Wiki
Wiki

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

Updated


Statement

Setting (p. 1). ff is a real additive function: f(m1m2)=f(m1)+f(m2)f(m_1m_2)=f(m_1)+f(m_2) whenever (m1,m2)=1(m_1,m_2)=1. The truncation f′f' is f′(p)=f(p)f'(p)=f(p) when ∣f(p)∣≤1|f(p)|\le1 and f′(p)=1f'(p)=1 otherwise.

Theorem V (p. 3). Let ff be additive, and suppose there are constants c1,c2c_1,c_2 and infinitely many nn for each of which there are integers a1<a2<⋯<ax≤na_1<a_2<\cdots<a_x\le n with x>c1nx>c_1n and ∣f(ai)−f(aj)∣<c2|f(a_i)-f(a_j)|<c_2 for all i,ji,j. Then there is a constant cc such that, with f+(p)=f(p)−clog⁡pf^+(p)=f(p)-c\log p and (f+)′(f^+)' its truncation,

∑p((f+)′(p))2p<∞.\sum_p\frac{((f^+)'(p))^2}{p}<\infty.

The print writes the series as ∑p((f+)′(p)/p)2\sum_p((f^+)'(p)/p)^2, which converges for every cc because ∣(f+)′(p)∣≤1|(f^+)'(p)|\le1; the paper's converse (p. 3), its proof (which ends on p. 14 with ∑(φ′(p))2/p<∞\sum(\varphi'(p))^2/p<\infty) and its use in Theorems IV and XI all take the form above.

Converse (p. 3). If f(p)=clog⁡p+f+(p)f(p)=c\log p+f^+(p) with ∑p((f+)′(p))2/p<∞\sum_p((f^+)'(p))^2/p<\infty, then for every c1<1c_1<1 there is a c2c_2 such that for every nn there are a1<⋯<ax≤na_1<\cdots<a_x\le n with x>c1nx>c_1n and ∣f(ai)−f(aj)∣<c2|f(a_i)-f(a_j)|<c_2.

Definition (p. 3). ff is finitely distributed when it satisfies the hypothesis of Theorem V.

Proof pointer

The converse is proved first (pp. 8--10), by a second-moment estimate for f+(m)f^+(m) around An=∑p≤nf+(p)/pA_n=\sum_{p\le n}f^+(p)/p: for every c2<1c_2<1 there is a c4c_4 with ∣f(m)−clog⁡n−An∣<c4|f(m)-c\log n-A_n|<c_4 for more than c2nc_2n integers m≤nm\le n. The theorem itself (pp. 10--14) splits into three cases on pairs of prime sequences pi,qip_i,q_i with pi/qi→c∈(1,∞)p_i/q_i\to c\in(1,\infty) and ∑1/pi=∑1/qi=∞\sum1/p_i=\sum1/q_i=\infty: f(pi)−f(qi)→±∞f(p_i)-f(q_i)\to\pm\infty (Case 1), f(pi)−f(qi)→0f(p_i)-f(q_i)\to0 with ∑(f′(p))2/p=∞\sum(f'(p))^2/p=\infty (Case 2), and f(pi)−f(qi)→df(p_i)-f(q_i)\to d, printed with 1<d<∞1<d<\infty (Case 3, reduced to Case 2 by subtracting a multiple of log⁡m\log m); in each the finitely distributed hypothesis is contradicted, Case 2 following an earlier density lemma of Erdős.

Read depth

Claims checked: the statement, the converse and the definition read on the page image of p. 3; the proof on pp. 8--14 read for structure. Nothing here is independently reviewed.

Dependencies

None in the corpus. External inputs named by the paper: Turán's method and Lemma 2 of Erdős's 1937 paper on the density of some sequences of numbers.

Source. P. Erdős, On the distribution function of additive functions, Ann. of Math. (2) 47 (1946), 1--20, doi:10.2307/1969031; the edition read is named on the source card.

Bears on

  • Problem 491: the paper says (p. 3) that it deduces from Theorem V the results proved as Theorems XI and XIII, which give f(n)=clog⁡nf(n)=c\log n for nondecreasing ff and for ff with f(n+1)−f(n)→0f(n+1)-f(n)\to0; the proof of Theorem XI (pp. 17--18) uses Theorem V, while the written proof of Theorem XIII (pp. 18--19) does not. Theorem V itself does not address the problem's bounded-difference hypothesis.