The Peirce Decomposition in Two Dimensions
Introduction
A rank-one idempotent $e$ splits $M_2(\mathbb{C})$ into four Peirce spaces $A_{11}$, $A_{12}$, $A_{21}$, $A_{00}$ of dimension one each, and the product of two elements follows the index rule $A_{ij}A_{jk}\subseteq A_{ik}$.
The article defines the Peirce spaces as the joint eigenspaces of the two multiplication operators $L_{e}$ and $R_{e}$, proves the index rule, computes the four dimensions, writes the spaces explicitly for the idempotent $E_{11}$ and for a general rank-one idempotent $uv^{\mathsf T}$, and compares the decomposition with the splitting into the two halves. It assumes Lagrange Idempotents of a $2\times2$ Matrix with Simple Spectrum, The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$ and The Two Halves over $\mathbb{C}$: Hermitian and Anti-Hermitian at $n=2$.
It defers the basis adapted to the frame and the gauge to The Peirce Basis and the Corners at $n=2$, and the Jordan-algebra reading to The Peirce Decomposition of $\operatorname{Sym}_2$ as a Jordan Algebra.
The Two Operators of an Idempotent
Definition. For an element $e$ the left and right multiplication operators are $L_{e}(x)=ex$ and $R_{e}(x)=xe$.
Proposition. For an idempotent the two operators satisfy $L_{e}^{2}=L_{e}$ and $R_{e}^{2}=R_{e}$, and they commute.
Proof. $e(ex)=(e^{2})x=ex$ and $(xe)e=xe^{2}=xe$; and $e(xe)=(ex)e$ by associativity.
Corollary. Each operator is a projection of the eight-dimensional real space, so the space splits into the kernel and the image of each, and jointly into four pieces.
Proof. A projection with commuting partner gives the joint splitting into the four combinations of the two images, and each piece is the intersection of an image or a kernel with an image or a kernel.
Definition. The Peirce spaces are
$$ A_{11}=\operatorname{im}L_{e}\cap\operatorname{im}R_{e},\quad A_{12}=\operatorname{im}L_{e}\cap\ker R_{e},\quad A_{21}=\ker L_{e}\cap\operatorname{im}R_{e},\quad A_{00}=\ker L_{e}\cap\ker R_{e}. $$
Corollary. In coordinates the four spaces are the conditions $ex=x$ or $ex=0$ together with $xe=x$ or $xe=0$.
Proof. The image of $L_{e}$ is the set of fixed vectors of $L_{e}$ and its kernel is the set of vectors sent to zero; combining the two conditions gives the four displays.
The Dimensions
Theorem. For a rank-one idempotent the four Peirce spaces have dimension one each, and the dimensions sum to four, the complex dimension of the algebra.
Proof. Choose a basis with $e=E_{11}$, which is possible because two idempotents of the same rank are conjugate. Then the four spaces are the spans of $E_{11}$, $E_{12}$, $E_{21}$, $E_{22}$, each of dimension one, and $1+1+1+1=4$.
Corollary. The decomposition does not depend on the idempotent chosen, only on its rank.
Proof. Any two rank-one idempotents are conjugate, and conjugation carries the Peirce spaces of one to those of the other.
Corollary. For the idempotents $0$ and $I$ the decomposition degenerates: all of the space lies in one Peirce piece.
Proof. The operator $L_{0}$ is zero, so the only nonempty piece is the one with both indices zero; and $L_{I}$ is the identity, so the only nonempty piece is the one with both indices one.
The Multiplication Rules
Theorem. The Peirce spaces multiply by the index rule
$$ A_{ij}A_{jk}\subseteq A_{ik},\qquad A_{ij}A_{kl}=0\ \text{when}\ j\ne k . $$
Proof. Take $x\in A_{ij}$ and $y\in A_{kl}$. Then $e(xy)=(ex)y=(ix)y=ixy$ and $(xy)e=x(ye)=x(ly)=l\,xy$, so the product lies in $A_{il}$ when $j=k$; when $j\ne k$ one of the two computations gives $0$ because $j$ and $k$ are different eigenvalues of the same projection.
Corollary. The mixed spaces are square-zero: $A_{12}^{2}=0$ and $A_{21}^{2}=0$.
Proof. The index rule with $j=1$, $k=2$ gives a product in $A_{12}A_{21}\subseteq A_{11}$ for one order and $A_{21}A_{12}\subseteq A_{00}$ for the other; the products of two elements of the same mixed space vanish by the second rule.
Corollary. $A_{12}A_{21}\subseteq A_{11}$ and $A_{21}A_{12}\subseteq A_{00}$, and both products are nonzero at the standard idempotent.
Proof. $E_{12}E_{21}=E_{11}$ and $E_{21}E_{12}=E_{22}$ by the multiplication of the matrix units.
The Standard Idempotent
Proposition. At $e=E_{11}$ the four spaces are the four lines of the matrix units.
Proof. $E_{11}E_{11}=E_{11}$ and $E_{11}E_{22}=0$ give the diagonal pieces; $E_{11}E_{12}=E_{12}$ and $E_{12}E_{11}=0$ put $E_{12}$ in $A_{12}$; and $E_{21}$ is placed symmetrically in $A_{21}$.
Corollary. In the Peirce basis $(e,\upsilon_{+},\upsilon_{-},1-e)$ the matrix of left multiplication by $e$ is $\operatorname{diag}(1,1,0,0)$ and the matrix of right multiplication is $\operatorname{diag}(1,0,1,0)$.
Proof. The left multiplication fixes $A_{11}$ and $A_{12}$ and kills the other two, and the right multiplication fixes $A_{11}$ and $A_{21}$.
Corollary. An element $x$ decomposes as $x=x_{11}+x_{12}+x_{21}+x_{00}$ with $x_{11}=exe$, $x_{12}=ex(1-e)$, $x_{21}=(1-e)xe$, $x_{00}=(1-e)x(1-e)$.
Proof. Apply the two projections in the two orders, which gives the four combinations, and sum them, which gives back $x$.
A General Rank-One Idempotent
Proposition. For $e=uv^{\mathsf T}$ with $v^{\mathsf T}u=1$ the four spaces are the lines spanned by $uv^{\mathsf T}$, $u\,w^{\mathsf T}$, $z\,v^{\mathsf T}$ and $zw^{\mathsf T}$, where $w$ and $z$ are vectors with $v^{\mathsf T}w=0$ and $z^{\mathsf T}u=0$.
Proof. $e(uw^{\mathsf T})=u(v^{\mathsf T}u)w^{\mathsf T}=uw^{\mathsf T}$ and $(uw^{\mathsf T})e=u(w^{\mathsf T}u)v^{\mathsf T}=0$, so $uw^{\mathsf T}$ lies in $A_{12}$; the other three cases are similar, and each space is one-dimensional.
Corollary. The four spaces are the four corners $eM_2e$, $eM_2(1-e)$, $(1-e)M_2e$, $(1-e)M_2(1-e)$.
Proof. The first corner consists of the elements $exe$, which is the intersection of the two images, and the others are the same computation with one or both projections replaced by their complements.
Corollary. The corner $A_{12}$ consists of the matrices whose image lies in the line of $u$ and whose kernel contains the line of $v$, so it is the space of rank-one maps in a fixed direction.
Proof. $ex(1-e)=x(1-e)$ for $x\in A_{12}$, so the image of $x$ lies in the image of $e$ and the kernel contains the kernel of $e$.
The Comparison with the Two Halves
Proposition. The Peirce splitting refines neither half: the symmetric part of $A_{12}$ is not contained in $A_{12}$.
Proof. For $x=uw^{\mathsf T}$ the transpose is $wu^{\mathsf T}$, which lies in $A_{21}$ rather than in $A_{12}$, so the symmetric part $x+x^{\mathsf T}$ has components in both mixed spaces.
Corollary. The transpose exchanges $A_{12}$ and $A_{21}$ and fixes $A_{11}$ and $A_{00}$.
Proof. The transpose of $ex(1-e)$ is $(1-e)x^{\mathsf T}e$, which lies in the corner with the indices exchanged.
Corollary. The symmetric half is the graph of a map from $A_{11}\oplus A_{00}$ to the mixed spaces, of dimension three.
Proof. A symmetric element is determined by its two diagonal pieces, of dimension two, and one mixed component, of dimension one, which gives three.
The Orthogonality of the Four Spaces
Proposition. The four Peirce spaces are mutually orthogonal for the trace form, except the two mixed ones, whose pairing is $\operatorname{tr}(E_{12}E_{21})=1$.
Proof. $L_{e}$ is self-adjoint for the trace form because $\operatorname{tr}(exy)=\operatorname{tr}(xey)$, so its eigenspaces are orthogonal; the two mixed spaces share the eigenvalue of $L_{e}$ and are separated only by $R_{e}$, which is not self-adjoint, and the computation at the matrix units gives the value $1$.
Corollary. The two blocks $A_{11}\oplus A_{12}$ and $A_{21}\oplus A_{00}$ are orthogonal and of dimension two each.
Proof. They are the two eigenspaces of $L_{e}$, of eigenvalues $1$ and $0$, and $L_{e}$ is self-adjoint.
The Corpus Translation
Proposition. The four Peirce spaces are the four corners $\mathbb{B}_{st}$ of the biquaternion algebra.
Proof. The corpus writes the corners as the four products of the idempotent pair with the algebra on the two sides, which is the same four-corner pattern as the display of this article, under the matrix model of $M_2(\mathbb{C})$ Pauli Representation.
Corollary. The dimensions one, one, one, one are the dimensions of the corners, and their sum is the dimension of the algebra over the field.
Proof. Each corner is spanned by one matrix unit and its multiple by the scalar $i$, so it is one complex line, and the four lines add to dimension four.
Summary
A rank-one idempotent $e$ gives two commuting projections $L_{e}$ and $R_{e}$, and their joint eigenspaces are the four Peirce spaces $A_{11}$, $A_{12}$, $A_{21}$, $A_{00}$, of dimension one each. They are the four corners $eMe$, $eM(1-e)$, $(1-e)Me$, $(1-e)M(1-e)$, they multiply by the index rule $A_{ij}A_{jk}\subseteq A_{ik}$ and vanish when the middle indices differ, and the mixed spaces are square-zero.
At the standard idempotent the spaces are the four lines of the matrix units, and for a general idempotent $uv^{\mathsf T}$ they are the lines of $uv^{\mathsf T}$, $uw^{\mathsf T}$, $zv^{\mathsf T}$ and $zw^{\mathsf T}$. The four are mutually orthogonal for the trace form except the two mixed ones, and the transpose exchanges the mixed pair.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $e$ | a rank-one idempotent |
| $L_{e},R_{e}$ | left and right multiplication by $e$ |
| $A_{ij}$ | the Peirce spaces, $i,j\in\{1,0\}$ |
| $u,v$ | the vectors with $e=uv^{\mathsf T}$ and $v^{\mathsf T}u=1$ |
| $w,z$ | the vectors orthogonal to $v$ and to $u$ |
Further Reading
- Kevin McCrimmon, A Taste of Jordan Algebras (Springer, 2004), for the Peirce decomposition of an algebra by an idempotent.
- Richard S. Pierce, Associative Algebras (Springer, 1982), for the Peirce spaces of a finite-dimensional algebra and the corner notation.
- Lagrange Idempotents of a $2\times2$ Matrix with Simple Spectrum, for the idempotents used as the frame.
- The Peirce Basis of the Biquaternion Algebra, for the same decomposition in the biquaternion language.
- The Two Halves over $\mathbb{R}$: Symmetric and Antisymmetric at $n=2$, for the splitting that the Peirce splitting refines.