The only pattern that moves

The Miura fold moves. Take a sheet creased into identical parallelograms, push two opposite corners together, and the whole thing collapses in both directions at once, along a single path, without any panel bending anywhere.

That sentence gets said about the pattern as though it were a property of the family — as though a corrugation of roughly that shape would do roughly that thing. It is not a property of the family. It is a property of the exact numbers, and the pattern next door to it is a solid.

Three equations, and what they are about #

The folded position of a Miura is not posed and then checked. It is solved, from nothing but the panels being rigid, and the solution is short enough to write down.

Take the flat pattern as parallelograms with a horizontal side b and a slanted side offset by (s, c). In the folded state, put the vertices on a grid whose horizontal step is S, whose row spacing is L, whose alternate rows are shifted by p, and whose alternate columns are raised by z. Then ask only that nothing stretches and nothing bends:

S² + z² = b², so the horizontal edges are still b. p² + L² = s² + c², so the slanted edges are still as long. S·p = b·s, so the panel’s corner angle is unchanged.

The three do not pin the folded position down. One coordinate is left over, and running it from one end of its range to the other is the fold — which is why the whole of the Miura’s motion is these three lines arriving as arithmetic. Every panel is a parallelogram by construction, so planarity comes at no cost and nothing has to be checked for it.

The important word is identical. The equations have one b, one s and one c in them, and they can only be written at all because every panel in the sheet has the same three numbers. A pattern whose panels differ from one another does not have three equations; it has a great many, and they do not agree.

The neighbour that is a solid #

So the test is obvious once the equations are on the page. Take the exact pattern, move every vertex of the flat drawing a little, and ask how far the equations are from being satisfied.

Two things have to be got right or the answer is worthless. The displacements must be one fixed set of directions scaled up and down rather than a fresh scatter at each size, so that what is measured is a scaling rather than a lottery. And the moved pattern must be given the best chance available: its own edge lengths are measured, and the single panel whose three numbers best match their means is fitted to it before any folding happens. What is left after that is not a bad choice of panel. It is what remains when the best choice has been made.

The exact pattern misses its equations by 6.7 × 10⁻¹⁶, which is not a small error but the absence of one — it is where double-precision arithmetic stops and not where the geometry does. The same pattern moved by a hundredth of a panel misses them by 1.7 × 10⁻², which is roughly a sixtieth of a panel width of edge that has to appear from somewhere and cannot.

An isometry does not have a tolerance. Either every distance is preserved or the object is not a rigid folding of that sheet, and 1.7 × 10⁻² is not a near miss but a statement that no isometric folded position of this kind exists for the moved pattern.

First order, which is the whole finding #

A reader who accepts the last section will still expect the failure to fade. Halve the displacement and surely the residual falls faster than the displacement does — enough of a shrink and the pattern would be foldable for practical purposes.

It does not fall faster. It falls at exactly the same rate.

Across five displacements from a millionth of a panel to a hundredth, the residual rises as the displacement to the power 0.9999, and the exponent is fitted from the measured points rather than quoted from a theory. A move of a millionth leaves a residual of 1.7 × 10⁻⁶ — larger than the move that caused it. There is no threshold below which the failure becomes second order and quietly disappears; the shortfall is proportional to the displacement, all the way down, until the arithmetic itself runs out.

That is a much stronger statement than “perturbed patterns do not fold”, and it is the one worth carrying. A second-order failure would mean the pattern had a neighbourhood: a region of patterns near the exact one that fold to within any accuracy asked for, shrinking as the square of the displacement. A first-order failure means the pattern is a point.

The coefficient, which is the number a builder needs #

An exponent of one says the failure is proportional to the displacement and leaves the constant of proportionality unstated, and the constant is the half a workshop can use.

Read it off the two ends of the sweep. A displacement of a millionth of a panel leaves a residual of 1.7 × 10⁻⁶; a displacement of a hundredth leaves 1.7 × 10⁻². The coefficient is about 1.7 throughout, which is what a slope of one means — the residual is one and seven tenths times the displacement, at every size measured, in whatever units the displacement was given in.

That number is dimensionless and it is an amplification. Move a vertex by a thousandth of a panel and the sheet has a thousand and seven hundredths of a panel width of edge that has to appear from nowhere. The error does not merely fail to shrink; it comes out slightly larger than it went in.

Turn that into millimetres and it becomes a specification. A sheet a hundred and fifty millimetres across with five panels to a side has panels thirty millimetres wide, so a residual budget of a thousandth of an edge — which is about a thirtieth of a millimetre of unaccounted length — allows vertices placed to within six ten-thousandths of a panel, which is eighteen microns. That is a machining tolerance rather than a drawing one, and it is what the phrase the pattern is a point costs when it is priced.

What a second-order failure would have been worth #

The essay’s sharpest claim is that the failure is first order rather than second, and the difference between those two is worth putting a number on rather than leaving as a shape.

A first-order failure means the patterns folding to within a residual $\varepsilon$ form a ball of radius $\varepsilon/1.7$ around the exact one. A second-order failure would put them in a ball of radius proportional to $\sqrt{\varepsilon}$.

At a residual budget of a millionth, the first gives a radius of about six ten-millionths of a panel and the second gives a thousandth. A factor of nearly two thousand in how much room a builder has, at that budget, and the factor grows as the budget tightens — because a square root shrinks far more slowly than its argument.

So the distinction between the two exponents is not a technicality about how a curve is drawn. It is the difference between a pattern with a workable neighbourhood and a pattern with none, and the measurement puts the Miura firmly in the second case. The eighteen microns above would have been thirty-five millimetres if the failure had been second order — which is to say, no constraint at all.

Which claim was checked, and how #

The number the whole essay rests on is a fitted exponent, so the way it is obtained decides whether it means anything.

The exact pattern is required to satisfy its own equations to better than 10⁻¹², checked before any displaced case is computed. That is the control, and it fails loudly if the solver is wrong: a pattern that cannot satisfy its equations when it has been given no reason to fail would mean the residual being measured afterwards is the solver’s and not the geometry’s.

The slope is then fitted by least squares in the logarithms and required to come out at one to within 0.15. It is not compared against a value the calculation put in — nothing anywhere in the construction knows that the failure should be first order, and a quadratic failure, or a failure that saturated, would come out at two or at zero and the figure would refuse to draw.

The measurement also has to be immune to a cheat that would produce a slope of one for the wrong reason. Fitting the best single panel to the moved pattern is exactly that guard: without it, the residual would partly be the arbitrariness of which panel was chosen, and it would scale with the displacement for a reason having nothing to do with folding. The fit removes that term and the slope survives it.

The same result, one dimension down #

This is the rigid counterpart of a finding this site already has on the flat-folding side, and the pair is worth putting together because the mechanism is not the same and the conclusion is.

Almost no crease pattern folds flat. Kawasaki’s condition is one equation for every interior vertex, a drawing satisfies an equation with probability zero, and every pattern anybody has ever drawn folds because it was constructed to. The same statement can be made by nudging: take a working pattern, move its vertices, and watch the residual.

The two results have different arithmetic underneath. The flat-folding condition is an equation in the sector angles at one point, and its residual is an angle. The rigid condition is a demand that every distance in the sheet be preserved by a folded position, and its residual is a length. They agree on the conclusion because both are equalities, and an equality has no interior: the two conditions at a point and the three equations above are all statements that something is exactly so.

What is genuinely surprising is which way the two point when they are read as advice. The flat-folding version says a drawn pattern will not fold, so patterns must be constructed. The rigid version says something sharper about the constructed ones: having constructed a pattern that works, a builder cannot then round its dimensions. The Miura on a drawing with three decimal places is a different object from the Miura.

Where a count would go, and why there is not one #

At this point a reader who knows mechanism theory is waiting for a count. The classical next move sets a tally of constraints against a tally of what an assembly would need to be immobile, and reads the difference off as a verdict on whether the thing can move.

Rigid origami is entitled to the word: its patterns are overconstrained spatial linkages, and saying so is part of what the field is. It is not entitled to the argument. Setting a count against a measurement — and the machinery for explaining why some assemblies move despite the count saying they should not — belongs in this fleet to machinekinematics.xyz, under its ground on spatial mechanisms and overconstraint, and that site’s essays are where a reader should go for it. Nothing in this essay counts anything.

What is left when the count is set aside turns out to be sufficient, which is the point worth making. The measurement above never asks how many conditions there are. It asks whether an isometric folded position exists, gets a length back, and watches how that length behaves as the pattern is moved. A count would explain why the answer comes out this way; the measurement establishes that it does, on patterns of two sizes at two fold states, without borrowing anything.

What the residual cannot say #

Three limits, and the first is the one that would be dishonest to leave out.

The residual is measured against a folded position of one particular form — the grid the exact Miura’s motion lives on, with its four unknowns. A moved pattern might have some entirely different isometric folded position, of a shape nobody parametrised, and this measurement could not see it. What has been established is that the family the Miura moves in does not contain the neighbour, which is a narrower claim than the neighbour cannot fold at all and is the claim the figures support.

The second is that a residual is not a motion. Even a pattern satisfying every equation exactly might be blocked by something the equations do not mention — one panel arriving where another already is, for instance, which no equation about lengths can report.

The third is the idealisation, and it is why the finding does not match anybody’s experience of paper. A panel is exactly flat by definition and stays exactly flat. Paper does not: it takes up a misplaced crease as a curvature spread over a hand’s width, which is why a badly drawn Miura folds perfectly well in paper and a badly cut one jams in aluminium. That difference is not a manufacturing detail but a geometric one, and it is the whole reason this result is invisible until somebody builds something. Four idealisations are doing work here and this is the load-bearing one.

Who noticed it, and when #

Koryo Miura arrived at the pattern from the other direction entirely, which is part of why the exactness was not the headline.

His interest in the 1970s was a corrugated sheet’s buckling behaviour, and the herringbone pattern came out of that work rather than out of a search for a rigid folding. The property that made it famous — a sheet that packs in two directions from a single input and deploys the same way — was recognised afterwards, and the solar-array application in the 1980s is what put it in orbit. The pattern was found, then understood, then used.

What was understood along the way is that the identity of the panels is not decoration. Everyone who has written the closure conditions for the pattern has written them with one panel in them, and the derivation does not survive the panels being merely similar. That is stated in the literature as a condition of the derivation rather than as a result, which is exactly the sort of thing that goes unread — it is a hypothesis, and hypotheses are what a reader skips.

The reading offered here is that the hypothesis is the result. The identical-panel condition is not a simplification made for tractability; it is the whole of what makes the sheet a mechanism, and a measurement of how badly a non-identical pattern fails is a measurement of how much the condition is worth. It is worth everything: the failure is proportional to the departure, with no discount at small sizes.

Where the ladder goes next #

The immediate continuation is what the exactness costs to build. A pattern that forgives nothing has to be cut by something that misses nothing, and since nothing misses nothing, the room has to be put in at the hinges on purpose — which is the same finding read as a specification rather than as a fact about geometry.

The other direction is toward what is bought. A sheet whose behaviour is set by its creases rather than by its material is a material made of creases, with a property that can be chosen rather than looked up, and the price of that catalogue is the exactness measured above.

The wider question is what else sits at such a point. Rigid patterns are found rather than designed, and the ones in everything that gets built are a very short list. That list is short because a rigid folding is not something a pattern nearly has, and the search for more of them is a search over exact objects in a space where almost everything is a folded state no motion can reach.