One fold at a time, and there are exactly seven of them

A fold is a reflection. Everything on one side of the crease is mirrored across it, and the crease is the line of the mirror.

That is the whole of the mechanics, and it means a fold is completely specified once the crease line is. The interesting question is how a folder specifies one — because a fold is made by hand, by bringing things into alignment, and the things available to align are the points and lines already on the paper.

The list is short, it is complete, and it is not obvious that either of those should be true.

What “specify a fold” means #

A crease is a line, and a line has two degrees of freedom. Each alignment condition removes one. So a fold is determined by two conditions, or by one condition plus a constraint on direction.

The available conditions come in a small number of shapes:

• bring a point onto a point;
• bring a point onto a line;
• bring a line onto a line;
• pass the crease through a point;
• make the crease perpendicular to a line.

Combine those in pairs and enumerate. Some combinations are impossible, some are degenerate, and what survives is seven distinct operations. That is the whole derivation, and it is why the list is a theorem rather than a catalogue somebody stopped adding to.

The seven #

Axiom 1 — the fold through two given points. The unique crease joining them.

Axiom 2 — the fold bringing one point onto another. Their perpendicular bisector, which is exactly the mirror that swaps them.

Axiom 3 — the fold bringing one line onto another. The angle bisector; there are two of them when the lines cross, and one when they are parallel.

Axiom 4 — the fold through a point, perpendicular to a line. A right angle dropped onto the line.

Axiom 5 — the fold bringing a point onto a line, passing through another point. The crease must pass through the fixed point, so its distance from the moving point is fixed, and the question becomes where a circle of that radius meets the line. Zero, one or two solutions — which is a quadratic.

Axiom 6 — the fold bringing two points onto two lines simultaneously. This is the one that matters, and the rest of this site keeps returning to it.

Axiom 7 — the fold bringing a point onto a line, perpendicular to another line. Hatori’s addition, and the one everybody had missed.

Why the sixth is different #

Axioms 1 to 5 and 7 do nothing a compass could not. Their solutions are intersections of lines and circles, which is precisely what straightedge and compass construct, so anything foldable by those six is constructible in the classical sense, and offers nothing the compass does not already have.

Axiom 6 is not like that.

Bringing two points simultaneously onto two lines is, algebraically, the problem of finding a line tangent to two parabolas — each point-and-line pair defines a parabola with that point as focus and that line as directrix, and a fold placing the point on the line is a tangent to it. A common tangent to two parabolas is the root of a cubic.

That difference in degree is the whole story. Straightedge and compass reach exactly the numbers obtainable by repeated square roots — field extensions of degree $2^k$. Folding reaches extensions of degree $2^a 3^b$. Every classical impossibility that turns on a cubic falls to a fold.

The three that fall #

Three problems from Greek antiquity resisted straightedge and compass for two thousand years, and two of them are cubic.

Trisecting a general angle. Trisection satisfies $\cos 3\theta = 4\cos^3\theta - 3\cos\theta$, a cubic in $\cos\theta$, and it is done with one fold.

Doubling the cube. Constructing $\sqrt[3]{2}$ is the cube root problem by definition, and it takes a single crease.

Squaring the circle. This one does not fall, and the reason is instructive: $\pi$ is transcendental, so it is not the root of any polynomial of any degree. Folding gains a degree, and transcendence is not about degree at all.

The pattern is worth stating clearly, because “origami is more powerful than Euclid” is often said loosely. It is more powerful in one specific and bounded way — it climbs from degree 2 to degree 6 — and it is exactly as helpless as a compass against anything transcendental.

Completeness, which is the surprising part #

That there are seven operations is easy to believe. That there are only seven is a claim about everything, and it needs proving rather than checking.

Lang’s argument runs by exhaustion over the alignment conditions. A single fold can be specified by at most two independent conditions; the conditions available are the five shapes listed above; enumerating pairs of them gives a finite list; and every entry either reduces to one of the seven, is impossible, or is degenerate.

The subtlety is that “point onto line” can be applied twice — that is axiom 6 — and applying it three times overdetermines the fold. So the enumeration terminates, and it terminates at seven.

The enumeration, carried out #

The completeness argument was described above rather than performed, and performing it takes a paragraph, which is worth spending because the terminating step is the interesting one.

Sort the five alignment shapes by how much of the crease’s two degrees of freedom each removes. Through a point removes one. Perpendicular to a line removes one. Point onto a line removes one — the creases carrying a given point to a given line form a one-parameter family, the tangents to a parabola. Point onto point removes both at once, since the perpendicular bisector is unique. Line onto line removes both, being the bisector of the angle.

A crease needs exactly two removed, so the admissible recipes are one two-condition shape on its own, or two one-condition shapes together. That is a list of eight:

The last recipe asks a crease to be square to two lines at once, which is impossible unless the lines are parallel and vacuous when they are. Seven survive, and nothing else can be written down because three conditions overdetermine a line and one leaves it free.

How many creases each recipe yields #

The table also settles a question the list of names hides, which is how many answers each operation has — a matter of practice as well as of algebra.

Axioms 1, 2, 4 and 7 give exactly one crease. Axiom 3 gives two when the lines cross and one when they are parallel. Axiom 5 gives none, one or two, being a circle meeting a line. Axiom 6 gives one, two or three, being the real roots of a cubic.

So a single alignment step, in the most generous position, produces eleven distinct creases across the seven operations. That is the branching factor of a folding construction, and it is small — which is why constructions are found by hand rather than searched for, and why a round of alignments on a square reaches only nine new points.

It also explains where the extra power sits, in one line. Six of the seven have a solution count that is a power of two; only axiom 6 has three, and three is the whole of the difference between what a compass reaches and what a fold does.

Two folds, and what changes #

If one fold solves cubics, an obvious question is what two simultaneous folds do.

The answer is startling and slightly deflating. Allowing multiple simultaneous folds — several creases made at once, with alignments between all of them — reaches quintics and beyond, and in fact a construction with enough simultaneous folds solves polynomials of arbitrary degree. Alperin and Lang worked out the hierarchy.

It is deflating because simultaneous folding is not something a person does. One crease at a time is a physical description of folding; two creases made at exactly the same moment with a shared alignment condition is a mathematical device. The theory continues past the point where the paper does.

That boundary is worth marking, because it is where this subject stops being about folding and starts being about a formal system that happens to be inspired by it.

What the axioms are for #

A list of operations is only useful if something is built with it, and the everyday use is less glamorous than trisection.

Almost every folding sequence begins by manufacturing references. A blank square has four edges and four corners and nothing else, so the first creases exist to produce points and lines for later creases to align to. A diagonal is axiom 1 applied to two corners. A midline is axiom 2 applied to two corners. The crease that puts a corner on a midline is axiom 5.

Read a folding sequence that way and it stops being a list of instructions and becomes a construction: each step consumes references and produces new ones, and the model appears when enough of them exist.

That is also why an origami diagram has so many steps that appear to do nothing — folding and immediately unfolding. Those steps are producing references, and a reader who does not see them as constructions finds them arbitrary.

The axioms describe a person, not a machine #

There is something unusual about this axiom set, and it separates it from most of geometry.

Euclid’s postulates describe idealised instruments. The Huzita–Hatori axioms describe what a pair of hands can align by eye. Bringing a point onto a line is something a folder does by sliding the paper until the crease looks right, and the operation is on the list because it is physically natural, not because it is algebraically distinguished.

That the physically natural operations happen to constitute a clean algebraic system — closed, complete, exactly one degree more powerful than the classical one — is not something anybody designed. Huzita was cataloguing what folders already did.

Exactness, which is the practical payoff #

The axioms matter to a folder for a reason that has nothing to do with cubics: they are exact, and estimating is not.

Dividing a square into thirds by eye introduces an error of a per cent or so. That is invisible on its own and is not invisible after the sheet has been divided into thirds twice more and the resulting sixteenths have to line up. Errors in folding compound, because every later crease is aligned to an earlier one.

An exact construction has no error to compound. Haga’s theorem gives thirds, fifths and sevenths from a single fold apiece, with the only inaccuracy being the folder’s hands rather than the method’s. That is the difference between a design that closes at the end and one that does not.

What a fold cannot do #

The limitations are as clarifying as the powers.

A fold cannot draw a circle. The classical toolkit has a compass; folding has no operation that produces a curve. Every crease is straight, so every construction is an arrangement of lines and points, and the parabolas above are implicit rather than drawn.

A fold needs something to align to. Starting from a blank square, the only available references are its edges and corners. Every construction begins by manufacturing references, which is why so many origami constructions start with a diagonal or a midline that is not part of the final model.

Precision is physical. The axioms are exact and hands are not. A crease made by aligning two corners is accurate to whatever the paper and the folder permit, and errors compound — which is why box pleating on a grid is worth its inefficiency.

One fold at a time is a real constraint. Everything above assumes creases are made sequentially. Folding a base is not a sequence of independent creases but a collapse, and the axioms say nothing about whether the collapse is possible — that is a different question with different theorems, and the answer is not in this axiom set at all.

Who found it, and when #

The chronology is short and recent, which is surprising for a subject this old.

Margherita Piazzolla Beloch showed in 1936 that folding could solve cubics and extract cube roots — the earliest statement of the key result, largely unnoticed for decades. Jacques Justin listed the operations in 1989. Humiaki Huzita presented six axioms at the First International Meeting of Origami Science and Technology in 1991, and the set carried his name.

Koshiro Hatori found the seventh in 2001, and it turned out Justin had already had it. Robert Lang then proved the list complete, and also showed that Hatori’s operation, despite looking like a genuine addition, was the last.

Nobody was doing this in 1900. Paper folding as a subject with theorems is younger than quantum mechanics.

A fold is a reflection, and that is a strong statement #

Returning to the opening claim, because everything above depends on it and it deserves defending.

A fold maps the paper to itself by reflection across the crease. It is an isometry: distances within the sheet are unchanged, angles are unchanged, and nothing stretches. That is not an idealisation chosen for convenience — it is what paper does, to within a fraction of a per cent, and it is why a flat sheet can only become certain shapes.

Two consequences follow immediately and are used constantly.

Lengths along the paper are conserved. A path drawn on the sheet has the same length after folding as before. That is the conservation argument the entire design method rests on: a flap of length L consumes a circle of radius L, because every point that reaches the tip had to travel that far through the paper.

Angles at a point are conserved. The sectors round a vertex still sum to 360° however the sheet is folded, which is the first of the flat-folding conditions and is the reason it is a condition at all.

So the axioms are not a self-contained formal system that happens to resemble paper. They are consequences of one physical fact — that paper reflects and does not stretch — and every theorem downstream inherits it.

Where the model stops #

Zero-width creases. Every axiom treats a crease as a line. A real crease has a radius, and on thick paper the radius is a substantial fraction of a small construction.

Exact alignment. The axioms assume points can be brought into perfect coincidence. In practice the accuracy is a fraction of a millimetre and the errors compound with every fold.

An unbounded sheet. Several constructions need a fold whose crease leaves the square. The figures here clip to the sheet, which is honest about the paper and slightly dishonest about the geometry — the line continues, and the construction sometimes needs the part that is not there.

One fold at a time, and no unfolding. The axioms model a sequence of creases on a flat sheet. They say nothing about a sheet that is already folded, which is what every actual model is after the third step — and what makes collapsing a base a different subject.

The figures are schematic. The seven small squares show which alignment specifies each fold, not a construction carried out to scale. Only the trisection and the division figures are computed.

The ladder from here #

Later rungs: each axiom derived properly. The parabola formulation of axiom 6. Beloch’s fold and the cube root. The proof that there is no eighth axiom. Multi-fold constructions and the degree hierarchy. Constructible numbers, and the field-theoretic statement. The regular heptagon and the regular nonagon, both foldable and neither constructible. Squaring the circle, and why folding does not help. Accuracy in practice, and error propagation. And the question of what a curved fold would do to the axiom system, which nobody has answered.

Beloch published the cube-root result in 1936 in a journal nobody in the origami world read, and it was rediscovered independently more than once. The mathematics of paper folding has been found and lost at least three times.