Wiki
Wiki

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

Updated


Let NN be sufficiently large, L=log⁡NL=\log N, ℓ=log⁡log⁡N\ell=\log\log N, and S≥N.9999S\ge N^{.9999}. Let 1≤M≤N/21\le M\le N/2, and let AA consist of all integers in [M,N][M,N] whose prime-power divisors are at most SS and for which Ω~(n)≤5ℓ\widetilde\Omega(n)\le5\ell, Ω(n)≤10ℓ\Omega(n)\le10\ell. Suppose a positive integer bb satisfies

ω(b)≤5,Ω~(b)≤5ℓ,Ω(b)≤5ℓ+4,2000≤U:=N/b≤L9/40,\omega(b)\le5,\quad \widetilde\Omega(b)\le5\ell,\quad \Omega(b)\le5\ell+4, \quad 2000\le U:=N/b\le L^9/40,

and all prime-power divisors of bb are at most SS. Then there is n∈A∩[N/2,N]n\in A\cap[N/2,N] divisible by bb. Here ω\omega counts distinct prime factors; Ω\Omega counts them with multiplicity, and Ω~\widetilde\Omega is the largest exponent.

Source. This expands the implicit multiple-selection step in Liu–Sawhney, arXiv:2404.07113v1, Proposition 3.2, p. 12. The argument is a compilation-supplied justification of that step. The precise prime reciprocal estimate is external at Theorem 2.1.

Bears on. Problem 297, through the nonempty fibers in the counting minor-arc proof.

Proof

First find an integer u∈[U/2,U]u\in[U/2,U] coprime to bb and satisfying Ω(u)≤ℓ\Omega(u)\le\ell. For every positive integer dd, the number of its multiples in this closed real interval differs from U/(2d)U/(2d) by at most one. Inclusion–exclusion over the at most five prime divisors of bb therefore gives at least

U2∏p∣b(1−1/p)−32≥U21677−32>U10−32≥U12\frac U2\prod_{p\mid b}(1-1/p)-32 \ge\frac U2\frac{16}{77}-32 >\frac U{10}-32\ge\frac U{12}

integers coprime to bb. The product 16/7716/77 is that for the five smallest primes, and the last inequality holds for U≥2000U\ge2000.

The identity

∑u≤UΩ(u)=∑pj≤U⌊Upj⌋\sum_{u\le U}\Omega(u) =\sum_{p^j\le U}\left\lfloor\frac U{p^j}\right\rfloor

counts each prime-power divisor once. The prime reciprocal estimate and convergence of ∑p∑j≥2p−j=∑p1/(p(p−1))\sum_p\sum_{j\ge2}p^{-j}=\sum_p1/(p(p-1)) give an absolute constant C1C_1 with

∑u≤UΩ(u)≤U(log⁡log⁡U+C1)(U≥2000).\sum_{u\le U}\Omega(u)\le U(\log\log U+C_1) \qquad(U\ge2000).

Thus the number of integers at most UU with Ω(u)>ℓ\Omega(u)>\ell is at most U(log⁡log⁡U+C1)/ℓU(\log\log U+C_1)/\ell. Uniformly for 2000≤U≤L9/402000\le U\le L^9/40, this is o(U)o(U), since

log⁡log⁡U+C1ℓ≤log⁡(9ℓ)+C1ℓ⟶0.\frac{\log\log U+C_1}{\ell} \le\frac{\log(9\ell)+C_1}{\ell}\longrightarrow0.

For large NN it is less than U/12U/12. Consequently one of the coprime integers in the interval has Ω(u)≤ℓ\Omega(u)\le\ell.

Set n=bun=bu. Then n∈[N/2,N]⊆[M,N]n\in[N/2,N]\subseteq[M,N]. Coprimality prevents combining prime exponents across the two factors, so

Ω~(n)=max⁡{Ω~(b),Ω~(u)}≤5ℓ,Ω(n)≤6ℓ+4≤10ℓ\widetilde\Omega(n) =\max\{\widetilde\Omega(b),\widetilde\Omega(u)\}\le5\ell, \qquad \Omega(n)\le6\ell+4\le10\ell

for large NN. All prime-power divisors of bb are at most SS by hypothesis; those of uu are at most u≤L9/40<Su\le L^9/40<S. Therefore n∈An\in A, as required.

Scope

The restricted Proposition 3.2 constructs b=qp′rpb=qp'rp and verifies every hypothesis above. This lemma does not assume a prime in an interval whose lower endpoint may stay bounded.