Sparse supports of lattice eigenfunctions: quantitative growth and algebraic rigidity

Abstract

We study how sparsely a nonzero discrete harmonic function on the standard lattice ${\mathbb{Z}}^d$ can be supported. Let $Q_n^{(d)}=\{-n,\cdots ,n{\}}^d$, and let $m_d(n)$ denote the least possible value of $|\mathrm{supp}(u)\cap Q_n^{(d)}|$ among discrete harmonic functions $u:{\mathbb{Z}}^d\to \mathbb{C}$ with $u(0)\ne 0$. For all $n\ge 1$, we prove \begin{equation*} m_3(n)\asymp n^2, \quad c_dn^{d^2/(2d-1)} \leq m_d(n)\leq (2n+1)^{\lfloor d/2\rfloor+1} \quad d\ge 4. \end{equation*} These estimates extend the two-dimensional support estimate of Buhovsky, Logunov, Malinnikova, and Sodin [Duke Math. J. 171 (2022), 1349--1378] to higher dimensions and obtain sharpness in dimension three. For $d\ge 4$, the lower exponent and the upper one differ by less than $3/4$ in even dimensions and $1/4$ in odd dimensions. The proof combines Hilbert functions of finite support sets with a position-translation uncertainty principle. The sharp three-dimensional bound additionally uses Cayley--Bacharach relations and rigidity of algebraic curves. Finally, for every nonzero lattice eigenfunction with eigenvalue $\lambda$, the Zariski closure of its full support has dimension at least $\lceil d/2\rceil$, and at least $\lfloor d/2\rfloor +1$ when $\lambda \ne 0$. Both bounds are optimal. All proofs resulted from human-guided exploration by GPT-5.6 Sol in Ultra mode and checked by the author.

33 pages. The previous multiscale and endpoint collision arguments in the first version have been replaced by a nearly optimal Hilbert function approach. Material included only for proofreading has been removed