Wiki
Wiki

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

Updated


For 2≤z≤y2\le z\le y, some absolute c>0c>0 gives

∑z≤p≤y1p≪e−clog⁡z+log⁡(y/z)log⁡z.(1)\sum_{z\le p\le y}\frac1p \ll\frac{e^{-c\sqrt{\log z}}+\log(y/z)}{\log z}. \tag{1}

Proof. Partial summation of the PNT in notation gives the more precise Mertens form

∑p≤t1p=log⁡log⁡t+B+O(1/log⁡t)(t≥10).(2)\sum_{p\le t}\frac1p =\log\log t+B+O(1/\log t)\qquad(t\ge10). \tag{2}

Indeed, the prime sum equals π(t)/t+∫2tπ(u)u−2du\pi(t)/t+\int_2^t\pi(u)u^{-2}du. Substituting π(u)=li⁡(u)+E(u)\pi(u)=\operatorname{li}(u)+E(u) makes the main part log⁡log⁡t\log\log t plus a constant. The integral of E(u)/u2E(u)/u^2 converges, and its tail, together with E(t)/tE(t)/t, is O(1/log⁡t)O(1/\log t) by the exponential PNT error. Bounded smaller tt can be absorbed.

If y>2zy>2z, (2), with the possible endpoint term 1/z1/z, gives

∑z≤p≤y1p≤log⁡log⁡y−log⁡log⁡z+O(1/log⁡z)≪log⁡(y/z)log⁡z.\sum_{z\le p\le y}\frac1p \le\log\log y-\log\log z+O(1/\log z) \ll\frac{\log(y/z)}{\log z}.

Here log⁡(1+u)≤u\log(1+u)\le u and log⁡(y/z)>log⁡2\log(y/z)>\log2 absorb the error. If z≤y≤2zz\le y\le2z and z≥10z\ge10, Lemma 1.6 gives

∑z≤p≤y1p≤π([z,y])z≪y−zzlog⁡z+e−c0log⁡z.\sum_{z\le p\le y}\frac1p \le\frac{\pi([z,y])}{z} \ll\frac{y-z}{z\log z}+e^{-c_0\sqrt{\log z}}.

On this range (y−z)/z≪log⁡(y/z)(y-z)/z\ll\log(y/z). Decrease c0c_0 to c>0c>0 so e−c0log⁡z≪e−clog⁡z/log⁡ze^{-c_0\sqrt{\log z}}\ll e^{-c\sqrt{\log z}}/\log z. For 2≤z<102\le z<10 and y≤2zy\le2z, the left side is bounded, while the exponential term divided by log⁡z\log z has a positive lower bound on that compact range. This completes every case. □\square

Source. Tao, published paper, published p.799, Lemma 1.7. This page uses that published version.

Bears on. Problem 49.