Composition Algebra¶

A unital algebra with a nondegenerate quadratic form whose value is multiplicative under the algebra product.

Version: v1 · 2026-10-03 · History
Domain-specific #: 13074
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomains: Nonassociative Algebra, Quadratic Forms → Mathematics
Aliases: Hurwitz Algebra

Core Idea¶

A composition algebra, in the unital finite-dimensional tier used here, is an algebra $A$ over a field $K$ with bilinear multiplication and identity $1$, together with a nondegenerate quadratic form $q:A\to K$ such that

\[q(xy)=q(x)q(y)\qquad\text{for every }x,y\in A.\]

The polar bilinear form of $q$ has zero radical; this is what nondegenerate means. It does not mean that $q(x)=0$ only when $x=0$. This entry explicitly takes $\operatorname{char}K\ne2$, so its standard conjugation and doubling discussion can be stated without silently changing characteristic. Some authors also call nonunital algebras satisfying the norm identity “composition algebras”; Hall begins with that broader convention and later fixes an identity. The present entry uses the unital/Hurwitz convention throughout.^[1]

The identity has two strikingly unlike realizations over the reals. Hamilton's quaternions have a positive-definite, multiplicative four-square norm and nonzero elements are invertible. Split-complex numbers have $q(a+bL)=a^2-b^2$ with $L^2=1$; their form is nondegenerate but isotropic, and $(1+L)(1-L)=0$. Thus “composition” describes compatibility between multiplication and a quadratic form, not necessarily division, associativity, or positive norm.^[1]^[2]^[3]

Structural Signature¶

Sig role-phrases: declared base field and convention tier → bilinear unital product → nondegenerate quadratic form → multiplicative composition law → isotropy profile and derived consequences.

Base field and tier. $K$ is fixed, here of characteristic not two; $A$ is finite-dimensional and has a two-sided identity. Field changes can change which norm forms and algebra forms occur. These assumptions precede any dimension or classification claim.^[1]
Bilinear internal product. Multiplication $A\times A\to A$ is linear in each argument over $K$. Neither commutativity nor associativity is an axiom of the genus; real quaternions are noncommutative, and octonions show why associativity cannot be required universally.^[1]^[3]
Quadratic form. $q(\lambda x)=\lambda^2q(x)$, and $b_q(x,y)=q(x+y)-q(x)-q(y)$ is bilinear. Nondegeneracy means no nonzero vector is orthogonal under $b_q$ to all of $A$. It permits nonzero isotropic vectors $x$ with $q(x)=0$.^[1]
Composition identity. $q(xy)=q(x)q(y)$ for all pairs is the indispensable extra relation. An ordinary algebra may have a quadratic form without this compatibility; a quadratic space may have no internal product at all.^[1]
Isotropy and inverse diagnostic. In the unital tier, conjugation yields $x\bar x=\bar x x=q(x)1$. If $q(x)\ne0$, then $\bar x/q(x)$ is an inverse. If a nonzero $x$ has $q(x)=0$, it cannot be inverted by this formula and split examples can have zero divisors. This profile distinguishes variants rather than defining the genus.^[1]^[2]

What It Is Not¶

It is not every algebra equipped with a norm. The form must be quadratic, nondegenerate, and multiplicative for every product pair. A familiar Euclidean length placed on an arbitrary bilinear algebra does not suffice. Conversely, it is not merely a quadratic form: $A$ must also have its internal bilinear product.^[1]

It is not synonymous with normed division algebra. Over $\mathbb R$, positive-definite real composition examples $\mathbb R,\mathbb C,\mathbb H,\mathbb O$ are normed division algebras in Baez's sense. But real split-complex numbers are a composition algebra with indefinite norm and zero divisors. Therefore the live Normed Division Algebra node is a narrower or intersecting specialist neighbor, not a universal parent.^[3]^[2]

It is not necessarily associative or commutative. Quaternions break commutativity, and octonions break associativity while retaining multiplicative norm. Nor is “Cayley–Dickson algebra” a blanket synonym: doubling is a construction; continued doubling eventually produces algebras outside this composition class.^[3]^[1]

It is not a universal assertion that every field has exactly the same four algebras. The $1,2,4,8$ theorem restricts dimensions under its hypotheses. It does not say an arbitrary field has only the four positive-definite real isomorphism types. Hall separately discusses split forms and characteristic-dependent starting points.^[1]

Scope of Application¶

The identity organizes quadratic and nonassociative algebra. For real positive-definite examples it links multiplicative Euclidean norms to $\mathbb R$, $\mathbb C$, $\mathbb H$ and $\mathbb O$. The real quaternion algebra is associative but noncommutative; the octonion algebra is nonassociative. Their norm behavior is not an accident of notation but a structural constraint on multiplication.^[3]

Split forms widen the scope. Hall's two-dimensional split algebra has orthogonal idempotents and norm $q(\alpha z+\delta\bar z)=\alpha\delta$; the four-dimensional split quaternion example is $2\times2$ matrices with determinant as its multiplicative quadratic form. Split-complex numbers give a concrete real presentation. The quadratic form remains nondegenerate even where nonzero isotropic elements and zero divisors appear.^[1]^[2]

For the stated finite-dimensional nondegenerate tier, Hall's Hurwitz theorem restricts possible dimensions to $1,2,4,8$. That result is a constraint on the entire package, not on arbitrary bilinear algebras or arbitrary quadratic spaces. The draft does not classify all isomorphism forms over every field, nor does it infer a real geometric interpretation for every field.^[1]

Clarity¶

Three properties that are often compressed into the word “norm” must be separated. Nondegenerate says the polar bilinear form has no radical. Anisotropic says no nonzero vector has quadratic value zero. Positive definite is an ordered-real-field condition and implies anisotropy in the real example. Split complex $q=a^2-b^2$ is nondegenerate but not anisotropic; ordinary quaternion $q=a^2+b^2+c^2+d^2$ is positive definite.^[1]^[2]

The formula $q(xy)=q(x)q(y)$ also identifies what “composition” composes. It is values of a quadratic form under an algebra product, not function composition and not simply construction by doubling. When comparing two candidate nodes, the mathematical object is the algebra-plus-form package; a theorem about its possible dimensions is a consequence, and Cayley–Dickson is a method of building examples.^[1]^[3]

Finally, the unit and field convention are visible rather than tacit. Hall's initial definition permits nonunital composition algebras; his later results assume a unit. A curator cannot swap the definitions midway to make an inverse or classification argument work. Characteristic two likewise requires different technical handling and is outside this entry's stated working tier.^[1]

Manages Complexity¶

A bilinear multiplication table on a finite-dimensional vector space can contain many structure constants. The multiplicative quadratic form constrains all of them simultaneously. It gives a single equality to test on products and, after polarization, relations among multiplication operators and the associated bilinear form. Hall expresses left and right multiplication by a fixed element as similarities of $q$ with scale $q(x)$.^[1]

The package organizes variants without pretending they are identical. “Division” and “split” do not require separate definitions of the norm-composition law; their difference lies in whether nonzero isotropic elements occur. That permits a coherent comparison of Hamilton quaternions, split complex numbers, and split matrix quaternions while preserving decisive distinctions about inverses and zero divisors.^[1]^[2]

The compression has limits. Merely checking that a norm is multiplicative on a few basis units does not prove the identity for every pair. The dimension theorem is a strong filter but is not a sufficient test: an arbitrary four-dimensional algebra need not admit a qualifying $q$. Choice of field and norm form remains data, not a detail that the label can erase.^[1]

Abstract Reasoning¶

To recognize the structure, first specify $K$, characteristic, dimension and whether the algebra has an identity. Verify bilinearity of multiplication. Next exhibit a quadratic $q$, compute its polar form, and establish nondegeneracy. Then verify $q(xy)=q(x)q(y)$ for arbitrary $x,y$—not just on a selected example pair. Only after these conditions hold should one invoke the dimension restriction or discuss conjugation and inverses.^[1]

For the unital case, Hall's conjugation satisfies $x\bar x=\bar x x=q(x)1$. If $q(x)\ne0$, division by $q(x)$ produces an inverse. The logical converse must not be blurred into “nondegenerate means every nonzero element invertible”: split complex numbers have a nondegenerate form and $q(1+L)=0$. Their product with $1-L$ is zero.^[1]^[2]

Cayley–Dickson doubling helps explain the dimensions and changing product laws. Baez uses it for the real sequence $\mathbb R\to\mathbb C\to\mathbb H\to\mathbb O$; Hall gives an appropriate nondegenerate-subalgebra doubling proposition. But the generated 16-dimensional stage does not inherit the full qualifying composition structure. Construction is evidence only when the norm condition is still checked.^[1]^[3]

Knowledge Transfer¶

The literal transfer is mathematical: the same relation $q(xy)=q(x)q(y)$ diagnoses a positive-definite quaternion algebra and an indefinite split-complex algebra. The names, dimensions, commutativity and zero-divisor behavior change; the carrier, quadratic form and compatibility roles survive. This is why the abstraction is more general than any one familiar number system.^[1]^[2]^[3]

The norm-composition relation also connects to multiplicative sums-of-squares identities in real positive-definite coordinates: for quaternions, the four-square norm of a product equals the product of two four-square norms. That statement should not be inflated into a claim that every composition form over every field is a positive sum of squares. The broader theorem concerns quadratic forms over the declared base field.^[1]^[3]

Transfer stops at the mathematical boundary. A slogan about “composing structures” in another field is analogy unless one can specify an algebra, quadratic form, nondegeneracy and the universal multiplicativity equation. The broad live Algebra over a Ring parent travels further because it requires only a compatible bilinear product.^[1]

Examples¶

Real Hamilton quaternions. In $\mathbb H$, an element is $x=a+bi+cj+dk$ with $i^2=j^2=k^2=-1$ and $ij=k=-ji$. Its conjugate is $\bar x=a-bi-cj-dk$ and $q(x)=x\bar x=a^2+b^2+c^2+d^2$. Direct quaternion multiplication gives $q(xy)=q(x)q(y)$; since $q(x)>0$ for $x\ne0$, $x^{-1}=\bar x/q(x)$. This is an anisotropic, noncommutative, associative member of the genus.^[1]^[3]

Mapped back: base field/tier = $\mathbb R$, unital finite-dimensional, characteristic zero; product = Hamilton bilinear multiplication; form = positive four-square $q$; law = multiplicativity; isotropy profile = anisotropic division; dimension consequence = four.

Real split-complex algebra. Let $A=\mathbb R[L]/(L^2-1)$ and $q(a+bL)=a^2-b^2$. For $x=a+bL$ and $y=c+dL$, multiplication gives $xy=(ac+bd)+(ad+bc)L$; expansion yields

\[q(xy)=(ac+bd)^2-(ad+bc)^2=(a^2-b^2)(c^2-d^2)=q(x)q(y).\]

The polar form has diagonal values $2,-2$, so it is nondegenerate. Yet $q(1+L)=0$ and $(1+L)(1-L)=0$: the nonzero null elements are zero divisors. This is a split, associative, commutative member, not a normed division algebra.^[2]^[1]

Mapped back: base field/tier = $\mathbb R$, unital finite-dimensional, characteristic zero; product = $L^2=1$ bilinear multiplication; form = indefinite $a^2-b^2$; law = displayed identity; isotropy profile = null vectors and zero divisors; dimension consequence = two.

Boundary negative. A three-dimensional real quadratic space with Euclidean squared length, but without a unital internal bilinear product obeying the multiplicative law, is not a composition algebra. A qualifying dimension alone would not suffice either.^[1]

Structural Tensions¶

Norm multiplicativity versus division. Positive-definite real norm makes every nonzero quaternion invertible; split-complex norm remains multiplicative and nondegenerate yet has nonzero null elements. Requiring division for conceptual simplicity would discard genuine composition algebras; ignoring isotropy would make false inverse claims. Diagnostic: Is $q$ merely nondegenerate, or is it also anisotropic on this carrier?^[1]^[2]^[3]

Doubling reach versus invariant preservation. The Cayley–Dickson story offers a compact chain through dimensions $1,2,4,8$, but iterating a construction is not the same as preserving the nondegenerate composition identity indefinitely. Stopping the story too early hides octonions; extending it without a property check wrongly admits later stages. Diagnostic: After the proposed doubling, does the new product still make the specified nondegenerate quadratic form multiplicative?^[1]^[3]

Broad algebra membership versus diagnostic sharpness. Every object in this entry is a bilinear algebra over a field and hence fits the live broad algebra-over-a-ring genus. That parent is useful for placement, but does not predict a multiplicative quadratic form; a division-only parent is sharper but excludes split cases. Diagnostic: Which feature is inherited from the parent and which additional invariant defines this child?^[1]

Structural–Framed Character¶

Evaluative weight: the definition is an equation and nondegeneracy condition, not a judgment that one algebra is better. Human-practice dependence: mathematicians choose whether “composition algebra” includes nonunital variants and which field/characteristic tier to study; once those hypotheses are fixed, the product identity is a formal property rather than an opinion. Institutional origin: Hurwitz's theorem and the Cayley–Dickson construction are historical mathematical work, not institutional rules that bring the object into existence.^[1]^[3]

Vocabulary travel: the same identity holds for real division and split examples and is studied over other fields, but the real positive-definite classification cannot simply be exported to arbitrary $K$. Import versus recognition: calling a novel algebra a composition algebra is recognition only after its bilinear product and nondegenerate quadratic $q$ satisfy the universal equation. Reusing the phrase for an unrelated “composition” activity is metaphor. Its character: a structural mathematical object with an explicit convention tier and formal invariant, domain-specific rather than prime.

Structural Core vs. Domain Accent¶

The core is the coupling of a product to a quadratic form through $q(xy)=q(x)q(y)$. The base algebra and nondegeneracy conditions make that coupling mathematically testable. The live Algebra over a Ring parent captures the bilinear-product part; it does not supply the norm identity. This entry's additional structure is not an optional application accent but the reason it exists.^[1]

The domain accent is exact: vector spaces over a field, quadratic forms, isotropy, conjugation and Hurwitz dimension restrictions. None should be turned into an all-purpose prime merely because “preserving a measure under composition” sounds portable. The split-complex counterexample also blocks reducing the core to normed division alone.^[1]^[2]

This entry is a kind of Algebra over a Ring. A composition algebra is a bilinear algebra enriched by a multiplicative quadratic form.

Relationships to Other Abstractions¶

Current abstraction Composition Algebra Domain-specific

Parents (1) — more general patterns this builds on

Composition Algebra is a kind of Algebra over a Ring Domain-specific

A composition algebra is a bilinear algebra enriched by a multiplicative quadratic form.

Hierarchy path (1) — routes to 1 parentless root

Composition Algebra → Algebra over a Ring → Linearity

Neighborhood in Abstraction Space¶

Composition Algebra sits in a moderately populated region (49^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Categorical & Representation-Theoretic Structures (8 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Function composition: combining maps by substitution does not by itself create a bilinear algebra and multiplicative quadratic form.
A generic algebra over a ring: the live parent has bilinearity but does not demand $q(xy)=q(x)q(y)$.
A normed division algebra: positive-definite real division examples are a special branch; split-complex zero divisors disprove equivalence.
An arbitrary Cayley–Dickson iterate: the defining norm-composition condition must survive each construction stage.
A sum-of-squares formula without hypotheses: only appropriate forms and products under a declared field yield the relevant identity.^[1]^[2]^[3]

References¶

[1] J. I. Hall, Notes on Composition Algebras, original author notes, revised 19 April 2012; §§1–3 for definition, theorem and conjugation, §§6.1–6.4 for split examples and doubling. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28 ↩29 ↩30 ↩31 ↩32 ↩33 ↩34

[2] Oregon State University, The Split Complex Numbers, original open textbook §2.8, equations (2.8.1)–(2.8.5). registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l

[3] John C. Baez, The Octonions, original author PDF, §§1.1 and 2.2 for real normed division algebras and Cayley–Dickson construction. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n