The Involutions of $M_4(\mathbb{C})$ over $\mathbb{R}$ and over $\mathbb{C}$
Introduction
Four involutions generate the whole picture — the transpose, the entrywise conjugation, the conjugate transpose and the negative transpose — and they split into the $\mathbb{C}$-linear ones, whose fixed sets are $\operatorname{Sym}_4$ and $\operatorname{Skew}_4$, and the semilinear ones, whose fixed sets are $M_4(\mathbb{R})$, $\operatorname{Herm}_4$ and $\mathfrak{u}(4)$.
The article classifies the involutions of $M_4(\mathbb{C})$ into the two kinds, computes the fixed and anti-fixed spaces of the four basic maps, describes the inner twists $A\mapsto SA^{\mathsf T}S^{-1}$ and their two similarity classes, and shows that the quaternion conjugation $\natural$ is the transpose in the axis-free frame but not in the Peirce frame, where a correction matrix is needed.
It assumes The Matrix Algebra $M_4(\mathbb{C})$ and $M_4(\mathbb{C})$ Pauli Representation for the two frames. The general theory is The Involutions of a Matrix Algebra; the case $n=2$ is The Involutions of $M_2(\mathbb{C})$ over $\mathbb{R}$ and over $\mathbb{C}$. The two halves and their inertias are The Trace Form and Its Inertia on the Two Halves of $M_4(\mathbb{C})$.
The Four Basic Maps
Definition. On $M_4(\mathbb{C})$ consider
$$ \mathsf T(A)=A^{\mathsf T},\qquad \mathsf C(A)=\bar A,\qquad \ast(A)=A^{\ast}=\bar A^{\mathsf T},\qquad -\mathsf T(A)=-A^{\mathsf T}. $$
Proposition. Each of the four is an involution, that is a bijection of order two. The maps $\mathsf T$ and $-\mathsf T$ are $\mathbb{C}$-linear antiautomorphisms; $\mathsf C$ is a semilinear automorphism; $\ast$ is a semilinear antiautomorphism.
Proof. Transposition reverses products, $(AB)^{\mathsf T}=B^{\mathsf T}A^{\mathsf T}$, and conjugating the entries also reverses them, $\overline{AB}=\bar A\bar B$; the two commute, so $\ast=\mathsf T\circ\mathsf C=\mathsf C\circ\mathsf T$. Each map is its own inverse, and linearity is read from the formulas.
Terminology. An involution of the first kind is $\mathbb{C}$-linear; one of the second kind is semilinear. So $\mathsf T$ and $-\mathsf T$ are of the first kind, $\mathsf C$ and $\ast$ of the second.
The two identities that organize the rest. First, $\mathsf T$ and $\mathsf C$ commute, so $\ast$ is their composite in either order. Second, a first-kind involution is an antiautomorphism, so its fixed set is closed under the anticommutator and under the commutator but not under the product; this is why $\operatorname{Sym}_4$ is a Jordan algebra and $\operatorname{Skew}_4$ a Lie algebra rather than associative algebras.
The Fixed Sets of the Four Maps
Theorem. The fixed and anti-fixed spaces are as follows.
| Map | Kind | Fixed space | $\dim_{\mathbb{R}}$ | Anti-fixed | $\dim_{\mathbb{R}}$ |
|---|---|---|---|---|---|
| $\mathsf T$ | first, anti | $\operatorname{Sym}_4$ | $10$ | $\operatorname{Skew}_4$ | $6$ |
| $-\mathsf T$ | first, anti | $\operatorname{Skew}_4$ | $6$ | $\operatorname{Sym}_4$ | $10$ |
| $\mathsf C$ | second, auto | $M_4(\mathbb{R})$ | $16$ | $iM_4(\mathbb{R})$ | $16$ |
| $\ast$ | second, anti | $\operatorname{Herm}_4$ | $16$ | $\mathfrak{u}(4)$ | $16$ |
Proof. $A=A^{\mathsf T}$ is the definition of symmetry, with $\binom{4+1}{2}=10$ free entries, and $A=-A^{\mathsf T}$ leaves $\binom{4}{2}=6$. The fixed set of $\mathsf C$ is the real matrices, of real dimension $16$, and the anti-fixed set is $i$ times it. For the dagger, $A=\bar A^{\mathsf T}$ means $A_{ij}=\bar A_{ji}$, which is $16$ real conditions on $32$ real parameters, leaving $16$; the anti-fixed set $A=-\bar A^{\mathsf T}$ is the same count.
Corollary. In every row the two dimensions sum to $32$.
Proof. An involution of a complex vector space diagonalises with eigenvalues $\pm1$, and both eigenspaces are real subspaces.
The two splitting statements. The first kind splits the algebra into two complex subspaces of complex dimensions $10$ and $6$; the second kind splits it into two real subspaces of real dimension $16$ each. The complex decomposition $M_4=\operatorname{Herm}_4\oplus\mathfrak{u}(4)$ is the one that makes $\mathrm{GL}_4\simeq U(4)$, and the complex decomposition $M_4=\operatorname{Sym}_4\oplus\operatorname{Skew}_4$ is the one that makes the trace form diagonal.
The First Kind and the Inner Twists
Definition. For invertible $S$ the inner twist of the transpose is $\varphi_S(A)=SA^{\mathsf T}S^{-1}$.
Proposition. $\varphi_S$ is an involution exactly when $S^{\mathsf T}=\varepsilon S$ with $\varepsilon=\pm1$.
Proof. Compute $\varphi_S^{2}(A)=SS^{-{\mathsf T}}AS^{\mathsf T}S^{-1}$; this equals $A$ for all $A$ exactly when $S^{\mathsf T}$ is a scalar multiple of $S$, and rescaling forces the scalar to be $\pm1$.
Proposition. For $S$ symmetric the fixed space is $S^{-1}\operatorname{Sym}_4$, of dimension $10$; for $S$ antisymmetric the fixed space is $S^{-1}\operatorname{Skew}_4$, of dimension $6$.
Proof. $\varphi_S(A)=A$ reads $SA^{\mathsf T}=(SA)^{\mathsf T}$, that is $SA$ symmetric when $S$ is symmetric, and antisymmetric when $S$ is antisymmetric; multiplication by $S^{-1}$ is a linear isomorphism.
Corollary (two similarity classes). Over $\mathbb{C}$ every $\mathbb{C}$-linear involution is a twist of $\mathsf T$ by some $S$, and up to conjugation by an invertible matrix there are exactly two: the orthogonal type, with fixed algebra of dimension $6$ isomorphic to $\mathfrak{so}(4)$, and the symplectic type, with fixed algebra of dimension $10$.
Proof. Conjugating $\varphi_S$ by $P$ gives $\varphi_{PSP^{\mathsf T}}$, so the class of the involution is the congruence class of $S$; over $\mathbb{C}$ all nondegenerate symmetric forms are congruent, and all nondegenerate alternating forms are congruent.
The reading through the halves. The two classes are the two halves read as fixed algebras: the orthogonal type returns $\operatorname{Skew}_4$, the symplectic type returns a conjugate of $\operatorname{Sym}_4$. The decomposition of the algebra into the two halves is therefore the same fact as the existence of two similarity classes of first-kind involutions.
The Second Kind and the Real Forms
Definition. For invertible $S$ the twist of the dagger is $\psi_S(A)=SA^{\ast}S^{-1}$; it is an involution exactly when $S^{\ast}=\varepsilon S$.
Proof. $\psi_S^{2}(A)=SS^{-{\ast}}AS^{\ast}S^{-1}$, which is $A$ for all $A$ exactly when $S^{\ast}$ is a scalar multiple of $S$; rescaling gives $\pm1$.
Proposition. With $S=\mathrm I_4$ the involution $\ast$ fixes $\operatorname{Herm}_4$ and anti-fixes $\mathfrak{u}(4)$, both of real dimension $16$.
Proof. This is the fourth row of the table.
Proposition. The automorphism $\mathsf C$ fixes $M_4(\mathbb{R})$, a real form: a real subalgebra of real dimension $16$ whose complexification is the algebra.
Proof. The real matrices are closed under products, and $\mathbb{C}\otimes_{\mathbb{R}}M_4(\mathbb{R})\cong M_4(\mathbb{C})$ by the real and imaginary parts of the entries.
The two real forms. A semilinear automorphism of order two fixes a real form, and the complex algebra has two of them: $M_4(\mathbb{R})$ and the quaternionic form $M_2(\mathbb{H})$, of real dimension $16$. The first is the fixed set of the entrywise conjugation, the second of a conjugation composed with a suitable twist. The classification is The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$.
Faithfulness of the picture. The first kind gives complex subspaces and the second kind real subspaces. Reading a dimension without its kind is the commonest error at this order: $\operatorname{Herm}_4$ has complex dimension $16$ only if one forgets that it is not a complex subspace at all.
The Transpose in the Two Frames
In the axis-free frame the transpose is the conjugation. With $L^{\mathrm{axf}}$ the left family of the coefficient basis,
$$ L^{\mathrm{axf}}(\tilde Q)^{\mathsf T}=L^{\mathrm{axf}}(\tilde Q^{\natural}),\qquad \tilde Q^{\natural}=Q_te_t-Q_x\vec e_x-Q_y\vec e_y-Q_z\vec e_z . $$
Proof. Both sides are signed permutations of the same four columns; the identity is checked on the four units, where transposition fixes $L^{\mathrm{axf}}_{e_t}=\mathrm I_4$ and exchanges the entries of the three vector matrices, which is exactly the sign change of $\natural$.
In the Peirce frame it is not. There the transpose needs a correction matrix on each side,
$$ L(\tilde Q)^{\mathsf T}=C_LL(\tilde Q^{\natural})C_L,\qquad C_L=\mathrm I_2\otimes\sigma_2, $$
and the same formula holds on the right with $C_R=\sigma_2\otimes\mathrm I_2$.
Proof. The Peirce frame is a conjugate of the axis-free one, $L=C_\Pi^{-1}L^{\mathrm{axf}}C_\Pi$, and transposition turns a conjugation into a congruence: $(C_\Pi^{-1}MC_\Pi)^{\mathsf T}=C_\Pi^{\mathsf T}M^{\mathsf T}C_\Pi^{-{\mathsf T}}$. The matrix $C_\Pi^{\mathsf T}C_\Pi^{-{\mathsf T}}$ reduces to the two copies of $\sigma_2$ displayed.
What this means. The identity "the transpose is the quaternion conjugation" is a property of the frame and not of the algebra: it holds in the axis-free basis, where the frame vectors are the four units, and fails in the Peirce basis, where they are built from a root. The adjoint, by contrast, is the conjugate transpose in both frames, because the frames are orthogonal with equal norms and the adjoint is frame-independent up to that common scale.
The moral for the halves. $\operatorname{Sym}_4$ and $\operatorname{Skew}_4$ are defined by the transpose and hence by the frame. In the axis-free basis they are the two index patterns, the ten symmetric slots and the six antisymmetric ones; in the Peirce basis the same two spaces are described by the two correction matrices. The spaces are frame-independent; only their description is not.
Summary
The four maps $\mathsf T$, $\mathsf C$, $\ast$ and $-\mathsf T$ are involutions, and the first two commute so that $\ast$ is their composite. The first kind, $\mathbb{C}$-linear, fixes $\operatorname{Sym}_4$ of dimension $10$ or $\operatorname{Skew}_4$ of dimension $6$; the second kind, semilinear, fixes $M_4(\mathbb{R})$, $\operatorname{Herm}_4$ or $\mathfrak{u}(4)$, each of real dimension $16$.
The inner twists $SA^{\mathsf T}S^{-1}$ are involutions exactly when $S$ is symmetric or anti-Hermitian, and up to conjugation there are only the two classes, the orthogonal and the symplectic, which are the two halves. The twists of the dagger behave the same way with $S^{\ast}=\pm S$, and the semilinear automorphisms fix the two real forms.
The quaternion conjugation $\natural$ is the transpose in the axis-free frame, exactly and with no correction, and is the transpose composed with two copies of $\sigma_2$ in the Peirce frame. The difference is a change of frame, not of the algebra.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\mathsf T$, $\mathsf C$, $\ast$ | transpose, entrywise conjugation, conjugate transpose |
| $\natural$ | the quaternion conjugation of $\mathbb{B}$ |
| $\operatorname{Sym}_4$, $\operatorname{Skew}_4$ | the fixed and anti-fixed spaces of $\mathsf T$ |
| $\operatorname{Herm}_4$, $\mathfrak{u}(4)$ | the fixed and anti-fixed spaces of $\ast$ |
| $\varphi_S$, $\psi_S$ | the inner twists of $\mathsf T$ and of $\ast$ |
| $C_L$, $C_R$ | $\mathrm I_2\otimes\sigma_2$ and $\sigma_2\otimes\mathrm I_2$, the Peirce dictionaries |
| $L^{\mathrm{axf}}$, $L$ | the left families of the coefficient and Peirce frames |
Further Reading
- Max-Albert Knus, Alexander Merkurjev, Markus Rost and Jean-Pierre Tignol, The Book of Involutions (AMS, 1998), for the classification of involutions of the first and second kind.
- Involutions of a Central Simple Algebra, for the general classification this article instantiates.
- The Involution and the Hermitian Structure of $M_4(\mathbb{C})$, for the dagger, the Hermitian form and the unitary group.
- The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$, for the two real forms and their conjugations.
- $M_4(\mathbb{C})$ Pauli Representation, for the Peirce frame and the matrices $C_L$ and $C_R$.
- The Involutions of $M_2(\mathbb{C})$ over $\mathbb{R}$ and over $\mathbb{C}$, for the same four maps at order two.