Maths Olympiad Prep

Track / Stage 7 / 272 of 300 #2152 of 2444

Problem 2152

National Olympiad second round; IMO P1/P4
Number theory Difficulty 7.9 Prove it Nordic Mathematical Contest · Nordic Mathematical Olympiad

Alice and Bob are playing a game. First, Alice chooses a partition C\mathcal{C} of the positive integers into a (not necessarily finite) set of sets, such that each positive integer is in exactly one of the sets in C\mathcal{C}. Then Bob does the following operation a finite number of times.

Choose a set SCS \in \mathcal{C} not previously chosen, and let DD be the set of all positive integers dividing at least one element in SS. Then add the set D\SD \backslash S (possibly the empty set) to C\mathcal{C}.

Bob wins if there are two equal sets in C\mathcal{C} after he has done all his moves, otherwise, Alice wins. Determine which player has a winning strategy.

Answer: Alice has a winning strategy.

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.

Next problem →

Official solutions — 2

Solution 1

Solution:

We will use the following notation. If SZ+S \subseteq \mathbb{Z}^{+} is a subset of the positive integers and DD is defined as in the problem statement, define d(S)=D\Sd(S) = D \backslash S. In addition, the dd-sequence of SS, written {dS}nZ+\{d^{S}\}_{n \in \mathbb{Z}^{+}}, is defined recursively as d0S=Sd_{0}^{S} = S, and dn+1S=d(dnS)d_{n+1}^{S} = d(d_{n}^{S}) for each nZ+n \in \mathbb{Z}^{+}.

Let p1,p2,p_{1}, p_{2}, \ldots be an enumeration of all the prime numbers. Alice chooses the sets
A={pmnnmmod3, 0<nm},B={pmnnm1mod3, 0<nm1} A = \{p_{m}^{n} \mid n \equiv m \bmod 3,\ 0 < n \leq m\}, \quad B = \{p_{m}^{n} \mid n \equiv m-1 \bmod 3,\ 0 < n \leq m-1\}
and
C=Z+\(AB) C = \mathbb{Z}^{+} \backslash (A \cup B)
It is clear that each positive integer belongs to exactly one of the sets AA, BB or CC. By induction we readily get that
dk(A)={pmnnmmod3, nm3k/2}{pmnn≢mmod3, nm13(k1)/2} if k is even and k0 d^{k}(A) = \begin{array}{ll} \{p_{m}^{n} \mid n \equiv m \bmod 3,\ n \leq m-3k/2\} \\ \{p_{m}^{n} \mid n \not\equiv m \bmod 3,\ n \leq m-1-3(k-1)/2\} \end{array} \quad \text{ if } k \text{ is even and } k \neq 0
and
dk(B)={pmnnm1mod3, nm13k/2} if k is even and k0{pmnn≢m1mod3, nm23(k1)/2} if k is odd.  d^{k}(B) = \begin{array}{lll} \{p_{m}^{n} \mid n \equiv m-1 \bmod 3,\ n \leq m-1-3k/2\} & \text{ if } k \text{ is even and } k \neq 0 \\ \{p_{m}^{n} \mid n \not\equiv m-1 \bmod 3,\ n \leq m-2-3(k-1)/2\} & \text{ if } k \text{ is odd. } \end{array}
If mZ+m \in \mathbb{Z}^{+} then 6m6m is definitely in CC as 6m6m is not a power of a prime. This means that all positive integers divide at least one number in CC. It follows that
d(C)=Z+\C=Z+\(Z+\(AB))=AB={pmnn≢m2mod3, 0<nm} d(C) = \mathbb{Z}^{+} \backslash C = \mathbb{Z}^{+} \backslash (\mathbb{Z}^{+} \backslash (A \cup B)) = A \cup B = \{p_{m}^{n} \mid n \not\equiv m-2 \bmod 3,\ 0 < n \leq m\}
By induction it follows that all of the sets dk(M)d^{k}(M) are all distinct. Hence Bob can never create the same set in two different ways.

Remark. In fact Alice has to partition the positive integers in at least three parts if she wishes to beat Bob. It is clear that selecting a single set, namely Z+\mathbb{Z}^{+}, will not do as d(Z+)=d(\mathbb{Z}^{+}) = \emptyset and d()=d(\emptyset) = \emptyset. Next assume Alice has partitioned the positive integers in two sets AA and BB. It can be assumed that d(A)Bd(A) \neq B and d(B)Ad(B) \neq A as Bob would win in either case. As d(A)d(A) is disjoint from AA and BB is the complement of AA this means that d(A)d(A) is a strict subset of BB. Hence there is some bb in BB that does not belong to d(A)d(A). The implication is that bb does not divide any number in AA. Similarly there exists a number aa in AA that does not divide any number in BB. Their product abab belongs to either of the sets AA or BB and is divisible by both aa and bb. This contradicts the existence of said numbers, aa and bb. In fact we have proven for any subset AA of Z+\mathbb{Z}^{+} that d(A)=Z+\Ad(A) = \mathbb{Z}^{+} \backslash A or d(Z+\A)=Ad(\mathbb{Z}^{+} \backslash A) = A.

Solution 2

Solution:

Use the same notation as in solution 1. We wish to construct a partition iZ+Ai\bigcup_{i \in \mathbb{Z}^{+}} A_{i} of the positive integers, such that no two sets among all sets in {dA1}nZ+,{dA2}nZ+,\{d^{A_{1}}\}_{n \in \mathbb{Z}^{+}}, \{d^{A_{2}}\}_{n \in \mathbb{Z}^{+}}, \ldots are equal.

To begin with, consider a partition iZ+Pi\bigcup_{i \in \mathbb{Z}^{+}} P_{i} of the primes, such that each Pi={pi1,pi2,}P_{i} = \{p_{i1}, p_{i2}, \ldots\} is infinite. We may assume each set is infinite since the set of primes and Z+×Z+\mathbb{Z}^{+} \times \mathbb{Z}^{+} both are countable, so there exists a bijection between them. Define the sets
Qi={pi1,pi22,pi3,pi33,pi42,pi44,pi5,pi53,pi55,} Q_{i} = \{p_{i1}, p_{i2}^{2}, p_{i3}, p_{i3}^{3}, p_{i4}^{2}, p_{i4}^{4}, p_{i5}, p_{i5}^{3}, p_{i5}^{5}, \ldots\}
consisting of pijj,pijj2,,pij(1,2)p_{ij}^{j}, p_{ij}^{j-2}, \ldots, p_{ij}^{(1,2)} for each pijPip_{ij} \in P_{i}. By the exponent (1,2)(1,2) we indicate that the last power is 1 when jj is odd and 2 when it is even.

We will include the remaining integers {r1,r2,}=Z+\(iZ+Qi)\{r_{1}, r_{2}, \ldots\} = \mathbb{Z}^{+} \backslash (\bigcup_{i \in \mathbb{Z}^{+}} Q_{i}) to the partition as follows. First, add r1r_{1} to some set Qi1Q_{i_{1}}, such that gcd(r1,q)=1\gcd(r_{1}, q) = 1 for each qQi1q \in Q_{i_{1}}. Then, for each kZ+k \in \mathbb{Z}^{+}, add rk+1r_{k+1} to a set Qik+1Q_{i_{k+1}} where ik+1>iki_{k+1} > i_{k}, such that gcd(rk+1,q)=1\gcd(r_{k+1}, q) = 1 for each qQik+1q \in Q_{i_{k+1}}. Note that such an ik+1i_{k+1} always exists, since there always is an infinite amount of prime divisors among Qik+1,Qik+2,Q_{i_{k}+1}, Q_{i_{k}+2}, \ldots, and rk+1r_{k+1} only has a finite amount of prime divisors.

The collection of QiQ_{i} now forms a partition of Z+\mathbb{Z}^{+}. What remains is to show that it satisfies Alice's winning condition.

We first see that we may ignore all the rkr_{k}. By definition, no divisor of rkr_{k} coincides with any divisor of QikQ_{i_{k}}, and their contributions in the dd-sequence will therefore be completely disjoint. As d(d({rk}))=d(d(\{r_{k}\})) = \emptyset, the dd-sequence will show no trace of the rkr_{k} after the second element. Hence, we will work with the original QiQ_{i}.

We observe that d(Qi)={1,pi2,pi32,pi4,pi43,pi52,pi54,pi6,pi63,pi65,}d(Q_{i}) = \{1, p_{i2}, p_{i3}^{2}, p_{i4}, p_{i4}^{3}, p_{i5}^{2}, p_{i5}^{4}, p_{i6}, p_{i6}^{3}, p_{i6}^{5}, \ldots\}. This is simply QiQ_{i} with the indices shifted by 1 and with an added 1. As this set essentially is on the same form as QiQ_{i}, we see that {dQi}n\{d^{Q_{i}}\}_{n} consists of every possible shift in indices of the first set QiQ_{i} (except the element 1 appearing in every other set). It is therefore obvious that the sets in the sequence are pairwise distinct. Additionally, the set of primes dividing some element in some set of {dQi}n\{d^{Q_{i}}\}_{n} is PiP_{i}. Since the PiP_{i} partition the primes, it is clear that all the sets in all the dd-sequences of the QiQ_{i} are distinct, which is what we wanted.

Remark. The main idea of the problem is to construct a single set SS, such that the sets in the dd-sequence of SS are pairwise distinct. Any set with similar properties to the QiQ_{i} should give a valid initial collection C\mathcal{C} such that Alice wins.

Source: MathNet, licensed CC-BY-4.0. Statement reproduced verbatim; metadata (topic, difficulty, ordering) added by this project.