The eleven-sided one nobody can fold

The heptagon is the first place folding beats the compass, and it is a satisfying result partly because the gain is so cheap: one axiom that solves a cubic, and a whole family of polygons comes into reach.

The gain is also bounded, and the boundary arrives sooner than most accounts mention. The regular hendecagon — eleven sides — cannot be folded, and the reason is a single prime factor.

Why a tool’s reach is a statement about degrees #

Both results come from the same machinery, and it is worth restating compactly because everything else follows from it.

A construction produces points, and the coordinates of those points are numbers. Each step of a construction produces coordinates that satisfy a polynomial equation over the numbers already available, so the whole construction builds a tower of field extensions, and what a tool can reach is decided by the degrees of the steps it allows.

Straightedge and compass: a step intersects two lines, a line and a circle, or two circles. The resulting coordinates satisfy at worst a quadratic, so every step has degree one or two, and every reachable number has degree a power of two over the rationals.

Folding: the axiom that solves a cubic is the one that folds two points onto two lines simultaneously, and it introduces degree three. Every step therefore has degree one, two or three, and every reachable number has degree 2ᵃ·3ᵇ.

That is the whole difference between the tools, and everything about which polygons they build is a corollary.

From degrees to polygons #

The regular n-gon is constructible with a given tool exactly when the number cos(2π/n) is reachable by that tool, and the degree of that number over the rationals is φ(n)/2, where φ is Euler’s totient — the count of numbers below n sharing no factor with it.

So the condition on n is a condition on φ(n):

• compass: φ(n) a power of two — Gauss’s criterion, and Pierre Wantzel’s proof in 1837 that nothing else works.
• folding: φ(n) of the form 2ᵃ·3ᵇ.

For n = 7, φ(7) = 6 = 2·3. The compass fails and a fold succeeds, which is the heptagon result.

For n = 11, φ(11) = 10 = 2·5. A five. No product of twos and threes is ten, so the hendecagon is out of reach of both tools, and it is the smallest regular polygon of which that is true.

The primes that matter #

The condition on a prime p is that p − 1 be smooth in the relevant sense, and the two families have names.

Fermat primes are primes one more than a power of two: 3, 5, 17, 257, 65537. Five of them are known and it is an open question whether there are any more. These are the primes whose regular polygons the compass builds.

Pierpont primes are primes one more than 2ᵃ·3ᵇ: 2, 3, 5, 7, 13, 17, 19, 37, 73, 97, 109, 163 and onward. These are the ones a fold builds, and unlike the Fermat primes there are conjecturally infinitely many — the density is thin but the supply does not obviously run out.

Eleven is neither. Twenty-three is neither, since 22 = 2·11. Twenty-nine is neither, since 28 = 2²·7. The primes a fold cannot reach are not exotic; they start at eleven and there are plenty of them.

Reading the table downward #

The table at the top of this essay repays being read as a whole rather than for its one interesting row, because the pattern of marks says something the hendecagon alone does not.

Up to forty there are twelve primes. The compass reaches four of them — 3, 5, 17 and, in the range shown, nothing else, because Fermat primes are desperately scarce. A fold reaches eight: 3, 5, 7, 13, 17, 19, 37 and 2. So the fold roughly doubles the haul in this range, and the numbers it adds are exactly the ones with a three in them.

The primes neither tool reaches — 11, 23, 29, 31, 41 — have factorisations with a five, a seven or an eleven in them, and those factors are simply unavailable. A tool does not fail on a prime because the prime is large; it fails because of what is inside it. Thirty-seven is reachable and eleven is not.

That single axiom is doing all of the extra work in the table, and it is worth appreciating how little it is. One move out of seven, one degree of extension, and the reachable set goes from a handful of Fermat primes to a conjecturally infinite family.

What “cannot” means here #

The word is doing precise work and deserves the same care the completeness of the axiom list does.

It does not mean nobody has found a construction. It means no construction exists, and the argument is the same shape as the classical impossibility proofs: a construction produces numbers in a field of degree 2ᵃ·3ᵇ, cos(2π/11) has degree five, five does not divide any 2ᵃ·3ᵇ, and a subfield’s degree divides the field’s. The obstruction is arithmetic and it is absolute.

It also does not mean an eleven-sided figure cannot be folded from paper. It can be, to any accuracy anybody wants, by the same kind of iterative method that divides a strip into thirds — guess and correct, with the error contracting geometrically. What cannot be done is to reach it exactly, in finitely many folds, with the vertices at their true positions.

That distinction matters more here than in the compass case, because folding has a strong tradition of approximate methods and they are excellent. The impossibility is about exactness, and the practice mostly does not need exactness. Where it does — a tessellation whose errors accumulate across a sheet — the distinction bites, which is why exactness is worth its awkwardness.

What a second fold would buy #

The natural question is whether some richer folding operation reaches further, and the answer is yes, with a known price.

The seven axioms describe folds made one at a time, each aligning points and lines that already exist. Allowing two creases to be made simultaneously, with an alignment condition linking them, gives a strictly stronger tool: the two-fold operations solve quintics and higher, and Roger Alperin and Robert Lang worked out in 2006 what an n-fold operation reaches.

That would put the hendecagon in range, since it needs degree five. It would also change what “folding” means, and not in a way anybody’s hands recognise: a simultaneous two-fold is not a move a person makes, it is a constraint satisfied by two creases at once, and executing one means finding the configuration rather than performing a step.

So the hierarchy is genuine and the practical statement is unchanged. Single-fold origami reaches 2ᵃ·3ᵇ, and that is the tool the tradition uses. This site’s axiom list is the single-fold one, its constructions live inside it, and the hendecagon is outside.

There is a reason to be careful with the multifold results beyond their unfamiliarity. A construction is only as good as the operation it assumes can be performed, and the seven single-fold axioms are chosen precisely because each of them describes an alignment a person can see: bring this point onto that line, and stop when it lands. A two-fold operation has no such moment. The alignment is a joint condition on two creases, and there is no partial state in which a folder can tell whether they are getting closer. What the arithmetic gains, the hand loses.

That is a good illustration of why the axiom list is where it is rather than somewhere more powerful. It is not a list of everything paper can do. It is a list of moves with a visible success criterion, and the argument that it is complete is an argument about that class of moves.

The advantage is a small-numbers effect #

The table’s tally — eight primes of twelve for a fold against four for a compass — reads as a doubling of reach, and it is worth extending the count, because the ratio does not hold.

Below forty a fold reaches two thirds of the primes. Below a thousand the picture is different: there are a hundred and sixty-eight primes, and the ones of the form two-to-the-a times three-to-the-b plus one number eighteen. That is under eleven per cent.

The list is short enough to write out: 2, 3, 5, 7, 13, 17, 19, 37, 73, 97, 109, 163, 193, 257, 433, 487, 577 and 769. Everything else below a thousand is out of reach — a hundred and fifty primes, and every regular polygon built on one of them.

So the fold’s advantage over the compass is real and it is concentrated at the very bottom of the range. Both tools reach almost nothing, and the interval where the difference is visible is the interval where the numbers are small enough for a product of twos and threes to land on a prime by luck.

Which is what makes the hendecagon representative #

That changes what the hendecagon is an example of. Read from the table it looks like a near miss — the first failure, immediately past a run of successes, with a single five spoiling it.

Read against the density it is the ordinary case. Eleven is the first prime a fold misses and it is one of a hundred and fifty below a thousand; the seven or eight it reaches are the anomalies. The interesting thing about the heptagon is not that folding reaches it but that folding reaches anything at all, and the reason it does is that the smallest numbers have the most factorisations into twos and threes.

That also settles a question the essay raises about scarcity. It is an open problem whether there are finitely many Fermat primes and conjectured that there are infinitely many Pierpont primes — but infinitely many at this density is a very thin infinity. The count of Pierpont primes below $N$ grows like the square of the logarithm of $N$, which is to say it barely grows at all, and a folder’s supply of constructible polygons is exhausted long before their patience is.

What a designer loses, which is almost nothing #

It is worth asking whether the hendecagon’s unreachability costs anybody anything, because the honest answer sharpens what impossibility results are for.

Regular polygons enter folding design in two places. They are the reference frames for tessellations — a twist needs a polygon to twist around — and they are the outlines that division methods produce. In the first role only 3, 4 and 6 are used, because those are the ones that tile, and all three are reachable by a compass let alone by a fold. In the second, what is wanted is usually a division of an edge rather than a polygon.

That last point is the one worth carrying away, and it is easy to get backwards. Dividing a segment into eleven equal parts is easy; building an eleven-sided polygon is impossible. Division is rational arithmetic and the axioms do it without effort. The polygon needs cos(2π/11), which is where the fifth-degree obstruction lives.

So the impossibility is real, absolute, and almost entirely without practical consequence — which is the usual fate of this family of results and no reason to think less of them. What they settle is what a tool is, and knowing that is worth having whether or not anybody was going to fold a hendecagon.

What no figure can show #

Every figure on this page is a table, and that is not laziness. An impossibility has no picture. There is no drawing of the hendecagon-that-cannot-be-folded which differs from a drawing of the hendecagon, because the object exists perfectly well — what does not exist is a finite folding sequence reaching it.

That is the same difficulty that makes a checker’s blind spot undrawable, and it recurs whenever a claim is about the absence of something. The honest response is to draw the arithmetic, which is what the figures here do: the factorisations, the totients, and the marks that follow from them.

The second thing no figure shows is how close an approximate hendecagon gets. That is a real and answerable question — the error of Fujimoto’s method after k folds is 2⁻ᵏ of the initial guess — and it is a question about a method rather than about the polygon, so it belongs to the essay about the method.

The gap between the tools, and the gap after them #

There is a tidy way to summarise all of this and it is worth stating, because it makes the hendecagon’s position clear rather than merely unfortunate.

Each tool corresponds to a set of degrees, and the sets nest: {2ᵃ} ⊂ {2ᵃ3ᵇ} ⊂ {2ᵃ3ᵇ5ᶜ…}. A compass sits in the first, single-fold origami in the second, and the multifold operations reach into the third. Every polygon lives at some degree and is built by every tool from its own level upward.

The hendecagon is not near the boundary of what folding does. It is at degree five, which is the very first degree folding misses, so it sits immediately outside — the way the heptagon sits immediately outside the compass’s reach. The two failures are the same failure one level apart, and an angle trisected by a fold is the move that separates them.

The idealisation, named #

The whole argument treats a fold as an exact reflection of the plane in a line, and a crease as that line.

Neither is true. A crease is a region rather than a line and it has a radius; a fold is exact to whatever accuracy the folder’s alignment achieves, which for a careful hand on good paper is perhaps a tenth of a millimetre. On a 150 mm sheet that is one part in fifteen hundred, and it swamps the distinction between a constructible number and a very good approximation to one.

So the impossibility result is a statement about an idealised operation, and its practical content is not that the hendecagon is unfoldable — it is that no sequence of exact folds reaches it, which means the error of an approximate construction cannot be driven to zero by being more careful about the geometry. It can only be driven down by iterating, which is a different kind of method with a different kind of guarantee.

Who proved this, and when #

Gauss gave the sufficiency half of the compass criterion in 1796, at eighteen, with the seventeen-gon; Wantzel proved the necessity in 1837 and thereby closed a problem that had been open since antiquity.

The folding criterion is much more recent. That folding solves cubics was established through the twentieth century in pieces — Margherita Piazzolla Beloch’s fold in 1936 is the crucial one — and the characterisation of origami-constructible numbers as those of degree 2ᵃ·3ᵇ was set out by Roger Alperin in 2000. Pierpont primes are James Pierpont’s, from 1895, and were defined for exactly this kind of question before folding was thought of as a construction tool at all.

The pleasing part of that history is the order. The number theory was in place a century before anybody asked what a sheet of paper could build, and when the question was finally asked the answer was waiting. Pierpont was interested in exactly this kind of question — which regular polygons yield to which class of construction — and considered angle trisection as an allowed operation, which is the same extension by degree three that a fold turns out to provide. He was not thinking about paper and the criterion is the paper’s anyway.

Where this goes next #

The reach of a tool is one kind of boundary. Another is the boundary a pattern runs into: every regular polygon admits a twist, and only three of them tile the plane — so the restriction that produces the familiar twist tessellations turns out to be about the plane rather than about the paper, which is the opposite of where anybody would look for it.