Wiki
Wiki

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

Updated


Source: arXiv v2, pp. 5--6, Lemma 3.3 and its proof.

Statement

Let s=m(A)s=m(\mathcal A) and 1≤j≤J1\le j\le J.

(a) If δi=0\delta_i=0 for every i<ji<j, then

Mj(1)≤s∑d≥d1P+(d)=pj1d.(1)M_j^{(1)}\le s \sum_{\substack{d\ge d_1\\P^+(d)=p_j}}\frac1d. \tag{1}

(b) Uniformly in all choices 0≤δi≤1/20\le\delta_i\le1/2,

Mj(2)≪s2(log⁡pj)6pj2,(2)M_j^{(2)}\ll\frac{s^2(\log p_j)^6}{p_j^2}, \tag{2}

with an absolute implied constant.

Full proof relative to the external distortion bound

Write p=pjp=p_j and let t∈{1,2}t\in\{1,2\}. Raise the nonnegative inequality of Lemma 3.2 to the ttth power, expand the ordered product, and take expectation. This gives

Ej−1αjt≤∑1≤r1,…,rt≤νjg1,…,gt∣Qj−1∑1≤i1,…,it≤ndiℓ=gℓprℓ (1≤ℓ≤t)Pj−1 ⁣(⋂ℓ=1t(aiℓ+gℓZ))pr1+⋯+rt.(3)\mathbb E_{j-1}\alpha_j^t\le \sum_{\substack{1\le r_1,\ldots,r_t\le\nu_j\\ g_1,\ldots,g_t\mid Q_{j-1}}} \sum_{\substack{1\le i_1,\ldots,i_t\le n\\ d_{i_\ell}=g_\ell p^{r_\ell}\ (1\le\ell\le t)}} \frac{\mathbb P_{j-1}\!\left( \bigcap_{\ell=1}^t(a_{i_\ell}+g_\ell\mathbb Z)\right)} {p^{r_1+\cdots+r_t}}. \tag{3}

Every occurring gℓprℓ=diℓg_\ell p^{r_\ell}=d_{i_\ell} is at least d1d_1. Once gℓ,rℓg_\ell,r_\ell are fixed, multiplicity gives at most sts^t ordered choices of the indices. A compatible intersection is one progression of modulus [g1,…,gt][g_1,\ldots,g_t]; an incompatible one is empty. Applying the exact external distortion bound to every nonempty intersection yields

Ej−1αjt≤st∑1≤r1,…,rt≤νjg1,…,gt∣Qj−1gℓprℓ≥d1 (1≤ℓ≤t)∏ph∣[g1,…,gt](1−δh)−1[g1,…,gt]pr1+⋯+rt.(4)\mathbb E_{j-1}\alpha_j^t\le s^t \sum_{\substack{1\le r_1,\ldots,r_t\le\nu_j\\ g_1,\ldots,g_t\mid Q_{j-1}\\ g_\ell p^{r_\ell}\ge d_1\ (1\le\ell\le t)}} \frac{\displaystyle\prod_{p_h\mid[g_1,\ldots,g_t]} (1-\delta_h)^{-1}} {[g_1,\ldots,g_t]p^{r_1+\cdots+r_t}}. \tag{4}

For t=1t=1 and δh=0\delta_h=0 for h<jh<j, the distortion product is 11. Each pair (g,r)(g,r) determines d=gprd=gp^r with P+(d)=pP^+(d)=p, so (4) is bounded by the enlarged sum in (1).

For t=2t=2, each distortion factor is at most 22, whence

Mj(2)≤s2(p−1)2∑g1,g2∣Qj−12ω([g1,g2])[g1,g2].(5)M_j^{(2)}\le\frac{s^2}{(p-1)^2} \sum_{g_1,g_2\mid Q_{j-1}} \frac{2^{\omega([g_1,g_2])}}{[g_1,g_2]}. \tag{5}

Indeed, the two geometric sums in r1,r2r_1,r_2 are each at most ∑r≥1p−r=1/(p−1)\sum_{r\ge1}p^{-r}=1/(p-1).

The remaining sum is multiplicative. If a prime power qeq^e with e≥1e\ge1 is the exact qq-part of [g1,g2][g_1,g_2], there are

(e+1)2−e2=2e+1(e+1)^2-e^2=2e+1

ordered pairs of exponents having maximum ee, and the factor 2ω([g1,g2])2^{\omega([g_1,g_2])} contributes 22. Therefore

∑g1,g2∣Qj−12ω([g1,g2])[g1,g2]=∏h<j(1+2∑e=1νh2e+1phe).(6)\sum_{g_1,g_2\mid Q_{j-1}} \frac{2^{\omega([g_1,g_2])}}{[g_1,g_2]} =\prod_{h<j}\left( 1+2\sum_{e=1}^{\nu_h}\frac{2e+1}{p_h^e}\right). \tag{6}

Each local factor is 1+6/ph+O(ph−2)1+6/p_h+O(p_h^{-2}). Enlarging the absolute constant handles the finitely many small primes, and 1+z≤ez1+z\le e^z gives

(6)≪exp⁡(6∑ph<p1ph)≪(log⁡p)6(6)\ll\exp\left(6\sum_{p_h<p}\frac1{p_h}\right) \ll(\log p)^6

by Mertens' estimate for the prime reciprocal sum. Finally (p−1)−2≪p−2(p-1)^{-2}\ll p^{-2}, and (2) follows. The Chinese remainder theorem and Mertens' estimate are standard external inputs; the distortion estimate is the separately identified imported lemma above.