Wiki
Wiki

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

Updated


Source context: published paper, printed pp. 413–414 (PDF pp. 3–4), the proof of Theorem 3. The elementary details below expand its numerical monotonicity assertions.

Statement

Write

a=2.1454,c=0.0077629,g(t)=log⁡t−at,h(t)=log⁡t−at(t+log⁡t).a=2.1454,\quad c=0.0077629,\quad g(t)=\frac{\log t-a}{t},\quad h(t)=\frac{\log t-a}{t(t+\log t)}.

Then

g(t)>c,20≤t≤500,h(t)>1.6⋅10−6,493≤t≤1800.(1)\begin{array}{ll} g(t)>c,&20\le t\le500,\\ h(t)>1.6\cdot10^{-6},&493\le t\le1800. \end{array} \tag{1}

Also

log⁡500<7,log⁡1800<8,log⁡1792>7.49,a+1657000018002<7.26,(2)\log500<7,\quad \log1800<8,\quad \log1792>7.49, \quad a+\frac{16570000}{1800^2}<7.26, \tag{2}

and

e20(20+log⁡20)<1011.(3)e^{20}(20+\log20)<10^{11}. \tag{3}

Full proof

The derivative

g′(t)=1+a−log⁡tt2g'(t)=\frac{1+a-\log t}{t^2}

changes sign at most once, from positive to negative. Thus the minimum of gg on [20,500][20,500] is at an endpoint. The directed rational certificate proves g(20)>cg(20)>c and g(500)>cg(500)>c, so the first line of (1) holds throughout the interval; it does not assume that gg decreases on the whole interval.

Let N(t)=log⁡t−aN(t)=\log t-a and D(t)=t(t+log⁡t)D(t)=t(t+\log t). On t≥493t\ge493, the certified inequality N(493)>1N(493)>1 implies N(t)>1N(t)>1. Moreover

N′(t)D(t)=t+log⁡t,D′(t)=2t+log⁡t+1.N'(t)D(t)=t+\log t,\qquad D'(t)=2t+\log t+1.

Therefore N′D−ND′<0N'D-ND'<0, and hh is strictly decreasing. Its minimum on [493,1800][493,1800] is h(1800)h(1800), which the same exact checker proves exceeds 1.6⋅10−61.6\cdot10^{-6}. This proves the second line.

Every inequality in (2)–(3) is separately checked by the same rational logarithm/exponential enclosures or direct rational arithmetic, with strict endpoint margins. The complete acceptance comparisons are printed by the checker.

Source precision

The first branch of the printed proof assumes 1011≤pk≤e50010^{11}\le p_k\le e^{500}, but the sentence bounding the minimum of g(log⁡k)g(\log k) refers instead to pk≤e1800p_k\le e^{1800}. That enlargement is not supported: the checker also verifies g(1800)<cg(1800)<c. The complete proof uses only the intended first branch through e500e^{500}. The intermediate range through e1800e^{1800} is treated separately with Theorem 2, exactly as the paper proceeds to do.