$L^1$ means of exponential sums with multiplicative coefficients. II
Mayank Pandey, Maksym Radziwiłł
TL;DR
The paper proves a sharp rigidity phenomenon for real-valued, 1-bounded multiplicative functions: if the L^1 mass of the exponential sum $\sum_{n\le N} f(n) e(n\alpha)$ is sufficiently small, then on large primes $p$ the function $f(p)$ must approximate a quadratic Dirichlet character $\chi(p)$, with the conductor bounded in terms of the L^1 bound. The authors develop a uniform, major/minor arc framework, employ Gallagher's lemma and Mellin transform techniques to pass from additive to multiplicative structures, and prove a Halász-type large sieve bound controlling the major arcs via a pretentious distance $M_{f,\psi,T}$. Beyond the real-valued case, they extend to L^p norms and derive a general, uniform obstruction result: if $f$ is close to a non-quadratic character on primes, the pretentious distance cannot stay small; this yields uniform classification results and corollaries for automorphic L-functions. The work provides sharp, scale-uniform obstructions to additive structure in multiplicative sequences and offers new avenues for understanding when multiplicative coefficients mimic characters on primes.
Abstract
Let $f$ be a real-valued $1$-bounded multiplicative function. Suppose that the mean-value of $f^{2}$ exists, and $$\int_{0}^{1} \Big | \sum_{n \leq N} f(n)e^{2πi n α} \Big | d α\leq N^{o(1)}$$ as $N \rightarrow \infty$, then there exists a quadratic character $χ$ such that for every $δ> 0$ the (logarithmic) proportion of primes $p \leq N$ such that $|f(p) - χ(p)| < δ$ tends to $1$ as $N \rightarrow \infty$. More generally we show that for all $N, Δ\geq 1$ and $1$-bounded multiplicative functions $f$, if $$\int_{0}^{1} \Big | \sum_{n \leq N} f(n) e^{2πi n α} \Big | d α\leq Δ$$ and the $L^{2}$ norm of $f$ over $[1, N]$ is $\geq N / 100$, then $f$ pretends to be a multiplicative character of conductor $\leq Δ^{2}$ on primes in $[Δ^{2}, N]$. We highlight that the result is uniform in $f$, $N$ and $Δ$ and sharp as far as the size of the conductor goes. Moreover, the restriction to primes $p \in [Δ^{2}, N]$ turns out to be sharp in a suitably generalized version of this result, concerning sequences $f$ that are close $1\%$ of the time to multiplicative functions.
