The sculptors got there first

Nearly everything on this site follows the same order: a theorem, then a figure, then something a reader can fold. Curved creases invert it. The objects came first, they were made by people who could not have computed them, and the mathematics that describes them arrived forty years later.

That inversion is worth an essay, because it is the one part of the subject where the craft was demonstrably ahead.

The Bauhaus exercise #

The story starts in a preliminary course rather than in a laboratory.

At the Bauhaus in the 1920s, Josef Albers set his students a paper exercise: take a disc, score it with concentric circles, and fold along them. What comes out is a saddle-shaped ring that stands up on its own, holds its shape without glue, and looks nothing like the flat disc it came from.

The exercise was about material and about form, not about geometry. Albers’s point was that paper has a nature and that working with it teaches something no drawing can. Whether anybody in the room could have said why the result was a saddle is doubtful.

Variations of the same exercise have been set ever since, and the object it produces is one of the few things in this subject that a complete beginner can make in five minutes and that a specialist still finds interesting.

What the paper is actually doing #

The reason the disc becomes a saddle is a consequence of the sheet refusing to stretch, and it is worth working through because it explains the whole family.

A crease that is curved does something a straight crease does not. A straight crease leaves the paper on either side flat, free to be anything. A curved one does not: the paper next to it is forced into a specific ruled surface, and the ruling directions are determined by the crease’s curvature and the fold angle together.

The paper cannot stretch, so it can only take shapes with zero Gaussian curvature — cylinders, cones and the general developable. The curved crease picks out which developable, and the band between two curved creases has no freedom left.

So a curved crease does not make a shape. It forces one, and the surface either side of it is an output rather than a choice.

The rulings are the content #

That brings the account to the object which took forty years to appear in a drawing.

A developable surface is ruled: through every point there is a straight line lying entirely in the surface. On a cone the rulings are the generators through the apex; on a cylinder they are the parallel lines along it; on a general developable they twist.

For a curved fold, the rulings on the two sides are what the crease determines, and they carry all the information. Two curved folds with the same crease curve and different fold angles have different rulings and are different surfaces.

Nobody drew them for the classic pieces. The sculptures were made, they were photographed, and the ruling fields that describe them were not computed until the 2000s. That is the sense in which the sculptors got there first: they produced the surfaces without ever having access to the object that specifies them.

Why the disc has to buckle #

The construction in these figures carries its own explanation of the Albers exercise, and it is worth reading out, because the exercise’s result is the one thing about curved folding that everybody has seen and nobody derives.

A cone of half-angle $\beta$ has, at slant distance $r$ from its apex, a circle of radius $r\sin\beta$ and therefore a circumference of $2\pi r \sin\beta$. The flat sheet at radius $r$ has a circumference of $2\pi r$. Since folding cannot stretch or compress the paper, the flat material at that radius is $1/\sin\beta$ times what one turn of the cone will hold.

So a full disc has too much paper in it. The cone map is injective only on a sector of angular width $2\pi \sin\beta$, and a whole disc exceeds that by $2\pi(1 - \sin\beta)$ — at the $\beta = 62°$ of the figure above, a sector of 318° fits and the remaining 42° has nowhere to go.

The excess cannot be absorbed by folding harder, because $\beta$ is what folding harder changes and every value of it leaves a deficit except $\beta = 90°$, which is the unfolded sheet. It cannot be absorbed by stretching, because the sheet will not stretch. So the surface has to leave the cone family altogether, and a surface carrying more circumference at its rim than a cone can hold is a saddle.

That is the Bauhaus result, derived from the isometry rather than observed: a creased disc becomes a saddle because a disc is exactly $2\pi(1-\sin\beta)$ too big to be a cone.

And why the figures are sectors #

It also explains a choice in these drawings that would otherwise look like caution. The construction here is exact, and it is exact because it is applied to a sector narrow enough to fit — which is the same statement as the self-intersection caveat below, arrived at from the arithmetic instead of from a failed render.

The prediction that comes with it is one a reader can test with a compass and five minutes. The deeper the fold, the greater the excess, and the wavier the ring. A shallow crease leaves $\beta$ near $90°$ and almost no surplus, so the disc stays nearly flat; a hard crease drives $\beta$ down, the surplus grows toward a full turn, and the rim gathers into more and deeper lobes.

What the isometry does not fix is how many lobes. The surplus has to go somewhere and the geometry does not say where, so the material decides by minimising its bending energy — which is why two people folding the same curved pattern at the same angle can get rings with different numbers of waves, and why a single curved crease is much better behaved than a closed ring of them.

Which theorem was checked, and how #

The surface in these figures is an exact isometry, and the check is on the metric rather than on distances between points.

The construction maps a flat annulus onto a cone: a point at radius r and angle θ goes to the point at slant r and angle θ/sin β on a cone of half-angle β. Bands are stacked by alternating β with its supplement, which is the condition for two cones to join along a circle.

The verification computes the first fundamental form numerically. Moving along r must cover ground at unit rate; moving along θ must cover it at rate r; the two directions must stay perpendicular. All three are exactly what they are on the flat sheet, and the generator throws if any is off by more than a part in a hundred thousand.

Comparing chords does not work, and the first version tried it. A straight line in space between two points of a cone is shorter than the arc joining them, so measuring the distance between sampled points reports a stretch of about a thousandth on a map that is exact. Isometry is a statement about distances along the surface, and the metric is the right thing to compare.

What the figures cannot show is a general curved fold. The construction here is exact because concentric circles map to cones and cones are the easy case; a crease of arbitrary curvature produces a developable whose rulings must be solved for, and this site does not solve them.

Where the theorems fall silent #

It is worth being blunt about how little of this subject the flat-folding theory reaches.

Kawasaki and Maekawa are statements about straight creases meeting at a point. A curved crease has no vertex and no sectors; there is nothing for either condition to be evaluated on. The big-little-big lemma is likewise about angles at a vertex.

That is not a matter of the theorems being hard to extend. They are about a different object. A curved-crease piece has no flat-folded state at all — the Albers disc does not fold flat and never will — so the entire question the theorems answer does not arise.

What replaces them is differential geometry, and the replacement is much weaker. There are necessary conditions relating the crease’s curvature in the flat sheet to its curvature in space and the fold angle. There is no analogue of “these conditions are jointly sufficient”, and there is no algorithm that takes a curved crease pattern and decides whether it can be folded.

So the best-developed part of this subject and its oldest hand-made objects have almost no theory in common.

Huffman, who did both #

The exception to the separation is one person, and his work is the reason the subject exists as a subject.

David Huffman is known outside folding for the coding algorithm he devised as a graduate student in 1952. From the 1970s until his death in 1999 he made curved-crease sculptures — hundreds of them, in vinyl and paper — and worked out much of the mathematics of curved folding along the way.

He published almost none of it. A single paper in 1976 sets out the relation between the ruling directions and the crease’s curvature, and the rest existed as notes, models and an unpublished manuscript. The community that could have used it did not know it was there.

The reconstruction of his work — by Erik and Martin Demaine, Duks Koschitz and others, from the 2000s onward — involved measuring his physical models and re-deriving the geometry. Several of his pieces have been folded again from the recovered patterns.

That is an unusual position for a field to be in: its most productive practitioner was also its best theorist, and the two halves were lost together and had to be recovered by measuring the objects.

Reconstructing a sculpture #

The recovery of Huffman’s work is a piece of methodology worth describing, because it is the reverse of how this subject usually goes.

The starting point is a physical object: a folded piece of vinyl, sitting in a collection, with no pattern and no notes. The crease curves are visible on it, and their positions can be measured — but they are measured in the folded state, and what is wanted is where they were on the flat sheet.

Recovering that means unfolding the object mathematically. Each band of the surface is developable, so it can be flattened exactly; flattening every band and reassembling them along the shared creases reconstructs the flat pattern. The reconstruction is only as good as the measurement of the surface, and the surfaces are not exactly developable because the material has thickness and has relaxed.

Several of the pieces have been re-folded from patterns recovered this way, which is the only real test: if the pattern is right, the fold reproduces the object.

That is an archaeology rather than a derivation, and it is what happens when the theory is lost with its author. It is also a good argument for the discipline this site tries to keep — a pattern generated from code is a pattern that survives, and a pattern that exists only as a folded object needs somebody with a scanner and a fortnight.

The material is part of the design #

Curved-crease work is unusually sensitive to what it is made of, and the sensitivity is geometric rather than aesthetic.

A curved crease is not a line in any real material. It is a strip of high curvature whose width is set by the material’s thickness and stiffness, and everything about the resulting surface is affected by it. Thin paper gives a crease close to the ideal and a piece that is floppy; thick vinyl gives a broad crease and a piece that holds itself up.

Huffman worked mostly in vinyl for exactly that reason. The pieces are large, they are self-supporting, and the crease radius is visible and deliberate. He also worked in paper, and the paper versions of the same designs are different objects.

There is a second effect. Where the isometry leaves freedom — and for a curved fold it often does, because the fold angle is not pinned down at every point — the material chooses, by minimising its bending energy. So the finished surface is the one that satisfies the geometry and is easiest for the sheet to be in, and the second condition is not in any of the mathematics above.

That is why two people folding the same curved pattern get slightly different objects, and it is the third idealisation doing more work here than anywhere else on this site.

Where the model stops #

Cones only, here. The exact isometry above works because concentric circular creases give cones. A crease of varying curvature gives a developable whose rulings twist, and computing it is a differential equation rather than a formula.

No self-intersection check. The surface is drawn from the isometry. Whether the bands actually clear one another as the piece closes is not tested, and for tightly nested arcs they do not.

Zero thickness, and it shows. A curved crease in real paper is not a line; it is a region of high curvature with a radius set by the material. On a tight curve that region is a substantial fraction of the band, and the folded shape departs visibly from the ideal.

The theorems do not apply. Not “apply weakly” — the flat-folding conditions are statements about straight creases at a vertex and there is neither here.

No elasticity. The surface here is the one the isometry allows. A real sheet also minimises bending energy, and where the isometry leaves freedom the material chooses — which is why two people folding the same curved pattern get slightly different objects.

Nothing about how to fold it. A curved crease has to be formed gradually and all at once, since the surface either side is forced by it. There is no folding sequence in the usual sense, and this is why curved work is difficult in a way straight work is not.

No decision procedure. There is no algorithm that takes a curved crease pattern and reports whether it folds. The straight-crease case is at least decidable in principle, even where it is intractable; here the question is open.

The surprise: it is stiffer than it should be #

There is a mechanical fact about curved-crease pieces that is obvious the moment one is picked up and is rarely stated.

An Albers ring made from thin paper is remarkably rigid. It resists being flattened, it holds its shape under its own weight, and it springs back. A flat disc of the same paper does none of those things.

The reason is that flattening it would require the surface to stop being developable somewhere — the curved creases hold the ruling directions, and changing the overall shape means changing the rulings, which means stretching. So the structure is stiff because deforming it is not merely resisted but geometrically excluded.

That is a different mechanism from the stiffness of a Miura-folded sheet, which is a mechanism with a soft mode and is compliant in one direction. Curved-crease structures have no soft mode at all, and the interest in them for architecture and for deployables follows directly.

It also explains the Bauhaus exercise’s staying power. A student who makes one has produced a self-supporting structure from a flat sheet and a pair of scissors, and the reason it stands up is a theorem about curvature that nobody in the room needs to have heard of.

Where it is going #

The subject is unusually active for one whose canonical objects are fifty years old, and the reason is fabrication.

A curved crease is difficult to make by hand and straightforward to make by machine. A scoring head on a plotter, a rolling tool on sheet metal, a laser at low power on acrylic — all of these produce a controlled crease along an arbitrary curve, repeatably, at whatever scale the machine spans. What was a craft technique requiring years of practice became a manufacturing operation.

That has opened two directions. One is architectural: curved-crease panels are self-supporting, stiff for the reason described above, and can be made from flat stock with no forming dies. The other is computational: with fabrication cheap, the bottleneck moves to design, and design needs the ruling solver that does not yet exist in general.

So the field is in an unusual position. It has better tools than theory, which is the opposite of where it was in 1975 and the opposite of where most of this subject is now.

The ladder from here #

Later rungs against this anchor: the ruling equations for a general curved fold, and what they look like as a differential equation. Huffman’s 1976 relation, derived. The reconstruction of his sculptures, and what measuring a physical model can and cannot recover. Curved-crease tessellations, which exist and are barely studied. The stiffness argument made quantitative. Fabrication at architectural scale, where the sheet is metal and the crease is a rolled line. And the decidability question — whether there is any algorithm that takes a curved crease pattern and says whether it folds — which is open and does not look close.

Every other part of this subject has the theory in front and the objects behind. This one has it the other way round, and the objects are better.