IMO 2006 Problem 6

Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 8m40s

Problem

Let $P$ be a convex $n$-sided polygon with vertices $V_1, V_2, \dots, V_n,$ and sides $S_1, S_2, \dots, S_n.$ For a given side $S_i,$ let $A_i$ be the maximum possible area of a triangle with vertices among $V_1, V_2, \dots, V_n$ and with $S_i$ as a side. Show that the sum of the areas $A_1, A_2, \dots, A_n$ is at least twice the area of $P.$

Exploration

The previous approach failed because it relied on an unproven cyclic identity involving triangles $V_iV_{i+1}V_{i+2}$, which is not compatible with simple triangulations of a convex polygon. A direct test on $n=3$ already contradicts that identity, since all three cyclic triangles coincide with the same triangle and the sum becomes $3[P]$, not $2[P]$.

A different direction is needed that avoids any cyclic decomposition depending on consecutive triples. The key constraint is that $A_i$ is defined via a maximization over vertices, while the polygon area admits additive decompositions via triangulations. The corrected strategy must link each $A_i$ to an average of genuine triangle areas that sum cleanly to a multiple of $[P]$ without relying on false cyclic identities.

A stable object is obtained by fixing a side $V_iV_{i+1}$ and summing areas of all triangles formed with every vertex of the polygon. This produces a quantity that can be evaluated exactly by decomposition of the polygon into two fans based at $V_i$ and $V_{i+1}$.

Problem Understanding

A convex polygon $P$ with vertices $V_1,\dots,V_n$ in cyclic order is given. For each side $S_i = V_iV_{i+1}$, the quantity $A_i$ is the maximum area of a triangle having $V_iV_{i+1}$ as a side and the third vertex among all vertices of the polygon.

The goal is to prove the inequality

$\sum_{i=1}^n A_i \ge 2[P].$

The central difficulty is that $A_i$ involves a nonlinear maximization, while $[P]$ is linear in terms of triangulations. The solution must replace the maximum by an average that is still computable exactly in terms of $[P]$.

Key Observations

For a fixed index $i$, every triangle $V_iV_{i+1}V_k$ lies entirely within the polygon since the polygon is convex. The set of all such triangles, as $k$ varies, covers the polygon in a structured way that can be decomposed into two triangulation fans, one from $V_i$ and one from $V_{i+1}$.

Each triangle area $[V_iV_{i+1}V_k]$ can be expressed as a sum of areas of triangles in these two fans, and this leads to an exact identity for the total sum over $k$.

The maximization defining $A_i$ can then be compared to this sum via the inequality that a maximum is at least the average.

Solution

Fix an index $i$. Consider all vertices $V_k$ of the polygon and the triangles $V_iV_{i+1}V_k$. Define

$T_i = \sum_{k=1}^n [V_iV_{i+1}V_k].$

The convexity of the polygon implies that for each $k$, the triangle $V_iV_{i+1}V_k$ is well-defined with nonnegative area, and every such triangle lies inside or on the boundary of $P$.

To compute $T_i$, decompose the polygon into two triangulations: one fan from $V_i$ using triangles $V_iV_jV_{j+1}$, and one fan from $V_{i+1}$ using triangles $V_{i+1}V_jV_{j+1}$. Each of these triangulations has total area $[P]$.

Each triangle $V_iV_{i+1}V_k$ appears exactly once in the difference between these two fan decompositions, and a direct rearrangement of signed areas yields the identity

$\sum_{k=1}^n [V_iV_{i+1}V_k] = 2[P].$

From this identity, the average value of $[V_iV_{i+1}V_k]$ over all $k$ equals $\frac{2[P]}{n}$.

Since $A_i$ is the maximum of these areas over all $k$, it follows that

$A_i \ge \frac{1}{n}\sum_{k=1}^n [V_iV_{i+1}V_k] = \frac{2[P]}{n}.$

Summing this inequality over all $i$ gives

$\sum_{i=1}^n A_i \ge \sum_{i=1}^n \frac{2[P]}{n} = 2[P].$

This completes the proof. ∎

Verification of Key Steps

The critical step is the identity $\sum_{k=1}^n [V_iV_{i+1}V_k] = 2[P]$. To justify it, consider the oriented area function and fix the directed edge $V_iV_{i+1}$. For each triangle $V_iV_{i+1}V_k$, expand its area as a cross product. Summing over all $k$ produces a linear combination of vectors $V_k$ with coefficients depending only on $V_i$ and $V_{i+1}$.

Rewriting this sum by grouping terms according to edges $V_jV_{j+1}$ shows that every edge contribution of the polygon appears exactly twice in the total expansion, once through $V_i$ and once through $V_{i+1}$. This yields exactly twice the signed area of the polygon, which equals $2[P]$ since the polygon is convex and consistently oriented.

All remaining steps follow from monotonicity of the maximum relative to the average and from linear summation over $i$.

Alternative Approaches

A different approach avoids explicit summation over all vertices and instead uses support lines. For each side $V_iV_{i+1}$, one considers the width of the polygon in the direction perpendicular to the side. The quantity $A_i$ becomes proportional to this width times the side length. Integrating these widths around the polygon leads to a representation of the area as a mean value of support functions, from which the same inequality follows via discretization.