A generalized PGL(2) Petersson/Bruggeman/Kuznetsov formula for analytic applications
Yueke Hu, Ian Petrow, Matthew P. Young
TL;DR
The paper develops generalized Petersson/Bruggeman/Kuznetsov (PBK) formulas at finite places for PGL2 by employing the adelic relative trace formula and introducing geometric and spectral hypotheses on non-archimedean test-function pairs. It proves a refined PBK identity (theorem theoGeomSpec) that isolates spectral contributions from representations with prescribed local components, and expresses the diagonal term via local Plancherel data, while the geometric side is governed by generalized Kloosterman sums $H(m,n;c)$ with controlled moduli and conductors. The authors provide several explicit local constructions (e.g., supercuspidal and principal-series test functions) that realize the narrow spectral support predicted by the hypotheses, and they demonstrate powerful applications including a harmonically-weighted Weyl-Selberg law, optimal large sieve inequalities, and moments bounds for L-functions. The framework yields precise arithmetic information about the Kloosterman sums and spectral weights, enabling equidistribution statements and robust analytic applications in the non-archimedean setting with potential for broad generalization. The results thus connect local representation-theoretic data with global automorphic phenomena, offering a versatile tool for analytic number theory on GL2/QQ.
Abstract
We develop generalized Petersson/Bruggeman/Kuznetsov (PBK) formulas for specified local components at non-archimedean places. In fact, we introduce two hypotheses on non-archimedean test function pairs $f \leftrightarrow π(f)$, called geometric and spectral hypotheses, under which one obtains `nice' PBK formulas by the adelic relative trace function approach. Then, given a supercuspidal representation $σ$ of ${\rm PGL}_2(\mathbb{Q}_p)$, we study extensively the case that $π(f)$ is a projection onto the line of the newform if $π$ is isomorphc to $σ$ or its unramified quadratic twist, and $π(f) = 0$ otherwise. As a first application, we prove an optimal large sieve inequality for families of automorphic representations that arise in our framework.
