Composition algebras (GPR 19.5) #
A composition algebra over a commutative ring k is a conic algebra C (GPR 16.1) which is
finitely generated projective as a k-module and whose norm is
- non-singular in
GPR's sense (QuadraticForm.IsGPRNonsingular,GN C.8(e): the radical vanishes after base change to every field), and - multiplicative (
ConicAlgebra.IsMultiplicative,GPR 19.1:n(xy) = n(x) n(y)).
Over a field these are exactly k, the quadratic étale algebras, the quaternion algebras and the
octonion algebras, of dimensions 1, 2, 4, 8.
Main definitions #
ConicAlgebra.IsComposition:GPR 19.5.
Main results #
ConicAlgebra.isComposition_baseRing:kitself, with norm the squaring map, is the rank-one composition algebra.ConicAlgebra.IsComposition.isGPRRegular: when2 ∈ kˣthe norm of a composition algebra is regular, byGN (C.8.2). This is false without a hypothesis on2— seeQuadraticForm.rankOne, which iskitself in characteristic 2.
Implementation notes #
Why a predicate and not a structure extending ConicAlgebra. The norm of a conic algebra is
data; being a composition algebra is a property of that data, so IsComposition is Prop-valued
and takes C : ConicAlgebra k A as an argument. This keeps the three axes of overview §2
separate: the algebra structure is in A's instances, the module regularity is in the
[Module.Finite k A] [Module.Projective k A] instance arguments, and only the two genuinely
norm-theoretic clauses are fields of IsComposition.
⚠ Finite generation is assumed here, where GPR derives it. This is a genuine, if slight,
divergence from GPR 19.5, and it goes in the direction of less generality. GPR assumes only
that the underlying module is projective of locally constant rank — allowing that rank to be
infinite a priori — and then proves the rank is finite, indeed one of 1, 2, 4, 8 (GPR 19.18).
We assume [Module.Finite k A] outright. The two definitions therefore pick out the same objects,
but only a posteriori: ours builds in a conclusion of the theory rather than deriving it. So a
GPR composition algebra is an IsComposition only by way of GPR 19.18, and until that theorem
is formalized here, a statement about "every composition algebra" is strictly a statement about the
finitely generated ones.
No separate "locally constant rank" clause, although GPR 19.5 states one: over a commutative
ring the rank of a finitely generated projective module is automatically locally constant on
Spec k (this project proves as much in Module/RankDecomposition.lean,
Module.finite_range_rankAtStalk and Module.rankIdempotent), so the clause would be redundant.
The constancy of the rank — which is what makes the rank dichotomy 1, 2, 4, 8 a statement about
a single number — is a hypothesis on theorems, not part of the definition.
No PreCompositionAlgebra. GPR defines one; we deliberately do not (overview, Design
decisions 1): the bundle "projective + non-degenerate + multiplicative" is not stable under base
change (GPR 19.7(a), with GPR 19.4 as the counterexample), and the one theorem that motivates
the notion (GPR 19.3) is stated with those three as separate hypotheses instead.
Universes. QuadraticForm.IsGPRNonsingular quantifies over all fields in a universe, so it
carries a third universe parameter beyond those of k and A; IsComposition inherits it, and
names it the same way QuadraticForm.IsEKMNondegenerate does.
References #
- [GPR] Garibaldi, Petersson, Racine, Albert Algebras over Commutative Rings, 19.1, 19.5.
- [GN] Gille, Neher, Group schemes over LG-rings…, European J. Math. 11 (2025), art. 42, C.8 (non-singularity).
GPR 19.5: C is a composition algebra. A conic algebra whose underlying module is
finitely generated projective (an instance argument, not a field) and whose norm is non-singular
in GPR's sense and multiplicative.
⚠ Finite generation is a hypothesis here and a theorem in GPR, which assumes only projective
of locally constant rank and derives finiteness of the rank from GPR 19.18. This definition is
correspondingly, and very slightly, less general; see the module docstring.
⚠ Non-singular, not regular and not non-degenerate. QuadraticForm.IsGPRNonsingular is the
base-change-stable vanishing of the radical (GN C.8(e)), which is what makes composition algebras
stable under base change. It is strictly weaker than QuadraticForm.IsGPRRegular unless 2 ∈ kˣ
(QuadraticForm.isGPRRegular_iff_isGPRNonsingular), and it is not mathlib's
LinearMap.BilinForm.Nondegenerate; see overview §6(b).
- isGPRNonsingular : C.norm.IsGPRNonsingular
The norm is non-singular in
GPR's sense (GN C.8(e)): its radical vanishes after base change to every field. - isMultiplicative : C.IsMultiplicative
The norm permits composition:
n(xy) = n(x) n(y)(GPR 19.1).
Instances For
Norm-associativity of a composition algebra (GPR 16.12(4) and 17.2(a)):
n(xy, z) = n(y, (conj x)z).
Norm-associativity of a composition algebra (GPR 16.12(3) and 17.2(a)):
n(xy, z) = n(x, z(conj y)).
When 2 ∈ kˣ, the norm of a composition algebra is regular (GN (C.8.2)), i.e. its polar
form is an isomorphism onto the dual and not merely injective.
⚠ The hypothesis on 2 is not removable: ConicAlgebra.baseRing k is a composition algebra whose
norm is QuadraticForm.rankOne 1, and over ℤ that form is non-singular but not regular
(QuadraticForm.not_isGPRRegular_rankOne_one_int).
The rank-one composition algebra. k itself, with norm the squaring map
(ConicAlgebra.baseRing), is a composition algebra: the norm is QuadraticForm.rankOne 1, which is
non-singular over any commutative ring (QuadraticForm.isGPRNonsingular_rankOne), and (xy)² = x²y².
Note that this holds with no hypothesis on 2, even though ⟨1⟩ is regular only when 2 is a
unit.