Tiling the Plane

Cover the plane with copies of one shape, and try to make it repeat

Two Rules, and Everything Else Is Allowed

A tiling of the plane is a family of shapes that covers every point and overlaps nowhere. That is the whole definition. The shapes may be rotated, reflected, repeated or not; there is no requirement that the arrangement be tidy, and no requirement that it repeat.

In practice almost every tiling anyone draws does repeat, because repeating is how you get an infinite object out of a finite instruction. This page is about what repeating means precisely, and about a tiling that declines to.

What Repeating Means

A tiling is periodic if some translation — a slide, with no turning and no flipping — carries the whole thing exactly onto itself. The slides that work are never isolated: if u and v both work then so does any whole-number combination of them, so the periods of a tiling form a lattice.

That is a period. The outlined copy sits exactly on the original; every tile has landed on a tile.

One tile per cell, and the lattice is the obvious one. The translations that work form a lattice: all the integer combinations of (1.00, 0.00) and (0.00, 1.00). A tiling with even one non-zero period is called periodic, and every tiling most people can draw from memory is.

Try it: Slide the copy around and watch the distance to the nearest genuine period. It hits zero only at the lattice points, and between them the outlined copy is visibly out of step. Switch to the triangles: the two triangles in a cell point opposite ways, so no translation swaps them — only a half turn does, and a half turn is not a period.

Covering Everything, Overlapping Nothing

Before asking whether a tiling repeats it is worth being able to check that it is a tiling at all, and by eye that is surprisingly hard — a hairline gap and a hairline overlap look identical at any zoom you are likely to try.

The patch below is checked by arithmetic instead. Every tile in this module sits on a triangular lattice with integer coordinates, so its area is an exact integer count of unit triangles. Add those up. Then, separately, trace the outside boundary by cancelling every edge that two tiles walk in opposite directions, and measure the area it encloses. An overlap or a gap would make the two disagree.

seed
100
tiles
2,010
area by adding up tiles
2,010
area inside the traced boundary
1
boundary loops
A tiling. The two areas agree exactly, so nothing is covered twice and nothing is left uncovered, and the single loop means no hole. In Cartesian units that area is 870.356.

Areas are counted in unit lattice triangles, which is what makes the comparison exact: a tiling is defined by covering everything and overlapping nothing, and both halves of that definition are integer arithmetic here rather than a judgement about pixels.

Try it: Grow the patch and watch the two area columns stay locked together. They are computed by completely different routes — one per tile, one from the boundary alone — and they agree at every size, on the nose, with no tolerance permitted.

Hunting for a Period That Is Not There

Here is the same kind of patch, and the question of the whole module: is there a translation carrying it onto itself? Slide it and count. A tile is green when it lands on a tile of the same shape in the same orientation, and red when it does not.

seed
100
tiles in the patch
0
landing on a tile
100
landing on nothing
0.0%
of the patch matched
100 tiles have nowhere to go. Green tiles land on a tile of the same type and orientation; red ones land on a different tile, or off the patch entirely.

Search over all 2400 non-zero lattice shifts within 24 steps and the best still leaves tiles stranded. It is worth being careful about what that does and does not show: a finite patch always has a boundary, so some failures are boundary effects, and enlarging the patch raises the best score rather than lowering it. The tiling really is full of repetition. What it has no room for is a single shift that works everywhere at once.

Try it: Press find the best one. It gets a long way — this tiling is stuffed with repetition, and any finite arrangement you find in it turns up again and again. But something is always left over, and a search on a finite patch can never rule out a period that shows up further out. That gap in the argument is exactly what the hierarchy closes.

The Instrument

One canvas runs through this whole module, and it gains a control on every page as the ideas that justify the control arrive. Right now it does two things: it pans and it zooms. That is enough to make the point it is here to make — look as far as you like, at any scale you like, and the pattern never settles into a repeat.

HTPF
Drag to pan, scroll or pinch to zoom. The colours are the four cluster shapes the tiling is built from — the next page explains where they come from.

Try it: Zoom into a corner and look for a patch you recognise. You will find it again elsewhere — that much is guaranteed. What you will not find is the same arrangement continuing in the same way, because the surroundings of a repeated patch differ every time.

The Question That Took Sixty Years

A set of shapes is called aperiodic when it tiles the plane and every tiling it admits fails to be periodic. That is a strong condition: it is not enough to arrange the tiles badly, the tiles have to make good arrangement impossible. In 1961 Hao Wang conjectured no such set existed. Four years later his student Robert Berger found one — with 20,426 tiles. The rest of this module is the story of that number coming down, and of what happened when it reached one.

Key Takeaways

  • A tiling covers everything and overlaps nothing — both halves are checkable by exact integer arithmetic when the tiles sit on a lattice, and both are checked here
  • Periodic means a translation maps it onto itself — and the translations that work always form a lattice, so a tiling has either no periods or infinitely many
  • Repetition is not periodicity — the hat tiling repeats every finite arrangement infinitely often, and still has no period at all
  • A finite search cannot settle it — the best shift on a small patch matches most of it, and the fraction rises as the patch grows; ruling out every period needs a structural argument
  • Aperiodic is a property of a tile set — it tiles the plane, and no tiling it admits is periodic; the first known set had 20,426 tiles