The Cayley–Dickson tower: rank 2, 4 and 8 (GPR 19.18) #
Iterating the Cayley–Dickson process three times from the base ring produces, for scalars
μ₁, μ₂, μ₃ ∈ k, conic algebras of ranks 2, 4 and 8:
Cay₂ k μ₁ = Cay(k, μ₁),
Cay₄ k μ₁ μ₂ = Cay(Cay₂, μ₂) — a quaternion algebra,
Cay₈ k μ₁ μ₂ μ₃ = Cay(Cay₄, μ₃) — an octonion algebra, which mathlib does
not have.
Everything is over an arbitrary commutative ring k, with arbitrary scalars; the hypotheses
2 ∈ kˣ and μᵢ ∈ kˣ appear only on the theorem that the tower consists of composition
algebras.
Why the tower stops at 8 #
Each rung is built by a theorem whose hypothesis is a property of the rung below.
ConicAlgebra.mul_comm_cay needs its input's conjugation to be the identity, which holds for the
base ring; ConicAlgebra.mul_assoc_cay needs its input commutative; and
ConicAlgebra.isMultiplicative_cay (GPR Exc. 18.18) and ConicAlgebra.isAlternative_cay
(GPR 18.14(c)) need it associative. So each property may be lost one rung after the property
that implies it, and the fourth double — the sedenions — is in general neither alternative
nor
multiplicative. That is the mechanism behind the rank dichotomy 1, 2, 4, 8 of GPR 19.18.
What is proved positively here is that Cay₂ is commutative, Cay₄ is associative, and
Cay₈
is alternative with multiplicative norm — each over an arbitrary base.
⚠ The negative half is genuinely characteristic-dependent and is therefore not stated in
this file. Over a base in which 2 = 0 the conjugation of Cay₂ is again the identity, so
ConicAlgebra.mul_comm_cay applies once more and the tower stays commutative — the ranks still
double, but the algebras do not become non-commutative and then non-associative in the classical
way. The explicit witnesses that ℍ is not commutative and 𝕆 is not associative are proved
over
ℝ in Composition/Examples.lean, from the general
ConicAlgebra.exists_mul_ne_mul_cay and ConicAlgebra.exists_mul_mul_ne_mul_mul_cay.
Once such a witness is in hand for a rung, ConicAlgebra.not_isAlternative_cay_of_mul_mul_ne —
the converse half of GPR 18.14(c) — turns it into the failure of alternativity one rung up, with
no hypothesis on the scalar. That is the route to the sedenions of GPR 19.23.
Main definitions #
ConicAlgebra.Cay₂,ConicAlgebra.Cay₄,ConicAlgebra.Cay₈: the underlying modules.ConicAlgebra.cay₂,ConicAlgebra.cay₄,ConicAlgebra.cay₈: their conic structures.ConicAlgebra.Octonion,ConicAlgebra.octonion: notation for the rank-8 case.
Main results #
ConicAlgebra.isMultiplicative_cay₈: the octonion norm permits composition.ConicAlgebra.isAlternative_cay₈: octonion algebras are alternative (directly, fromGPR 18.14(c), without going throughGPR 19.3).ConicAlgebra.isComposition_cay₈: for2 ∈ kˣandμᵢunits, the octonion algebra is a composition algebra; likewiseisComposition_cay₂andisComposition_cay₄.ConicAlgebra.finrank_cay₂,.finrank_cay₄,.finrank_cay₈: the ranks are2,4,8.
Implementation notes #
The Ring, CommRing and Algebra structures on the rungs are declared as genuine instances
here, unlike the defs they come from in Conic/CayleyDicksonAssociative.lean: at each rung the
side conditions are discharged by theorems (ConicAlgebra.baseRing_conj_apply at the bottom,
ConicAlgebra.ConjIsInvolution.cay above it) rather than assumed, so there is nothing for instance
search to guess. This is what makes the tower usable without letI at every step.
⚠ Cay₄ is not mathlib's QuaternionAlgebra, and the two are incomparable in general —
see
overview trap 3. QuaternionAlgebra takes arbitrary structure constants and always produces a
free module; Cay₄ produces a composition algebra exactly under the hypotheses below. Relating
them is the job of a later file.
References #
- [GPR] Garibaldi, Petersson, Racine, Albert Algebras over Commutative Rings, 18.14, Exc. 18.18, 19.18.
Rank 2 #
The rank-2 Cayley–Dickson algebra Cay(k, μ₁), as a k-module.
Equations
- ConicAlgebra.Cay₂ k μ₁ = (ConicAlgebra.baseRing k).Cay μ₁
Instances For
The conic structure on Cay₂, i.e. GPR 18.4 applied to the base ring.
Equations
- ConicAlgebra.cay₂ k μ₁ = (ConicAlgebra.baseRing k).cay μ₁
Instances For
Cay₂ is commutative: the conjugation of the base ring is the identity
(ConicAlgebra.baseRing_conj_apply), which is exactly the hypothesis of
ConicAlgebra.mul_comm_cay. This is the last commutative rung.
Equations
Equations
The rank-2 norm is multiplicative: GPR Exc. 18.18 applied to the base ring, which is
associative
with multiplicative norm.
Rank 4 #
The rank-4 Cayley–Dickson algebra Cay(Cay(k, μ₁), μ₂) — a quaternion algebra.
⚠ Not mathlib's QuaternionAlgebra; see the module docstring.
Equations
- ConicAlgebra.Cay₄ k μ₁ μ₂ = (ConicAlgebra.cay₂ k μ₁).Cay μ₂
Instances For
The conic structure on Cay₄.
Equations
- ConicAlgebra.cay₄ k μ₁ μ₂ = (ConicAlgebra.cay₂ k μ₁).cay μ₂
Instances For
Cay₄ is associative, by GPR 18.14(b) applied to the commutative Cay₂. It is the
last
associative rung, which is why the tower produces composition algebras only up to rank 8.
Equations
- ConicAlgebra.instRingCay₄ k μ₁ μ₂ = ConicAlgebra.Cay.ring (ConicAlgebra.cay₂ k μ₁) μ₂ ⋯
Equations
- ConicAlgebra.instAlgebraCay₄ k μ₁ μ₂ = ConicAlgebra.Cay.algebra (ConicAlgebra.cay₂ k μ₁) μ₂ ⋯
Rank 8: octonion algebras #
The rank-8 Cayley–Dickson algebra Cay(Cay(Cay(k, μ₁), μ₂), μ₃): an
octonion algebra.
mathlib has no octonions, so this is the definition of record here.
Equations
- ConicAlgebra.Cay₈ k μ₁ μ₂ μ₃ = (ConicAlgebra.cay₄ k μ₁ μ₂).Cay μ₃
Instances For
The conic structure on Cay₈.
Equations
- ConicAlgebra.cay₈ k μ₁ μ₂ μ₃ = (ConicAlgebra.cay₄ k μ₁ μ₂).cay μ₃
Instances For
The underlying module of an octonion algebra, under its usual name.
Equations
- ConicAlgebra.Octonion k μ₁ μ₂ μ₃ = ConicAlgebra.Cay₈ k μ₁ μ₂ μ₃
Instances For
An octonion algebra, under its usual name.
Equations
- ConicAlgebra.octonion k μ₁ μ₂ μ₃ = ConicAlgebra.cay₈ k μ₁ μ₂ μ₃
Instances For
Octonion algebras are alternative — GPR 18.14(c) applied to the associative Cay₄.
This is a direct proof, independent of GPR 19.3: it needs no regularity of the norm, no
hypothesis on 2, and no hypothesis on the μᵢ, whereas the general
ConicAlgebra.IsComposition.isAlternative needs the norm's radical to vanish.
The octonion norm permits composition, n(xy) = n(x)n(y) — GPR Exc. 18.18 applied to the
associative Cay₄. Again no hypothesis on 2 or on the μᵢ.
This is the last rung covered uniformly by the associative-input theorem used here. For a
further double, GPR Exc. 18.18 gives the sharp criterion μ₄[B, B, B] ⊆ Rad(Dn_B).
Non-associativity alone does not force failure of norm multiplicativity; the criterion is
automatic when μ₄ = 0.
The ranks are 2, 4, 8 #
ConicAlgebra.Cay.finrank_cay doubles the rank at each step, starting from finrank k k = 1.
These need k nontrivial (otherwise finrank is 0 throughout) but nothing else — no unit
scalars, no hypothesis on 2, and no composition.
An octonion algebra has rank 8.
The tower consists of composition algebras #
Here, and only here, the scalars must be units and 2 must be invertible. The hypothesis on 2
is what turns non-singularity of each norm into the regularity that
ConicAlgebra.IsComposition.cay consumes; without it the very first rung already fails, since
Cay(k, μ) in characteristic 2 is inseparable and not a composition algebra.
An octonion algebra with unit scalars, over a base in which 2 is invertible, is a
composition algebra.