Olympiad Maths Prep

Track / Stage 9 / 53 of 80 #1933 of 2000

Problem 1933

IMO P2/P5; hard shortlist
Number theory Difficulty 9.1 Prove it China-TST-2023B · China

Does there exist an irrational number xx such that there are at most finitely many positive integers nn satisfying
{kx}1n+1 \{kx\} \geq \frac{1}{n+1}
for every k{1,,n}k \in \{1, \dots, n\}?

Note: Here, for a positive real number yy, {y}\{y\} denotes the fractional part of yy.

This one wants a proof. Work it on paper, read the official solution, then mark yourself honestly — the ladder only means something if the record is true.

Official solution

*Proof* 1. Nonexistence. Assume that there exists such a positive integer nn. Let xx be an irrational number. Since the terms of the sequence {x}\{x\}, {2x}\{2x\}, \dots are distinct and densely distributed in the interval (0,1)(0, 1), there exist infinitely many positive integers dd satisfying
{dx}=min{{ax}a=1,2,,d}. \{dx\} = \min \{ \{ax\} \mid a = 1, 2, \dots, d \}.
Arrange these dd in an infinite increasing sequence d1<d2<d_1 < d_2 < \dots.
We will now show that for all n=di1n = d_i - 1 (i2i \ge 2), we always have
min{{kx}k=1,,n}={di1x}1di. \min \{ \{kx\} \mid k = 1, \dots, n \} = \{d_{i-1}x\} \ge \frac{1}{d_i}.
Suppose that this inequality does not hold, then di{di1x}1d_i \cdot \{d_{i-1}x\} \le 1. Thus,
{di1dix}={didi1x+di{di1x}}=di{di1x}. \{d_{i-1}d_i x\} = \{d_i \cdot \lfloor d_{i-1}x \rfloor + d_i \cdot \{d_{i-1}x\}\} = d_i \cdot \{d_{i-1}x\}.
On the other hand,
{didi1x}={di1dix+di1{dix}}di1{dix}. \{d_i d_{i-1} x\} = \{d_{i-1} \lfloor d_i x \rfloor + d_{i-1} \{d_i x\}\} \le d_{i-1} \{d_i x\}.
Combining the above two equations, we obtain di1{dix}di{di1x}d_{i-1} \cdot \{d_i x\} \ge d_i \cdot \{d_{i-1} x\}. However, di1<did_{i-1} < d_i and {dix}<{di1x}\{d_i x\} < \{d_{i-1} x\}, which leads to a contradiction!
Therefore, n=di1n = d_i - 1 satisfies the given condition, and there are infinitely many such nn. \square

*Proof* 2. Nonexistent. If there is no irrational number xx satisfying the condition, then there exists n0Z+n_0 \in \mathbb{Z}_+ such that for any nn0n \ge n_0, there exists k{1,,n}k \in \{1, \dots, n\} such that {kx}<1n+1\{kx\} < \frac{1}{n+1}. Since xx is an irrational number, for any kZ+k \in \mathbb{Z}_+, {kx}0\{kx\} \ne 0. For n0n_0, there exist k0{1,,n0}k_0 \in \{1, \dots, n_0\} and 0Z\ell_0 \in \mathbb{Z} such that
0<k0x0<1n0+1. 0 < k_0 x - \ell_0 < \frac{1}{n_0 + 1}.
Without loss of generality, assume that k0k_0 and 0\ell_0 are coprime. Otherwise, we can replace k0k_0 and 0\ell_0 with k0gcd(k0,0)\frac{k_0}{\gcd(k_0, \ell_0)} and 0gcd(k0,0)\frac{\ell_0}{\gcd(k_0, \ell_0)} respectively, and the condition still holds.
Let n1=1k0x0>n0n_1 = \left\lfloor \frac{1}{k_0 x - \ell_0} \right\rfloor > n_0. Then there exist k1{1,,n1}k_1 \in \{1, \dots, n_1\} and 1Z\ell_1 \in \mathbb{Z} such that
0<k1x1<1n1+1<k0x0. 0 < k_1 x - \ell_1 < \frac{1}{n_1 + 1} < k_0 x - \ell_0.

Similarly, we can assume that k1k_1 and 1\ell_1 are coprime. Since k0k_0 is coprime with 0\ell_0 and k1k_1 is coprime with 1\ell_1, we have 0k01k1\frac{\ell_0}{k_0} \neq \frac{\ell_1}{k_1}. Therefore,
1k01k10=k1(k0x0)k0(k1x1)<max{k1(k0x0),k0(k1x1)}(because k1(k0x0)>0,k0(k1x1)>0)max{k1n1,k0n1+1}1(because 1k0x0n1, hence k0x01n1.) \begin{aligned} 1 \le |k_0\ell_1 - k_1\ell_0| &= |k_1(k_0x - \ell_0) - k_0(k_1x - \ell_1)| \\ &< \max \{k_1(k_0x - \ell_0), k_0(k_1x - \ell_1)\} \quad (\text{because } k_1(k_0x - \ell_0) > 0, k_0(k_1x - \ell_1) > 0) \\ &\le \max \left\{\frac{k_1}{n_1}, \frac{k_0}{n_1 + 1}\right\} \le 1 \quad (\text{because } \frac{1}{k_0x - \ell_0} \ge n_1, \text{ hence } k_0x - \ell_0 \le \frac{1}{n_1}.) \end{aligned}
This leads to a contradiction. \Box

*Proof 3.* For any irrational number xx, the fractional part {kx}\{kx\} is distinct and densely distributed in the interval (0,1)(0,1). Hence, there exist infinitely many positive integers mm satisfying:
{mx}<{kx},k=1,2,,m1. \{mx\} < \{kx\}, \quad \forall k = 1, 2, \dots, m-1.
For each such m2m \ge 2, let β\beta be the smallest value among {x}\{x\}, {2x}\{2x\}, ..., {(m1)x}\{(m-1)x\}. Thus, for k=1,2,,m1k = 1, 2, \dots, m-1, we have {kx}β\{kx\} \ge \beta, which implies that there are no integers in the open intervals (kxβ,kx)(kx - \beta, kx).
Consider the points O:(0,0)O : (0,0), A:(m,mx)A : (m,mx), B:(m,mxβ)B : (m,mx - \beta), C:(0,β)C : (0,-\beta) on the coordinate plane. The parallelogram OABCOABC does not contain any lattice points in its interior, and on its boundary, there are exactly three lattice points: OO, D:(m,mx)D : (m, \lfloor mx \rfloor), and E:(r,rxβ)=(r,rx)E : (r, rx - \beta) = (r, \lfloor rx \rfloor).
The lattice triangle ODE\triangle ODE has no lattice points inside or on its boundary, except for the vertices. By Pick's theorem, its area is 12\frac{1}{2}. Therefore, the area of the parallelogram OABCOABC is 2SODE=1\ge 2S_{\triangle ODE} = 1. On the other hand, the area of this parallelogram is m×β=m×{rx}1m \times \beta = m \times \{rx\} \ge 1. Hence, β1m\beta \ge \frac{1}{m}, and for k=1,2,,m1k = 1, 2, \dots, m-1, we have
{kx}{rx}1m. \{kx\} \ge \{rx\} \ge \frac{1}{m}.
Therefore, n=m1n = m - 1 satisfies the given condition. There are infinitely many such nn, implying that no irrational number xx satisfies the condition. \Box

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic, difficulty and ordering added by this site.