Wiki
Wiki

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

Updated


For y≥10y\ge10 and nonnegative real numbers ada_d indexed by d∈N≤yd\in\mathbb N_{\le y},

∑d∈N≤yad≪(log⁡y)sup⁡d∈N≤yd1−1/log⁡yad.(1)\sum_{d\in\mathbb N_{\le y}}a_d \ll(\log y)\sup_{d\in\mathbb N_{\le y}} d^{1-1/\log y}a_d. \tag{1}

Either side may be infinite. The notation and exact classical inputs are fixed in Definitions and analytic inputs.

Proof. Set α=1−1/log⁡y>0\alpha=1-1/\log y>0. Unique factorization and monotone limits of positive geometric series give

∑d∈N≤yd−α=∏p≤y(1−p−α)−1.\sum_{d\in\mathbb N_{\le y}}d^{-\alpha} =\prod_{p\le y}(1-p^{-\alpha})^{-1}.

For y≥e4y\ge e^4, α≥3/4\alpha\ge3/4, so uniformly in p≤yp\le y,

(1−p−α)−1=1+p1/log⁡yp+O(p−3/2).(1-p^{-\alpha})^{-1} =1+\frac{p^{1/\log y}}p+O(p^{-3/2}).

Since 0≤log⁡p/log⁡y≤10\le\log p/\log y\le1, the inequality eu=1+O(u)e^u=1+O(u) on [0,1][0,1] yields

log⁡∏p≤y(1−p−α)−1=∑p≤y1p+O ⁣(1log⁡y∑p≤ylog⁡pp)+O(1)=log⁡log⁡y+O(1).\log\prod_{p\le y}(1-p^{-\alpha})^{-1} =\sum_{p\le y}\frac1p+ O\!\left(\frac1{\log y}\sum_{p\le y}\frac{\log p}{p}\right)+O(1) =\log\log y+O(1).

The error ∑p−3/2\sum p^{-3/2} converges, and the two prime sums are (2) in the input page. For 10≤y≤e410\le y\le e^4, only finitely many primes occur and α≥1−1/log⁡10>0\alpha\ge1-1/\log10>0. The product is uniformly bounded; enlarging the constant therefore proves the same O(log⁡y)O(\log y) bound.

Let TT be the supremum in (1). If T=∞T=\infty, the assertion is immediate. Otherwise ad≤Td−αa_d\le T d^{-\alpha} for every dd, and summing proves (1). This includes T=0T=0 and the term d=1d=1. □\square

Source. Tao, published paper, published pp.798–799, Lemma 1.5. This page uses that published version.

Bears on. Problem 49.