Left and Right Multiplication on the Diagonal Peirce Part
Introduction
Left and right multiplication by an element $a$ are the operators $L_a(x)=ax$ and $R_a(x)=xa$. They commute, and their product is the two-sided operator $L_{a}R_{b}$:
$$ L_aR_b=R_bL_a=L_{a}R_{b}. $$
An idempotent $e$ gives the two operators $L_e$ and $R_e$, and the Peirce decomposition of $A$ is the joint eigenspace decomposition of this commuting pair. The diagonal Peirce part is
$$ D=A_{00}\oplus A_{11}=eAe\oplus fAf, \qquad f=1-e, $$
and it is exactly the centraliser of $e$: the elements on which $L_e$ and $R_e$ agree. The off-diagonal part is $N=A_{01}\oplus A_{10}$, and the commutator $L_e-R_e=\operatorname{ad}_e$ vanishes on $D$, acts as $+1$ on $A_{01}$ and as $-1$ on $A_{10}$, and satisfies
$$ \boxed{\;(L_e-R_e)^2=P_N,\qquad P_D=I-L_e-R_e+2L_eR_e\;}, $$
where $P_D$ and $P_N$ are the projections onto $D$ and $N$. The article proves these identities, computes the restrictions of $L_d$ and $R_d$ to the diagonal part for $d\in D$, and reads the diagonal part as the algebra $eAe\times fAf$ acting on itself by left and right multiplication.
The article is the diagonal companion of Left and Right Multiplication, The Commutator Operator and The Two-Sided Operators on the Peirce Components. The Peirce decomposition is that of Unital Algebras, §Idempotents and the Peirce Decomposition; the centraliser theorem is The Centraliser and the Diagonal Peirce Part; the off-diagonal part is Off-Diagonal Peirce Components and Noncentral Idempotents; the derivations are Derivations and the Peirce Decomposition; and the two-sided multiplication is The Two-Sided Operator on an Algebra. No form, involution or topology is used.
Throughout, $k$ is a field and $A$ is a unital associative $k$-algebra with $1\neq0$, not assumed commutative and not assumed finite-dimensional unless a statement says so. The idempotent is $e$, its complement is $f=1-e$, and the corners are $A_{ij}=e_iAe_j$ with $e_0=e$, $e_1=f$. The Peirce decomposition is $A=A_{00}\oplus A_{01}\oplus A_{10}\oplus A_{11}$, with the multiplication table $A_{ij}A_{kl}\subseteq\delta_{jk}A_{il}$. The diagonal part is $D=A_{00}\oplus A_{11}$ and the off-diagonal part is $N=A_{01}\oplus A_{10}$; the projections onto them are $P_D$ and $P_N$.
Left and Right Multiplication
Definitions and elementary identities
Definition. For $a\in A$, the left multiplication $L_a$ and the right multiplication $R_a$ are the $k$-linear operators
$$ L_a:A\to A,\quad x\mapsto ax, \qquad R_a:A\to A,\quad x\mapsto xa. $$
They are the regular representations of $A$ on itself.
Proposition. For all $a,b\in A$:
(a) $L_{a+b}=L_a+L_b$ and $R_{a+b}=R_a+R_b$;
(b) $L_{ab}=L_aL_b$ and $R_{ab}=R_bR_a$;
(c) $L_aR_b=R_bL_a=L_{a}R_{b}$;
(d) $L_1=R_1=\mathrm{id}_A$;
(e) $L_a=0$ exactly when $a=0$, and $R_a=0$ exactly when $a=0$.
Proof. (a), (b) and (c) are associativity: $L_{ab}(x)=(ab)x=a(bx)=L_aL_b(x)$, $R_{ab}(x)=x(ab)=(xa)b=R_bR_a(x)$, and $L_aR_b(x)=a(xb)=(ax)b=R_bL_a(x)$. (d) is the unit. (e) $L_a=0$ gives $a=L_a(1)=0$, and $R_a=0$ gives $a=R_a(1)=0$.
Remark. The left regular representation $a\mapsto L_a$ is an algebra homomorphism $A\to\operatorname{End}_k(A)$, and the right regular representation $a\mapsto R_a$ is an algebra anti-homomorphism. The two representations commute, and their product is the two-sided representation of the enveloping algebra $A\otimes A_{\mathrm{opp}}$. The identities are those of Left and Right Multiplication; the article uses them without further proof.
The operators of an idempotent
Proposition. Let $e$ be an idempotent and $f=1-e$.
(a) $L_e$ and $R_e$ are idempotent operators: $L_e^2=L_e$, $R_e^2=R_e$.
(b) $L_e$ and $R_e$ commute, and $L_eR_e=R_eL_e=P_e=L_{e}R_{e}$.
(c) $L_e+L_f=\mathrm{id}_A$ and $R_e+R_f=\mathrm{id}_A$.
(d) The images and kernels are
$$ \boxed{\;\operatorname{im}L_e=eA, \quad \ker L_e=fA, \quad \operatorname{im}R_e=Ae, \quad \ker R_e=Af\;}. $$
(e) The image of $P_e=L_eR_e$ is the corner $eAe$, and its kernel is $eAf\oplus fAe\oplus fAf$.
Proof. (a) $L_e^2=L_{e^2}=L_e$ and $R_e^2=R_{e^2}=R_e$. (b) $L_eR_e=L_{e}R_{e}=R_eL_e$. (c) $L_e+L_f=L_{e+f}=L_1=\mathrm{id}$, and similarly for $R$. (d) $\operatorname{im}L_e=\{ex:x\in A\}=eA$, and the Peirce decomposition gives $\ker L_e=\{x:ex=0\}=fA$; the right-hand statements are $\operatorname{im}R_e=Ae$ and $\ker R_e=Af$. (e) $P_e(x)=exe$ lies in $eAe$, and $P_e$ fixes $eAe$; its kernel is the sum of the other three corners.
Remark (the two projections). $L_e$ is the projection onto the right ideal $eA$ along $fA$, and $R_e$ is the projection onto the left ideal $Ae$ along $Af$. They are not the Peirce projections: the Peirce corner $A_{00}=eAe$ is the intersection of their images, and the Peirce projection onto it is $P_e=L_eR_e$, not $L_e$ or $R_e$. The four Peirce projections are the elementary two-sided operators $E_{ij}=L_{e_i}R_{e_j}$ of The Two-Sided Operators on the Peirce Components.
The joint eigenspaces
Because $L_e$ and $R_e$ commute and are idempotent, they are simultaneously diagonalisable, with eigenvalues in $\{0,1\}$. The joint eigenspaces are the four Peirce corners:
| Corner | $L_e$ | $R_e$ | $L_e-R_e$ | $L_e+R_e$ |
|---|---|---|---|---|
| $A_{00}=eAe$ | $1$ | $1$ | $0$ | $2$ |
| $A_{01}=eAf$ | $1$ | $0$ | $1$ | $1$ |
| $A_{10}=fAe$ | $0$ | $1$ | $-1$ | $1$ |
| $A_{11}=fAf$ | $0$ | $0$ | $0$ | $0$ |
Proof. For $x\in A_{ij}$, $L_e(x)=e x$ is nonzero only when $i=0$, and then equals $x$; $R_e(x)=x e$ is nonzero only when $j=0$, and then equals $x$. The table follows.
Corollary. The pair $(L_e,R_e)$ determines the Peirce decomposition: the corner $A_{ij}$ is the joint eigenspace with eigenvalues $(1-i,1-j)$, where $i,j\in\{0,1\}$. The diagonal part $D$ is the sum of the joint eigenspaces with pairs $(1,1)$ and $(0,0)$, and the off-diagonal part $N$ is the sum of the joint eigenspaces with pairs $(1,0)$ and $(0,1)$.
Remark. The table is the operator form of the Peirce decomposition. The diagonal corners are the two extremes $(1,1)$ and $(0,0)$, and the mixed corners are the two mixed pairs. The operator $L_e-R_e$ separates the diagonal from the off-diagonal, and the operator $L_e+R_e$ separates the two diagonal corners from each other only through the value $2$ versus $0$; in characteristic two the value $2$ is zero, and $L_e+R_e$ no longer distinguishes $A_{00}$ from $A_{11}$. The corner projections $P_e=L_eR_e$ and $P_f=L_fR_f$ distinguish them in every characteristic.
The Diagonal Peirce Part
Definition and the commutator identity
Definition. The diagonal Peirce part is $D=A_{00}\oplus A_{11}$, and the off-diagonal part is $N=A_{01}\oplus A_{10}$. The projections onto them are $P_D$ and $P_N=I-P_D$.
The diagonal part is the set of elements whose Peirce decomposition has no mixed component, and the off-diagonal part is the set of elements whose decomposition has only mixed components.
Theorem. The following hold.
(a) $D=\ker(L_e-R_e)=\ker(\operatorname{ad}_e)$, where $\operatorname{ad}_e=L_e-R_e$ is the commutator with $e$.
(b) $N=\operatorname{im}(\operatorname{ad}_e)$, and $\operatorname{ad}_e$ acts as $+1$ on $A_{01}$ and as $-1$ on $A_{10}$.
(c) $\operatorname{ad}_e^2=P_N$, and $P_D=I-L_e-R_e+2L_eR_e$.
(d) $D$ is a subalgebra with identity $1$, isomorphic to $A_{00}\times A_{11}$; $N$ is a $D$-bimodule, and $N^2\subseteq D$. Hence $A=D\oplus N$ is a $\mathbb{Z}/2$-graded algebra with even part $D$ and odd part $N$.
Proof. (a) For $a=\sum a_{ij}$, $e a=a_{00}+a_{01}$ and $a e=a_{00}+a_{10}$, so $(L_e-R_e)(a)=a_{01}-a_{10}$. This vanishes exactly when $a_{01}=a_{10}=0$, that is, when $a\in D$. (b) The same computation shows that the image is $N$, with $+1$ on $A_{01}$ and $-1$ on $A_{10}$. (c) On $D$, $\operatorname{ad}_e=0$; on $N$, $\operatorname{ad}_e^2=1$. Hence $\operatorname{ad}_e^2=P_N$. The formula for $P_D$ follows from $P_D=I-P_N$ and $P_N=(L_e-R_e)^2=L_e+R_e-2L_eR_e$, using $L_e^2=L_e$, $R_e^2=R_e$ and $L_eR_e=R_eL_e$. (d) $D$ is closed under multiplication because $A_{00}^2\subseteq A_{00}$, $A_{11}^2\subseteq A_{11}$ and the cross products $A_{00}A_{11}$, $A_{11}A_{00}$ vanish; it contains $1=e+f$. The map $D\to A_{00}\times A_{11}$, $d\mapsto(e d e,f d f)$, is an algebra isomorphism with inverse $(x,y)\mapsto x+y$. The bimodule and grading statements follow from the multiplication table: $D N\subseteq N$, $N D\subseteq N$, $N^2\subseteq D$.
Corollary (the centraliser). $D=C_A(e)$.
Proof. $a\in C_A(e)$ exactly when $ea=ae$, that is, when $\operatorname{ad}_e(a)=0$; by (a) this is $D$. The computation is the centraliser theorem of The Centraliser and the Diagonal Peirce Part.
Remark (the meaning of the identities). The commutator $\operatorname{ad}_e$ is a partial involution: it is zero on the diagonal part and squares to the identity on the off-diagonal part. The operator $P_N=(L_e-R_e)^2$ is the projection onto $N$, and $P_D=I-P_N$ is the projection onto the diagonal. The diagonal part is the kernel of the commutator, and the off-diagonal part is its image; the decomposition $A=D\oplus N$ is the eigenspace decomposition of $\operatorname{ad}_e$ with eigenvalues $0,\pm1$, with the two off-diagonal corners carrying the two signs. The identities are characteristic-free; in characteristic two the signs $+1$ and $-1$ coincide, and $\operatorname{ad}_e$ is the projection onto $N$.
The diagonal operators
Proposition. Let $d=d_0+d_1\in D$ with $d_0\in A_{00}$, $d_1\in A_{11}$.
(a) $L_d$ and $R_d$ preserve each Peirce corner.
(b) Relative to the ordered corners $(A_{00},A_{01},A_{10},A_{11})$, the block matrices are
$$ \boxed{\;L_d=\operatorname{diag}(L_{d_0},L_{d_0},L_{d_1},L_{d_1}), \qquad R_d=\operatorname{diag}(R_{d_0},R_{d_1},R_{d_0},R_{d_1})\;}, $$
where each entry is the left or right multiplication of the corresponding corner algebra.
(c) $L_d$ and $R_d$ preserve $D$ and $N$, and their restrictions to $D$ are the left and right regular representations of $D$; in particular the maps $d\mapsto L_d|_D$ and $d\mapsto R_d|_D$ are faithful.
(d) For $d,d'\in D$, the two-sided operator $L_{d}R_{d'}$ preserves each corner and acts on $A_{ij}$ by the two-sided operator of the corner elements $d_i$ and $d'_j$.
Proof. (a) For $x\in A_{ij}$, $d x=(d_0+d_1)x$; the term $d_0x$ is nonzero only when $i=0$, and $d_1x$ only when $i=1$, so $dx\in A_{ij}$. The same for $xd$. (b) On $A_{00}$ and $A_{01}$, the left factor is $d_0$; on $A_{10}$ and $A_{11}$, it is $d_1$. On $A_{00}$ and $A_{10}$, the right factor is $d_0$; on $A_{01}$ and $A_{11}$, it is $d_1$. (c) The preservation is (a); the restrictions are the regular representations because $D$ is a subalgebra with the same multiplication. Faithfulness follows from $L_d(1)=d$ and $R_d(1)=d$. (d) is the corner computation $d_i x d'_j$ for $x\in A_{ij}$.
Remark (the asymmetry). The left multiplication $L_d$ uses $d_0$ on the left of the two corners with first index $0$, and $d_1$ on the left of the two corners with first index $1$; the right multiplication $R_d$ uses $d_0$ on the right of the two corners with second index $0$, and $d_1$ on the right of the two corners with second index $1$. The two block matrices are not the same, and they differ exactly on the mixed corners: on $A_{01}$, $L_d$ uses $d_0$ and $R_d$ uses $d_1$; on $A_{10}$, $L_d$ uses $d_1$ and $R_d$ uses $d_0$. The diagonal corners see the same factor on both sides.
The compression of a general multiplication
For a general $a\in A$, the left multiplication does not preserve the corners, but its compression to $D$ has a clean form.
Proposition. Let $a=\sum a_{ij}\in A$, and let $P_D$ be the projection onto $D$ along $N$.
(a) $P_D L_a|_D=L_{a_{00}}\oplus L_{a_{11}}$ and $P_D R_a|_D=R_{a_{00}}\oplus R_{a_{11}}$.
(b) $P_N L_a|_D$ sends $A_{00}$ to $A_{10}$ by $x\mapsto a_{10}x$ and $A_{11}$ to $A_{01}$ by $x\mapsto a_{01}x$; $P_N R_a|_D$ sends $A_{00}$ to $A_{01}$ by $x\mapsto x a_{01}$ and $A_{11}$ to $A_{10}$ by $x\mapsto x a_{10}$.
Proof. For $x=x_0+x_1\in D$, $a x=a_{00}x_0+a_{01}x_1+a_{10}x_0+a_{11}x_1$. The first and last terms lie in $D$, and the middle two in $N$; the components are as stated. The right-hand computation is the same with $x a$.
Corollary. The diagonal compression of $L_a$ depends only on the diagonal part $a_{00}+a_{11}$ of $a$; the off-diagonal part of $a$ acts on $D$ only through the mixed compressions, and it maps $D$ to $N$.
Remark. The proposition is the operator form of the Peirce multiplication table. The diagonal part of $a$ acts on the diagonal part of $A$ by left multiplication, the off-diagonal part of $a$ maps the diagonal part to the off-diagonal part, and the mixed corners carry the two directions. The formula is the first block column of the full block matrix of $L_a$ relative to $A=D\oplus N$; the other columns are the same computation on $N$.
Examples
The matrix algebra $M_2(k)$
Let $e=E_{11}$, $f=E_{22}$, and use the corner-adapted basis $(E_{11},E_{12},E_{21},E_{22})$. The operators are diagonal in this basis:
$$ L_e=\operatorname{diag}(1,1,0,0), \qquad R_e=\operatorname{diag}(1,0,1,0), \qquad P_e=L_eR_e=\operatorname{diag}(1,0,0,0), $$
$$ \operatorname{ad}_e=L_e-R_e=\operatorname{diag}(0,1,-1,0), \qquad \operatorname{ad}_e^2=\operatorname{diag}(0,1,1,0)=P_N. $$
The diagonal part is $D=kE_{11}\oplus kE_{22}$, the algebra of diagonal matrices; the off-diagonal part is $N=kE_{12}\oplus kE_{21}$. The centraliser of $e$ is $D$, and $\operatorname{ad}_e$ is the commutator with $e$. For $d=\operatorname{diag}(d_0,d_1)$, the diagonal operators are
$$ L_d=\operatorname{diag}(d_0,d_0,d_1,d_1), \qquad R_d=\operatorname{diag}(d_0,d_1,d_0,d_1). $$
The example is the model of the whole article: the two diagonal entries of $d$ act on the two rows and the two columns of the matrix, and the mixed corners see the row factor on the left and the column factor on the right.
The split algebra $k\times k$
Let $A=k\times k$ and $e=(1,0)$, $f=(0,1)$. The mixed corners vanish, so $N=0$ and $D=A$. The operators are
$$ L_e=R_e=P_e=\operatorname{diag}(1,0) $$
on the two coordinates, and $\operatorname{ad}_e=0$. The idempotent is central, the diagonal part is the whole algebra, and the Peirce decomposition is the product decomposition $A=A_{00}\times A_{11}$. The example is the extreme in which $L_e=R_e$ and the commutator vanishes identically.
The upper triangular algebra
Let $A=T_2(k)$ and $e=E_{11}$. The corners are
$$ A_{00}=kE_{11}, \qquad A_{01}=kE_{12}, \qquad A_{10}=0, \qquad A_{11}=kE_{22}. $$
The diagonal part is $D=kE_{11}\oplus kE_{22}$, the diagonal matrices; the off-diagonal part is $N=kE_{12}$. The commutator is
$$ \operatorname{ad}_e(E_{11})=0, \qquad \operatorname{ad}_e(E_{12})=E_{12}, \qquad \operatorname{ad}_e(E_{22})=0, $$
so $\operatorname{ad}_e$ is the projection onto $N$, and $\operatorname{ad}_e^2=\operatorname{ad}_e=P_N$ on this algebra. The example is the degenerate case in which one mixed corner vanishes and the commutator has only the eigenvalue $+1$ on the off-diagonal part. The diagonal operators $L_d$ and $R_d$ preserve the three nonzero corners by the block formula of §The diagonal operators.
The biquaternion algebra
Let $A=\mathbb{B}=M_2(\mathbb{C})$ and let $e=E_{11}$, $f=E_{22}$. The diagonal part is $D=\mathbb{C}e\oplus\mathbb{C}f\cong\mathbb{C}\times\mathbb{C}$; the off-diagonal part is $N=\mathbb{C}E_{12}\oplus\mathbb{C}E_{21}$. The commutator $\operatorname{ad}_e$ has eigenvalues $0,\pm1$, and its kernel is $D$, the centraliser of $e$. The example is the biquaternion case of The Centraliser and the Diagonal Peirce Part and Off-Diagonal Peirce Components and Noncentral Idempotents, read through the left and right multiplications; the diagonal part is the algebra of the two complex lines, and the off-diagonal part is the complement. The quaternion division algebra $\mathbb{H}$ itself has no nontrivial idempotent, so it has no such decomposition; the biquaternion algebra is the matrix model in which the idempotent exists.
Summary
Left and right multiplication commute, and their product is the two-sided operator: $L_aR_b=R_bL_a=L_{a}R_{b}$. For an idempotent $e$, the operators $L_e$ and $R_e$ are commuting idempotents whose joint eigenspaces are the four Peirce corners: $A_{00}$ with $(1,1)$, $A_{01}$ with $(1,0)$, $A_{10}$ with $(0,1)$, and $A_{11}$ with $(0,0)$. The diagonal Peirce part $D=A_{00}\oplus A_{11}$ is the kernel of the commutator $\operatorname{ad}_e=L_e-R_e$, the centraliser of $e$, and a subalgebra with identity $1$ isomorphic to $A_{00}\times A_{11}$. The off-diagonal part $N=A_{01}\oplus A_{10}$ is the image of $\operatorname{ad}_e$; the commutator acts as $+1$ on $A_{01}$ and $-1$ on $A_{10}$, and $\operatorname{ad}_e^2=P_N$, $P_D=I-L_e-R_e+2L_eR_e$. The decomposition $A=D\oplus N$ is a $\mathbb{Z}/2$-grading: $D$ is a subalgebra, $N$ is a $D$-bimodule, and $N^2\subseteq D$. For $d=d_0+d_1\in D$, the operators $L_d$ and $R_d$ preserve each corner and have block matrices $\operatorname{diag}(L_{d_0},L_{d_0},L_{d_1},L_{d_1})$ and $\operatorname{diag}(R_{d_0},R_{d_1},R_{d_0},R_{d_1})$; their restrictions to $D$ are the faithful left and right regular representations of $D$. The matrix algebra $M_2(k)$, the split algebra $k\times k$, the upper triangular algebra and the biquaternion algebra are the worked examples. The article uses no form, involution or topology.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $L_a,R_a$ | left and right multiplication by $a$ |
| $L_{a}R_{b}$ | the two-sided operator $x\mapsto axb$ |
| $e,f$ | idempotent and complement $f=1-e$ |
| $A_{ij}$ | the corner $e_iAe_j$, $e_0=e$, $e_1=f$ |
| $D=A_{00}\oplus A_{11}$ | the diagonal Peirce part |
| $N=A_{01}\oplus A_{10}$ | the off-diagonal part |
| $P_D,P_N$ | the projections onto $D$ and $N$ |
| $\operatorname{ad}_e=L_e-R_e$ | the commutator with $e$ |
| $P_e=L_eR_e$ | the projection onto $A_{00}$ |
Further Reading
- Richard S. Pierce, Associative Algebras (Springer, 1982), for the Peirce decomposition and the regular representations.
- Nathan Jacobson, Basic Algebra II (Freeman, 1989), for the left and right regular representations and the centraliser theorem.
- Frank W. Anderson and Kent R. Fuller, Rings and Categories of Modules (Springer, 1992), for the module-theoretic reading of the left and right multiplications.
- John Voight, Quaternion Algebras (Springer, 2021), for the biquaternion and split-quaternion examples and the absence of idempotents in the division quaternion algebra.