The test measures the rim

A crease carries no curvature closes this ladder’s first four rungs with a test. A leaf that corrugates and a leaf that ruffles are doing categorically different things — one has a flat metric throughout and the other has a metric that forbids flatness — and they can look similar from a distance. Flattening the specimen will separate them, and that is a better test than any picture.

The test is right. What is done in its name is not quite it, and the gap has a computable shape.

What a flattened specimen actually gives #

Press a leaf and the thing that is visible is what happens at the margin. A flat lamina lies down; a grown one buckles, and the buckling shows as a wavy edge, as tearing along the margin, or as the sheet refusing to make contact near its rim.

So the measurement anybody makes on a pressed specimen is a boundary measurement — the arc length of the margin against the chord it spans, or the count of ripples along it, or simply whether the edge lies down. All of them are quantities of the rim.

That is not carelessness; it is what a pressed specimen offers. The interior of a lamina that has been flattened is flat because it has been flattened, and reading a metric off it requires measuring distances between marked material points before and after, which is a different experiment on a different specimen.

The integral is a boundary quantity, exactly #

Here is the difficulty, and it is a theorem rather than a limitation of technique.

Write Ω for the linear growth factor as a function of the flat sheet’s radius. The metric it produces has a curvature at every point, and the total curvature over the grown surface is ∫K dA with the area element Ω² r dr dθ.

Because K Ω² = −Δ ln Ω, that integral collapses by parts to

∫K dA = −2πR (ln Ω)′®

and that is the whole of it. The total depends on the growth profile’s slope at the rim and on nothing else. Every detail of how the growth was distributed across the interior — where it was fast, where it was slow, whether it changed sign — cancels out on the way.

This collection has had both forms of the integral since the field was founded, and has used them as a check on each other: the differencing scheme has to reproduce the by-parts value, and a boundary mistake shows up as a disagreement. Read as a check it is routine. Read as a statement it says something sharper than the check was meant to establish.

A profile that is curved everywhere and integrates to nothing #

The theorem invites an example and one is easy to build. Take

ln Ω® = a( r² − r⁴ ⁄ 2R² )

whose derivative is 2ar(1 − r²⁄R²) and therefore vanishes at r = R for every a. Its total curvature is zero, exactly, whatever a is.

It is not flat, and growth is what forbids it. Its Laplacian is 4a(1 − 2r²⁄R²), which is positive at the centre and negative outside r = R⁄√2 — so the curvature changes sign there, and the sheet is a negatively curved disc surrounded by a positively curved annulus, or the reverse, depending on the sign of a. At a = 0.35 the curvature reaches minus one and four tenths at the middle and about plus one at the rim, on a disc of unit radius. The grown sheet has 27% more area than the flat one it started as.

A sheet like that cannot be laid in a plane. Not approximately and not nearly: the check that decides it — every sampled curvature within a billionth of zero — fails at almost every radius.

And a rim measurement reports it as flat, because at the rim the growth has levelled off and the margin has exactly the arc length its chord allows.

What the test can and cannot separate #

Putting the two together gives the honest scope of the field’s own separating test.

What flattening at the margin catches. A profile whose growth is concentrated at the rim, which is the standard ruffled-edge case and the one the field’s second rung is about. Its rim slope is large and positive, its total curvature is large and negative, and the margin will not lie down. The test works, and it works for the case that motivated it.

What it misses. Any profile whose growth has levelled off at the margin, whatever it did inside. The interior can be curved as strongly as one likes and the margin will lie flat, so the specimen passes the test and the metric is not flat.

That is a false negative with a name and a shape. It is not noise and it is not a matter of resolution: the class of profiles the test cannot see is exactly the class with vanishing rim slope, and it is a large class rather than a knife edge — any profile that levels off is in it, and levelling off is what a growth process that runs out of time or substrate does — including one held inside a container it has to fit.

Which is a statement about a class of measurements #

The result is stated about pressing a leaf and it is not really about leaves. It is about what any boundary measurement can determine, and the same argument applies to several others this field would like to make.

A measurement that returns ∫K dA cannot see the interior. That includes the margin’s overrun, the count of ripples along the edge, the angle at which the sheet leaves a flat table near its rim, and any quantity computed from the boundary alone. All of them are the same number and all of them have the same blind spot.

What would see it is any measurement made inside. Two marked points a known distance apart on the flat sheet, measured again after growth, give the metric directly and would separate the cancelling profile from a uniform one immediately. So would cutting the disc into annuli and seeing which of them lie flat on their own — a positively curved annulus and a negatively curved disc each refuse the plane separately, and the refusals do not cancel because cutting removes the boundary condition that made them cancel.

Cutting the specimen is the test the theorem suggests, and it is a stronger test than flattening it whole for exactly the reason the theorem gives: the integral over a piece is a boundary quantity of that piece, and moving the boundary inwards is how the interior is interrogated.

What is unusual about this failure #

Most measurement limitations are a matter of degree. A quantity is estimated with an error, the error can be made smaller, and knowing its size bounds the conclusions. This one is not of that kind, and the difference is worth stating because it changes what to do about it.

The blind spot here is a kernel. There is a whole family of growth fields the boundary measurement maps to the same reading, and the family is not small — it is parametrised by everything the profile does in the interior subject to one condition at the edge. No improvement in precision reduces it, because the measurement is not approximating the quantity that would distinguish them; it is computing a different quantity exactly.

That makes the correct response a different measurement rather than a better one. And it makes a particular error easy to fall into: a very precise boundary measurement that returns zero looks like strong evidence for flatness, and it is no evidence at all against the profiles in the kernel.

The same shape appears elsewhere in this collection and it is worth the cross-reference, because seeing it twice is what makes it recognisable. A crease pattern’s layer ordering is underdetermined by the local conditions in exactly this way: the conditions leave a set of possibilities, the set is not small, and the physical answer needs something the conditions do not contain. Precision in checking the conditions does not help.

Curvature spread and curvature at a point, again #

The cancelling profile has a counterpart one rung down that makes it easier to hold, and putting the two side by side is worth doing because they are the same arithmetic used for opposite purposes.

A crease carries no curvature builds a growth field carrying exactly the deficit of a cone: a surface with all of its curvature at one point and a surface with the same total smeared over its whole area, matched on one number and alike in nothing else. The matching works because the total is computable both ways and can be made to agree.

The cancelling profile is that construction run to a total of zero. A cone with no wedge removed is a plane; a growth field with a vanishing rim slope is not a plane and has the same total. So the pair that rung constructed — same total, different distribution — has a degenerate member in which the total is nothing and one of the two is genuinely flat while the other is not.

Which is what makes the total such a poor summary. It matched a cone to a disc where the match was informative; here it matches a flat sheet to a curved one, and the matching is the failure rather than the finding.

What the sign change is for #

The cancelling profile changes the sign of its curvature at R over root two, and the radius is not a free parameter — it is fixed by the family, whatever the growth rate. That is worth taking seriously because it says what such a sheet would look like if it were built.

Inside R over root two the disc is curved one way; outside it, the other. So the surface is a dome with a ruffled skirt, or a saddle with a dished rim, and the two regions are not similar in area: the inner disc is half the flat sheet’s area and the outer annulus is the other half, exactly, because the sign change is at the radius that halves the area.

That is a shape somebody has certainly seen and not named as this. A leaf that is dished in the middle and wavy at the edge is a common enough thing, and a boundary measurement on it reports flat while a hand feels immediately that it is not.

So the blind spot is not exotic. It is a shape a growth process reaches by doing something perfectly ordinary — growing fast early and slowing down towards the margin — and the family with the vanishing rim slope is exactly the family of processes that finished.

That is the uncomfortable version of this rung, and it is the reason for writing it. The class the test cannot see is not a mathematical curiosity constructed to defeat it. It is the class of completed growth, and the class it can see is the class still in progress.

Which theorem was checked and how #

The two forms of the integral are computed independently and required to agree, on every profile drawn, to within a part in a million of the largest total. That agreement is what licenses reading the by-parts form as a statement about the boundary rather than as an identity that happens to hold.

The cancelling profile’s curvature at the centre is checked against a closed form the grid never sees. Its Laplacian at the origin is 4a, so K(0) is −4a exactly; the differenced value has to reproduce it, which it does to better than a part in a million.

The profile is required to be curved. The check that decides flatness — every sampled curvature within a billionth of zero — must fail on it, and the figure refuses to draw if it passes. A cancelling profile that came out flat would make the essay’s point vacuous rather than wrong, which is the harder kind of error to notice.

And the total is required to be small beside the peak, at better than one per cent, so the claim “curved everywhere and integrating to nothing” is a measured statement rather than a description of what was intended.

Where the model stops #

The growth is radial, isotropic and prescribed. Real tissue grows anisotropically, along directions the tissue itself sets, on a schedule responding to the shape already produced. None of that is here, and what survives the simplification is the part that is geometry: a metric decides a curvature, and the integral of that curvature is decided at the boundary.

The disc has one boundary. A lamina with a midrib, a lobe or a notch has more, and the integral is then a sum of boundary terms — which does not make the blind spot smaller, but changes which profiles fall into it.

Flattening is modelled as a question about the metric and it is a mechanical process. A pressed leaf tears, creases and slips, and a specimen that lies flat may be lying flat because it stretched rather than because its metric permitted it. That is a second way for the test to return a false negative and it is not the one computed here.

And the cancelling family is exhibited rather than characterised. The essay shows one family with vanishing rim slope; the class of profiles a boundary measurement cannot see is every profile with that property, and no attempt is made here to say what else is in it.

What the picture cannot show #

The figures draw growth factors and curvatures against radius, and neither is a surface. What a sheet with this metric would look like is not computed anywhere — this ladder has been careful throughout to draw the curvature rather than an embedding, because a picture of a shape is exactly where this field would start lying.

The consequence is that the essay cannot show what the reader most wants to see, which is a picture of the cancelling sheet next to a flat one. There is a good reason for the refusal and it is the same one: the two would have to be drawn as surfaces, drawing them as surfaces would mean solving for an embedding, and an embedding is exactly the thing that is not determined by the metric.

Nor can the figure show the measurement it is about. A pressed specimen, a ruler along a margin and a count of ripples are not quantities this collection computes, and the essay’s argument is that a quantity of that kind equals the boundary term — which is arithmetic on the model rather than a claim about anybody’s laboratory.

The idealisation, named #

The sheet is a disc of radius R with a metric prescribed by a positive radial function, and the curvature is computed from that metric alone. No material property appears anywhere: given how much each ring grew, the curvature is determined, and a curvature that is not zero is a sheet that cannot lie in a plane.

That is the ladder’s founding idealisation and it is what makes this rung’s finding a theorem rather than an observation about specimens. The integral of a curvature computed from a metric is a boundary quantity, exactly, and every failure mode above follows from that and from nothing else.

What is left out is everything about how hard it is to make the sheet lie flat. A metric with small curvature and a metric with large curvature both forbid flatness, and a specimen with the first will lie down under a press while one with the second will not. So the practical test has a threshold in it that the geometry does not, and the cancelling profile at small a will pass a real press even where the interior curvature is nominally nonzero.

That widens the blind spot rather than narrowing it, which is the direction worth noticing.

Where the ladder goes next #

This ladder has now spent six rungs on the same object read six ways, and what it owes next is the measurement rather than another computation.

The measurement is an interior one: two material points, a known distance apart on the flat sheet, measured again after growth. That single number is the metric where it matters and it is invisible to everything this rung is about. It is not a computation this collection can make; it is an experiment, and it is the one the ladder’s whole argument has been implicitly asking for since its first rung established that a set of distances determines a curvature.

Sideways from here, the field’s two halves meet at the thing this rung failed to see. A corrugated leaf and a cancelling grown one both lie flat at the margin, both refuse to lie flat as a whole for different reasons, and separating them needs the interior. So the test the folding half of the field would run — does the pattern have creases in it — is the one that works, and it works because a crease is an interior feature.

The habit worth carrying out of this rung is a question. When a measurement is proposed as a test, ask what it is an integral of. A quantity that reduces to a boundary term is blind to everything inside, and the class it is blind to is a kernel rather than an error bar — so no amount of care with the instrument will recover it.