Wiki
Wiki

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

Updated


Source. Laczkovich (1984), printed p. 109 (PDF p. 1), crediting J. Lawrence, reference [3]. Laczkovich supplies the formula and states that it satisfies the stronger inequality. The verification below is expanded here. Lawrence's separately cited paper is not used as an independently inspected source.

Statement

For positive rational rr, let ν(r)\nu(r) be the least positive integer jj for which rj!∈Zrj!\in\mathbb Z. Define F:Q→[0,∞)F:\mathbb Q\to[0,\infty) by

F(r)={0,r≤0,2 rν(r)!,r>0.F(r)= \begin{cases} 0,&r\le0,\\ 2^{\,r\nu(r)!},&r>0. \end{cases}

Then for every x∈Qx\in\mathbb Q and positive h∈Qh\in\mathbb Q,

2F(x)≤max⁡{F(x+h),F(x+2h)}.(1)2F(x)\le\max\{F(x+h),F(x+2h)\}. \tag{1}

In particular FF satisfies (K), but it is neither nondecreasing nor nonincreasing. It is unbounded on every rational interval (u,v)(u,v) with 0<u<v0<u<v.

Bears on. Problem 1125: this separates the rational domain from the real-line theorem. It is not a counterexample to that theorem.

Proof

The level ν(r)\nu(r) exists because the positive denominator of a reduced fraction divides a sufficiently large factorial. Its exponent rν(r)!r\nu(r)! is a positive integer.

If x≤0x\le0, (1) follows from nonnegativity. Suppose x>0x>0, and write n=ν(x)n=\nu(x). At least one of x+hx+h and x+2hx+2h has level at least nn. This is automatic when n=1n=1. For n≥2n\ge2, if both levels were at most n−1n-1, their products with (n−1)!(n-1)! would be integers. The identity

x=2(x+h)−(x+2h)x=2(x+h)-(x+2h)

would make x(n−1)!x(n-1)! an integer, contradicting minimality of nn.

Choose such a later point yy, and put m=ν(y)≥nm=\nu(y)\ge n. Since y>x>0y>x>0 and m!≥n!m!\ge n!,

ym!>xn!.ym!>xn!.

Both sides are integers, so ym!≥xn!+1ym!\ge xn!+1. Therefore F(y)≥2F(x)F(y)\ge2F(x), proving (1). As F≥0F\ge0, its maximum at the two later points is at most their sum, so (K) follows.

For explicit failures of the two monotonicity directions, observe

F(0)=0<F(1)=2,F(2/3)=24=16>F(1).F(0)=0<F(1)=2,\qquad F(2/3)=2^4=16>F(1).

Here ν(2/3)=3\nu(2/3)=3. Thus FF is neither nonincreasing nor nondecreasing.

Finally, fix 0<u<v0<u<v. For every sufficiently large integer mm, the interval (2mu,2mv)(2^m u,2^m v) has length greater than 22, and so contains an odd integer kmk_m. Put rm=km/2m∈(u,v)r_m=k_m/2^m\in(u,v). Its reduced denominator is 2m2^m. For each fixed integer JJ, this denominator fails to divide J!J! for all sufficiently large mm, so ν(rm)→∞\nu(r_m)\to\infty. Consequently

F(rm)=2rmν(rm)!≥2uν(rm)!⟶∞.F(r_m)=2^{r_m\nu(r_m)!}\ge2^{u\nu(r_m)!}\longrightarrow\infty.

This proves the stated local unboundedness.