Why the list of perfect solids stops at five

Regular polygons go on forever. Triangle, square, pentagon, hexagon, and onward through shapes with a hundred or a million equal sides, each perfectly well defined and perfectly regular. There is no largest one and no obstruction anywhere.

Regular solids stop at five.

A regular solid — Platonic solid, in the usual name — is one whose faces are all copies of the same regular polygon, with the same number of them meeting at every corner. The Greeks were as interested in these as they were in the figurate numbers, and for the same reason: both are places where counting and shape turn out to be the same subject. That is not much of a demand. Two dimensions grants it infinitely often. Three dimensions grants it five times and then refuses.

The argument fits in one row #

The whole obstruction is local. Forget the solid; look at a single corner.

At a corner, some number of identical regular polygons meet. Three at minimum — two faces meeting along an edge make a flap, not a corner. And crucially, the angles of those faces must add up to less than a full turn. If they sum to exactly $360°$, the faces lie flat and there is no corner, just a piece of tiled floor. If they sum to more than $360°$, they cannot be assembled without overlapping.

That single constraint decides everything.

Walk it through:

• Triangles, interior angle $60°$. Three give $180°$, four give $240°$, five give $300°$ — all under budget, all real corners. Six give exactly $360°$: flat, the familiar triangular tiling of the plane. Seven or more overshoot. So triangles allow three possibilities.
• Squares, $90°$. Three give $270°$: a corner, and it is the corner of a cube. Four give $360°$: flat, the graph-paper tiling. So squares allow one.
• Pentagons, $108°$. Three give $324°$: a corner. Four give $432°$, hopelessly over. Pentagons allow one.
• Hexagons, $120°$. Three already give exactly $360°$ — the honeycomb, flat. There is no room even for the minimum. Hexagons allow none.
• Everything beyond: interior angles only grow, so three copies always exceed $360°$. Nothing.

Three plus one plus one is five. The three triangle cases are the tetrahedron, octahedron and icosahedron; the square case is the cube; the pentagon case is the dodecahedron.

Notice that the budget is a local constraint doing global work. Nothing in the argument looks at the solid as a whole; it only ever examines one corner. That a purely local check settles a global question is unusual and worth flagging — the same thing happens in Königsberg, where counting the edges at each landmass settles whether a route exists across the entire city.

What is satisfying about this is how little it assumes. It never asks whether the solids can actually be built — it only rules out corners that cannot exist. That the five surviving corner types each do close up into a genuine solid is a separate fact, and Euclid spends the whole of Book XIII of the Elements constructing them and proving the list complete, ending the work he is most famous for on exactly this note.

The same five, paired off #

Set a point at the centre of each face of a cube and join points on neighbouring faces. The result is an octahedron. Do the same to an octahedron and a cube comes back.

Exchanged counts are a hint rather than a construction, and the construction is short enough to draw: put a point at the centre of each face and join the points whose faces share an edge. Doing that to one of the pair produces the other, with faces and corners swapping roles because a face centre becomes a corner.

The dodecahedron and icosahedron pair up the same way: 12 faces and 20 corners against 20 faces and 12 corners. The tetrahedron is its own dual, being the one that pairs with itself.

Duality explains why the list has the shape it does. The five solids are not five unrelated objects; they are two pairs and a self-paired singleton. Anything true of one member of a pair has a mirrored statement about the other, obtained by swapping the words face and vertex throughout. Whole arguments can be reused for free.

A second route #

There is another way to reach five, using Euler’s formula: for any convex polyhedron,

$$V - E + F = 2,$$

where $V$, $E$ and $F$ count vertices, edges and faces. Suppose every face has $p$ sides and every vertex has $q$ edges meeting it. Counting edge-ends two ways gives $pF = 2E$ and $qV = 2E$, so substituting into Euler’s formula and dividing through by $2E$:

$$\frac{1}{p} + \frac{1}{q} = \frac{1}{2} + \frac{1}{E}.$$

Since $E$ is positive, the left side must exceed $\tfrac12$. With $p, q \geq 3$, the whole-number solutions are $(3,3)$, $(4,3)$, $(3,4)$, $(5,3)$ and $(3,5)$. Five again, and the duality is now visible in the notation: swapping $p$ and $q$ swaps the members of each pair.

That equation delivers more than the list, because $E$ is still in it. Solve for it — $E = 1/(1/p + 1/q - \tfrac12)$ — and then $F = 2E/p$ and $V = 2E/q$ follow from the two counting identities. Feeding the five admissible pairs through gives $6$, $12$, $12$, $30$ and $30$ edges, and with them the entire census: $4$ faces and $4$ corners for $(3,3)$, six faces and eight corners for $(4,3)$, eight and six for $(3,4)$, twelve and twenty for $(5,3)$, twenty and twelve for $(3,5)$. Every count in this essay falls out of one equation, with nothing constructed and nothing counted by hand, and the duality appears as the transposition it is.

The same inequality also accounts for the cases the angle budget discarded, which is the part worth dwelling on. Three hexagons at a corner sum to exactly $360°$; in this notation that is $1/p + 1/q = \tfrac12$ exactly, and the equation then demands $1/E = 0$. That is not “no solution” — it is infinitely many edges. The equality cases are $\{3,6\}$, $\{4,4\}$ and $\{6,3\}$, which are exactly the three regular tilings of the plane, arriving as the polyhedra with infinitely many faces. Below the line, where $1/p + 1/q < \tfrac12$ — seven triangles at a corner, four pentagons — $E$ comes out negative, which is the equation’s way of reporting that the corner has more angle than the plane can hold. Those configurations exist as well, as regular tilings of the hyperbolic plane, and unlike the other two cases there are infinitely many of them.

So one quantity, $1/p + 1/q$ measured against $\tfrac12$, sorts every regular configuration there is into three classes: above gives the five solids, equal gives the three flat tilings, below gives the hyperbolic ones. The list is not short because the question is hard. Five is the finite case of a trichotomy, and the reason it is five is that there are only five ways to beat a half.

Two completely different arguments, one about angles at a corner and one about counting edges, land on the same list — the same-thing-twice situation that usually signals something structural underneath. That is usually a sign that the answer is about the structure rather than about the method — and comparing the two is worthwhile, because the angle argument makes it obvious why the list is finite while the Euler argument makes it obvious why the entries pair up. Neither picture contains both facts comfortably.

Where checking corners is not enough #

The angle argument works because a local condition happens to settle a global question. It is worth seeing a nearby case where the same move fails, because the failure was overlooked for four hundred years.

Relax the definition slightly. Allow more than one kind of regular face, but keep the requirement that every corner look identical — the same polygons meeting in the same cyclic order all the way round. That is the standard description of the Archimedean solids, and the standard answer is thirteen.

The standard answer is thirteen and the standard description is wrong. There is a fourteenth shape satisfying it.

Take a rhombicuboctahedron — twenty-four corners, at each of which three squares and a triangle meet — and cut it along a band, rotate the top cap by an eighth of a turn, and set it back down. Every corner still has three squares and a triangle in the same cyclic order. Every face is still a square or an equilateral triangle. Nothing local has changed anywhere, because the twist happened along a seam where squares meet squares either way.

Globally it is a different solid. In the original, any corner can be carried to any other by a symmetry of the whole shape; in the twisted one it cannot, because a corner in the rotated cap sits in a different relationship to the far side than a corner below the seam. The object is a Johnson solid, the elongated square gyrobicupola, and it is the standard counterexample to the assumption that identical corners force a uniform solid.

The error persisted a long time. The shape appears in a 1930 model collection by J. C. P. Miller and was for years assumed to be a version of the ordinary rhombicuboctahedron; it is still occasionally called the pseudo-rhombicuboctahedron or Miller’s solid. Getting the Archimedean list right requires stating the condition globally — that the symmetry group acts transitively on the vertices — rather than as a check performed one corner at a time.

None of which damages the argument for five, and that is the point of raising it. The angle budget is not a local check that turns out to be sufficient; it is a local check used only to rule things out, which is a job local checks can always do. Ruling things in is the direction that fails, and the essay’s own five-corners-five-solids match is a separate theorem precisely because of it.

The same budget, one dimension up #

The section above notes that the corner argument would not have predicted the four-dimensional answer. It would have, run properly, and running it is worth the paragraph because the answer is stranger than five.

In three dimensions the budget is a full turn of plane angle around a corner, spent by polygons. In four, the thing being surrounded is an edge rather than a corner, and what is spent around it is the dihedral angle — the angle between two faces of a solid meeting along an edge. So the question becomes: which Platonic solids can be packed around an edge, several at a time, without their dihedral angles reaching a full turn?

The five dihedral angles are about $70.5°$ for the tetrahedron, $90°$ for the cube, $109.5°$ for the octahedron, $116.6°$ for the dodecahedron and $138.2°$ for the icosahedron. Three of anything is again the minimum, and the budget is again $360°$.

Tetrahedra are cheap enough to allow three, four or five around an edge — five spend about $353°$, just inside — and six overshoot. Cubes allow three, at $3 \times 90 = 270$ degrees, and four spend exactly $360°$, which is flat and is the ordinary stacking of cubes filling space. Octahedra allow three, at about $328°$. Dodecahedra allow three, at about $350°$, with almost nothing to spare. Icosahedra allow none: three already spend over $414°$.

Three plus one plus one plus one is six, and those six exist — the four-dimensional analogues of the tetrahedron and cube and octahedron, plus the 24-cell, the 120-cell and the 600-cell. The extra one, the 24-cell, is built from octahedra and has no counterpart in any other dimension.

Run the same accounting again in five dimensions and it collapses. The cells are now the six four-dimensional polytopes, and their dihedral angles are large: only the simplex and the hypercube are shallow enough to admit three around a face, with the 16-cell and the 24-cell landing on exactly $360°$ — which is why both of them tile four-dimensional space — and the last two far over. What survives is the simplex, the hypercube and the cross-polytope: three, in every dimension from five upward, forever.

So the sequence of counts runs $5, 6, 3, 3, 3, \ldots$ and the peak is at four. That is a genuinely odd shape for an answer, and the budget explains it exactly: four dimensions is the last place where the shallow-angled cells are still numerous enough to make anything exotic, and after that only the three families that exist in every dimension by construction are left. The list is short for the same reason it was short in three dimensions, and it is longest where the accident of the octahedron’s angle leaves just enough room.

Kepler’s mistake, which was a good one #

In 1596 Johannes Kepler published a model of the solar system built on this list. There were six known planets, therefore five gaps between their orbital spheres, therefore — since there are exactly five regular solids — the spheres must be separated by the five solids nested inside one another in the right order. Octahedron, icosahedron, dodecahedron, tetrahedron, cube.

The coincidence of counts is genuinely striking, the model is beautiful, and it is entirely wrong. Kepler eventually abandoned it, in favour of the observation that the orbits are ellipses — which, as it happens, is another shape that comes from slicing a cone. He seems to have regarded the trade as a loss for most of his life — swapping a picture with five perfect objects in it for one with a tilted plane.

It is worth being clear about what went wrong, because it was not the mathematics. The five solids really are exactly five, for exactly the reason given above. What failed was the assumption that a number showing up in two places means the two places are connected. Six planets and five solids was a coincidence about which objects happened to have been discovered by 1596, and the discovery of Uranus in 1781 would have ended the model even if nothing else had. Matching counts are the weakest kind of evidence there is; a pattern that is visibly present is not thereby explained.

What the picture cannot show #

The angle-budget figure rules out corners. It does not construct solids, and the difference matters more than it looks.

Showing that three pentagons can meet at a corner is not the same as showing that twelve pentagons close up into a dodecahedron. The corner argument leaves open the possibility of a configuration that starts assembling happily and then fails to meet itself on the far side — and nothing in the picture would detect that. Euclid spends most of Book XIII on exactly this gap, constructing each solid explicitly and inscribing it in a sphere.

That the two halves happen to match — five permitted corners, five actual solids — is a genuine theorem, not a corollary of the drawing. When the same style of argument is run in four dimensions the counts do not match so tidily: six regular polytopes exist there, and in five dimensions and above there are only ever three. The corner budget alone would not have predicted that.

What the constraint really is #

The reason three dimensions is stingy where two is generous comes down to a single thing: there is a fixed amount of angle at a corner, and only so many ways to spend it.

A plane tiling has no such budget — it has all the room it needs, forever, which is why regular polygons never run out. A solid has to close up, and closing up costs angle. Every corner of a Platonic solid is a place where the faces fall short of flat, and that shortfall is the quantity worth naming: the angle defect at a corner is $360°$ minus whatever the faces there actually contribute.

The remarkable part is what the defects add up to.

Run the same sum on the others. The tetrahedron has four corners, three triangles each, defect $360° - 180° = 180°$: four times $180°$ is $720°$. The cube has eight corners, three squares each, defect $90°$: eight times $90°$ is $720°$. The octahedron has six corners of four triangles, defect $120°$: $720°$ again. The dodecahedron has twenty corners of three pentagons, defect $360° - 324° = 36°$: twenty times $36°$ is $720°$.

Five solids with completely different corner counts and completely different corners, and the total is the same number every time — and it is $720°$, which is two full turns, which is the $2$ in $V - E + F = 2$. That is Descartes’ theorem, and it predates Euler’s formula by a century in an unpublished manuscript that was not found until long after Euler had done it again.

The result is not confined to the regular solids, which is what makes it more than a coincidence of five arithmetics. Every convex polyhedron whatsoever has total defect exactly $720°$ — a squashed box, a pyramid on a heptagon, anything at all. Sharpening one corner takes defect from somewhere else; the total is fixed before the shape is chosen.

What that says is that the defect is not measuring the corner. It is measuring the sphere the corners are wrapped around, distributed over however many corners the polyhedron happens to have. A shape with a hole through it, wrapped instead around a doughnut, totals $0°$; two holes total $-720°$. The angle at a corner is local and arbitrary; the sum is global and forced, and the continuous version of that sentence is the Gauss–Bonnet theorem.

The list is short because the budget is small. There is nothing more to it than that, and there is no sixth solid hiding anywhere.

The ladder from here #

Beyond this rung: Euclid’s actual constructions from Book XIII, drawn. The Archimedean solids, which relax “all faces identical” and immediately give thirteen more. The Kepler–Poinsot star polyhedra, which relax “convex” and give four. Duality made precise, as a correspondence rather than an observation. The regular polytopes in four dimensions, where the count rises to six before collapsing to three forever. The symmetry groups of the five solids, and their unexpected appearance in the roots of polynomials. Angle defect and the discrete Gauss–Bonnet theorem, which is Euler’s formula in geometric clothing. And sphere packing, where the icosahedron very nearly wins and then does not.