Multiplicative conic algebras and norm-associativity #
A conic algebra is multiplicative (GPR 19.1; the older literature says its norm permits
composition) if n(xy) = n(x) n(y) for all x, y. This is the second half of the definition of a
composition algebra (GPR 19.5, Layer 7), the first half being a regularity condition on the norm.
The point of this file is that multiplicativity is much stronger than it looks: linearizing it
twice turns the quadratic statement n(xy) = n(x)n(y) into a bilinear one, and that bilinear
identity is exactly GPR's norm-associativity
n(xy, z) = n(y, (conj x)z), n(xy, z) = n(x, z(conj y)),
which is the workhorse of the whole composition-algebra theory. Everything here holds in an
arbitrary conic algebra over an arbitrary commutative ring: no alternativity, no regularity, no
projectivity, and nothing on 2. Only additivity of the norm's polar form is used.
Main definitions #
ConicAlgebra.IsMultiplicative:n(xy) = n(x) n(y).ConicAlgebra.HasTwoSidedInverses: every nonzero element has a two-sided multiplicative inverse. ⚠ This is weaker than being a division algebra in the sense ofGPR 8.6; see its docstring.
Main results #
QuadraticForm.polar_mul_mul_right_of_map_mul,QuadraticForm.polar_mul_mul_left_of_map_mul,QuadraticForm.polar_mul_add_polar_mul_of_map_mul: the linearizationsGPR (19.1.3)–(19.1.5)for any quadratic form permitting composition on a non-associative ring, with no conic structure.ConicAlgebra.IsMultiplicative.polar_mul_mul_right,ConicAlgebra.IsMultiplicative.polar_mul_mul_left: the first linearization, in either variable,n(xz, yz) = n(x, y) n(z)andn(xy, xz) = n(x) n(y, z).ConicAlgebra.IsMultiplicative.polar_mul_add_polar_mul: the second linearization,n(xz, yw) + n(xw, yz) = n(x, y) n(z, w).ConicAlgebra.IsMultiplicative.polar_mul_left: norm-associativity,n(xy, z) = n(y, (conj x)z).ConicAlgebra.IsMultiplicative.polar_mul_right: norm-associativity on the other side,n(xy, z) = n(x, z(conj y)).ConicAlgebra.IsMultiplicative.trace_mul:t(xy) = t(x)t(y) - n(x, y).ConicAlgebra.IsMultiplicative.anisotropic_iff_noZeroDivisors_of_isDomain: over a domain, the norm is anisotropic exactly when the algebra has no zero divisors.ConicAlgebra.isMultiplicative_baseRing: the base ring, with norm the squaring map, is multiplicative.
Implementation notes #
n(x, y) above is QuadraticMap.polar C.norm x y, the un-halved polar form
(overview §6(a)); the two linearizations below are obtained by additivity alone, so every
statement here is valid in characteristic 2.
The two norm-associativity identities are genuinely two statements, not one applied to the
opposite algebra: A is not assumed associative, commutative or alternative, so there is no
symmetry to exploit. They are proved the same way, by specializing the second linearization at
y = 1 and at w = 1 respectively.
References #
- [GPR] Garibaldi, Petersson, Racine, Albert Algebras over Commutative Rings, 16.5, 16.12--16.13, 19.1, 22.18.
Quadratic forms permitting composition #
GPR (19.1.4): a quadratic form permitting composition satisfies
n(xz, yz) = n(x, y) n(z). Only right distributivity is used; compare
ConicAlgebra.IsMultiplicative.polar_mul_mul_right, the same identity for a conic norm.
GPR (19.1.3): a quadratic form permitting composition satisfies
n(xy, xz) = n(x) n(y, z). Only left distributivity is used; compare
ConicAlgebra.IsMultiplicative.polar_mul_mul_left.
GPR (19.1.5): a quadratic form permitting composition satisfies
n(xz, yw) + n(xw, yz) = n(x, y) n(z, w). This is GPR's
n(xy, wz) + n(wy, xz) = n(x, w) n(y, z) with the variables renamed, in the shape of
ConicAlgebra.IsMultiplicative.polar_mul_add_polar_mul; it is (19.1.4) linearized in z.
A conic algebra is multiplicative (GPR 19.1) if its norm permits composition:
n(xy) = n(x) n(y).
This is a property of the pair (A, n), not of A alone — the norm is data — so it is a
predicate on C : ConicAlgebra k A rather than a typeclass on A. It carries no regularity and
no projectivity: those are separate axes (overview §2), and a composition algebra
(GPR 19.5) is a conic algebra having all three.
Instances For
Every nonzero element has a two-sided inverse: for each x ≠ 0 there is y with
x * y = 1 and y * x = 1.
⚠ This property is weaker than being a division algebra as defined in GPR 8.6, and is
deliberately not called IsDivision. GPR 8.6 calls A a division algebra when A ≠ 0 and, for
every u ≠ 0, the left and right multiplications L_u and R_u are bijective. Without
associativity a two-sided inverse does not give that: the real sedenions have an anisotropic norm,
so every nonzero x has the two-sided inverse n(x)⁻¹ (conj x), yet they are not a division
algebra (GPR Exc. 19.29(e)). For a unital alternative algebra the two notions agree, given
1 ≠ 0 (GPR 13.8); in particular they agree for composition algebras, which are alternative. Both
facts are formalized in Conic/Division.lean and Composition/Sedenion.lean:
ConicAlgebra.isDivision_iff_hasTwoSidedInverses is GPR 13.8 for alternative conic algebras over
a field, and ConicAlgebra.hasTwoSidedInverses_realSedenion with
ConicAlgebra.not_isDivision_realSedenion is the sedenion example. Theorems stated with
HasTwoSidedInverses are about this weaker property; the classification theorems over a field are
also stated with IsDivision.
The property depends only on the multiplication, but is placed in the conic-algebra namespace because it is used here in the classification of composition algebras. It does not assert associativity.
Instances For
Multiplicativity of the norm, as a rewrite rule. ConicAlgebra.IsMultiplicative is a def,
so it has to be unfolded before rw/simp can use it.
First linearization of multiplicativity, in the left variable: n(xz, yz) = n(x, y) n(z).
Obtained from n((x + y)z) = n(x + y) n(z) by subtracting n(xz) = n(x)n(z) and
n(yz) = n(y)n(z); only right distributivity and additivity are used.
First linearization of multiplicativity, in the right variable: n(xy, xz) = n(x) n(y, z).
Not a consequence of ConicAlgebra.IsMultiplicative.polar_mul_mul_right — A is not assumed
commutative — but proved the same way, from n(x(y + z)) = n(x) n(y + z).
Second linearization of multiplicativity: n(xz, yw) + n(xw, yz) = n(x, y) n(z, w).
This is the bilinear shadow of n(xy) = n(x)n(y), and the identity from which both halves of
norm-associativity below are read off. It is
ConicAlgebra.IsMultiplicative.polar_mul_mul_right linearized in its second variable.
Norm-associativity (GPR 16.12(4), obtained from multiplicativity by GPR 17.2(a)):
n(xy, z) = n(y, (conj x)z).
The left factor may be moved across the polar form at the cost of conjugating it. Specialize the
second linearization at y = 1, where n(x, 1) = t(x), and use conj x = t(x)·1 - x.
Norm-associativity, the other side (GPR 16.12(3) and 17.2(a)):
n(xy, z) = n(x, z(conj y)).
Not a consequence of ConicAlgebra.IsMultiplicative.polar_mul_left — A is not assumed
associative or commutative — but proved the same way, specializing the second linearization at
w = 1.
In a multiplicative conic algebra, the trace of a product is
t(xy) = t(x)t(y) - n(x, y) [GPR, 16.12(5), 16.13].
The norm of a product of conjugates: n((conj x)(conj y)) = n(x) n(y).
Over a domain, a multiplicative conic algebra has anisotropic norm if and only if it has no
zero divisors. Indeed, multiplicativity sends a zero product to a zero product of norms; in the
other direction, a null vector annihilates its conjugate, which is zero only when the original
vector is zero. This is the zero-divisor step used in GPR 22.19.
Over a field, every nonzero element of a multiplicative conic algebra has a two-sided inverse
if and only if its norm is anisotropic. Forwards, a null nonzero element cannot have an inverse
because norms multiply. Backwards, n(x)⁻¹ (conj x) is a two-sided inverse. This is the step of
GPR 22.18 that GPR phrases with division algebras; HasTwoSidedInverses is weaker than that
notion in general, and agrees with it for composition algebras
(ConicAlgebra.IsComposition.isDivision_iff_hasTwoSidedInverses).
The base ring is a multiplicative conic algebra: (xy)² = x² y². Together with
QuadraticForm.isGPRNonsingular_rankOne this makes k the rank-one composition algebra.