commutative and associative algebra: , with v∙2 = v + μ1, with 4μ +1 ≠ 0, and n(∊ + δv) = ∊2 − μδ2 + ∊δ, for ∊, ;
3 3) a quaternion algebra for as in (2) and ;
4 4) a Cayley (or octonion) algebra , for as in (3) and .
In particular, the dimension of a Hurwitz algebra is restricted to 1, 2, 4 or 8.
PROOF.– The only Hurwitz algebra of dimension 1 is, up to isomorphism, the ground field. If (
If
Finally, if
Note that if char
COROLLARY 2.1.– Every Hurwitz algebra over a field
1 1) the ground field ;
2 2) a two-dimensional algebra for a non-zero scalar α;
3 3) a quaternion algebra for as in (2) and ;
4 4) a Cayley (or octonion) algebra , for as in (3) and .
REMARK 2.2.– Over the real field ℝ, the scalars α, β and γ in corollary 2.1 can be taken to be ±1. Note that [2.3] and the analogous equation for ℂ and
REMARK 2.3.– Hurwitz (1898) only considered the real case with a positive definite norm. Over the years, this was extended in several ways. The actual version of the generalized Hurwitz theorem seems to appear for the first time in Jacobson (1958) (if char
The problem of isomorphism between Hurwitz algebras of the same dimension relies on the norms:
PROPOSITION 2.3.– Two Hurwitz algebras over a field are isomorphic if and only if their norms are isometric.
PROOF.– Any isomorphism of Hurwitz algebras is, in particular, an isometry of the corresponding norms, due to the Cayley–Hamilton equation. The converse follows from Witt’s cancellation theorem (see Elman et al. (2008, theorem 8.4)). □
A natural question is whether the restriction of the dimension of a Hurwitz algebra to be 1, 2, 4 or 8 is still valid for arbitrary composition algebras. The answer is that this is the case for finite-dimensional composition algebras.
COROLLARY 2.2.– Let (
PROOF.– Let
Note that since the left and right multiplications by a norm 1 element are isometries, we still have n(x ◊ y) = n(x)n(y), so (
However, contrary to the thoughts expressed in Kaplansky (1953), there are examples of infinite-dimensional composition algebras. For example (see Urbanik and Wright (1960)), let φ : ℕ × ℕ → ℕ be a bijection (for instance, φ(n, m) = 2n−1 (2m − 1)), and let