Wiki
Wiki

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

Updated


Statement

After the proof of Theorem 3 the paper writes: "We can improve the theorem slightly. Let rr be the solution of the equation e−r=re^{-r}=r and let c=r/(1−r)=1.7398…c=\sqrt r/(1-r)=1.7398\ldots. Then

f(n)≤(c+o(1)) nlog⁡n.(11)f(n)\le(c+o(1))\,n\sqrt{\log n}. \tag{11}

We now sketch a proof of (11)." Here f(n)f(n) is the least integer such that (n,f(n)](n,f(n)] contains distinct a1,…,ana_1,\ldots,a_n with i∣aii\mid a_i (p. 147). The constant was recomputed here: r=0.567143…r=0.567143\ldots and c=1.73981…c=1.73981\ldots.

Source. P. Erdős and C. Pomerance, Matching the natural numbers up to nn with distinct multiples in another interval, Indag. Math. (Proc.) 83 (1980), no. 2, 147--161, DOI 10.1016/1385-7258(80)90018-9; display (11) and its sketch on printed pp. 154--155 (PDF pp. 8--9 of the 15-page scan read for this page), read on the page images.

Read depth. Claims checked: the statement of (11) was read clause by clause on the page image. The paper itself calls its argument a sketch; the sketch (pp. 154--155) was read for its structure and not checked. Nothing here is independently reviewed.

Proof pointer

Pages 154--155. With γ∈(0,1)\gamma\in(0,1) and an integer kk, the indices [1,n][1,n] are split into the ranges Ij=(γjn/log⁡n,γj−1n/log⁡n]I_j=(\gamma^jn/\sqrt{\log n},\gamma^{j-1}n/\sqrt{\log n}] (j=−k+1,…,kj=-k+1,\ldots,k) together with I−kI_{-k} and Ik+1I_{k+1}; the indices in I−kI_{-k} are matched directly by ai=i([γklog⁡n]+1)a_i=i([\gamma^k\sqrt{\log n}]+1) into J1=(n,γkn(log⁡n+1)]J_1=(n,\gamma^kn(\sqrt{\log n}+1)], and the rest into J2=(γkn(log⁡n+1),(γk+b)nlog⁡n]J_2=(\gamma^kn(\sqrt{\log n}+1),(\gamma^k+b)n\sqrt{\log n}] through the graph in which (i,j)(i,j) is an edge when j/ij/i is prime. A failure of the König–Hall condition is analyzed range by range, giving display (12) and then the sufficient condition (13), β>2γk+1/(−γ2k+1+(2k+2)γ−2k)\beta>2\gamma^{k+1}/(-\gamma^{2k+1}+(2k+2)\gamma-2k) for f(n)≤βnlog⁡nf(n)\le\beta n\sqrt{\log n}; the choice γ=1−r/2k\gamma=1-r/2k with e−r=re^{-r}=r makes the right side r/(1−r)+o(1/k)\sqrt r/(1-r)+o(1/k), and k→∞k\to\infty gives (11).

Dependencies

The König–Hall matching theorem and the prime number theorem, as in Theorem 3.

Bears on

  • Problem 710: the upper bound the site prints, (1.7398⋯+o(1))n(log⁡n)1/2(1.7398\cdots+o(1))n(\log n)^{1/2}, is this display; the theorem the paper proves in full is Theorem 3's (2+o(1))nlog⁡n(2+o(1))n\sqrt{\log n}.