---
title: Quadratic Orthogonal Pairs in Algebra & Geometry
url: https://www.emergentmind.com/topics/quadratic-orthogonal-pairs
type: topic
---

# Quadratic Orthogonal Pairs in Algebra & Geometry

“Quadratic orthogonal pairs” is not a single universally fixed term across the cited literature. In the theory of Azumaya algebras over schemes, the central object is a quadratic pair on an algebra with orthogonal involution, namely a semi-trace compatible with the reduced trace [2209.07107]. In complex projective geometry, orthogonal pairs are pairs of holomorphic maps preserving orthogonality for possibly degenerate Hermitian forms [2109.09340]. In hyperelliptic function theory, a quadratic orthogonal pair is a pair of functions $(Y_0,\tilde Y_0)$ satisfying $\tilde Y_0Y_0^*=-1$, introduced to resolve a continued-fraction “mismatch” problem and to construct bilateral Hankel-determinant solutions of Somos recurrences [2603.11670]. A further scheme-theoretic development studies canonical quadratic pairs on Clifford algebras attached to Azumaya algebras with quadratic pair [2309.03077].

## 1. Terminological scope and ambient structures

The phrase occurs in several adjacent but distinct settings.

| Context | Basic object | Defining condition |
|---|---|---|
| Azumaya algebras over a scheme | $(A,\sigma,f)$ with $\sigma$ orthogonal | $f(a+\sigma(a))=\operatorname{Trd}_A(a)$ |
| Projective spaces with Hermitian forms | holomorphic pair $(U,V)$ | $H(x,y)=0 \Rightarrow H'(U(x),V(y))=0$ |
| Hyperelliptic function fields | nonzero pair $(Y_0,\tilde Y_0)$ | $\tilde Y_0\cdot Y_0^*=-1$ |

In the Azumaya-algebra setting, Gille–Neher–Ruether investigate quadratic pairs for Azumaya algebras with involutions over a base scheme $S$, generalizing the case of quadratic pairs on central simple algebras over a field due to Knus, Merkurjev, Rost, and Tignol [2209.07107]. In the hyperelliptic setting, the terminology “orthogonal” refers to the analogue of a bilinear pairing $(f,g)=fg^*$, for which a quadratic orthogonal pair satisfies $(Y_0,\tilde Y_0)=-1$ [2603.11670]. In projective geometry, Gao–Ng define orthogonal pairs between projective spaces equipped with possibly degenerate Hermitian forms and connect them to Segre maps between real hyperquadrics [2109.09340].

This suggests that the terminology is context-sensitive. What remains common is the presence of an involutive or Hermitian notion of orthogonality together with an additional compatibility condition that is linear, semi-tracial, or multiplicative.

## 2. Quadratic pairs on Azumaya algebras with orthogonal involution

Let $S$ be a scheme with structure sheaf $\mathcal O_S$. An $\mathcal O_S$-algebra $A$ is Azumaya of degree $n$ if $A$ is finite locally free of rank $n^2$ over $\mathcal O_S$ and the sandwich map
$$
A\otimes_{\mathcal O}A^{op}\to \operatorname{End}_{\mathcal O}(A),\qquad a\otimes b\mapsto (x\mapsto axb)
$$
is an isomorphism. An involution of the first kind on $A$ is an anti-automorphism $\sigma:A\to A$ with $\sigma^2=\operatorname{Id}_A$. Étale-locally on $S$, one has
$$
(A,\sigma)\simeq (\operatorname{End}_{\mathcal O}(M),\operatorname{ad}_b)
$$
for some locally free $M$ and regular bilinear form $b$ on $M$; $\sigma$ is called orthogonal if $b$ is symmetric, weakly symplectic if $b$ is skew, and symplectic if in addition $b$ is alternating [2209.07107].

For an orthogonal involution, the standard submodules are
$$
\operatorname{Sym}(A,\sigma)=\ker(\sigma-\operatorname{Id}),\qquad
\operatorname{Alt}(A,\sigma)=\operatorname{Im}(\sigma-\operatorname{Id}),
$$
together with
$$
\operatorname{Symd}(A,\sigma)=\operatorname{Im}(\sigma+\operatorname{Id}),\qquad
\operatorname{Skew}(A,\sigma)=\ker(\sigma+\operatorname{Id}).
$$
A quadratic pair, or quadratic triple, on $(A,\sigma)$ is a linear form
$$
f:\operatorname{Sym}(A,\sigma)\to \mathcal O_S
$$
such that for every local section $a\in A(T)$ one has
$$
f(a+\sigma(a))=\operatorname{Trd}_A(a).
$$
This is the semi-trace property. When $2$ is invertible in $\mathcal O_S$, the assignment
$$
f(s)=\tfrac12\operatorname{Trd}_A(s)
$$
on symmetric elements automatically satisfies the defining condition, and hence there is a unique quadratic pair [2209.07107].

The same framework underlies the Clifford-algebra construction over schemes. Ruether formulates an algebra with quadratic pair as a triple $(\mathcal A,\sigma,f)$, where $\sigma$ is orthogonal and $f$ is a semi-trace on $\operatorname{Sym}(\mathcal A,\sigma)$ [2309.03077]. In this sense, “quadratic orthogonal pair” in the scheme-theoretic literature refers not to a pair of maps or functions, but to an orthogonal involution together with a compatible semi-trace.

A classification statement is available when $\sigma$ is locally quadratic, meaning $1_A\in \operatorname{Symd}(A,\sigma)(S)$. Then the set of all quadratic pairs
$$
\{\,f:\operatorname{Sym}(A,\sigma)\to \mathcal O_S \mid (A,\sigma,f)\ \text{quadratic}\,\}
$$
is in bijection with the fibre of $\sigma+\operatorname{Id}$ over $1_A$,
$$
\{\,\ell\in (A/\operatorname{Alt}(A,\sigma))(S)\mid \ell+\sigma(\ell)=1_A\,\},
$$
so the existence and multiplicity of quadratic pairs are controlled by a quotient modulo alternating elements [2209.07107].

## 3. Cohomological obstructions, affine vanishing, and non-affine counterexamples

The obstruction theory of Gille–Neher–Ruether is formulated from the exact rows and columns involving $\operatorname{Skew}(A,\sigma)$, $\operatorname{Symd}(A,\sigma)$, $\operatorname{Alt}(A,\sigma)$, and $A/\operatorname{Alt}(A,\sigma)$. Passing to fppf or étale cohomology yields connecting maps
$$
A(S)\xrightarrow{\sigma+\operatorname{Id}} \operatorname{Symd}(A,\sigma)(S)\xrightarrow{\delta} H^1\!\bigl(S,\operatorname{Skew}(A,\sigma)\bigr)
$$
and
$$
\delta':\operatorname{Symd}(A,\sigma)(S)\to H^1\!\bigl(S,\operatorname{Skew}(A,\sigma)/\operatorname{Alt}(A,\sigma)\bigr).
$$
Since $1_A\in \operatorname{Symd}(A,\sigma)(S)$, one defines
$$
o(A,\sigma):=\delta(1_A)\in H^1(S,\operatorname{Skew}(A,\sigma)),
$$
the strong obstruction, and
$$
w(A,\sigma):=\delta'(1_A)\in H^1(S,\operatorname{Skew}(A,\sigma)/\operatorname{Alt}(A,\sigma)),
$$
the weak obstruction [2209.07107].

The existence theorem separates two levels of extension. If $(A,\sigma)$ is Azumaya with $\sigma$ locally quadratic, then there exists $f$ on all of $A$ extending a quadratic pair if and only if $o(A,\sigma)=0$. There exists $f$ defined only on $\operatorname{Sym}(A,\sigma)$ making $(A,\sigma,f)$ a quadratic pair if and only if $w(A,\sigma)=0$. Constructively, choosing local sections $\ell_i$ with $\ell_i+\sigma(\ell_i)=1$ on a cover $\{U_i\}$ produces local semi-traces
$$
f_i(s)=\operatorname{Trd}_A(\ell_i\cdot s),\qquad s\in \operatorname{Sym}(A,\sigma)(U_i),
$$
and the Čech $1$-cocycle $\{\ell_j-\ell_i\}$ represents the obstruction class [2209.07107].

On affine schemes the obstruction vanishes. If $U=\operatorname{Spec}R$ is affine, then any quasi-coherent $\mathcal O$-module $M$ satisfies $H^1_{et}(U,M)=0$. In particular, $\operatorname{Skew}(A,\sigma)$ and $\operatorname{Skew}(A,\sigma)/\operatorname{Alt}(A,\sigma)$ have vanishing $H^1$, so
$$
\operatorname{Symd}(A,\sigma)(U)=(\sigma+\operatorname{Id})(A(U))
$$
and $\delta(1_A)=0$. Consequently, on any affine open $U\subset S$, every locally quadratic involution extends to a quadratic pair, and in fact there is an $R$-linear form $f:A(U)\to R$ with the semi-trace property [2209.07107].

The non-affine case is genuinely different. In characteristic $2$, Gille–Neher–Ruether construct a nontrivial strong obstruction using an ordinary elliptic curve $E$ over a field $k$ of $\operatorname{char}2$, the $2$-multiplication torsor $[2]:E\to E$, and a twisted quaternion Azumaya algebra $Q$ of degree $2$ with canonical orthogonal involution $\sigma$. Here $Q(E)=k$ only, and Lemma 6.4 gives $o(Q,\sigma)\neq 0$. Tensoring $n$ copies of $(Q,\sigma)$ yields degree $2^n$ examples with nonzero strong obstruction. They also construct a nontrivial weak obstruction in characteristic $2$ from a Galois torsor $Y\to S$ with group $PGL_2(\mathbb F_4)$ over a connected smooth projective $k$-variety $S$; a Čech-cocycle check shows that no global $f$ on $\operatorname{Sym}(Q,\sigma)$ can satisfy the semi-trace property, so $w(Q,\sigma)\neq 0$ [2209.07107].

These results sharply separate local and global behavior: the affine case is unobstructed, whereas non-affine schemes may carry precisely the cohomological obstructions that prevent the existence of quadratic pairs.

## 4. Canonical quadratic pairs on Clifford algebras

Ruether studies the Clifford algebra attached to an Azumaya algebra with quadratic pair over an arbitrary base scheme $S$ [2309.03077]. For an algebra with quadratic pair $(\mathcal A,\sigma,f)$ of even degree $2n$, one forms the tensor algebra $T(\mathcal A)$ and two two-sided ideals
$$
J_1(\sigma,f),\qquad J_2(\sigma,f),
$$
where $J_1(\sigma,f)$ is generated by $s-f(s)$ for $s\in \operatorname{Sym}(\mathcal A,\sigma)$ and $J_2(\sigma,f)$ is generated by elements $u-D_u(\Delta_f)$ for $u\in \mathcal A\otimes \mathcal A$ satisfying $(\operatorname{Id}\otimes \sigma)u=u$. The Clifford algebra is then
$$
\operatorname{Cl}(\mathcal A,\sigma,f)=T(\mathcal A)/(J_1(\sigma,f)+J_2(\sigma,f)),
$$
with canonical map $c:\mathcal A\to \operatorname{Cl}(\mathcal A,\sigma,f)$ and canonical involution
$$
\gamma\bigl(c(a)\bigr)=c\bigl(\sigma(a)\bigr).
$$

The orthogonality of $\gamma$ is exceptional. Writing $\deg(\mathcal A)=2n$, the canonical involution on the Clifford algebra is orthogonal precisely in the following two situations: $n\equiv 0\pmod 4$, or $n\equiv 2\pmod 4$ and $2=0$ in the structure sheaf. Under these hypotheses, and assuming $\deg(\mathcal A)\ge 8$, there is a unique semi-trace
$$
f_{\operatorname{Cl}}:\operatorname{Sym}\bigl(\operatorname{Cl}(\mathcal A,\sigma,f),\gamma\bigr)\to \mathcal O
$$
such that the canonical homomorphism
$$
\operatorname{PGO}(\mathcal A,\sigma,f)\to \operatorname{Aut}\bigl(\operatorname{Cl}(\mathcal A,\sigma,f),\gamma\bigr)
$$
factors through
$$
\operatorname{Aut}\bigl(\operatorname{Cl}(\mathcal A,\sigma,f),\gamma,f_{\operatorname{Cl}}\bigr).
$$
Explicitly,
$$
f_{\operatorname{Cl}}(s)=\operatorname{Trd}_{\operatorname{Cl}}\!\bigl(p(1)\cdot s\bigr),
$$
where $p(1)$ is the distinguished class in $\operatorname{Cl}(\mathcal A,\sigma,f)/\operatorname{Alt}(\operatorname{Cl},\gamma)$ with $(\operatorname{Id}+\gamma)(p(1))=1$ [2309.03077].

The construction is functorial for isomorphisms of quadratic pairs and stable under base-change. Over a field of characteristic $2$, if $\mathcal A$ admits an element $x$ with $\operatorname{Trd}_{\mathcal A}(x)=1$, then $p(1)=\overline{c(x)}$ and
$$
f_{\operatorname{Cl}}(s)=\operatorname{Trd}_{\operatorname{Cl}}(c(x)s),
$$
recovering the construction of Dolphin–Quéguiner-Mathieu [2309.03077].

Degree $4$ is an exceptional failure mode. When $\deg(\mathcal A)=4$ and $2=0$ in $\mathcal O$, the involution $\gamma$ is orthogonal, but there is no semi-trace
$$
f':\operatorname{Sym}(\operatorname{Cl}(\mathcal A,\sigma,f),\gamma)\to \mathcal O
$$
for which the natural map
$$
\operatorname{PGO}(\mathcal A,\sigma,f)\to \operatorname{Aut}\bigl(\operatorname{Cl}(\mathcal A,\sigma,f),\gamma\bigr)
$$
factors through $\operatorname{Aut}(\operatorname{Cl},\gamma,f')$. The paper states this equivalently as the nonexistence of a canonical semi-trace in degree $4$ [2309.03077].

## 5. Orthogonal pairs of holomorphic maps and Segre rigidity

Gao–Ng define a local orthogonal pair for projective spaces equipped with possibly degenerate Hermitian forms [2109.09340]. Let
$$
V\simeq \mathbb C^{n+1},\qquad W\simeq \mathbb C^{N+1}
$$
carry Hermitian forms $H$ and $H'$ of signatures $(r,s,t)$ and $(r',s',t')$, with projectivizations
$$
P(V)=P^{r,s,t},\qquad P(W)=P^{r',s',t'}.
$$
A local orthogonal pair is a pair of holomorphic maps
$$
U:\Omega\subseteq P(V)\to P(W),\qquad V:\Omega\subseteq P(V)\to P(W),
$$
such that for any lifts $x,y\in V$ of points in $\Omega$,
$$
H(x,y)=0 \quad\Rightarrow\quad H'(U(x),V(y))=0.
$$
Equivalently,
$$
U([x])\perp_{H'}V([y])\qquad\text{whenever}\qquad [x]\perp_H [y].
$$

The basic dichotomy is between degenerate and non-degenerate orthogonal pairs. The pair $(U,V)$ is degenerate if one branch, say $U$, has image contained in a proper linear subspace of $P(W)$ of dimension strictly less than $\dim P(V)$; if neither branch lies in a proper linear subspace, the pair is non-degenerate. Gao–Ng prove that every degenerate local orthogonal pair is in fact null:
$$
U([x])\perp_{H'}V([y])\qquad\text{for all }[x],[y]\in \Omega.
$$
Thus degeneracy forces global orthogonality rather than merely orthogonality along incident pairs [2109.09340].

The main rigidity theorem imposes the codimension restriction
$$
r'+s'\le 2\cdot (r+s+t-1)-1,
$$
equivalently $N\le 2n-2$. Under this hypothesis, every local orthogonal pair
$$
(U,V):P^{r,s,t}\supset \Omega\to P^{r',s',t'}
$$
is either null or quasi-standard. Here quasi-standard means that, after possibly projecting onto a non-degenerate subspace in the target and splitting off a null summand, the pair is induced by a single complex-linear isometry up to an overall constant factor $\lambda\neq 0$. For a linear orthogonal pair there exist linear $T,S$ with
$$
S^*H'T=\lambda H,\qquad \lambda\in \mathbb C^*,
$$
and
$$
U([x])=[Tx],\qquad V([x])=[Sx].
$$
The proof uses a hyperplane-restriction lemma, projection away from null directions, and reduction until one branch becomes a linear embedding [2109.09340].

This framework is directly related to Segre maps of real hyperquadrics. The Heisenberg hypersurface
$$
H_n=\{(z,w)\in \mathbb C^n\times \mathbb C:\operatorname{Im}w=|z|^2\}
$$
has complexification identified with the null cone in $P^{1,n}$. By Lemma 4.1 of Gao–Ng, a holomorphic Segre map is exactly an orthogonal pair, and Zhang’s rigidity theorem becomes the special case $t=t'=0$, $r=1$, $s=n$, $r'=1$, $s'=N$ with $N\le 2n-2$: every local Segre map $H_n\to H_N$ is either null or quasi-standard [2109.09340].

## 6. Hyperelliptic quadratic orthogonal pairs, continued fractions, and Hankel determinants

In the hyperelliptic-function setting, a quadratic orthogonal pair is defined in a function field with involution [2603.11670]. Let $C$ be a hyperelliptic curve of genus $g$ over $\mathbb C$ given by
$$
Y^2=A(X)^2+4R(X),
$$
where
$$
A(X)=X^{g+1}+k_{g-1}X^{g-1}+\cdots +k_0,\qquad
R(X)=u(X^g+l_{g-1}X^{g-1}+\cdots +l_0),
$$
and the coefficients are generic so that the right-hand side has $2g+2$ distinct zeros. In the function field
$$
F=\mathbb C(X)[Y]/(Y^2-(A^2+4R)),
$$
the nontrivial Galois involution is $*:F\to F$, given by $X\mapsto X$ and $Y\mapsto -Y$. A pair of nonzero functions
$$
Y_0,\tilde Y_0\in F
$$
is called a quadratic orthogonal pair if
$$
\tilde Y_0(X)\cdot Y_0^*(X)=-1.
$$
Equivalently,
$$
\tilde Y_0=-1/(Y_0^*).
$$

The notion is introduced to resolve the “mismatch” problem that remained unsolved in Hone’s work on continued-fraction expansions and Hankel determinants from hyperelliptic curves. There, the positive and negative halves of a Somos-4 or Somos-5 sequence produce two independent families of Hankel determinants, and the difficulty is to glue them into a single bi-infinite solution. The single algebraic relation
$$
\tilde Y_0Y_0^*=-1
$$
forces the positive and negative continued-fraction expansions to be compatible and hence yields a coherent bi-infinite Hankel-determinant solution [2603.11670].

Existence and uniqueness are immediate: for any $Y_0\in F$ with $Y_0^*\neq 0$, there is a unique partner
$$
\tilde Y_0:=-1/(Y_0^*).
$$
If $Y_0$ admits a nonterminating continued-fraction expansion as a J-fraction in $\mathbb C((X^{-1}))$, then so does $\tilde Y_0$, and their partial quotients satisfy
$$
\tilde a_n=a_{n-1}\qquad\text{for all }n.
$$
The paper also gives a constructive realization from polynomial recurrences. Choosing initial polynomials $P_0(X),Q_0(X),Q_{-1}(X)$ of degrees $\le g$ so that
$$
P_0^2+Q_0Q_{-1}=Y^2,
$$
one defines sequences $\{P_n\},\{Q_n\},\{a_n\}$ for all $n\in \mathbb Z$ by
$$
P_{n+1}=a_nQ_n-P_n,\qquad
Q_{n+1}=Q_{n-1}+2a_nP_n-a_n^2Q_n,
$$
with
$$
a_n=\text{coefficient of }X^g\text{ in }Q_n^{-1}\cdot 2X^g.
$$
Then
$$
P_n^2+Q_nQ_{n-1}=Y^2
$$
for all $n$, and
$$
Y_n=\frac{Y+P_n}{Q_n},\qquad
\tilde Y_n=\frac{Y+P_n}{Q_{n-1}}
$$
form a quadratic orthogonal pair for each $n$:
$$
\tilde Y_n\cdot Y_n^*=-1.
$$

The associated generating functions
$$
G^{(n)}(X)=Y_n^{-1}=\sum_{k\ge 1}s_{k-1}^{(n)}X^{-k},\qquad
\tilde G^{(n)}(X)=-(\tilde Y_n)^{-1}=\sum_{k\ge 1}\tilde s_{k-1}^{(n)}X^{-k}
$$
define moment Hankel determinants
$$
\Delta_k^{(n)}=\det[s_{i+j-2}^{(n)}]_{1\le i,j\le k},\qquad
\tilde\Delta_k^{(n)}=\det[\tilde s_{i+j-2}^{(n)}]_{1\le i,j\le k},
$$
and the tau-sequence
$$
\tau_m^{(n)}=
\begin{cases}
\Delta_m^{(n)}, & m\ge 1,\\
1, & m=0,\\
\tilde\Delta_{-m}^{(n)}, & m\le -1.
\end{cases}
$$
In genus $1$, the theorem states that $\tau_m^{(n)}$ satisfies the bilinear Somos-4 recurrence
$$
\tau_{m+2}\tau_{m-2}=\alpha\,\tau_{m+1}\tau_{m-1}+\beta\,\tau_m^2,\qquad \forall\, m\in \mathbb Z,
$$
with
$$
\alpha=u^2,\qquad \beta=u^2(v^2+f).
$$
For Somos-5, if
$$
\tau_{n+5}\tau_n=\alpha\,\tau_{n+4}\tau_{n+1}+\beta\,\tau_{n+3}\tau_{n+2},
$$
then the even and odd subsequences each satisfy a Somos-4 relation with
$$
\tilde\alpha=\beta^2,\qquad
\tilde\beta=\alpha(2\beta^2+\alpha\beta J+\alpha^3),
$$
where $J$ is the conserved quantity of the bilinear form. In this way, the QOP condition is the mechanism that reconstructs a bi-infinite sequence from its two half-sequences [2603.11670].

Source: https://www.emergentmind.com/topics/quadratic-orthogonal-pairs