Orthogonality, Isotropy and the Radical
Introduction
A form of degree two organises the space it acts on by pairing elements, and the pairing yields a second, derived structure: the relation of orthogonality, the isotropic elements that the form cannot separate, the totally isotropic subspaces, and the radical, the part of the space that the form does not see at all. This article develops that structure for the forms of Sesqualgebras with a Form.
Three facts govern the layer. Orthogonality is a relation and not an inner product: it is symmetric because the form is Hermitian, and a subspace $W$ has two orthogonal complements which coincide, $W^{\perp} = {}^{\perp}W$, and satisfy $W \subseteq W^{\perp\perp}$, with equality for every $W$ as soon as the form is non-degenerate; the radical of the restriction is the intersection $W \cap W^{\perp}$. Isotropy and the radical are different: an isotropic element has $h(x,x) = 0$ and the radical is the set of the elements orthogonal to everything, and a totally isotropic subspace need not lie in the radical — the hyperbolic plane is the standard witness. And the radical is the total orthogonal complement, $A^{\perp} = \operatorname{rad}$, so that the form descends to the quotient $A/\operatorname{rad}$, where it is non-degenerate.
This is the counterpart of the orthogonality theory of Bilinear Forms and of the inertia theory of The Indefinite Case and the Signature; the classification of the forms by their Witt index is The Hyperbolic Form and the Witt Index, and the cancellation theorems are Witt's Theorems. Throughout, $(A,*,h)$ is a sesqualgebra with a form over a base $(R,\varsigma)$; the rank statements are made over a field $k$ with an involution $\varsigma$, where the Riesz maps are isomorphisms exactly when the form is non-degenerate, and the general case is stated with the hypothesis that the form be nonsingular, that is, that the Riesz map $x \mapsto h(\cdot, x)$ be an isomorphism.
Orthogonality
Definition. Two elements $x, y$ are orthogonal when $h(x,y) = 0$; for a subset $W \subseteq A$ the right complement and the left complement are
$$ \boxed{\;W^{\perp} = \{x : h(w,x) = 0 \ \text{for all } w \in W\}, \qquad {}^{\perp}W = \{x : h(x,w) = 0 \ \text{for all } w \in W\}\;} . $$
Proposition. For a Hermitian form the two complements coincide, $W^{\perp} = {}^{\perp}W$, and orthogonality is a symmetric relation.
Proof. $h(x,w) = 0$ is equivalent to $\varsigma(h(w,x)) = 0$ and $\varsigma$ is injective, so the two sets are equal; the same computation gives symmetry.
Proposition (the radical of a restriction). For every subset $W$,
$$ \boxed{\;W \cap W^{\perp} = \operatorname{rad}(h|_{W})\;} , $$
and the restriction $h|_{W}$ is non-degenerate exactly when $W \cap W^{\perp} = 0$. For every $W$ one has $W \subseteq W^{\perp\perp}$.
Proof. $x \in W \cap W^{\perp}$ means $x \in W$ and $h(x,w) = 0$ for every $w \in W$, which is exactly $x \in \operatorname{rad}(h|_{W})$; the restriction is non-degenerate exactly when that radical vanishes. For the inclusion, if $w \in W$ and $y \in W^{\perp}$ then $h(w,y) = 0$, so every element of $W$ is orthogonal to every element of $W^{\perp}$, which is $W \subseteq W^{\perp\perp}$.
Corollary. For a non-degenerate form the complement is an involution on the subspaces, $W^{\perp\perp} = W$ for every $W$. The radical is a different object: for a totally isotropic $W$ one has $\operatorname{rad}(h|_{W}) = W$, the whole of $W$, so the restriction is as degenerate as possible while the double complement is still $W$ and carries no information about the isotropy.
Proposition (the orthogonal decomposition). Let $A$ be a vector space of finite dimension over a field and $h$ non-degenerate. A subspace $W$ is non-degenerate exactly when $A = W \oplus W^{\perp}$.
Proof. If $W$ is non-degenerate then $W \cap W^{\perp} = \operatorname{rad}(h|_{W}) = 0$. The map $W \to W^{*}$, $w \mapsto h(\cdot,w)|_{W}$, is an isomorphism in finite dimension — injective by the vanishing of the radical and an isomorphism of vector spaces of the same dimension $\dim W$ — so the restriction map $A \to W^{*}$, $x \mapsto h(\cdot,x)|_{W}$, is onto; its kernel is $W^{\perp}$ and the rank–nullity theorem gives $\dim W^{\perp} = \dim A - \dim W$. The sum $W + W^{\perp}$ is then direct and of dimension $\dim A$, hence all of $A$. Conversely, if $A = W \oplus W^{\perp}$ and $x \in W$ is orthogonal to $W$, then $x$ is orthogonal to $W + W^{\perp} = A$, so $x = 0$ by non-degeneracy.
Over a general ring the rank argument is replaced by the hypothesis that $h$ be nonsingular, and the decomposition holds as a direct summand statement.
Isotropy
Definition. An element $x$ is isotropic when $h(x,x) = 0$ and anisotropic otherwise; a subspace $W$ is totally isotropic when $h(x,y) = 0$ for all $x, y \in W$. A form with no non-zero isotropic element is anisotropic.
Proposition. Every element of the radical is isotropic, and the radical is a totally isotropic subspace. A totally isotropic subspace need not lie in the radical.
Proof. If $x \in \operatorname{rad}$ then $h(x,x) = 0$, and $h(x,y) = 0$ for all $x, y \in \operatorname{rad}$ by definition, so the radical is totally isotropic. For the failure, let $A = k^{2}$ with the form $h((a,b),(c,d)) = ad + bc$, the hyperbolic plane of the next article: it is non-degenerate, hence has zero radical, while $e_{1} = (1,0)$ is isotropic with $h(e_{1},e_{1}) = 0$ and the line $ke_{1}$ is totally isotropic.
Remark (the null set is not the radical). The set $\{x : h(x,x) = 0\}$ is the null set of the form and it is generally much larger than the radical; the two agree only in degenerate situations. The indefinite case is where the difference is visible, and it is the subject of The Indefinite Case and the Signature, where the inertia is defined through maximal definite subspaces precisely because the null set is not an invariant.
The Radical and the Quotient
Proposition. The radical is the total orthogonal complement,
$$ \boxed{\;\operatorname{rad} = A^{\perp} = {}^{\perp}A\;} , $$
and it is invariant under the isometries of the form.
Proof. The first clause is the definition of the radical as the set of the elements orthogonal to every element of $A$; the Hermitian property identifies the two sides. If $T$ is an invertible isometry — in finite dimension and for a non-degenerate form every isometry is invertible, by Isometries and the Unitary Group of a Form — and $x \in \operatorname{rad}$ then $h(Tx, Ty) = h(x,y) = 0$ for every $y$, and $T$ being onto, $h(Tx, z) = 0$ for every $z$, so $Tx \in \operatorname{rad}$.
Proposition (the descent). The form descends to the quotient $A/\operatorname{rad}$ by $h_{\mathrm{red}}(x + \operatorname{rad}, y + \operatorname{rad}) = h(x,y)$, and the descended form is non-degenerate.
Proof. Well defined: if $z \in \operatorname{rad}$ then $h(x + z, y) = h(x,y) + h(z,y) = h(x,y)$ and likewise in the second slot. Non-degenerate: if $h(x,y) = 0$ for every $y$ then $x \in \operatorname{rad}$, so the class of $x$ is zero.
The descent is the algebraic reduction of the layer, and the radical is a left ideal of the algebra by Sesqualgebras with a Form; when the form is associative the radical is a two-sided ideal and the quotient is again an algebra, which is the quotient construction of The Completion of a Sesqualgebra with a Form read without the completion.
The Witt Index
Definition. Let $h$ be a non-degenerate Hermitian form on a vector space of finite dimension over a field with an involution. The Witt index of $h$ is the largest dimension of a totally isotropic subspace, $\operatorname{ind}(h) = \max\{\dim W : W$ totally isotropic$\}$.
Proposition. Every totally isotropic subspace of a non-degenerate form is contained in a maximal one, and the Witt index is the number of hyperbolic planes in a Witt decomposition $A = H \perp A_{\mathrm{an}}$ with $H$ a sum of hyperbolic planes and $A_{\mathrm{an}}$ anisotropic.
Proof. The extension of a totally isotropic subspace to a maximal one is the finite-dimensional case of the extension theorem; the decomposition is the Witt decomposition, which the hyperbolic-plane reduction constructs pair by pair, and the number of pairs is the index. The details, with the cancellation theorem that makes the index and the anisotropic part invariants, are Witt's Theorems and The Hyperbolic Form and the Witt Index.
Corollary. Over an algebraically closed field $k$ with $\varsigma = \mathrm{id}$ the index of a non-degenerate form of dimension $n$ is $\lfloor n/2\rfloor$: the form is the sum of $\lfloor n/2\rfloor$ hyperbolic planes and, for $n$ odd, one anisotropic line. Over $\mathbb{R}$ the index is $\min(n,m)$ where $(n,m)$ is the signature, and the inertial invariant is The Indefinite Case and the Signature.
Examples
The Hyperbolic Plane
On $k^{2}$ with $h((a,b),(c,d)) = ad + bc$ the form is non-degenerate of Witt index one: the line $ke_{1}$ and the line $ke_{2}$ are totally isotropic, the radical is zero, and the matrix of the form is $\begin{pmatrix}0 & 1\\ 1 & 0\end{pmatrix}$ of determinant $-1$.
The Matrix Algebra with an Indefinite Form
On $M_{2}(\mathbb{C})$ with $h(X,Y) = \tau(XGY^{*})$, $G = \operatorname{diag}(1,-1)$, the rank-one matrix $E_{11}$ has the positive diagonal $h(E_{11},E_{11}) = 1$ and is not isotropic; the sum $E_{11} + E_{22}$ is isotropic, $h(E_{11}+E_{22}, E_{11}+E_{22}) = \tau(G) = 0$, so the null set is not empty although the radical is zero, because the form is non-degenerate. The comparison with the rank-one functional form of the layer, $h_{\ell}(X,Y) = \ell(Y^{*}X)$ with $\ell(Z) = \operatorname{tr}(ZE_{11})$, whose radical is the left ideal $\{X : XE_{11} = 0\}$ of the matrices with the first column zero, separates the two notions.
The Biquaternion Algebra
On $\mathbb{B}$ with the form $\operatorname{Sc}(\bar{\tilde Q}\tilde Q')$ of signature $(2,6)$ over $\mathbb{R}$, the totally isotropic subspaces have dimension at most $2$; the two null directions are the null cone of the Lorentzian reading, and the passage to the definite quotient is The Form on the Biquaternion Algebra as a General Plain Sesquilinear Form.
Summary
- Orthogonality is the symmetric relation $h(x,y) = 0$; the two orthogonal complements coincide for a Hermitian form and $W \subseteq W^{\perp\perp}$, with equality for every $W$ when the form is non-degenerate, while $W \cap W^{\perp} = \operatorname{rad}(h|_{W})$.
- Over a field and for a non-degenerate form, $W$ is non-degenerate exactly when $A = W \oplus W^{\perp}$.
- An element is isotropic when $h(x,x) = 0$; the radical is totally isotropic, and a totally isotropic subspace need not lie in the radical, the hyperbolic plane being the witness.
- The null set is not the radical, and the inertia must be defined through maximal definite subspaces rather than through the null set.
- The radical is the total orthogonal complement $A^{\perp}$ and is isometry-invariant; the form descends to $A/\operatorname{rad}$, where it is non-degenerate.
- The Witt index is the largest dimension of a totally isotropic subspace and the number of hyperbolic planes in a Witt decomposition.
- The bilinear originals are Bilinear Forms and Witt's Theorems; the index classification is The Hyperbolic Form and the Witt Index.
Summary of Notation
| symbol | meaning |
|---|---|
| $W^{\perp}$, ${}^{\perp}W$ | the right and the left orthogonal complement of $W$ |
| $W^{\perp\perp}$ | the double complement, containing $W$ |
| $\operatorname{rad}$ | the radical $A^{\perp} = {}^{\perp}A$ |
| $h_{\mathrm{red}}$ | the form descended to $A/\operatorname{rad}$ |
| $\operatorname{ind}(h)$ | the Witt index of a non-degenerate form |
| $A_{\mathrm{an}}$ | the anisotropic part of a Witt decomposition |
| $H$ | the hyperbolic part of a Witt decomposition |
Further Reading
- Winfried Scharlau, Quadratic and Hermitian Forms (Springer, 1985), for orthogonality, the radical, the descent and the Witt decomposition.
- T. Y. Lam, Introduction to Quadratic Forms over Fields, Graduate Studies in Mathematics 67 (American Mathematical Society, 2005), for totally isotropic subspaces and the Witt index over a field.
- Max-Albert Knus, Quadratic and Hermitian Forms over Rings, Grundlehren der mathematischen Wissenschaften 294 (Springer, 1991), for the non-singular form over a ring and the orthogonal summand statements.