The Derivations and the Automorphisms of $M_4(\mathbb{C})$
Introduction
A derivation of $M_4(\mathbb{C})$ is a linear map $D$ with the Leibniz rule
$$ D(AB)=D(A)B+AD(B). $$
An automorphism is an invertible linear map preserving the product. For the full matrix algebra both structures are as small as they can be: every derivation is inner, $D=\operatorname{ad}_X$ for some $X$, and every automorphism is inner, $A\mapsto UAU^{-1}$ for some invertible $U$. The two statements are the theorems of Wedderburn and Skolem–Noether, and they give
$$ \operatorname{Der}(M_4)=\operatorname{ad}\mathfrak{sl}_4\cong\mathfrak{sl}_4,\qquad \operatorname{Aut}(M_4)=PGL_4(\mathbb{C}), $$
both of dimension $15$. There are no outer derivations and no outer automorphisms.
This article states the two theorems, the kernel of the adjoint map, the $*$-structures, the complex-antilinear automorphisms, and the comparison with the biquaternion algebra, whose derivations are only three-dimensional. The Lie algebra itself is The Lie Algebra $\mathfrak{gl}_4(\mathbb{C})$ and $\mathfrak{sl}_4(\mathbb{C})$ of $M_4(\mathbb{C})$; the inner automorphisms as a group are The Adjoint Action, the Orbits and the Flag Variety of $M_4(\mathbb{C})$; the operators of the Peirce decomposition are The Operators of the Peirce Decomposition of $M_4(\mathbb{C})$.
Derivations
Definition. Let $\operatorname{Der}(M_4)$ be the vector space of derivations of $M_4(\mathbb{C})$, with the bracket $[D_1,D_2]=D_1D_2-D_2D_1$.
Proposition. $\operatorname{Der}(M_4)$ is a Lie algebra, and for every $X\in M_4$ the map
$$ \operatorname{ad}_X(Y)=[X,Y] $$
is a derivation.
Proof. The Jacobi identity gives
$$ \operatorname{ad}_X(YZ)=[X,YZ]=[X,Y]Z+Y[X,Z]=\operatorname{ad}_X(Y)Z+Y\operatorname{ad}_X(Z). $$
The bracket of two derivations is a derivation because the commutator of two endomorphisms satisfying the Leibniz rule satisfies it as well.
Proposition. The kernel of $\operatorname{ad}:\mathfrak{gl}_4\to\operatorname{Der}(M_4)$ is the centre $\mathbb{C}I$, and the image is $\operatorname{ad}\mathfrak{sl}_4$, of dimension $15$.
Proof. $\operatorname{ad}_X=0$ exactly when $X$ commutes with every matrix, that is, when $X$ is scalar. The image of $\operatorname{ad}$ is isomorphic to $\mathfrak{gl}_4/\mathbb{C}I\cong\mathfrak{sl}_4$, of dimension $15$.
Theorem (Wedderburn). Every derivation of $M_4(\mathbb{C})$ is inner:
$$ \operatorname{Der}(M_4)=\operatorname{ad}\mathfrak{sl}_4,\qquad \dim\operatorname{Der}(M_4)=15. $$
Proof. The theorem is the case $n=4$ of Wedderburn's theorem: every derivation of a central simple algebra over a field is inner. The elementary proof reconstructs $X$ from the images of the matrix units. Fix the units $E_{ij}$ and a derivation $D$. The relation $E_{ij}E_{kl}=\delta_{jk}E_{il}$ forces the entries of $D(E_{ij})$ to be the entries of a single matrix $X$: the entry $(1,k)$ of $D(E_{1j})$ is $X_{jk}$ for $k\ne j$, the entry $(1,j)$ of $D(E_{1j})$ is $X_{jj}-X_{11}$, and the entries of $D(E_{ij})$ in the other positions are the remaining entries of $X$ and $-X$; the consistency of these readings is the Leibniz rule applied to the three-factor identities $E_{ij}E_{jk}E_{ki}=E_{ii}$. With $X$ so reconstructed and normalised by $X_{11}=0$, the map $D-\operatorname{ad}_X$ is a derivation vanishing on every $E_{1j}$, hence on every matrix unit, hence everywhere. So $D=\operatorname{ad}_X$.
The inner derivations. The map $\operatorname{ad}$ is an isomorphism $\mathfrak{sl}_4\to\operatorname{Der}(M_4)$, and the derivation $\operatorname{ad}_X$ depends only on the traceless part of $X$. The fifteen-dimensional space of derivations is therefore a copy of $\mathfrak{sl}_4$, and the scalar matrices act trivially.
The two bases. The argument reconstructs $X$ in the matrix units $E_{ij}$, the basis of the statement. The derivations are read as well on the biquaternion algebra $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ in the coefficient basis $e_t,\vec e_x,\vec e_y,\vec e_z$: the algebra of the operators is $\mathcal{E}=\mathbb{B}\otimes_{\mathbb{C}}\mathbb{B}_{\mathrm{opp}}$ of dimension $16$, the four left matrices are $M_4(\mathbb{C})$ Axis-Free Representation and the sixteen products $T_{\mu\nu}=L_{\mathrm{axf},e_\mu}R_{\mathrm{axf},e_\nu}$ are $M_4(\mathbb{C})$ Axis-Free Representation: Operators, and $\operatorname{ad}$ there is the commutator with those matrices.
The grading. The derivations act on the Peirce blocks of an idempotent by the rule of The Operators of the Peirce Decomposition of $M_4(\mathbb{C})$: $\operatorname{ad}_X$ carries $A_{rs}$ to $A_{rs}$ for the diagonal blocks and mixes the off-diagonal blocks according to the entries of $X$. The derivation is determined by its values on the four blocks.
Automorphisms
Definition. An automorphism of $M_4(\mathbb{C})$ is an invertible linear map $\varphi$ with $\varphi(AB)=\varphi(A)\varphi(B)$ for all $A,B$.
Proposition. For every $U\in GL_4(\mathbb{C})$ the map $\operatorname{Inn}_U(A)=UAU^{-1}$ is an automorphism, and $\operatorname{Inn}_U=\operatorname{Inn}_V$ exactly when $V=\lambda U$ for some $\lambda\in\mathbb{C}^{\times}$.
Proof. Conjugation preserves products and is invertible. If $UAU^{-1}=VAV^{-1}$ for all $A$, then $V^{-1}U$ commutes with every $A$, hence is scalar.
Theorem (Skolem–Noether). Every automorphism of $M_4(\mathbb{C})$ is inner:
$$ \operatorname{Aut}(M_4)=PGL_4(\mathbb{C})=GL_4(\mathbb{C})/\mathbb{C}^{\times}I, $$
of dimension $15$. In particular there is no outer automorphism.
Proof. The theorem is the case $n=4$ of the Skolem–Noether theorem: every isomorphism of central simple algebras is conjugation by an invertible element. The elementary proof uses the module: an automorphism $\varphi$ makes $\mathbb{C}^{4}$ into a second module for $M_4$ by $A\cdot v=\varphi(A)v$, and the two modules are isomorphic because $M_4$ has a unique irreducible module up to isomorphism; an intertwiner $U$ satisfies $\varphi(A)U=UA$, that is, $\varphi(A)=UAU^{-1}$.
The centre. The scalar matrices act trivially, so the automorphism group is the quotient $PGL_4$, not $GL_4$. The determinant of $A$ is preserved by every inner automorphism, but the group is not reduced to $SL_4$: conjugation by $U$ and by $\lambda U$ agree for every $\lambda\ne0$, and the correct statement is $PGL_4\cong SL_4/\mu_4$, where $\mu_4$ is the group of fourth roots of unity acting by scalars.
The Lie algebra. The Lie algebra of $\operatorname{Aut}(M_4)$ is $\operatorname{Der}(M_4)=\mathfrak{sl}_4$, and the exponential of a derivation is an automorphism: $\exp(\operatorname{ad}_X)(A)=e^{X}Ae^{-X}$. The exponential map of $PGL_4$ is the exponential of $\mathfrak{sl}_4$ modulo the kernel of the adjoint representation.
The $*$-Structures
Definition. A $*$-derivation is a derivation $D$ with $D(A^{*})=D(A)^{*}$ for all $A$. A $*$-automorphism is an automorphism $\varphi$ with $\varphi(A^{*})=\varphi(A)^{*}$.
Proposition. A derivation $\operatorname{ad}_X$ is a $*$-derivation exactly when $X$ is anti-Hermitian, $X^{*}=-X$. The $*$-derivations form the real Lie algebra $\mathfrak{u}(4)$ modulo its centre, isomorphic to $\mathfrak{su}(4)$, of real dimension $15$.
Proof. For $Y$ arbitrary,
$$ \operatorname{ad}_X(Y)^{*}=[Y,X]^{*}=[X^{*},Y^{*}]=-\operatorname{ad}_{X^{*}}(Y^{*}). $$
The condition $\operatorname{ad}_X(Y^{*})=\operatorname{ad}_X(Y)^{*}$ for all $Y$ is therefore $\operatorname{ad}_X=\operatorname{ad}_{-X^{*}}$, that is, $X+X^{*}$ is scalar. The scalar part acts trivially, so the condition is that the traceless part of $X$ is anti-Hermitian; the anti-Hermitian matrices form $\mathfrak{u}(4)$ of real dimension $16$, and the scalars $i\theta I$ act trivially, leaving $\mathfrak{su}(4)$ of dimension $15$.
The $*$-automorphisms. The automorphism $\operatorname{Inn}_U$ is a $*$-automorphism exactly when $U$ is unitary:
$$ (UAU^{-1})^{*}=UAU^{-1}\quad\text{for all }A \iff U^{*}U\in\mathbb{C}^{\times}I. $$
The group of $*$-automorphisms is therefore $PU(4)=U(4)/U(1)$, of real dimension $15$, the compact real form of $PGL_4$.
The real structure. For the real form $M_4(\mathbb{R})$ the derivations are $\operatorname{ad}\mathfrak{sl}_4(\mathbb{R})$, of real dimension $15$, and the automorphisms are $PGL_4(\mathbb{R})$. For the quaternionic form $M_2(\mathbb{H})$ the derivations are $\operatorname{ad}\mathfrak{sl}_2(\mathbb{H})$, and the automorphisms are $PGL_2(\mathbb{H})$; the two real forms are The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$.
The Antilinear Automorphisms
Definition. A complex-antilinear automorphism is a bijective map $\varphi$ with
$$ \varphi(A+B)=\varphi(A)+\varphi(B),\qquad \varphi(\lambda A)=\bar\lambda\,\varphi(A),\qquad \varphi(AB)=\varphi(A)\varphi(B). $$
Proposition. Every antilinear automorphism is of the form
$$ \varphi(A)=U\bar A U^{-1} $$
for some $U\in GL_4(\mathbb{C})$, and two matrices $U,V$ give the same map exactly when $V=\lambda U$ with $\lambda\in\mathbb{C}^{\times}$.
Proof. Composing $\varphi$ with the conjugation $\bar A$ gives a complex-linear automorphism $A\mapsto \varphi(\bar A)$, which is inner by Skolem–Noether: $\varphi(\bar A)=UAU^{-1}$, hence $\varphi(A)=U\bar A U^{-1}$. The ambiguity is the scalar, as for the linear case.
The group. The antilinear automorphisms form the coset $PGL_4\cdot\bar{\phantom{A}}$, and
$$ \operatorname{Aut}_\mathbb{R}(M_4)=PGL_4(\mathbb{C})\rtimes\mathbb{Z}/2, $$
where the nontrivial element acts by $\bar{\phantom{A}}$. The subgroup of complex-linear automorphisms has index two, and the coset is the set of antilinear ones.
The real forms again. An antilinear automorphism of order two is a real structure on the algebra; its fixed algebra is a real form. The two possibilities, $\varphi^{2}=\mathrm{id}$ and $\varphi^{2}$ an inner automorphism of order two, correspond to the real and quaternionic structures. The classification is The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$; the point here is that the antilinear automorphisms are a single coset, and the real forms are the fixed algebras of its involutions.
The Biquaternion Comparison
The derivations of $\mathbb{B}$. The biquaternion algebra $\mathbb{B}\cong M_2(\mathbb{C})$ has derivations
$$ \operatorname{Der}(\mathbb{B})=\operatorname{ad}\mathbb{B}\cong\mathfrak{sl}_2(\mathbb{C}), $$
of complex dimension $3$, and automorphisms $PGL_2(\mathbb{C})$, of dimension $3$. The inner derivation $\operatorname{ad}_u$ for $u\in\mathbb{B}$ is the infinitesimal rotation of the biquaternion algebra; the three-dimensional algebra is the one generated by the three imaginary units.
The embedding. The left factor of the Kronecker decomposition gives an inclusion
$$ \mathfrak{sl}_2(\mathbb{C})\cong\operatorname{ad}(\mathbb{B}\otimes I)\subset\operatorname{ad}\mathfrak{sl}_4=\operatorname{Der}(M_4), $$
and the right factor gives a second, commuting copy. The two copies are the two factors of The Kronecker Factorisation $M_4(\mathbb{C})\cong M_2(\mathbb{C})\otimes M_2(\mathbb{C})$, and their sum is a six-dimensional subalgebra of the fifteen derivations. The remaining nine derivations mix the two factors: they are the ones that do not preserve the tensor decomposition.
The physical interpretation. The derivations of $\mathbb{B}$ are the gauge rotations of the biquaternion algebra, of dimension three; the derivations of $M_4$ are the gauge rotations of the algebra of two-sided operators, of dimension fifteen. The difference of twelve is the price of the passage from the four-dimensional algebra to its sixteen-dimensional endomorphism algebra, and it is the same difference that separates $\mathbb{B}$ from the gamma set — five of the sixteen basis operators, up to $i$ — of The Gamma Frame and the Complexified Dirac Algebra of $M_4(\mathbb{C})$.
Summary
Every derivation of $M_4(\mathbb{C})$ is inner, $\operatorname{Der}(M_4)=\operatorname{ad}\mathfrak{sl}_4\cong\mathfrak{sl}_4$ of dimension $15$, with kernel $\mathbb{C}I$; every automorphism is inner, $\operatorname{Aut}(M_4)=PGL_4(\mathbb{C})$ of dimension $15$, with no outer automorphisms. The $*$-derivations are the anti-Hermitian ones, forming $\mathfrak{su}(4)$, and the $*$-automorphisms are $PU(4)$; the antilinear automorphisms form the coset $PGL_4\rtimes\mathbb{Z}/2$. The biquaternion algebra has only a three-dimensional derivation algebra, embedded twice in the fifteen, and the remaining nine derivations mix the two Kronecker factors.
| Quantity | Value |
|---|---|
| $\dim\operatorname{Der}(M_4)$ | $15$ |
| $\operatorname{Der}(M_4)$ | $\operatorname{ad}\mathfrak{sl}_4$ |
| Kernel of $\operatorname{ad}$ | $\mathbb{C}I$ |
| $\dim\operatorname{Aut}(M_4)$ | $15$ |
| $\operatorname{Aut}(M_4)$ | $PGL_4(\mathbb{C})$ |
| Outer automorphisms | none |
| $*$-derivations | $\mathfrak{su}(4)$, dimension $15$ |
| $*$-automorphisms | $PU(4)$, dimension $15$ |
| Antilinear automorphisms | $PGL_4\rtimes\mathbb{Z}/2$ |
| $\dim\operatorname{Der}(\mathbb{B})$ | $3$ |
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\operatorname{Der}(M_4)$ | the Lie algebra of derivations |
| $\operatorname{ad}_X$ | the inner derivation $Y\mapsto [X,Y]$ |
| $\operatorname{Inn}_U$ | the inner automorphism $A\mapsto UAU^{-1}$ |
| $PGL_4$ | $GL_4/\mathbb{C}^{\times}I$, the automorphism group |
| $PU(4)$ | $U(4)/U(1)$, the $*$-automorphism group |
| $\operatorname{Aut}_\mathbb{R}$ | the group of linear and antilinear automorphisms |
| $\mu_4$ | the fourth roots of unity, the kernel of $SL_4\to PGL_4$ |
Further Reading
- $M_4(\mathbb{C})$ Axis-Free Representation, for the coefficient basis, the four left matrices and the transpose as the quaternion conjugation.
- $M_4(\mathbb{C})$ Axis-Free Representation: Operators, for the sixteen two-sided operators in the coefficient frame, their signs and their composition.
- The Lie Algebra $\mathfrak{gl}_4(\mathbb{C})$ and $\mathfrak{sl}_4(\mathbb{C})$ of $M_4(\mathbb{C})$, for $\mathfrak{sl}_4$ and its root decomposition.
- The Operators of the Peirce Decomposition of $M_4(\mathbb{C})$, for the derivations on the Peirce blocks.
- The Adjoint Action, the Orbits and the Flag Variety of $M_4(\mathbb{C})$, for the action of $PGL_4$.
- The Kronecker Factorisation $M_4(\mathbb{C})\cong M_2(\mathbb{C})\otimes M_2(\mathbb{C})$, for the two commuting $\mathfrak{sl}_2$'s.
- The Quaternionic Structure and the Real Forms of $M_4(\mathbb{C})$, for the real and quaternionic structures.
- The Involution and the Hermitian Structure of $M_4(\mathbb{C})$, for the $*$-structures.
- Richard S. Pierce, Associative Algebras (Springer, 1982), for Wedderburn's and Skolem–Noether's theorems.
- Nathan Jacobson, Lie Algebras (Dover, 1979), for the derivations of matrix algebras.