The Derivation $X\mapsto[A,X]$ and Its Anti-Self-Adjointness for the Trace Form at $n=4$

Introduction

Every commutator map $\operatorname{ad}_A(X)=[A,X]$ is a derivation of $M_4(\mathbb{C})$, every derivation is of this form because the algebra is central simple, and the space of them is $\mathfrak{sl}_4$ of dimension $15$; the map is anti-self-adjoint for the trace form for every $A$ and anti-self-adjoint for the Hilbert–Schmidt form only when $A$ is anti-Hermitian, and the difference between the two statements is the involution, not the derivation.

The article proves the Leibniz rule and the representation property, records the inner-derivation theorem with its dimension count, computes the two anti-self-adjointness conditions and their adjoints, determines the spectrum of $\operatorname{ad}_A$, and explains why the trace form cannot distinguish the symmetric from the antisymmetric directions while the Hilbert–Schmidt form can.

It assumes The Trace Form and Its Inertia on the Two Halves of $M_4(\mathbb{C})$ and The Derivations and the Automorphisms of $M_4(\mathbb{C})$, where the inner-derivation theorem is proved in full. The case $n=2$ is The Derivation Commutator and Its Anti-Self-Adjointness for the Trace Form; the operator form is The Adjoints and the Polar Decomposition of an Operator at $n=4$.

The Commutator Is a Derivation

Proposition. For every $A$, the map $\operatorname{ad}_A(X)=AX-XA$ satisfies the Leibniz rule $\operatorname{ad}_A(XY)=\operatorname{ad}_A(X)Y+X\operatorname{ad}_A(Y)$.

Proof. $A(XY)-(XY)A=(AX-XA)Y+X(AY-YA)$, by expanding both sides.

Corollary. $\operatorname{ad}_A$ is a derivation of $M_4$ for every $A$, and $\operatorname{ad}$ is linear in $A$.

Proof. The Leibniz rule is the definition of a derivation, and the map $A\mapsto\operatorname{ad}_A$ is linear by the bilinearity of the commutator.

Proposition. $\operatorname{ad}_{[A,B]}=[\operatorname{ad}_A,\operatorname{ad}_B]$, so $\operatorname{ad}$ is a homomorphism of Lie algebras from $\mathfrak{gl}_4$ into $\operatorname{Der}(M_4)$.

Proof. The Jacobi identity: $[[A,B],X]=[A,[B,X]]-[B,[A,X]]$, which is the commutator of the two maps applied to $X$.

Corollary. The image of $\operatorname{ad}$ is a Lie subalgebra of the derivations, isomorphic to $\mathfrak{gl}_4$ modulo its kernel.

Proof. The first isomorphism theorem for Lie algebras.

The spectrum. If $A$ is diagonalisable with eigenvalues $\lambda_1,\dots,\lambda_4$, the eigenvalues of $\operatorname{ad}_A$ on $M_4$ are the sixteen differences $\lambda_i-\lambda_j$.

Proof. In the eigenbasis of $A$, $\operatorname{ad}_A(E_{ij})=(\lambda_i-\lambda_j)E_{ij}$, and the sixteen units form a basis.

Corollary. $\operatorname{ad}_A$ is nilpotent exactly when $A$ has a single eigenvalue, and it is zero exactly when $A$ is a scalar.

Proof. All differences vanish exactly when all eigenvalues are equal, which over $\mathbb{C}$ means $A=\lambda\mathrm I+N$ with $N$ nilpotent; then $\operatorname{ad}_A=\operatorname{ad}_N$ is nilpotent.

Every Derivation Is Inner

Theorem. Every derivation of $M_4(\mathbb{C})$ is inner: $\operatorname{Der}(M_4)=\operatorname{ad}(M_4)$, and the kernel of $\operatorname{ad}$ is the centre $\mathbb{C}\mathrm I_4$.

Proof. The kernel is the set of $A$ commuting with every matrix, which is the centre. Surjectivity is the theorem of the inner derivation, proved from the matrix units: a derivation is determined by its values on the sixteen units, and the element $a=\sum_{i,j}D(E_{ij})E_{ji}$ reproduces those values, the verification being the identity $D(E_{ij})=[a,E_{ij}]$ read off from $E_{ik}E_{kj}=E_{ij}$ summed over $k$.

Corollary. $\operatorname{Der}(M_4)\cong\mathfrak{gl}_4/\mathbb{C}\mathrm I_4\cong\mathfrak{sl}_4$, of dimension $15$.

Proof. The kernel has dimension $1$ and the source has dimension $16$.

Corollary. There are no outer derivations, and the Lie algebra of derivations is simple.

Proof. The quotient of $\mathfrak{gl}_4$ by the scalars is $\mathfrak{sl}_4$, which is simple.

Why the theorem needs the algebra to be simple. For a direct sum $A\oplus B$ the map sending a pair to the derivation $(a,b)\mapsto(0,\delta(b))$ with $\delta$ a derivation of $B$ is not inner in general; simplicity removes the summands, and centrality removes the scalar ambiguity.

Proof. A derivation of a direct sum acts on each summand and can be inner in one and outer in the other; the centre supplies the only ambiguity for a simple algebra.

Anti-Self-Adjointness for the Trace Form

Theorem. For every $A$ and all $X,Y$,

$$ \beta(\operatorname{ad}_A X,Y)+\beta(X,\operatorname{ad}_A Y)=0,\qquad \beta(X,Y)=\operatorname{tr}(XY). $$

Proof. Expand: $\operatorname{tr}((AX-XA)Y)+\operatorname{tr}(X(AY-YA))=\operatorname{tr}(AXY)-\operatorname{tr}(XAY)+\operatorname{tr}(XAY)-\operatorname{tr}(XYA)$, and the two middle terms cancel, leaving $\operatorname{tr}(AXY)-\operatorname{tr}(XYA)=0$ by the cyclic invariance of the trace.

Corollary. Every inner derivation lies in the orthogonal Lie algebra of $\beta$, which is $\mathfrak{so}(10,6)$ of dimension $120$; the fifteen inner derivations form a subalgebra of it.

Proof. Anti-Self-Adjointness for a bilinear form is the defining equation of its orthogonal Lie algebra, and $\mathfrak{so}(p,q)$ with $p+q=16$ has dimension $16\cdot15/2=120$.

The trace form cannot detect the condition. No hypothesis on $A$ is needed: the anti-self-adjointness is produced by the cyclic invariance of the trace alone, which holds for every triple of factors. The form is invariant under the whole of $\mathrm{GL}_4$, since $\operatorname{tr}(PXP^{-1}\cdot PYP^{-1})=\operatorname{tr}(XY)$, and the invariance is what produces the anti-self-adjointness.

Proof. Conjugation invariance of the trace, applied to the invariance identity $\beta(gXg^{-1},gYg^{-1})=\beta(X,Y)$ differentiated at the identity.

Anti-Self-Adjointness for the Hilbert–Schmidt Form

Definition. $\langle X,Y\rangle=\operatorname{tr}(X^{\ast}Y)$, positive definite.

Proposition. $\operatorname{ad}_A^{\ast}=\operatorname{ad}_{A^{\ast}}$ for the Hilbert–Schmidt form.

Proof. $\langle\operatorname{ad}_A X,Y\rangle=\operatorname{tr}((AX-XA)^{\ast}Y)=\operatorname{tr}(X^{\ast}A^{\ast}Y)-\operatorname{tr}(A^{\ast}X^{\ast}Y)=\langle X,A^{\ast}Y\rangle-\langle X,YA^{\ast}\rangle=\langle X,\operatorname{ad}_{A^{\ast}}Y\rangle$ after moving $A^{\ast}$ to the right by cyclicity.

Corollary. $\operatorname{ad}_A$ is anti-self-adjoint for the Hilbert–Schmidt form exactly when $A$ differs from an anti-Hermitian matrix by a real scalar.

Proof. Anti-Self-Adjointness is $\operatorname{ad}_{A^{\ast}}=-\operatorname{ad}_A$, that is $A^{\ast}+A\in\mathbb{C}\mathrm I_4$; writing $A^{\ast}+A=\mu\mathrm I$ and taking adjoints gives $\bar\mu=\mu$, so $\mu$ is real and $A-\tfrac{\mu}{2}\mathrm I_4$ is anti-Hermitian, with the same adjoint map since the scalar part is central.

Corollary. The derivations that are anti-self-adjoint for the Hilbert–Schmidt form are exactly those of $\operatorname{ad}\mathfrak{u}(4)$, of dimension $15$.

Proof. The anti-Hermitian matrices form $\mathfrak{u}(4)$ of real dimension $16$, and the scalars in it are the single imaginary line, so the image has dimension $15$.

The adjoint is a different operator. For the trace form the adjoint of $\operatorname{ad}_A$ is $-\operatorname{ad}_A$; for the Hilbert–Schmidt form it is $\operatorname{ad}_{A^{\ast}}$, which is a different derivation unless $A$ is anti-Hermitian.

Proof. Compare the two displayed formulas.

The Difference Is the Involution

Proposition. The condition for anti-self-adjointness for $\beta$ involves no involution; the condition for anti-self-adjointness for $\langle\cdot,\cdot\rangle$ involves the adjoint ${}^{\ast}$.

Proof. The first condition is an identity in $A$, the second is the equation $A^{\ast}+A\in\mathbb{C}\mathrm I$.

Corollary. The trace form sees the whole of $\mathfrak{gl}_4$ inside $\mathfrak{so}(10,6)$, while the Hilbert–Schmidt form sees only $\mathfrak{u}(4)$ inside $\mathfrak{so}(16)$.

Proof. The two invariance groups are $\mathrm{GL}_4$ and $U(4)$, and the Lie algebras of anti-self-adjoint endomorphisms contain the respective images of $\operatorname{ad}$.

The moral. Every derivation is anti-self-adjoint for the trace form because that form is invariant under the largest possible group; a derivation is anti-self-adjoint for the Hilbert–Schmidt form only when its generator lies in the Lie algebra of the smaller group. The involution ${}^{\ast}$ chooses the smaller group, and the derivation is the same object in both statements.

The two halves of the statement. For $A$ symmetric, $\operatorname{ad}_A$ is anti-self-adjoint for $\beta$ and self-adjoint for $\langle\cdot,\cdot\rangle$; for $A$ antisymmetric, it is anti-self-adjoint for both.

Proof. For symmetric $A$, $A^{\ast}=A$, so $\operatorname{ad}_A^{\ast}=\operatorname{ad}_A$; for antisymmetric $A$, $A^{\ast}=-A$, so the adjoint is $-\operatorname{ad}_A$.

Corollary. The space of inner derivations has dimension $15$, the dimension of $\mathfrak{sl}_4$, since the scalars act by zero and every traceless matrix gives a nonzero derivation.

Proof. The kernel of $A\mapsto\operatorname{ad}_A$ is the centre, which is the scalars.

Summary

The commutator map is a derivation for every element, the representation property holds by the Jacobi identity, and every derivation of the algebra is inner, so the derivations form $\mathfrak{sl}_4$ of dimension $15$.

The map is anti-self-adjoint for the trace pairing for every $A$, with no condition, because the trace is cyclically invariant; it is anti-self-adjoint for the Hilbert–Schmidt form exactly when $A$ is anti-Hermitian up to a real scalar, that is when the derivation lies in $\operatorname{ad}\mathfrak{u}(4)$.

The difference between the two statements is the involution and not the derivation: the trace form is invariant under all of $\mathrm{GL}_4$, the Hilbert–Schmidt form only under $U(4)$, and the adjoint needed for the second statement is built from the conjugate transpose. The spectrum of $\operatorname{ad}_A$ is the sixteen differences of eigenvalues, and the kernel is the centraliser, of dimension at least four.

Summary of Notation

Symbol Meaning
$\operatorname{ad}_A(X)=[A,X]$ the inner derivation
$\operatorname{Der}(M_4)$ the derivations, isomorphic to $\mathfrak{sl}_4$
$\beta(X,Y)=\operatorname{tr}(XY)$ the trace pairing
$\langle X,Y\rangle=\operatorname{tr}(X^{\ast}Y)$ the Hilbert–Schmidt form
$\mathfrak{so}(10,6)$ the orthogonal Lie algebra of $\beta$
$\operatorname{ad}\mathfrak{u}(4)$ the derivations anti-self-adjoint for the Hilbert–Schmidt form

Further Reading