What the vertex does on the way

Flat-foldability asks whether a vertex can reach a flat state. For anything that gets built, that is the wrong question: a solar array spends almost none of its life flat and almost all of it somewhere in between, and what matters is the path.

The path is computable, and computing it is more informative than it sounds, because two theorems that are normally proved about the destination turn out to be visible all the way along.

Forget the paper, keep the directions #

The standard move in rigid origami is to throw away everything except the crease directions, and it is worth seeing why it loses nothing.

Draw a small sphere around the vertex. Each crease pierces it at a point. Each sector of paper between two creases becomes an arc joining two of those points, and the arc’s length is the sector angle.

Folding moves the points around the sphere. It cannot change the arc lengths, because the sectors are rigid panels and their angles are fixed. So the whole motion of the vertex is the motion of a closed polygon on a sphere with fixed side lengths — a linkage, on a sphere instead of on a table.

That reduction is exact for a rigid vertex. Everything about the folding, including which crease is a mountain and how far each one has turned, is in the spherical polygon.

Solving the closure #

The polygon has to close, and closing it is the computation.

Put the first crease along a fixed direction and the second at the sector angle from it, in a plane — that is the flat state. Now rotate everything beyond the second crease about it by an angle: that is the drive parameter, and it is the one thing a person or a motor controls.

The third crease has moved with the rotation and its position is known. The fourth is not free: it has to sit at a prescribed angle from the third and at another prescribed angle from the first, which puts it on the intersection of two circles on the sphere.

Two circles meet in nothing, once, or twice. Twice is the ordinary case, and the two solutions are the two branches the vertex can pop into. Nothing at all means the drive has been pushed past the point where the linkage can close, which is a mechanism reaching the end of its travel.

So the vertex has one continuous parameter and one binary choice. That is what “one degree of freedom” means for a real mechanism, and it is why a degree-four vertex can be built out of panels and hinges.

Maekawa, arriving uninvited #

With the configuration solved, the fold angle at each crease can be measured: flatten both of its neighbours into the plane square to it, take the angle between them, and see how far short of a straight line it falls. The sign says which way the paper turns, which is the mountain-valley distinction.

Do that at every point of the motion and the signs come out three of one and one of the other. Always. Not at the end, not at the flat-folded state — throughout.

Maekawa’s theorem says exactly that, and it is normally proved by a winding argument about the folded cross-section: walk round the flat-folded vertex, count half-turns, observe that the walk closes, conclude that the counts differ by two. That proof is about a flat state and says nothing about the journey.

Here the same statement falls out of a linkage closure with no reference to a flat state at all. Nothing in the computation knows about winding numbers or about mountains and valleys; it solves for four directions in space and the sign pattern is read off afterwards.

Which is a stronger result than the theorem as usually stated. The assignment is not something the vertex acquires when it arrives flat. It is fixed the instant the vertex leaves the plane and it does not change.

Kawasaki, as the condition for arriving #

The second theorem appears as an endpoint rather than as a pattern.

Drive a vertex whose sectors satisfy Kawasaki’s condition — opposite sectors supplementary — and every fold angle reaches a full half-turn together. The vertex arrives flat.

Drive one whose sectors do not, and it stops early. Either the linkage runs out of configurations, or it keeps going while one crease is still barely folded, and either way the vertex never reaches a state where all four creases are fully turned.

The generator asserts the equivalence rather than illustrating it. Before drawing, it evaluates Kawasaki on the sector angles and separately asks the linkage whether the least-folded crease reaches a half-turn. If the two disagree it throws, because two independent computations of the same fact have no business disagreeing.

That check has a useful property: it would fail loudly if either the theorem were misstated or the solver were wrong, and it passes for every sector list the figures use.

Where the two circles stop meeting #

The closure can be pushed one step further, and doing so turns “the linkage runs out of configurations” into a formula.

Call the sectors $\alpha_1 \dots \alpha_4$ in order and let $d$ be the spherical distance between the first and third creases — the polygon’s diagonal. The first two arcs reach across it, so the triangle inequality on the sphere requires $|\alpha_1 - \alpha_2| \leq d \leq \alpha_1 + \alpha_2$; the other two arcs impose the same with $\alpha_3, \alpha_4$.

Developability makes those two intervals meet in a way they otherwise would not. Since the sectors sum to a full turn, $\alpha_1 + \alpha_2 = 360° - (\alpha_3 + \alpha_4)$, and a spherical distance is at most $180°$, so both upper limits are the same number, $\min(\alpha_1+\alpha_2, \alpha_3+\alpha_4)$ — which is the unfolded sheet. Folding drives $d$ down from there, and it stops at

$$d_{\min} = \max\big(|\alpha_1 - \alpha_2|,\; |\alpha_3 - \alpha_4|\big).$$

That is the jam, and the formula says what jamming is: one half of the linkage has folded onto itself while the other still has travel left, so the two circles pull apart and there is no configuration to be in.

Which gives Kawasaki a second derivation #

A vertex reaches a flat-folded state only if both halves run out at once — all four creases on one great circle — so it needs

$$|\alpha_1 - \alpha_2| = |\alpha_3 - \alpha_4|.$$

Kawasaki’s condition supplies that identically. With $\alpha_1 + \alpha_3 = \alpha_2 + \alpha_4 = 180°$, the difference is $\alpha_3 - \alpha_4 = (180° - \alpha_1) - (180° - \alpha_2) = \alpha_2 - \alpha_1$, and the two magnitudes agree. So the flat state is where the two halves of the mechanism exhaust their travel together, and the alternating-sum condition is the arithmetic that arranges it — derived here from a distance rather than from a walk round a folded cross-section.

The reverse does not hold, which is worth stating because it is the kind of thing a reader would otherwise assume. The equality of magnitudes has a second solution, $\alpha_1 + \alpha_4 = \alpha_2 + \alpha_3 = 180°$, and it is not Kawasaki. Sectors of $100°, 90°, 90°, 80°$ satisfy it: both halves run out at $d = 10°$, and the alternating sum is $20°$ rather than zero, so the vertex does not fold flat. The two halves arrive at the same distance in opposite orientations, and the distance test cannot see an orientation.

So the linkage supplies a necessary condition and Kawasaki is the sign-consistent half of it — which is the same relationship the two theorems have everywhere else in this subject, arriving here as a triangle inequality.

The three-to-one split is not universal #

There is a limit to the previous section that is worth stating, because getting it wrong is easy and this site did.

The sign split is a property of vertices that can fold flat. A developable degree-four vertex that fails Kawasaki is a working mechanism whose fold angles can come out two and two, and asserting a three-to-one split for every developable vertex is simply false.

That was in an earlier draft of the generator as a hard assertion, and it fired on the first non-Kawasaki example. The fix was not to weaken the assertion but to scope it: the check now applies exactly when the vertex reaches a flat state, which is the case the theorem is about.

The distinction is the essay’s point in miniature. Maekawa is a theorem about flat-foldable vertices, and a mechanism that never folds flat is outside its scope while remaining entirely legitimate as a mechanism. Flat-foldability and rigid-foldability are different properties and the theorems attached to one do not transfer.

The two branches are two different models #

The binary choice deserves more attention than it usually gets, because it is where self-folding fails.

At any point of the motion the vertex sits on one of two branches. Both are legal, both keep every panel rigid, both reach a flat state if the sectors permit. They differ in the assignment: one gives mountain-valley-mountain-mountain, the other mountain-valley-valley-valley, and folded flat they are two different objects.

A degree-four vertex therefore has four flat-folded states in all — two branches, each reachable by driving the parameter either way — and enumerating the valid assignments by the local conditions gives exactly four as well. The two computations share no code and no idea: one is combinatorics over letters, the other is a closed spherical linkage. Their agreeing on both the count and the identity of the four is the strongest single check on this site.

For an engineer the branch is the problem. A mechanism that can pop into either of two configurations under the same actuation is a mechanism that needs to be told which, and telling it is not a matter of applying more force.

Driving a whole pattern #

One vertex is a mechanism with one degree of freedom. A pattern of many is not simply more of the same, and the difference is the reason rigid origami is a subject.

Every vertex in a pattern is its own linkage, and adjacent vertices share creases. A shared crease has one fold angle, so the two linkages have to agree about it — which is a constraint linking the two mechanisms. Chase those constraints across a pattern and the degrees of freedom fall fast.

The usual outcome is zero. A generic pattern of several vertices has no continuous motion at all: it is rigid, and the only way to fold it is to let the panels bend. That is why rigid-foldability is a much stronger property than flat-foldability and why so few patterns have it.

The patterns that survive are the ones with enough symmetry for the constraints to be redundant. A Miura fold is one degree of freedom however large it is, because every vertex is a translate of every other and their constraints all say the same thing. Break the symmetry — vary the panel angles across the sheet — and the motion usually disappears.

So the useful summary is that rigid-foldable patterns are rare, and the ones in use are periodic almost without exception. Tachi’s solvers exist precisely to find the exceptions.

What a generic vertex does #

There is a case worth naming that neither of the two theorems covers, because it is the one an engineer meets most often.

Take a developable degree-four vertex with sectors chosen at random. It is a mechanism: it has one degree of freedom, it moves, no panel bends. It does not satisfy Kawasaki, so it never reaches a flat state, and it stops somewhere in the middle of its travel.

Where it stops is a useful number. It is the deployment limit of anything built on that vertex — the angle past which the linkage has no configuration and the structure jams. For the seventy-eighty-one-twenty-ninety vertex drawn above that is a hundred and nine degrees, and no amount of force will get past it.

That number is invisible to the flat-folding theory, which reports only that the vertex fails Kawasaki and stops there. The kinematics gives the angle, and the angle is what a designer needs.

It also reframes what Kawasaki’s condition is for. Read as a flat-folding condition it is a yes-or-no test. Read as a kinematic one it is the boundary case of a continuous quantity: how far a developable vertex can fold, with Kawasaki-satisfying vertices at the maximum.

Where the model stops #

Degree four only. Vertices of higher degree have more than one degree of freedom in general, and the closure argument above does not extend directly — there is no two-circle intersection to solve.

One vertex. A pattern of many vertices is a system of linkages sharing creases, and the number of degrees of freedom is not the sum of the individual ones. It is usually much smaller and sometimes zero.

Rigid panels. The whole reduction assumes the sectors do not bend. Paper bends a little, which is why paper folds patterns that panels cannot.

No thickness, no collisions. The solver finds configurations that satisfy the angles. It does not check whether the panels pass through each other on the way, which for a vertex is rarely a problem and for a pattern frequently is.

Nothing about which branch. The solver returns both, and choosing between them is the whole difficulty of self-folding.

Nothing about thickness. The panels have none, so the accommodation problem is invisible here and unavoidable in anything built.

Nothing about force. Kinematics says where things can be, not what it takes to put them there. Stiffness, torque and the energy stored in a crease are a separate model.

The surprise: the theorems are about the wrong thing #

The conventional order of presentation puts the flat-folding theorems first and rigid origami afterwards, as an application with extra constraints. The kinematics suggests the opposite reading.

Maekawa’s sign split is a fact about the mechanism, visible from the moment the vertex leaves the plane. Kawasaki’s condition is a fact about where the mechanism can get to. Neither is fundamentally about a flat state; the flat state is where both happen to be easiest to see and where they were first noticed.

Read that way, the local theory of flat-folding is a set of statements about spherical linkages that happen to have been discovered by people looking at folded paper. The winding proof of Maekawa and the alternating-sum proof of Kawasaki are proofs about the endpoint because the endpoint is what a folder holds in their hand.

Which is not a criticism of the theorems. It is an observation that the same facts have two derivations of quite different character, and the mechanical one says more.

What a designer takes from it #

Three things come out of the kinematics that a designer can use directly, and none of them is visible in the flat-folding theory.

The first is the travel: how far a vertex folds before it jams, which sets the deployment range of anything built on it. For a Kawasaki-satisfying vertex that is the full half-turn; for anything else it is a number the solver reports and the theorems do not.

The second is the coupling: how far the other three creases have moved when one has moved a given amount. That is what a mechanism’s designer needs to size an actuator, and it is not linear — near the flat state a small drive produces a large response at some creases and almost none at others.

The third is the branch, which is where self-folding fails and which the flat theory records only as “there are four valid assignments” without saying that two of them are on the far side of a choice the hardware has to make.

Who found it, and when #

The spherical-linkage view is old and the origami application is recent.

Spherical mechanisms are nineteenth-century kinematics, and the closure equations for a spherical four-bar are standard material in the theory of machines. Nothing in the computation above would have surprised anybody in 1900 — except that anybody would want to do it.

The application to folding came with rigid origami as a subject. Tomohiro Tachi’s work from the mid-2000s onward put it on a computational footing, with a solver that handles whole patterns rather than single vertices; Thomas Hull and others developed the mathematical account, including the explicit relations between the fold angles at a degree-four vertex.

The Maekawa connection is not usually presented as it is here. The standard statement is the flat-state one, and the observation that the sign pattern is constant along the motion is available in the kinematics without being the point anybody was making.

The ladder from here #

Later rungs against this anchor: the explicit closed forms for the degree-four fold-angle relations, and what makes them the shape they are. Vertices of degree six and higher, where the degree of freedom count is not one. Whole-pattern rigid folding, and how a solver propagates. Multi-degree-of-freedom patterns and how they are constrained back to one. Self-intersection during the motion, which the closure ignores. And the inverse question — given a desired motion, find the vertex — which is a design problem and much harder.

The vertex conditions are three lines of arithmetic about a flat sheet. The same facts, read off a linkage, describe a machine.