Next to nothing
At the end of 1610 Johannes Kepler was imperial mathematician in Prague, and he owed a New Year's present to his friend and patron, the court counsellor Johannes Matthäus Wackher von Wackenfels. He had no money for a good one. What he gave him was a pamphlet of twenty-four pages, printed at Frankfurt in 1611 as Strena, seu de nive sexangula: a New Year's gift, or on the six-cornered snow. The joke that opens it is that Wackher loved Nothing, and nix, Latin for snow, sounds like the German nichts. Kepler had been crossing the bridge over the Vltava and saw flakes settle on his coat, every one of them with six corners. That would do as a gift. It was nearly nothing, it came down from the sky, and it raised a question he couldn't answer.
The Wellcome Collection has a copy of the 1611 printing. The English most people read is Colin Hardie's translation, The Six-Cornered Snowflake, Oxford, 1966.
Why six, every time? Kepler went looking for other places where nature makes six, and the most useful one was full of bees. A honeycomb cell is a six-sided tube, and its closed end is not flat. It is three rhombuses meeting at a point, so that the cells of the two faces of the comb fit back to back. To Kepler it looked less like the bees' invention than like what you get when equal soft cells are packed as closely as they will go and pressed. The second example was a pomegranate. Its seeds grow inside a hard rind until they push on one another, and they come out with flat faces that are rhombuses too.
To explain these, he stacked balls. Lay a flat layer in squares and set the next layer in the hollows, or lay it in triangles and do the same, and you come to the same arrangement either way. Each ball touches twelve others, and squeezed evenly it becomes a solid with twelve rhombic faces, the pomegranate seed's shape. Kepler said in passing that this packing would be the tightest possible, and gave no proof. That sentence became the Kepler conjecture. Thomas Hales and Samuel Ferguson announced a proof in 1998, so long and so dependent on computer calculation that the referees could not certify all of it. Hales then led a project to rewrite the whole argument in logic a machine could check, and the paper reporting it, A formal proof of the Kepler conjecture, appeared in 2017.
He couldn't get it to work for snow. Snow isn't cells pressed together in a rind. It forms in open air and grows outwards, and Kepler couldn't see why packing would make it flat or give it feathered arms. He tried heat and cold, and the nature of the vapour, and in the end put the cause down to a formative faculty in nature that made shapes for beauty as well as for use. He finished by saying that he had knocked on the door of chemistry, and left it there, with nothing settled.
I think the answer, once it came, really was in two parts, one for the six and one for the arms. The six belongs to ice itself. Twentieth-century X-ray work showed that frozen water molecules sit in a lattice of hexagonal rings, so a crystal of ice has six equivalent directions in its plane, much as Kepler's packed balls have their twelve. Kepler had been looking in the right place for the six. The arms don't come from the lattice. They come from the air. Ukichiro Nakaya grew the first snow crystal in a laboratory at Hokkaido University on 12 March 1936, on a single rabbit hair. He then grew them in cold boxes under controlled conditions and found that temperature and the wetness of the air decided whether the ice made plates, needles or ferns.
Two rules, and nothing about branches. In 2005 the mathematician Clifford Reiter published a very small model of this (A Local Cellular Model for Snow Crystal Growth). The plane is a grid of hexagonal cells, each holding some amount of water. A cell with one unit or more is ice. Any cell that is ice or touches ice keeps what it has and gains a small constant amount each step. Every other cell is vapour, and on each step it becomes the average of itself and its neighbours. Far off, the air is held at a fixed level. There are just two numbers to choose, the background level and the constant, and Reiter set them out as a diagram. Low on both, the crystal grows ferns. Raise the constant and you get sectors and then plates.
Nothing in either rule mentions a branch. A bump on the edge reaches further into air that still has water in it, so it collects faster than the hollow beside it and the bump gets longer. The arms grow along the six directions of the grid because those are the only directions there are. Change the constant halfway through, as a falling crystal does when it drops into a wetter layer of cloud, and a plate puts out ferns from its corners. The piece Six corners runs Reiter's two rules and draws the vapour as well as the ice, so the pale halo the crystal has already drained is visible around it.
Kepler's gift was a question. Half the answer turned out to be near where he looked first, in the way small things pack. The other half was in the air he was walking through.