8
$\begingroup$

Let $\Lambda \subset\mathbb{R}^n$ be a lattice satisfying $\|x-y\|_2^2 \in \mathbb{Z}$ for all $x,y\in\Lambda$. How small can $\text{vol}(\Lambda)=\det(\Lambda)$ be?

For example, in dimension $2$, the hexagonal lattice has smallest volume of any integral lattice, equal to $\frac{\sqrt{3}}{2}$. In higher dimensions, rescaling an even unimodular lattice by a factor of $\sqrt{2}$ yields a lattice with integral distances with $\text{vol}(\Lambda) = 2^{-\frac{n}{2}}$.

What is the smallest volume possible as $n\rightarrow\infty$?

$\endgroup$
2
  • 4
    $\begingroup$ Both your examples satisfy only $\langle x,x \rangle \in \mathbb Z$, not $\langle x, y\rangle \in \mathbb Z$ - is that what you meant to ask? $\endgroup$
    – Will Sawin
    Commented Apr 19, 2022 at 21:06
  • $\begingroup$ @WillSawin: Thank you, yes I meant $\|x-y\|_2^2 \in \mathbb{Z}$ rather than $\langle x,y\rangle\in\mathbb{Z}$. If $\Lambda$ is an even unimodular lattice, then $\|x-y\|_2^2 = \langle x,x\rangle+2\langle x,y\rangle+\langle y,y\rangle \in 2\cdot\mathbb{Z}$, and so the $\frac{1}{\sqrt{2}}$ scaling has $\|x-y\|^2 \in \mathbb{Z}$. $\endgroup$ Commented Apr 19, 2022 at 21:38

2 Answers 2

10
$\begingroup$

After scaling your lattice by $\sqrt{2}$, the Gram matrix has integral entries, so the absolute value of its determinant, being a nonzero integer, is at least $1$. This gives a lower bound of $2^{-n/2}$.

$\endgroup$
3
  • 1
    $\begingroup$ And this bound is achieved whenever n is divisible by 8. See: en.wikipedia.org/wiki/Unimodular_lattice $\endgroup$ Commented Apr 20, 2022 at 0:22
  • 1
    $\begingroup$ What about an upper bound? If $n$ is not divisible by 8, then $D_n$ is above the lower bound by a factor of $2$. When $n$ is 1 mod 8, then $kE_8\oplus A_2$ is above by a factor of $\sqrt{3}$. Can this be improved? $\endgroup$
    – Ben Mares
    Commented Apr 20, 2022 at 8:00
  • 1
    $\begingroup$ @BenMares The way I understood the question, the OP wanted an order of magnitude. But I agree that the question of the exact bound is not solved for all $n$. $\endgroup$
    – Aurel
    Commented Apr 20, 2022 at 8:36
5
$\begingroup$

Note that $\sqrt2\Lambda$ is even, and $\mathrm{vol}(\sqrt2\Lambda)^2=\det(A)$, where $A$ is the Gram matrix of $\sqrt2\Lambda$. Thus $\mathrm{vol}(\Lambda)=2^{-n/2}\sqrt{\det(A)}$, so the smallest possible volume in dimension $n$ is determined by the smallest determinant of an even $n$-dimensional lattice.

The smallest determinant question is answered in the lemma of section 5 in Conway & Sloane's Lattices with Few Distances. The answer is $8$-periodic in $n$, given by

$n\ (\mathrm{mod}\ 8)$ $0$ $1$ $2$ $3$ $4$ $5$ $6$ $7$ $8$
$\Lambda_0$ $\{0\}$ $A_1$ $A_2$ $D_3$ $D_4$ $D_5$ $E_6$ $E_7$ $E_8$
$\det A$ $1$ $2$ $3$ $4$ $4$ $4$ $3$ $2$ $1$

where the minimum for a particular $n$ is achieved (usually not uniquely!) by the orthogonal sum of $\Lambda_0$ with some number of copies of $E_8$.

That one cannot do better follows from the classification of even forms of small determinant, SPLAG 15.8, table 15.4: if $\det A<4$ then the mod-8 signature is $\pm (\det A-1)$.

Returning to the original question, this shows \begin{align} \liminf_{n\to\infty}\ 2^{n/2}\mathrm{vol}(\Lambda) =1,\\ \limsup_{n\to\infty}\ 2^{n/2}\mathrm{vol}(\Lambda) =2. \end{align}

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .