Boundary Delocalization and Spectral Packets for Dirichlet Eigenfunctions
Anton Alexa
TL;DR
On a bounded $C^ abla$ strictly convex domain $\Omega$, high-frequency Dirichlet eigenfunctions cannot accumulate most of their boundary energy into a single mode when averaged over a sublinear spectral window. The authors develop a mode-to-packet comparison using the Rellich identity on the boundary and, for the multi-mode case, the boundary local Weyl law to obtain zero-mean cancellation. They prove that for any $N_k\to\infty$ with $N_k=o(k)$, $E_k/\sum_{m=k}^{k+N_k-1}E_m\to 0$, and that no positive fraction of modes in a sublinear packet can maintain a fixed zero-mean boundary bias, independent of eigenvalue monotonicity and stable under crossings. This boundary-delocalization principle provides a deterministic tool for boundary spectral statistics, highlighting that averaging over short spectral packets controls single-mode boundary concentration without invoking microlocal or dynamical assumptions.
Abstract
We establish a boundary delocalization principle for high-frequency Dirichlet eigenfunctions on smooth strictly convex domains. The main result excludes persistent boundary concentration at the level of individual eigenmodes when compared to short spectral packets of sublinear length. Quantitatively, we compare boundary energies of single eigenfunctions to packet sums over frequency windows of size N_k = o(k), without asserting any asymptotic gain in magnitude. The main mode-to-packet estimate relies only on the Rellich identity. For the multi-mode bias exclusion we additionally use the boundary local Weyl law to obtain a packet zero-mean cancellation estimate. This mode-to-packet comparison is independent of eigenvalue monotonicity and is stable under eigenvalue crossings.
