A Penrose tiling covers the plane forever and never repeats — no matter how far you slide it, it never lands back on itself. That raises two questions that sound almost impossible to answer. First, how would you ever build such a thing, tile by tile, out to infinity? And second — the harder one — how could anyone prove that a pattern reaching to infinity never repeats, when you can only ever draw a finite patch of it?

Both answers come from a single, deceptively simple move: a substitution rule. Replace every tile by a fixed little cluster of smaller tiles; then do it again, and again. Repeated forever, this one rule grows a tiling with perfect self-similarity — the same structure at every scale — and, as we'll see, it hands us a clean proof of aperiodicity built out of one irrational number: the golden ratio \varphi.

What a substitution rule does

A substitution rule is a dictionary: for each type of tile it specifies a fixed arrangement of smaller tiles to replace it with. Apply the rule to a starting tile and you get a small cluster; apply it to every tile of that cluster and you get a bigger one; keep going and the pattern floods outward across the plane. Nothing is ever chosen by hand — the whole infinite tiling is forced by iterating the rule.

Because each stage is literally a scaled-down copy of the rule applied to the stage before, the result is self-similar: zoom in and you meet the same local arrangements you saw at full size. This is exactly the mechanism behind fractals — detail that regenerates at every scale — now put to work paving the plane.

Worked example 1 — the Fibonacci substitution (in one dimension)

The cleanest place to see the machinery is a one-dimensional "tiling": a row of two tile types, L (long) and S (short). The substitution rule is just two lines:

Start with a single L and apply the rule to every symbol, over and over:

Count the tiles in each generation: 1, 2, 3, 5, 8, 13, 21, \dots — the Fibonacci numbers. And count the two types separately: the numbers of L's and S's are consecutive Fibonacci numbers, so their ratio L/S = 1, 2, \tfrac32, \tfrac53, \tfrac85, \tfrac{13}{8}, \dots marches straight towards the golden ratio \varphi \approx 1.618.

In the picture below, each row is one substitution step, drawn as coloured bars — L long, S short — and every row is rescaled to the same width. Watch what that rescaling reveals: each row looks like a magnified view of the one above it. That is self-similarity, on screen.

The bookkeeping is captured by a tiny substitution matrix. Each L produces one L and one S; each S produces one L and no S. Stacking the counts as columns:

Its characteristic equation is \lambda^2 - \lambda - 1 = 0, whose largest root is \lambda = \tfrac{1+\sqrt5}{2} = \varphi. That eigenvalue is the factor by which the tiling grows at each step, and — because the tile counts settle into the direction of its eigenvector — it also fixes the limiting ratio of L's to S's at exactly \varphi : 1. One little matrix, and the golden ratio falls out.

Inflation and deflation

In two dimensions the same idea wears a second name. A Penrose tiling can be deflated: cut every tile along fixed lines into smaller Penrose tiles — and the smaller tiling is again a perfectly valid Penrose tiling. Or run it backwards and inflate: group the tiles into larger ones. The remarkable fact is that inflation returns the same kind of tiling, scaled up by the golden ratio \varphi.

That invariance under \varphi-scaling — the tiling is similar to itself blown up by \varphi — is the fingerprint of aperiodic order. A periodic wallpaper is invariant under a translation; an aperiodic Penrose tiling is instead invariant under a scaling. It repeats in scale, not in place.

Worked example 2 — the Penrose substitution matrix

The kite-and-dart form of the Penrose tiling deflates by the rule: a kite becomes two kites and one dart, and a dart becomes one kite and one dart. Its substitution matrix is therefore

The largest (Perron–Frobenius) eigenvalue is \lambda = \tfrac{3+\sqrt5}{2} = \varphi^2 \approx 2.618. This is now an area scale factor: deflation shrinks every length by 1/\varphi, so areas shrink by 1/\varphi^2 and the number of tiles multiplies by \varphi^2 each step — precisely the eigenvalue.

And its eigenvector, as before, gives the relative frequencies of the two tile types. Solving (M - \varphi^2 I)\vec{v} = \vec 0 gives a kite-to-dart ratio of

So the largest eigenvalue tells you the scale factor (\varphi for length in 1-D, \varphi^2 for area in 2-D), and its eigenvector tells you the tile frequencies. Both point at \varphi.

The proof of aperiodicity, in one line

Here is the payoff. Suppose, for contradiction, that a Penrose tiling were periodic — that it had a repeating unit cell. A unit cell contains a whole number of kites and a whole number of darts, and the whole tiling is just copies of that cell, so the overall kite-to-dart frequency would have to be a ratio of two integers — a rational number.

But the substitution matrix has just forced that frequency to be \varphi, which is irrational. A rational number cannot equal an irrational one. So no unit cell can exist: the tiling is aperiodic. The irrationality of a single eigenvector ratio, handed to us by the largest eigenvalue of a 2\times2 matrix, rules out every possible period at once.