Flat-folding

The smallest sector decides

Two assignments can satisfy both flat-folding theorems and only one of them folds. What separates them is a condition about the smallest angle, and it is the first rule in the subject that is not about counting.

Kawasaki settles the angles. Maekawa settles the counts. Between them they look like a complete answer, and they are not.

The smallest sector decidesTwo assignments of the same four creases. Both satisfy Kawasaki and Maekawa. The left one folds; the right one does not, because the strictly smallest sector has the same assignment on both sides and the paper either side of it has nowhere to go.MVMM40°foldsopposite across the small sectorMMVM40°does not foldthe same on both sidesboth satisfy Kawasaki and Maekawa — the angles and the counts are identical
Fig. 1 Two assignments of the same four creases. Both satisfy Kawasaki — the angles are identical. Both satisfy Maekawa — three of one and one of the other. The left one folds and the right one does not.

Something else is going on, and it concerns the smallest angle at the vertex.

The condition

If a sector is strictly smaller than both of its neighbours, the two creases bounding it must have opposite assignment.

That is the big-little-big lemma. It is a statement about three consecutive sectors — big, little, big — and about the two creases on either side of the little one.

The reason is mechanical rather than arithmetic. Fold the two large sectors toward each other and the small one between them has to go somewhere. If both bounding creases fold the same way, the small sector is trapped: the paper on either side of it arrives from the same direction and the sector has to occupy space that the neighbouring panels already own.

If the bounding creases fold oppositely, the small sector tucks between the two large ones and there is room.

What “does not fold” means here

It is worth being precise, because “does not fold” can mean several things.

The right-hand vertex in the figure is not merely difficult, or a matter of technique. It is geometrically impossible: pressing it flat would require two panels of paper to occupy the same region of the plane, and paper does not do that.

Try it and the vertex will fold to nearly flat, with a small pocket of paper that will not lie down. Force it and the sheet tears or a crease shifts. That failure mode is familiar to anyone who has folded from a crease pattern with an assignment error, and this lemma is what it is a symptom of.

Where it bites, and where it does not

The condition only applies when a sector is strictly smaller than both neighbours. Ties let it off.

That exemption does a great deal of work. In the preliminary base all eight sectors are 45°, so no sector is strictly smallest, the lemma is vacuous, and the local conditions reduce to Maekawa alone — which is why the count comes out at exactly (83)+(85)=112\binom{8}{3} + \binom{8}{5} = 112.

The preliminary baseBoth diagonals and both midlines of a square, with the assignment that folds flat. Eight creases meet at the centre in equal sectors, so Kawasaki is satisfied by any assignment and Maekawa is the binding condition — five of one and three of the other, never four and four.at the centre8 creases, all sectors 45°3 mountain, 5 valleydifference 2 — Maekawa holdsfour and four would fail,which is what most people drawfold every line, then collapse — the four corners meetmountainvalleyraw edge
Preliminary base — sheet 150×150 mm — 3 mountain, 5 valley, 724.26 mm of crease
Fig. 2 Equal sectors throughout, so the big-little-big lemma has nothing to bite on. Everything constraining this vertex is the mountain-and-valley count.

In the Miura fold the four sectors come in two equal pairs, and the two small ones are adjacent — so neither is strictly smaller than both of its neighbours, and again the lemma is silent. That is not a coincidence: the Miura’s assignment has freedom precisely because its geometry avoids the condition.

How much it removes

Where it does apply, the effect is substantial.

A degree-four vertex with unequal sectors has 16 assignments. Maekawa allows 8 of them — the three-and-one splits. Big-little-big requires the two creases bounding the smallest sector to differ, which halves it again to 4.

How many assignments actually foldFor a fixed set of crease lines, the number of mountain-and-valley assignments that satisfy the local conditions, against the number of assignments there are. The valid ones are a small and shrinking fraction, which is the quantitative form of the claim that flat-foldability is rare.one degree-4 vertex4 of 164 creases · 25.0% surviveone degree-6 vertex8 of 646 creases · 12.5% survivethe preliminary base112 of 2568 creases · 43.8% surviveand these are only the local tests — a pattern can pass every vertexand still collide once the layers stack, which is the hard part
Fig. 3 Locally valid assignments against all assignments. For the degree-four vertex the count is 4 of 16; Maekawa alone would have allowed 8, and the other four are removed by the smallest sector.

That is the third condition doing as much work as the second. And it explains a fact about degree-four vertices that is otherwise mysterious: the odd crease out is not free to be any of the four, it must be one of the two bounding the smallest sector, so a flat-foldable degree-four vertex has essentially one shape and two mirror images.

Every vertex passes, which is not enoughThe local conditions are checked at each vertex independently, and a pattern can satisfy all of them and still fail to fold, because the layers have to stack without passing through one another. Deciding that for a general pattern is NP-hard, so no figure can settle it.6 interior vertices, every one satisfying both theoremswhat the local tests seeangles at each vertexassignment at each vertexwhat they cannot seewhether layer 3 passes through layer 7whether a flap has room to existwhether the order is consistent everywhereBern and Hayes, 1996: NP-hardso this pattern is checked, not proved
Fig. 4 Where the local conditions run out. Every vertex here passes all three, and that is necessary rather than sufficient.

The generalisation

Big-little-big is the simplest member of a family, and the family is what a general algorithm needs.

The full condition — Justin’s — concerns any sequence of sectors that could collide, not merely the strictly-smallest single one. Equal adjacent sectors can also cause trouble in combination, and the general statement is about which sequences of creases can be “cancelled” as the sheet collapses.

That generalisation is the basis of the standard single-vertex algorithm: repeatedly find a sector that is a local minimum, check the condition, cancel it by folding the two neighbours together, and recurse on the smaller vertex. A vertex folds flat exactly when the recursion terminates.

The algorithm is linear in the number of creases, which is why the single-vertex problem is easy and the whole-sheet problem is not.

Why this one is different in kind

Kawasaki and Maekawa are equalities. Big-little-big is an inequality with a case analysis, and that difference matters.

An equality is checkable by arithmetic and generalises cleanly. An inequality about local minima is a combinatorial condition — it depends on the ordering of the sectors, not just their values, and it is the first place in this subject where the answer depends on arrangement rather than on quantity.

That is also why it is the last of the three to be found, and the one folders are least likely to know. A crease pattern with a Kawasaki error is obviously wrong; one with a Maekawa error is obviously wrong once counted; one with a big-little-big error looks entirely reasonable and fails in the hands.

A vertex that folds flatFour creases at one point, with the sectors between them measured and both flat-folding conditions evaluated. Kawasaki constrains the angles and Maekawa constrains the assignment; a vertex needs both, and they are independent of one another.VMMM60°90°120°90°Kawasaki60° + 120° = 180°90° + 90° = 180°both 180° — satisfiedMaekawa3 mountains, 1 valleysdifference 2exactly 2 — satisfiedangles sum to 360°which is what a flat sheet requiresmountainvalley
Fig. 5 The two arithmetic conditions, evaluated. Neither of them looks at the ordering of the sectors, which is exactly the information the third condition needs.

The physical picture

It helps to see what the collision actually is.

Consider three consecutive sectors of sizes big, little, big. When the vertex folds flat, the little sector ends up sandwiched. The two creases bounding it fold the large sectors down over it — and the question is whether they fold down onto the same side of the little sector or onto opposite sides.

Same side: both large panels arrive from above, and the little sector is squeezed from one direction with nowhere to escape. The two large panels then have to overlap each other in the region where the small sector should be, which is the self-intersection.

Opposite sides: one large panel comes from above and one from below, and the little sector lies between them as a layer. That is a legitimate stack.

So the condition is a statement about layer order in disguise, which is a hint about what the hard part of the general problem is.

Ties, and why they are load-bearing

The lemma exempts sectors that are merely equal to a neighbour rather than strictly smaller, and that exemption deserves more than a footnote, because the two most important patterns in the subject both rely on it.

The preliminary base has eight equal sectors. Nothing is strictly smallest, the lemma is silent, and 112 assignments survive rather than a handful.

The Miura fold has four sectors in two equal pairs, arranged so the two small ones are adjacent. Neither is strictly smaller than both neighbours — each has the other as one neighbour — so again the lemma says nothing, and the pattern has the assignment freedom it needs to tile.

Neither of those is a coincidence. Symmetric patterns tile, tiling requires assignment freedom, and assignment freedom is what the lemma removes. A tessellation that violated the ties would be constrained at every vertex and would have nowhere to go.

So the exemption is not a technicality in the statement — it is the reason a large part of the subject exists.

Who found it, and when

Jacques Justin stated the general condition in 1994, in a paper that also contains a good deal of the rest of the local theory. Thomas Hull worked out the single-vertex algorithm and the counting results through the late 1990s.

The name “big-little-big” is folklore rather than anybody’s coinage, and it is descriptive enough that it has stuck. The formal literature usually calls it a case of Justin’s non-crossing condition, which is more accurate and less memorable.

It is later than the other two theorems by about five years, which is roughly the time it took the field to move from “here are two conditions” to “here is a complete algorithm”.

Trying it

This condition is the easiest thing in the subject to verify by hand, and doing so is worth more than reading about it.

Take a square of paper. Mark a point in the middle and draw four creases from it at 40°, 110°, 140° and 70° — the angles in the figure above. Kawasaki holds: 40 + 140 = 180 and 110 + 70 = 180.

Now fold it two ways. First with the two creases bounding the 40° sector assigned oppositely: one mountain, one valley, and the other two mountains. It collapses flat without complaint.

Then reverse one of them so the small sector is bounded by two mountains, keeping three-and-one so Maekawa still holds. It will not go down. There is a pocket of paper at the middle that refuses to lie flat, and pressing harder produces a new crease somewhere the pattern did not ask for.

That is the condition, felt rather than read, and it takes about two minutes. It is also the clearest demonstration available that these theorems describe something physical — the paper enforces the lemma whether or not anybody knows it.

The order the layers end up in

The physical argument above hints at something the condition does not state, and it is worth following.

When the small sector tucks between its two neighbours, it ends up as a layer — one sheet in a stack of three at that point. Which of the three is on top is determined by the assignment, and for a single vertex the answer is forced.

For two vertices sharing a crease, each imposes a layer ordering on the shared region, and the two orderings have to agree. That is a constraint the local conditions never see, because each is evaluated at one vertex with no knowledge of the other.

And with many vertices the orderings form a system of constraints that can be inconsistent in ways no individual vertex reveals — which is exactly where the problem stops being easy and starts being NP-hard. Big-little-big is the last condition that can be checked without thinking about layers, and it is already halfway into the question.

Where the model stops

Strictly smaller, only. Ties are exempt, and the exemption is used constantly — both the preliminary base and the Miura fold escape the condition entirely through equal sectors.

One vertex. Like the other two, this is a local condition. Satisfying it everywhere does not make a pattern foldable.

It hides layer ordering. The physical argument is about which side each panel arrives from, which is a layer question. The condition summarises the simplest case of that and does not solve it in general.

Zero thickness. With real paper the sandwiched sector has a stack of finite height above and below it, and a vertex that is geometrically fine can still be physically impossible in thick material.

The figure shows one pair. Two assignments of one vertex is enough to show the condition exists and not enough to show how much of the space it removes; that is what the counting figure is for.

Three filters, and what survives

With the third condition in hand it is worth collecting what the local theory removes.

Start with 2n2^n assignments of nn creases at a vertex. Maekawa keeps the splits differing by two, which is a little under half. Kawasaki is a condition on angles rather than assignment, so it either passes everything or nothing — it filters geometries, not assignments. Big-little-big removes those where the smallest sector is badly bounded, roughly halving what is left where it applies.

So for a degree-four vertex: 16 assignments, 8 after Maekawa, 4 after big-little-big. A quarter survive.

How many assignments actually foldFor a fixed set of crease lines, the number of mountain-and-valley assignments that satisfy the local conditions, against the number of assignments there are. The valid ones are a small and shrinking fraction, which is the quantitative form of the claim that flat-foldability is rare.one degree-4 vertex4 of 164 creases · 25.0% surviveone degree-6 vertex8 of 646 creases · 12.5% survivethe preliminary base112 of 2568 creases · 43.8% surviveand these are only the local tests — a pattern can pass every vertexand still collide once the layers stack, which is the hard part
Fig. 6 The survival rate for three patterns. It is the local conditions only, and even so most assignments are gone.

Now put many such vertices in a sheet with shared creases, and the survivors have to agree with each other. That compounding is where flat-foldability becomes genuinely rare — not because any one vertex is hard to satisfy, but because they all have to be satisfied at once with a shared assignment.

The condition inside a real fold

Anyone who has folded a squash fold or a petal fold has performed this lemma without naming it.

A squash fold takes a flap, opens it, and flattens it symmetrically — and the reason it works is that the small sector between the two creases being opened is bounded by creases of opposite assignment. A petal fold is a pair of them.

Reverse one of those creases and the manoeuvre stops working. The paper will not sit down, and a folder discovers this in the first thirty seconds of trying, long before consulting a theorem.

That is the pleasant thing about this condition in particular: it is the one that experienced folders already know in their hands. Kawasaki and Maekawa are facts about a pattern on a table; big-little-big is a fact about what happens when the paper is pressed.

Three conditions, and a fourth that is not there

Having assembled the local theory it is worth asking what is missing, because the absence is structural rather than accidental.

Developability constrains the angles to sum correctly. Kawasaki constrains their alternating sums. Maekawa constrains the counts. Big-little-big constrains the arrangement around the smallest sector.

Together they are sufficient at a single vertex — there is a constructive algorithm and a proof. So for one vertex the theory is complete and there is nothing missing.

For two vertices there is no fourth condition, and there cannot be a clean one, because the extra constraint is that the two vertices’ layer orderings agree — and layer ordering is where the intractability lives.

That is why the list stops at three. It is not that nobody has found the next one; it is that the next thing to check is a global consistency condition on an object the vertex conditions never mention, and checking it in general is NP-hard.

Why it is called that

A note on the name, because it is unusually descriptive for a mathematical condition and the description is doing work.

“Big-little-big” names the configuration the condition applies to: three consecutive sectors where the middle one is smaller than both neighbours. That is the situation in which the paper can collide, and naming the situation rather than the conclusion is what makes the condition memorable.

Compare Kawasaki and Maekawa, which are named after people and give no hint of what they say. A folder who has heard “big-little-big” once can reconstruct roughly what it must be about; nobody reconstructs Maekawa’s theorem from its name.

The formal literature calls it a case of Justin’s non-crossing condition, which is more accurate and which nobody uses. That is usually how it goes when a descriptive folk name competes with a correct one.

The ladder from here

Later rungs: Justin’s general non-crossing condition. The single-vertex flat-foldability algorithm, and its linear running time. Counting flat-foldable assignments at a vertex of degree nn. The interaction of the three conditions. Layer ordering as the underlying question. Generic versus non-generic vertices, and why ties matter so much. The two-vertex case, which is already harder. And the point at which local conditions stop being sufficient, which is exactly two vertices.

The condition is a sentence long, it decides half the assignments a degree-four vertex could have, and it took the field a decade after Maekawa to write down. It is not a difficult idea; it is one that nobody thought to look for until an algorithm needed it.