Wiki
Wiki

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

Updated


Let M<(N)M_{<}(N) be the maximum size of {a1<⋯<at}⊂{1,…,N}\{a_1<\cdots<a_t\}\subset\{1,\ldots,N\} such that φ(a1)<⋯<φ(at)\varphi(a_1)<\cdots<\varphi(a_t). Then, for N≥10N\ge10,

π(N)≤M<(N)≤(1+O ⁣((log⁡2N)5log⁡N))π(N),\pi(N)\le M_{<}(N)\le \left(1+O\!\left(\frac{(\log_2N)^5}{\log N}\right)\right)\pi(N),

and hence M<(N)∼π(N)M_{<}(N)\sim\pi(N) and M<(N)=o(N)M_{<}(N)=o(N).

Proof. Every strict sequence is weak, so M<(N)≤M(N)M_{<}(N)\le M(N). The primes up to NN form a strict sequence because φ(p)=p−1\varphi(p)=p-1. Apply Theorem 1.1 between these lower and upper bounds. Its relative error tends to zero, and the PNT gives π(N)/N→0\pi(N)/N\to0. □\square

This elementary compilation consequence answers the asymptotic and o(N)o(N) clauses of Problem 49. It does not show that the primes are exactly a largest strict example for every NN, and it does not identify strict and weak finite maxima.

Source. Tao, published paper, published pp.793–794, definitions and Theorem 1.1; the transfer is explicit compilation. This page uses that published version.

Bears on. Problem 49.