Why the list stops at seven

A list of seven looks like a list somebody made. Seven operations, catalogued by Humiaki Huzita in 1991, with a seventh added by Koshiro Hatori a decade later — the shape of the story is a collection growing as people notice things, and it invites the obvious question of what turns up next.

Nothing turns up next. The list is not a collection; it is an enumeration, and it is complete for a reason that fits on a postcard.

A fold line has two degrees of freedom #

Start with the thing being specified rather than with the ways of specifying it.

A fold is a reflection of the sheet in a line, so the fold is exactly as determined as the line is. A line in the plane takes two numbers — a direction and an offset, or an intercept and a slope, or any other pair. Two numbers, so two constraints, and a construction that supplies fewer leaves a family of folds while one that supplies more usually has no solution at all.

That is the whole of the counting argument, and everything below is bookkeeping.

It is worth pausing on how unusual that is. Most of the results on this site are about crease patterns with many creases, where the interesting questions are combinatorial and the answers are hard — whether a pattern folds flat at all is not even decidable in polynomial time. Here the object has two degrees of freedom and the whole theory is a paragraph. The single fold is the one place in the subject where completeness is cheap.

What an alignment costs #

The other half is a list of the constraints available, and here the list really is short, because a fold can only be told about things that already exist on the sheet.

A point can be asked to land on another point. That fixes both coordinates of its image, so it is worth two.

A line can be asked to land on another line. Two again.

A point can be asked to land somewhere on a line. Its image is then free to slide along that line, so only one condition has been imposed: worth one.

A point can be asked to stay where it is, which is the same as saying the fold passes through it. Worth one.

A line can be asked to stay where it is, which is the same as saying the fold is square to it. Worth one.

There is no sixth kind. An alignment is an incidence between something and the image of something, and the only objects in play are points and lines.

The enumeration #

Now combine to a total of two, and count.

One two-constraint alignment on its own gives two combinations, because there are two of that kind: point onto point, and line onto line. Those are axioms 2 and 3.

Two one-constraint alignments give six, because there are three of that kind and they may be paired with repetition. Through a point and through a point is axiom 1. Through a point and square to a line is axiom 4. Through a point and a point onto a line is axiom 5. Square to a line and a point onto a line is axiom 7. A point onto a line, twice over, is axiom 6.

That is five of the six pairs. The sixth is square to a line and square to a line — a fold at right angles to two different lines at once — and it is the one that does not work.

Eight combinations, seven folds. The count is closed because the ingredients are, and no cleverness can add a ninth to a list whose length was fixed by the number of ways to write two as a sum of ones and twos.

The one that fails, and why it keeps coming back #

The failing combination deserves more than a shrug, because it is not obviously absurd and it has been proposed as an eighth axiom more than once.

Asking a fold to be square to two lines is not meaningless. If the two lines happen to be parallel, plenty of folds satisfy it — every fold perpendicular to their common direction — and if they are not parallel, none does. So the operation returns either an infinite family or nothing, and never a fold.

That is precisely what it means for a specification to fail to determine its object. The other seven return a small, finite set: one fold, or two, or at most three. This one returns a set whose size depends on the configuration rather than on the operation, which is a different kind of thing entirely.

Jorge Lucero’s 2017 analysis makes the same point from the other end and adds the case that is genuinely missing from most retellings: folding a line onto itself, which is a legitimate operation and is the degenerate member of axiom 3 rather than a new one. The list absorbs it; it does not grow.

The count in closed form #

The enumeration is short enough to write as a formula, which is worth doing because the formula says what would have to change for the list to have a different length.

With $q$ alignment types costing two and $p$ costing one, the combinations reaching a total of two are the $q$ single two-cost alignments plus the unordered pairs, with repetition, of the one-cost ones:

$$q + \frac{p(p+1)}{2}.$$

Here $q = 2$ and $p = 3$, giving $2 + 6 = 8$. Seven is $8 - 1$, and the $-1$ is one specific combination failing rather than a fact about the number seven.

The inputs are what a fold can be told about. The sheet carries points and lines and nothing else, an alignment is an incidence between an object and the image of an object, and the five that exist are all the table holds. A sheet that carried circles would have a longer list — which is the honest form of the observation that a compass reaches further in some directions and less far in others.

And the axiom the whole subject rests on #

Reading the seven by how many folds each returns puts the list in a different order, and the order has a consequence.

Axioms 1 to 4 and 7 return one fold apiece: each is a linear condition. Axiom 5 returns up to two, being a tangent to a parabola through a given point. Axiom 6 returns up to three, being a common tangent to two parabolas, and three is the largest number anywhere in the list.

That maximum is not a curiosity. The degree of the equation an operation solves is what it adds to the field of constructible lengths: quadratics give the powers of two that straightedge and compass already reach, and only axiom 6 contributes a cubic. Origami’s constructible numbers have degree $2^a 3^b$, and every factor of three in that expression comes from one operation.

So the list has a single load-bearing member. Delete axiom 6 and the remaining six construct exactly what a compass constructs — no trisection, no doubled cube — while deleting any other leaves the reach unchanged. A list of seven with one indispensable entry is a better description of the subject than a list of seven.

Which theorem was checked, and how #

The figure at the head of this essay is generated by enumeration rather than drawn from the published list, which matters because the two could disagree.

The generator declares the alignments and their costs, forms every unordered pair of the one-cost kind together with every single two-cost kind, and then looks each combination up in a table of verdicts. If a combination comes out with no verdict, the generator throws; if the number of combinations that determine a fold is anything other than seven, it throws. So the picture cannot quietly drift out of agreement with the argument beside it — a change to either one breaks the build.

What the figure cannot show is the proof that the five alignment types are all of them. That is an argument about what an incidence between a point, a line and a reflection can be, and it is a sentence rather than a picture: the objects are points and lines, the relation is incidence, and the enumeration of incidences between two kinds of object is a two-by-two table with nothing outside it.

The idealisation underneath #

Everything above assumes a fold is an exact reflection in an exact line, which is the idealisation this whole field is built on and which is one of four that are not true.

The consequence here is specific rather than general. Axiom 6 places two points onto two lines simultaneously, and “simultaneously” is doing real work: the fold has to be positioned so that both incidences hold at the same instant, and a fold made by hand achieves that by sliding the paper until both look right. What the eye is doing is solving a cubic by successive approximation, and it converges because the residual is smooth. The idealised statement is exact; the physical procedure is a numerical method with a human in the loop.

The sixth is the one that matters #

The enumeration says seven exist. It says nothing about their being equally interesting, and they are not.

Six of the seven can be reproduced with straightedge and compass. Perpendicular bisectors, angle bisectors, perpendiculars through a point — these are the first constructions in Euclid, and folding does them more conveniently rather than more powerfully. Convenience is not nothing: a fold carries its own straightedge and needs no marking, which is why exact division by folding accumulates no error where a ruler does. But convenience is not new mathematics.

The sixth cannot. Placing two points onto two lines at once is a common tangent to two conics, which is a cubic condition, and a compass never leaves the quadratics. That single step is the entire difference between what a fold can build and what Euclid’s tools can, and everything downstream of it — trisection, the cube root of two, the regular heptagon — comes out of that one line of the enumeration.

There is a pleasing irony in it. The completeness argument is the reason nobody should expect an eighth axiom, and the same argument shows that the interesting content of the list is one entry.

How many folds each operation returns #

The enumeration says which alignments determine a fold. It is silent on how many folds each one determines, and the counts are not all one.

Axioms 1, 2, 4 and 7 return exactly one. Axiom 3 returns two, because a pair of crossing lines has two bisectors — a fact that surfaces in practice as the choice between two ways of bringing an edge onto another. Axiom 5 returns up to two, since it is a line meeting a parabola. Axiom 6 returns up to three, because it is a line touching two parabolas.

The pattern is exactly the degree of the equation each operation solves, which is the sense in which the axioms are algebra rather than geometry. One solution for a linear condition, two for a quadratic, three for a cubic. A construction that needs a particular one of the three has to say which, and the trisection is a case in point: two of its three roots place the marked point on the wrong half of the ray, and the generator that draws it selects by testing which side the image lands on rather than by comparing the answer with the one expected.

That distinction — solving the equation against solving the construction — is where most retellings of the axioms go quietly wrong. The equation has three roots; the paper has one fold that does what was asked.

Where the model stops #

One fold at a time. The enumeration counts the ways to specify a fold. A construction that makes several folds simultaneously — a multifold — is a different object with more degrees of freedom, and the enumeration says nothing about it.

Alignments of existing objects only. A fold specified by, say, an angle measured with a protractor is not in the list, because a protractor is not paper. The list describes what a sheet can do to itself.

Existence, not construction. Axioms 5 and 6 can return more than one fold, and in the case of axiom 6 as many as three. The axiom asserts that such folds exist and gives no procedure for finding them. Solving for one is a numerical problem, and every figure on this site that uses axiom 6 solves it by root-finding rather than by construction.

Nothing about sequences. A single fold is completely characterised. What a sequence of folds can reach is a much harder question, and the answer — the field of origami numbers — is the subject of the constructibility results rather than of the axiom list. Dividing a square into any number of equal parts is a sequence result, not an axiom, and it needs the axioms only in the way that arithmetic needs addition.

Nothing about the crease that results. The axioms specify where a fold line goes. They say nothing about whether the sheet can actually be brought into that position without passing through itself, which is a question about layers and is the one the vertex conditions begin to answer. A single fold on a flat sheet is always physically realisable, so the gap does not bite here; it opens the moment there are two creases.

What a second simultaneous fold buys #

The most interesting thing the enumeration leaves open is what happens when the “one fold at a time” restriction is lifted, and the answer is surprisingly large.

A multifold makes several creases at once, all in a single motion, with the alignments imposed on the final state. Two simultaneous folds have four degrees of freedom rather than two, so four constraints can be imposed, and the resulting conditions are of higher degree.

Roger Alperin and Robert Lang showed in 2009 that the n-fold operations reach every polynomial of degree n + 2, so allowing two folds at once solves quartics, three solves quintics, and the tower does not stop. Every algebraic number becomes constructible if enough creases are allowed to be made in one motion.

That result puts the seven axioms in their place rather neatly. They are not the theory of folding; they are the theory of folding once, and the restriction to one is what makes the list finite. Lift it and the list becomes an infinite hierarchy whose first non-trivial level is the one everybody works in.

It also explains why the restriction is worth keeping. A multifold is not something a person does; it is something a person approximates, by making several creases and then collapsing them together while hoping the alignments come out. Every practical construction in the literature stays at one fold at a time precisely because one fold at a time is what hands can do reliably — the same reason designing on a grid beats designing on a free packing once a model has hundreds of creases. The mathematics is happy with the hierarchy; the paper is not.

Who found it, and when #

The history is unusually tidy, and it is worth stating because the list is often attributed to one person.

Margherita Piazzolla Beloch established the essential result in 1936, decades before anybody wrote an axiom list. She showed that a single fold placing two points onto two lines solves a general cubic — the operation now numbered six — and used it to double the cube. Her work sat almost unread for fifty years.

Humiaki Huzita presented six operations at the First International Meeting of Origami Science and Technology in 1991, and Jacques Justin published an equivalent set the same year, which is why the list is sometimes called Huzita–Justin. Koshiro Hatori added the seventh in 2001. Robert Lang proved the list complete shortly afterwards, by exactly the enumeration above.

So the sequence runs backwards from the usual telling. The powerful operation was found first and forgotten; the catalogue came half a century later; the proof that the catalogue was finished came later still.

The ladder from here #

Later rungs on this anchor: the completeness proof stated properly, with the incidence table it rests on. Beloch’s fold and Lill’s method, which is where the cubic really comes from. The multifold hierarchy and Alperin and Lang’s theorem. What a sequence of folds reaches, which is the field of origami numbers. The operations that are not alignments — cutting, marking, using a reference — and why each one is excluded. And the question of what an axiom list would look like for folding on a sphere or on a cone, where the sheet is not flat and the reflections are not reflections.

Seven is a small number, arrived at by counting to two. The surprise is not that the list is short but that anybody had to look for it: the enumeration was available from the moment somebody wrote down that a fold is a reflection, and it took sixty-five years.