Wiki
Wiki

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

Updated


Source. Published p. 6, Lemma 10 (canonical PDF); arXiv Lemma 3.8. Karamanlis gives a proof pointer to Schoenberg and Frankl–Rödl. The deduction below is expanded relative to the complete finite negative-type criterion already compiled with Frankl–Rödl's 1990 source. That finite Gram proof is not duplicated here.

Statement. Every nonempty simplex Y={y1,…,yn}Y=\{y_1,\ldots,y_n\} is a regular expansion of another simplex X={x1,…,xn}X=\{x_1,\ldots,x_n\}. For n≥2n\ge2, the amount subtracted from every off-diagonal squared distance may be chosen positive and sufficiently small.

Proof. Suppose n≥2n\ge2 and put eij=∥yi−yj∥2e_{ij}=\|y_i-y_j\|^2. For every nonzero zero-sum vector cc, affine independence gives

Qe(c):=∑i<jcicjeij=−∥∑iciyi∥2<0.Q_e(c):=\sum_{i<j}c_ic_je_{ij} =-\left\|\sum_i c_i y_i\right\|^2<0.

The zero-sum unit sphere in Rn\mathbb R^n is compact. Consequently there is a γ>0\gamma>0 with Qe(c)≤−γ∥c∥2Q_e(c)\le-\gamma\|c\|^2 on the whole zero-sum subspace. Choose 0<α2<2γ0<\alpha^2<2\gamma, set dii=0d_{ii}=0, and set dij=eij−α2d_{ij}=e_{ij}-\alpha^2 for i≠ji\ne j. Since ∑i<jcicj=−∥c∥2/2\sum_{i<j}c_ic_j=-\|c\|^2/2 on that subspace,

Qd(c)=Qe(c)+α22∥c∥2≤−(γ−α22)∥c∥2<0Q_d(c)=Q_e(c)+\frac{\alpha^2}{2}\|c\|^2 \le-\left(\gamma-\frac{\alpha^2}{2}\right)\|c\|^2<0

for every nonzero zero-sum cc. The linked finite Gram criterion therefore realizes (dij)(d_{ij}) as the squared distances of an affinely independent set XX in Rn−1\mathbb R^{n-1}. In particular these off-diagonal entries are positive; this also follows by applying the strict inequality to ci=1,cj=−1c_i=1,c_j=-1, with all other entries zero. We obtain ∥yi−yj∥2=∥xi−xj∥2+α2\|y_i-y_j\|^2=\|x_i-x_j\|^2+\alpha^2 for i≠ji\ne j, as required. For n=1n=1, choose any singleton XX and any α>0\alpha>0; the off-diagonal condition has no instances. □\square

Dependency scope. This proves the source's essential reduction, relative only to the precise elementary criterion linked above. It does not assume that an arbitrary difference of two Euclidean distance matrices is Euclidean. Neither a general spherical Ramsey theorem nor the Matoušek–Rödl spread-vector theorem is an input to this deduction.

Use. Theorem 2. The same contraction mechanism appears in Frankl–Rödl's original simplex proof, whose subsequent approximation and density argument are different.