Divisors of Fourier coefficients of two newforms (Q6182832)

Let \(f\) and \(g\) be non-CM holomorphic newforms with integer Fourier coefficients \(a(n)\) and \(b(n)\), respectively. The forms \(f\) and \(g\) can have different levels and different weights, but the weights must be at least \(2\). This paper studies the numbers of prime divisors of differences \(a(p) - b(p)\) as \(p\) ranges across primes up to some bound \(X\). The primary result is their Theorem 1.1, which states that if \(k\) is the maximum of the weights of \(f\) and \(g\) (and where \(f\) and \(g\) are not character twists of each other), then under GRH they have that \[ \big \lvert \{ p \leq X : a(p) \neq b(p), \omega\big(a(p) - b(p) \big) \leq 7k + \frac{1}{2} + k^{1/5} \} \big \rvert \gg \frac{X}{\log^2 X}. \] Here, \(\omega(n)\) counts the number of distinct prime divisors of \(n\). The authors also show that analogous work proves a similar result with \(\Omega\), which counts the number of prime divisors with multiplicity, in place of \(\omega\). The restriction for \(f\) and \(g\) to not be character twists of each other is natural. In the trivial case when \(f = g\), clearly \(\omega(a(p) - b(p)) = 0\) always; if \(f\) and \(g\) are related by character twists, then a positive proportion of the time, \(\omega(a(p) - b(p)) = 0\). These observations are closely related to multiplicity one results. For example, if \(a(p) = b(p)\) for all but finitely many primes, then \(f = g\). The main theorem can be used to give multiplicity one results based on counting prime divisors of differences. This naturally has a statement in terms of congruences, as \(\ell \mid \big(a(p) - b(p) \big) \iff a(p) \equiv b(p) \bmod \ell\). Corollary 1.4 in this paper gives the corresponding result, which states that if there are primes \(\ell_1, \ldots, \ell_n\) with \(n > 7k + \frac{1}{2} + k^{1/5}\) such that \(a(p) \equiv b(p) \bmod \ell_i\) for all \(p\) with at most \(o(X/\log^2 X)\) exceptions, then in fact \(f\) is a character twist of \(g\). In order to prove these results, the authors apply a one-dimensional sieve. As with most sieves, the most challenging aspect is understanding the ``local densities'' and their associated remainders. In this paper, the authors study aspects of the mod \(\ell\) Galois representations associated to \(f\) and \(g\). As traces of Frobenius elements give the Fourier coefficients \(a(p)\) and \(b(p)\), it suffices to study how often traces of certain matrices in \(\mathrm{GL}(2, \mathbb{Z}/\ell \mathbb{Z})\) agree. The authors show that there is a strong analogy with the Chebotarev density theorem in \S2.