Nothing in a body folds on a line

Every figure in this field so far has drawn a crease as a line. That is the idealisation the whole site runs on, it is a good one for paper, and it is worse for an organism than for almost any other material anybody folds.

A fold in a wing, a leaf or an organ is not a crease. It is a compliant region: a strip of thinner or differently structured material that bends because it is easier to bend there than elsewhere. Compliant regions have a width, and the width is not small compared with anything else about the structure.

What a radius costs, in one line #

The arithmetic was already on this site and it is worth restating because it is short.

A fold that goes round a radius rather than through a point carries the material along an arc of half a circle, which is π times the radius. The stack it produces advances by twice the radius. The difference — π minus two, times the radius — is material that went into the fold and did not come out the other side as usable surface.

That is a fixed loss per fold. It does not depend on the panel length, on how the pattern is arranged, or on how many other folds there are. So a pattern with twice as many folds loses exactly twice as much, and the loss as a share of the sheet is proportional to the count.

The generator checks that proportionality rather than assuming it. It computes the share at each fold count from the geometry and then asserts that the ratio of the largest share to the smallest equals the ratio of the counts, to within two percent. An arithmetic error that made the loss superlinear or sublinear stops the build.

Why this is worse in a body than in paper #

For paper the correction is a correction. For biological material it is closer to the dominant term, for three reasons that compound.

The relative thickness is larger. A sheet of paper is a tenth of a millimetre thick and a fold spacing might be ten millimetres, a ratio of a hundred. An insect wing membrane is thin, but its fold regions are not membrane — they are structured cuticle with veins nearby — and the ratio of hinge width to panel length is nothing like a hundred.

The hinge is deliberately wide. A compliant hinge works by distributing the bending over a length; concentrating it would exceed the material’s strain limit and break it. So the width is not a manufacturing imperfection to be minimised, it is a design requirement, and making it narrower makes the structure fail sooner.

And the fold has to survive repetition. A paper crease is made once. A wing hinge cycles thousands of times, and fatigue in a bent region scales sharply with the strain, which scales inversely with the radius. Every consideration pushes the radius up.

The ceiling, and what it does to the packing numbers #

The packing comparison two rungs back reported fractions that improve without limit as each geometry’s parameter is pushed. Every one of those numbers is now known to be optimistic, and by a knowable amount.

A roll at n turns packs to one over n, which goes to zero. With a minimum bend radius it does not: the innermost turn cannot be tighter than the radius, so the packed cylinder has a hole in it and the fraction bottoms out.

A corrugation at a finer angle packs better without limit. With a hinge width it does not: past the count where the hinges consume the sheet, adding folds removes surface faster than it removes footprint.

Every packing number in this field is an upper bound, and the bound is approached only in the limit of a material that does not exist. That is the right thing for those numbers to be, provided it is said, and this is where it is said.

The habit worth taking from it generalises past folding. A quantity that improves monotonically as a parameter is pushed, with no turning point anywhere, is nearly always a model that has left something out — because real systems have costs that grow as well as benefits, and a benefit with no matching cost is a sign that the cost was not modelled rather than that it does not exist. The corrugation’s packing fraction going to zero as the angle closes is exactly that signature.

Applying the test to this site’s own results is instructive. The circle-packing bound on a base’s efficiency has a turning point and is honest. The packing fractions in this field did not, and now do.

The same reading applies to any claim about a folded structure that reports a compaction ratio with no fold count attached. The ratio is only meaningful with the count, because the count is what determines how much of the ratio the hinges have already spent.

The number that gets the ceiling right #

It is worth extracting the ceiling as a formula, because it is the one design-usable output of the essay and it is short.

The loss per fold is π minus two, times the hinge radius — about 1.14 radii. Set a tolerance for how much of the sheet may go into hinges, multiply the sheet’s size by that tolerance, and divide by 1.14 radii: that is the largest number of folds worth having. At a tenth-of-a-unit radius on a ten-unit sheet with a quarter of the surface allowed to go into hinges, it is about twenty-two folds.

Everything else about the pattern is absent from that calculation, which is what makes it useful early. It needs no crease pattern, no arrangement, no packing analysis and no simulation — a radius, a size, and a tolerance — and it says whether an ambition is available before any work is done on how to achieve it.

This is the same shape of result as the bud’s interior optimum and it arrives from a different term — that one is about the depth of a stack and this one about the surface consumed by hinges — which is worth noticing because it means the two bounds are independent and a real design has to clear both.

The ceiling is really an optimum #

The formula above needs a tolerance supplied from outside, which is the usual shape of a design rule and is weaker than it needs to be. Asking what the folds are for removes the tolerance and replaces the ceiling with a maximum.

A corrugation of $n$ folds reduces a sheet of size $L$ to a packed footprint of about $L/n$, and leaves usable surface $L - (\pi-2)n\rho$. The quantity a deployable is actually judged on is the ratio of the two — how much surface it delivers per unit of stowed space:

$$\frac{L - (\pi-2)n\rho}{L/n} = \frac{n\big(L - 1.1416\,n\rho\big)}{L}$$

which is a quadratic in $n$ and therefore has a peak. Differentiating, the peak sits at

$$n^{*} = \frac{L}{2 \times 1.1416\,\rho} \approx 0.438\,\frac{L}{\rho}$$

On the worked example — a sheet ten units across with a hinge radius of a tenth — that is forty-four folds, and past it every additional fold makes the structure worse by this measure rather than merely less efficient.

Half the sheet, again #

Substitute $n^{*}$ back into the loss and the answer is worth noticing: the hinges consume $(\pi-2)n^{*}\rho / L = 1/2$.

At the optimum, exactly half the sheet has gone into the folds. That is the third time this fraction has turned up in this field — the bud’s optimum bundle, the surface a corrugation holds in a fixed box, and now the fold count a deployable should carry — and it is the same arithmetic each time. A benefit proportional to a count, a cost proportional to that count subtracted from a fixed budget, and a product that peaks where the budget is evenly split.

The rule is worth carrying because it needs no tolerance, no simulation and no pattern. Fold until half the material has gone into hinges, then stop, and the resulting count is what the geometry can support.

It also gives the figure of merit at the peak, which is $n^{*}/2$ — twenty-two on the worked example. A corrugation of ideal creases would report an unbounded compaction; the same corrugation with a real hinge radius delivers about twenty-two times its stowed footprint in usable surface and no more, whatever the fold count.

That number is the honest version of every packing fraction two rungs back. It depends on one ratio, the sheet’s size to the hinge’s radius, and a designer who knows nothing else about a structure can compute it.

The other correction, which points the same way #

There is a second thickness effect and it is worth distinguishing because the two are often conflated.

The hinge radius costs surface: material used going round a corner. The layer thickness costs depth: a stack of n layers is n thicknesses deep, and the outer layers of a fold have further to travel than the inner ones.

Both scale with the fold count and both make fine patterns worse, so they cannot be told apart by looking at how a packing number degrades. They can be told apart by what fixes them: the depth cost is addressed by offsetting the panels, which is a geometric construction, and the surface cost is not addressable at all.

What engineering does about it, and what a body cannot #

The engineered response to thickness is a body of technique, and none of it is available to an organism.

Every one of those techniques assumes a manufactured structure: parts that can be given different thicknesses, hinge axes placed away from the material’s mid-plane, tolerances specified. A wing grows as one piece and its hinges are where its material is thinner, which places the hinge axis at the mid-surface by construction.

So the organism is working with the least favourable version of the problem and the engineering literature’s solutions do not transfer. What an organism has instead is a material with graded properties — stiffness that varies continuously rather than in steps — which is a capability engineering has only recently started to reach for.

Where the compliant hinge is better #

The account so far is all cost, which is unbalanced, and the compensating advantage is real.

A compliant hinge has no bearing, no pin, no clearance and no wear surface. It cannot come apart, it does not need lubrication, it seals by default, and it has zero backlash — the panels either side are in a defined relationship at every angle rather than within a tolerance.

Those are the properties engineered compliant mechanisms are built for, and they are the reason the technique exists in engineering at all. The trade is that a compliant hinge stores energy when bent, which a pin joint does not, so the structure pushes back and a deployed state has to be held or latched.

For a wing that storage is not a cost but the mechanism: a hinge that pushes toward open is a hinge that helps deploy. The same property that makes the fold expensive in surface makes the deployment cheap in muscle, which is the kind of trade that only looks like a trade when both halves are on the page.

It also resolves something the previous rung left open. A single degree of freedom explains how one input can move the whole surface; it does not explain where the energy comes from to move it quickly, or what holds the deployed state against aerodynamic load. Stored strain in the hinges answers both — the fold releases toward open, and the same stiffness resists closing again under load.

That is a different arrangement from the engineered deployables, where the pattern provides the kinematics and a separate spring or latch provides the energy and the retention. In a wing the hinge is doing all three jobs, which is why the compliant hinge is not merely a cheaper joint but a different design philosophy. Self-folding sheets are the engineered attempt at the same consolidation, and they are recent.

The cost of consolidating is that the three requirements can no longer be tuned independently. A hinge stiff enough to hold the deployed wing is a hinge that fights being folded, and there is one material property setting both.

What the picture cannot show #

The figure computes a share of a sheet from a radius and a count. It does not model any hinge.

Real compliant hinges are not arcs of constant radius. They are regions with a stiffness that varies along them, they bend into shapes with varying curvature, and the effective length lost is a property of that shape rather than of a single number. The π-minus-two arithmetic is the simplest possible model, it is the same one this site uses for paper, and it is a lower bound on the loss rather than an estimate of it.

Nor does the figure say anything about where on the sheet the loss falls. A pattern whose hinges are concentrated in one region loses that region; one whose hinges are spread loses a little everywhere. For a wing, where the fold regions are placed relative to the veins is most of the design, and none of that is here.

The idealisation, named #

The sheet is uniform. That is the assumption this whole essay is arguing against for the material and still making for the geometry — the loss is computed as though every fold were the same and every part of the sheet were identical.

A real wing is graded on purpose. Its fold regions are where they are because the material there was made to be compliant, and the rest is stiff for reasons of its own. A model that prices hinges uniformly across a uniform sheet is describing the constraint rather than the solution.

And the count is treated as a free variable. In a wing it is not: the fold lines are where the veins allow, and the number of them is a developmental outcome rather than a choice. The ceiling computed here is a bound the animal has to live under, not a menu it selects from.

The last idealisation is the flattest one and it runs through every figure in this field: the fold is treated as reversible and lossless. It is neither. Material that has been bent past its elastic range does not come back to where it was, and a fold that has been cycled thousands of times is not the fold it was on the first cycle. Nothing modelled here accounts for fatigue, and a packing number computed for a new structure says nothing about an old one.

Where this ladder goes next #

This closes the wings ladder, and the field turns from patterns that pack to patterns that route.

Two rungs remain in the field’s other half. A sheet that routes itself is folding at a scale where there is no sheet at all — a single strand held into a shape by short complements — and the design problem turns out to have a parity obstruction of exactly the kind Maekawa’s condition is. And two things called folding asks whether the word survives the change of scale, and concludes that it does not.

The connection worth carrying from here is about where models fail. Every rung of this ladder has been a correction to a zero-thickness idealisation, and the corrections have all pointed the same way — the real thing is worse than the model and the gap grows with the fold count. That is a useful prior for reading any packing claim about anything.

The broader version of that prior is worth stating plainly, because it is what three rungs of corrections add up to. A folding claim made in the zero-thickness model is a claim about what the arrangement permits, and a folding claim made about an object is a claim about what a material allows. The two are routinely reported in the same units and they are not the same quantity, and the gap between them is not a small percentage — it is the difference between a bound that improves forever and one that turns over.