Wiki
Wiki

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

Updated


Write ω(n)\omega(n) for the number of distinct prime divisors of nn, with ω(1)=0\omega(1)=0. The external input is Erdős–Pomerance–Sárközy Lemma 1: with modulus m=1m=1, additive function ω\omega and K=1K=1, it gives for X≥2X\ge2

∑n≤X(ω(n)−A(X))2≪X(A(X)+1),A(X)=∑p≤X1p.(1)\sum_{n\le X}(\omega(n)-A(X))^2\ll X(A(X)+1), \qquad A(X)=\sum_{p\le X}\frac1p. \tag{1}

The condition concerning primes dividing the modulus is vacuous. The function is nonnegative and additive, and ω(pa)=1\omega(p^a)=1. Euler's divergence of the prime reciprocal sum implies A(X)→∞A(X)\to\infty; the elementary bound A(X)≤1+log⁡XA(X)\le1+\log X will also suffice below.

For sufficiently large integers jj, put

Xj=2j+2,Aj=A(Xj),Hj=⌈Aj3/4⌉,Dj={1≤n≤Xj:∣ω(n)−Aj∣>Hj}.X_j=2^{j+2},\quad A_j=A(X_j),\quad H_j=\lceil A_j^{3/4}\rceil, \quad D_j=\{1\le n\le X_j:|\omega(n)-A_j|>H_j\}.

By (1) and Chebyshev's inequality,

∣Dj∣≪XjAj+1Hj2=o(2j).(2)|D_j|\ll X_j\frac{A_j+1}{H_j^2}=o(2^j). \tag{2}

For an integer u∈[2j,2j+1)u\in[2^j,2^{j+1}), let bj(u)b_j(u) count those n∈Djn\in D_j whose image n+ω(n)n+\omega(n) belongs to (u,u+Hj+1)(u,u+H_j+1). Each integer image belongs to at most Hj+1H_j+1 of these windows. Hence

∑2j≤u<2j+1bj(u)≤(Hj+1)∣Dj∣.\sum_{2^j\le u<2^{j+1}} b_j(u)\le(H_j+1)|D_j|.

Set Ej={u∈[2j,2j+1)∩Z:bj(u)>Hj+1}E_j=\{u\in[2^j,2^{j+1})\cap\mathbb Z:b_j(u)>H_j+1\}. Then ∣Ej∣≤∣Dj∣=o(2j)|E_j|\le|D_j|=o(2^j). Both the union E=⋃jEjE=\bigcup_jE_j and

D=⋃j(Dj∩[2j,2j+1))D=\bigcup_j(D_j\cap[2^j,2^{j+1}))

have natural density zero. Indeed, for any ε>0\varepsilon>0 their dyadic counts are at most ε2j\varepsilon2^j for all sufficiently large jj. Summing complete dyadic intervals and bounding the last partial one by its whole interval gives at most 2εx+O(1)2\varepsilon x+O(1) elements below xx.

If u∉Eju\notin E_j, consider all positive integers nn with n+ω(n)∈(u,u+Hj+1)n+\omega(n)\in(u,u+H_j+1). They satisfy n≤Xjn\le X_j for sufficiently large jj, since ω(n)≥0\omega(n)\ge0 and 2j+1+Hj+1<2j+22^{j+1}+H_j+1<2^{j+2}. The latter inequality follows from Hj=O((j+1)3/4+1)H_j=O((j+1)^{3/4}+1). Those nn outside DjD_j obey

u−Aj−Hj<n<u+2Hj+1−Aj.u-A_j-H_j<n<u+2H_j+1-A_j.

There are at most 3Hj+23H_j+2 integers in this interval. The exceptional inputs contribute at most bj(u)≤Hj+1b_j(u)\le H_j+1. The whole count is thus at most 4Hj+34H_j+3.

For real x≥1x\ge1, put u=⌊x⌋u=\lfloor x\rfloor and choose the dyadic index jj with 2j≤u<2j+12^j\le u<2^{j+1}. Define

F(x)={Hj,u∉Ej,2(u+1),u∈Ej.(3)F(x)= \begin{cases} H_j,&u\notin E_j,\\ 2(u+1),&u\in E_j. \end{cases} \tag{3}

For the finitely many initial uu not covered by the construction use the second alternative. Both alternatives tend to infinity with xx. At a nonexceptional location, (x,x+F(x))⊂(u,u+Hj+1)(x,x+F(x))\subset(u,u+H_j+1), so the normalized count is at most 4+3/Hj≤74+3/H_j\le7. At an exceptional location, every counted positive integer satisfies n<x+F(x)<3(u+1)n<x+F(x)<3(u+1). Its normalized count is therefore less than 3/23/2. This proves the asserted bound at every sufficiently large real location, including nonintegral ones.

Finally, let GG be the positive integers outside E∪DE\cup D, omitting the finitely many initial values. This set has natural density one. For u∈G∩[2j,2j+1)u\in G\cap[2^j,2^{j+1}), eventually

ω(u)≥Aj−Hj≥Aj/2,F(u)ω(u)≤2HjAj⟶0.\omega(u)\ge A_j-H_j\ge A_j/2, \qquad \frac{F(u)}{\omega(u)}\le\frac{2H_j}{A_j}\longrightarrow0.

Thus (3) satisfies the density-one width condition but its normalized image count cannot tend to infinity along any sequence of locations. This establishes L34 exactly as stated.