Math Art

Seifert Surfaces

Dip a knotted wire in soapy water and a film spans it. That the film exists at all is not obvious: a knot is a closed loop tangled in space, and there is no reason a surface should be able to fill it in without passing through itself. In 1934 Herbert Seifert settled the question for good, and did it constructively — he gave a recipe that takes any knot and hands back a surface whose only edge is that knot.

The recipe is almost embarrassingly simple. Point the knot in a direction and walk along it. At every crossing, instead of carrying straight on, take the turn that keeps you going the same way round. Do that everywhere and the knot falls apart into a handful of plain circles that never meet. Fill each with a disk, then put back one twisted band at every crossing you cut. The result is a surface, and its edge is exactly the knot you started with.

What makes it more than a trick is the genus — the number of handles. It comes straight out of the counting, and the smallest genus any spanning surface can have is a property of the knot itself. A trefoil cannot be spanned by anything simpler than a one-handled surface, and that single fact proves the trefoil is knotted.

Using it

  1. Pick a knot from the list, or type a braid word of your own. The letters name crossings: A twists the first two strands one way, a the other way, B the second and third, and so on. A trailing number repeats, so A3 is the trefoil and A5 the next torus knot along.
  2. Surface chooses which spanning surface to build. Seifert's is the famous one and is always two-sided. The others are what you get by resolving the crossings differently, and most of them are one-sided.
  3. Smoothing rounds the disks and bands off by Catmull-Clark subdivision. Each level quadruples the face count, so two is usually plenty.
  4. Knot tube draws the boundary itself as a wire. That wire is the knot the surface was built to span — the whole point of the construction, and worth switching on at least once.
  5. The line under the controls reports what was built: disks, bands, Euler characteristic, and the genus or crosscap number. It also says whether the finished mesh still agrees with the combinatorics it was built from, which is the check that the geometry did not quietly go wrong.

How it works

Start with the everyday picture. Imagine the knot drawn on paper with little breaks where one strand passes under another, and imagine walking the loop with an arrow painted on it. At a crossing you have a choice: carry on under the bridge, or turn the corner and follow the other strand. Seifert's rule says always take the turn that keeps your arrow pointing the same way round as the strand you are joining. Because the choice is forced by the arrows, nothing is left to taste. Follow it everywhere and the single tangled loop is replaced by a few simple circles, each flat and unknotted. Those are the Seifert circles.

Now rebuild. Stack the circles at different heights and fill each one in with a disk. At every crossing you cut, glue in a band joining the two disks, giving it a half twist so the surfaces meet the right way round. You have a surface made of flat parts and twisted ribbons, and running your finger along its edge traces the original knot.

The counting is where it becomes mathematics. A surface built from d disks and b bands has Euler characteristic χ = d − b, and for a surface with one boundary circle and g handles, χ = 1 − 2g. Equate the two and the genus falls out of the diagram without any geometry at all. The trefoil as the braid AAA gives two disks and three bands, so χ = −1 and g = 1: one handle. Since a surface of genus 0 spanning a knot would force it to be unknotted, that is a proof the trefoil is tied.

Seifert's choice is one among many. At each crossing there are two ways to cut, so a diagram with c crossings carries 2c spanning surfaces — Kauffman's states — and all of them have the same knot as their boundary. The striking part is how cleanly they split. Build the state's graph, a node per circle and an edge per band; the surface is two-sided exactly when that graph is two-colourable. Seifert's state always is. Most others are not, and those are genuinely one-sided: resolve all three crossings of the trefoil the other way and you get a Möbius band with three half twists, whose single edge is the trefoil.

Two details of the construction are worth naming, because the obvious versions of both fail. The disks are meshed as an O-grid — a square patch with annular rings around it — rather than as a fan of triangles from the centre; a fan puts one vertex at the hub with as many neighbours as the rim has samples, and both the smoothing and the subdivision misbehave around it, leaving a starburst crease that never goes away. And the bands are swept with a rotation-minimising frame rather than the curve's natural Frenet frame, which is undefined wherever the curve is momentarily straight and flips through such a point — exactly what a gently bowed band does.

Where this comes from