The sheet has a thickness
An ideal fold brings two panels into contact along a line. Both panels are surfaces with no thickness, they meet exactly, and the crease is a line about which one rotates onto the other.
Take a piece of card and try it. The two halves do not meet along a line; they meet along a curve of finite radius, the outer surface has to travel further than the inner one, and the fold does not close completely.
For one fold in thin paper this is negligible. For forty layers in aluminium it is the whole problem.
Where the error comes from
The mid-surface of a folded sheet is what a crease pattern describes. The material extends a distance either side of it, and when the sheet folds, the outer face is on a larger radius than the inner one.
The consequence is that the two panels, folded to 180°, are offset by roughly the thickness. Their edges no longer coincide with the crease line — they are displaced by an amount comparable to , and the geometry of the pattern is wrong by that much.
One fold, one thickness of error. Two folds in the same region, two. And the errors do not cancel: each fold pushes the material further from where the pattern said it would be.
Layers accumulate
The multiplication is what makes this severe rather than fiddly.
A flat-folded model has a layer count at every point — how many sheets of material are stacked there. For a simple model that is two or four. For a complex box-pleated design it reaches thirty or forty where several flaps meet.
Forty layers of ordinary 80 gsm paper is about four millimetres. If the flaps at that point are five millimetres wide, the model is nearly as thick as it is wide, and the geometry that the crease pattern described has essentially nothing to do with the object.
This is why complex origami is folded from very thin paper — tissue laminated to foil, at perhaps a fifth the thickness of copier paper — and from very large squares. Both moves reduce the ratio of thickness to feature size, and neither eliminates it.
The problem is worst where the design is best
There is an unpleasant coupling here, and it constrains design more than anything else at high complexity.
A packing that is efficient is one where the circles are tight and the leftover paper is minimal. Tight packing means flaps meeting closely, which means layers concentrating, which means thickness.
So the more efficiently a design uses its paper, the worse its thickness problem. Designers routinely detune a packing — accepting eighty per cent of the achievable efficiency — specifically to spread the layers out.
That is an optimisation problem the theory does not describe. Efficiency is computable and layer distribution is computable, and nobody has a method that trades them off well.
Counting layers
Before the problem can be managed it has to be measured, and the measurement is a property of the crease pattern rather than of the material.
Every point of a flat-folded model has a layer count: how many sheets lie above it in the stack. The count is determined entirely by the pattern and the assignment, and it can be computed — for patterns where the layer ordering is known, which is not the general case but covers most designed models.
Typical numbers are worth having. A crane has four layers at the body and two at the wings. A traditional frog base reaches eight. A moderately complex insect reaches twenty, and the most complex published designs reach forty or more where several appendages meet.
The distribution matters as much as the maximum. A model with forty layers in one spot and two everywhere else folds much worse than one with a uniform twelve, because the thick region resists while the thin region gives, and the whole thing distorts.
The material response
Folders respond to thickness by changing the material rather than the design, and the range of options is wider than it looks.
Thinner paper. Ordinary copier paper is about 0.1 mm. Kami is thinner. Tissue is 0.03 mm or less, and forty layers of it is a millimetre rather than four.
Foil lamination. Tissue bonded to kitchen foil gives a sheet that is very thin, holds a crease absolutely, and does not spring back. It is the standard material for complex work and it looks like nothing anybody would call paper.
Larger squares. Doubling the sheet halves the ratio of thickness to feature size at the cost of a model twice as big. Complex models are large for this reason rather than for effect.
Wet-folding. Damp paper compresses at the crease more than dry paper, which reduces the effective thickness at the fold and lets thicker stock be used.
Each is a way of changing the thickness-to-feature ratio, which is the number that actually governs.
Accommodating it in hardware
For engineered folds the problem is severe enough to have its own literature, and the techniques share a strategy: move the hinge away from the mid-surface.
Offset panels. Thicken each panel away from the mid-surface and place the hinge at the surface where the two panels will meet. The panels then rotate about the right line and close properly. It works well and it makes the panels asymmetric, which complicates manufacture.
Membrane hinges. Replace the crease with a thin flexible strip joining two thick rigid panels. The strip absorbs the thickness by bending over a finite length, and the panels stay planar. This is the most common approach in practice and is what most deployable hardware uses.
Tapered panels. Cut the panels so their edges are angled rather than square, letting them nest. Elegant, and it constrains the fold angle to the value the taper was cut for.
Hinge shifting. Move each hinge to whichever surface the fold requires — outer for a mountain, inner for a valley. Simple in principle and it means the two faces of a panel are no longer equivalent.
Each restores the ideal kinematics at the cost of manufacturing complexity, and each works only for a rigid-foldable pattern — one that relies on the material bending has no hinge line to move.
Thickness changes what is foldable
The important consequence is not that thick folds are imprecise. It is that thickness changes the set of foldable patterns.
A pattern that folds flat as a zero-thickness surface may have no valid configuration at all once the material has a depth, because two panels that should be adjacent now overlap. The flat-folding theorems say nothing about this — they are statements about a surface.
So there is a genuine gap between “this pattern folds flat” and “this pattern can be built”, and the gap widens with thickness and with layer count. A great deal of the engineering literature is about closing it for particular patterns.
What paper does instead
Paper has a thickness too, and it copes by cheating in ways rigid panels cannot.
It compresses at the crease, so the material is thinner there than elsewhere. It stretches slightly on the outside of a fold, which no rigid panel does. And the layers slide against one another as the model closes, redistributing the error.
That is why paper models tolerate layer counts that would be impossible in any other material, and why a folder’s intuition about what is buildable is misleading for engineering. Twenty layers of paper is a thick model; twenty layers of anything rigid is a design failure.
The ratio that governs
There is one number that predicts whether a fold pattern will survive being built, and it is not the thickness.
It is thickness divided by feature size — the depth of the material against the smallest dimension the pattern asks for. A pattern with 5 mm features in 0.1 mm paper has a ratio of 1:50 and folds comfortably. The same pattern in 1 mm card has a ratio of 1:5 and does not.
That is why every response to the thickness problem is a way of changing the ratio. Thinner paper reduces the numerator. Bigger squares increase the denominator. Detuning a packing spreads the layers, which increases the effective denominator where it matters most.
And it is why the problem is scale-free in a useful way: a microscale folded device in silicon and a metre-scale solar array face the same difficulty if their ratios match, and completely different difficulties if they do not.
Layers are not evenly distributed
A statistic that misleads: the maximum layer count of a model.
What governs foldability is the distribution. A model with forty layers in one small region and two everywhere else behaves far worse than one with a uniform twelve, because the thick region is stiff, the thin region is compliant, and the mismatch distorts everything between them.
Designers therefore care about the layer map rather than the peak, and the good ones develop an intuition for it that the packing algorithms do not have. The efficiency an optimiser maximises is directly at odds with the evenness a folder wants, and nobody has a method that trades them off well.
There is a related effect on appearance. A region with many layers is visibly thicker in the finished model, and complex designs are often criticised for looking lumpy where the paper concentrates — a purely geometric consequence that reads as a stylistic failing.
What thickness does to the theorems
A precise statement of the damage is more useful than a general lament.
Kawasaki and Maekawa are statements about a surface, and they remain exactly true of the mid-surface of a thick sheet. Nothing about them fails.
What fails is the conclusion drawn from them. The theorems say the mid-surface can reach a flat configuration. They say nothing about whether the material either side of that mid-surface can occupy the space that configuration requires, and for a thick sheet with many layers it often cannot.
So thickness does not break the mathematics; it breaks the correspondence between the mathematics and the object. That is a subtler failure than an incorrect theorem and it is harder to notice, because every check passes.
Where the model stops
Uniform thickness. All the accommodation techniques assume a constant material depth. Laminates, honeycombs and anything with a varying section need their own treatment.
Flat panels. The techniques assume the panels are flat plates. Curved panels are a different problem again.
Two panels at a time. The analysis is usually of a single fold between two panels. A vertex where four thick panels meet has interference conditions that pairwise analysis misses.
The figure is a cross-section. Everything drawn here is a slice through a fold, which shows the thickness problem clearly and shows nothing about how it propagates across a pattern.
No material behaviour. Thickness here is geometry. Real folding also involves the material yielding, the fibres breaking on the outside of a crease, and the crease acquiring a permanent set — none of which is drawn.
Thickness in a tessellation
The problem compounds in a way worth spelling out, because it sets the size limit on engineered folded sheets.
A Miura sheet with rows folds flat into a stack layers deep. Not four, not twenty — , the number of rows, because each row folds onto the one before it.
So a 40-row Miura in 0.1 mm paper is 4 mm thick when packed, and in 1 mm panels it is 40 mm. The packed thickness scales linearly with the pattern’s extent, which means the packing ratio in area is excellent and the packing ratio in volume is not.
For a spacecraft array that is the binding constraint: the fairing has a diameter and a length, the array packs to a stack, and the stack’s depth is what has to fit. Designers therefore split large arrays into several smaller folded panels rather than one large one, purely to keep the stack depth down.
The gap between the theorems and the object
It is worth being precise about what thickness breaks, because the usual phrasing is imprecise.
Kawasaki and Maekawa remain exactly true of the mid-surface of a thick sheet. Nothing about the theorems fails; the mid-surface is a surface and it satisfies them.
What fails is the inference. The theorems establish that the mid-surface can reach a flat configuration. They say nothing about whether the material either side of that surface can occupy the space the configuration requires — and for many layers it cannot.
So thickness does not falsify the mathematics; it breaks the correspondence between the mathematics and the object. That is a subtler failure and a more dangerous one, because every check passes and the object still does not exist.
It is also the reason engineering origami is a separate discipline rather than an application of the geometry. The geometry is correct and insufficient, and the additional work is not a refinement but a different problem.
What gets built instead
When accommodation is not enough, designers change the pattern rather than the material, and the moves are worth knowing.
Fewer layers. Redesign so that no region accumulates more than a set count. That costs efficiency and is the most common response.
Split the structure. A large Miura array becomes several smaller folded panels, each with a shallow stack, joined by conventional hinges. The fold does the area reduction and the hinges do the rest.
Different pattern. Some patterns concentrate layers and others distribute them. Choosing on that basis rather than on packing ratio is common in hardware and rare in the geometry literature.
Accept a thicker packed state. Sometimes the volume is available and the constraint was imaginary.
All four are ways of saying that the crease pattern is a starting point rather than a specification, which is the working attitude in the field.
The number to quote
If one figure is going to be carried away from this, it should be the ratio rather than any thickness.
Thickness divided by the smallest feature the pattern asks for. Under about 1:50 and a design folds comfortably; around 1:20 it becomes difficult; below 1:10 it does not work in any material without accommodation.
That single number explains the practices. Complex models use tissue-foil because it moves the ratio. They use large squares because it moves the ratio. Designers detune packings because it moves the ratio where the layers concentrate. Engineers use membrane hinges because they cannot move the ratio and must change the geometry instead.
It also explains why the same crease pattern can be trivial at one scale and impossible at another, which is otherwise puzzling — the geometry has no length scale in it and the material does.
The ladder from here
Later rungs: the offset-panel method derived. Membrane hinges and their design. Tapered and nested panels. Hinge shifting. Thickness in tessellations, where it multiplies. Layer counting as a design objective. Compliant mechanisms. Materials for thick origami. Manufacturing methods. And the question of whether thickness accommodation can be automated, which is partly solved for particular pattern families and not in general.
The crease patterns in this subject are exact statements about a surface that has no thickness, and every object anybody has ever folded is made of something. The whole engineering field is the correction term.