On Penrose tilings and the patterns that never repeat
There is a question hidden inside every bathroom floor. Squares tile the plane. So do hexagons. So do triangles. They tile it periodically—shift the whole pattern by some fixed distance and it lands exactly on itself. Translational symmetry. This is the kind of order we expect: repetition, rhythm, a pattern you can grab and slide.
For a long time, mathematicians assumed that any set of tiles capable of covering the plane must be capable of covering it periodically. If they tile at all, they tile with repetition. This turns out to be wrong. In the 1970s, Roger Penrose found two shapes—just two—that tile the plane but only aperiodically. They cover every inch of an infinite surface, leaving no gaps, yet the pattern never repeats. Not because repetition is unlikely, but because the geometry makes it impossible.
This is not randomness. The tiling is deterministic, lawful, exquisitely structured. It has five-fold rotational symmetry—something forbidden in periodic crystals. It has a kind of long-range order that reaches across the entire plane without ever settling into a repeating unit cell. It is a third possibility nobody expected: ordered without being periodic.
Penrose's P3 tiling uses two rhombi. The thick rhombus has interior angles of 72° and 108°. The thin rhombus has angles of 36° and 144°. Both have edges of equal length. Every proportion in the tiling—every ratio of diagonals, every relationship between areas—is governed by φ = (1+√5)/2, the golden ratio.
The shapes alone are not enough. Two rhombi of these angles can tile the plane periodically if you let them. What prevents periodicity is the matching rules—constraints on how tiles may be joined. Penrose encoded these as arcs or arrows on the tile edges: adjacent tiles must have their markings align. These local rules, obeyed at every edge, are sufficient to guarantee that the global pattern never repeats.
Computationally, the most elegant way to generate the tiling is not to place tiles one by one but to subdivide. Start with a simple configuration—ten triangles arranged in a star—and split each triangle into smaller triangles according to precise rules involving φ. Each level of subdivision produces a finer, more detailed tiling. The rhombi emerge by pairing adjacent triangles that share a long edge.
Each thick rhombus contains, at the next level of subdivision, two thick rhombi and one thin. Each thin rhombus contains one thick and one thin. Apply the subdivision again and the pattern repeats at a smaller scale. The tiling is self-similar—it contains rescaled copies of itself at every level of magnification. Zoom in and you find the same two shapes, the same relationships, the same proportions, forever.
The ratio of thick to thin rhombi converges to φ. This is not coincidence. The subdivision matrix has eigenvalue φ, and the tile counts grow as powers of φ. The golden ratio is not merely a parameter in this system—it is the system's eigenvalue, the number that governs how it scales.
Watch a single subdivision step below. Each triangle splits according to the golden ratio, dividing its edges at the point that produces smaller copies of both triangle types. The operation is reversible in principle—you can inflate as well as deflate—which means the tiling has no preferred scale. It is all scales at once.
Here is the strangeness. Every finite patch in a Penrose tiling appears infinitely many times. Take any region of any size—a cluster of a hundred tiles, a thousand, a million—and that exact configuration occurs again elsewhere in the tiling. Infinitely often. In every direction. This is proved and not subtle.
Yet the tiling as a whole never repeats. No translation maps it onto itself. The global property—aperiodicity—is not visible at any finite local scale. A creature living on the tiling, able to examine any neighborhood however large, could never prove from local inspection that the pattern is aperiodic. Every finite sample is consistent with a periodic tiling. The aperiodicity is a property of the infinite whole that leaves no trace in any finite part.
This is emergence in its purest mathematical form. Local rules—the matching constraints at each edge—produce a global property that no local observation can detect. The tiles do not “know” they are part of an aperiodic pattern. They follow their rules, edge by edge, and the aperiodicity happens above them, around them, through them. It is not encoded in any tile or any finite assembly of tiles. It exists only in the limit.
I find this disorienting in the best possible way. We are trained to think that local and global are connected—that if you understand the parts well enough, the whole follows. Penrose tilings say: not always. Some properties belong only to the infinite. They are real, provable, consequential—and completely invisible from the inside.
Periodic order is a promise of return. You walk far enough and you arrive where you started. Aperiodic order is a different promise: you will never return, but you will never be lost. Every neighborhood is familiar. Every neighborhood is unique. The pattern holds without repeating. This, it turns out, is not less ordered than periodicity. It is more.