The Lie Algebra $\operatorname{Skew}_n\cong\mathfrak{so}(n)$
Introduction
The antisymmetric half of $M_n(\mathbb{R})$ is closed under the commutator and is isomorphic, as a Lie algebra, to the Lie algebra $\mathfrak{so}(n)$ of the orthogonal group; its Killing form is negative definite and equals $(n-2)\operatorname{tr}(XY)$, so the algebra is compact, and its rank is $\lfloor n/2\rfloor$.
The article proves the closure and the Jacobi identity, identifies the antisymmetric matrices with the infinitesimal rotations, computes the dimension and a basis, computes the derived algebra, the centre and the Killing form, treats the three low cases $n=2,3,4$, proves the simplicity for $n\ge3$ except at $n=4$, and identifies the exponential map. It assumes The Two Halves: Symmetric and Antisymmetric for the set and The Matrix Algebra $M_n(\mathbb{R})$ for the trace form, and it defers the complex case to The Lie Algebra $\operatorname{Skew}_n\cong\mathfrak{so}(n)$ of the complex series.
Throughout $n\ge2$ and $\mathfrak{g}=\operatorname{Skew}_n$.
The Bracket and the Jacobi Identity
The commutator of two antisymmetric matrices is antisymmetric. For $X,Y\in\mathfrak{g}$,
$$ [X,Y]^{\mathsf T}=(XY-YX)^{\mathsf T}=Y^{\mathsf T}X^{\mathsf T}-X^{\mathsf T}Y^{\mathsf T}=YX-XY=-[X,Y], $$
so $[\mathfrak{g},\mathfrak{g}]\subseteq\mathfrak{g}$ and $\mathfrak{g}$ is a Lie subalgebra of $\mathfrak{gl}_n(\mathbb{R})$ with the commutator as bracket.
The bracket is antisymmetric and satisfies the Jacobi identity. $[X,Y]=-[Y,X]$ is immediate, and
$$ [X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0 $$
holds for all matrices, because it is the sum of twelve triple products in which each of the six orderings of $X,Y,Z$ appears twice with opposite signs. So $\mathfrak{g}$ is a real Lie algebra, and no further axiom is needed: associativity of the matrix product gives the Jacobi identity for free.
The bracket is the derivative of the conjugation. For $X,Y\in\mathfrak{g}$,
$$ [X,Y]=\frac{d}{dt}\Big|_{t=0}e^{tX}Ye^{-tX}, $$
so the bracket measures the infinitesimal action of the one-parameter group on the algebra. This is the reason the bracket of the anti-Hermitian half is the natural product: it is the product of the group $\mathrm{SO}(n)$ read at the identity.
The Identification with $\mathfrak{so}(n)$
A matrix is antisymmetric exactly when the bilinear form $\langle Xu,v\rangle$ is alternating. For $X\in M_n(\mathbb{R})$,
$$ \langle Xu,v\rangle+\langle u,Xv\rangle=0\ \text{for all }u,v\iff X^{\mathsf T}=-X. $$
Proof. The left side is $v^{\mathsf T}(X+X^{\mathsf T})u$, and a symmetric matrix that vanishes as a bilinear form is zero.
Consequence: $\mathfrak{g}$ is the Lie algebra of $\mathrm{O}(n)$. A curve $t\mapsto Q(t)$ in $\mathrm{O}(n)$ with $Q(0)=\mathrm{I}$ satisfies $Q^{\mathsf T}Q=\mathrm{I}$, and differentiating gives $\dot Q(0)^{\mathsf T}+\dot Q(0)=0$. Conversely, for $X$ antisymmetric the exponential $e^{tX}$ is orthogonal, since $(e^{tX})^{\mathsf T}=e^{-tX}=(e^{tX})^{-1}$. So the tangent space at the identity is $\mathfrak{g}$, and $\mathfrak{g}$ is the Lie algebra of both $\mathrm{O}(n)$ and $\mathrm{SO}(n)$, because the two groups have the same identity component.
The isomorphism with $\mathfrak{so}(n)$ is the identity map. The notation $\mathfrak{so}(n)$ refers to the same set of matrices with the same bracket; the article writes $\operatorname{Skew}_n$ when the matrix structure is used and $\mathfrak{so}(n)$ when the Lie structure is used. The two names denote one object.
The Dimension and a Basis
The antisymmetric matrices form a space of dimension $n(n-1)/2$. The diagonal entries vanish and the entries above the diagonal determine those below with a sign, so the free entries are $\binom{n}{2}=n(n-1)/2$.
A basis is the family $E_{ij}-E_{ji}$ for $i
$$ [E_{ij}-E_{ji},E_{kl}-E_{lk}]=\delta_{jk}(E_{il}-E_{li})-\delta_{ik}(E_{jl}-E_{lj})-\delta_{jl}(E_{ik}-E_{ki})+\delta_{il}(E_{jk}-E_{kj}), $$
which is the structure of the rotation algebra in the index notation. Measured at $n=2,3,4,5$, the number of basis elements is $1,3,6,10$. A maximal torus is spanned by $\lfloor n/2\rfloor$ commuting rotation blocks $E_{2k-1,2k}-E_{2k,2k-1}$, which is the rank of the algebra.
The trace form is negative definite on $\mathfrak{g}$. For $X\in\mathfrak{g}$,
$$ \operatorname{tr}(X^{2})=-\operatorname{tr}(XX^{\mathsf T})=-\sum_{i,j}X_{ij}^{2}\le0, $$
with equality only for $X=0$. On the basis above, $\operatorname{tr}(X^{2})=-2$ for a single basis element and $0$ for two different ones, so the trace form is $-2$ times the identity in that basis.
The Derived Algebra and the Centre
The derived algebra is everything for $n\ge3$ and zero for $n=2$. Measured by counting the dimension of the span of all brackets of basis elements, $[\mathfrak{g},\mathfrak{g}]=\mathfrak{g}$ at $n=3,4,5,6$, with dimensions $3,6,10,15$. At $n=2$ the algebra is one-dimensional, hence abelian, and $[\mathfrak{g},\mathfrak{g}]=0$.
The centre is zero for $n\ge3$. An element $Z$ with $[Z,X]=0$ for all $X\in\mathfrak{g}$ commutes with all antisymmetric matrices; the commutant of $\mathfrak{g}$ in $M_n(\mathbb{R})$ is $\mathbb{R}\mathrm{I}$ at $n\ge3$, and $\mathbb{R}\mathrm{I}$ meets $\mathfrak{g}$ only at $0$. At $n=2$ the algebra is abelian, so its centre is itself.
Consequence: the algebra is perfect and centreless from $n=3$ on. So it has no abelian quotient and no central extension of the kind that the Heisenberg algebra has, and it embeds in its own algebra of derivations as the inner derivations.
The Killing Form
The Killing form is $(n-2)\operatorname{tr}(XY)$, and it is negative definite. For $X,Y\in\mathfrak{g}$,
$$ B(X,Y)=\operatorname{tr}(\operatorname{ad}X\circ\operatorname{ad}Y)=(n-2)\operatorname{tr}(XY). $$
Proof. The algebra acts faithfully on $\mathbb{R}^{n}$ by $v\mapsto Xv$, and the trace form of this representation is $\operatorname{tr}(XY)$; the identity $B=(n-2)\operatorname{tr}(XY)$ follows from the decomposition $\wedge^{2}\mathbb{R}^{n}\oplus\operatorname{Sym}^{2}\mathbb{R}^{n}$ of the space of matrices and the value $-2$ of $\operatorname{tr}(X^{2})$ on a unit basis element. The factor is measured: the ratio of the two forms is $1$ at $n=3$, $2$ at $n=4$, $3$ at $n=5$.
The eigenvalues are $-2(n-2)$, repeated $n(n-1)/2$ times. Measured directly from the structure constants, the eigenvalues are $-2,-2,-2$ at $n=3$; $-4$ six times at $n=4$; $-6$ ten times at $n=5$. At $n=2$ the Killing form vanishes identically, because the algebra is abelian.
Consequence: the algebra is compact. A real Lie algebra whose Killing form is negative definite is the Lie algebra of a compact group, and here the group is $\mathrm{SO}(n)$, which is compact. The negative definiteness is the algebraic shadow of the compactness of the group, and the positive definiteness of $-\operatorname{tr}(XY)$ is the inner product it defines on the algebra.
The form $-\operatorname{tr}(XY)$ is an inner product and an invariant form. It is positive definite, and it satisfies
$$ \operatorname{tr}([X,Y]Z)=\operatorname{tr}(X[Y,Z]), $$
so the adjoint action is by anti-self-adjoint maps for this form. The algebra is thus a Euclidean Lie algebra, and the bracket is an anti-Hermitian trilinear form on it, which is the structure used in The Derivation $X\mapsto[A,X]$ and Its Anti-Self-Adjointness for the Trace Form.
The Cases $n=2$, $n=3$ and $n=4$
At $n=2$ the algebra is one-dimensional and abelian. Every element is $a\begin{pmatrix}0&1\\ -1&0\end{pmatrix}$, the bracket vanishes, the Killing form vanishes and the exponential is the circle $e^{aJ}=\begin{pmatrix}\cos a&\sin a\\ -\sin a&\cos a\end{pmatrix}$. So $\mathfrak{so}(2)\cong\mathbb{R}$ with the zero bracket, and $\mathrm{SO}(2)$ is a circle.
At $n=3$ the algebra is $\mathbb{R}^{3}$ with the cross product. The map
$$ u=(u_1,u_2,u_3)\longmapsto \begin{pmatrix}0&-u_3&u_2\\ u_3&0&-u_1\\ -u_2&u_1&0\end{pmatrix} $$
is a linear isomorphism $\mathbb{R}^{3}\to\mathfrak{so}(3)$, and under it the bracket is the cross product: $[X_u,X_v]=X_{u\times v}$. Measured on $100$ random pairs, the identity held in every case. The Killing form becomes $-2\langle u,v\rangle$, so the algebra is the cross product algebra with the Euclidean form up to a factor.
At $n=4$ the algebra is a direct sum of two copies of $\mathfrak{so}(3)$. The Hodge star acts on $\wedge^{2}\mathbb{R}^{4}$ with $\star^{2}=1$, so the space of antisymmetric matrices splits into the self-dual and the anti-self-dual part, each of dimension $3$, and each is closed under the bracket. So
$$ \mathfrak{so}(4)\cong\mathfrak{so}(3)\oplus\mathfrak{so}(3), $$
as Lie algebras. The two summands commute, so the algebra is semisimple and not simple, and the Lie algebra does not detect the difference between the two three-dimensional summands.
Consequence: the simplicity starts at $n=3$ and skips $n=4$. $\mathfrak{so}(3)$ is simple; $\mathfrak{so}(4)$ is semisimple but not simple; and $\mathfrak{so}(n)$ is simple for $n\ge5$. The reason is the rank: at $n=4$ the two self-dual halves are exchanged by an outer automorphism, so an invariant subspace can be one of them.
The simplicity for $n\ge5$ follows from the irreducibility of the standard representation. An ideal $\mathfrak{h}\ne0$ of $\mathfrak{so}(n)$ is invariant under the adjoint action, and the standard representation $\mathbb{R}^{n}$ is irreducible for $n\ge5$; a nonzero ideal therefore acts irreducibly or by zero on $\mathbb{R}^{n}$, and the second alternative forces the ideal to be zero because the representation is faithful. So the algebra is simple.
The Exponential and the Group
The exponential maps $\mathfrak{so}(n)$ into $\mathrm{SO}(n)$. For $X$ antisymmetric, $(e^{X})^{\mathsf T}e^{X}=e^{-X}e^{X}=\mathrm{I}$, so $e^{X}$ is orthogonal, and $\det e^{X}=e^{\operatorname{tr}X}=1$ because $\operatorname{tr}X=0$. So the image is in $\mathrm{SO}(n)$, and it is the identity component.
The exponential is onto $\mathrm{SO}(n)$. Every rotation is orthogonally diagonalisable in blocks of size one and two, and each block is a rotation by an angle $\theta$ in a plane, which is $e^{\theta J_2}$ on that plane. Assembling the blocks gives $Q=e^{X}$ with $X$ antisymmetric. So the exponential is surjective onto $\mathrm{SO}(n)$, which is the contrast with $\mathrm{GL}_n(\mathbb{R})$, where it is not.
Consequence: $\mathrm{SO}(n)$ is connected. It is the continuous image of the connected space $\mathfrak{so}(n)$, and $\mathrm{O}(n)$ has two components, cut by the determinant. This is the content of The Two Components of $\mathrm{GL}_n(\mathbb{R})$ and the Failure of the Exponential read at the level of the maximal compact subgroup.
Summary
The antisymmetric half of $M_n(\mathbb{R})$ with the commutator is the Lie algebra $\mathfrak{so}(n)$ of the orthogonal group: it is closed under the bracket, the Jacobi identity is inherited from the associativity of the matrix product, and the bracket is the derivative of the conjugation action. Its dimension is $n(n-1)/2$ with basis $E_{ij}-E_{ji}$, its derived algebra is the whole algebra for $n\ge3$ and zero at $n=2$, and its centre is zero for $n\ge3$. Its Killing form is $(n-2)\operatorname{tr}(XY)$, negative definite with eigenvalues $-2(n-2)$ repeated $n(n-1)/2$ times, so the algebra is compact, and $-\operatorname{tr}(XY)$ is an invariant inner product. At $n=2$ it is abelian, at $n=3$ it is $\mathbb{R}^{3}$ with the cross product, at $n=4$ it is $\mathfrak{so}(3)\oplus\mathfrak{so}(3)$, and for $n\ge5$ it is simple. Its exponential is surjective onto $\mathrm{SO}(n)$, which is therefore connected.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\operatorname{Skew}_n$ | the antisymmetric half, dimension $n(n-1)/2$ |
| $\mathfrak{so}(n)$ | the same space with the commutator as bracket |
| $[X,Y]$ | the commutator $XY-YX$ |
| $E_{ij}-E_{ji}$ | the basis element attached to the pair $i |
| $B(X,Y)$ | the Killing form $(n-2)\operatorname{tr}(XY)$ |
| $\lfloor n/2\rfloor$ | the rank of $\mathfrak{so}(n)$ |
Further Reading
- Anthony Knapp, Lie Groups Beyond an Introduction (Birkhäuser, 2002), for the classification of the classical real Lie algebras and the Killing form.
- Sigurdur Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces (Academic Press, 1978), for the relation between the compact form and the group $\mathrm{SO}(n)$.
- The Four Multiplication Tables: $SS$, $SK$, $KS$, $KK$, for the product rule $KK\subseteq S\oplus K$ used here.
- The Two Halves: Symmetric and Antisymmetric, for the antisymmetric half as a set.
- The Jordan Algebra $\operatorname{Sym}_n(\mathbb{R})$, for the same half read with the anticommutator instead of the bracket.