Maths Olympiad Prep

Track / Stage 6 / 239 of 400 #1239 of 1964

Problem 1239

National olympiad, first round
Number theory Difficulty 6.4 Prove it

Is it true that among five irrational numbers, there are always three such that the sum of any two of them is irrational?

This one wants a proof. Work it on paper, then read the official solution and mark yourself. Be honest about it: the record is only any use to you if it is.

Official solution

Among the numbers under consideration, we call aa and bb acquaintances if there exist numbers

a=c0,c1,c2,,cn1,cn=b a=c_{0}, c_{1}, c_{2}, \ldots, c_{n-1}, c_{n}=b

such that

c0c1=3k1,c1c2=3k2,,cn1cn=3kn \left|c_{0}-c_{1}\right|=3^{k_{1}},\left|c_{1}-c_{2}\right|=3^{k_{2}}, \ldots,\left|c_{n-1}-c_{n}\right|=3^{k_{n}}

where k1,k2,,knk_{1}, k_{2}, \ldots, k_{n} are integers. The numbers (c0,c1,c2,,cn)\left(c_{0}, c_{1}, c_{2}, \ldots, c_{n}\right) can also be equal. If aa and bb are acquaintances, we will call them friends or enemies depending on whether the above nn is even or odd. (Specifically, if ab=3k|a-b| = 3^{k}, then aa and bb are enemies.) It is easily seen that the following rules hold:

(1) If aa and bb are acquaintances, and bb and cc are acquaintances, then aa and cc are also acquaintances.

(2) If aa and bb are friends, and aa and cc are friends, then bb and cc are also friends.

(3) If aa and bb are enemies, and aa and cc are enemies, then bb and cc are friends.

It is far from obvious that the following statement holds:

(4) If aa and bb are friends, they cannot be enemies.

Suppose, however, that aa and bb are friends and also enemies. This means that among the given numbers, we can select c1,c2,,c2m1c_{1}, c_{2}, \ldots, c_{2 m-1} and d1,d2,,d2td_{1}, d_{2}, \ldots, d_{2 t} such that

ac1=3k1,c1c2=3k2,,c2m1b=3k2mad1=3l1,d1d2=3l2,,d2tb=3l2t+1 \begin{aligned} & \left|a-c_{1}\right|=3^{k_{1}},\left|c_{1}-c_{2}\right|=3^{k_{2}}, \ldots,\left|c_{2 m-1}-b\right|=3^{k_{2 m}} \\ & \quad\left|a-d_{1}\right|=3^{l_{1}},\left|d_{1}-d_{2}\right|=3^{l_{2}}, \ldots,\left|d_{2 t}-b\right|=3^{l_{2 t+1}} \end{aligned}

(with appropriate integers ki,ljk_{i}, l_{j}). Then

ac1=s13k1,,c2m1b=s2m3k2md1a=f13l1,,d2tb=f2t+13l2t+1 \begin{aligned} & \left|a-c_{1}\right|=s_{1} \cdot 3^{k_{1}}, \ldots,\left|c_{2 m-1}-b\right|=s_{2 m} \cdot 3^{k_{2 m}} \\ & \left|d_{1}-a\right|=f_{1} \cdot 3^{l_{1}}, \ldots,\left|d_{2 t}-b\right|=f_{2 t+1} \cdot 3^{l_{2 t+1}} \end{aligned}

where each si,fjs_{i}, f_{j} is either 1 or (1)(-1). Adding the above equalities, we get

0=i=12msi3ki+j=12t+1fj3lj 0=\sum_{i=1}^{2 m} s_{i} \cdot 3^{k_{i}}+\sum_{j=1}^{2 t+1} f_{j} \cdot 3^{l_{j}}

Let MM be the maximum of the absolute values of ki,ljk_{i}, l_{j}. Multiplying both sides of (5) by 3M3^{M}, we obtain

0=i=12msi3ki+M+j=12t+1fj3lj+M 0=\sum_{i=1}^{2 m} s_{i} \cdot 3^{k_{i}+M}+\sum_{j=1}^{2 t+1} f_{j} \cdot 3^{l_{j}+M}

Since here ki+M,lj+M0k_{i}+M, l_{j}+M \geqq 0, each term in the sums in (6) is an odd integer. The number of terms, 2n+2t+12 n+2 t+1, is also odd, so the sum cannot be even (and therefore not 0); this is a contradiction, which proves (4).

To be able to divide our numbers into two groups as desired, we first classify them into classes as follows. Arbitrarily select a number a1a_{1} from among them, and place a1a_{1} and its acquaintances in the first class. If there are any numbers left, select one of them, a2a_{2}, and place a2a_{2} and its acquaintances in the second class. Continuing this process, we can eventually classify all numbers into rr classes, where the classes consist of a1,a2,,ara_{1}, a_{2}, \ldots, a_{r} and their acquaintances. Numbers in different classes never have a difference of the form 3k3^{k}; if aa is in the ii-th class and bb is in the jj-th class, and ab=3k(ij,k|a-b|=3^{k} (i \neq j, k integer), then aa and bb are acquaintances. However, aia_{i} and aa, as well as bb and aja_{j}, are also acquaintances, so by (1), aia_{i} and aja_{j} are acquaintances, which is impossible.

It is therefore sufficient to divide the numbers in the same class into two groups as desired. For the pp-th class, place apa_{p} and the numbers friendly with apa_{p} in the first group, and the numbers hostile to apa_{p} in the second group. By (2), any two (different) numbers in the first group are friends, and by (3), the same is true for any two numbers in the second group. By (4), numbers in the same group cannot be enemies, so their difference is certainly not of the form 3k3^{k}.

Remarks. 1. The statement of the problem is also true if 3k3^{k} is replaced by aka^{k} for any odd aa. With some modifications to the given proof, it can also be shown that not only a finite number, but any number (e.g., all!) real numbers can be divided into two groups in the desired manner.

2. Consider the graph whose vertices are the real numbers in the problem, and two vertices are connected by an edge if the difference of the corresponding numbers is of the form 3k3^{k}. According to the problem, this graph is bipartite. It is well known that a graph is bipartite if and only if every cycle in it has an even number of edges. In the course of the above proof, we essentially showed that this graph has no cycles with an odd number of edges (1. (4)), and then used this to prove the bipartiteness of the graph.

Source: NuminaMath-1.5, licensed Apache-2.0. Statement and solution reproduced as published; topic, difficulty and ordering added by this site.