Wiki
Wiki

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

Updated


Source statement. Mathialagan, published 2021 PDF, p. 1, Theorem 1, cites Erdős for the ordinary planar distinct-distance bound Dordinary(N)=O(N/log⁡N)D_{\rm ordinary}(N)=O(N/\sqrt{\log N}). Its p. 3 applies D(m,n)≤Dordinary(m+n)D(m,n)\leq D_{\rm ordinary}(m+n); p. 4, Table 1 and Question 5, record the resulting balanced comparison.

The primary construction is Erdős, On sets of distances of nn points, American Mathematical Monthly 53 (1946), 248--250, Theorem 1 and the third paragraph of its proof on printed p. 248 (physical p. 1 of the scan read). That scan states weak inequalities in Theorem 1. The asymptotic upper bound, not its unrelated lower inequality, is used here.

Imported counting premise. Let

R(X)=∣{k∈Z:1≤k≤X,k=u2+v2 for some u,v∈Z}∣.R(X)=|\{k\in\mathbb Z:1\leq k\leq X,\quad k=u^2+v^2\text{ for some }u,v\in\mathbb Z\}|.

There are constants C,X0>0C,X_0>0 such that for all real X≥X0X\geq X_0,

R(X)≤C X/log⁡X.(1)R(X)\leq C\,X/\sqrt{\log X}. \tag{1}

This is the sum-of-two-squares counting fact used in Erdős's proof on p. 248, which cites Landau, Verteilung der Primzahlen, vol. 2, in its footnote. The premise is imported at that precise source interface; Landau's original text and the number-theoretic proof are not compiled here. No asymptotic constant or error term is assumed.

Own-words construction. For an integer N≥2N\geq2, put k=⌈N⌉k=\lceil\sqrt N\rceil and take the grid G={0,1,…,k−1}2G=\{0,1,\ldots,k-1\}^2. It contains k2≥Nk^2\geq N points; select any NN of them. Every positive squared distance is an integer u2+v2u^2+v^2 with ∣u∣,∣v∣≤k−1|u|,|v|\leq k-1, hence is at most 2(k−1)2<2N2(k-1)^2<2N. Taking the positive square root is injective, so the number of distances is at most R(2N)R(2N). For all sufficiently large NN, (1) gives R(2N)≤2CN/log⁡(2N)=O(N/log⁡N)R(2N)\leq2CN/\sqrt{\log(2N)}=O(N/\sqrt{\log N}). Discarding grid points introduces no distances, so the estimate holds for the selected set of exactly NN points. Finitely many smaller N≥2N\geq2 are covered by increasing the absolute constant.

For the balanced bipartite case take N=2nN=2n and partition the selected 2n2n distinct grid points into two disjoint sets P,QP,Q of exactly nn points each. Every cross-distance is one of the ordinary distances of their union, so

D(P,Q)≤R(4n)=O(n/log⁡n).D(P,Q)\leq R(4n)=O(n/\sqrt{\log n}).

This construction works for every sufficiently large integer nn, without a square-size restriction. It also meets a convention requiring the two color classes to be disjoint.

Current verification. Verified at the stated scope, retained in the final review and finalization-delta review. An independent source-based reviewer, distinct from the compiler, checked the complete lattice selection, squared-distance count and balanced partition against the published Mathialagan statements identified above and Erdős's 1946 scan, printed p. 248. The compiler supplied the exact-size selection and bipartite application; the reviewer checked their deductions and the stated counting interface. No unresolved local proof gap remains within this scope.

The sole imported nontrivial premise is (1) at Erdős's cited Landau interface. Its statement, locator and use were independently checked, but no original Landau proof or unrelated Erdős lower-bound proof is compiled or reviewed here. This record is separate from the Theorem 3 lower-bound record and gives no solution, status or literature-freshness conclusion.

A substantive change to the source version, statement, construction, counting premise or relied-on dependency returns the affected scope and its applications to Needs review until independently checked again.

Application. The construction supplies big-O. It does not show that the ratio to n/log⁡nn/\sqrt{\log n} tends to zero, and therefore does not resolve Problem 661.

Bears on. Problem 661.