Wiki
Wiki

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

Updated


Source. The seed family and equation (21), printed pp. 89–90 (PDF pp. 5–6). The source states that the weighted estimate is not hard to prove and omits its details. The following is a full construction of a subfamily satisfying its conditions.

Statement. There is a constant C0>0C_0>0 such that, for every sufficiently large xx, a family Ax\mathcal A_x of distinct odd square-free integers exists with:

  1. x1/2<a<x3/4x^{1/2}<a<x^{3/4} for all a∈Axa\in\mathcal A_x;
  2. every aa is divisible by 3,5,7,113,5,7,11;
  3. any consecutive prime factors q<q′q<q' with q′>11q'>11 satisfy q′<q5/4q'<q^{5/4};
  4. ∑a∈Ax1/a>(log⁡x)−C0\sum_{a\in\mathcal A_x}1/a>(\log x)^{-C_0}.

Every such integer belongs to the sequence in equation (19).

A fixed initial chain

Put λ=9/8\lambda=9/8 and ρ=11/10\rho=11/10, so 1<ρ<λ1<\rho<\lambda and ρλ=99/80<5/4\rho\lambda=99/80<5/4. Choose a fixed prime Q≥17Q\ge17 sufficiently large for

∑Y<p≤Yρ1p≥b:=12log⁡ρ>0(Y≥Qλ),(1)\sum_{Y<p\le Y^\rho}\frac1p\ge b:=\frac12\log\rho>0 \qquad(Y\ge Q^\lambda), \tag{1}

which follows from the exact prime estimates on the inputs page. In particular b<1b<1.

Let a0a_0 be the product of all odd primes at most QQ. Its prime factors start with 3,5,7,113,5,7,11. The next two steps are 13<115/413<11^{5/4} and 17<135/417<13^{5/4}. Thereafter Bertrand's postulate gives the next prime q′<2q<q5/4q'<2q<q^{5/4} for q≥17q\ge17. Thus this fixed prefix satisfies the required consecutive-prime condition.

For j≥1j\ge1 let

uj=λjlog⁡Q,Pj={p prime:euj<p≤eρuj}.u_j=\lambda^j\log Q,\qquad \mathcal P_j=\{p\text{ prime}:e^{u_j}<p\le e^{\rho u_j}\}.

These intervals are pairwise disjoint and above QQ, because ρ<λ\rho<\lambda. If pj∈Pjp_j\in\mathcal P_j, then p1<Q5/4p_1<Q^{5/4}, and

log⁡pj+1≤ρλuj<54uj<54log⁡pj.\log p_{j+1}\le\rho\lambda u_j<\frac54u_j <\frac54\log p_j.

So appending one prime from each successive interval preserves the consecutive-prime condition.

Choosing the number of intervals

Write X=log⁡xX=\log x and define

Am=log⁡a0+∑j=1muj.A_m=\log a_0+\sum_{j=1}^m u_j.

Choose the least integer m≥1m\ge1 such that Am>X/2A_m>X/2; it exists, and for large xx the fixed value A0A_0 is below X/2X/2. The sum is geometric, so

m=log⁡Xlog⁡λ+O(1).m=\frac{\log X}{\log\lambda}+O(1).

Every product a=a0p1⋯pma=a_0p_1\cdots p_m has Am<log⁡a≤ρAmA_m<\log a\le\rho A_m. Also

Am=λAm−1+λlog⁡Q−(λ−1)log⁡a0≤λX2+O(1).A_m=\lambda A_{m-1}+\lambda\log Q-(\lambda-1)\log a_0 \le\frac{\lambda X}{2}+O(1).

Consequently log⁡a≤(ρλ/2)X+O(1)<3X/4\log a\le(\rho\lambda/2)X+O(1)<3X/4 eventually. The products lie in (x1/2,x3/4)(x^{1/2},x^{3/4}), are odd and square-free, and their prime factors identify the tuple uniquely.

By (1), their reciprocal sum is

1a0∏j=1m(∑p∈Pj1p)≥bma0=(log⁡x)log⁡b/log⁡λ+o(1).\frac1{a_0}\prod_{j=1}^m\left(\sum_{p\in\mathcal P_j}\frac1p\right) \ge\frac{b^m}{a_0} = (\log x)^{\log b/\log\lambda+o(1)}.

Any fixed C0>−log⁡b/log⁡λC_0>-\log b/\log\lambda proves the required bound for all sufficiently large xx.

Membership in the original sequence

The first four primes satisfy equation (19): 7<3⋅57<3\cdot5 and 11<3⋅5⋅711<3\cdot5\cdot7. Their product 11551155 exceeds 11211^2. Inductively, suppose the product DD through the previous largest prime qq exceeds q2q^2. The next prime satisfies q′<q5/4<Dq'<q^{5/4}<D, so its equation (19) inequality holds, and the new product Dq′Dq' exceeds (q′)2(q')^2. This proves every later inequality and membership in N\mathcal N.

Source precision. The source's seed description does not explicitly exclude a factor 2, although its assertion that all seeds satisfy p1=3,p2=5p_1=3,p_2=5 requires this. The odd subfamily constructed here has the stated reciprocal mass and supplies the needed correction. The explicit intervals fill the omitted same-paper count; they are not claimed to reproduce a construction written in the source.