Euler's Four-Square Identity¶

A bilinear four-component composition law whose output represents the product of two sums of four squares as another sum of four squares.

Version: v2 · 2026-09-06 · History
Domain-specific #: 1795
Origin domain: algebra and number theory
Subdomain: composition of quadratic forms
Aliases: Euler Four Square Identity

Core Idea¶

Euler's Four-Square Identity is an explicit bilinear rule showing that the product of two values represented as sums of four squares is again represented as a sum of four squares. Let

\[ Q(x_0,x_1,x_2,x_3)=x_0^2+x_1^2+x_2^2+x_3^2. \]

For two quadruples $a=(a_0,a_1,a_2,a_3)$ and $b=(b_0,b_1,b_2,b_3)$, define

\[ \begin{aligned} c_0&=a_0b_0-a_1b_1-a_2b_2-a_3b_3,\\ c_1&=a_0b_1+a_1b_0+a_2b_3-a_3b_2,\\ c_2&=a_0b_2-a_1b_3+a_2b_0+a_3b_1,\\ c_3&=a_0b_3+a_1b_2-a_2b_1+a_3b_0. \end{aligned} \]

Then

\[ Q(a)Q(b)=Q(c). \]

Because each $c_i$ is bilinear—linear in $a$ for fixed $b$ and linear in $b$ for fixed $a$—the identity is more than an existential assertion. It is a constructive composition algorithm: two witnesses of four-square representability are transformed into a witness for their product.^[1]^[2]

The historical name is well anchored but does not define the mathematical boundary. Lemmermeyer's study of the Euler–Goldbach correspondence records Euler stating the four-square product formula in his letter of May 4, 1748.^[3] Later quaternion language exposed the formula's structural source. A draft must therefore retain the explicit bilinear law even when discussing its history; “an identity associated with Euler” is not a sufficient recognition test.

The formula is a polynomial identity with integer coefficients. Direct expansion verifies it in $\mathbb Z[a_0,\ldots,a_3,b_0,\ldots,b_3]$, so substitution makes it valid for elements of any commutative ring. Over the real numbers it is also the coordinate expression of multiplicativity of the quaternion norm. If

\[ \alpha=a_0+a_1i+a_2j+a_3k,\qquad \beta=b_0+b_1i+b_2j+b_3k, \]

with $i^2=j^2=k^2=ijk=-1$, then the coordinates of the Hamilton product $\alpha\beta$ are precisely $c_0,c_1,c_2,c_3$ in the displayed convention. Quaternion conjugation gives

\[ N(\alpha)=\alpha\bar\alpha=Q(a),\qquad N(\alpha\beta)=N(\alpha)N(\beta), \]

which is exactly the four-square identity after coordinates are restored.^[2]

The identity therefore joins three stable mathematical roles: a universal polynomial equality, a bilinear composition of four-component representations, and a norm-multiplication law for quaternions. In number theory it proves that the set of values represented by $x_0^2+x_1^2+x_2^2+x_3^2$ is closed under multiplication. That closure supplies the multiplicative step used in four-square arguments, but it does not by itself prove that every positive integer is represented.^[4]

This is an autonomous domain-specific abstraction. Generic Closure names its broad structural effect, but does not specify the four-square quadratic form or its bilinear witness map. Factorization can organize a reduction from primes to composite integers, but it does not produce the product representation. Quaternion multiplication explains the formula without making the coordinate identity an alias for the quaternion algebra as a whole.

Structural Signature¶

A well-formed instance contains the following roles:

Two input quadruples. The inputs $a,b\in R^4$ are defined over one commutative coefficient ring $R$.
The four-square quadratic form. Both inputs are evaluated by the same form $Q(x)=\sum_{r=0}^3x_r^2$.
A four-component bilinear recombination. The output coordinates $c_i$ use the displayed signed pairings. Each term is a product of one $a$-coordinate and one $b$-coordinate.
The multiplicative invariant. The output satisfies $Q(c)=Q(a)Q(b)$ identically, not approximately or only for selected numerical inputs.
A witness-producing interpretation. If $m=Q(a)$ and $n=Q(b)$, then $c$ explicitly witnesses $mn=Q(c)$.
A convention boundary. Equivalent sign, coordinate, conjugation, or operand-order conventions may produce a different output quadruple while preserving the norm identity. The chosen formula must be declared before coordinates are compared.

The operational chain is

\[ (a,Q(a))+(b,Q(b)) \longrightarrow c=B(a,b) \longrightarrow Q(c)=Q(a)Q(b), \]

where $B:R^4\times R^4\to R^4$ is the specific bilinear map above. The identity is recognized by the whole pair $(Q,B)$, not merely by the phrase “a product remains representable.”

The quaternion derivation provides a compressed proof. Quaternion conjugation reverses products, so

\[ N(\alpha\beta) =(\alpha\beta)\overline{(\alpha\beta)} =\alpha\beta\bar\beta\bar\alpha =\alpha N(\beta)\bar\alpha =N(\beta)N(\alpha). \]

The norm values are central scalars, so the last product is $N(\alpha)N(\beta)$. Expanding $\alpha\beta$ recovers the four $c_i$.

What It Is Not¶

Not Lagrange's Four-Square Theorem. Lagrange's theorem says every nonnegative integer is a sum of four integer squares. Euler's identity says how to multiply two already known representations. The latter supplies closure, not universal existence.
Not merely the statement that products exist. The identity gives four exact bilinear output polynomials. An argument establishing representability without those or an equivalent composition law is weaker and is not this named identity.
Not quaternion algebra in full. Quaternions include addition, noncommutative multiplication, conjugation, inverses, rotations, orders, and much more. The four-square identity is the norm-multiplicative coordinate shadow of one operation.
Not the two-square identity. The Brahmagupta–Fibonacci identity composes $x_0^2+x_1^2$ through complex-number multiplication. It embeds as the special case with the last two coordinates zero but does not cover the genuinely four-component formula.
Not Degen's eight-square identity. The eight-square law corresponds to octonionic composition and has eight bilinear outputs. Dimension and associativity behavior differ.
Not Pfister's general power-of-two theorem. Pfister's rational-function identities relax the polynomial-bilinear requirement and can introduce denominators. They belong to a broader sums-of-squares theory, not to Euler's fixed bilinear four-square formula.^[5]
Not unique factorization. The identity neither factors integers nor asserts unique four-square representations. Different input representations and sign conventions may yield different output representations.
Not the parallelogram law. The parallelogram law is an additive norm identity characterizing inner-product norms; Euler's identity is multiplicative and uses a bilinear product on two quadruples.

Scope of Application¶

The home domains are algebra, elementary number theory, quadratic forms, and quaternion theory. Over $\mathbb Z$, the identity composes integral representations. Over $\mathbb Q$, $\mathbb R$, or another commutative ring, it remains a polynomial equality. Over $\mathbb R$, it expresses the Euclidean squared norm on $\mathbb H\cong\mathbb R^4$ as a multiplicative quadratic form.

In elementary number theory, let

\[ S_4=\{x_0^2+x_1^2+x_2^2+x_3^2:x_i\in\mathbb Z\}. \]

The identity proves $m,n\in S_4\Rightarrow mn\in S_4$, and $1=1^2+0^2+0^2+0^2$ supplies a multiplicative identity. Thus $S_4$ is a multiplicative monoid. Before Lagrange's theorem is invoked, this is substantive closure information; after the theorem identifies $S_4$ with the nonnegative integers, the witness-level formula remains useful.^[4]

In quaternion algebra, the identity is the coordinate form of reduced-norm multiplicativity for Hamilton's quaternions. It helps connect quadratic forms with noncommutative multiplication and with arithmetic orders. Voight presents the four-square formula as implicit quaternion multiplication, then develops the Hurwitz order and a quaternionic proof of Lagrange's theorem.^[2]^[6]

In composition-of-quadratic-forms theory, Euler's law is the $n=4$ member of the bilinear sums-of-squares family. Hurwitz's theorem sharply limits equal-size bilinear identities to dimensions $1,2,4,8$ over fields of characteristic not two; characteristic two requires separate qualification.^[1] This makes the four-dimensional construction a special structural possibility, not an arbitrary coordinate trick.

The scope does not include any equality that happens to contain four squared terms. The same quadratic form, bilinear output roles, and multiplicative value relation must be present.

Clarity¶

A quick diagnostic asks:

Are there two four-component inputs evaluated by the sum-of-four-squares form?
Are four output components given bilinearly in the two inputs?
Does the output's four-square value equal the product of the input values as a universal identity?
Is the claim about composition of representations rather than universal representability of integers?

If only the last scalar equality is stated with no output rule, the witness-producing center is missing. If the output functions contain denominators or higher-degree terms, the construction may belong to Pfister-type sums-of-squares identities rather than Euler's bilinear law. If eight coordinates are used, the likely neighbor is Degen's eight-square identity. If the statement is $Q(x+y)+Q(x-y)=2Q(x)+2Q(y)$, it is the parallelogram law.

Signs are conventional but not arbitrary. Switching the convention for $i,j,k$, conjugating one factor, permuting coordinates coherently, or reversing operand order changes the coordinate expressions while retaining an equivalent norm composition. Comparing formulas term by term without aligning those choices can create false contradictions. The safest test is to reconstruct the declared bilinear product and verify $Q(B(a,b))=Q(a)Q(b)$.

Manages Complexity¶

A bare closure proof would show only that some four-square representation of $mn$ exists whenever representations of $m$ and $n$ exist. Euler's identity compresses that search into four fixed formulas. No new Diophantine search is required: insert the two witnesses and evaluate the bilinear map.

That compression scales to products of many represented integers. Repeated composition transports witnesses along a factorization, while quaternion associativity explains why different parenthesizations of the standard Hamilton product agree. The numeric values commute even though quaternion multiplication does not; this separates scalar multiplicativity from coordinate-level operand order.

The identity also compresses a long polynomial expansion into the conceptual equation $N(\alpha\beta)=N(\alpha)N(\beta)$. Conversely, expanding the quaternion statement makes every output coordinate explicit and shows that the number-theoretic consequence uses integral coefficients. The two views let practitioners choose the proof language appropriate to the problem without changing the identity.

Abstract Reasoning¶

The identity licenses a closure inference: for any commutative ring $R$, the value set $Q(R^4)$ is closed under multiplication. It also licenses witness propagation: given actual coordinate witnesses $a$ and $b$, the bilinear output $B(a,b)$ is a computable witness for the product.

For positive integers, prime factorization plus closure gives a conditional reduction. If every prime has a four-square representation, then any positive integer $n=\prod p_r^{e_r}$ has one by repeated application of the identity, with $1$ handling the empty product. The identity does not prove the antecedent. This distinction prevents a common overstatement that the product formula alone proves Lagrange's theorem.

The quaternion interpretation yields further checks. Swapping $a$ and $b$ need not preserve the output quadruple because $\alpha\beta$ may differ from $\beta\alpha$, but both outputs have the same norm $Q(a)Q(b)$. Conjugating the output changes signs of three coordinates without changing $Q$. These are genuine variant symmetries, not failures of the identity.

Hurwitz's dimension result also predicts a boundary. A request for an equal-size bilinear identity in three, five, or sixteen squares cannot be satisfied under the characteristic-not-two hypotheses. A rational-function or unequal-size construction may evade those hypotheses, so the impossibility conclusion must carry its exact scope.^[1]^[5]

Knowledge Transfer¶

Within mathematics, the exact transferable pattern is quadratic-form witnesses + bilinear composition → multiplicative norm → closure of represented values. It transfers from complex numbers and two squares to quaternions and four squares, and—with nonassociative qualifications—to octonions and eight squares.^[7]

The identity also transfers between coordinate algebra and structural algebra. A polynomial-expansion proof works over arbitrary commutative rings; a quaternion proof reveals why the signs and pairings fit together. In arithmetic, the same structure propagates integral representations. In quadratic-form theory, it is an example of composition. Each transfer preserves the output map and norm law rather than merely borrowing the phrase “four squares.”

Outside these mathematical substrates, only the broader Closure pattern remains: combine two valid witnesses and stay within a class. That residue is already captured by the Closure prime. Without a quadratic form and bilinear norm composition, an analogy is not an instance of Euler's identity, so the candidate does not qualify as a prime.

Examples¶

A fully recomputed product. Let

\[ a=(1,2,3,4),\qquad b=(2,-1,0,1). \]

Then $Q(a)=1+4+9+16=30$ and $Q(b)=4+1+0+1=6$. The displayed bilinear formulas give

\[ c=(0,6,0,12), \]

because

\[ \begin{aligned} c_0&=2+2-0-4=0,\\ c_1&=-1+4+3-0=6,\\ c_2&=0-2+6-4=0,\\ c_3&=1+0+3+8=12. \end{aligned} \]

Therefore $Q(c)=0^2+6^2+0^2+12^2=180=30\cdot6$.

Composing representations of 3 and 5. Take $a=(1,1,1,0)$, so $Q(a)=3$, and $b=(2,1,0,0)$, so $Q(b)=5$. The formula yields

\[ c=(1,3,2,-1),qquad 15=1^2+3^2+2^2+(-1)^2. \]

This demonstrates exactly what closure supplies: a representation of a product from two existing representations. It says nothing about whether a previously unrepresented prime must have a witness.

The two-square identity inside the four-square law. Set $a_2=a_3=b_2=b_3=0$. Then

\[ (a_0^2+a_1^2)(b_0^2+b_1^2) =(a_0b_0-a_1b_1)^2+(a_0b_1+a_1b_0)^2+0^2+0^2. \]

This is the Brahmagupta–Fibonacci identity embedded in four coordinates. The embedding shows family resemblance without collapsing the two named laws.

Structural Tensions¶

Coordinate formula vs. conceptual proof. Direct expansion proves the identity over every commutative ring but obscures its design. Quaternion norm multiplicativity explains the design elegantly but imports algebraic structure. Diagnostic: use the structural proof for explanation and the polynomial proof to justify ring-general substitution.

Scalar commutativity vs. noncommutative witnesses. $Q(a)Q(b)=Q(b)Q(a)$ in a commutative ring, yet the standard quaternion outputs for $B(a,b)$ and $B(b,a)$ can differ. Diagnostic: compare norms separately from ordered output coordinates.

Closure power vs. existence limits. The identity propagates known representations without creating the initial representations needed for all primes. Diagnostic: in a four-square theorem proof, identify the separate lemma that supplies or descends prime witnesses.

Canonical identity vs. equivalent conventions. Sign changes and coordinate symmetries generate visibly different formulas with the same norm law. Diagnostic: align multiplication, basis, conjugation, and operand order before declaring formulas unequal.

Bilinear rarity vs. relaxed formulas. Hurwitz's restriction makes dimensions $1,2,4,8$ special for equal-size bilinear composition, while Pfister-type rational identities exist under different hypotheses. Diagnostic: record whether output coordinates are bilinear polynomials, arbitrary polynomials, or rational functions.

Structural–Framed Character¶

Euler's Four-Square Identity is highly structural. Once $Q$ and the signed bilinear map are fixed, the equality is exact, mechanically verifiable, and independent of historical interpretation. Its quaternion, matrix, and polynomial presentations expose the same composition law.

The mathematical frame is nevertheless indispensable. “Square,” “commutative ring,” “bilinear,” “quadratic form,” “norm,” and “quaternion multiplication” carry exact algebraic meanings. Removing them leaves the generic idea of closure under combination, which is already a prime. The candidate is therefore a precise domain-specific abstraction rather than a substrate-independent prime or a merely framed historical topic.

Structural Core vs. Domain Accent¶

The portable core is two witnesses in a representability class → a structured combination rule → a witness for the product that preserves the defining measure. This instantiates Closure and relates to Monoid formation.

The domain accent contains nearly all of the candidate's differentiating information: four coordinates, the positive sum-of-squares form, signed bilinear pairings, quaternion norm, integral polynomial coefficients, and the number-theoretic interpretation of represented values. These are not decorative labels. They determine whether the equation holds.

After catalog stripping, a substantial residual remains: the exact map $B:R^4\times R^4\to R^4$, $Q(B(a,b))=Q(a)Q(b)$, the witness-production semantics, the quaternion coordinate equivalence, the prime-factor propagation role, and the bilinear-versus-rational family boundary. No live or accepted-overlay node entails that bundle.

Closure is the proposed minimal parent. The identity is a domain-specific constructive closure law: the carrier is the set $Q(R^4)$, the operation is multiplication in $R$, and the bilinear output proves the result remains in the carrier. The child adds an explicit witness map absent from generic Closure.

Monoid is a derived related structure. Because $Q(R^4)$ is multiplicatively closed and contains $1$, it inherits an associative identity-bearing operation from the ring. Euler's identity proves the closure component but is not itself the whole monoid structure.

Factorization explains one use: representations can be propagated through a prime factorization. The identity does not find the factorization or assert uniqueness, so Factorization is not a parent.

Invariance is related through preservation of the norm equation under coherent sign and coordinate changes, but the central claim is multiplicativity, not an unchanged quantity under a transformation group.

Composition in the catalog has a broad design-oriented meaning. Mathematical composition of quadratic forms is terminologically relevant, but generic visual/conceptual arrangement does not supply a literal parent edge.

Relationships to Other Abstractions¶

Current abstraction Euler's Four-Square Identity Domain-specific

Parents (1) — more general patterns this builds on

Euler's Four-Square Identity is a kind of Closure Prime

Closure is the proposed minimal parent.

Hierarchy path (1) — routes to 1 parentless root

Euler's Four-Square Identity → Closure

Neighborhood in Abstraction Space¶

Euler's Four-Square Identity sits in a sparse region of the domain-specific corpus (84^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

The frozen top semantic match, accepted-workspace domain_specific:sq_universal_group, is a group-theoretic property and shares only abstract algebra vocabulary. It has no four-square quadratic form or bilinear composition identity.

Accepted-workspace domain_specific:parallelogram_law is also a norm identity, but it relates $Q(x+y)$ and $Q(x-y)$ additively and characterizes inner-product norms. It does not multiply two represented values. domain_specific:euler_sequence is an unrelated namesake. domain_specific:tensor_representation can encode bilinear maps but is a representation-theoretic construction, not this identity. domain_specific:gelfand_naimark_segal_construction and domain_specific:quartic_reciprocity are unrelated operator-algebraic and number-theoretic constructs.

Live domain_specific:matrix can package the bilinear map into $4\times4$ matrices, but the matrix target does not entail the norm constraint or sum-of-squares interpretation. Live domain_specific:planarity is retrieval noise. Live prime:factorization, prime:equivalence_relation, and prime:monoid cover broad surrounding structures without the formula.

External neighbors must also remain separate: Lagrange's Four-Square Theorem is a universal-existence theorem; the two- and eight-square identities are other dimensions; Hurwitz's theorem classifies possible bilinear dimensions; quaternion algebra is the ambient algebra; and Pfister identities allow a different formula class. None is an unrestricted alias.

References¶

[1] Keith Conrad, “The Hurwitz Theorem on Sums of Squares by Linear Algebra,” University of Connecticut expository notes, especially equations 1.2 and Theorem 1.1. Official PDF. registry ↩a ↩b ↩c

[2] John Voight, Quaternion Algebras, Graduate Texts in Mathematics 288, Springer (2021), especially section 1.1 and equation 1.1.2. doi:10.1007/978-3-030-56694-4. registry ↩a ↩b ↩c

[3] Franz Lemmermeyer, “Euler, Goldbach, and ‘Fermat's Theorem’,” Elemente der Mathematik 65(4), 144–153 (2010), documenting the May 4, 1748 product formula in the correspondence. doi:10.4171/EM/154. registry ↩

[4] Keith Conrad, “Pythagorean Descent,” section 6, “The Four-Square Theorem,” University of Connecticut expository notes, especially Lemma 6.2. Official PDF. registry ↩a ↩b

[5] Keith Conrad, “Pfister's Theorem on Sums of Squares,” University of Connecticut expository notes. Official PDF. registry ↩a ↩b

[6] John Voight, “The Hurwitz Order,” chapter 11 of Quaternion Algebras, pp. 165–179, Springer (2021). Open-access publisher chapter. registry ↩

[7] John C. Baez, “The Octonions,” Bulletin of the American Mathematical Society 39, 145–205 (2002). doi:10.1090/S0273-0979-01-00934-X. registry ↩