Work

Mean squares and oscillations of the error term for the Jacobi-symbol sums Σ_{x mod n} ((x²−m)/n)

Year: 2026
Type: preprint
Abstract: For odd n let S(n) = Σ_{x mod n} ((x²−1)/n) (Jacobi symbol). The function S is multiplicative, sign S(n) = λ(n) is the Liouville function, and Σ_{n≤x, n odd} S(n) = C·x^{3/2} + E(x) with C = 0.0586383825…, E(x) ≪ x log³x and E(x) ≠ o(x). We prove that the error term has logarithmic mean square exactly 1/π²:∫₁^Y (E(x)/x)² dx/x = (1/π²)·log Y + o(log Y).More generally, for squarefree m ≠ 0 and S_m(n) = Σ_{x mod n} ((x²−m)/n), the logarithmic mean square equals (1/π²)·∏_{p|m, p>2} 2p(p−1)/(p²+1); the values χ_m(p) do not enter the constant. The proof uses Plancherel's formula, the functional equation of ζ, and a uniform mean value lemma on the line Re s = 1. For m = 1, assuming the Riemann hypothesis, simplicity of the zeros and a mild bound for Σ|ζ′(ρ)|^{−3}, the logarithmic mean square of R₂(X)/(X^{3/4} log X) for the Riesz-smoothed error term equals Λ = 2·Σ_{γ>0}|B_ρ|² = 7.507…·10⁻⁶, an analogue of Ng's theorem for the Mertens function. Unconditionally, for every ν ∈ (0,1), R₂ has ≫ log T sign-alternating oscillations of size X^{3/4}(log X)^ν in [2,T], because every zero of ζ produces a pole of order at least two. Adapting Montgomery's method for Euler's function, we prove E(x) = Ω±(x log log x), and E_m(x) = Ω±(x √(log log x)) for squarefree m ≠ 1. Numerical data up to 4·10⁷ agree with the results. Version 4: adds Theorem D_m (the family m ≠ 1), Section 5.1 and Remarks (3)–(4); corrects the constant C₃.
Language: English
Open Access status: green
OpenAlex ID: W7233949792
Cites:
Cited by: 1
Related to: 10