The 3 of the tilings, the 5 of the Platonic solids, the 12 of the fullerenes, the 0 of the torus and the orders 1, 2, 3, 4 and 6 of crystallography do not constitute a collection of privileged numbers. They appear as the results of different constraints that reduce the space of possibilities until only certain compatible configurations remain. Their value becomes intelligible only once the mechanism producing it has been identified. The topological development that follows is restricted to closed surfaces without boundary. The case of open structures, which requires additional boundary terms, is taken up in the extensions.
1. One local equation: from the plane to convex closure
The 3 of the regular tilings and the 5 of the Platonic solids arise from the same local problem of assembling angles. If regular polygons with p sides meet q at a time around a vertex, the sum of the angles gathered at that point is:
Sum of angles = q × ((p − 2) × 180° / p)
The regime of equality: flatness. When this sum equals exactly 360°, there is no local angular deficit. The network can remain planar. The corresponding integer equation has only three regular solutions:
(p, q) = (3, 6), (4, 4), (6, 3)
They give the only three regular tilings of the Euclidean plane by a single type of polygon: six equilateral triangles, four squares or three regular hexagons around each vertex. The number 3 does not designate an abstract preference. It counts the three admissible solutions of a precise geometric equality.
The regime of deficit: convex closure. When the sum of the angles remains strictly below 360°, an angular deficit appears. The surface can fold around the vertex and take part in closing a convex polyhedron. The condition becomes:
1/p + 1/q > 1/2
With p ≥ 3 and q ≥ 3, this inequality leaves only five integer pairs: (3,3), (4,3), (3,4), (5,3) and (3,5). They correspond exactly to the tetrahedron, the cube, the octahedron, the dodecahedron and the icosahedron. The 3 of the tilings and the 5 of the Platonic solids are not two independent regularities. They represent two regimes of the same mechanism: angular equality maintains the plane, while angular deficit permits convex closure.
2. From local to global: the topological balance
The 12 of the fullerenes and the 0 of the torus mark the passage from a local constraint of assembly to a global constraint of topology. For a closed trivalent network, that is, a network in which three edges meet at every vertex, Euler's formula leads to the general combinatorial balance:
Σₙ (6 − n) fₙ = 6χ
The value on the right depends on the Euler characteristic χ of the surface. Topology imposes neither a particular geometry nor a precise metric form. It imposes a global invariant that the combinatorics of the network must satisfy.
Spherical topology: a balance of 12. For a surface of spherical topology, χ = 2. The balance becomes:
Σₙ (6 − n) fₙ = 12
In a network composed solely of pentagons and hexagons, each pentagon contributes +1 and each hexagon 0. Exactly twelve pentagons are required, whatever the number of hexagons. The 12 is not a primary property. It results from the conjunction of spherical topology, the trivalence of the network and the type of faces allowed.
Toroidal topology: a null balance. For a torus, χ = 0. The same formula gives:
Σₙ (6 − n) fₙ = 0
An exclusively hexagonal network satisfies this balance, since the hexagon has a null combinatorial contribution. Pentagons and heptagons may also appear if their contributions cancel out. Where only these three families of faces are present, one obtains f₅ = f₇. This result describes the combinatorics and the topology of the network. It does not mean that an ordinary torus embedded in three-dimensional space can be covered with regular hexagons without metric deformation.
3. The intrusion of translation: the crystallographic restriction
The orders 1, 2, 3, 4 and 6 of crystallography introduce a third family of constraints. It is no longer enough for a local motif to be geometrically coherent. It must also remain compatible with a periodic repetition by translation. A rotation that preserves a periodic lattice must satisfy the condition:
2 cos(2π/n) ∈ ℤ
This condition leaves only five orders of rotation:
n ∈ {1, 2, 3, 4, 6}
Symmetries of order 5 or 7 are not forbidden to matter. They are incompatible with one precise type of organisation, the classical periodic crystal lattice. Quasicrystals can in particular display a symmetry of order 5 because they possess long-range order without classical translational periodicity.
4. Synthesis: constraint as the matrix of form
These three families of results expose one and the same principle. A constraint organises the space of accessible solutions and makes certain forms possible, stable or necessary, beyond the mere reduction of an already constituted set.
| Regime or structure | Structural constraint | Compatible solutions |
|---|---|---|
| Euclidean plane | Sum of angles = 360° | 3 regular tilings |
| Convex closure | 1/p + 1/q > 1/2 | 5 regular polyhedra |
| Spherical topology | χ = 2 (Euler + trivalence) | Combinatorial balance: Σ(6 − n)fₙ = 12 |
| Toroidal topology | χ = 0 (Euler + trivalence) | Combinatorial balance: Σ(6 − n)fₙ = 0 |
| Crystalline periodicity | 2 cos(2π/n) ∈ ℤ | Orders 1, 2, 3, 4, 6 |
Overall reading. The plane retains three tilings because the angles must fill exactly one full turn. Convexity retains five polyhedra because the sum must stay below that same turn. Topology then transforms the local departures from the hexagon into a global balance, positive on the sphere and null on the torus. Crystalline periodicity finally adds a translational constraint that selects five orders of rotation.
The same values may reappear in different contexts, but they do not carry their explanation within themselves. The observed number is the visible output of a problem, an invariant or a rule of compatibility. Understanding an architecture consists in moving back from the value toward the relations that made it necessary or accessible.
Constraint is not an external limit added to form. It is part of the relations that make that form possible.
5. Scope and limits of constraints
The preceding results establish that the observed numerical value depends on the mechanism producing it. They nevertheless open a second inquiry. A topology, a combinatorics or a rule of periodicity does not always suffice to determine the geometric form actually realised. Between the abstract structure and the material object, lengths, angles, deformations, forces and conditions of fabrication also intervene. Three extensions make it possible to explore this articulation without abandoning the guiding thread of the text.
5.1. Topology and metric: the same network, different geometric forms. Topology describes relations of neighbourhood, connectivity and the properties that resist continuous deformation. The metric, by contrast, describes the distances, angles, areas and geometric curvatures actually realised. Two objects may possess the same topology without having the same geometry, and one and the same combinatorics may be materialised by very different metric forms.
The torus provides the most direct example. A hexagonal tiling of the plane can be periodically identified in two directions to form an abstract flat torus. In this construction, each vertex keeps exactly the neighbourhood of the hexagonal network and the combinatorial balance remains null. An ordinary torus embedded in three-dimensional space nevertheless does not have zero metric curvature everywhere: its outer part shows positive curvature, while its inner region shows negative curvature. Topology imposes the global balance, but it fixes neither the local distribution of curvature nor the precise form of the object embedded in space.
The Gauss–Bonnet theorem makes this articulation precise: on a closed surface, the integral of the Gaussian curvature equals 2πχ. For the torus, where χ = 0, the total curvature vanishes exactly, so that the regions of positive and negative curvature compensate one another whatever the metric realisation adopted. The same topological invariant governs both the combinatorial balance of the network and the global curvature balance of the surface that carries it.
This distinction becomes essential as soon as one moves from mathematical demonstration to physical modelling. A membrane, a shell, a foam or a reticulated material must simultaneously satisfy a connectivity, edge lengths, rigidities, tensions and boundary conditions. Topology determines what must be preserved through deformations. Metric and mechanics determine the cost of the possible realisations and select the one the system actually adopts.
Topology imposes a balance. The metric determines how that balance takes shape in space.
Scope of the formula. The identities used here concern closed surfaces without boundary. In the presence of a boundary, some edges belong to a single face and the double-incidence relations are modified. Additional terms, bound up with the contour and with the valence of its vertices, must then be introduced. The substitution χ = 2 − 2g − b alone does not suffice.
5.2. Quasicrystals: what becomes possible when periodicity is relaxed. The crystallographic restriction does not forbid symmetries of order 5, 8, 10 or 12 to matter. It forbids their compatibility with a periodic lattice of translations. Quasicrystals show what happens when this precise requirement is abandoned without giving up all order.
Quasicrystals showed experimentally that long-range order could exist without translational periodicity. A quasicrystal produces sharp diffraction patterns, yet its structure does not repeat by translation according to a single unit cell. Relaxing periodicity then opens a space of solutions inaccessible to the classical crystal. Symmetries forbidden by the crystallographic restriction theorem become compatible with a non-periodic order, notably the symmetry of order 5.
This case directly illustrates constraint as the matrix of form. The passage from the periodic crystal to the quasicrystal does not consist in removing all rules. It replaces one constraint by another: strict translational repetition gives way to a non-periodic global order. The permitted forms change because the space of constraints has changed. The 5 was not impossible in itself. It was incompatible with one determinate regime of organisation.
Relaxing a constraint does not produce disorder: it can bring another type of order into being.
5.3. Rereading the classical theorems as frontiers of possibility. Many classical theorems can be reread from the same angle. They do not merely state that a construction exists or does not exist. They describe the frontiers between several regimes of possibility, identifying the constraint that permits certain structures and excludes others.
The classification of regular tilings shows that equality of angles around a vertex leaves only three solutions. The classification of the Platonic solids shows that an angular deficit compatible with convexity leaves only five. The crystallographic restriction shows that translational periodicity retains only rotations of order 1, 2, 3, 4 and 6. Euler's formula shows that closing a trivalent network on a sphere imposes a combinatorial balance equal to 12.
The absence of a regular tiling of the plane by regular pentagons belongs to the same family of results. It does not mean that the number 5 is geometrically deficient or that every pentagon is incapable of tiling the plane. It means that the interior angle of the regular pentagon cannot fill exactly 360 degrees around a vertex with a whole number of identical copies. Non-regular pentagons, for their part, can tile the plane because their angles and their assemblies satisfy other relations.
This reading transforms the impossibility theorems. A mathematical impossibility is never an absence of explanation. It always reveals the presence of a constraint. It localises the constraint responsible for the exclusion and often indicates which assumption must be modified to open a new space of forms: abandoning regularity, changing the topology, relaxing periodicity, introducing several types of face or moving to another geometry.
6. A general method of reading
These extensions make it possible to reformulate the approach as a method. Faced with a numerical regularity or a recurrent form, the first step consists in relating the value to the problem it answers. One must identify the object counted, the relations preserved, the metric assumptions, the topology of the support, the symmetries required and the constraints possibly relaxed.
| Question | What must be identified | What it reveals |
|---|---|---|
| What is the regime of constraints? | Geometric, topological, metric, dynamic | Why some forms exist and others do not |
| What is the support? | Plane, sphere, torus, surface with boundary, three-dimensional space | The Euler characteristic and the possibilities of closure |
| What is preserved? | Connectivity, lengths, angles, periodicity, symmetry | The type of constraint that filters the solutions |
| What is being counted? | Faces, vertices, neighbours, defects, orders of rotation | The real meaning of the numerical value |
| Which assumption can be relaxed? | Regularity, convexity, periodicity, Euclidean metric | The new architectures that become accessible |
A form is explained by the system of constraints that makes the associated value necessary, possible or probable. The numerical value takes on its meaning within that system.