Wiki
Wiki

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

Updated


Fix 0<ε<10<\varepsilon<1 and δ>0\delta>0. For every sufficiently large prime power qq and every

I⊆[qε,2qε]∩Z,∣I∣≥δqε,(i,q)=1 (i∈I),I\subseteq[q^\varepsilon,2q^\varepsilon]\cap\mathbb Z,\qquad |I|\ge\delta q^\varepsilon,\qquad (i,q)=1\ (i\in I),

each residue modulo qq is s(B)s(B) for a subset B⊆IB\subseteq I with ∣B∣≤qε/2|B|\le q^{\varepsilon/2}. The threshold depends only on ε,δ\varepsilon,\delta, so the statement is uniform in II and the residue.

Source: published PDF, Theorem 2, pp. 7–8. The printed statement allows every ε>0\varepsilon>0. This page certifies the complete range 0<ε<10<\varepsilon<1 used by the paper's absorption theorem; it does not certify the unused larger range. The proof is complete relative to the exact external CFP and divisor inputs, with the restricted modular approximation and proper-dilate reduction explicitly repaired.

Bears on. Problem 297.

Proof

Use the progression, sets, and parameters from gap_symmetrization. If its dimension kk is at least 2, its volume estimate first gives A<qA<q. Apply the corrected Lemma 4 to its integer steps; they need not individually be units. Then claim_1 gives

A≥(ρ16k)kq1−kκ(q),ρ=δ/4,κ(q)⟶0.A\ge\left(\frac{\rho}{16k}\right)^k q^{1-k\kappa(q)}, \qquad \rho=\delta/4,\quad \kappa(q)\longrightarrow0.

On the other hand, A≤Ckqs1−k≤Ck′q1−(k−1)ε/2A\le C_k q s^{1-k}\le C_k' q^{1-(k-1)\varepsilon/2}. For every fixed 2≤k≤d2\le k\le d these inequalities contradict each other for large qq, since kκ(q)<(k−1)ε/4k\kappa(q)<(k-1)\varepsilon/4 eventually. Thus the actual dimension is 1.

Now P∗=[−a1,a1]d1P_*=[-a_1,a_1]d_1 contains a member of JJ coprime to qq. It follows that d1d_1 is coprime to qq. Choose TT to be its inverse modulo qq and take d1′=1d_1'=1. The bound required by Claim 1 is valid because 2(q/a1)(a1/q)=22(q/a_1)(a_1/q)=2. This proves the same volume lower bound for k=1k=1 without any assumption a1≤qa_1\le q.

The proper progression contained in Σ(B0)\Sigma(B_0) has at least (λ/4)a1(\lambda/4)a_1 points in an arithmetic progression with step d1d_1. Since λ≍qε/2\lambda\asymp q^{\varepsilon/2} and a1≥(ρ/16)q1−κ(q)a_1\ge(\rho/16)q^{1-\kappa(q)}, this number exceeds qq for large qq. Any qq consecutive terms of a progression with step coprime to qq give every residue. Every such term is a subset sum of B0B_0, and ∣B0∣≤s≤qε/2|B_0|\le s\le q^{\varepsilon/2}. Each element of B0B_0 is the inverse of a distinct element of II. Lifting a subset of B0B_0 to those original denominators gives the required subset B⊆IB\subseteq I.