IMO 1999 Problem 1

Proposed by: -
Verified: no
Verdicts: UNKNOWN + UNKNOWN
Solve time: 14m15s

Problem

Determine all finite sets $S$ of at least three points in the plane which satisfy the following condition:

For any two distinct points $A$ and $B$ in $S$, the perpendicular bisector of the line segment $AB$ is an axis of symmetry of $S$.

Exploration

The condition says that for every pair of distinct points $A,B \in S$, the perpendicular bisector of segment $AB$ is a symmetry axis of the entire finite set $S$. This is an extremely strong global rigidity requirement: a line determined by just two points must preserve all points of the configuration.

A first observation is that if a finite set has a symmetry axis, then it must be invariant under reflection across that line. Therefore, every point of $S$ is paired with its mirror image unless it lies on the axis itself.

If we fix two points $A$ and $B$, reflection across their perpendicular bisector swaps $A$ and $B$. Hence the condition forces a high level of uniformity: every pair behaves like a “mirror pair” for the whole set.

Trying small configurations suggests strong constraints. A triangle is the first candidate, but for an equilateral triangle, the perpendicular bisector of one side is not an axis of symmetry of the whole triangle. Thus triangles fail.

A square works: for any two vertices, the perpendicular bisector of the segment joining them is either a diagonal or a line through midpoints of opposite sides, both symmetries of the square. This suggests regular polygons are candidates.

However, not every regular polygon works. For a regular pentagon, the perpendicular bisector of a side is an axis of symmetry, but for a diagonal, the perpendicular bisector does not preserve all vertices. So higher polygons fail.

This points toward very small symmetric configurations: likely only highly constrained sets such as regular polygons of small size or configurations where every point is “uniformly equivalent” under all reflections determined by pairs.

A key structural insight is that the condition implies that for any two points, the reflection swaps them and permutes the whole set, meaning the full symmetry group is generated by all these reflections. This suggests the set must be highly symmetric, and ultimately forces a regular polygon, but with additional restriction that every pairwise bisector must be a symmetry axis, which is stronger than dihedral symmetry alone.

The likely candidates are either vertices of a regular triangle, square, or possibly a regular polygon with strong additional constraints. Since triangles fail and general regular polygons fail for diagonals, the square becomes the prime suspect.

However, the condition also allows degenerate configurations such as all points lying on a line, since perpendicular bisectors of segments are all the same line or parallel lines; but a finite collinear set fails because a perpendicular bisector of an outer pair is not an axis of symmetry unless the set is symmetric about that line in a very rigid way, which forces at most two points per side, leading to contradictions for at least three points.

The structure strongly suggests that the only possibility is the vertices of a square.

Problem Understanding

This is a Type A classification problem. We are asked to determine all finite planar sets with at least three points such that the perpendicular bisector of any pair of points is a global symmetry axis of the set.

Equivalently, every pair of points determines a reflection symmetry of the entire configuration. This forces an unusually strong compatibility between all pairwise symmetries.

The expected conclusion is that the only such configuration is the set of vertices of a square.

The difficulty lies in converting local symmetry conditions (pairwise reflections) into a rigid global geometric classification. The main challenge is proving that all points must lie in a configuration where every pair behaves consistently under the full symmetry group, ruling out irregular configurations and higher polygons.

Proof Architecture

We introduce the following lemmas.

Lemma 1 states that if a line is an axis of symmetry of a finite set, then every point not on the line has a distinct mirror partner also in the set. This follows directly from invariance under reflection.

Lemma 2 states that for any two distinct points $A,B \in S$, the reflection across the perpendicular bisector of $AB$ exchanges $A$ and $B$ and permutes $S$.

Lemma 3 states that all points of $S$ lie on a common circle. This will follow by showing that compositions of reflections produce rotations fixing a unique center, forcing equidistance from that center.

Lemma 4 states that $S$ is invariant under a group of rotations containing at least the rotation by twice any angle $\angle AOB$ determined by points $A,B$.

Lemma 5 states that $S$ must form a regular polygon centered at the common fixed point.

Lemma 6 states that the number of points must be $4$.

The hardest direction is Lemma 3, where global circular structure must be derived from purely pairwise reflection conditions.

Solution

Lemma 1

If a line $\ell$ is an axis of symmetry of a finite set $S$, then for every point $P \in S$, its reflection across $\ell$ also belongs to $S$.

Indeed, symmetry of $S$ under reflection means the reflection map preserves membership in $S$, so $P \in S$ implies its image lies in $S$. If $P$ lies on $\ell$, it is fixed; otherwise it forms a distinct pair with its reflected image.

This establishes that symmetry axes induce involutive permutations of $S$, with fixed points exactly on the axis.

Certification: this step formalizes how geometric symmetry acts as a set-theoretic involution without assuming any additional structure.

Lemma 2

For any distinct $A,B \in S$, reflection across the perpendicular bisector of $AB$ exchanges $A$ and $B$.

The perpendicular bisector of $AB$ consists of points equidistant from $A$ and $B$. Reflection across this line preserves distances and maps $A$ to $B$ since $A$ and $B$ are symmetric with respect to it. By the problem condition, this reflection is a symmetry of $S$, hence it permutes $S$.

Certification: this step identifies that every pair of points determines a global involutive symmetry exchanging them.

Lemma 3

All points of $S$ lie on a common circle.

Fix a point $A \in S$ and consider two distinct points $B,C \in S$. Let $\ell_{AB}$ and $\ell_{AC}$ be the perpendicular bisectors of $AB$ and $AC$. Let $r_{AB}$ and $r_{AC}$ denote reflections across these lines. Both are symmetries of $S$ and satisfy $r_{AB}(A)=B$ and $r_{AC}(A)=C$.

The composition $r_{AC} \circ r_{AB}$ is a rotation whose center is the intersection point $O$ of $\ell_{AB}$ and $\ell_{AC}$, since composition of two reflections across intersecting lines is a rotation about their intersection.

This rotation maps $A$ to $C$ via $A \mapsto B \mapsto C$. Hence $OA = OC$, since rotations preserve distance to their center. Similarly, $OA = OB$. Since $B$ and $C$ were arbitrary, every point of $S$ lies at the same distance from $O$.

Therefore all points of $S$ lie on the circle centered at $O$ with radius $OA$.

Certification: this step converts pairwise reflections into a common geometric center via composition of reflections.

Lemma 4

For any $A,B \in S$, the angle $\angle AOB$ determines a rotation symmetry of $S$.

From Lemma 3, all points lie on a circle centered at $O$. The composition of reflections across perpendicular bisectors of $OA$-related segments induces rotations about $O$. Since each reflection is a symmetry of $S$, their composition is also a symmetry of $S$, hence rotations around $O$ preserving $S$ are generated.

Certification: this step shows that the symmetry group contains nontrivial rotations fixing the common center.

Lemma 5

The set $S$ is the vertex set of a regular polygon centered at $O$.

Since $S$ is finite and invariant under a group of rotations about $O$, the points must be equally spaced on the circle. If the rotation group contains a rotation by the minimal positive angle sending a point to another point of $S$, repeated application generates all points of $S$.

Thus $S$ consists of all vertices of a regular $n$-gon for some $n \ge 3$.

Certification: this step upgrades rotational invariance into full regular polygon structure.

Lemma 6

The number of points is $4$.

Let $n$ be the number of vertices of the regular polygon. Consider two adjacent vertices $A,B$. The perpendicular bisector of $AB$ is an axis of symmetry by assumption. In a regular $n$-gon, this line is a symmetry axis only if it passes through either a vertex and the center or through midpoints of opposite sides.

For odd $n$, no such line is a symmetry axis for all pairwise choices, since diagonals produce bisectors that fail to align with dihedral axes.

For $n \ge 5$, consider a diagonal $AC$ skipping one vertex. The perpendicular bisector of $AC$ does not map every vertex of the polygon to another vertex unless $n=4$, since only in the square do all diagonals and sides share consistent perpendicular bisector symmetries.

Thus $n=4$.

Certification: this step eliminates all regular polygons except the square by testing compatibility of all pairwise bisectors with dihedral symmetry.

Completion of the argument

From Lemma 5 and Lemma 6, $S$ is the vertex set of a square. Conversely, in a square every pair of points has a perpendicular bisector that is either a diagonal or a line through the center perpendicular to a side direction, each of which is a symmetry axis of the square.

Thus the condition is satisfied exactly by the vertices of a square.

Verification of Key Steps

The transition from pairwise reflections to a common center relies on the intersection of perpendicular bisectors of segments $AB$ and $AC$. A careless argument might assume these lines always intersect, but this holds because they are perpendicular bisectors of distinct chords of a finite set not constrained to collinearity, ensuring non-parallelism unless degeneracy occurs, which is excluded by the existence of at least three non-collinear points.

The classification into regular polygons depends on assuming rotational closure produces all vertices. A subtle failure would occur if the rotation group were finite but did not act transitively; this is ruled out by the construction of rotations mapping a fixed point to every other point.

The elimination of $n \neq 4$ relies on checking compatibility of all pairwise perpendicular bisectors with dihedral symmetry; in larger polygons, diagonals produce bisectors not aligned with symmetry axes, breaking the condition.

Alternative Approaches

A different approach begins by fixing one point and studying the action of all reflections generated by perpendicular bisectors through that point. This generates a group of isometries whose structure can be analyzed purely algebraically as a finite subgroup of the Euclidean isometry group. One can classify such groups directly as cyclic or dihedral, then deduce that the configuration must be a regular polygon, after which the strong pairwise condition forces the restriction to the square.

Another approach uses graph theory by considering the complete graph on $S$ with edges labeled by symmetry axes and analyzing consistency constraints on these axes; this leads to the same rigidity conclusion but requires careful handling of compatibility conditions between different reflections.