Olympiad Maths Prep

Track / Stage 6 / 268 of 400 #1268 of 2000

Problem 1268

National olympiad, first round
Combinatorics Difficulty 6.5 Prove it

Is it true that if HH and AA are bounded subsets of the real line, then HH can be decomposed into pairwise disjoint translated copies of AA in at most one way? (We allow for infinitely many translated copies.)

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

I. Solution. We will show that the questioned conclusion is not true. We will recursively construct the sets At,HtA-t, H-t, and the EEE \neq E' translation sets (all non-empty subsets of R\mathbb{R}), such that

H=eE(A+e)=eE(A+e) H=\bigcup_{e \in E}(A+e)=\bigcup_{e' \in E'}\left(A+e'\right)

and all the translations A+e={a+e:aA}(eE)A+e=\{a+e: a \in A\} (e \in E) and A+e={a+e:aA}(eE)A+e'=\left\{a+e': a \in A\right\} (e' \in E') are pairwise disjoint.

First, let A1={0},E1={0},E1={1}A_1=\{0\}, E_1=\{0\}, E_1'=\{1\}, and H1=(A1+E1)(A1+E1)={0,1}H_1=\left(A_1+E_1\right) \cup\left(A_1+E_1'\right)=\{0,1\}. From here, we proceed recursively. Suppose we have already constructed the finite sets An,En,EnA_n, E_n, E_n' such that EnEn=E_n \cap E_n'=\emptyset, and An+EnA_n+E_n and An+EnA_n+E_n' cover each element they cover exactly once (i.e., the sets A+e(eEn)A+e (e \in E_n) are pairwise disjoint, and the sets A+e(eEn)A+e' (e' \in E_n') are also pairwise disjoint). Let Hn=(An+En)(An+En)H_n=\left(A_n+E_n\right) \cup\left(A_n+E_n'\right). Now, for each hHnh \in H_n, we do the following: if hh has not yet been in An+EnA_n+E_n, we add an element aa to AnA_n and an element ee to EnE_n such that their sum is exactly hh, and to ensure that the previous properties do not break, aa should not be of the form a1+e1e2a_1+e_1-e_2 (where these are previous elements: a1An,e1,e2Ena_1 \in A_n, e_1, e_2 \in E_n), and ee should not be of the form e1+a1a2e_1+a_1-a_2 (where e1En,a1,a2Ane_1 \in E_n, a_1, a_2 \in A_n), and the new ee should not be in EnE_n'. Each of these conditions represents a finite number of forbidden elements. We proceed similarly for An+EnA_n+E_n'. Thus, we obtain the sets An+1,En+1,En+1A_{n+1}, E_{n+1}, E_{n+1}', clearly HnAn+1+En+1H_n \subseteq A_{n+1}+E_{n+1}, HnAn+1+En+1H_n \subseteq A_{n+1}+E_{n+1}'.

It is clear that A=n=1An,E=n=1En,E=n=1En,H=n=1HnA=\bigcup_{n=1}^{\infty} A_n, E=\bigcup_{n=1}^{\infty} E_n, E'=\bigcup_{n=1}^{\infty} E_n', H=\bigcup_{n=1}^{\infty} H_n satisfy the conditions, provided that the boundedness also holds. However, we can easily ensure the boundedness by choosing each new element from the interval (2,2)(-2,2) as follows: in a typical step, we want to write a given hHnh(2,2)+(2,2)=(4,4)h \in H_n \ni h \in (-2,2)+(-2,2)=(-4,4) in the form a+ea+e, where aAn+1a \in A_{n+1}, and eEn+1e \in E_{n+1} (or En+1E_{n+1}'). It is clear that since only a finite number of forbidden elements exist, there exist suitable a,e(2,2)a, e \in (-2,2) numbers.

II. Solution (based on the work of Matolcsi Dávid). Let HH be the set of rational numbers with 3-power denominators that fall into the open interval (2,2)(-2,2). Let A(1,1)A \subset (-1,1) be the set of numbers of the form ±(13r)\pm(1-3^{-r}), where rr is a non-negative integer. We will show that HH can be decomposed into pairwise disjoint translates of AA in more than one way.

Let EE be the set of rational numbers with 3-power denominators that fall into the closed interval [1,1][-1,1]. Then A+E={a+e:aA,eE}HA+E=\{a+e: a \in A, e \in E\} \subseteq H (in fact, equality holds). Since AA and A+2/3A+2/3 both contain the number 2/32/3, these two translates of AA are not disjoint. It suffices to show that both can be extended to a decomposition of HH into pairwise disjoint translates of AA. Since HH is countable, it suffices to show that if the translates A+eiA+e_i for some e1,,enEe_1, \ldots, e_n \in E do not contain the number hHh \in H, then there exists a number eEe \in E such that the translate A+eA+e contains the number hh and is disjoint from each of A+e1,,A+enA+e_1, \ldots, A+e_n.

Choose an integer r2r \geq 2 large enough so that 3r2(hei)3^{r-2}(h-e_i) is an integer for all i=1,,ni=1, \ldots, n, and 3r2h3^{-r} \leq 2-|h|. Then h(13r)1|h|-(1-3^{-r}) \leq 1, so we can choose a sign such that with a=±(13r)a= \pm(1-3^{-r}) and e=hae=h-a, we have e1|e| \leq 1, i.e., eEe \in E. Then h=a+eA+eh=a+e \in A+e.

We only need to show that for b,cAb, c \in A and 1in1 \leq i \leq n, b+ec+eib+e \neq c+e_i, i.e., b+hac+eib+h-a \neq c+e_i, or ab+cheia-b+c \neq h-e_i. Since hA+eih \notin A+e_i, we have heiAh-e_i \notin A, so we are done if a=ba=b or a=ca=-c, since in these cases ab+cAa-b+c \in A. (In the latter case, we use the fact that the set AA is symmetric about 0.) In other cases, we show that 3r2(ab+c)3^{r-2}(a-b+c) is not an integer, from which the desired inequality immediately follows.

Let b=±(13s)b= \pm(1-3^{-s}) and c=±(13t)c= \pm(1-3^{-t}), then 3r(ab+c)=±(3r1)(3r3rs)±(3r3rt)3^{r}(a-b+c)= \pm(3^{r}-1) \mp(3^{r}-3^{r-s}) \pm(3^{r}-3^{r-t}), where 3r3^{r} is an integer divisible by 9, and 1±3rs3rt\mp 1 \pm 3^{r-s} \mp 3^{r-t} is not, because if max(s,t)>r\max(s, t)>r then it is either 1\mp 1 or not an integer, and if max(s,t)r\max(s, t) \leq r then it is either 3\mp 3 or not divisible by 3. Thus, 3r(ab+c)3^{r}(a-b+c) cannot be an integer divisible by 9.

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