The Operators of the Peirce Decomposition of $M_3(\mathbb{C})$
Introduction
The Peirce decomposition is acted on by four families of linear operators: the left multiplication $L_a$, the right multiplication $R_a$, the two-sided operator $L_{a}R_{b}$ and the derivation $\operatorname{ad}_a$. On the matrix units, $L_a$ moves the row index, $R_a$ moves the column index, $L_{a}R_{b}$ moves both, and $\operatorname{ad}_a$ is the difference of the two. The derivations of $M_3(\mathbb{C})$ are exactly the inner derivations, $\operatorname{Der}(M_3)\cong\mathfrak{sl}_3(\mathbb{C})$, of complex dimension $8$; the automorphisms are exactly the inner automorphisms, $\operatorname{Aut}(M_3)\cong PGL_3(\mathbb{C})$, of complex dimension $8$. The transpose, the conjugate transpose and the entrywise conjugation are not complex-linear automorphisms: they are anti-automorphisms or conjugate-linear maps.
The article defines the four operators, computes their matrices and their action on the Peirce corners, proves the derivation and automorphism theorems, and explains the two-sided operator as a map of corners. The Peirce decomposition and the corners $A_{ij}=\mathbb{C}E_{ij}$ are The Peirce Decomposition of $M_3(\mathbb{C})$; the coarse split by a rank-one idempotent is The Rank-Two Split of $M_3(\mathbb{C})$; the conjugation action and its orbits are The Adjoint Action, the Orbits and the Flag Variety of $M_3(\mathbb{C})$; the frame and the gauge are The Frame, the Simple Nilpotents and the Gauge of $M_3(\mathbb{C})$.
The Four Operators
Definition. For $a,b\in M_3(\mathbb{C})$, define the linear maps $M_3(\mathbb{C})\to M_3(\mathbb{C})$ by
$$ L_a(X)=aX,\qquad R_a(X)=Xa,\qquad L_{a}R_{b}(X)=aXb,\qquad \operatorname{ad}_a(X)=[a,X]=aX-Xa . $$
The operator $L_{a}R_{b}$ is the two-sided operator; $\operatorname{ad}_a$ is the adjoint or inner derivation.
Proposition. The operators satisfy
$$ L_{a}R_{b}=R_bL_a,\qquad \operatorname{ad}_a=L_a-R_a, $$
and
$$ L_aL_b=L_{ab},\qquad R_aR_b=R_{ba},\qquad L_aR_b=R_bL_a . $$
Proof. The identities are the associativity of the product: $a(Xb)=(aX)b$ and $(ab)X=a(bX)$, while $X(ba)=(Xb)a$. The adjoint is the difference of the left and the right multiplications.
The matrices. Vectorise $M_3$ by rows, so that $X$ is a column of nine entries, ordered $X_{11},X_{12},X_{13},X_{21},\dots,X_{33}$. Then
$$ L_a=a\otimes I_3,\qquad R_b=I_3\otimes b^{\mathsf T},\qquad L_{a}R_{b}=a\otimes b^{\mathsf T}, $$
$$ \operatorname{ad}_a=a\otimes I_3-I_3\otimes a^{\mathsf T}. $$
Proof. The row vectorisation of $aX$ is $(a\otimes I_3)\operatorname{vec}X$, because the $(i,j)$ entry of $aX$ is $\sum_k a_{ik}X_{kj}$. The row vectorisation of $Xb$ is $(I_3\otimes b^{\mathsf T})\operatorname{vec}X$, because the $(i,j)$ entry of $Xb$ is $\sum_l X_{il}b_{lj}$. The two-sided operator is the composition, and the adjoint is the difference.
The eigenvalues and the trace. Let $\lambda_1,\lambda_2,\lambda_3$ be the eigenvalues of $a$, counted with algebraic multiplicity, and $\mu_1,\mu_2,\mu_3$ those of $b$. The eigenvalues of $L_a$ are $\lambda_1,\lambda_2,\lambda_3$, each with multiplicity $3$; those of $R_b$ are $\mu_1,\mu_2,\mu_3$, each with multiplicity $3$; those of $L_{a}R_{b}$ are the nine products $\lambda_i\mu_j$; and those of $\operatorname{ad}_a$ are the nine differences $\lambda_i-\lambda_j$. Hence
$$ \operatorname{tr}L_a=3\operatorname{tr}a,\qquad \operatorname{tr}R_b=3\operatorname{tr}b,\qquad \operatorname{tr}L_{a}R_{b}=\operatorname{tr}a\operatorname{tr}b,\qquad \operatorname{tr}\operatorname{ad}_a=0, $$
$$ \det L_a=(\det a)^3,\qquad \det R_b=(\det b)^3,\qquad \det L_{a}R_{b}=(\det a)^3(\det b)^3. $$
Proof. The eigenvalues of $a\otimes b^{\mathsf T}$ are the pairwise products of the eigenvalues of $a$ and of $b^{\mathsf T}$, which are those of $b$; the trace is the sum of the products, $\operatorname{tr}a\operatorname{tr}b$, and the determinant is their product, $(\det a)^3(\det b)^3$. The assertions for $L_a$ and $R_b$ are the cases $b=I$ and $a=I$. For the adjoint, $a\otimes I$ and $I\otimes a^{\mathsf T}$ commute, and their eigenvalues on the common eigenbasis are $\lambda_i$ and $\lambda_j$, so the difference has eigenvalues $\lambda_i-\lambda_j$. The trace is $\sum_{i,j}(\lambda_i-\lambda_j)=3\operatorname{tr}a-3\operatorname{tr}a=0$.
The rank of the adjoint. The kernel of $\operatorname{ad}_a$ is the centraliser
$$ Z(a)=\{X\in M_3:aX=Xa\}. $$
For a semisimple element with eigenvalue multiplicities $m_1,m_2,m_3$, the centraliser is the direct sum of the full matrix algebras on the eigenspaces, and $\dim Z(a)=\sum_i m_i^2$. In general the centraliser of a Jordan form is computed block by block, and
$$ \operatorname{rank}\operatorname{ad}_a=9-\dim Z(a). $$
For an element with three distinct eigenvalues the centraliser is the algebra of polynomials in $a$, of dimension $3$, and the rank is $6$. For a scalar the centraliser is all of $M_3$, of dimension $9$, and the rank is $0$. For the principal nilpotent $J_3(0)$ the centraliser is the three-dimensional algebra $\mathbb{C}[N]$, and the rank is again $6$. For the subregular nilpotent $E_{12}$ the centraliser is five-dimensional, so the rank is $4$ and the orbit is the minimal nilpotent orbit.
Proof. A matrix commuting with a Jordan block of size $d$ is upper triangular Toeplitz on that block; a block mapping one Jordan block to another with the same eigenvalue is Toeplitz of the smaller size. The count is the standard one, in Horn and Johnson, Matrix Analysis, Chapter 4. The rank is $9-\dim\ker$.
The rank of the two-sided operator. The two-sided operator $L_{a}R_{b}=a\otimes b^{\mathsf T}$ has rank
$$ \operatorname{rank}L_{a}R_{b}=\operatorname{rank}a\cdot\operatorname{rank}b . $$
Its kernel is the set of $X$ with $aXb=0$, and its image is the set of matrices whose columns lie in $\operatorname{im}a$ and whose rows lie in the row space of $b$.
Worked example: a diagonal element. Take $a=\operatorname{diag}(1,2,4)$ and $b=\operatorname{diag}(3,5,7)$. Then $L_{a}R_{b}$ is diagonal in the basis of matrix units, with the nine entries $3,5,7,6,10,14,12,20,28$; its trace is $105=7\cdot15$, and its determinant is $(\det a)^{3}(\det b)^{3}=8^{3}\cdot105^{3}=592\,704\,000$. The adjoint $\operatorname{ad}_a$ is diagonal with the nine entries
$$ 0,\ -1,\ -3,\ 1,\ 0,\ -2,\ 3,\ 2,\ 0, $$
in the order $E_{11},E_{12},E_{13},E_{21},E_{22},E_{23},E_{31},E_{32},E_{33}$. Its rank is $6$, its kernel is the diagonal algebra of dimension $3$, and its trace is $0$.
Worked example: a non-diagonal element. Take
$$ a=\begin{pmatrix}1&2&0\\0&3&1\\0&0&1\end{pmatrix}, \qquad b=\begin{pmatrix}0&1&0\\1&0&0\\0&0&2\end{pmatrix}. $$
Then $aE_{12}=\begin{pmatrix}0&1&0\\0&0&0\\0&0&0\end{pmatrix}$ and
$$ L_{a}R_{b}(E_{12})=aE_{12}b=E_{11}, \qquad L_{a}R_{b}(E_{13})=2E_{13}. $$
The first identity is an elementary corner map: the unit $E_{12}$ is carried to $E_{11}$, not to a multiple of itself. The adjoint gives
$$ \operatorname{ad}_a(E_{12})=aE_{12}-E_{12}a =\begin{pmatrix}0&-2&-1\\0&0&0\\0&0&0\end{pmatrix} =-2E_{12}-E_{13}, $$
which is the difference of a row move and a column move.
The Action on the Matrix Units
The left multiplication. On the matrix units,
$$ L_a(E_{ij})=aE_{ij}=\sum_{k=1}^{3}a_{ki}E_{kj}. $$
The column index $j$ is fixed, and the row index is transformed by $a$. The left multiplication acts on the row index of the unit.
The right multiplication. On the matrix units,
$$ R_b(E_{ij})=E_{ij}b=\sum_{l=1}^{3}b_{jl}E_{il}. $$
The row index $i$ is fixed, and the column index is transformed by $b$. The right multiplication acts on the column index of the unit.
The two-sided operator. On the matrix units,
$$ L_{a}R_{b}(E_{ij})=aE_{ij}b=\sum_{k,l=1}^{3}a_{ki}b_{jl}E_{kl}. $$
The coefficient of $E_{kl}$ is $a_{ki}b_{jl}$: the row index is transformed by $a$ and the column index by $b$, independently. This is the corner map: the unit $E_{ij}$ is sent to a linear combination of the units $E_{kl}$, and the coefficient factorises into a row factor and a column factor.
The elementary two-sided operator. For $a=E_{ir}$ and $b=E_{sl}$,
$$ L_{a}R_{b}(E_{rs})=E_{ir}E_{rs}E_{sl}=E_{il}. $$
The two-sided operator with two elementary matrices carries the corner $A_{rs}$ to the corner $A_{il}$:
$$ E_{ir}E_{rs}E_{sl}=E_{il}. $$
The general two-sided operator is a linear combination of these elementary corner maps, with coefficients $a_{ir}b_{sl}$.
The corner matrix. The three-by-three grid of Peirce corners $A_{ij}=\mathbb{C}E_{ij}$ is acted on by the two-sided operator as
$$ A_{rs}\longmapsto\sum_{i,l}a_{ir}b_{sl}A_{il}. $$
The row index of the corner is transformed by $a$, the column index by $b$. The two-sided operator is the tensor product of the two actions, and its matrix on the nine-dimensional space of corners is $a\otimes b^{\mathsf T}$.
The adjoint on the units. The adjoint is
$$ \operatorname{ad}_a(E_{ij})=aE_{ij}-E_{ij}a =\sum_{k}a_{ki}E_{kj}-\sum_{l}a_{jl}E_{il}. $$
The first term changes the row index, the second changes the column index; the difference is the commutator. On the diagonal units, $\operatorname{ad}_a(E_{ii})=aE_{ii}-E_{ii}a$ has zero diagonal; its off-diagonal entries are $a_{ki}$ in row $k$ and $-a_{ik}$ in column $k$.
The Peirce table. The multiplication rule $A_{ij}A_{kl}\subseteq\delta_{jk}A_{il}$ is read through the operators: $L_a$ carries the corner $A_{ij}$ into the sum of the corners $A_{kj}$ weighted by $a_{ki}$, and $R_b$ carries it into the sum of the corners $A_{il}$ weighted by $b_{jl}$. The two-sided operator is the two moves composed, and the adjoint is their difference. The coarse split of The Rank-Two Split of $M_3(\mathbb{C})$ replaces the nine lines by the four corners $1+2+2+4$, and the same operators act on the four blocks.
The Derivations
Definition. A derivation of $M_3$ is a linear map $D:M_3\to M_3$ with
$$ D(XY)=D(X)Y+XD(Y). $$
The derivations form a Lie algebra $\operatorname{Der}(M_3)$ under the commutator.
Proposition. Every inner derivation $\operatorname{ad}_a$ is a derivation:
$$ \operatorname{ad}_a(XY)=aXY-XYa =(aX-Xa)Y+X(aY-Ya) =\operatorname{ad}_a(X)Y+X\operatorname{ad}_a(Y). $$
The map $a\mapsto\operatorname{ad}_a$ is linear, with kernel the centre $\mathbb{C}I$, so the inner derivations form a Lie algebra isomorphic to $M_3/\mathbb{C}I\cong\mathfrak{sl}_3(\mathbb{C})$, of complex dimension $8$.
Theorem (inner derivations). Every derivation of $M_3(\mathbb{C})$ is inner:
$$ \operatorname{Der}(M_3)=\operatorname{ad}_{M_3}\cong\mathfrak{sl}_3(\mathbb{C}), \qquad \dim_\mathbb{C}\operatorname{Der}(M_3)=8. $$
Proof. A derivation is determined by its values on the matrix units. For any derivation $D$, the element
$$ a=\frac{1}{3}\sum_{i,j=1}^{3}D(E_{ij})E_{ji} $$
satisfies $D=\operatorname{ad}_a$; the verification is the standard computation for central simple algebras, in Pierce, Associative Algebras, §Derivations. The kernel of $a\mapsto\operatorname{ad}_a$ is the centre $\mathbb{C}I$, so the image has dimension $9-1=8$, which is $\dim\mathfrak{sl}_3$. The inner derivations are contained in $\operatorname{Der}(M_3)$ and the two spaces have the same dimension, so they are equal.
There are no outer derivations. The space of outer derivations is zero. The Lie algebra structure is
$$ [\operatorname{ad}_a,\operatorname{ad}_b]=\operatorname{ad}_{[a,b]}, $$
so $\mathfrak{sl}_3\to\operatorname{Der}(M_3)$ is an isomorphism of Lie algebras. The derivations act on $M_3$ by the adjoint action, and the decomposition of $M_3$ under $\mathfrak{sl}_3$ is the adjoint decomposition $1+3+5$ under the principal $\mathfrak{sl}_2$.
Worked example. For $a=\operatorname{diag}(1,2,4)$ the adjoint has rank $6$ and kernel the diagonal matrices, of dimension $3$. The image is the six-dimensional space of matrices with zero diagonal, the span of the six off-diagonal units. The element $a$ is regular semisimple: its centraliser is the diagonal torus, and the derivation reaches every off-diagonal direction. For $a=I$ the adjoint is zero and the kernel is the whole algebra.
The biquaternion case. For $M_2(\mathbb{C})$ the same theorem gives $\operatorname{Der}(M_2)\cong\mathfrak{sl}_2(\mathbb{C})$, of dimension $3$. The passage from $n=2$ to $n=3$ raises the derivation dimension from $3$ to $8$.
The Automorphisms
Definition. An automorphism of $M_3$ is an invertible linear map $\Phi$ with $\Phi(XY)=\Phi(X)\Phi(Y)$.
Theorem (Skolem–Noether). Every automorphism of $M_3(\mathbb{C})$ is inner:
$$ \operatorname{Aut}(M_3)\cong PGL_3(\mathbb{C})=GL_3(\mathbb{C})/\mathbb{C}^\times I, \qquad \dim_\mathbb{C}=8. $$
Proof. The automorphism sends the matrix units to a complete system of $3\times3$ matrix units. By the Skolem–Noether theorem for central simple algebras, any two complete systems of matrix units are conjugate by an invertible matrix. Hence $\Phi(X)=SXS^{-1}$ for some $S\in GL_3$, and $S$ is determined up to a scalar by the centre theorem. The group is $PGL_3$, of dimension $9-1=8$.
The inner automorphisms. The conjugation $\Phi_S(X)=SXS^{-1}$ is an automorphism for every $S\in GL_3$. It preserves the trace, the determinant, the characteristic polynomial, the rank and the Jordan type. The kernel of $S\mapsto\Phi_S$ is the centre $\mathbb{C}^\times I$, so the automorphism group is $PGL_3$.
There are no complex-linear outer automorphisms. The "outer" operations that appear in the theory are not complex-linear automorphisms:
- the transpose $X\mapsto X^{\mathsf T}$ is an anti-automorphism, $\Phi(XY)=\Phi(Y)\Phi(X)$;
- the conjugate transpose $X\mapsto X^*$ is a conjugate-linear anti-automorphism;
- the entrywise conjugation $X\mapsto\bar X$ is a conjugate-linear automorphism.
These are real-linear or conjugate-linear maps, not elements of $\operatorname{Aut}_\mathbb{C}(M_3)$. The star is the involution of The Involution and the Hermitian Structure of $M_3(\mathbb{C})$.
The Lie algebra. The group $PGL_3$ has Lie algebra $\mathfrak{sl}_3$, and the adjoint action of the group on $M_3$ is the exponentiation of the adjoint action of the Lie algebra. The orbit of an element is its conjugacy class; the classification is The Adjoint Action, the Orbits and the Flag Variety of $M_3(\mathbb{C})$.
The Two-Sided Operator as a Map of Corners
The corner map. The two-sided operator $L_{a}R_{b}$ is the most general linear map built from one left multiplication and one right multiplication. On the corners,
$$ L_{a}R_{b}(A_{rs})=\sum_{i,l}a_{ir}b_{sl}A_{il}. $$
The coefficient of $A_{il}$ is the product of the $(i,r)$ entry of $a$ and the $(s,l)$ entry of $b$. The map is the tensor product of the row action and the column action; it sends the corner $(r,s)$ to a linear combination of the corners $(i,l)$.
The elementary corner maps. The elementary two-sided operator $L_{E_{ir}}R_{E_{sl}}$ sends the corner $A_{rs}$ to the corner $A_{il}$:
$$ L_{E_{ir}}R_{E_{sl}}(A_{rs})=A_{il}. $$
The elementary corner maps are the matrix units of the space of corner maps; the general two-sided operator is a linear combination with coefficients $a_{ir}b_{sl}$. In the worked example above, $L_{a}R_{b}(E_{12})=E_{11}$ is the elementary map with $i=1$, $r=1$, $s=2$, $l=1$ and coefficients $a_{11}=1$, $b_{21}=1$.
The inner automorphism. For $b=a^{-1}$,
$$ L_{a}R_{a^{-1}}(X)=aXa^{-1} $$
is the inner automorphism $\Phi_a$. On the corners, it sends $A_{rs}$ to a linear combination of the corners $A_{il}$ with coefficients $a_{ir}(a^{-1})_{sl}$. The inner automorphism preserves the trace and the determinant but not the individual corners; it permutes the spectral lines only when $a$ is diagonal in the frame.
The two-sided operator and the coarse Peirce table. For the rank-two split with $\Pi=Q$, the two-sided operator with block-diagonal $a,b$ preserves the four corners; the two-sided operator with off-diagonal $a$ or $b$ mixes the corners. The mass block $uv$ is carried to the corner determined by the row of $a$ and the column of $b$.
The operators span the endomorphisms. The two-sided operators $L_{a}R_{b}$ span the space of linear maps $X\mapsto aXb$: the nine-by-nine matrix $a\otimes b^{\mathsf T}$ ranges over a space of dimension $81$, which is the whole $\operatorname{End}(M_3)$. The left and right multiplications alone span the subspace of maps that move only one index, of dimension $18$; the two-sided operators add the mixed moves.
Summary
The four operators on $M_3$ are the left multiplication $L_a(X)=aX$, the right multiplication $R_a(X)=Xa$, the two-sided operator $L_{a}R_{b}(X)=aXb$ and the adjoint $\operatorname{ad}_a(X)=[a,X]$. They satisfy $L_{a}R_{b}=R_bL_a$ and $\operatorname{ad}_a=L_a-R_a$. On the matrix units, $L_a$ moves the row index, $R_a$ moves the column index, $L_{a}R_{b}$ moves both, and $\operatorname{ad}_a$ is the difference. The matrices are $L_a=a\otimes I_3$, $R_b=I_3\otimes b^{\mathsf T}$, $L_{a}R_{b}=a\otimes b^{\mathsf T}$ and $\operatorname{ad}_a=a\otimes I_3-I_3\otimes a^{\mathsf T}$; the eigenvalues are the eigenvalues of $a$, of $b$, the products $\lambda_i\mu_j$, and the differences $\lambda_i-\lambda_j$. The derivations of $M_3$ are exactly the inner derivations, $\operatorname{Der}(M_3)\cong\mathfrak{sl}_3$, of dimension $8$; the automorphisms are exactly the inner automorphisms, $\operatorname{Aut}(M_3)\cong PGL_3$, of dimension $8$. There are no outer derivations and no complex-linear outer automorphisms; the transpose and the conjugations are anti-automorphisms or conjugate-linear maps. The two-sided operator is the corner map $A_{rs}\mapsto\sum a_{ir}b_{sl}A_{il}$; the elementary two-sided operator $L_{E_{ir}}R_{E_{sl}}$ carries the corner $A_{rs}$ to the corner $A_{il}$.
$$ \boxed{\ \operatorname{Der}(M_3)\cong\mathfrak{sl}_3,\ \dim=8;\qquad \operatorname{Aut}(M_3)\cong PGL_3,\ \dim=8;\qquad \operatorname{rank}\operatorname{ad}_a=9-\dim Z(a).\ } $$
| Operator | Matrix | Action on $E_{ij}$ | Trace |
|---|---|---|---|
| $L_a$ | $a\otimes I_3$ | row index by $a$ | $3\operatorname{tr}a$ |
| $R_b$ | $I_3\otimes b^{\mathsf T}$ | column index by $b$ | $3\operatorname{tr}b$ |
| $L_{a}R_{b}$ | $a\otimes b^{\mathsf T}$ | both indices | $\operatorname{tr}a\operatorname{tr}b$ |
| $\operatorname{ad}_a$ | $a\otimes I_3-I_3\otimes a^{\mathsf T}$ | commutator | $0$ |
Summary of Notation
| Symbol | Meaning |
|---|---|
| $L_a$ | left multiplication $X\mapsto aX$ |
| $R_a$ | right multiplication $X\mapsto Xa$ |
| $L_{a}R_{b}$ | two-sided operator $X\mapsto aXb$ |
| $\operatorname{ad}_a$ | adjoint $X\mapsto[a,X]$ |
| $\operatorname{Der}(M_3)$ | the derivation algebra, $\cong\mathfrak{sl}_3$ |
| $\operatorname{Aut}(M_3)$ | the automorphism group, $\cong PGL_3$ |
| $A_{ij}=\mathbb{C}E_{ij}$ | the Peirce corner $(i,j)$ |
| $Z(a)$ | the centraliser of $a$ |
Further Reading
- Richard S. Pierce, Associative Algebras (Springer, 1982), for derivations, automorphisms and the Skolem–Noether theorem.
- Tsit-Yuen Lam, A First Course in Noncommutative Rings (Springer, 2001), for central simple algebras and their automorphisms.
- Roger A. Horn and Charles R. Johnson, Matrix Analysis (Cambridge, 2013), for the Kronecker product, vectorisation and the centraliser of a Jordan form.
- The Peirce Decomposition of $M_3(\mathbb{C})$, for the corners and the multiplication rule.
- The Rank-Two Split of $M_3(\mathbb{C})$, for the two-sided Peirce table.
- The Adjoint Action, the Orbits and the Flag Variety of $M_3(\mathbb{C})$, for the conjugation action and its orbits.