English

On the Liouville function at polynomial arguments

Number Theory 2024-08-19 v4

Abstract

Let λ\lambda denote the Liouville function. A problem posed by Chowla and by Cassaigne-Ferenczi-Mauduit-Rivat-S\'ark\"ozy asks to show that if P(x)Z[x]P(x)\in \mathbb{Z}[x], then the sequence λ(P(n))\lambda(P(n)) changes sign infinitely often, assuming only that P(x)P(x) is not the square of another polynomial. We show that the sequence λ(P(n))\lambda(P(n)) indeed changes sign infinitely often, provided that either (i) PP factorizes into linear factors over the rationals; or (ii) PP is a reducible cubic polynomial; or (iii) PP factorizes into a product of any number of quadratics of a certain type; or (iv) PP is any polynomial not belonging to an exceptional set of density zero. Concerning (i), we prove more generally that the partial sums of g(P(n))g(P(n)) for gg a bounded multiplicative function exhibit nontrivial cancellation under necessary and sufficient conditions on gg. This establishes a "99% version" of Elliott's conjecture for multiplicative functions taking values in the roots of unity of some order. Part (iv) also generalizes to the setting of g(P(n))g(P(n)) and provides a multiplicative function analogue of a recent result of Skorobogatov and Sofos on almost all polynomials attaining a prime value.

Keywords

Cite

@article{arxiv.2010.07924,
  title  = {On the Liouville function at polynomial arguments},
  author = {Joni Teräväinen},
  journal= {arXiv preprint arXiv:2010.07924},
  year   = {2024}
}

Comments

43 pages; further referee comments incorporated; to appear in Amer. J. Math

R2 v1 2026-06-23T19:23:03.554Z