IMO 2018 · Problem 2

Find all integers \(n \ge 3\) for which there exist real numbers \(a_1,a_2,\ldots,a_{n+2}\), such that \(a_{n+1}=a_1\) and \(a_{n+2}=a_2\), and \[ a_i a_{i+1}+1=a_{i+2} \] for \(i=1,2,\ldots,n\).

Official text of the problem. Official problem paper (PDF)

Answer

All positive multiples of \(3\): \[ \boxed{n=3,6,9,\ldots}. \]

Solution

Extend the sequence periodically by \(a_{i+n}=a_i\) for all integers \(i\). The given closing equalities ensure that \[ a_i a_{i+1}+1=a_{i+2} \] then holds for every integer \(i\).

First, two consecutive terms cannot both be positive. Otherwise the recurrence makes every subsequent term positive, so periodicity makes every term positive. The recurrence would then give \(a_i>1\) for every \(i\), and consequently \[ a_{i+2}=a_i a_{i+1}+1>a_{i+1} \] for every \(i\), contradicting periodicity.

Also, no term can be zero: if \(a_i=0\), then \[ a_{i+1}=a_{i-1}a_i+1=1, \qquad a_{i+2}=a_i a_{i+1}+1=1, \] giving two consecutive positive terms.

We next prove that every positive term is greater than \(1\). There is at least one positive term, since otherwise all terms would be negative, whereas \(a_3=a_1a_2+1>1\). Let \(p=a_j\) be the smallest positive value among the terms. Its two neighbours are negative. Write \(a_{j-1}=-b\), where \(b>0\). Since \[ a_{j+1}=1-bp<0, \] we have \(bp>1\).

If \(p=1\), the equality \(a_{j-2}a_{j-1}+1=p\) forces a zero term, which is impossible. If \(0<p<1\), the same equality gives \[ a_{j-2}=\frac{1-p}{b}, \qquad 0<a_{j-2}<\frac1b<p, \] contradicting the minimality of \(p\). Thus \(p>1\), as claimed.

Now start at any positive term \(a_j\). We have \(a_{j+1}<0\), and hence \[ a_{j+2}=a_j a_{j+1}+1<1. \] Since zero is impossible and every positive term exceeds \(1\), it follows that \(a_{j+2}<0\). Therefore \[ a_{j+3}=a_{j+1}a_{j+2}+1>1. \] Repeating this argument shows that the signs, starting at \(a_j\), are \[ +,-,-,+,-,-,+,-,-,\ldots. \] But \(a_{j+n}=a_j>0\), so necessarily \(3\mid n\).

Conversely, whenever \(3\mid n\), take the sequence obtained by repeating the block \[ (-1,-1,2). \] The identities \[ (-1)(-1)+1=2, \qquad (-1)\cdot2+1=2\cdot(-1)+1=-1 \] verify every recurrence equation. Since \(n\) is divisible by \(3\), the closing equalities \(a_{n+1}=a_1\) and \(a_{n+2}=a_2\) also hold. This proves the answer.

The solution was checked against the official IMO shortlist solutions.