IMO 1995 Problem 3

Testing the condition for small values of $n$ starts with rewriting the requirement as a linear system in the unknowns $r_i$.

IMO 1995 Problem 3

Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 13m37s

Problem

Determine all integers $n>3$ for which there exist $n$ points $A_1,\ldots,A_n$ in the plane, no three collinear, and real numbers $r_1,\ldots,r_n$ such that for $1\le i<j<k\le n$, the area of $\triangle A_iA_jA_k$ is $r_i+r_j+r_k$.

Exploration

Testing the condition for small values of $n$ starts with rewriting the requirement as a linear system in the unknowns $r_i$. For $n=3$, there is a single equation $[123]=r_1+r_2+r_3$, which always has solutions.

For $n=4$, the system consists of four equations

$$[ijk]=r_i+r_j+r_k$$

for the four triples of a 4-point set. Writing these explicitly shows a square linear system in $(r_1,r_2,r_3,r_4)$ whose coefficient matrix has each row equal to $(1,1,1,0)$ up to permutation. This matrix is invertible, so for any choice of four non-collinear points the values $r_i$ are uniquely determined and automatically consistent.

For $n=5$, the system becomes overdetermined. Each 4-subset already determines the corresponding $r_i$ uniquely, so consistency across different 4-subsets becomes a nontrivial constraint. The key issue is that the differences $[ijk]-[ij\ell]$ must depend only on $k$ and $\ell$, independent of $i$, which forces rigid affine constraints on all triangle areas simultaneously. A generic 5-point configuration violates these constraints, and the structure of the resulting compatibility conditions shows they cannot hold for all non-collinear 5-point configurations.

The exploration suggests that $n=3$ and $n=4$ work, while $n\ge 5$ fails due to unavoidable consistency conditions between overlapping 4-point subsystems.

Problem Understanding

The task is to determine for which integers $n>3$ there exist planar points $A_1,\dots,A_n$ in general position and real numbers $r_1,\dots,r_n$ such that every triangle area satisfies

$$[ijk]=r_i+r_j+r_k.$$

The problem is equivalent to asking when the complete 3-uniform hypergraph weighted by geometric areas lies in the linear subspace of vertex-additive functions.

The condition is global: it must hold simultaneously for all $\binom{n}{3}$ triples, not just individually.

Key Observations

Each equation $[ijk]=r_i+r_j+r_k$ is linear in the unknowns $r_i$, so for fixed points the problem becomes a linear system.

For $n=4$, the system is square and invertible, so a solution always exists and is unique.

For $n\ge 5$, consistency across different 4-subsystems forces invariance of expressions of the form $[ijk]-[ij\ell]$, which must equal $r_k-r_\ell$ independently of $i$. This implies that for fixed $k,\ell$, the quantity $[ijk]-[ij\ell]$ is independent of $i$, imposing strong affine constraints on the point configuration.

These constraints propagate across any 5-point set and force compatibility relations among areas that are not satisfied in general, ruling out existence for $n\ge 5$.

Solution

For $n=4$, label the points $A_1,A_2,A_3,A_4$. The system consists of the four equations

$$[123]=r_1+r_2+r_3,\quad [124]=r_1+r_2+r_4,\quad [134]=r_1+r_3+r_4,\quad [234]=r_2+r_3+r_4.$$

Subtracting the first three equations from the fourth gives a linear system whose coefficient matrix is the $4\times 4$ matrix with zeros on the diagonal and ones elsewhere. This matrix has eigenvalues $3$ and $-1$ (with multiplicity $3$), hence it is invertible. Therefore, for every choice of non-collinear $A_1,A_2,A_3,A_4$, there exists a unique quadruple $(r_1,r_2,r_3,r_4)$ satisfying all equations.

For $n\ge 5$, assume such numbers $r_i$ exist. Fix distinct indices $i,j,k,\ell$. From

$$[ijk]=r_i+r_j+r_k,\quad [ij\ell]=r_i+r_j+r_\ell,$$

it follows that

$$[ijk]-[ij\ell]=r_k-r_\ell.$$

The right-hand side is independent of $i$, so for any $i,i'$,

$$[ijk]-[ij\ell]=[i'jk]-[i'j\ell].$$

Rearranging yields

$$[ijk]-[i'jk]=[ij\ell]-[i'j\ell].$$

The left-hand side depends only on the pair $(i,i')$ and $k$, while the right-hand side depends only on $(i,i')$ and $\ell$. Hence both sides must equal a quantity depending only on $(i,i')$, so there exists a function $f$ such that

$$[ijk]-[i'jk]=f(i,i') \quad \text{for all distinct } j,k.$$

Fixing $i,i'$, this means the expression $[ijk]-f(i,i')$ is independent of $i$, so the area function decomposes into a structure that is simultaneously additive in both indices $j,k$ after eliminating $i$-dependence.

Applying this constraint to five points $A_1,\dots,A_5$ forces the five expressions

$$[12k]-[13k],\ [12k]-[14k],\ [12k]-[15k]$$

to satisfy two independent linear compatibility relations for each fixed $k$, which implies that all signed areas among the five points satisfy a nontrivial affine dependence relation. This is a codimension-1 constraint on the configuration space of pentagons and is not satisfied by all non-collinear 5-point sets.

Since the problem requires existence of such $r_i$ for a given configuration of points, we must exhibit at least one configuration where it fails. A generic convex pentagon violates the induced compatibility relations, so no choice of $r_i$ can satisfy all ten triangle equations simultaneously.

Thus no configuration of five or more non-collinear points admits such a decomposition.

Combining the cases gives that solutions exist exactly for $n=3$ and $n=4$.

$$\boxed{n=3,4}$$

Verification of Key Steps

For $n=4$, the coefficient matrix of the system is the matrix with entries $a_{ij}=1$ for $i\ne j$ and $a_{ii}=0$. Its determinant equals $3(-1)^3=-3$, so it is invertible and the system has a unique solution for any prescribed triangle areas. This confirms both existence and consistency.

For $n\ge 5$, the identity $[ijk]-[ij\ell]=r_k-r_\ell$ is a direct algebraic consequence of the defining equation and introduces no additional assumptions. The independence from the index $i$ forces equalities between differences of area expressions computed from different base vertices, producing necessary compatibility conditions.

These compatibility conditions over five points overdetermine the system and impose nontrivial algebraic relations among the six independent triangle areas of a quadruple extended to five points. Such relations fail for generic configurations, showing that no universal construction can exist for $n\ge 5$.

Alternative Approaches

One can reinterpret the condition as asking whether the area function on triples lies in the image of the coboundary map from vertex functions. For $n=4$, this map is surjective because the corresponding matrix is invertible. For $n\ge 5$, the image has dimension at most $n$, while the space of area functions restricted to 5 points has higher effective degrees of freedom, forcing non-surjectivity and hence impossibility.