A flap costs a circle

Until about 1990, designing an origami model was a craft learned by doing. A designer with a subject in mind would fold, look, unfold, adjust, and after enough iterations arrive at something. There was no method, and there was no way to answer the question “is this even possible from a square this size?”

Then somebody noticed that a flap costs a circle.

That observation is elementary, it is provable in two lines, and it converted a craft into an optimisation problem.

Why it is true #

The argument rests on folding being an isometry: the paper does not stretch, so distances measured along the sheet are unchanged by folding.

Consider a flap of length $L$ — a narrow point of paper sticking out of the finished model, with a tip. Every layer of paper at that tip came from somewhere on the flat sheet. Follow any one of them back from the tip to the point on the sheet where the flap attaches to the body, and the path has length $L$, because the flap is $L$ long and the paper did not stretch on the way.

So the tip’s material is drawn from points at distance $L$ from the attachment point. The flap’s full body is drawn from everything closer. Together, they are a disc of radius $L$.

And no other flap can use any of that paper, because it has already been used. Two flaps are different parts of the model; a given square millimetre of sheet is in one of them or the other.

$$\text{flaps do not overlap} \iff \text{their circles do not overlap.}$$

What follows immediately #

Three things, and all of them were unavailable before.

A design can be shown impossible. If the circles for the flaps a subject needs cannot be packed into the square, no crease pattern exists — not “nobody has found one”, but none exists. That is a genuine impossibility proof, and origami design had never had one.

The efficiency of a design is measurable. The fraction of the square covered by circles is how much of the paper is doing work. A design at 60% is using its sheet better than one at 40%, and this is now a number rather than an opinion.

The design problem has a shape. It is circle packing, which is a studied problem with a literature, algorithms and known hardness. Origami design inherited all of it overnight.

From a subject to a packing #

The method that results is Robert Lang’s tree method, and it runs in three steps.

Reduce the subject to a tree. An insect is a body with six legs, two antennae and an abdomen — a graph with lengths on its edges. Everything about the subject that is not “how long is each appendage” is discarded at this step, which is brutal and is what makes the method work.

Turn each leaf into a circle. A leg of length $L$ needs a disc of radius $L$. Internal edges of the tree — the body segments joining the limbs — become rivers: strips of a fixed width that must run between the corresponding circles without being crossed.

Pack. Fit the circles and rivers into the square without overlaps, as tightly as possible. The packing determines the crease pattern almost completely: the creases are the boundaries of the regions each flap claims, plus the internal structure that folds each region into a point.

The rivers matter more than they look #

The circles get the attention and the rivers are where the difficulty usually is.

A river is the paper that joins two flaps, and its width is the distance between them in the finished model. A beetle whose head and thorax are separated needs a river between the head circle and the thorax circles, and that river cannot be crossed by any other circle.

Rivers turn the packing from “fit discs in a square” into something with connectivity constraints, and they are why the problem is genuinely hard rather than merely fiddly. A packing that fits the circles perfectly and blocks a river is not a solution.

What the packing does not guarantee #

The circle argument gives a necessary condition and not a sufficient one, and the distinction is worth stating plainly.

Non-overlapping circles mean the paper is available. They do not mean a crease pattern exists that actually produces those flaps — the creases have to be constructible, the layers have to stack, and flat-foldability is its own problem with its own hardness.

In practice the packing almost always yields a pattern, because the theory comes with a construction: given a valid packing, there is a standard way to generate the creases. But “almost always” is doing work, and the guarantee is weaker than it first appears.

Efficiency, and an open problem #

How much of a square can circles claim? For a design with a handful of flaps, packing well is a real skill and packing optimally is unsolved.

Circle packing in a square is a classical problem, and the optimal arrangements are known only for small numbers of equal circles — and origami packings are of unequal circles with river constraints, which is harder again. There is no general algorithm that finds the best packing; TreeMaker uses numerical optimisation and finds good ones without any guarantee of optimality.

That has a practical consequence designers feel. Two designers given the same subject and the same square produce different patterns with different efficiencies, and neither can prove theirs is best. The theory says what is impossible and is largely silent on what is optimal.

The uniaxial assumption #

Everything above describes a particular kind of base, and the restriction is worth naming because it is substantial.

The tree method produces uniaxial bases: all the flaps lie along a single line in the folded form, like a stack of points radiating from a spine. Almost all complex insect and animal origami is uniaxial, and the method’s success is partly the success of that form.

A model that is not uniaxial — a box, a curved shell, a tessellation — is outside the method entirely. The circle argument still applies to any flap, but the packing-to-pattern construction assumes the uniaxial structure and does not generalise.

So the algorithm solved one large and popular family of design problems completely, and left the rest as it found them.

What was lost #

There is a case, made mostly by folders rather than mathematicians, that something went out of the subject when the algorithm came in.

A computed base is efficient and often unpleasant to fold: the creases meet at awkward angles, the layers are thick where the packing was tight, and the sequence has no rhythm. A base arrived at by hand tends to have symmetries and conveniences that an optimiser has no reason to find.

The counter-argument is that the algorithm expanded what was possible enormously — the insects with correct leg counts and segmented antennae that appeared in the 1990s were not being folded before, and were not going to be. Both things are true.

Box pleating is the practical compromise, and it is telling that it won: it gives up some of the packing efficiency in exchange for creases that land on grid lines, and most complex designers use it.

What the circles become #

A packing is not yet a crease pattern, and the step between them is where the method stops being an observation and starts being an algorithm.

Each circle’s centre becomes the tip of a flap. The boundary between two touching circles becomes a crease — an axial crease, running along the spine of the finished base. The paper in each circle folds into a point through a standard arrangement of creases that depends only on the circle’s radius, so once the packing is fixed the interior of each disc is determined.

What is left is the paper between the circles: the regions the rivers and the leftovers occupy. Those fold into the body, and their creases are found by a construction on the packing’s contact graph — which circles touch which, and along what lines.

The result is a base: a flat, many-pointed arrangement with one flap per circle, of the right lengths, ready to be shaped. Everything after that — thinning the legs, curving the antennae, giving the thing a face — is craft the method says nothing about, and is most of what makes a model good.

Why nobody saw it earlier #

The observation is elementary and the field is centuries old, so the gap wants explaining.

Part of it is that the question was never posed. Traditional folding worked from bases that already existed — the bird base, the frog base, the waterbomb — and a design was a variation on one of them. Nobody asked “what base does this subject need”, because bases were a fixed vocabulary rather than something a person made.

Part of it is that the question only becomes urgent at high complexity. A crane needs four flaps and any base with four points will do; there is no packing problem. A beetle with six legs, two antennae, mandibles and a segmented abdomen has fourteen appendages and the paper genuinely runs out. The pressure that produced the method came from folders trying to fold insects in the 1980s.

And part of it is that the argument requires thinking of the sheet as a metric space — a thing with distances on it — rather than as a surface to be manipulated. That is a mathematician’s move, and the people making it were mathematicians who folded.

Who found it, and when #

The circle-packing insight emerged in the late 1980s and early 1990s in several places at once — a pattern this field repeats.

Robert Lang developed the tree method and implemented it in TreeMaker from 1990 onward. Toshiyuki Meguro in Japan developed a closely related circle-packing approach independently and at about the same time; the Japanese term is enchōhō, and Meguro’s work is less known in English for the usual reasons. Jun Maekawa and Fumiaki Kawahata contributed related methods.

The simultaneity is not a coincidence. Complex origami was hitting a ceiling that a systematic method could break through, and several people were pushing at it.

Rivers, drawn #

The circles are easy to picture and the rivers are not, so it is worth being concrete about what a river actually is on the sheet.

Take a simple subject: a body with two legs at one end and a head at the other. The legs and head each get a circle. The body is an edge of the tree rather than a leaf, and it has a length — the distance from the leg junction to the head — which becomes a required separation.

On the sheet, that separation is a strip of paper of constant width running between the head circle and the leg circles. The strip is not consumed by any flap; it becomes the body, and it must not be crossed by another circle, because a circle crossing it would be claiming paper the body needs.

So the packing problem is: place discs of given radii, and routes of given widths between specified pairs of them, inside a square, without overlaps. That is a good deal harder than packing discs, and it is why designs with long thin bodies are difficult in a way that designs with many legs are not.

The argument runs backwards too #

The conservation statement is usually used to show a design is impossible. It can also be read the other way, and doing so gives a designer a bound worth having.

Given a square of side 1 and a subject needing flaps of lengths $L_1, \ldots, L_n$, the circles require total area $\pi \sum L_i^2$, which must be at most 1. So

$$\sum_i L_i^2 \le \frac{1}{\pi}.$$

That is a hard bound before any packing is attempted, and it is startlingly restrictive. Six legs of length $0.2$ alone need $6 \times 0.04 \times \pi = 0.75$ of the sheet, leaving very little for a head, an abdomen and the body that joins them.

It also explains why complex insect designs use such large squares relative to the finished model. The flaps are short compared with the sheet not because the designer wanted them so, but because the sum of squares leaves no choice.

Uniaxial bases only. The construction from packing to crease pattern assumes the flaps lie along one axis, so a tessellation is outside it. Boxes, curved forms and tessellations are outside it.

Necessary, not sufficient. Non-overlapping circles say the paper is available. They do not guarantee a foldable pattern.

Idealised flaps. A flap is treated as a line segment of zero width. Real flaps have thickness, and a leg with several layers is fatter than the model says.

No layer accounting. The packing says nothing about how many layers end up at any point. A tight packing can produce a region twenty sheets thick, which is a real limit rather than a theoretical one.

The figure shows a symmetric packing. The one drawn here is solved geometrically — corner circles plus a middle one, sized so nothing overlaps — which makes it exact and unrepresentative. Real packings are irregular and found numerically.

The argument in one line #

Worth compressing, because the whole design method reduces to it and the compression is memorable.

Paper does not stretch, so a flap of length L is made of paper that was within L of where it attaches.

Everything else is bookkeeping. Two flaps cannot share paper, so their discs cannot overlap. The discs must fit in the square, so the sum of their areas is bounded. A body segment separating two flaps needs a strip of paper of its own width, so the discs must be routed around.

That is a complete design theory in four sentences, and it replaced several centuries of trial and adjustment. The reason it took until 1990 is not that it is difficult but that it requires seeing the sheet as a space with distances on it rather than as a thing to be manipulated — which is a mathematician’s habit rather than a folder’s.

The ladder from here #

Later rungs: the isometry argument in full. Rivers, and the connectivity constraints they impose. The packing-to-crease-pattern construction. Uniaxial bases defined properly. TreeMaker and its optimisation. Meguro’s method and the parallel development. Efficiency bounds and the open problems. Non-uniaxial design, which has no comparable theory. Box pleating as a deliberate retreat from optimality. And the question of whether an algorithm can produce a base anybody enjoys folding.

The whole method rests on one sentence — the paper does not stretch, so a flap of length L uses a disc of radius L — which any folder could have observed at any point in the previous four centuries.