Why five-fold is impossible

The trace argument settles the question in one line, and it settles it so quickly that it can leave a reader unsatisfied. Nothing was constructed, no case was examined, and the impossibility arrived as a remark about integers.

Here is a second proof of the five-fold case that constructs the contradiction explicitly. It is longer, it uses distances instead of matrices, and it shows why the impossibility is about how close lattice points can get.

The one property being used #

A lattice is discrete. Its points are separated, not packed arbitrarily close, and the consequence is that there is a shortest non-zero lattice vector.

That is the entire input to the argument. Discreteness is what distinguishes a lattice from an arbitrary set of points closed under addition — the set of all rational multiples of a vector is closed under addition and has no shortest member, and it is not a lattice.

The existence of a shortest vector gives the proof something to contradict. Construct a shorter one and something has gone wrong, and the only assumption available to blame is the one made at the start.

The construction #

Assume the pattern has a five-fold rotation centre. Let $\mathbf{a}$ be a shortest non-zero lattice vector, and let $R$ be the rotation by $72°$ about the centre.

Because $R$ is a symmetry, it maps the lattice onto itself, so $R\mathbf{a}$ is a lattice vector, and so is $R^{-1}\mathbf{a}$. Both have the same length as $\mathbf{a}$, since rotations preserve length.

Now form the sum $R\mathbf{a} + R^{-1}\mathbf{a}$. It is a lattice vector, being the sum of two of them. Its length is computable directly: the two vectors each make an angle of $72°$ with $\mathbf{a}$, one on each side, so their sum lies along $\mathbf{a}$ and has length

$$2|\mathbf{a}|\cos 72° \approx 0.618\,|\mathbf{a}|.$$

That is shorter than $\mathbf{a}$, and it is not zero. So the lattice contains a non-zero vector shorter than its shortest non-zero vector, which is a contradiction, and the five-fold centre cannot exist.

Why the sum is a lattice vector at all #

The step that does the work is easy to read past, so it is worth slowing down on.

$R\mathbf{a}$ is a lattice vector because $R$ is assumed to be a symmetry of the pattern, and a symmetry maps the pattern’s translation set onto itself. $R^{-1}\mathbf{a}$ is a lattice vector for the same reason, since the inverse of a symmetry is a symmetry. And their sum is a lattice vector because the lattice is closed under addition — which is not an extra assumption but part of what being a lattice means.

So all three facts come from the same place: the translations of a pattern form a group, and a group is closed under its operation. Take away the closure and the construction has nothing to construct with.

It is also worth noticing what is not assumed. Nothing is said about the motif, about the pattern’s other symmetries, or about which of the five lattice types is underneath. The argument runs on the existence of a shortest vector and on closure, and those are properties every lattice has.

Where the golden ratio walks in #

The factor $2\cos 72° = 0.618\ldots$ is not an accidental decimal. It is $1/\varphi$, the reciprocal of the golden ratio, and it is the same number that governs the inflation of a Penrose tiling.

That is a pleasing coincidence at first sight and it is not a coincidence at all. The golden ratio is the fundamental unit of the algebra generated by five-fold rotation, so any quantity built out of $72°$ angles tends to be an expression in $\varphi$. The number that makes five-fold periodicity impossible is the same number that makes five-fold aperiodic order self-similar, and the connection runs deeper than the shared symbol: it is because $0.618$ is irrational that the tiling never repeats, and because it is a quadratic irrational that the tiling has an inflation rule at all.

So the shrinking construction is not merely a proof of impossibility. It is the first sighting of the structure that replaces periodicity when periodicity is unavailable.

The same argument at seven, eight and above #

The construction adapts, and adapting it is the best way to see what the number $2\cos\theta$ is doing.

For a rotation of order $n$, the sum $R\mathbf{a} + R^{-1}\mathbf{a}$ has length $2|\mathbf{a}|\cos(2\pi/n)$. Whenever that factor lies strictly between zero and one, the sum is a non-zero lattice vector shorter than the shortest, and the same contradiction follows.

For $n = 7$ the factor exceeds one, so a slightly different combination is needed: subtract rather than add, or take the difference $R\mathbf{a} - \mathbf{a}$, and a short vector appears again. For $n \geq 7$ generally the difference $R\mathbf{a} - \mathbf{a}$ has length $2|\mathbf{a}|\sin(\pi/n)$, which is less than $|\mathbf{a}|$ as soon as $\pi/n < \pi/6$, that is as soon as $n > 6$.

So above six every order fails by the difference construction, and at five the sum construction does it. Orders one, two, three, four and six survive because in each of those cases both combinations land exactly on a lattice vector of permitted length rather than in between.

Three cases is not a theorem, and the theorem is a statement about all of them. Both combinations are available at every order, so both can be computed at every order and the pair plotted together.

The two curves cross at six, and that is the whole reason the answer is the list it is. Below six the sum is the dangerous combination and only five is close enough to the middle for it to bite; above six the difference takes over and never lets go, because $2\sin(\pi/n)$ falls to zero as the rotation gets finer. Six is where both are exactly one — the last order at which a lattice can hold a rotation without being forced to hold something shorter than its own shortest vector — and there is no arithmetic anywhere in the argument, only two lengths and a comparison.

The construction, checked rather than drawn #

The figures on this page are not illustrations of a known result. Each one computes the descent and reports what it finds.

The generator takes the order, builds the star of rotated copies of the shortest vector, forms both candidate combinations — the sum of the two neighbours and the difference of a neighbour with the original — measures whichever is shorter, and declares a contradiction when that length falls below the vector it started from. It then asserts that the verdict matches the crystallographic restriction: a contradiction for every order except one, two, three, four and six, and none for those five.

That assertion earned its place immediately. An earlier version of the generator computed the shrink factor by a formula that fires at threefold and stays silent at fivefold — the result exactly inverted. The picture it produced was entirely convincing: a five-pointed star, some vectors, a confident label. Nothing about it looked wrong, and nothing would have caught it except making the figure state a claim that could be tested against an independently known answer.

That is the case worth staring at, because it is where the answer changes. At six the constructed vector is exactly as long as the one it came from; at seven it is shorter and the order dies. There is nothing gradual about the transition and no tolerance involved in detecting it.

The shape of the argument is descent #

One step is enough here, because a lattice has a shortest vector and the construction produces a shorter one. It is worth noticing that the construction does not stop after one step, and that the version which does not stop is an argument of a recognisable kind.

Feed the new vector back in. The combination is itself a lattice vector, and the five-fold rotation is still assumed, so the same construction applies to it and returns something shorter again — by the same factor. Iterating gives an infinite sequence of lattice vectors whose lengths fall geometrically towards zero and never reach it, since no step can produce the zero vector.

That is Fermat’s infinite descent, and it is the same shape as the geometric proof that the diagonal of a square is incommensurable with its side: assume a common measure, construct a smaller square whose side and diagonal are measured by the same unit, and repeat forever inside a figure that cannot contain an infinite decreasing sequence of whole multiples.

The two versions assume slightly different things, which is the reason to have both. The one-step version needs a shortest vector to exist — that is discreteness, stated as a minimum. The descent version needs only that lengths cannot decrease forever without reaching zero, which is discreteness stated as the absence of an accumulation point at the origin. They are equivalent for a lattice and they are not equivalent in general, and the distinction is exactly what a quasicrystal exploits: its diffraction peaks do accumulate, densely, and it survives the argument by having no lattice of translations for the rotation to act on in the first place.

What the picture makes visible that the algebra does not #

The two proofs are equivalent and they leave a reader with different intuitions, which is the argument for meeting both.

The trace proof says: a rotation compatible with a lattice must have an integer trace, and only five angles manage it. It is exact, complete and slightly opaque. Nothing in it suggests why a lattice should care.

The shrinking proof says: a lattice has a minimum spacing, and a forbidden rotation manufactures points closer together than the minimum. That is a statement about crowding, and it makes the result feel inevitable rather than arithmetical. It also explains what goes wrong physically: an attempt to build a five-fold periodic structure produces atoms that would have to sit impossibly close together.

Neither intuition is complete. The trace argument generalises immediately to any dimension and the shrinking argument does not, at least not without work. The shrinking argument explains the failure and the trace argument merely certifies it.

The physical reading #

The algebra says a forbidden rotation manufactures points too close together. It is worth asking what that means for actual matter, since the restriction is a statement about crystals and not only about drawings.

Atoms have effective sizes, and a structure that places two of them at a distance much shorter than the sum of their radii is not a structure — it is an arrangement no material adopts, because the energy cost is enormous. The descent construction, read physically, says that a five-fold periodic arrangement would demand exactly that: given any repeat distance, the five-fold symmetry generates a shorter one, and iterating generates shorter ones still without limit.

That last point is the sharpest version. The construction can be applied to its own output. Starting from a shortest vector of length $\ell$ it produces one of length $0.618\ell$; applying it again gives $0.382\ell$; and the sequence descends to zero. A periodic five-fold structure would need lattice points arbitrarily close together, which is not a matter of being energetically expensive but of not being discrete at all.

Discreteness is the property that fails, and it fails catastrophically rather than marginally. That is why the restriction admits no near misses and no exceptions at high pressure or low temperature. It is not a statement about what is favourable. It is a statement about what is a lattice.

Two routes matter more than one route checked twice #

There is a methodological point here that this site applies well beyond this essay.

A fact carrying as much weight as the crystallographic restriction should be reachable by more than one road. Not because either proof is doubted, but because independent routes fail differently: an error in a chain of reasoning tends to be invisible from inside that chain and obvious from outside it.

The same principle governs the diffraction figures here. A pattern figure asserts a group directly from its point set; a diffraction figure computes what the same point set would scatter, and reads the symmetry off the reflections that vanish. Two calculations sharing nothing but the atom positions. When they agree, the agreement is evidence; when a single calculation agrees with itself, that is not evidence of anything.

What the theorem forbids, and what it does not #

The result is stated so often in the compressed form “crystals cannot be five-fold” that it is worth spending a paragraph on what the sentence actually excludes, because the commonest misreading is a strong one.

The theorem forbids a global operation: there is no rotation by $72°$ about any point that maps the whole infinite pattern onto itself. It says nothing whatever about local arrangement. A periodic pattern may be built entirely out of regular pentagons, may have five-fold clusters at every lattice point, and may look five-fold everywhere a reader chooses to look. What it cannot have is a five-fold symmetry of the pattern, and that is a claim about a map of the plane rather than about the shapes in it.

The distinction is not a quibble; it is the difference between a property of a motif and a property of a group. The detector used throughout this site makes it concrete: it enumerates operations that map the whole point set to itself, so a five-fold cluster placed at every lattice point is simply never a candidate. The rotation that fixes one cluster moves its neighbours, and an operation that moves any point off the set is rejected. Nothing in the enumeration inspects shapes.

Real materials use this loophole constantly. Molecules with five-fold symmetry — ferrocene, cyclopentadienyl rings, many porphyrins — crystallise perfectly well. Their crystals are periodic and their space groups contain no five-fold operation; the molecule’s own symmetry is simply not a symmetry of the crystal, and the molecule sits at a general position where it is required to be invariant under nothing at all.

So the price of the theorem is narrower than it sounds. A crystal pays for periodicity by giving up five-fold symmetry as an operation of the pattern. It does not give up five-fold objects, five-fold neighbourhoods, or the appearance of five-foldness at any scale short of the whole. The thing that is genuinely forbidden is the one thing a diffraction experiment measures, which is why the 1982 result was startling rather than merely surprising.

Where the argument stops #

The proof has a hypothesis, and it is worth naming precisely, because the famous counterexample of 1982 attacks exactly that hypothesis and nothing else.

The proof assumes a lattice — a discrete set of translations, with a shortest vector. Remove that and every step fails. There is no shortest vector to contradict, the sum $R\mathbf{a} + R^{-1}\mathbf{a}$ need not be anything in particular, and the argument has no purchase.

A Penrose tiling has no translational symmetry whatever. No slide, however large, maps it onto itself. So it is not restricted, and its five-fold symmetry costs it nothing.

What the tiling does instead #

If five-fold periodicity is impossible, and five-fold order is nevertheless observed, something has to take periodicity’s place. The something has a name and a mechanism.

A Penrose tiling is built from two rhombi with matching rules that forbid any periodic arrangement. It has five-fold rotational symmetry about certain points, it fills the plane completely, and it never repeats. What it has instead of translations is repetitivity: every finite patch that occurs anywhere occurs infinitely often, and within a bounded distance of any point. That is enough to make the pattern determinate — knowing a large enough patch constrains the rest — without giving it a single translation.

The scaling symmetry is the substitute. A Penrose tiling maps onto itself not under any slide, but under an inflation: rescale by $\varphi$, redraw, and the same tiling reappears. That is a symmetry of a kind the four plane motions do not include, which is why the classification of the seventeen has nothing to say about it and why the restriction does not bind it.

What Shechtman measured #

In April 1982 Dan Shechtman obtained an electron diffraction pattern from a rapidly cooled aluminium–manganese alloy with sharp spots arranged in tenfold symmetry.

What that pattern looked like matters less here than what it was taken to mean, and the reasoning that made it a crisis can be set out in three lines. Sharp spots had always been read as evidence of periodicity. Periodicity forbids tenfold — by the argument on this page, and by the trace argument beside it. The alloy’s spots were sharp and their arrangement was tenfold. One of those three had to give, and the drawing of the pattern is where the alloy’s own essay begins rather than something this page needs in order to say which one it was.

The proof above was never in danger, and the belief that gave way was one nobody had written down: that sharp diffraction requires a lattice. It does not. It requires long-range order, which a pattern can have without repeating, and separating those two ideas is what the episode forced.

The rest of that story — how twinning was excluded, why publication took thirty-one months, and what the material turned out to be — is its own essay. What belongs here is only the relationship to the theorem: the alloy has no lattice, so the first line of this page’s argument does not apply to it, and nothing on this page needs revising.

Where the ladder goes next #

The immediate companion is the trace proof, if it has not already been read, since the two arguments illuminate different halves of the same fact.

The immediate consequence is the seventeen, which is what the five permitted orders generate when combined with everything compatible.

And the sequel is the exception that turns out not to be one: Penrose tilings, the inflation that generates them, and what a quasicrystal replaces periodicity with.

What the pictures here cannot show. The construction on this page is drawn at one scale with one starting vector; the proof is about every lattice and every choice. A drawing can illustrate a contradiction and cannot establish it, and the establishing is done by the algebra rather than by the picture.