The bottle that needs a fourth dimension

The Möbius band is a rectangle with one pair of edges glued after a flip. It has one side, and it has an edge — a single closed curve running all the way round, which is what stops it being a closed surface in the sense the classification uses.

The obvious next move is to get rid of the edge by gluing the other pair too. There are only two ways to do that, and one of them gives a torus. The other gives an object that has no edge, has one side, and cannot be built.

The word, and the two ways to close the rectangle #

The gluing rules are written as words round the boundary of the polygon. Walk the square’s four edges in order, name each edge by the letter of the edge it is glued to, and write an inverse when the arrow points backwards. Four rules exhaust the possibilities.

$aba^{-1}b^{-1}$ glues both pairs straight across and gives a torus. $abab^{-1}$ glues one pair straight and the other with a flip, and gives the Klein bottle. The difference between them is one exponent, and it is the whole of the difference between a surface with two sides and a surface with one.

The reason the flip matters is exactly the reason it mattered on the band. A letter used twice the same way round means that walking across that edge and out the other side arrives with left and right exchanged. There is no way to choose a consistent sense of turning, and the closing rung of this ladder makes that statement precise without mentioning sides at all.

Why it is a Möbius band twice #

The most useful description of the Klein bottle is not the bottle at all. Cut it along a suitable closed curve and it falls into two Möbius bands.

That is worth taking slowly, because it explains the picture. The Möbius band has one boundary curve. Two of them have two boundary curves, and gluing those two curves together closes the surface with nothing left over. The result has no edge, and it is one-sided because each half already was.

So the Klein bottle is not a strange new object. It is the smallest thing that can be made out of two copies of the object the rung below is about, and every property it has is inherited.

The counting agrees. A Möbius band has Euler characteristic zero, and gluing two surfaces along a circle adds their characteristics — a circle contributes nothing, since it has as many vertices as edges. Zero plus zero is zero, which is what the gluing diagram computed.

What the drawing is doing wrong, precisely #

The picture at the top of this essay is not a Klein bottle. It is the image of one under a map that is not one to one.

The distinction has a name on each side. An embedding is a map that is one to one: distinct points of the surface go to distinct points of space, so the image is a faithful copy. An immersion is weaker: the map is smooth and its derivative never collapses, so the surface is locally faithful — every small patch is a genuine patch of surface — but two patches far apart on the surface may land on the same place in space.

The Klein bottle immerses in three dimensions and does not embed. The circle in the figure is where two far-apart patches meet, and the assertion behind the drawing is exactly that: at every angle round the tube, two different parameter values produce the same point of space, and the figure checks all of them rather than drawing a crossing and calling it one.

Nothing is wrong with the surface at those points. A creature living on the Klein bottle would find nothing there — no crease, no boundary, no edge. The self-intersection is a feature of the map, and it is the price of the room.

The reason three dimensions is not enough #

It is tempting to think the obstruction is about cleverness — that a sufficiently ingenious construction would avoid the crossing. It is not, and the reason is a theorem with a short statement.

A closed surface embedded in three-dimensional space separates it into an inside and an outside, and is therefore two-sided. This is the surface version of the Jordan curve theorem, and the argument has the same shape: a closed surface with no edge divides the space it sits in, a point just off the surface is either in the bounded piece or the unbounded one, and following the surface around never gets from one to the other. Consistently choosing “the side facing the outside” gives a global choice of side.

So an embedded closed surface in three dimensions has two sides. The Klein bottle has one. It cannot be embedded, and no amount of ingenuity changes that.

Two things about the argument are worth pinning down. It needs the surface to be closed — the Möbius band embeds in three dimensions perfectly well, and it is one-sided, because it has an edge and does not separate anything. And it needs codimension one: a surface in three-space misses being space by exactly one dimension, which is what gives “just off the surface” its two options. In four dimensions, just off the surface is a whole circle of directions, and there is no inside and outside to be consistent about.

That last observation is the whole of why the Klein bottle embeds in four dimensions. The self-intersection in the drawing is two sheets crossing; lift one of them a small distance in the fourth coordinate near the crossing, and they no longer meet. There is nowhere to lift it to in three dimensions, and one more coordinate is enough.

The figure-eight cross-section, and why not a circle #

The familiar picture of a Klein bottle is the glass one: a tube that narrows, bends back, passes through its own wall and flares out to meet the wide end. The drawing above is the other standard immersion, and it is the one worth computing.

Take a circle of radius $a$ in the horizontal plane, and at each point of it place a figure eight standing on end. Carry the figure eight once round the circle while rotating it by a half turn. The half turn is the Möbius band’s twist; the figure eight is what makes the surface close up without a boundary.

Written out, with $u$ running round the figure eight and $v$ round the circle:

$$\bigl(a + \cos\tfrac{v}{2}\sin u - \sin\tfrac{v}{2}\sin 2u\bigr)\bigl(\cos v,\ \sin v\bigr), \qquad z = \sin\tfrac{v}{2}\sin u + \cos\tfrac{v}{2}\sin 2u.$$

Two identities decide what this is, and both are exact. Increasing $u$ by $2\pi$ returns to the same point, which says the cross-section closes. Increasing $v$ by $2\pi$ returns to the point with $u$ replaced by $-u$ — not the same point — which says that going round the long way arrives with the cross-section turned over. That reversal is the flip in $abab^{-1}$, and it is why the surface is this one rather than a torus.

The self-intersection follows from the same formula without any searching. At $u = 0$ and at $u = \pi$ the two sine terms both vanish, so both parameter values give the point at distance $a$ from the axis in the horizontal plane. Two places on the surface, one place in space, at every $v$ — which is the circle drawn.

The glass version is a different immersion of the same surface, and its crossing is a circle too. Neither is more correct; the figure eight is easier to check.

What has to be given up #

An immersion is a compromise, and it is worth being explicit about what survives it and what does not, because a reader who is told “the picture crosses itself, ignore that” is being asked to ignore something specific.

What survives: every local measurement. Angles, lengths along the surface, curvature, the local pattern of neighbourhoods. The immersion is a genuine surface everywhere, and every small piece of the drawing is a small piece of a real Klein bottle.

What does not survive: anything that asks whether two points are the same point. Distance measured through space rather than along the surface is wrong near the crossing. Whether a closed curve on the surface is one curve or two cannot be read off the drawing. And the inside is not the inside of anything: the crossing means the drawn object does not separate space, so pouring liquid into a glass Klein bottle fills a region that is genuinely connected to the outside.

The last is the source of the standing joke about the bottle holding no volume, and the joke is a correct statement about the immersion rather than about the surface. The surface has no inside because it has no sides at all.

The name, which is a mistranslation that stuck #

The surface is Felix Klein’s, from 1882, and in German he called it a Fläche — a surface. Somewhere in the passage into English that became Flasche, a bottle, and the glass models followed the name rather than the other way round.

The accident is worth a paragraph because it did real damage to how the object is understood. A bottle has an inside; the whole point of this surface is that it does not. Generations of readers have been introduced to it as a container with a peculiar neck, which is a picture of the immersion’s defect rather than of the surface, and the first thing that has to be unlearned.

The gluing diagram has no such problem. It is a square with four arrows on it, it fits on a page, and nothing about it suggests a vessel. The same is true one rung down, where the band’s rectangle is the real object and the paper strip is a model of it, and it is true again in the classification, where the surfaces are polygons with words and the blobs with handles are illustrations.

A picture that has to be argued down is worse than a diagram that has to be learned. That is a general lesson about this subject and the bottle is its clearest case.

Where a one-sided closed surface actually turns up #

The Klein bottle is not only a counterexample, though it earns its keep as one. Three places it appears as the answer rather than as the exception:

The space of unoriented directions in the plane, with a length attached. A direction with no arrow on it is a point of a circle with opposite points identified; adding a length that may be positive or negative and asking for the pairs that describe the same object produces exactly the Klein bottle’s gluing. It is the natural home of a quantity that is periodic in one variable and reverses sign in another.

Phase spaces with a reversal. A system whose state is an angle and a velocity lives on a cylinder; if the physics is unchanged by reflecting the angle and the velocity together, the states that are genuinely different form the quotient, and for the right reflection that quotient is a Klein bottle. The reversal in the gluing word is the reflection in the physics.

The two-dimensional shadow of a lattice question. Certain periodic patterns in the plane are unchanged by a glide — a translation combined with a reflection — and the surface of distinct patterns is the plane divided by the group the glide generates. That group is the Klein bottle’s fundamental group, and the surface is what unrolling a loop produces when the deck transformations reverse.

None of the three needs the bottle to sit in a room, which is the point: every one of them is a quotient, described by a rule for identifying points, and the rule is where the surface lives.

The rest of the family #

The Klein bottle is the second one-sided closed surface, and the list continues. Attaching another cross-cap gives a surface with characteristic $-1$, then $-2$, and so on without end. Every one of them fails to embed in three dimensions for the same reason, and every one of them embeds in four.

The classification is the statement that this list and the list of spheres with handles between them contain every closed surface exactly once. Two numbers name a surface: the Euler characteristic, and whether it is one-sided. Neither alone is enough, and the Klein bottle is the standing proof — it shares its characteristic with the torus and shares nothing else.

There is a pleasing consequence for anyone who has cut a Möbius band. The Klein bottle can be cut into two Möbius bands, and it can also be cut — along a different curve — into a single cylinder. Which pieces fall out depends on the curve, exactly as cutting the band depends on where the scissors go, and for the same reason: the curve either respects the flip or does not.

What the picture cannot show #

The drawing cannot show the surface. It shows a shadow of it — the image of a map that loses information at one circle — and every honest picture of a Klein bottle in three dimensions has that defect.

It cannot show the fourth dimension that would repair it, either. The repair is a small displacement in a direction the page does not have, applied near the crossing, and there is no way to draw the displacement that does not immediately look like a displacement in one of the three directions already used.

And it cannot show one-sidedness directly, which is the deeper limitation. Sidedness is a property of how a surface sits in a space, and the drawing sits it in the wrong space — where, because of the crossing, “which side” is not even well defined. The property being claimed is intrinsic, the drawing is extrinsic, and the gap between them is what the gluing diagram exists to close.

Where the ladder goes next #

Above this rung: the projective plane, which is the smallest one-sided closed surface and is what the fourth gluing of the square produces. The orientation double cover, which puts a two-sided surface over every one-sided one and turns the question into a question about covering spaces. And orientation stated as a sign rather than as a side, which is where the word finally stops depending on there being a room to sit in.

Two debts this essay leaves open. The claim that a closed surface in three-space separates it is quoted here and not proved; its proof is the Alexander duality argument and it is genuinely harder than the curve version. And the embedding of the Klein bottle in four dimensions is described as a lift near the crossing, which is right but is not a construction — writing one down means giving four coordinates as functions of $u$ and $v$, and the natural choice is the flat torus’s cousin rather than anything the eye recognises.

What the flip was doing all along #

One exponent in a four-letter word decides whether a surface has two sides.

That is the sentence the rung below and this one share, and it is worth carrying because it says where the property lives. Not in the paper, not in the room, not in the picture — in the rule for gluing, which is a statement about arrows and can be written down in four characters. Everything else the Klein bottle does, including the crossing it cannot avoid, follows from that one letter being used twice the same way round.