The container picks the member

The excess does not choose its waves ends at a handover. A rim that has grown longer than its span has to put the extra length somewhere; two large waves and eight small ones carry identical arc length and identical excess; and the metric is completely indifferent between them. Geometry runs out of things to say, and the essay names the handover as the thing to watch because it is usually made silently.

This rung is what happens after the handover. The question is what does the choosing, and the first two candidates both turn out to be unavailable.

Bending cannot choose, and the reason is that it is monotone #

The obvious candidate is energy, and the obvious energy is bending. A sheet resists being curved, the resistance is quadratic in curvature, and the natural rule is that the rim settles into whichever member of the family costs it least. A sheet that grows cannot lie flat, so something has to be spent whatever it does.

That rule has an answer and the answer is always the same one.

The family’s total squared curvature goes as the square of the wave count, exactly — it is one of the things asserted whenever the family is drawn, and it follows from the invariant. Amplitude times wave count is constant, so doubling the number of waves halves each one’s height; a sinusoid’s curvature goes as amplitude times the square of the wave number; so the curvature per wave rises as the wave number, the length is unchanged, and the integral of the square rises as the square of the count.

Rising means monotone, and monotone means the minimum is at the end. A least-bending rule selects the coarsest member on offer, at every excess, for every span. Two waves, always. Not two rather than eight because two is better suited to something, but two because the objective has no interior minimum at all.

So a leaf edge with a dozen ripples in it is not at the bottom of a bending energy. Something else is holding it there, and bending is the thing pushing the other way.

The invariant, read as a constraint #

The thing that does the choosing has to have a length in it, because every quantity so far — the metric, the excess, the bending integral — is scale-free in the relevant sense and none of them prefers a number of waves.

A container has a length in it. A rim confined inside a bud has a depth it must not exceed — and which way the disc curves has already fixed that it is the rim rather than the middle doing the growing — and the family’s own invariant turns that into a statement about the count.

Amplitude × wave count is constant across the family. That is asserted every time the family is drawn — it holds to within a fraction of a per cent across the members computed — and it says twice the waves is exactly half the height. Read the other way it says a ceiling on the height is a floor on the count:

waves ≥ (amplitude × waves) ⁄ depth

with the numerator a constant fixed by the excess and the span. A container half as deep needs twice as many waves. Not roughly twice: exactly, because the invariant is exact.

The figure computes the floor for several depths and checks that floor × depth is the same number for all of them, which is the invariant tested at the place it is being used rather than where it was established.

What the container costs #

A floor on the wave count and a bending energy rising as the square of the count together price the container, and the price is steeper than the container’s own dimensions suggest.

If the confined rim sits at its floor — the coarsest arrangement that fits — then the wave count is proportional to one over the depth, and the bending is proportional to the square of the count. So

bending ∝ 1 ⁄ depth²

A bud half as deep costs four times the bending energy. A bud a third as deep costs nine times.

That is a strong statement about what confinement is worth and it is worth being clear about what it is a statement of. It is not a claim that any bud is at any depth, or that any leaf minimises anything. It is the shape of the trade the geometry sets up: the container sets a floor, the floor sets a bending cost, and the cost is quadratic in how tight the container is.

Which makes the container the expensive constraint rather than the incidental one. The bud chooses the pattern is the same finding one field over and about folding rather than growing, and the two arrive at it by different routes: there, the container decides which crease pattern fits; here, it decides which member of a metrically indistinguishable family the sheet must take.

The two mechanisms are pulling opposite ways #

Setting the two side by side gives the argument its shape, and the shape is a familiar one that this collection keeps meeting.

Bending prefers few waves, monotonically, with no interior optimum.

The container requires many, with a hard floor and nothing above it.

So the selected member is the floor — the coarsest arrangement the container admits — and not an interior optimum at all. That is a different kind of answer from the ones the neighbouring ladders produce. A hinge radius against a packing benefit gives a parabola with a peak in the middle; a sheet’s thickness against a fold count gives another. Here the objective is monotone and the constraint is a bound, so the answer sits on the constraint.

Which has a diagnostic consequence worth stating. A structure whose answer sits on a constraint is a measurement of the constraint and not of the objective. Measuring the wave count of a confined rim measures the depth of its container, to within the excess, and says nothing whatever about the sheet’s stiffness — because the stiffness only sets the direction of a preference that the constraint has already overruled.

What the sheet’s own thickness contributes, and why it is not this #

There is a third candidate for what selects the wave number and it deserves to be dealt with rather than ignored, because it is the one the literature on rippled edges is actually about.

A sheet with a thickness has a bending stiffness that goes as the cube of it and a stretching stiffness that goes as the thickness itself. A rim that has grown too long can spend the excess by rippling — which costs bending — or by stretching the material near it — which costs stretching. Balancing the two gives a selected wavelength with the thickness in it, and that is the standard account of why a torn plastic sheet or a lettuce leaf ripples at the scale it does.

That mechanism is real and it is not in this model, for a reason that is worth stating plainly rather than apologising for. Everything in this ladder is a metric, and a metric has no material in it: the family here is a family of isometric profiles, all of which carry the excess exactly, none of which stretches anything. A stretching term cannot be computed from a metric, because a sheet that stretches has changed its metric and is a different sheet.

So the honest account is that there are two selections and this rung is about one of them. The stretching balance sets a wavelength for an unconfined rim in a material with stated stiffnesses. The container sets a floor for a confined one in any material at all. Which of them binds is a question about numbers, and for a rim in a bud — where the depth is much smaller than the span — the floor is the harder constraint by a long way.

The excess and the depth enter as a ratio #

The floor has two quantities in it and they do not enter separately, which sharpens what a wave count can be read as.

The constant that amplitude and wave count multiply to is set by the excess and the span: a rim carrying more excess has taller waves at every count, in proportion. So the floor is that constant over the depth, and doubling the excess and doubling the depth together leave the floor exactly where it was.

A wave count is therefore a measurement of excess-over-depth and of nothing else. Two rims with the same number of ripples might be a barely-overgrown rim in a tight bud or a badly-overgrown one in a loose bud, and no count distinguishes them.

That has a use, because one of the two is much easier to measure than the other. The depth of a bud is a dissection; the excess of a rim is the thing a botanist would like to know and cannot easily get, since it requires the flattened margin’s arc length against its chord. Counting ripples and measuring the bud gives the excess without ever flattening anything.

Whether that is worth doing depends on how good the model is, and the honest answer is that it is good for the direction and rough for the number: the sinusoid is a choice, the container is a bound rather than a shape, and the constant multiplying the ratio depends on both. What the reading is robust to is scale, which is often what matters — a lineage whose buds are half as deep and whose margins ripple twice as finely has the same excess as one whose buds are loose and whose margins are smooth, and that is a comparison the ratio makes and neither measurement makes alone.

What a monotone objective does to an argument #

The finding that bending is monotone across the family is worth separating from the leaf, because it is the part that generalises and it is the part that is easiest to get wrong.

An objective with an interior optimum is what most arguments about form assume. A structure is described as balancing two costs, the balance is drawn as a curve with a minimum, and the structure is placed at the minimum. That picture licenses a particular kind of inference: measure where the structure is, and read off the ratio of the two costs.

An objective with no interior optimum licenses nothing of the kind. Bending across this family is monotone, so a rim’s wave count carries no information about its stiffness — every stiffness gives the same preference, and the preference is overruled. A reader who assumed an interior optimum and read a stiffness off a wave count would get a number, and the number would be meaningless rather than merely imprecise.

The check that separates the two cases is cheap and almost never made. Evaluate the objective across the family and see whether it turns. Here it does not, and the fact that it does not is asserted every time the family is drawn, so a change to the family that introduced an interior minimum would break the figure rather than quietly changing what the essay means.

The same check is worth running on the neighbouring ladders’ claims, and there it comes out the other way: a hinge radius against a packing benefit really does turn, and so does a sheet’s thickness against a fold count. Those are genuine interior optima and the inferences they license are sound. This one is not, and the difference is a property of the arithmetic rather than of how carefully anybody argued.

Which theorem was checked and how #

The invariant is checked where it is used, not only where it was established. The floor is computed for each depth and floor × depth is required to be the same number across all of them, to within a part in a billion — which is the statement “the floor is exactly a reciprocal” in a form a wrong power breaks.

The monotonicity is asserted rather than described. Every member’s bending integral must exceed the one before it, so a figure in which the bending had an interior minimum would refuse to draw — and that figure would falsify this essay’s central claim rather than merely looking odd.

The amplitudes are solved rather than approximated. Each member’s height is found by bisection on its own arc length, and the arc length is then measured and required to match the target — so the family really does share one excess rather than sharing it to the accuracy of a small-slope formula.

And the containers admit different numbers of members, which is checked: the tightest depth must accept strictly fewer members than the loosest. A comparison in which every container admitted the same set would be a figure with no selection in it.

Where the model stops #

The profiles are sinusoids and a real rim is not. The family is a one-parameter family of sine waves chosen because the arc length is computable and the invariant is exact; a rim free to take any profile carrying the excess has more members, and the invariant becomes an approximation rather than an identity.

The container is a depth and a bud is a shape. Bounding the amplitude is the crudest possible model of confinement and it ignores everything about how the rim is packed against itself, which is what the folding half of this field spends its rungs on.

The rim is treated alone. A growing lamina ripples at its edge and does other things in its interior — nothing grown has a seam to relieve it at — and the interior constrains the edge — a ripple has to decay inwards, and how far it takes to do so is a length the model here does not have.

And nothing here is a measurement of a leaf. No growth rate is modelled and no specimen is traced. What is computed is a family of profiles carrying an excess, and what the essay claims is about what a container does to such a family.

What the picture cannot show #

The figure draws amplitudes against a depth and cannot show a rim arriving at one. Nothing here is a dynamics: there is no account of how a sheet growing inside a bud finds the member it ends at, only of which members are available.

That gap is the one worth naming, because a floor is not a mechanism. A rim could reach its floor by growing gradually inside a container that squeezes it, or by buckling into a fine mode from the start, or by taking a coarse mode and then splitting it — and all three end at the same member without the picture distinguishing them.

Nor can the figure show what happens when the excess grows past what any admissible member can carry. The family is drawn at a fixed excess and a real rim’s excess keeps increasing, so the floor keeps rising, and something has to move from one member to the next. The transition between members is not in this model at all, and it is where the interesting behaviour of a real rippled edge lives.

The idealisation, named #

The rim is a curve of prescribed arc length across a fixed span, spending its excess in a sinusoid of some number of periods, inside a container that bounds its amplitude. Nothing bends anything back; nothing touches anything; nothing stretches.

The most consequential of those is the last. A profile that stretches is a profile with a different metric, so an inextensible model is not an approximation to a slightly extensible one — it is a different problem. What survives is everything computed from arc length: the invariant, the floor, and the quadratic cost.

The least consequential is the sinusoid, which is a choice of coordinates more than a claim. Any family of profiles with a fixed excess and a wave count has amplitude falling as the count rises; the exact reciprocal is a property of sinusoids and the direction is not.

And a crease carries no curvature is the reason the whole family is a family of smooth profiles. A rim that folded rather than rippled would carry its excess at points instead of over an interval, and the two are categorically different things that can look alike from a distance.

Where the ladder goes next #

This rung says a container selects, and it says so about a family the metric had left open. The rung after it turns on the test that would tell the two halves of this field apart in a specimen.

A crease carries no curvature ends by proposing one: a leaf that corrugates and a leaf that ruffles are doing categorically different things, they can look similar from a distance, and flattening the specimen will separate them. That is the right test and it is not the test that gets run. What gets run is a measurement at the margin, and what a margin measurement can see turns out to be a boundary quantity with a computable blind spot in it — a growth field can be curved everywhere and integrate to nothing, and a rim measurement reports it as flat.

The thing worth carrying from this rung is the diagnostic sentence. When a monotone objective meets a hard constraint, the answer is the constraint, and measuring the answer measures the constraint rather than the objective. That is a good deal if the constraint is what was wanted and a trap if it is not, and telling which is which requires knowing that the objective was monotone — which is exactly the thing nobody checks.