Wiki
Wiki

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

Updated


Statement

The Dickman function ϱ\varrho (p. 42) is the continuous function with ϱ(u)=1\varrho(u)=1 for 0≤u≤10\le u\le1 and ϱ(u)=ϱ(N)−∫Nuv−1ϱ(v−1) dv\varrho(u)=\varrho(N)-\int_N^u v^{-1}\varrho(v-1)\,dv for N<u≤N+1N<u\le N+1, N=1,2,…N=1,2,\ldots.

Theorem 2 (p. 43). Let ε\varepsilon and θ\theta be real numbers with 0<ε<10<\varepsilon<1 and 0<θ<10<\theta<1. Let kk and ll be positive integers with 2≤l≤θlog⁡k2\le l\le\theta\log k, where kk exceeds a number effectively computable in terms of ε\varepsilon and θ\theta. Then there are distinct positive integers a1,…,aka_1,\ldots,a_k and distinct non-negative integers b1,…,blb_1,\ldots,b_l with

(7)P(∏i=1k∏j=1l(ai+bj))<kh(θ)+ε,h(θ)=min⁡u≥1(1−θlog⁡ϱ(u)u).\text{(7)}\qquad P\Bigl(\prod_{i=1}^k\prod_{j=1}^l(a_i+b_j)\Bigr)<k^{h(\theta)+\varepsilon}, \qquad h(\theta)=\min_{u\ge1}\Bigl(\frac{1-\theta\log\varrho(u)}{u}\Bigr).

Here P(n)P(n) is the greatest prime factor of nn.

Remarks of the paper (pp. 43--44). The minimum defining h(θ)h(\theta) is attained, and h(θ)<1h(\theta)<1 for 0<θ<10<\theta<1, so (7) improves by a power on the trivial bound k+lk+l given by ai=ia_i=i, bj=j−1b_j=j-1. For θ≤1/6\theta\le1/6, Buchstab's lower bound for ϱ\varrho gives h(θ)≤θ(1+log⁡(1/θ)+log⁡log⁡(1/θ)+6log⁡log⁡(1/θ)/log⁡(1/θ))h(\theta)\le\theta\bigl(1+\log(1/\theta)+\log\log(1/\theta)+6\log\log(1/\theta)/\log(1/\theta)\bigr). The bound (7) also holds with ω\omega, the number of distinct prime factors, in place of PP. In the introduction (p. 39) the authors state that it follows from Theorem 2 that, even for ll of the form δlog⁡k\delta\log k with 0<δ<10<\delta<1, the bound (2), P(a+b)>C3log⁡klog⁡log⁡kP(a+b)>C_3\log k\log\log k (p. 38), cannot be replaced by k1−εk^{1-\varepsilon} for every ε>0\varepsilon>0 and k≥k0(δ,ε)k\ge k_0(\delta,\varepsilon).

Conjectures stated (pp. 39 and 44). For l>log⁡kl>\log k and every ε>0\varepsilon>0 the authors conjecture that some a∈Aa\in A, b∈Bb\in B have P(a+b)>k1−εP(a+b)>k^{1-\varepsilon} for k≥k1(ε)k\ge k_1(\varepsilon) (p. 39); and that there is no positive real γ<1\gamma<1 with arbitrarily large l,kl,k, l>log⁡kl>\log k, admitting distinct positive a1,…,aka_1,\ldots,a_k and distinct non-negative b1,…,blb_1,\ldots,b_l with ω(∏i,j(ai+bj))<(π(k+l))γ\omega\bigl(\prod_{i,j}(a_i+b_j)\bigr)<(\pi(k+l))^\gamma (p. 44).

Proof pointer

Pp. 47--48. Lemma 6 (p. 46) is the analogue of Lemma 3 with the smooth-number count taken from Dickman's asymptotic ψ(x,x1/u)∼xϱ(u)\psi(x,x^{1/u})\sim x\varrho(u) (Lemma 5, p. 46): applying Lemma 1 to the N1/uN^{1/u}-smooth integers up to NN gives about (N/l)ϱ(u)l(N/l)\varrho(u)^l integers aa with every a+bja+b_j smooth. The proof of Theorem 2 takes u=u0u=u_0, a point where (1−θlog⁡ϱ(u))/u(1-\theta\log\varrho(u))/u is minimal, and N=⌈kl((1−ε/2)ϱ(u0))−l⌉N=\lceil kl((1-\varepsilon/2)\varrho(u_0))^{-l}\rceil.

Read depth

Claims checked: the definition of ϱ\varrho, the statement and the remarks above were read clause by clause on the page images of the print. The proof was followed for the outline above and is not independently verified.

Dependencies

  • Lemma 1 (p. 39), through Lemma 6 (p. 46).

Source. P. Erdős, C. L. Stewart and R. Tijdeman, Some diophantine equations with many solutions, Compositio Mathematica 66 (1988), 37--56; the edition read is named on the source card.

Bears on

No problem page of this corpus.