Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Published pp. 6–7, Proposition 11 (canonical PDF); arXiv Proposition 3.9.

Statement. If YY is a regular expansion of a finite Euclidean configuration XX, then YY embeds into an mm-regular polygonal torus for some integer m≥2m\ge2. The radii of its factors need not be equal.

Proof. A singleton is immediate, so write X={x1,…,xn}X=\{x_1,\ldots,x_n\} with n≥2n\ge2, and choose α>0\alpha>0 such that

∥yi−yj∥2=∥xi−xj∥2+α2(i≠j).\|y_i-y_j\|^2=\|x_i-x_j\|^2+\alpha^2\qquad(i\ne j).

Put δ=α2/n2\delta=\alpha^2/n^2. By Proposition 8, there are m≥2m\ge2, r>0r>0 and an injective δ\delta-embedding f:X→Tm,rkf:X\to T_{m,r}^{k}. Define the symmetric errors

eij=∥xi−xj∥2−∥f(xi)−f(xj)∥2,∣eij∣<δ.e_{ij}=\|x_i-x_j\|^2-\|f(x_i)-f(x_j)\|^2, \qquad |e_{ij}|<\delta.

Set aii=0a_{ii}=0 and aij=α2+eija_{ij}=\sqrt{\alpha^2+e_{ij}} for i≠ji\ne j. These roots are positive because δ<α2\delta<\alpha^2. To verify the hypothesis of Lemma 4, let A2=max⁡i<jaij2A^2=\max_{i<j}a_{ij}^2. Then

A2>α2−δ,0≤A2−aij2<2δ,A^2>\alpha^2-\delta, \qquad 0\le A^2-a_{ij}^2<2\delta,

and therefore

∑i<j(A2−aij2)<n(n−1)δ=α2n−1n<α2(1−1n2)=α2−δ<A2.(1)\begin{aligned} \sum_{i<j}(A^2-a_{ij}^2) &<n(n-1)\delta =\alpha^2\frac{n-1}{n}\\ &<\alpha^2\left(1-\frac1{n^2}\right) =\alpha^2-\delta<A^2. \end{aligned} \tag{1}

The middle strict inequality uses n>1n>1. Lemma 4 constructs an almost-regular simplex Z={z1,…,zn}Z=\{z_1,\ldots,z_n\} with distances aija_{ij}. Proposition 5 embeds ZZ into an mm-regular torus T0T_0, with the same mm already chosen above; write this isometry as hh.

For 1≤i≤n1\le i\le n, put yi′=(f(xi),h(zi))y'_i=(f(x_i),h(z_i)) in Tm,rk×T0T_{m,r}^{k}\times T_0. This product is mm-regular. For i≠ji\ne j,

∥yi′−yj′∥2=∥f(xi)−f(xj)∥2+aij2=∥f(xi)−f(xj)∥2+eij+α2=∥xi−xj∥2+α2=∥yi−yj∥2.\begin{aligned} \|y'_i-y'_j\|^2 &=\|f(x_i)-f(x_j)\|^2+a_{ij}^2\\ &=\|f(x_i)-f(x_j)\|^2+e_{ij}+\alpha^2\\ &=\|x_i-x_j\|^2+\alpha^2=\|y_i-y_j\|^2. \end{aligned}

For i=ji=j both sides are zero. The matching of labels is thus an isometric embedding of YY. □\square

Source precision. The source calls the residual array almost regular without giving inequality (1); this is the full bound for its exact choice δ=α2/n2\delta=\alpha^2/n^2. Its final display is introduced for all i,ji,j, but the added α2\alpha^2 applies only when i≠ji\ne j. The diagonal case is separated here. The corrected sufficient parameter choice in Lemma 7 changes neither the tolerance nor the conclusion of this argument. No Euclidean realization of the residual is assumed before Lemma 4 supplies it.

Use. Theorem 2.