Most people meet the golden ratio \(\phi = \tfrac{1+\sqrt5}{2}\) as a proportion - the ratio of a regular pentagon's diagonal to its side, or the shape of a rectangle that reproduces itself when you cut off a square. It's a number with a geometric personality.
What's less advertised is that the same number is the arithmetic backbone of one of the most rigid objects in algebra: a subring1 of quaternions called the icosians. And that ring is not just golden-flavored bookkeeping - it is the rotational symmetry of the icosahedron, written in the language of quaternion multiplication.
This post walks the short path from \(\phi\), through the quaternions and the field \(\mathbb{Q}(\sqrt5)\), to the icosians, and shows how they encode the icosahedron.
Quaternions, in one paragraph
A quaternion is a number
where \(i,j,k\) multiply by Hamilton's rule \(i^2 = j^2 = k^2 = ijk = -1\) (so \(ij = k\), \(ji = -k\); the multiplication is associative but not commutative). The norm is \(N(q) = a^2+b^2+c^2+d^2\), and it is multiplicative: \(N(pq) = N(p)N(q)\). The quaternions of norm \(1\) form the unit sphere \(S^3 \subset \mathbb{R}^4\), and they are exactly the object we want for symmetry, because of one classical fact:
A unit quaternion \(q\) acts on \(\mathbb{R}^3\) (the imaginary part \(bi+cj+dk\)) by \(v \mapsto q\,v\,q^{-1}\), and this map is a rotation. Every rotation arises this way, and \(q\) and \(-q\) give the same rotation.
So the unit quaternions are a two-to-one cover of the rotation group \(\mathrm{SO}(3)\). That is to say, \(\pm q\) give the same rotation. Keep that in your back pocket - it's the bridge that will carry 120 unit icosians onto the 60 rotations of the icosahedron; which 60 is the golden arithmetic's doing, below.
The golden ratio and the field \(\mathbb{Q}(\sqrt5)\)
Take a rectangle and cut off the largest square that fits inside it. For almost every rectangle the leftover strip is nothing special - but for one proportion, the strip is a smaller copy of the rectangle you started with, the same shape turned a quarter-turn. That self-similar proportion is the golden ratio, and it pins the number down with no trigonometry at all.
Write the long side as \(a+b\) and the short side as \(a\), so that removing the \(a\times a\) square leaves an \(a\times b\) strip. "Same shape" means the strip's long-to-short ratio matches the original's:
Call that common ratio \(\phi\). The left-hand side is \(1 + \tfrac{b}{a} = 1 + \tfrac{1}{\phi}\), so \(\phi = 1 + \tfrac{1}{\phi}\) - clear the fraction and you get
The same self-reproducing property - \(\phi = 1 + 1/\phi\), the rectangle that keeps a smaller copy of itself - is why the golden ratio turns up in growing things. When a plant adds new elements around a stem by turning through a fixed fraction of a full circle each step, one angle never lets successive elements fall into lines or leave gaps: the golden angle, \(360^\circ/\phi^2 \approx 137.5^\circ\). It wins precisely because \(\phi\) is the "most irrational" number - the one whose continued fraction is \(1 + 1/(1 + 1/(1+\cdots))\), so its multiples spread the most evenly of anything. That single rule sets the seed spirals of a sunflower, the scales of a pinecone, and the nested florets of a head of romanesco2, whose spiral counts come out as consecutive Fibonacci numbers - ratios that march straight toward \(\phi\).
The same quadratic \(x^2 = x + 1\) has a second root, \(\phi' = \tfrac{1-\sqrt5}{2}\), the Galois conjugate of \(\phi\). The two are related by the only nontrivial symmetry of the field \(\mathbb{Q}(\sqrt5)\) - the automorphism that sends \(\sqrt5 \mapsto -\sqrt5\), and hence \(\phi \mapsto \phi'\). Concretely, \(\mathbb{Q}(\sqrt5)\) has two real embeddings (two "places"): one where \(\sqrt5\) is the usual \(+2.236\ldots\), one where you read it as \(-2.236\ldots\). Every statement about \(\phi\) has a mirror statement about \(\phi'\), and they are equally true.
Inside \(\mathbb{Q}(\sqrt5)\) sits the ring \(\mathbb{Z}[\phi]\) - the integers extended by \(\phi\), i.e. all \(m + n\phi\) with \(m,n \in \mathbb{Z}\). Because \(\phi^2 = \phi+1\), this set is closed under multiplication; it's the natural home for "golden integers."
The connection between this arithmetic and geometry is a pair of trigonometric identities:
These say that the angles of fivefold symmetry - i.e. the pentagon's - are golden on the nose. Anything built from fivefold rotations will end up with coordinates in \(\mathbb{Z}[\phi]\), and its "conjugate" version (read \(\sqrt5\) the other way) will be equally valid. That is exactly what happens next.
The icosians
An icosian is a quaternion whose four coordinates all lie in the golden ring:
These form the icosian ring - in the language of number theory, a maximal order in the quaternion algebra over \(\mathbb{Q}(\sqrt5)\), the golden analogue of the Hurwitz integers over \(\mathbb{Q}\). What makes them special is the group hiding among the units.
Here "unit" means norm \(1\), in the quaternion norm3 from the first section:
Its multiplicativity, \(N(pq) = N(p)N(q)\), is exactly what makes the units close under multiplication - and geometrically \(N(q) = 1\) just places \(q\) on the unit sphere \(S^3\).
There are exactly 120 unit icosians, and they fall into three families, sorted by how much of \(\sqrt5\) they carry:
| how many | coordinates (all divided by 2) | type |
|---|---|---|
| 8 | \((\pm 2, 0, 0, 0)\) and its rearrangements \(=\ \pm 1, \pm i, \pm j, \pm k\) | rational |
| 16 | \((\pm 1, \pm 1, \pm 1, \pm 1)\) | rational |
| 96 | even rearrangements of \((0,\ \pm 1,\ \pm 1/\phi,\ \pm\phi)\) | golden (carry \(\sqrt5\)) |
The first \(24\) are the ordinary quaternion units (the Hurwitz units); they'd be there with or without the golden ratio. The last \(96\) are the new content: they cannot be written without \(\sqrt5\), and applying the Galois automorphism \(\sqrt5 \mapsto -\sqrt5\) permutes them among themselves. Together the three families are exactly the \(120\) vertices of the 600-cell, the regular four-dimensional polytope whose symmetry is the \(H_4\) reflection group.
This is not hand-waving: you can build all \(120\) from the table above and check by direct quaternion multiplication that they are closed - the product of two unit icosians is again a unit icosian - and that the resulting group has order exactly \(120\).4
How the icosians represent the icosahedron
As a finite group under multiplication, the \(120\) unit icosians are the
the perfect double cover of \(A_5\), the group of the \(60\) rotational symmetries of the icosahedron. Its center is \(\{\pm 1\}\), and quotienting by it collapses the \(120\) icosians two-to-one onto those \(60\) rotations:
This is the "\(\pm q\) give the same rotation" fact from the first section, now made exact. The rotation group of the icosahedron is \(A_5\); lift each rotation to the unit quaternions implementing it and you pick up a sign ambiguity, doubling \(60\) to \(120\). Those \(120\) lifted quaternions are precisely the unit icosians. The icosahedron's symmetry and the golden arithmetic are two views of one object.
And the golden family is where the icosahedron literally lives. The twelve vertices of a regular icosahedron are the corners of three mutually perpendicular golden rectangles:
Those vertices are cyclic rearrangements of \((0, \pm 1, \pm\phi)\) - the same golden coordinates, now in \(\mathbb{R}^3\), that the identity \(2\cos\tfrac{\pi}{5} = \phi\) forced on us. The fivefold axes of the icosahedron are golden angles; the quaternions that rotate about them have golden coordinates; the closed system of all such quaternions is the icosian ring. Golden arithmetic \(\rightarrow\) icosian ring \(\rightarrow\) icosahedral symmetry, and back again.
Where this goes
The icosians are a small, completely concrete set of numbers: a list of \(120\) quaternions you could fit on a napkin. They tie together a field extension, a regular polytope, and the symmetry group of a Platonic solid. That's already a generous amount of structure for one ring.
But of course, there's more: the \(\mathbb{Z}[\phi]\)-lattice spanned by the icosians is, up to scale, the \(E_8\) root lattice. The binary icosahedral group \(2\cdot A_5\) is precisely the group that labels \(E_8\) in the famous McKay correspondence... but that is a story for another post.
-
To be more precise, the icosian ring - the full \(\mathbb{Z}[\phi]\)-span of the 120 unit icosians - is closed under \(+\), \(-\), and \(\times\) and contains \(1\), so it's a genuine (noncommutative, unital) subring of Hamilton's quaternions. More than that, it's an order: a subring that is simultaneously a full lattice. In fact it's a maximal order in the quaternion algebra \(\left(\frac{-1,-1}{\mathbb{Q}(\sqrt5)}\right)\): the golden-field analogue of the Hurwitz quaternions inside \(\mathbb{H}(\mathbb{Q})\). ↩
-
Romanesco's florets are arranged in phyllotactic spirals whose counts are consecutive Fibonacci numbers, set by the golden-angle (~137.5°) packing - exactly the same mechanism as the sunflower and pinecone. This self-similarity approximates a fractal, but should not be confused with one. The nautilus shell, the usual poster child for the golden ratio, is a total red herring: it's a logarithmic spiral, but its growth rate is nowhere near \(\phi\). ↩
-
In quaternion algebra and number theory, "the norm" is the reduced norm \(N(q) = q\bar q\), which is the square of the Euclidean length. The Euclidean length itself is \(\lVert q\rVert = \sqrt{N(q)}\), so indeed \(N(q) = \lVert q\rVert^2\). ↩
-
If you'd like to check the closure and the order-\(120\) claim yourself, it's about fifteen lines: build the three families as quaternions with coordinates stored exactly as pairs \((p, q)\) meaning \(p + q\sqrt5\), multiply with Hamilton's rule, and confirm the set is closed and has size \(120\). ↩