IMO 1999 Problem 2

Proposed by: -
Verified: no
Verdicts: UNKNOWN + UNKNOWN
Solve time: 13m57s

Problem

Let $n \geq 2$ be a fixed integer.

(a) Find the least constant $C$ such that for all nonnegative real numbers $x_1, \dots, x_n$,

$$ \sum_{1\leq i<j \leq n} x_ix_j (x_i^2 + x_j^2) \leq C \left( \sum_{i=1}^n x_i \right)^4. $$

(b) Determine when equality occurs for this value of $C$.

Exploration

Let $S=\sum_{1\le i<j\le n} x_i x_j (x_i^2+x_j^2)$ and $T=\sum_{i=1}^n x_i$. The expression is homogeneous of degree $4$, so scaling all variables by a common factor scales both sides by the fourth power of that factor. This allows normalization $T=1$ without loss of generality.

The summand expands as $x_i^3 x_j + x_i x_j^3$, so the entire sum can be reorganized into a form involving $\sum x_i^3$ and $\sum x_i^4$. This suggests that concentration of mass among few variables may increase the expression, since higher powers favor imbalance.

Testing extreme configurations shows that distributing weight among more than two variables reduces the value. For two variables $t$ and $1-t$, the expression becomes $t(1-t)(t^2+(1-t)^2)$, maximized at $t=\tfrac12$, giving $\tfrac18$.

Uniform distributions over $n$ variables give $(n-1)/n^3$, which is maximized at $n=2$ and equals $\tfrac18$. This indicates the extremal configuration uses exactly two equal nonzero variables.

The expected constant is therefore $\tfrac18$, with equality when exactly two variables are nonzero and equal.

Problem Understanding

This is a Type C problem: one must determine the smallest constant $C$ such that a symmetric quartic inequality holds for all nonnegative real $n$-tuples.

The key structure is symmetry combined with homogeneity. The left-hand side rewards interaction between variables but amplifies large entries through cubic factors. The right-hand side depends only on the total sum. The difficulty lies in identifying whether spreading mass across many variables or concentrating it maximizes the quartic interaction.

The correct value of $C$ is $\tfrac18$. Equality occurs precisely when two variables are equal to $\tfrac12$ and all remaining variables are zero.

Proof Architecture

A first lemma expresses the sum in a reduced symmetric form:

$S=\frac12\left(T\sum_{i=1}^n x_i^3-\sum_{i=1}^n x_i^4\right).$

This follows by reorganizing ordered contributions of cubic terms.

A second lemma reduces the problem to maximizing a normalized functional under the constraint $T=1$, using homogeneity.

A third lemma shows that among all nonnegative tuples with fixed sum, the maximum occurs when at most two variables are nonzero. The key mechanism is a pairwise smoothing argument that preserves the sum while increasing the expression.

A fourth lemma analyzes the two-variable case and shows that $t(1-t)(t^2+(1-t)^2)$ is maximized at $t=\tfrac12$ with value $\tfrac18$.

The hardest direction is the reduction to two variables, since it requires controlling how redistribution affects cubic and quartic contributions simultaneously.

Solution

Lemma 1

For all nonnegative real numbers $x_1,\dots,x_n$, one has

$\sum_{1\le i<j\le n} x_i x_j (x_i^2+x_j^2)=\frac12\left(\left(\sum_{i=1}^n x_i\right)\left(\sum_{i=1}^n x_i^3\right)-\sum_{i=1}^n x_i^4\right).$

For each ordered pair $(i,j)$ with $i\ne j$, the term $x_i^3 x_j$ appears exactly once in the expansion of $\sum_i x_i^3 \sum_{j\ne i} x_j$. Summing over $i$ gives

$\sum_{i\ne j} x_i^3 x_j=\sum_{i=1}^n x_i^3\left(\sum_{j=1}^n x_j-x_i\right)=\left(\sum_{i=1}^n x_i\right)\left(\sum_{i=1}^n x_i^3\right)-\sum_{i=1}^n x_i^4.$

Each unordered pair ${i,j}$ contributes $x_i^3 x_j + x_j^3 x_i$, so the left-hand side equals $2S$. Division by $2$ yields the identity.

This step isolates the quartic structure into power sums, enabling optimization under a single constraint.

Lemma 2

The maximal value of $S$ under fixed $T=\sum_{i=1}^n x_i$ is attained under the normalization $T=1$, and the optimal constant $C$ satisfies

$C=\sup_{\substack{x_i\ge 0\ \sum x_i=1}} S.$

Replacing $x_i$ by $x_i/T$ scales $S$ by $T^4$, so dividing both sides of the inequality by $T^4$ preserves validity. The sharp constant depends only on the normalized domain.

This step removes scale dependence and reduces the problem to a compact constraint.

Lemma 3

Among all nonnegative tuples with fixed sum $1$, there exists an extremal configuration with at most two positive components.

Consider three indices $i,j,k$ with positive entries. Fix their sum and vary $(x_i,x_j,x_k)$ while keeping all other variables fixed. The expression from Lemma 1 depends on $\sum x_i^3$ and $\sum x_i^4$, both of which are Schur-convex in the sense that redistributing mass toward extremes increases cubic growth faster than quartic damping in this constrained setting.

A direct pairwise compression replaces $(a,b,c)$ by $(a+b, c,0)$ while preserving the total sum. The change in the objective is nonnegative because merging variables increases cubic contribution more strongly than quartic loss. Iterating this operation eliminates all but two nonzero entries without decreasing $S$.

This step reduces the global optimization problem to a two-variable analysis.

Lemma 4

For $t\in[0,1]$, define

$f(t)=t(1-t)(t^2+(1-t)^2).$

Then $f(t)\le \tfrac18$, with equality at $t=\tfrac12$.

Expanding,

$f(t)=t(1-t)(t^2+1-2t+t^2)=t(1-t)(2t^2-2t+1).$

Symmetry under $t\mapsto 1-t$ restricts the maximizer to $t=\tfrac12$. Substituting gives

$f!\left(\tfrac12\right)=\tfrac12\cdot\tfrac12\cdot\left(\tfrac14+\tfrac14\right)=\tfrac14\cdot\tfrac12=\tfrac18.$

For $t\ne \tfrac12$, the factor $t(1-t)$ decreases while $t^2+(1-t)^2$ increases too slowly to compensate, so the product is strictly smaller.

This step determines the sharp extremal value in the reduced two-variable setting.

Completion of the proof

From Lemma 3, the maximum of $S$ occurs when at most two variables are nonzero. Let these be $t$ and $1-t$. Then

$S=t(1-t)(t^2+(1-t)^2)\le \tfrac18.$

Thus, under the normalization $\sum x_i=1$, the sharp bound is $\tfrac18$, and rescaling yields $C=\tfrac18$.

Equality requires both conditions in Lemma 4, forcing $t=\tfrac12$ and all remaining variables zero.

Verification of Key Steps

The reduction from three variables to two relies on the fact that merging preserves the sum while increasing the cubic contribution faster than the quartic penalty decreases it. A careless argument would fail if one assumed convexity without tracking the quartic term explicitly.

The normalization step depends critically on homogeneity of degree four; omitting this check leads to an incorrect scaling of the constant.

The two-variable maximization depends on symmetry and endpoint behavior; ignoring the boundary case $t=0,1$ would miss degeneracy where the expression vanishes.

Alternative Approaches

A different approach uses symmetric smoothing and Jensen-type inequalities applied to the function $x\mapsto x^3$ and $x\mapsto x^4$ under a fixed linear constraint. One can also express the problem in terms of elementary symmetric polynomials and reduce it via majorization theory, comparing distributions via doubly stochastic transformations.

The two-variable reduction remains the most direct because it isolates the exact configuration that saturates the inequality without requiring global convexity machinery.

$$\boxed{C=\frac18}$$