A|X is O(L). It follows immediately that, for any bounded f,g:NC, 1NL|nN;pP:p|nf(n)¯g(n±p)nN;pPf(n)¯g(n±p)p|=O(1L)

This bound is sharp up to constant factors.

Specializing the above bound to f(n)=g(n)=λ(n) with λ(n) the Liouville function, and using a result in (Matomäki-Radziwiłł-Tao, 2015), we obtain 1logxnxλ(n)λ(n+1)n=O(1loglogx), improving on a result of Tao’s. Tao’s result relied on a different approach (entropy decrement), requiring H(logN)o(1) and leading to weaker bounds.

We also prove the stronger statement that Chowla’s conjecture is true at almost all scales with an error term as in (2), improving on a result by Tao and Teraväinen.

Harald Helfgott