The Unitary Group of a Form and Its Lie Algebra
Introduction
The unitary group $\mathrm{U}(A,h)$ of Isometries and the Unitary Group of a Form is an algebraic group, and every algebraic group has a Lie algebra: the tangent space at the identity, computed by the dual numbers. This article computes it, and the answer is the space of the anti-self-adjoint operators with the commutator:
$$ \mathfrak{u}(A,h) = \{T : T^{*} = -T\} = \operatorname{Lie}\,\mathrm{U}(A,h) . $$
The computation is algebraic and it uses no limit and no topology. The tangent space is the kernel of the map $U(R[\varepsilon]/\varepsilon^{2}) \to U(R)$, the infinitesimal isometries are the operators whose quadratic form $h(Tx,x)$ vanishes to first order, and the first-order condition is exactly the antisymmetry $T^{*} = -T$; this is the algebraic form of the classical statement that the Lie algebra of the isometry group of a form is the space of the anti-self-adjoint operators of that form. The bracket is the commutator, and it closes on the anti-self-adjoint operators by the sign rule of Self-Adjoint and Anti-Self-Adjoint Operators of a Form.
The corresponding structure on the other eigenspace of the involution is the Jordan algebra of The Hermitian Jordan Algebra of a Form; the bilinear original is The Orthogonal Lie Algebra, the completed reading is The Lie Algebra and the Exponential Map and Unitary and Isometric Operators of the Form of Part II, and the exponential map that joins the algebra to the group is The Lie Algebra and the Exponential Map in Part II. Throughout, $(A,*,h)$ is a sesqualgebra with a form over a base $(R,\varsigma)$ with $h$ nonsingular, $T^{*}$ is the adjoint of $T$ for $h$, and $R[\varepsilon] = R[\varepsilon]/\varepsilon^{2}$ is the ring of the dual numbers over the base.
The Lie Algebra by the Dual Numbers
Definition. The Lie algebra of the unitary group is the set of the operators $T$ with
$$ 1 + \varepsilon T \in U(A\otimes_{R}R[\varepsilon], h) , $$
that is, the set of the first-order deformations of the identity that remain isometries over the dual numbers.
Proposition. The Lie algebra is the space of the anti-self-adjoint operators:
$$ \boxed{\;\mathfrak{u}(A,h) = \{T : T^{*} = -T\}\;} . $$
Proof. Compute over the dual numbers with $\varepsilon^{2} = 0$:
$$ h\bigl((1+\varepsilon T)x, (1+\varepsilon T)y\bigr) = h(x,y) + \varepsilon\bigl(h(Tx,y) + h(x,Ty)\bigr) . $$
The first-order condition for the isometry is therefore $h(Tx,y) + h(x,Ty) = 0$ for all $x, y$. Now $h(x,Ty) = h(T^{*}x, y)$ by the Hermitian property and the definition of the adjoint, so the condition reads $h\bigl((T + T^{*})x, y\bigr) = 0$ for all $x,y$, which the nonsingularity of $h$ converts into $T + T^{*} = 0$, that is $T^{*} = -T$.
Corollary. The Lie algebra is an $R^{\varsigma}$-submodule of the operators, and it is the kernel of the restriction of the isometry equation to the first order; it is cut out by the linear equations $T^{*} + T = 0$, so it is a module of the same nature as the operators and needs no completion.
Corollary (the quadratic reading). When $2$ is invertible and the diagonal determines the Hermitian form, the condition is equivalent to the vanishing of the trace part of the quadratic form, $h(Tx,x) + \varsigma(h(Tx,x)) = 0$ for every $x$; over the trivial base involution this is the vanishing of $h(Tx,x)$ itself, and then the infinitesimal isometries are exactly the operators whose quadratic form vanishes identically.
Proof. If $T^{*} = -T$ then $h(Tx,x) = -h(x,Tx) = -\varsigma(h(Tx,x))$, so $h(Tx,x) + \varsigma(h(Tx,x)) = 0$; over the trivial involution the fixed ring is the whole ring and $h(Tx,x) = 0$ as well. Conversely, the diagonal of the Hermitian form $B(x,y) = h(Tx,y) + h(x,Ty)$ is $h(Tx,x) + \varsigma(h(Tx,x))$, which vanishes under the hypothesis; a Hermitian form with vanishing diagonal vanishes when the diagonal determines the forms, so $B = 0$, which is the antisymmetry $T^{*} = -T$. The last clause is the case $\varsigma = \mathrm{id}$.
The Bracket
Proposition. The space $\mathfrak{u}(A,h)$ closes under the commutator,
$$ \boxed{\;S, T \in \mathfrak{u}(A,h) \implies [S,T] = ST - TS \in \mathfrak{u}(A,h)\;} , $$
and the commutator makes it a Lie algebra over the fixed ring $R^{\varsigma}$.
Proof. The sign rule of Self-Adjoint and Anti-Self-Adjoint Operators of a Form gives $(ST - TS)^{*} = -\varepsilon\delta(ST-TS)$ with $\varepsilon = \delta = -1$, hence $(ST-TS)^{*} = -(ST-TS)$, which is the closure. The bilinearity, the antisymmetry and the Jacobi identity of the commutator are the associativity of the product of the operators.
Corollary. The Lie algebra of the unitary group is the anti-Hermitian part of the algebra of the operators, the self-adjoint part is its complement under the involution of The Form-Adjoint of an Operator, and the product of two anti-self-adjoint operators splits as
$$ \boxed{\;ST = \tfrac{1}{2}[S,T] + \tfrac{1}{2}(ST + TS)\;} , $$
with the commutator inside the Lie algebra and the anticommutator leaving it; the anticommutator is the Jordan structure of The Hermitian Jordan Algebra of a Form.
The Algebraic Nature of the Construction
Proposition. The tangent space is computed with the dual numbers and not with a limit: the map $U(R[\varepsilon]) \to U(R)$ is the reduction $\varepsilon \mapsto 0$, and its kernel is the Lie algebra. In particular the construction is defined over an arbitrary commutative base with an involution, with no ordered field, no norm and no completion.
Proof. The group over the dual numbers is the set of the operators $1 + \varepsilon T$ satisfying the isometry equation with the Gram matrix read over $R[\varepsilon]$; reducing modulo $\varepsilon$ gives the identity, so every element of the kernel is of the shape $1 + \varepsilon T$, and the first-order expansion of the previous section is the computation of the kernel. The reduction map is a homomorphism of groups by the functoriality of the construction of the group.
Remark (the exponential). Over $\mathbb{R}$ or $\mathbb{C}$ the exponential of an anti-self-adjoint operator is unitary, and the differential of the exponential at $0$ is the identity, so the Lie algebra is the tangent space of the group in the analytic sense as well; the series is analytic, and the exponential and its image are The Lie Algebra and the Exponential Map in Part II. In the algebraic layer the Cayley transform of Self-Adjoint and Anti-Self-Adjoint Operators of a Form plays the role of the exponential without the series.
Examples
The Unitary Group
On $M_n(\mathbb{C})$ with the trace form the Lie algebra is the space of the anti-Hermitian matrices, $\mathfrak{u}(n) = \{T : T^{\dagger} = -T\}$, of real dimension $n^{2}$; its traceless part is $\mathfrak{su}(n)$ of dimension $n^{2}-1$, the Lie algebra of the special unitary group.
The Orthogonal Group
At the trivial base involution and $* = \mathrm{id}$ the Lie algebra is the space of the matrices with $T^{\mathrm{t}}G + GT = 0$, of dimension $n(n-1)/2$, the orthogonal Lie algebra $\mathfrak{so}(G)$ of The Orthogonal Lie Algebra; its structure and its root system are the subject of that article.
The Symplectic Group
For an alternating form with Gram matrix $J$, the Lie algebra is $\{T : T^{\mathrm{t}}J + JT = 0\}$, the symplectic Lie algebra $\mathfrak{sp}(2n)$, of dimension $n(2n+1)$; the form is alternating and the algebra is the one of the classical groups of the layer.
The Biquaternion Algebra
On $\mathbb{B}$ with the dagger the Lie algebra of the unitary slice is the space of the anti-Hermitian biquaternions, and the Lie algebra of the Lorentzian form $\operatorname{Sc}(\bar{\tilde Q}\tilde Q')$ is $\mathfrak{so}(2,6)$; the computations are those of The Form on the Biquaternion Algebra as a General Plain Sesquilinear Form and The Unitary Slice and the Compact Real Form with Hermitian Adjoint.
Summary
- The Lie algebra of the unitary group is the space of the first-order isometries, computed by the dual numbers: $\mathfrak{u}(A,h) = \{T : T^{*} = -T\}$, the anti-self-adjoint operators.
- Equivalently, when $2$ is invertible and the diagonal determines the form, it is the set of the operators whose quadratic form $h(Tx,x)$ has vanishing trace part; over the trivial involution this is $h(Tx,x) = 0$ for every $x$.
- It closes under the commutator by the sign rule, and it is a Lie algebra over the fixed ring $R^{\varsigma}$.
- The construction is algebraic: the kernel of $U(R[\varepsilon]) \to U(R)$, with no limit, no norm and no completion; the analytic exponential is Part II and the Cayley transform is the algebraic substitute.
- The examples are $\mathfrak{u}(n)$, $\mathfrak{su}(n)$, $\mathfrak{so}(G)$, $\mathfrak{sp}(2n)$ and the Lorentzian $\mathfrak{so}(2,6)$ of the biquaternions.
- The complementary structure is the Jordan algebra of the self-adjoint operators, The Hermitian Jordan Algebra of a Form.
Summary of Notation
| symbol | meaning |
|---|---|
| $R[\varepsilon] = R[\varepsilon]/\varepsilon^{2}$ | the ring of the dual numbers over the base |
| $\mathfrak{u}(A,h)$ | the Lie algebra of the anti-self-adjoint operators |
| $[S,T] = ST - TS$ | the commutator of the operators |
| $\mathfrak{u}(n)$, $\mathfrak{su}(n)$ | the unitary and special unitary Lie algebras |
| $\mathfrak{so}(G)$, $\mathfrak{sp}(2n)$ | the orthogonal and the symplectic Lie algebras |
Further Reading
- Jean Dieudonné, La géométrie des groupes classiques, Ergebnisse der Mathematik 5 (Springer, 1955), for the classical Lie algebras as the anti-self-adjoint operators of a form.
- Nathan Jacobson, Lie Algebras (Interscience, 1962; Dover reprint, 1979), for the Lie algebra of an algebraic group and the structure of the classical algebras.
- Max-Albert Knus, Quadratic and Hermitian Forms over Rings, Grundlehren der mathematischen Wissenschaften 294 (Springer, 1991), for the Lie algebra of the unitary group of a form over a ring with an involution.