How short a paper Möbius band can be

The surface with one side began this shelf with a strip of paper, half a twist and a join. Every essay since has treated the Möbius band as a topological object: a rectangle with one pair of edges glued after a flip, the piece that sewn to a disc closes up into a projective plane, a surface whose orientation is a sign that comes back reversed, a line bundle over a circle on which every section must vanish, and finally a polynomial of signs deciding how much room a projective space needs. None of them cared about the paper. A topological band can be any shape, stretched as far as convenient.

Real paper cannot be stretched. It bends, it creases, and it keeps every length drawn on it. That one physical constraint turns the band into a question with a number for an answer. Take a strip of width 1 and length $\lambda$. If $\lambda$ is ten, twist it and join the ends: a Möbius band. If $\lambda$ is one, the strip is a square, and trying to join two opposite sides with a half-twist crumples or tears it. Somewhere between, the half-twist stops being possible. Where?

Benjamin Halpern and Charles Weaver posed the question in 1977, proved that $\lambda$ must exceed $\pi/2$, and conjectured that for a band that does not pass through itself the true threshold is $\sqrt 3 \approx 1.732$. Richard Schwartz proved the conjecture in 2023. The extremal shape behind that number can be made in a minute from a strip of paper, and it is the subject of the figure below.

Three folds and a triangle #

Draw three lines across the strip, each at sixty degrees to its length, leaning alternately one way and the other, so that they cut the strip into a row of equilateral triangles: half a triangle at the left end, two whole triangles, and half a triangle at the right end. Each triangle has height 1, the width of the strip, and side $2/\sqrt3$. Three triangles’ worth of area fit in the strip, and its length is three half-sides, $3 \cdot 1/\sqrt3 = \sqrt3$.

Fold along all three lines, each fold turning the paper fully over. Each fold reflects everything beyond it across the fold line, so the piece after the $j$-th fold is placed by a composite of $j$ reflections. Every piece lands on the same triangle; the strip lies flat in three layers. And the right end, carried by three reflections, arrives on the left end with its top and bottom exchanged: the corner marked A at the top left meets the corner marked A at the bottom right. Gluing the ends there joins the strip with a half-twist. The result is a Möbius band, made of real paper, never stretched, and creased along three lines.

The figure checks this by doing it. The three reflections are applied to the right-hand end of the strip, and the end lands on the left-hand end, flipped, to within the rounding of the arithmetic — about $10^{-16}$. The folded pieces fill a triangle of height exactly 1 and width $2/\sqrt3$. No fact about Möbius bands was used, only the reflections; the half-twist is a consequence of the arithmetic. The same device of following a path through a chain of reflections is the one that turns a billiard ball’s bounces into a straight line: there the table is reflected and the path unfolds, here the paper is reflected and the strip folds up, and in both the composite of the reflections is the only thing that has to be computed.

This is the triangular Möbius band, and it is not a curiosity at the edge of the problem. Schwartz’s theorem says more than that $\sqrt3$ is a lower bound: any sequence of smooth paper bands, free of self-crossings, whose length-to-width ratios approach $\sqrt3$ must approach this triangle. The best possible paper Möbius band is a creased, flattened object, and the smooth ones can only approximate it. That is unusual among extremal problems about surfaces: the answer is not a smooth shape that balances competing pressures, as a soap film is, but a degenerate one at the boundary of what counts as a surface at all. One-sidedness is a property of the whole loop, and it turns out to be bought most cheaply by a surface that gives up smoothness along three lines.

Why parity decides the twist #

The reason three folds give a half-twist is a sign. A reflection reverses orientation — it turns the paper over — and a composite of $j$ reflections reverses it when $j$ is odd and preserves it when $j$ is even. Orientation is a sign defined one-sidedness as exactly this: carry a frame round the loop and see whether it comes back reversed. Here the loop is the strip and the frame is the paper itself.

The signs in the figure alternate along each strip, one change per fold. After three folds the far end is face down, and joining it to the face-up near end requires a flip: Möbius. After six folds the far end is face up again, the strip lies in the same triangle twice round, and the ends join without a flip: a cylinder, folded flat. After nine, a Möbius band again, three times round.

The same bookkeeping is what two sheets over a one-sided surface built a whole cover out of: each point of a one-sided surface carries two possible senses of turning, and a loop either swaps them or does not. A folded strip makes the swap physical. Each piece of paper has a front and a back, the folded map shows one of them, and going once round the band carries the front to the back exactly when the number of folds is odd.

The cylinder in the middle is a reminder of why the question is only about Möbius bands. A paper cylinder exists at every length. Roll the strip into a tube of any circumference, or for a very short strip fold it flat in half and glue the ends — two layers, no twist, no stretching. Nothing forces a minimum. The half-twist is what costs length, and the parity argument locates the cost: the paper has to be turned over an odd number of times on its way round, and turning paper over takes room.

The polygons that more folds make #

Three folds at sixty degrees is one solution of a family. Fold lines can cross the strip at any angle $\theta$, and the question is which angles make the far end land on the near end turned over. A composite of reflections across lines alternately at angles $\theta$ and $-\theta$ is again a reflection when there are an odd number $k$ of them, and its line lies at angle $k\theta$. For the far end to meet the near end flipped, that line must run along the strip, so $k\theta$ must be a whole number of half-turns. The steepest solution, with each fold as close to straight across as the count allows, is

$$\theta = \frac{(k-1)\pi}{2k},$$

which is sixty degrees for $k = 3$, seventy-two for $k = 5$, and approaches a right angle as $k$ grows. The pieces are again triangles of height 1, now thinner, and the length of the strip is $k\cot\theta = k \tan(\pi/2k)$.

Folded flat, each band fills a regular polygon with an odd number of sides and width 1 — the distance from any corner to the opposite side. Each piece of the strip is a thin triangle reaching from one side of the polygon to the corner opposite, and the creases are the polygon’s diagonals. The strip’s two long edges become the band’s single edge, and that edge runs exactly once round the polygon. So the strip’s length is half the polygon’s perimeter: the triangle of height 1 has perimeter $2\sqrt3$, the pentagon of width 1 has perimeter $3.2492$, and as the number of sides grows the polygon rounds towards a circle of diameter 1, whose perimeter is $\pi$.

That gives a sequence of paper Möbius bands with lengths $\sqrt3$, $1.6246$, $1.5977$, $1.5869$ and so on, falling towards $\pi/2$. Every one of them is checked in the figures by composing the reflections and confirming that the end lands where a Möbius join needs it. On the face of it the triangle is not the best; the pentagon beats it by six per cent.

The catch is in the picture. In the triangle, each crease is a side of the triangle, and no two creases cross. In the pentagon the creases are the five diagonals, and they cross one another in a pentagram. Where two creases cross, two different pairs of layers are each folded round a line, and the paper of one fold would have to pass through the paper of the other. The pentagon and every polygon after it describe paper bands that exist only if the paper may pass through itself. Mathematicians call such a thing an immersion, as against an embedding, which may not cross itself; the projective plane of the room a projective space needs is the classic example of a surface that has the first in three dimensions and not the second, and the bottle that needs a fourth dimension is another. The distinction there was topological — whether any surface of that kind fits — and here it is metric: a Möbius band of any shape embeds in space, and only the demand that lengths be kept makes the crossing unavoidable for the shorter strips.

Two thresholds, a half of pi and the root of three #

The figure below sets the lengths of the folded bands against the two numbers in the problem.

The lower line, $\pi/2 \approx 1.5708$, is Halpern and Weaver’s bound: no paper Möbius band, even one allowed to pass through itself, can be made from a shorter strip. The folded polygons show the bound cannot be improved for that looser problem, since their lengths come as close to it as desired — $1.5729$ at twenty-five folds — and smoothing their creases slightly costs as little extra length as one likes. For bands allowed to cross themselves the answer is $\pi/2$, and it is approached by flattened polygons with ever more sides.

Why should the looser problem have a round answer? As the number of folds grows, the folded polygon approaches a disc of diameter 1, the creases approach diameters, and the strip approaches a fan of infinitely thin triangles all turning about the centre. The band’s edge runs once round the disc, a length of $\pi$, so the strip has length $\pi/2$. Halpern and Weaver’s theorem says that this limiting fan, which is no longer a band at all, is as economical as paper can be.

The upper line, $\sqrt3$, is the answer when the paper may not cross itself, and the gap between the two lines is the whole difficulty of the problem. The triangle shows that any length above $\sqrt3$ suffices, once its creases are softened a little; Halpern and Weaver’s argument shows that nothing below $\pi/2$ does. For more than forty years the region between them was open. Schwartz raised the lower bound partway, to about 1.69, in 2021, and closed the gap two years later.

His argument is elementary in its tools and delicate in its bookkeeping. A smooth paper surface is ruled: through every point where it bends there runs a straight segment of the paper, reaching from one edge to the other, along which the surface does not bend at all — the same lines a sheet of paper shows when it is curled in one hand. Schwartz showed that every embedded paper Möbius band contains two such segments that are perpendicular to each other and lie in a common plane, so that together they look like the letter T. Finding the T is the hard part and uses the embedding essentially; once it is found, the bound follows from a careful but elementary estimate of how much paper the rest of the band needs to connect the two segments without crossing itself.

The edge carries the length #

The folded pictures suggest a cleaner way to see what the question is about. Every band made from a strip of length $\lambda$ has a single edge — the strip’s two long sides, joined end to end by the flip — and that edge has length $2\lambda$. Asking for the shortest strip is asking for the shortest edge.

In the triangular band the edge is the perimeter of the triangle, traced once: $3 \times 2/\sqrt3 = 2\sqrt3$, twice the length of the strip, as it must be. The strip’s centre line, halfway between its edges, is a closed curve of length $\sqrt3$, and it runs round the smaller triangle joining the midpoints of the sides, meeting each crease at the crease’s own midpoint. Both computations are made in the figure from the folded pieces, and the edge length is confirmed by tracing it rather than assumed.

The picture also explains why a sequence of good bands should converge to this one. A paper band must have room for its edge to go once round while the paper turns over an odd number of times. The triangle uses the minimum number of turns, three creases, and fits them along the three sides of the smallest region they can bound. More creases would make the edge shorter, but once there are more than three of them they cannot all lie along the outside of the folded shape, and an inside crease meeting another inside crease is a collision. That is a heuristic rather than the proof, but it points at the same obstruction the proof has to overcome.

Why the textbook picture is not paper #

Every textbook draws the Möbius band the same way: a short segment swung once round a circle, turning through half a revolution as it goes, giving a smooth, round, gently twisted loop. It is natural to assume that this is what a strip of paper looks like when its ends are joined with a twist. It is not, and the reason is curvature.

The Gaussian curvature of a surface at a point is the product of its two principal curvatures, and Gauss proved that it cannot be changed by bending without stretching — the theorem that makes every flat map of the globe wrong somewhere. A flat sheet has curvature zero everywhere, so every surface made from paper has curvature zero everywhere too. Paper can be rolled into a cylinder or a cone, both of curvature zero, but it cannot be shaped into a saddle or a dome.

The textbook band is a saddle at every point. Along its centre circle the curvature is exactly $-1/4$, and towards its edges it falls as low as $-1.56$, the curves on the two sides exchanging as the band comes round with its half-twist. No part of it could be made of paper. The figure computes this from the formula that defines the textbook band, with no physical model involved: the familiar picture of the most famous paper object in mathematics is not one paper can take.

A real paper band looks different. Its surface is flat in places and bent along straight rulings elsewhere; it is made of pieces of planes, cylinders and cones, joined smoothly; and when it is short its shape is pulled ever closer to the flattened triangle. A long paper band can look nearly like the textbook one, because its curvature can be spread thinly along its length, but the closer the strip comes to $\sqrt3$, the less room there is for curvature anywhere except the creases.

What the folding does and does not prove #

Every figure here is a computation with reflections and with a parametrised surface, and every one checks its claims: that the folds close with a flip, that the lengths are $k\tan(\pi/2k)$ and fall, that the edge runs once round the triangle, that the textbook band’s curvature is negative everywhere sampled. Those checks establish that the folded bands are what they appear to be — paper Möbius bands, with the lengths shown — and that the textbook band is not one.

They do not establish the two bounds. That nothing shorter than $\pi/2$ can be folded or bent into a Möbius band at all is Halpern and Weaver’s theorem, and that nothing shorter than $\sqrt3$ can be done without self-crossing is Schwartz’s; neither is a statement a finite computation can reach, since both quantify over every smooth surface. Nor does the crossing of the pentagon’s creases, drawn in the figure, by itself prove that its layers cannot be stacked. That follows from Schwartz’s theorem: a stacked pentagon could be eased into a smooth embedded band only slightly longer than $1.6246$, and the theorem forbids it.

Still open: other twists and other shapes #

Schwartz’s theorem is about a single half-twist. A band joined with three half-twists is also one-sided, but its edge is a knot — a trefoil — and its paper version raises the same question with a new answer. The same question has been asked for bands with more half-twists and for paper cylinders joined with a full twist, and for most of these the optimal shape and the exact threshold are not known. The guess that the extremal shapes are always flattened, creased polygons, as the triangle is, is natural and unproved.

A second question is about stability rather than possibility. A physical paper band, released, settles into a shape that minimises its bending energy, and Sadowsky computed in 1930 how that energy is distributed along a long, narrow band. For bands of moderate length the equilibrium shapes were computed numerically by Eugene Starostin and Gert van der Heijden in 2007, and they found the bending concentrating along lines, with regions of the band nearly flat, as the strip is made wider relative to its length. Whether the energy-minimising bands converge to the triangle as their length falls to $\sqrt3$, as Schwartz’s theorem suggests they must, is not proved.

What a half-twist costs #

The topological Möbius band costs nothing. Glue the ends of any rectangle with a flip and it exists. The paper Möbius band costs length — at least $\sqrt3$ times its width if the paper is not to pass through itself, and $\pi/2$ times its width if it may. The difference between the topological object and the physical one is a single constraint, that lengths are kept, and that constraint converts a question with a yes-or-no answer into one with a number. The number is attained by three folds and an equilateral triangle, and the proof that nothing does better took forty-six years.