On the geometry of nearly orthogonal lattices

TL;DR

This paper investigates the geometric and optimization properties of well-rounded nearly orthogonal lattices in high dimensions, revealing their packing density landscape, lattice types they contain, and bounds on minimal vectors.

Contribution

It proves the absence of local maxima for the packing density on nearly orthogonal lattices and shows these sets do not include perfect lattices, highlighting their unique geometric structure.

Findings

Sphere packing density has no local maxima on nearly orthogonal lattices for n≥3.

Nearly orthogonal lattices do not contain any perfect lattices for n≥3.

Such lattices have at most 4n-2 minimal vectors.

Abstract

Nearly orthogonal lattices were formally defined in [4], where their applications to image compression were also discussed. The idea of ``near orthogonality" in $2$-dimensions goes back to the work of Gauss. In this paper, we focus on well-rounded nearly orthogonal lattices in~$\mathbb R^n$ and investigate their geometric and optimization properties. Specifically, we prove that the sphere packing density function on the space of well-rounded lattices in dimension $n\geq 3$ does not have any local maxima on the nearly orthogonal set and has only one local minimum there: at the integer lattice~$\mathbb Z^n$. Further, we show that the nearly orthogonal set cannot contain any perfect lattices for~$n \geq 3$, although it contains multiple eutactic (and even strongly eutactic) lattices in every dimension. This implies that eutactic lattices, while always critical points of the packing density function, are not necessarily local maxima or minima even among the well-rounded lattices. We also prove that a (weakly) nearly orthogonal lattice in~$\mathbb R^n$ contains no more than~$4n-2$ minimal vectors (with any smaller even number possible) and establish some bounds on coherence of these lattices.