IMO 1971 · Problem 1

Prove that the following assertion is true for \(n=3\) and \(n=5\), and that it is false for every other natural number \(n>2\):

If \(a_1,a_2,\ldots,a_n\) are arbitrary real numbers, then \[ (a_1-a_2)(a_1-a_3)\cdots(a_1-a_n) +(a_2-a_1)(a_2-a_3)\cdots(a_2-a_n) +\cdots+(a_n-a_1)(a_n-a_2)\cdots(a_n-a_{n-1})\geq 0 \]

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

Solution

Write \[ S_n=\sum_{i=1}^{n}\prod_{\substack{1\leq j\leq n\\j\ne i}}(a_i-a_j). \]

The case \(n=3\). Expanding gives \[ S_3=a_1^2+a_2^2+a_3^2-a_1a_2-a_1a_3-a_2a_3 =\frac12\bigl((a_1-a_2)^2+(a_1-a_3)^2+(a_2-a_3)^2\bigr)\geq0. \]

The case \(n=5\). Since \(S_5\) is unchanged by permuting the numbers, assume \[ a_1\leq a_2\leq a_3\leq a_4\leq a_5, \] and put \(P_i=\prod_{j\ne i}(a_i-a_j)\). We have \[ \begin{aligned} P_1+P_2 =(a_2-a_1)\bigl[& (a_3-a_1)(a_4-a_1)(a_5-a_1)\\ &-(a_3-a_2)(a_4-a_2)(a_5-a_2)\bigr]\geq0. \end{aligned} \] Indeed, \(a_2-a_1\geq0\), and each factor in the first product inside the brackets is at least the corresponding nonnegative factor in the second product.

Likewise, \[ \begin{aligned} P_4+P_5 =(a_5-a_4)\bigl[& (a_5-a_1)(a_5-a_2)(a_5-a_3)\\ &-(a_4-a_1)(a_4-a_2)(a_4-a_3)\bigr]\geq0 \end{aligned} \] by the same factorwise comparison. Finally, \[ P_3=(a_3-a_1)(a_3-a_2)(a_4-a_3)(a_5-a_3)\geq0. \] Thus \[ S_5=(P_1+P_2)+P_3+(P_4+P_5)\geq0. \] These arguments allow equality among the \(a_i\), so they also cover all repeated-value cases.

Every even \(n>2\). Choose \[ a_1=-1,\qquad a_2=\cdots=a_n=0. \] Every summand except the first contains a zero factor, because each zero has another zero among the other entries. Consequently, \[ S_n=(-1)^{n-1}=-1<0. \]

Every odd \(n\geq7\). Choose \[ a_1=a_2=a_3=-1,\qquad a_4=0,\qquad a_5=\cdots=a_n=1. \] Both \(-1\) and \(1\) occur at least three times. Hence every summand belonging to one of these entries contains a zero factor. Only the summand belonging to \(a_4=0\) remains, giving \[ S_n=(0-(-1))^3(0-1)^{n-4}=(-1)^{n-4}=-1<0. \]

These cases exhaust all natural numbers \(n>2\), proving the assertion.

The solution was checked against John Scholes' IMO solution archive (Kalva).