Every distance squared between two points in $\mathbb{Z}^n$ is clearly all $\in \mathbb{N}$. Is the converse true (upto isomorphism)?
The title was rather informal, so clarifying the question:
Is every countably infinite set $S \subset \mathbb{R}^n$ such that
$$\forall v_1, v_2 \in S, v_1 \neq v_2 \rightarrow d(v_1, v_2)^2 \in \mathbb{N}$$
subset of $L + \delta$, where $L$ is some lattice in $\mathbb{R}^n$ and $\delta \in \mathbb{R}^n$
?
I suspect it would be true, here is some approach I've took.
- Fix a point v. translate $S$, so that $v = 0$. Also transform so that $\dim S = n$. Define $S^*$ as maximal superset of $S$ that also satisfies such property. It would be enough to show that $S^*$ is a lattice.
- Fix some set of $k$ points, and denote them as $T$. Denote set of points $p$ in $\mathbb{R}^n$ such that $d^2(p, t) \in \mathbb{N}$ for every $t \in T$ as $P_T$. By definition, $S^* \subset P_T$.
- (choosing nice $k$ points). Since $S^*$ is countable, set of every size $k$ subset of $S^*$ is also countable. Therefore we select $k$ points $T$ that has maximal rank with respect to $S$. Formally this would mean selecting $k$ points that $\dim (Span(T) \cap Span(S^*))$ is maximum. Here dim is used in lattice sense.
- It would be enough to prove that for $k = n$, such $T$ satisfies $P_T = S^*$.
There are close problems, but non exact search results…
https://en.wikipedia.org/wiki/Erd%C5%91s%E2%80%93Ulam_problem
Seems to be lots of researches with "integral distances" but my google search skills limits me with finding non about distance squared being integral.
Best Answer
Following your approach, we translate so that $0\in S$, and define $S^*$ as a maximal superset of $S$ satisfying the requested property.
To prove that $S^*$ is a lattice, it suffices to show that it is a discrete subgroup of $\mathbb{R}^n$, i.e. that for any points $a,b \in S^*$, their sum and difference $a\pm b$ are also in $S^*$ (discreteness is obvious since each point is at least $1$ unit of distance away from the others).
First, we note that for any $x,y \in S^*$, we have
$$|x|^2 = d(x,0)^2 \in \mathbb{N}\cup\{0\}$$
and
$$2\,x\cdot y = d(x,0)^2+d(y,0)^2-d(x,y)^2 \in \mathbb{Z},$$
where $\cdot$ and $|\:\:|$ denote the standard Euclidean inner product and norm respectively.
Next, given an arbitrary $c \in S^*$ we compute
$$d(a\pm b,c)^2=|a\pm b-c|^2 = |a|^2+|b|^2+|c|^2\pm 2\,a\cdot b-2\,a\cdot c \mp 2\,b \cdot c \in \mathbb{N}\cup\{0\}.$$
We conclude that $a\pm b \in S^*$ by maximality.
Finally, we prove that the rank of the lattice $S^*$ must be $n$: if it were lower, say $m<n$, then any unit-length vector perpendicular to the hyperplane $S^*\otimes \mathbb{R} \simeq \mathbb{R}^m$ would have integral squared distance to any vector in $S^*$ by the Pythagorean theorem, contradicting maximality.