What a flat sheet can become

Wrap a sheet of paper round a bottle and it fits. Wrap one round a football and it does not — there are creases, or wrinkles, or a fight.

The distinction is exact, it has a name, and it is the constraint underneath everything else on this site.

The constraint, and whose result it is #

The rule is a consequence of one theorem, and this site does not re-derive it.

A sheet that cannot stretch preserves every distance measured within it. Gauss’s Theorema Egregium says that one particular measure of curvature — the product of the two principal curvatures at a point, the Gaussian curvature — can be computed from those internal distances alone. Put the two together and the conclusion is immediate: bending cannot change Gaussian curvature, a flat sheet has none, so anything it becomes without stretching has none either.

The surfaces with none everywhere are the developables, and their classification — cylinders, cones, tangent developables and combinations of the three — is the other half of the standard account.

Both halves belong to cartographic projection, where the theorem is used to close the map question permanently and the classification is used to show that calling a projection cylindrical does not make it faithful. That site derives the theorem two ways and states the classification properly; there is no reason for this one to do it again.

What this essay is about is the half that is folding’s own: given that the restriction holds, what a sheet does about it.

What it forbids #

A sphere has $K = 1/R^2$ everywhere, which is positive and not zero. A saddle has $K < 0$. Neither is reachable from a flat sheet.

It is also why sheet metal for a car body is pressed rather than bent. Pressing stretches the material, which changes the curvature, and that is exactly the operation bending cannot perform.

Folding has three answers to the restriction and they are the subject of this essay. Cut the surface into pieces each of which is nearly flat. Put the curvature into the crease rather than into the material. Or stretch the sheet a little and accept the consequences. Each pays for the same thing with a different currency.

The first answer: pay in seams #

Cut the surface into strips narrow enough that flattening each one costs almost nothing, and join them. That is what a paper globe is: a set of gores, the pointed almond shapes a globe’s printed surface is divided into, each of which is nearly developable on its own.

The arithmetic of how well it works is short. A surface of curvature K, developed along a central line, is out by about K u² / 6 at transverse distance u; a gore of an n-gore sphere is at most π/n wide either side of its meridian; so the worst strain left in a strip is about π²/6n². Two gores need forty percent, four need ten, eight need two and a half, twelve need one.

That is why twelve is the number on a globe. It is where the residual strain drops below what paper absorbs without complaint, and going further buys accuracy nobody notices at the cost of seams everybody sees.

The cost is the seam, and on this site the seam is a serious cost. A gore construction is a cut — the one operation the rest of the subject forbids — so a folded sheet that reaches a sphere by gores has stopped being one sheet. The technique belongs to sail-making, to balloon construction and to panelled architecture, and it is the answer for everybody who is allowed to cut.

The second answer: put the curvature in the crease #

The answer folding actually took is subtler and costs nothing at all.

A crease can be curved. The paper either side of a curved crease cannot stretch, so it is forced into a developable surface — but the two developables meeting along a curve produce a shape that reads as doubly curved from any distance. The material is flat everywhere and the object is not.

That is the whole of the curved-crease tradition, and the objects it produces are the best answer this subject has to “make something round out of a flat sheet”. The Bauhaus disc exercise is a ring of concentric creases that arrives at a saddle; Huffman’s sculptures are elaborations of the same idea.

The construction can be made exact rather than approximate. A flat annulus maps onto a cone without stretching, cones join along a circle when their half-angles are supplementary, and a set of concentric arcs therefore produces a surface whose every distance matches the flat sheet’s — which the generator below checks on the metric rather than asserting.

So this answer pays nothing. No cut, no strain, no approximation. What it gives up is control: the surface either side of a curved crease is forced, so a designer chooses the crease and accepts the shape, and solving the inverse problem is a piece of mathematics that does not yet exist in general.

How folding gets around it #

Folding does not violate the theorem; it exploits a loophole in how the curvature is distributed.

A crease is a place where the surface is not smooth, and Gaussian curvature is defined for smooth surfaces. At a crease the curvature is concentrated rather than absent — it is a distribution, not a function.

For a straight crease the concentrated curvature is still zero, and the surface either side stays developable. But at a vertex — where several creases meet — curvature can concentrate at the point, and the total concentrated there is $2\pi$ minus the sum of the angles around it.

Which is exactly the developability condition this site checks on every pattern: the angles must sum to $2\pi$, so that the concentrated curvature is zero and the vertex came from a flat sheet.

So folding lets a sheet reach shapes that are piecewise developable — flat pieces joined along creases — and the whole of origami lives in that space. What it cannot do is make a smoothly curved sphere.

Approximating what cannot be reached #

A great deal of practical folding is approximating a non-developable shape with a developable one, and the approximations are the interesting part.

Facets. Many small flat panels approximate a curved surface, and the approximation improves with the panel count. This is what a geodesic dome does and what a low-polygon model does.

Gores. Long thin strips, each nearly developable, joined along their edges. Globes, balloons, hot-air balloons and parachutes.

Corrugation. A tessellation can absorb the curvature mismatch: a corrugated sheet can be curved in a second direction because the corrugation opens or closes to accommodate it. This is why a folded sheet can take shapes a flat one cannot.

That last one is worth stating clearly, because it looks like a contradiction. A corrugated sheet is not doubly curved as a surface; it is a flat sheet arranged so that its coarse-grained shape appears doubly curved. The material is still developable everywhere.

The third answer: pay in strain #

The remaining option is to break the assumption, and it is the only one of the three that genuinely does.

Wet-folding dampens the paper so the fibres can move relative to one another, shapes it, and lets it dry. Distances within the surface change, the map stops being an isometry, and the theorem stops applying — so genuinely doubly-curved shapes become available.

How much of a sphere that buys is a calculation rather than an argument, and the number is about twenty degrees. A cap of angular radius α needs its rim to shrink by 1 − sin α / α, which is one percent at fourteen degrees, three at twenty-four, six at thirty-five, and thirty-six for a hemisphere. Dry paper supplies about one percent and damp paper a few more.

So the third answer buys a genuine but small piece of the forbidden territory. Akira Yoshizawa developed the technique and it is why his animals look like animals rather than like polyhedra; what it does not do is get anywhere near a sphere.

Two of the three trade against each other #

The answers are presented as alternatives and two of them are not: seams and strain buy the same thing, and the exchange rate is available from the numbers already given.

A gore construction leaves about $\pi^2/6n^2$ of residual strain in each strip. If the material will absorb $s$ of strain, the seams only have to reduce it to that, so the gore count required is

$$n \;\geq\; \frac{\pi}{\sqrt{6s}}.$$

Seams and stretch are therefore one currency at an exchange rate of $n \propto 1/\sqrt{s}$, and the square root is what makes the trade worth so little at the easy end and so much at the hard one.

Put the essay’s own two numbers in. Dry paper supplies about 1% of strain, which needs $n \geq \pi/\sqrt{0.06} = 12.8$ — thirteen gores, against the twelve everybody actually uses. Damp paper supplies about 3%, giving $n \geq 7.4$: eight gores. At 6%, six.

So the answer to “why does a globe have twelve gores” is not a printing convention or an aesthetic. It is paper’s own strain tolerance read through $\pi/\sqrt{6s}$, and the traditional number sits one seam below what the arithmetic asks — which is what a real material’s tolerance being a little over 1% would produce.

And it says what wet-folding is worth in the currency somebody making a globe cares about: damp paper halves the seam count, because tripling the strain budget divides the gore count by $\sqrt{3}$.

The second answer sits off this curve entirely. A curved crease pays neither seams nor strain — it pays in the inverse problem, and no amount of either of the other two buys any of it back.

Three answers, three currencies: a cut, a loss of control, or a few percent of stretch. Nothing else is on offer, because the restriction is not a difficulty to be engineered around but a consequence of the sheet’s metric.

Where the loophole is #

Given how strong the constraint is, it is worth being precise about how folding escapes it, because the escape is narrow.

Folding does not add curvature. Every point of a folded sheet away from a crease still has zero Gaussian curvature, and every crease is a place where the surface is not smooth so the curvature is undefined rather than nonzero.

What folding adds is concentration. Curvature can accumulate at isolated points — vertices — where the angle deficit is nonzero. A cone has all of its curvature at the apex; a polyhedron has all of its curvature at the corners.

But a vertex made by folding a flat sheet has angle deficit zero, which is exactly the developability condition every pattern on this site is checked against. So folding a flat sheet produces no curvature anywhere, concentrated or otherwise.

The apparent curvature of a folded model is entirely a matter of arrangement. The material is flat throughout and always was.

Approximation, quantified a little #

Since spheres are unreachable, everything spherical made from sheet is an approximation, and it is worth knowing roughly how good.

A gore of angular width $w$ on a sphere of radius $R$ has a maximum in-plane distortion of order $w^2/8$ when flattened. For a globe with 12 gores, $w$ is 30° and the distortion is around 3%. With 24 gores it is under 1%, which is why quality globes use many narrow gores and cheap ones visibly bulge at the seams.

The same arithmetic governs faceted approximations, sail panels, and the panelling of curved architectural surfaces. In every case the error falls with the square of the panel size, so halving the panel quarters the error — which is why the practical answer to “how curved can this be” is always “how many panels are affordable”.

Where the model stops #

Ideal inextensibility. Real paper stretches by a fraction of a per cent, which is enough to fudge a small amount of curvature. Many models that are geometrically impossible are physically fine.

Smoothness. The theorem is about smooth surfaces. Creases and vertices are where the interesting behaviour is, and they require the distributional version.

No thickness. As everywhere, the sheet is a surface. Real material has depth and a bent plate has different inner and outer geometry.

The figure’s surfaces are parameterised. The four shapes are drawn from closed-form parameterisations chosen to be recognisable. They illustrate the classification and are not solutions to a developability computation.

Approximation is not quantified. How many facets a given accuracy needs, or how narrow a gore must be, is a real question with real answers, and nothing here computes one.

Why this is the deepest constraint here #

Of everything on this site, this is the result the rest rests on, and it is worth saying why.

The circle-packing argument works because distances within the sheet are preserved. The flat-folding conditions are conditions because angles at a vertex are preserved. Curved creases force developable surfaces because curvature is preserved.

All three are consequences of one fact: folding is an isometry, and an isometry preserves everything intrinsic. Gauss’s theorem is the statement that curvature is intrinsic, which is what makes it the common ancestor.

That also explains why the exceptions are all material rather than geometric. Wet-folding breaks inextensibility; thickness breaks the surface model; crease radius breaks the assumption that folds are lines. Every departure from the theory is a departure from the paper being an ideal inextensible surface, and there is no geometric escape at all.

The constraint that is not a limitation #

It is easy to read this as a list of things paper cannot do, and the more useful reading is the opposite.

Developability is what makes folding predictable. Because the sheet cannot stretch, its behaviour is determined by its crease pattern and nothing else — the same pattern in the same material gives the same object, every time, in anybody’s hands.

A material that stretched would have no such property. The shape would depend on how hard each region was pulled, which is a continuum of possibilities rather than a determinate answer, and there would be no crease pattern worth publishing because it would not specify anything.

So the constraint is what makes the pattern the object. Everything this site relies on — that a figure can be printed and folded and come out right — depends on the sheet having no freedom to stretch, and would fail immediately if it did.

That is a general property of strong constraints in design. They remove possibilities and, in exchange, they make what remains reliable.

Testing it on a pizza #

The most accessible demonstration of the theorem requires no paper at all.

Hold a slice of pizza flat by the crust and the tip droops. Curl the crust upward — introducing curvature across the slice — and the tip lifts and stays rigid.

That is Gaussian curvature being conserved. The slice is flat, so $K = 0$, so the product of the two principal curvatures is zero everywhere. Force one of them to be nonzero by curling, and the other must be zero — the slice cannot bend along its length while it is curved across.

It is the same fact that stops a sheet becoming a sphere, applied in the opposite direction: rather than preventing a shape, it is being used to prevent a deformation.

Corrugated iron works this way, as do folded card structures and every tessellated sheet on this site. Curvature in one direction buys stiffness in the other, and the payment is exact.

The ladder from here #

Later rungs against this anchor, all of them on folding’s side of the boundary: concentrated curvature at a vertex, and the distributional statement that makes a crease legal. Gore computation done properly rather than to leading order. Faceted approximation and how its error falls with panel count. Corrugation as apparent curvature, quantified. The inverse problem for curved creases — given a target surface, find the crease — which is open. Wet-folding and the fibre mechanics underneath it. And the question of which non-developable surfaces are reachable by combinations of the three answers, which nobody appears to have asked.

The theorem itself, its proof and the classification of developable surfaces belong to cartographic projection, which uses them to settle a question this site does not ask.

What is left when the theorem is handed over is still a subject: a flat sheet, an impossibility, and three ways of paying for it.