UPDF AI

Multi-scale Vandermonde test kernels for spectral trace formulas

Stefan Horvath

2026 · ArXiv: 2602.11205
0 Citations

Abstract

We construct a family of test kernels for use in spectral trace formulas on locally symmetric spaces. The key innovation is the factorization hT=gT⋆g~Th_T = g_T \star \widetilde{g}_T, which simultaneously achieves: (i) automatic positive semi-definiteness of the spectral multiplier mhT(π)=∣mgT(π)∣2≥0m_{h_T}(\pi) = |m_{g_T}(\pi)|^2 \ge 0; (ii) JJ-fold moment annihilation via a multi-scale Vandermonde construction, yielding super-polynomial decay of all error terms; (iii) uniform spectral parameter bounds (Master-Bound) Etot(T)≪Td+1−δ\mathfrak{E}_{\mathrm{tot}}(T) \ll T^{d+1-\delta} with δ>0\delta>0 depending only on the symmetry order kk and the annihilation depth J≍(log⁡T)/kJ \asymp \sqrt{(\log T)/k}, representing a power saving over the main term ≍Td+1\asymp T^{d+1}. The cost is a controlled polynomial growth Tc02/2+o(1)T^{c_0^2/2+o(1)} in the Vandermonde coefficients (with exponent strictly less than 1), which is dominated by the super-polynomial decay of the off-diagonal terms. The construction is axiomatized over two analytic hypotheses -- a Weyl law and Bessel/Airy asymptotics -- making it applicable beyond the classical GL(2)\mathrm{GL}(2) setting.