Maths Olympiad Prep

Track / Stage 7 / 59 of 300 #1939 of 2444

Problem 1939

National Olympiad second round; IMO P1/P4
Number theory Difficulty 7.2 Prove it Taiwan IMO Selection Camp · Taiwan

Let kk be a positive integer. A sequence a0,a1,,ana_0, a_1, \dots, a_n (n>0n > 0) of positive integers satisfies the following conditions:
(i) a0=an=1a_0 = a_n = 1;
(ii) 2aik2 \le a_i \le k for each i=1,2,,n1i = 1, 2, \dots, n-1;
(iii) For each j=2,3,,kj = 2, 3, \dots, k, the number jj appears φ(j)\varphi(j) times in the sequence a0,a1,,ana_0, a_1, \dots, a_n (φ(j)\varphi(j) is the number of positive integers that do not exceed jj and are coprime to jj);
(iv) For any i=1,2,,n1i = 1, 2, \dots, n-1, gcd(ai1,ai)=1=gcd(ai,ai+1)\text{gcd}(a_{i-1}, a_i) = 1 = \text{gcd}(a_i, a_{i+1}), and aia_i divides ai1+ai+1a_{i-1} + a_{i+1}.
There is another sequence b0,b1,,bnb_0, b_1, \dots, b_n of integers such that bi+1ai+1>biai\frac{b_{i+1}}{a_{i+1}} > \frac{b_i}{a_i} for all i=0,1,,n1i = 0, 1, \dots, n-1. Find the minimum value for bnb0b_n - b_0.

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 solution

The minimum value of bnb0b_n - b_0 is 1.
For convenience, we call a sequence a0,a1,,ana_0, a_1, \dots, a_n satisfying the conditions of the problem a "kk-good sequence".
First, we prove that the kk-good sequence is unique. To do this, we strengthen the statement, proving simultaneously that the kk-good sequence satisfies the following condition:
If (a,b)=1,a+bk+1(a, b) = 1, a + b \ge k + 1 and 1a,bk1 \le a, b \le k, then there exists a unique positive integer ii satisfying ai=a,ai+1=ba_i = a, a_{i+1} = b. ...... (*)
We induct on kk. When k=1k = 1, if n2n \ge 2, then 11n11 \le 1 \le n-1, so by (1) we know 2ai12 \le a_i \le 1, a contradiction. Hence n=1n = 1, so this sequence can only be 1,11, 1, and is therefore unique.
Next we prove (*). If (a,b)=1,a+bk+1(a, b) = 1, a + b \ge k + 1 and a,bka, b \le k, then since k=1k = 1 we know a=b=1a = b = 1. Also a0=a1=1a_0 = a_1 = 1, so (*) holds.
Suppose the statement holds for k=t1k = t - 1 (t2t \ge 2). Then for k=tk = t, if ai=ta_i = t, then since a0=an=1ta_0 = a_n = 1 \ne t, we have 1in11 \le i \le n-1. By (3) we know ai1,ai+1ta_{i-1}, a_{i+1} \ne t (otherwise they would not be coprime), that is, tt cannot be adjacent to itself. So ai1,ai+1<ta_{i-1}, a_{i+1} < t. By (3) we know aiai1+ai+1a_i | a_{i-1} + a_{i+1}, but 0<ai1+ai+1<2t0 < a_{i-1} + a_{i+1} < 2t, so combined with ai=ta_i = t we know ai1+ai+1=ta_{i-1} + a_{i+1} = t.
Now remove all entries equal to tt from a0,,ana_0, \dots, a_n, forming a new sequence A0,,ANA_0, \dots, A_N. Since
t1t \neq 1, we have A0=AN=1A_0 = A_N = 1 and N>0N > 0. We now prove: A0ANA_0 \sim A_N is a (t1)(t-1)-good sequence.
Let f:[0,N][0,n]f: [0, N] \to [0, n] denote the original position of AiA_i in the sequence a0,,ana_0, \dots, a_n. Since the original a0,,ana_0, \dots, a_n satisfies conditions (1)(2), and all tt's have been removed from it, A0,,ANA_0, \dots, A_N satisfies conditions (1)(2). It remains to prove (3).
Set 0iN10 \le i \le N-1. If f(i+1)=f(i)+1f(i+1) = f(i)+1, then (Ai,Ai+1)=(af(i),af(i)+1)=1(A_i, A_{i+1}) = (a_{f(i)}, a_{f(i)+1}) = 1. Otherwise, between af(i)a_{f(i)} and af(i+1)a_{f(i+1)} a tt has been removed. Since tt's are not adjacent, only one tt is removed there, that is, af(i)=Ai,af(i)+1=t,af(i)+2=Ai+1a_{f(i)} = A_i, a_{f(i)+1} = t, a_{f(i)+2} = A_{i+1}.
By what was proved above, Ai+Ai+1=t=af(i)+1A_i + A_{i+1} = t = a_{f(i)+1}, and since aa satisfies condition (3), (Ai,Ai+1)=(Ai,Ai+Ai+1)=(af(i),af(i)+1)=1(A_i, A_{i+1}) = (A_i, A_i + A_{i+1}) = (a_{f(i)}, a_{f(i)+1}) = 1. In summary, in either case (Ai,Ai+1)=1(A_i, A_{i+1}) = 1, which also means (Ai1,Ai)=(Ai,Ai+1)=11iN1(A_{i-1}, A_i) = (A_i, A_{i+1}) = 1 \quad \forall 1 \le i \le N-1.
Next we prove AiAi1+Ai+11iN1A_i | A_{i-1} + A_{i+1} \quad \forall 1 \le i \le N-1. Since aa satisfies condition (3), it suffices to prove af(i)1Ai1modAi,af(i)+1Ai+1modAia_{f(i)-1} \equiv A_{i-1} \mod A_i, a_{f(i)+1} \equiv A_{i+1} \mod A_i. If f(i+1)=f(i)+1f(i+1) = f(i)+1, this is clear. Otherwise, as discussed before, af(i)=Ai,af(i)+2=Ai+1a_{f(i)} = A_i, a_{f(i)+2} = A_{i+1} and Ai+Ai+1=af(i)+1A_i + A_{i+1} = a_{f(i)+1}, so af(i)+1=Ai+Ai+1Ai+1modAia_{f(i)+1} = A_i + A_{i+1} \equiv A_{i+1} \mod A_i.
So in either case, af(i)+1Ai+1modAia_{f(i)+1} \equiv A_{i+1} \mod A_i, and similarly af(i)1Ai1modAia_{f(i)-1} \equiv A_{i-1} \mod A_i, combined with the fact that aa satisfies condition (3), we know AiAi1+Ai+1A_i | A_{i-1} + A_{i+1}. Thus, we have proved that AA satisfies condition (3), and therefore AA is a (t1)(t-1)-good sequence. By the induction hypothesis, AA is unique and satisfies (*). Next we prove aa is unique. Note that given the choice of AA, it suffices to prove that the way of inserting the ϕ(t)\phi(t) copies of tt into A0,,ANA_0, \dots, A_N is unique. However, by what was proved above, two tt's cannot appear simultaneously between some AiA_i and Ai+1A_{i+1}, and if tt is between Ai,Ai+1A_i, A_{i+1}, then Ai+Ai+1=tA_i + A_{i+1} = t. Also (Ai,Ai+1)=1(A_i, A_{i+1}) = 1 and Ai,Ai+1tA_i, A_{i+1} \le t, combined with (*) we know that if tt can be inserted both between Ai,Ai+1A_i, A_{i+1} and between Aj,Aj+1A_j, A_{j+1} (iji \neq j), then AiAjA_i \neq A_j (otherwise Ai+1=Aj+1A_{i+1} = A_{j+1}, contradicting the uniqueness in (*)). Also (Ai,t)=1(A_i, t) = 1 and AitA_i \le t, so AiA_i can take at most ϕ(t)\phi(t) values, that is, there are at most ϕ(t)\phi(t) positions where tt can be inserted. Therefore the way of inserting tt is unique, that is, aa is unique. Also note that if Ai+Ai+1=tA_i + A_{i+1} = t, then a tt must be inserted between Ai,Ai+1A_i, A_{i+1} (otherwise there would not be enough positions).
Next we prove that a0,,ana_0, \dots, a_n satisfies condition (*). Let (a,b)=1,a+bt+1(a, b) = 1, a+b \ge t+1 and a,bta, b \le t. If a,bta, b \neq t, then by the fact that AA satisfies condition (*) and a+bta+b \neq t, it is easy to see that there exists a unique positive integer ii satisfying ai=a,ai+1=ba_i = a, a_{i+1} = b. If a=ta=t, then since (tb,b)=(t,b)=1,(tb)+b=tt(t-b, b) = (t, b) = 1, (t-b)+b=t \ge t and tb,bt1t-b, b \le t-1 (note that tt's are not adjacent, so btb \neq t), combined with the fact that AA satisfies condition (*), we know there exists a positive integer ii satisfying Ai=tb,Ai+1=bA_i = t-b, A_{i+1} = b. Since Ai+Ai+1=tA_i+A_{i+1} = t, by what was proved above, af(i)+1=t,af(i)+2=Ai+1a_{f(i)+1} = t, a_{f(i)+2} = A_{i+1}, and the existence part of (*) is proved. It remains only to prove uniqueness. If ai=t,ai+1=ba_i = t, a_{i+1} = b, then ai1=tba_{i-1} = t-b and there exists a unique II satisfying f(I)=i1,f(I+1)=i+1f(I) = i-1, f(I+1) = i+1. Hence AI=tb,AI+1=bA_I = t-b, A_{I+1} = b. Also since AA satisfies (*), there exists a unique II satisfying AI=tb,AI+1=bA_I = t-b, A_{I+1} = b, from which we know ii is also unique. Uniqueness is proved. Hence a0,,ana_0, \dots, a_n also satisfies (*).
By mathematical induction, the kk-good sequence is unique and satisfies (*).

Let the numerators in order be b0,,bnb_0, \dots, b_n, and the denominators in order be c0,,cnc_0, \dots, c_n. For convenience, we call b,cb, c respectively the kk-numerator sequence and kk-denominator sequence. We now prove: c0,,cnc_0, \dots, c_n is a kk-good sequence and
cibi1ci1bi=11inc_i b_{i-1} - c_{i-1} b_i = 1 \quad \forall 1 \le i \le n. This, combined with the uniqueness of the kk-good sequence, shows that the numerators b0,,bnb_0, \dots, b_n satisfy the conditions of the problem. Since b0=0,bn=1b_0 = 0, b_n = 1, we get that the minimum value of bnb0b_n - b_0 is 1.
It is easy to see that c0,,cnc_0, \dots, c_n satisfies conditions (1)(2). Next, we use mathematical induction on kk to prove that cc also satisfies (3) and cibi1ci1bi=11inc_i b_{i-1} - c_{i-1} b_i = 1 \quad \forall 1 \le i \le n.
First, when k=1k=1 this is obviously true. If it holds for k=t1k=t-1, then for k=tk=t, let B0,,BNB_0, \dots, B_N and C0,,CNC_0, \dots, C_N be respectively the (t1)(t-1)-numerator sequence and denominator sequence. Consider the fraction xt\frac{x}{t} in lowest terms in [0,1][0,1], and suppose it falls in the interval (BiCi,Bi+1Ci+1)(\frac{B_i}{C_i}, \frac{B_{i+1}}{C_{i+1}}). By the induction hypothesis, it is easy to see that it suffices to prove (t,Ci)=(t,Ci+1)=1,Cit(modCi+1),Ci+1t(modCi),tCi+Ci+1,xCiBit=Bi+1txCi+1=1(t, C_i) = (t, C_{i+1}) = 1, C_i \equiv t \pmod{C_{i+1}}, C_{i+1} \equiv t \pmod{C_i}, t \nmid C_i + C_{i+1}, xC_i - B_it = B_{i+1}t - xC_{i+1} = 1. ... (Δ\Delta) Since (x,t)=1(x,t)=1, let q<tq<t and qx1(modt),p=qxtqx \equiv 1 \pmod t, p = \lfloor \frac{qx}{t} \rfloor. Then pq<xt\frac{p}{q} < \frac{x}{t}. By the definition of B, C and q<tq<t, we know pqBiCi\frac{p}{q} \le \frac{B_i}{C_i}. Also xqpt=qxtqxt=qxxq-pt = qx-t\lfloor \frac{qx}{t} \rfloor = qx divided by tt has remainder 1. Similarly, xCiBit=CixxC_i - B_it = C_ix divided by tt has remainder (call it rr) (note that here we use the fact that BiCi\frac{B_i}{C_i} is the closest fraction to xt\frac{x}{t}). Since rqxrCix(modt)rqx \equiv r \equiv C_ix \pmod t, so Cirq(modt)C_i \equiv rq \pmod t. Also CitC_i \le t and rq>0rq > 0, so CirqC_i \le rq. From pqBiCi\frac{p}{q} \le \frac{B_i}{C_i} we know xtBiCixtpq\frac{x}{t} - \frac{B_i}{C_i} \le \frac{x}{t} - \frac{p}{q}. However xtpq=xqptqt=1qt\frac{x}{t} - \frac{p}{q} = \frac{xq-pt}{qt} = \frac{1}{qt} and xtBiCi=xCiBitCitrrqt\frac{x}{t} - \frac{B_i}{C_i} = \frac{xC_i - B_it}{C_it} \ge \frac{r}{rqt} (since CirqC_i \le rq) =1qt=xtpq= \frac{1}{qt} = \frac{x}{t} - \frac{p}{q}, so equality must hold, that is, Bi=p,Ci=qB_i = p, C_i = q. Hence (t,Ci)=(t,q)=1,xCi1(modt)(t, C_i) = (t, q) = 1, xC_i \equiv 1 \pmod t and xCiBit=xqpt=1xC_i - B_it = xq-pt = 1. Similarly we can prove (t,Ci+1)=1,xCi+11(modt)(t, C_{i+1}) = 1, xC_{i+1} \equiv -1 \pmod t and Bi+1txCi+1=1B_{i+1}t - xC_{i+1} = 1. Hence x(Ci+Ci+1)0(modt)x(C_i + C_{i+1}) \equiv 0 \pmod t, that is, tCi+Ci+1t \nmid C_i + C_{i+1}. (Δ\Delta) is fully proved.
In summary, by mathematical induction we know c=ac=a and cibi1ci1bi=11inc_i b_{i-1} - c_{i-1} b_i = 1 \quad \forall 1 \le i \le n, and the proof of the whole problem is complete.

Source: MathNet, licensed CC-BY-4.0. Statement translated into English from zh; metadata (topic, difficulty, ordering) added by this project.