Nothing on a sphere can be combed flat

Give every point of a sphere an arrow, lying flat against the surface, with the arrows varying continuously from place to place. Somewhere, an arrow has length zero.

There is no arrangement that avoids it. Not a clever one, not a lopsided one, not one with the zeros pushed toward a corner — a sphere has no corners. The obstruction is the sphere.

The field drawn, and why its zeros are not its fault #

The field above flows from one pole toward the other along the meridians. It is the most obvious field to write down and its zeros are at the poles, which invites the response that a cleverer field would put them somewhere else, or nowhere.

Somewhere else, yes. Nowhere, no.

The poles are zeros here because a meridian has no direction at a pole — every direction from the north pole is south, so the rule “point along the meridian” has nothing to say. That is a property of the coordinate system rather than of the sphere, which is why the response is a reasonable one to make. Choosing different coordinates moves the poles. It does not remove them.

What follows is the argument for why.

What a zero looks like from close up #

Around an isolated zero, look at how the field turns while walking once around it in a small circle.

The field turns through some whole number of complete revolutions. It has to be whole, because the walk returns to where it started and so must the direction. That number is the index of the zero.

A source or a sink has index $+1$: the arrows point outward or inward everywhere, and turning once round the circle turns the arrows once round too. A saddle has index $-1$: the arrows come in along one axis and go out along the other, and walking anticlockwise around it turns the field clockwise. Higher indices are possible and less common.

The index is the crucial object because it is stable. Perturb the field slightly and a zero of index $+1$ may move, but it cannot vanish, and it cannot change its index — a small perturbation cannot change a whole number continuously. A zero can only disappear by colliding with another zero whose index cancels it, and then the total is unchanged.

That stability is what makes the total worth computing.

The number, and where it comes from #

Poincaré–Hopf: for any continuous tangent field on a closed surface with finitely many zeros, the sum of the indices of the zeros equals the Euler characteristic of the surface.

$$\sum \text{index} = \chi$$

For the sphere, $\chi = 2$. So the indices must add to two, and in particular they cannot add to zero, and in particular there must be at least one zero. That is the hairy ball theorem, and it is a corollary rather than a theorem in its own right.

That is the surprise worth stopping on. The Euler characteristic was introduced by counting the corners, edges and faces of a solid — a completely combinatorial business, with no notion of continuity, direction, or field anywhere in it. The theorem says that this counting number governs what continuous fields can do on the surface.

Two subjects with nothing in common turn out to be constrained by the same integer. There is no reason from either side to expect it.

Where it can be done #

A doughnut has Euler characteristic zero, so its indices must add to zero — and the easiest way to add to zero is to have no zeros at all.

That is achievable, and it is easy to picture: comb the whole surface in the direction that goes around the hole. Every point has an unambiguous such direction, nothing degenerates anywhere, and the field is nowhere zero. A torus can be combed flat.

So the difference between a sphere and a torus, as far as combing is concerned, is one number, and it is the number that already distinguished them as surfaces. The characteristic $2 - 2g$ falls with the genus, hits zero at one hole, and goes negative beyond — a two-holed surface has $\chi = -2$, so its fields must have zeros of total index $-2$, and saddles are unavoidable there in the same way sources are unavoidable on a sphere.

The classification is complete for closed orientable surfaces: exactly one can be combed, and it is the torus.

That is a sharper statement than it sounds. Combability is not a spectrum on which the torus happens to score well — it is a yes or no, decided by whether one integer is zero, and exactly one closed orientable surface has that integer equal to zero. Every other surface in the infinite family is obstructed, and the obstruction gets worse in both directions from the torus: spheres are forced to have index $+2$ and multi-holed surfaces increasingly negative totals.

The non-orientable surfaces behave differently again, and the Möbius band’s one-sidedness is why. A Klein bottle has characteristic zero and can be combed; the projective plane has characteristic one and cannot. So orientability is not what decides combability, which is worth saying because the two properties travel together often enough to be confused.

The rung below, and how it differs #

Brouwer’s theorem says a continuous map of a disc into itself has a fixed point. This one says a continuous tangent field on a sphere has a zero. The two are close enough to be confused and the difference is worth fixing.

A map sends points to points. A field attaches a direction to each point. On a disc the two are nearly interchangeable — a map’s displacement is a field, and the map’s fixed points are that field’s zeros — which is why the disc case can be argued either way.

On a closed surface they come apart. There is no “displacement” for a map of a sphere to itself unless the map is close to the identity, because two distant points of a sphere have no canonical direction between them. So Brouwer’s argument does not transfer, and the theorem that does transfer is the index one, which counts rather than constructs.

Both are non-constructive in the same way and to the same degree: each proves a point exists and neither produces it. What is different is what the proofs count. Brouwer counts a contradiction about retractions; this counts a contradiction about degrees.

Consequences worth stating carefully #

Two are commonly quoted and one of them is usually stated wrongly.

Somewhere on Earth the horizontal wind speed is zero. This is correct. Wind at the surface is a tangent field on a sphere; if it is continuous it must vanish somewhere, and the place it vanishes is the eye of a cyclone or a calm. What the theorem does not say is that there is a storm — only that horizontal motion stops somewhere, which is satisfied by a dead calm as well as by a hurricane.

A hairy ball cannot be combed without a cowlick. This is the name of the theorem and is fine.

Antipodal points have equal temperature. This is not this theorem — it is Borsuk–Ulam, a different result with a different proof — and the two are confused constantly.

The one genuine engineering consequence: a sphere cannot carry a continuous non-vanishing tangent field, so no scheme for orienting something smoothly over a whole spherical surface works. Global coordinates on a sphere always degenerate somewhere, which is why every map projection has a problem point and why spacecraft attitude systems avoid Euler angles near their singularities.

The proof, in outline, and what it costs #

The cleanest argument compares two things that must agree.

Suppose a nowhere-zero tangent field existed on the sphere. Normalise it to unit length. Then at every point there is a distinguished tangent direction, and that lets the field be used to slide the identity map of the sphere continuously to the antipodal map — push each point along its arrow by half a great circle. The identity has degree $+1$; the antipodal map on a sphere has degree $-1$, since it reverses orientation in an odd number of dimensions. Degree is unchanged by continuous deformation. So $+1 = -1$, which is false, so no such field exists.

The whole argument is a contradiction between two integers, and the integers are computed by counting rather than measuring. That is characteristic of the subject and it is the same shape as the argument that something always stays put, which is the rung below this one — there a retraction was constructed and shown to be impossible, here a deformation is constructed and shown to be impossible.

Computationally the theorem gives nothing. It says a zero exists and says nothing about where, and locating the zeros of a field is an ordinary numerical root-finding problem that the theorem does not help with. What it does supply is a check: sum the indices of the zeros a numerical method has found, and if they do not add to the characteristic, some zero has been missed. That is genuinely used in dynamical systems and in mesh processing, and it is the only practical use of the theorem this essay can honestly claim.

Where it needs a condition #

Continuity is doing everything. A field allowed one discontinuity can be non-vanishing: take the meridian field and, at the north pole only, define the arrow to point along some fixed meridian. The result is nowhere zero and it is not continuous at that point. The theorem is not about the existence of arrows; it is about the existence of a continuous choice of arrow.

The surface must be closed. Remove one point from the sphere and the remaining surface is a plane, which combs trivially — every arrow parallel. So a sphere with a puncture behaves entirely differently from a sphere, and that punctured sphere is the plane in the strongest available sense. The obstruction lives in the closing-up.

The dimension has to be even. A circle combs perfectly: point every arrow anticlockwise. So does any odd-dimensional sphere, by pairing coordinates and rotating each pair. Only the even-dimensional ones are obstructed, and the reason is exactly the degree computation above — the antipodal map has degree $(-1)^{n+1}$ on the $n$-sphere, which equals $+1$ when $n$ is odd and there is no contradiction to reach.

The parity is not a technicality of the proof; it is where the phenomenon lives. A circle is one-dimensional and combs; a sphere is two-dimensional and does not; the three-sphere combs; the four-sphere does not, and so on forever, alternating. Any intuition that says curvature or closure or compactness is what obstructs combing has to explain why it obstructs at two and four and not at one and three, and none of those notions has a parity in it.

What does have a parity in it is the antipodal map’s orientation behaviour, which is a determinant of $-1$ raised to the number of coordinates — and that is the same reason a reflection reverses orientation and a rotation does not. The whole theorem, in the end, rests on the sign of a determinant.

That last condition is the one with the most content and the least intuition behind it. A circle can be combed and a sphere cannot, and the difference is a parity.

What the picture cannot show #

The sphere figure shows one field. The theorem is about every continuous tangent field on a sphere, and no drawing rules out a cleverer one — the picture illustrates a case and the argument in The proof section is what excludes the rest.

Only the near hemisphere’s arrows are drawn. Arrows on the far side project onto the same disc pointing the opposite way and read as a second, contradictory field, so they are omitted; the reader is asked to accept that the field continues around the back. That omission is a genuine loss, because the theorem is about the whole closed surface and the picture shows half of it.

The index figures are plane fields rather than sphere fields, so they show what a zero looks like and not where one sits. The step that matters — that a field on a sphere has zeros whose indices sum to two — is not drawn anywhere on this page, because it would need the whole sphere and a field with its zeros visible at once, and the projection makes that unreadable.

And the torus, which is the case where combing succeeds and which carries half the essay’s argument, has no figure at all. That is the largest gap here.

The ladder from here #

Rungs on this anchor above this one: the degree of a map, drawn as a winding count. Borsuk–Ulam, and the antipodal-temperature statement done properly. The index theorem on surfaces of higher genus, where saddles are forced. The Gauss–Bonnet theorem, which is the same characteristic controlling total curvature instead of total index — the two together are the strongest statement that $\chi$ is the surface’s fundamental integer. Fields on non-orientable surfaces, where one-sidedness changes what a field can even mean. Morse theory, where a function’s critical points play the role of the zeros. And the Lefschetz fixed-point theorem, which contains both this and Brouwer as cases.

One integer, three subjects #

The lasting content of this essay is the reappearance.

$\chi$ was defined by counting corners, edges and faces of a polyhedron. It reappeared as a property of surfaces that survives any deformation. It reappears here as the total index of the zeros of every continuous tangent field, and it reappears again in Gauss–Bonnet as the total curvature divided by $2\pi$.

Four definitions, from combinatorics, topology, dynamics and differential geometry, none of which mentions the others, all producing the same integer for the same surface. When that happens the reasonable conclusion is not that a coincidence has occurred but that the integer is more fundamental than any of the four definitions — that each subject is measuring the same thing with its own instrument.

That is the strongest available evidence that a mathematical object is real rather than invented, and it is worth more than any single one of the theorems it explains.