9
$\begingroup$

Given an integer $n > 0$, let $f(n)$ denote the least dimension of the Euclidean space into which there exists a C isometric embedding of every Riemannian flat $n$-dimensional torus $\Bbb T^n =\Bbb R^n / L$, where $L$ is an $n$-dimensional lattice in $\Bbb R^n$.

What is known about $f(n)$ in terms of exact values, upper and lower bounds, or asymptotics?

$\endgroup$
6
  • 1
    $\begingroup$ Smooth isometric embedding, or just isometry of metric spaces? $\endgroup$
    – Ben McKay
    Commented Mar 22, 2022 at 18:54
  • $\begingroup$ I suspect there is not much research on either problem, but I would be delighted to be wrong. $\endgroup$ Commented Mar 22, 2022 at 21:31
  • $\begingroup$ A parameter count suggests that $f(n)\le n(n{+}1)$, and it's not hard to verify this for $n=1$ and $n=2$. This bound is probably not sharp, though. $\endgroup$ Commented Mar 22, 2022 at 22:04
  • 2
    $\begingroup$ Actually, since the n-torus is diffeomorphic to a submanifold of (n+1)-space, Nash-Kuiper shows that all Riemannian n-tori (flat or not) isometrically embed continuously differentiably in (n+1)-space. So it's really the case of smooth embeddings that is of interest. $\endgroup$ Commented Mar 23, 2022 at 17:37
  • 2
    $\begingroup$ I just noticed that the paper by C. Tompkins, "Isometric embedding of flat manifolds in Euclidean space" (Duke Math. J., v. 5, 1939) contains a proof that for any compact Riemannian flat n-manifold M, its isometric embedding dimension e(M) satisfies the inequality e(M) ≥ 2n. Hence f(n) ≥ 2n. $\endgroup$ Commented Mar 24, 2022 at 20:18

1 Answer 1

9
$\begingroup$

By the Nash Embedding Theorem, $f(n)\le \tfrac12n(3n+11)$.

Here's an idea of a way to show that $f(n)\le n(n{+}1)$ by constructing an explicit isometric embedding: Let $\mathbb{T}^n = \mathbb{R}^n/L$ where $L\subset\mathbb{R}^n$ is a lattice and $\mathbb{R}^n$ is endowed with the usual inner product and the standard Riemannian metric $g=\mathrm{d}x\cdot\mathrm{d} x$.

Let $L^*\subset\mathbb{R}^n$ be the dual lattice, i.e., the set of vectors $\xi\in \mathbb{R}^n$ such that $\xi\cdot x$ is an integer for all $x\in L$. Now choose $N = \tfrac12n(n{+}1)$ elements $\xi_1,\ldots,\xi_N$ that generate $L^*$ and $N$ nonnegative constants $r_1,\ldots, r_n$ and define a map $\Phi: \mathbb{T}^n\to \mathbb{C}^N\simeq \mathbb{R}^{n(n+1)}$ $$ \Phi(x) = \frac1{2\pi}\bigl(\sqrt{r_1}\,\mathrm{e}^{2\pi i\,\xi_1\cdot x},\ldots, \sqrt{r_N}\,\mathrm{e}^{2\pi i\,\xi_N\cdot x}\bigr). $$ Because of the hypothesis about the set $\xi_1,\ldots,\xi_N$, it follows that $\Phi$ is a well-defined injection, provided that the $r_i$ are all positive, and moreover it follows that, giving $\mathbb{R}^{n(n{+}1)} = \mathbb{C}^N$ the standard product metric with each copy of $\mathbb{C}$ given its standard metric, the metric pulled back by $\Phi$ is of the form $\mathrm{d}x\cdot G\,\mathrm{d}x$ where $G$ is the symmetric positive definite constant $n$-by-$n$ matrix $$ G = {r_1}\,\xi_1\,\xi_1^T + \cdots + {r_N}\,\xi_N\,\xi_N^T. $$ In particular, the induced metric on the image of $\Phi$ is flat.

Now, if it's possible to choose the $\xi_1,\ldots,\xi_N$ in $L^*$ so that a multiple of the identity matrix $I_n$ is in the convex hull of the $N$ rank 1 positive semi-definite symmetric matrices $$ \xi_1\,\xi_1^T,\ \ldots,\ \xi_N\,\xi_N^T, $$ then one can choose the $r_i\ge0$ so that $\Phi$ will be an isometric embedding.

This is trivial when $n=1$ (of course, $f(1) = 2$), and not difficult when $n=2$: After a rotation in $\mathbb{R}^2$, we can assume that $L^*$ is generated by $\xi_1 = (a,0)$ and $\xi_2 = (b,c)$ where $a>0$ and $c>0$ and $b^2+c^2>a^2$ and $0\le b\le \tfrac12 a$. Then, taking $\xi_3 = (b-a,c) = \xi_2-\xi_1$ works. Thus, $4\le f(2)\le 6$.

I'm pretty sure that I have a method for choosing $\xi_1,\ldots ,\xi_6$ when $n=3$ that works, but the details are messy. Probably, there's a cleaner way to do it.

$\endgroup$
2
  • $\begingroup$ One can do a bit better: $f(n)\leqslant\tfrac{n\cdot(n+3)}2$. In 2B of "Isometric Immersions with Controlled Curvatures", Gomov constructs a free isometric embedding of a flat $n$-torus in the $\tfrac{n\cdot(n+3)}2$-dimensional Euclidean space. By rescaling this example and passing to a cover, one can get a flat torus arbitrarily close to a given one. Now, by applying Nash's deformation, one gets an embedding of any flat torus. $\endgroup$ Commented Oct 8, 2023 at 17:23
  • $\begingroup$ I hadn't realized that rescaling and passing to a cover could get from one flat n-torus to (arbitrarily close to) any flat n-torus. I need to think about that. ... ... Okay, I think I see how. $\endgroup$ Commented Oct 8, 2023 at 17:29

You must log in to answer this question.

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