Wiki
Wiki

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

Updated


Statement

Theorem 4 (p. 114). Let a1<a2<⋯a_1<a_2<\cdots be an infinite sequence of integers with ak∣ak+1a_k\mid a_{k+1}. The integers p+akp+a_k, with pp prime, have positive density if and only if

lim sup⁡log⁡akk<∞(4)\limsup\frac{\log a_k}{k}<\infty \qquad (4)

and

∑d∣ai1d<c5.(5)\sum_{d\mid a_i}\frac1d<c_5. \qquad (5)

The print states (5) with the index ii free; under the paper's convention (p. 113) that the cc's are positive absolute constants, the corpus reads (5) as a bound uniform in ii, which is how the proof uses it.

What the proof gives (pp. 120--123). If (4) fails, the number of integers p+ak≤nip+a_k\le n_i is o(ni)o(n_i) along a suitable sequence nin_i; if (5) fails, it is less than n/A+o(n)n/A+o(n) for every AA and all large nn. If (4) and (5) hold, the number of distinct integers p+ak≤np+a_k\le n with k≤c18log⁡nk\le c_{18}\log n exceeds c19nc_{19}n for large nn. So the density in the theorem is positive lower density on the sufficiency side.

The paper closes (p. 123) by noting that the theorem generalizes Romanoff's result that the integers 2k+p2^k+p have positive density.

Proof pointer

Necessity, pp. 120--121: if (4) fails there are o(log⁡ni)o(\log n_i) terms ak≤nia_k\le n_i; if (5) fails, take jj with ∑d∣aj1/d>A\sum_{d\mid a_j}1/d>A and split the integers p+ak≤np+a_k\le n into those with k≤jk\le j, those with p∣ajp\mid a_j, and the rest, which are coprime to aja_j and so number less than n/An/A. Sufficiency, pp. 121--123: count the integers p+ak≤np+a_k\le n, k≤c18log⁡nk\le c_{18}\log n, that are not of the form p+ajp+a_j with j<kj<k, bounding the coincidences p2−p1=ak−ajp_2-p_1=a_k-a_j by Schnirelmann's bound (23) and the Lemma (p. 121): under the hypotheses of the theorem, with (4) and (5), ∑l<k∑d∣ak−al, (d,ak)=11/d<c22k\sum_{l<k}\sum_{d\mid a_k-a_l,\,(d,a_k)=1}1/d<c_{22}k for an absolute constant c22c_{22}. The lemma's proof (pp. 122--123) splits the dd by how many ll have d∣ak−ald\mid a_k-a_l and rules out the second class by a divisor count for a single integer al2/al1−1a_{l_2}/a_{l_1}-1.

Read depth

Claims checked: the statement on p. 114 and the proof with its Lemma on pp. 120--123 were read on the page images of the print; the estimates were followed in outline, not re-derived. Nothing here is independently reviewed.

Dependencies

None in the corpus. External inputs named by the paper: Schnirelmann's upper bound for the number of prime pairs with a given difference, cited through Landau's tract (its footnote 9), and Chebyshev's lower bound for π(n)\pi(n).

Source. P. Erdős, On integers of the form 2k+p2^k+p and some related problems, Summa Brasil. Math. 2 (1950), fasc. 8, 113--123; the edition read is named on the source card.

Bears on

  • Problem 244: for an integer C≥2C\ge2 the sequence ak=Cka_k=C^k satisfies ak∣ak+1a_k\mid a_{k+1}, (4), since log⁡ak/k=log⁡C\log a_k/k=\log C, and (5), since ∑d∣Ci1/d\sum_{d\mid C^i}1/d is at most ∏q∣Cq/(q−1)\prod_{q\mid C}q/(q-1) over the primes qq dividing CC; so the integers p+Ckp+C^k have positive lower density. This application is an observation of this page, not of the paper; it recovers the integer case that the problem page credits to Romanoff and says nothing about non-integer CC.