Wiki
Wiki

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

Updated


Let

Cx=[x]∖{2kp:k≥0, p prime}.\mathcal C_x=[x]\setminus\{2^kp:k\ge0,\ p\text{ prime}\}.

Then, for x≥10x\ge10,

M(Cx)≤(12+O ⁣((log⁡2x)5log⁡x))π(x).(1)M(\mathcal C_x) \le\left(\frac12+ O\!\left(\frac{(\log_2x)^5}{\log x}\right)\right)\pi(x). \tag{1}

Proof. Decompose Cx\mathcal C_x by Lemma 3.1. The exceptional and secondary parts retain the bounds in Proposition 3.2 and Proposition 3.3. For a primary representation n=dpn=dp, the ratio φ(d)/d=1\varphi(d)/d=1 would imply d=1d=1, and ratio 1/21/2 would imply d=2kd=2^k with k≥1k\ge1, by the support characterization. Both are excluded by the definition of Cx\mathcal C_x. Every remaining fiber therefore has reciprocal mass at most 1/21/2 by Remark 2.2.

Repeat the proof of Proposition 3.4 on this primary subfamily. Its hulls remain disjoint; the term 1/log⁡x1/\log x in each fiber integral is now multiplied by at most 1/21/2, while the logarithmic moment remains at most four. The same calculation bounds its size by

(12+O(log⁡2x/log⁡x))x/log⁡x.\left(\frac12+O(\log_2x/\log x)\right)x/\log x.

Adding the other two contributions and using the PNT gives (1) for sufficiently large xx. A larger absolute error constant covers the bounded range x≥10x\ge10, exactly as in Theorem 1.1. The integer one, if present, causes no exception: for large xx it lies in the small exceptional class. □\square

Source. Tao, published paper, published p.815, Section 4.3. This page uses that published version.

Bears on. Problem 49.