Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Source. Karamanlis, published pp. 5–6, Lemma 7 (canonical PDF). The corresponding arXiv v1–v3 result is Lemma 3.5.
The sufficient parameter range below repairs the source's unrestricted threshold. All later uses require only that the integer parameter can be chosen sufficiently large.
Statement. Let have at least two points, let , and let be an integer such that
For every integer
there is a -embedding into , with and . The hypothesis (1) itself forces to be finite, since a bounded interval contains only finitely many points separated by at least .
Proof. Translate into . Put
Thus , , and . Since , one has . Two distinct points in the same floor interval would be less than apart, contradicting (1). Consequently is injective. Because , all these indices satisfy and select distinct polygon vertices.
The difference between the two rounding errors and has absolute value less than . The original and rounded distances are both at most , and hence
Define
The map is injective. For distinct , write and . Then , so the chord length is .
For completeness, gives and , so . For , the bound gives the same strict inequality. It follows that
Also, since ,
Adding (3) and (4) proves the required strict squared-distance error bound. For , the error is zero.
Counterexample to the printed range. Published Lemma 7 assumes only . This does not imply injectivity, even with understood.
Proof. Take , , , and . The minimum and maximum distances are and , as required. The printed threshold is satisfied because . But has ten points, while has eight vertices. No injection exists, irrespective of the error tolerance.
Version and endpoint precision. All three arXiv versions retain that same threshold. Versions 1 and 2 call positive and write ; version 3 and the publication correctly use nonnegative and . The earlier rounding display is , whereas version 3 and the publication use . The proof above uses the latter estimate. At an equality endpoint in the printed threshold, the numerical upper bound in source (4) can equal when ; the strict sine estimate above still gives strictly smaller actual error. The added restrictions in (2) supply the missing injection and polygon-order conditions. This is a compilation repair, not a reported author correction.
Use. Proposition 8.